{\rtf1{\fonttbl {\f2 Times New Roman;} {\f3 Times New Roman;} {\f4 Times New Roman;} {\f5 Times

New Roman Italic;} {\f6 Times New Roman;} {\f7 Times New Roman;} {\f8 Times New Roman;} {\f9 Times New Roman Italic;} {\f10 Times New Roman;} {\f11 Times New Roman;} {\f12 Times New Roman Italic;} {\f13 Times New Roman Italic;} {\f14 Times New Roman;} {\f15 Times New Roman Italic;} {\f16 Times New Roman Italic;} {\f17 Times New Roman;} {\f18 Times New Roman Italic;} {\f19 Times New Roman;} {\f20 Times New Roman;} {\f21 Times New Roman;} {\f22 Times New Roman Italic;} {\f23 Times New Roman;} {\f24 Times New Roman Italic;} {\f25 Times New Roman;} {\f26 Times New Roman Italic;} {\f27 Times New Roman;} {\f28 Times New Roman Italic;} {\f29 Times New Roman;} {\f30 Times New Roman;} {\f31 Times New Roman;} {\f32 Times New Roman;} {\f33 Times New Roman Italic;} {\f34 Times New Roman;} {\f35 Times New Roman Italic;} {\f36 Times New Roman Italic;} {\f37 Times New Roman Italic;} {\f1000000 Times New Roman;} }{\colortbl; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0;

\red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; \red0\green0\blue0; }\viewkind1\viewscale100\margl0\margr0\margt0\margb0\deftab80\dntblnsbdb\expshrt n\paperw12240\paperh15840{\bkmkstart Pg1}{\bkmkend Pg1}\par\pard\li1420\sb0\sl-1 15\slmult0\par\pard\li1420\sb0\sl-115\slmult0\par\pard\li1420\sb0\sl-115\slmult0 \par\pard\li1420\sb0\sl-115\slmult0\par\pard\li1420\sb0\sl-115\slmult0\par\pard\ li1420\sb0\sl-115\slmult0\par\pard\li1420\sb34\sl-115\slmult0\fi0\tx10300 \up0 \ expndtw0\charscalex125 \ul0\nosupersub\cf1\f2\fs10 IEEE JOURNAL ON SELECTED AREA S IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\tab \dn2 \expndtw0\charscalex1 10 1353\par\pard\li1420\sb368\sl-437\slmult0\fi2107 \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf2\f3\fs38 Trafic Models for Wireless\par\pard\li1420\sb53\sl437\slmult0\fi2275 \up0 \expndtw0\charscalex110 Communication Networks\par\pard\ li1420\sb108\sl-184\slmult0\fi1569 \up0 \expndtw0\charscalex125 \ul0\nosupersub\ cf3\f4\fs16 Kin K. Leung, \ul0\nosupersub\cf4\f5\fs14 Senior Member, IEEE, \ul0\ nosupersub\cf3\f4\fs16 William A. Massey, and Ward Whitt\par\pard\sect\sectd\sbk none\cols2\colno1\colw5947\colsr110\colno2\colw6043\colsr160\qj \li1411\sb0\sl-1 75\slmult0 \par\pard\qj \li1411\sb0\sl-175\slmult0 \par\pard\qj \li1411\sb0\sl-1 75\slmult0 \par\pard\qj \li1411\ri0\sb91\sl-175\slmult0\fi148 \up0 \expndtw0\cha rscalex128 \ul0\nosupersub\cf4\f5\fs14 Abstract\u8212? In \ul0\nosupersub\cf5\f6 \fs14 this paper, we introduce a \ul0\nosupersub\cf4\f5\fs14 deterministic fluid \line \up0 \expndtw0\charscalex127 model \ul0\nosupersub\cf5\f6\fs14 and two \u l0\nosupersub\cf4\f5\fs14 stochastic traic models \ul0\nosupersub\cf5\f6\fs14 fo r wireless networks. \line \up0 \expndtw0\charscalex132 The setting is a highway with multiple entrances and exits. \line \up0 \expndtw0\charscalex122 Vehicles are classiied as \ul0\nosupersub\cf4\f5\fs14 calling \ul0\nosupersub\cf5\f6\fs14 or \ul0\nosupersub\cf4\f5\fs14 noncalling, \ul0\nosupersub\cf5\f6\fs14 dependin g upon \line \up0 \expndtw0\charscalex128 whether or not they have calls in prog ress. The main interest \line \up0 \expndtw0\charscalex124 is in the calling veh icles, but noncalling vehicles are important \line \up0 \expndtw0\charscalex118 because they can become calling vehicles if they initiate (place or \line \up0 \ expndtw0\charscalex122 receive) a call. The deterministic model ignores the beha vior of \line \up0 \expndtw0\charscalex119 individual vehicles and treats them a s a continuous luid, whereas \line \up0 \expndtw0\charscalex129 the stochastic t rafic models consider the random behavior of \line \up0 \expndtw0\charscalex116 each vehicle. However, all three models use the same two coupled \line \up0 \exp ndtw0\charscalex125 partial differential equations \up0 \expndtw0\charscalex128 (PDE's) or ordinary differential \line \up0 \expndtw0\charscalex123 equations (O DE's) to describe the evolution of the system. The \line \up0 \expndtw0\charscal ex127 call density and call handoff rate \up0 \expndtw0\charscalex126 (or their expected values in \line \up0 \expndtw0\charscalex123 the stochastic models) are readily computable by solving these \line \up0 \expndtw0\charscalex131 equation s. Since no capacity constraints are imposed in the \line \up0 \expndtw0\charsca lex126 models, these computed quantities can be regarded as offered \line \up0 \ expndtw0\charscalex125 traffic loads. The models complement each other, because the \line \up0 \expndtw0\charscalex122 fluid model can be extended to include ad ditional features such \line \up0 \expndtw0\charscalex119 as capacity constraint s and the interdependence between velocity \line \up0 \expndtw0\charscalex131 an

d vehicular density, while the stochastic trafic model can \line \up0 \expndtw0\ charscalex121 provide probability distributions.\par\pard\qj \li1406\ri6\sb3\sl176\slmult0\fi163 \up0 \expndtw0\charscalex118 Numerical examples are presented to illustrate how the models \up0 \expndtw0\charscalex130 can be used to investi gate various aspects of time and space \line \up0 \expndtw0\charscalex124 dynami cs in wireless networks. The numerical results indicate \line \up0 \expndtw0\cha rscalex123 that both the time-dependence and the mobility of vehicles can \line \up0 \expndtw0\charscalex123 play important roles in determining system performa nce. Even \line \up0 \expndtw0\charscalex124 for systems in steady state with re spect to time, the movement \line \up0 \expndtw0\charscalex127 of vehicles and t he calling patterns can signiicantly affect the \line \up0 \expndtw0\charscalex1 24 number of calls in a given region of the system. The examples \line \up0 \exp ndtw0\charscalex123 demonstrate that the proposed models can serve as useful too ls \line \up0 \expndtw0\charscalex121 for system engineering and planning. For i nstance, we calculate \up0 \expndtw0\charscalex120 approximate call blocking pro babilities.\par\pard\ql \li1800\sb0\sl-138\slmult0 \par\pard\ql \li1800\sb0\sl-1 38\slmult0 \par\pard\ql \li2995\sb65\sl-138\slmult0\tx3191 \up0 \expndtw0\charsc alex107 \ul0\nosupersub\cf6\f7\fs12 I.\tab \up0 \expndtw0\charscalex115 INTRODUC TION\par\pard\ql \li1800\sb107\sl-184\slmult0 \up0 \expndtw0\charscalex117 \ul0\ nosupersub\cf3\f4\fs16 NLIKE a ixed, terrestrial telephone network, a wire-\par\ pard\column \qj \li6072\sb0\sl-213\slmult0 \par\pard\qj \li6072\sb0\sl-213\slmul t0 \par\pard\qj \li30\ri1529\sb168\sl-213\slmult0\fi4 \up0 \expndtw0\charscalex1 08 according to a location stochastic process. The PALM is made \line \up0 \expn dtw0\charscalex110 tractable by assuming that different customers do not \ul0\no supersub\cf4\f5\fs14 interact, \line \up0 \expndtw0\charscalex111 \ul0\nosupersu b\cf3\f4\fs16 although this behavior can be approximated indirectly. Simi-\line \up0 \expndtw0\charscalex113 larly, the PALM can be used as an offered trafic mo del that \line \up0 \expndtw0\charscalex112 serves to approximate important syst em capacity constraints \line \up0 \expndtw0\charscalex110 (e.g., the number of available radio channels) indirectly. This \line \up0 \expndtw0\charscalex107 is achieved through applying methods like the modiied offered \line \up0 \expndtw0 \charscalex117 load technique developed in \up0 \expndtw0\charscalex117 [3] to a pproximate blocking \line \up0 \expndtw0\charscalex110 probabilities in these wi reless networks. The PALM also pro-\line \up0 \expndtw0\charscalex107 vides a us eful framework for representing both time-dependent \line \up0 \expndtw0\charsca lex114 behavior and customer mobility in wireless communication \line \up0 \expn dtw0\charscalex103 networks.\par\pard\qj \li20\ri1539\sb8\sl-214\slmult0\fi182 \ up0 \expndtw0\charscalex112 The general PALM in [6] is quite abstract. Further s pecii-\line \up0 \expndtw0\charscalex112 cation is needed in order to obtain pra ctical models. Toward \line \up0 \expndtw0\charscalex118 this end, a version of the PALM was constructed to study \line \up0 \expndtw0\charscalex113 communicati ng mobiles on a highway in [7]. In this \ul0\nosupersub\cf4\f5\fs14 highway \lin e \up0 \expndtw0\charscalex112 PALM, \ul0\nosupersub\cf3\f4\fs16 vehicles altena te between thinking and calling modes \line \up0 \expndtw0\charscalex121 as they move along on a one-way, semi-ininite highway \line \up0 \expndtw0\charscalex10 8 according to a deterministic location function. Two-way trafic \line \up0 \exp ndtw0\charscalex112 and more complicated highway networks are represented by \li ne \up0 \expndtw0\charscalex110 superposing independent versions of these highwa y PALM's. \line \up0 \expndtw0\charscalex116 As with the general PALM, the syste m is assumed to have \line \up0 \expndtw0\charscalex124 no capacity constraints in terms of the number of radio \line \up0 \expndtw0\charscalex115 channels avai lable for calls. Thus, the highway PALM can \line \up0 \expndtw0\charscalex112 a lso be regarded as an offered load model, which enables us \line \up0 \expndtw0\ charscalex118 to characterize key quantities such as the call density, the \line \up0 \expndtw0\charscalex115 handoff rate, the call-origination-rate density. a nd the call-\line \up0 \expndtw0\charscalex106 termination-rate density. In [7], it is shown that these quantities \line \up0 \expndtw0\charscalex107 are relate d by two fundamental conservation equations, similar \line \up0 \expndtw0\charsc alex114 to relations in vehicular trafic theory \up0 \expndtw0\charscalex111 [2] . These results thus \line \up0 \expndtw0\charscalex116 bring together teletrafi

c theory and vehicular trafic theory.\par\pard\sect\sectd\sbknone\cols3\colno1\c olw1784\colsr20\colno2\colw4103\colsr160\colno3\colw6043\colsr160\ql \li1411\sb0 \sl-384\slmult0 \up0 \expndtw0\charscalex100 \ul0\nosupersub\cf7\f8\fs48 U\par\p ard\column \ql \li20\sb1\sl-152\slmult0 \up0 \expndtw0\charscalex115 \ul0\nosupe rsub\cf3\f4\fs16 less network must support moving customers. Due to\par\pard\col umn \ql \li20\sb74\sl-128\slmult0 \up0 \expndtw0\charscalex110 Other researchers have also observed the need for combining\par\pard\sect\sectd\sbknone\cols2\col no1\colw5892\colsr160\colno2\colw6048\colsr160\qj \li1406\ri1\sb0\sl-186\slmult0 \up0 \expndtw0\charscalex113 customer mobility, both the location and the lengt h of a call \line \up0 \expndtw0\charscalex113 in progress affect the network re sources required to support \line \up0 \expndtw0\charscalex109 the call. Custome r mobility is presenting a major challenge to \line \up0 \expndtw0\charscalex109 system designers of wireless networks [4], \up0 \expndtw0\charscalex107 [5]. Si nce wireless \line \up0 \expndtw0\charscalex115 services are becoming more popul ar, there is an increasing \line \up0 \expndtw0\charscalex118 need for mathemati cal models to help understand system \line \up0 \expndtw0\charscalex109 dynamics and analyze the performance of wireless networks.\par\pard\qj \li1406\ri6\sb0\s l-209\slmult0\fi182 \up0 \expndtw0\charscalex112 Motivated primarily by this nee d, a \ul0\nosupersub\cf4\f5\fs14 Poisson-arrival-location \line \up0 \expndtw0\c harscalex120 model \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs16 (PAL M) was introduced in \up0 \expndtw0\charscalex115 [6], in which customers \line \up0 \expndtw0\charscalex113 arrive according to a nonhomogeneous Poisson proces s and \line \up0 \expndtw0\charscalex115 move independently through a general lo cation state space\par\pard\ql \li1396\sb0\sl-138\slmult0 \par\pard\ql \li1545\s b19\sl-138\slmult0\tx3479 \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf6\f7\fs 12 Manuscript received February \up0 \expndtw-5\charscalex89 1,\tab \up0 \expndt w0\charscalex102 1994.\par\pard\ql \li1550\sb25\sl-138\slmult0 \up0 \expndtw0\ch arscalex117 K. K. Leung is with AT&T Bell Laboratories, Holmdel, NJ 07733 USA.\p ar\pard\qj \li1396\ri11\sb5\sl-158\slmult0\fi148 \up0 \expndtw0\charscalex122 W. A. Massey and W. Whitt are with AT&T Bell Laboratories, Murray \line \up0 \expn dtw0\charscalex117 Hill, NJ 07974 USA.\par\pard\ql \li1540\sb12\sl-138\slmult0 \ up0 \expndtw0\charscalex116 IEEE Log Number 9403384.\par\pard\column \ql \li25\s b0\sl-144\slmult0 \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 these two theories to study wireless networks \up0 \expndtw-5\charscalex100 [8]-[ 1 \up0 \expndtw-7\charscalex72 1 ].\par\pard\qj \li25\ri1539\sb0\sl-209\slmult0\fi 177 \up0 \expndtw0\charscalex120 While much more concrete than the general PALM in \line \up0 \expndtw0\charscalex111 [6], the highway PALM in \up0 \expndtw0\ch arscalex113 [7] is still quite abstract. Further \line \up0 \expndtw0\charscalex 123 speciication is still required in order to produce readily \line \up0 \expnd tw0\charscalex114 computable performance-related quantities. The purpose of \lin e \up0 \expndtw0\charscalex126 this paper is to develop such a version of the hi ghway \line \up0 \expndtw0\charscalex118 PALM that is indeed substantially more tractable, so that \line \up0 \expndtw0\charscalex115 all the desired quantities are computable. We obtain such a \line \up0 \expndtw0\charscalex107 more tracta ble highway PALM primarily by making additional \line \up0 \expndtw0\charscalex1 06 Markov assumptions, in the spirit of Section 8 of [6]. Hence we \line \up0 \e xpndtw0\charscalex112 call the main stochastic model introduced here the \ul0\no supersub\cf4\f5\fs14 Markovian \line \up0 \expndtw0\charscalex120 highway PALM.\ par\pard\qj \li20\ri1539\sb7\sl-211\slmult0\fi187 \up0 \expndtw0\charscalex109 \ ul0\nosupersub\cf3\f4\fs16 In the Markovian highway PALM considered here, in add i-\line \up0 \expndtw0\charscalex110 tion to having arrivals occur according to a nonhomogeneous \line \up0 \expndtw0\charscalex123 Poisson process, the state o f each vehicle is assumed to \par\pard\sect\sectd\sbknone \ql \li4896\sb0\sl-115 \slmult0 \par\pard\ql\li4896\sb54\sl-115\slmult0\tx6172 \up0 \expndtw0\charscale x140 \ul0\nosupersub\cf1\f2\fs10 0733-8716/94\up0 \expndtw0\charscalex135 $04.00 \tab \up0 \expndtw0\charscalex145 1994 IEEE \par\pard\sect\sectd\fs24\paperw122 40\paperh15840{\bkmkstart Pg2}{\bkmkend Pg2}\par\pard\li1732\sb0\sl-184\slmult0\ par\pard\li1732\sb0\sl-184\slmult0\par\pard\li1732\sb0\sl-184\slmult0\par\pard\l i1732\sb67\sl-184\slmult0\fi0\tx5347 \up0 \expndtw-7\charscalex83 \ul0\nosupersu b\cf3\f4\fs16 1354\tab \up0 \expndtw-6\charscalex84 IEEE JOURNAL ON SELECTED ARE

AS IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\sect\sectd\sbknone\c ols2\colno1\colw6209\colsr160\colno2\colw5731\colsr160\qj \li1713\sb0\sl-211\slm ult0 \par\pard\qj \li1718\ri0\sb97\sl-211\slmult0 \up0 \expndtw0\charscalex111 e volve according to a nonstationary continuous-time Markov \line \up0 \expndtw0\c harscalex116 chain, while the vehicle moves deterministically along the \line \u p0 \expndtw0\charscalex114 highway. Each vehicle on the highway is classiied as either \line \up0 \expndtw0\charscalex114 a \ul0\nosupersub\cf8\f9\fs16 calling \ul0\nosupersub\cf3\f4\fs16 or \ul0\nosupersub\cf8\f9\fs16 noncalling \ul0\nosup ersub\cf3\f4\fs16 vehicle, depending upon whether or \line \up0 \expndtw0\charsc alex110 not it has a call in progress. Each calling (noncalling) vehicle \line \ up0 \expndtw0\charscalex108 becomes a noncalling (calling) vehicle randomly with speciied \line \up0 \expndtw0\charscalex107 deterministic intensity depending o n time and space. Similarly, \line \up0 \expndtw0\charscalex113 each calling \up 0 \expndtw0\charscalex116 (noncalling) vehicle leaves the highway with \line \up 0 \expndtw0\charscalex118 a deterministic intensity depending on time and space. Of \line \up0 \expndtw0\charscalex114 course, our main interest is in the calli ng vehicles; they use \line \up0 \expndtw0\charscalex111 the most resources in w ireless networks. However, it may be \line \up0 \expndtw0\charscalex111 necessar y to keep track of locations of noncaling vehicles so \line \up0 \expndtw0\chars calex108 that they can receive calls. They are also important as a source \line \up0 \expndtw0\charscalex111 of future calling vehicles. There can be dramatic i ncreases in \line \up0 \expndtw0\charscalex110 the number of calling vehicles wh en and where the density of \line \up0 \expndtw0\charscalex111 noncalling vehicl es is high and the calling rate is high.\par\pard\qj \li1728\ri1\sb0\sl-211\slmu lt0\fi177 \up0 \expndtw0\charscalex111 In the special case of stationary intensi ties, our Markovian \line \up0 \expndtw0\charscalex115 assumption for each vehic le means that call holding times, \line \up0 \expndtw0\charscalex111 think times \up0 \expndtw0\charscalex116 (before making a call), and highway residence \lin e \up0 \expndtw0\charscalex109 times (before leaving the highway) would be indep endent ex-\line \up0 \expndtw0\charscalex106 ponential random variables with mea ns equal to the reciprocals \line \up0 \expndtw0\charscalex107 of the intensitie s. Without stationarity, these random times can \line \up0 \expndtw0\charscalex1 08 be represented as time-ordered exponentials, as in (8.3) of [6].\par\pard\qj \li1728\ri0\sb0\sl-212\slmult0\fi177 \up0 \expndtw0\charscalex123 It turns out t hat the densities of the mean numbers of \line \up0 \expndtw0\charscalex119 vehi cles of each type are described by partial differential \line \up0 \expndtw0\cha rscalex105 equations \up0 \expndtw0\charscalex116 (PDE's), similar to the ones a rising in the classic \line \up0 \expndtw0\charscalex107 approach to modeling ve hicular trafic [2]. With the Markovian \line \up0 \expndtw0\charscalex118 assump tion, the two PDE's are coupled due to the calling \line \up0 \expndtw0\charscal ex110 activities; i.e., a noncalling vehicle becomes a calling vehicle \line \up 0 \expndtw0\charscalex107 and vice versa, if it initiates (places and receives) or terminates \line \up0 \expndtw0\charscalex108 a call. In this Markovian highw ay PALM, the PDE's relate the \line \up0 \expndtw0\charscalex112 derivatives of the expected number of vehicles of each type, \line \up0 \expndtw0\charscalex109 while the actual numbers of noncalling and calling vehicles in \line \up0 \expn dtw0\charscalex108 a given section of the highway have Poisson distributions, du e \line \up0 \expndtw0\charscalex112 to previous PALM results in [6], [7].\par\p ard\qj \li1713\ri2\sb0\sl-211\slmult0\fi187 \up0 \expndtw0\charscalex112 The PDE 's can also be interpreted in another way. Instead \line \up0 \expndtw0\charscal ex114 of characterizing the \ul0\nosupersub\cf8\f9\fs16 expected values \ul0\nos upersub\cf3\f4\fs16 in a \ul0\nosupersub\cf8\f9\fs16 stochastic model, \line \up 0 \expndtw0\charscalex115 \ul0\nosupersub\cf3\f4\fs16 they can be regarded as ch aracterizing the \ul0\nosupersub\cf8\f9\fs16 actual values \ul0\nosupersub\cf3\f 4\fs16 in \line \up0 \expndtw0\charscalex117 a \ul0\nosupersub\cf8\f9\fs16 deter ministic fluid model. \ul0\nosupersub\cf3\f4\fs16 This deterministic luid model \line \up0 \expndtw0\charscalex113 neglects the behavior of individual vehicles \up0 \expndtw0\charscalex109 (or customers), \line \up0 \expndtw0\charscalex115 but is still capable of capturing the overall dynamics of the \line \up0 \expndt w0\charscalex116 system. Calling and noncalling vehicles are treated as two \lin

e \up0 \expndtw0\charscalex117 types of continuous luid. Such a model is appropr iate and \line \up0 \expndtw0\charscalex120 justiiable if the system has a large number of calling and \line \up0 \expndtw0\charscalex111 noncalling vehicles, a s discussed in Section 9 of [6]. Indeed, \line \up0 \expndtw0\charscalex109 such deterministic diferential equation models are common in \line \up0 \expndtw0\ch arscalex110 vehicular trafic theory [2]. (Wright [11] also uses differential \li ne \up0 \expndtw0\charscalex111 equations to capture the vehicle movement in a d eterministic \line \up0 \expndtw0\charscalex109 way for a highway cellular syste m. The focus in [11] is on the \line \up0 \expndtw0\charscalex110 control strate gy, e.g., call throttling, to maintain transmission \line \up0 \expndtw0\charsca lex110 quality in spread spectrum systems such as CDMA systems.)\par\pard\qj \li 1718\ri0\sb4\sl-207\slmult0\fi187 \up0 \expndtw0\charscalex121 In fact, three di ferent models are considered here: the \line \up0 \expndtw0\charscalex117 determ inistic luid model, the Markovian highway PALM, \line \up0 \expndtw0\charscalex1 17 and a stochastic generalization of the Markovian highway \line \up0 \expndtw0 \charscalex113 PALM in which the arrival process need not be Poisson. As \line \ up0 \expndtw0\charscalex117 noted in Remark 2.3 of [6], the mean value formulas for a \line \up0 \expndtw0\charscalex114 general PALM remain valid when the arri val process is \ul0\nosupersub\cf8\f9\fs16 not \line \up0 \expndtw0\charscalex13 8 \ul0\nosupersub\cf5\f6\fs14 Poisson, provided that the arrival rate is still w ell deined\par\pard\column \qj \li6369\sb0\sl-211\slmult0 \par\pard\qj \li25\ri1 212\sb102\sl-211\slmult0 \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf3\f4\fs1 6 and that successive arrivals do not interact. We use the extra \up0 \expndtw0\ charscalex108 Poisson arrival assumption only to characterize the probability \u p0 \expndtw0\charscalex111 distributions of the quantities of interest; the mean values are \up0 \expndtw0\charscalex114 determined from the PDE's.\par\pard\qj \li20\ri1222\sb2\sl-210\slmult0\fi182 \up0 \expndtw0\charscalex115 We believe th at an interesting feature of this paper is the \line \up0 \expndtw0\charscalex11 0 identiication of the three related models. 'The important point \line \up0 \ex pndtw0\charscalex121 is that all three models lead to the same PDE's. We thus \l ine \up0 \expndtw0\charscalex117 have different possible interpretations of thes e PDE's. For \line \up0 \expndtw0\charscalex120 the deterministic luid model, we interpret the solution as \line \up0 \expndtw0\charscalex115 the actual numbers \up0 \expndtw0\charscalex117 (which need not be integers), whereas \line \up0 \ expndtw0\charscalex115 for the stochastic trafic model, the solutions are interp reted \line \up0 \expndtw0\charscalex115 as the expected values. With the additi onal Poisson-arrival \line \up0 \expndtw0\charscalex115 assumption in the Markov ian highway PALM, we are able \line \up0 \expndtw0\charscalex111 to say more abo ut the full probability distributions. For large \line \up0 \expndtw0\charscalex 111 systems, the three approaches tend to be almost fully consis-\line \up0 \exp ndtw0\charscalex107 tent, because in the stochastic model, the true distribution s will \line \up0 \expndtw0\charscalex109 typically cluster relatively tightly a bout their means, by virtue \line \up0 \expndtw0\charscalex113 of the law of lar ge numbers; see Section 9 of [6]. The mean \line \up0 \expndtw0\charscalex110 va lue interpretation in the stochastic models tends to be more \line \up0 \expndtw 0\charscalex110 general, because it does not require large populations.\par\pard \qj \li20\ri1236\sb6\sl-210\slmult0\fi177 \up0 \expndtw0\charscalex120 In both t he deterministic luid model and the stochastic \line \up0 \expndtw0\charscalex11 0 trafic models, the system quantities of interest such as calling \line \up0 \e xpndtw0\charscalex111 density and handof rate (or their expected values) are rea dily \line \up0 \expndtw0\charscalex107 computable. Thus, the models can provide insight into the time \line \up0 \expndtw0\charscalex109 and space dynamics of mobile customers. Hopefully, this new \line \up0 \expndtw0\charscalex111 insight and understanding can contribute to improved design \line \up0 \expndtw0\charsc alex109 and management of wireless communication networks.\par\pard\qj \li20\ri1 236\sb1\sl-210\slmult0\fi177 \up0 \expndtw0\charscalex115 The rest of this paper is organized as follows. In Section \line \up0 \expndtw0\charscalex114 II, we d evelop the deterministic luid model. The model has \line \up0 \expndtw0\charscal ex118 two versions: a time-nonhomogeneous model and a time-\line \up0 \expndtw0\ charscalex115 homogeneous model. The two PDE's become ordinary dif-\line \up0 \e

xpndtw0\charscalex111 ferential equations (ODE's) in the time-homogeneous model. \line \up0 \expndtw0\charscalex117 There are interesting spatial phenomena even in the time-\line \up0 \expndtw0\charscalex109 homogeneous case. In Section III , we present some numerical \line \up0 \expndtw0\charscalex112 examples to illus trate the time and space dynamics captured \line \up0 \expndtw0\charscalex107 by the models. These numerical examples demonstrate that the \line \up0 \expndtw0\ charscalex110 quantities of interest can readily be computed for our models. \li ne \up0 \expndtw0\charscalex107 In addition, they also show that the proposed mo dels can serve \line \up0 \expndtw0\charscalex110 as valuable tools for system e ngineering and planning.\par\pard\qj \li20\ri1236\sb6\sl-210\slmult0\fi177 \up0 \expndtw0\charscalex116 In Section IV, we introduce the Markovian trafic model \ line \up0 \expndtw0\charscalex111 and discuss its connection with the determinis tic luid model. \line \up0 \expndtw0\charscalex112 In Section V, we discuss the additional distributional results \line \up0 \expndtw0\charscalex123 that can be obtained when arrivals occur according to a \line \up0 \expndtw0\charscalex110 nonhomogeneous Poisson process. In Section VI, we indicate \line \up0 \expndtw0\ charscalex112 how to approximately represent interactions among vehicles \line \ up0 \expndtw0\charscalex122 in the stochastic model. The idea is to allow the ve hicle \line \up0 \expndtw0\charscalex116 velocity to depend on the expected numb ers of calling and \line \up0 \expndtw0\charscalex117 noncalling vehicles in var ious regions. Finally, in Section \line \up0 \expndtw0\charscalex111 VII, we pre sent our conclusions and discuss future work. An \line \up0 \expndtw0\charscalex 118 important component of this future work is a study of the \line \up0 \expndt w0\charscalex110 approximate representations of vehicle interactions discussed \ line \up0 \expndtw0\charscalex114 in Section VI.\par\pard\ql \li7166\sb0\sl-184\ slmult0 \par\pard\ql \li7166\sb0\sl-184\slmult0 \par\pard\ql \li817\sb34\sl-184\ slmult0 \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf6\f7\fs12 II. THE DETERMN ISTIC \ul0\nosupersub\cf3\f4\fs16 FLUID \ul0\nosupersub\cf6\f7\fs12 MODEL\par\pa rd\qj \li20\ri1227\sb80\sl-201\slmult0\fi182 \up0 \expndtw0\charscalex108 \ul0\n osupersub\cf3\f4\fs16 Our basic setting is a one-way, semi-ininite highway. (As in \line \up0 \expndtw0\charscalex107 [7], independent versions of these highway s can be superposed \line \up0 \expndtw0\charscalex122 to \ul0\nosupersub\cf5\f6 \fs14 make richer models.) Thus, we can regard the location space \par\pard\sect \sectd\fs24\paperw12240\paperh15840{\bkmkstart Pg3}{\bkmkend Pg3}\par\pard\sect\ sectd\sbknone\cols2\colno1\colw6066\colsr20\colno2\colw6014\colsr160\ql \li1435\ sb0\sl-184\slmult0 \par\pard\ql \li1435\sb0\sl-184\slmult0 \par\pard\ql \li1435\ sb0\sl-184\slmult0 \par\pard\ql \li1435\sb86\sl-184\slmult0 \up0 \expndtw0\chars calex116 \ul0\nosupersub\cf1\f2\fs10 LEUNG \ul0\nosupersub\cf8\f9\fs16 et at.: \ ul0\nosupersub\cf1\f2\fs10 TRAFFIC MODELS FOR WIRELESS COMMUNICATION NETWORKS\pa r\pard\qj \li1435\sb0\sl-210\slmult0 \par\pard\qj \li1435\ri1\sb94\sl-210\slmult 0 \up0 \expndtw0\charscalex109 \ul0\nosupersub\cf3\f4\fs16 as the interval [0, o o). There are two types of vehicles, calling \line \up0 \expndtw0\charscalex110 and noncalling. Each vehicle is assumed to make at most one \line \up0 \expndtw0 \charscalex117 call at a time and each call occupies one radio channel for \line \up0 \expndtw0\charscalex112 the duration of the call. As in [6], [7], we do no t impose any \line \up0 \expndtw0\charscalex113 capacity constraints, i.e., we a ssume that there are an ininite \line \up0 \expndtw0\charscalex117 number of cha nnels available so that all calls are accepted \line \up0 \expndtw0\charscalex11 3 without blocking. Thus our model can be regarded as a way \line \up0 \expndtw0 \charscalex120 to quantify the ofered load. It is possible to add capacity \line \up0 \expndtw0\charscalex114 constraints to the model in various ways, as we il lustrate in \line \up0 \expndtw0\charscalex110 Section III.\par\pard\column \ql \li10310\sb0\sl-115\slmult0 \par\pard\ql \li10310\sb0\sl-115\slmult0 \par\pard\q l \li10310\sb0\sl-115\slmult0 \par\pard\ql \li10310\sb0\sl-115\slmult0 \par\pard \ql \li10310\sb0\sl-115\slmult0 \par\pard\ql \li10310\sb0\sl-115\slmult0 \par\pa rd\ql \li4244\sb20\sl-115\slmult0 \up0 \expndtw0\charscalex110 \ul0\nosupersub\c f1\f2\fs10 1355\par\pard\ql \li1425\sb0\sl-184\slmult0 \par\pard\ql \li25\sb143\ sl-184\slmult0 \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf5\f6\fs14 Finally, let us \ul0\nosupersub\cf3\f4\fs16 deine the rate densities as\par\pard\ql \li1 316\sb76\sl-276\slmult0\tx3131 \up0 \expndtw-11\charscalex94 \ul0\nosupersub\cf9

\f10\fs24\ul a2cn-(s, t)\ul0\nosupersub\cf10\f11\fs24 c- (x, \ul0\nosupersub\cf 12\f13\fs24 0\tab \up0 \expndtw-11\charscalex96 \ul0\nosupersub\cf9\f10\fs24\ul a2c;\ul0\nosupersub\cf10\f11\fs24 \ul0\nosupersub\cf11\f12\fs24\ul (x,\par\pard \ql \li481\sb0\sl-160\slmult0\tx1547\tx2929 \up0 \expndtw0\charscalex115 \ul0\no supersub\cf12\f13\fs24 cvs, \ul0\nosupersub\cf8\f9\fs16 t)=\tab \up0 \expndtw0\c harscalex184 axat '\tab \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf10\f11\fs 24 7 \ul0\nosupersub\cf9\f10\fs24\ul \ul0\nosupersub\cf10\f11\fs24 axat\par\par d\ql \li2367\sb0\sl-74\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1 3\f14\fs8 n\par\pard\ql \li1196\sb106\sl-276\slmult0\tx3246 \up0 \expndtw-9\char scalex92 \ul0\nosupersub\cf10\f11\fs24 a2c--(x, \ul0\nosupersub\cf8\f9\fs16 t)\t ab \up0 \expndtw0\charscalex125 a2c-\ul0\nosupersub\cf12\f13\fs24 (x,\ul0\nosupe rsub\cf8\f9\fs16 t)\par\pard\ql \li366\sb42\sl-129\slmult0\tx2094\tx2617\tx3567 \up0 \expndtw0\charscalex136 c\ul0\nosupersub\cf15\f16\fs8 q \ul0\nosupersub\cf8 \f9\fs16 (x\tab \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 and \ul 0\nosupersub\cf8\f9\fs16 c\tab \up0 \expndtw0\charscalex144 (x,t)=\tab \up0 \exp ndtw-2\charscalex100 q\par\pard\ql \li1417\sb0\sl-118\slmult0\tx3471 \up0 \expnd tw0\charscalex238 \ul0\nosupersub\cf15\f16\fs8 aXat\tab \up0 \expndtw0\charscale x114 \ul0\nosupersub\cf8\f9\fs16 Ox8t\par\pard\qj \li20\ri1519\sb51\sl-211\slmul t0\fi172 \up0 \expndtw0\charscalex112 Lemma 2.1: \ul0\nosupersub\cf3\f4\fs16 The evolution of noncalling and calling vehi-\line \up0 \expndtw0\charscalex113 cle s on the highway is governed by the PDE's:\par\pard\ql \li221\sb91\sl-184\slmult 0\tx1119 \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf14\f15\fs16\ul On(x,t)\u l0\tab \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf8\f9\fs16 0\par\pard\sect \sectd\sbknone\cols2\colno1\colw6895\colsr40\colno2\colw5165\colsr160\ql \li1440 \ri68\sb0\sl-156\slmult0\fi177\tx6491 \up0 \expndtw0\charscalex107 \ul0\nosupers ub\cf3\f4\fs16 Vehicles of both types at location \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 and time \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupers ub\cf3\f4\fs16 move forward \line \up0 \expndtw0\charscalex119 on the highway ac cording to a deterministic velocity ield \tab \up0 \expndtw-2\charscalex100 \ul0 \nosupersub\cf8\f9\fs16 Ot\par\pard\ql \li1440\sb95\sl-184\slmult0\tx6081 \up0 \ expndtw0\charscalex124 v(x, t). \ul0\nosupersub\cf3\f4\fs16 To ensure vehicle lo w in a single direction, it is\tab \up0 \expndtw0\charscalex105 and\par\pard\ql \li1440\sb23\sl-184\slmult0\tx6302 \up0 \expndtw0\charscalex117 assumed that \ul 0\nosupersub\cf8\f9\fs16 v(x, t) \up0 \expndtw0\charscalex132 \ul0\nosupersub\c f3\f4\fs16 > 0 for all \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 with \ul0\nosuper sub\cf8\f9\fs16 x \up0 \expndtw0\charscalex124 > \ul0\nosupersub\cf3\f4\fs16 0 and\tab \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf8\f9\fs16 aq(x,\ul0\nosup ersub\cf14\f15\fs16\ul t)\par\pard\qj \li1440\ri58\sb5\sl-216\slmult0\fi14\tx64 99 \up0 \expndtw0\charscalex132 \ul0\nosupersub\cf3\f4\fs16 -oo \up0 \expndtw0\ charscalex159 < t \up0 \expndtw0\charscalex121 < oo. Additionally, in order to make sure that\tab \up0 \expndtw0\charscalex128 \ul0\nosupersub\cf8\f9\fs16 8t \ line \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf3\f4\fs16 vehicles low at al l, v(x, \ul0\nosupersub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex126 > \ul0\nosup ersub\cf3\f4\fs16 0 at least for some t \up0 \expndtw0\charscalex144 > \ul0\nosu persub\cf5\f6\fs14 to\par\pard\column \ql \li25\sb0\sl-144\slmult0\tx457\tx3280 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf8\f9\fs16 +\tab \up0 \expndtw0\ch arscalex125 [n(x,t)v(x,t)] \up0 \expndtw0\charscalex148 = c;(x,t)- c;(x,t)\tab \ up0 \expndtw0\charscalex105 (2.1)\par\pard\ql \li217\sb0\sl-128\slmult0 \up0 \ex pndtw0\charscalex102 Ox\par\pard\ql \li1425\sb0\sl-184\slmult0 \par\pard\ql \li1 425\sb0\sl-184\slmult0 \par\pard\ql \li20\sb32\sl-184\slmult0\tx2704\tx3280 \up0 \expndtw0\charscalex108 +\tab \up0 \expndtw0\charscalex116 (x, t)\tab \up0 \exp ndtw0\charscalex103 \ul0\nosupersub\cf3\f4\fs16 (2.2)\par\pard\ql \li217\sb0\sl128\slmult0 \up0 \expndtw0\charscalex153 \ul0\nosupersub\cf8\f9\fs16 Ox [q(x" t) ] = c:(x,t) -\par\pard\sect\sectd\sbknone\cols2\colno1\colw5921\colsr160\colno2\ colw6019\colsr160\qj \li1435\ri1\sb11\sl-211\slmult0 \up0 \expndtw0\charscalex12 0 \ul0\nosupersub\cf5\f6\fs14 for \ul0\nosupersub\cf3\f4\fs16 all times to at ea ch location \ul0\nosupersub\cf8\f9\fs16 x \up0 \expndtw0\charscalex116 > \ul0\n osupersub\cf3\f4\fs16 0. For full generality, \line \up0 \expndtw0\charscalex115 it is also assumed that both calling and noncalling vehicles \line \up0 \expndt

w0\charscalex116 can enter and leave the highway at any location. \up0 \expndtw0 \charscalex112 (Cases in \line \up0 \expndtw0\charscalex114 which vehicles can e nter or leave only at initely many ixed \line \up0 \expndtw0\charscalex109 entra nces and exits are considered at the end of Section II-A.)\par\pard\qj \li1435\r i1\sb1\sl-210\slmult0\fi182 \up0 \expndtw0\charscalex112 In Section II-A, we irs t present the model, which captures \line \up0 \expndtw0\charscalex110 both the time-dependent behavior (e.g., nonhomogeneous ar-\line \up0 \expndtw0\charscalex 106 rivals of vehicles) and vehicle movement on the highway. Then \line \up0 \ex pndtw0\charscalex113 in Section II-B, we simplify the time-nonhomogeneous luid \ line \up0 \expndtw0\charscalex115 model into a time-homogeneous model to capture only the \line \up0 \expndtw0\charscalex118 spatial dynamics. The PDE's then be come ODE's. As can \line \up0 \expndtw0\charscalex112 be seen in the numerical e xamples in Section III, even when \line \up0 \expndtw0\charscalex114 vehicles "b ehave" in a time-independent manner, the space \line \up0 \expndtw0\charscalex11 2 dynamics alone signiicantly affect system trafic loads.\par\pard\ql \li1425\sb 0\sl-221\slmult0 \par\pard\ql \li1430\ri1\sb78\sl-221\slmult0\tx1612 \up0 \expnd tw0\charscalex108 \ul0\nosupersub\cf8\f9\fs16 A. The Time-Nonhomogeneous Determi nistic Fluid Model \line\tab \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf3\f4 \fs16 To begin, we introduce some notation. Let \ul0\nosupersub\cf8\f9\fs16 N(x, t) \ul0\nosupersub\cf3\f4\fs16 and \line \up0 \expndtw0\charscalex121 \ul0\nosu persub\cf8\f9\fs16 Q(x, t) \ul0\nosupersub\cf3\f4\fs16 be the \ul0\nosupersub\cf 8\f9\fs16 number \ul0\nosupersub\cf3\f4\fs16 of noncalling and calling vehicles \line \up0 \expndtw0\charscalex113 in location \up0 \expndtw0\charscalex118 (0, \ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosupersub\cf3\f4\fs16 at time \ul0\nosupers ub\cf8\f9\fs16 t, \ul0\nosupersub\cf3\f4\fs16 respectively. The model treats \li ne \up0 \expndtw0\charscalex122 vehicles as a continuous luid; \ul0\nosupersub\c f8\f9\fs16 N(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 Q (x, t) \ul0\nosupersub\cf3\f4\fs16 are not \line \up0 \expndtw0\charscalex116 ne cessarily integers, but any nonnegative real numbers. In \line \up0 \expndtw0\ch arscalex113 addition, let \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf 3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x, t) \ul0\nosupersub\cf3\f4\fs16 be the \ul0\nosupersub\cf8\f9\fs16 noncalling \ul0\nosupersub\cf3\f4\fs16 density and \line \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf8\f9\fs16 calling \ul0\ nosupersub\cf3\f4\fs16 density at location \ul0\nosupersub\cf8\f9\fs16 x \ul0\no supersub\cf3\f4\fs16 and time t, respectively. That is,\par\pard\ql \li1425\sb0\ sl-184\slmult0 \par\pard\ql \li1425\sb0\sl-184\slmult0 \par\pard\ql \li1425\sb0\ sl-184\slmult0 \par\pard\ql \li1425\sb0\sl-184\slmult0 \par\pard\ql \li1425\sb0\ sl-184\slmult0 \par\pard\ql \li1435\sb3\sl-184\slmult0 \up0 \expndtw0\charscalex 115 Throughout this section, we assume that all derivatives are\par\pard\column \ql \li25\sb0\sl-144\slmult0\tx692 \up0 \expndtw0\charscalex121 for \ul0\nosuper sub\cf8\f9\fs16 x\tab \up0 \expndtw0\charscalex135 \ul0\nosupersub\cf3\f4\fs16 0 and -oo < \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 < oo.\par\p ard\qj \li25\ri1519\sb0\sl-199\slmult0\fi345 \up0 \expndtw0\charscalex122 \ul0\n osupersub\cf8\f9\fs16 Proof. \ul0\nosupersub\cf3\f4\fs16 We irst \ul0\nosupersub \cf8\f9\fs16 relate N(x, t + \ul0\nosupersub\cf3\f4\fs16 At) to \ul0\nosupersub\ cf8\f9\fs16 N(x. t). \ul0\nosupersub\cf3\f4\fs16 During \line \up0 \expndtw0\cha rscalex111 the time interval \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf8\f9 \fs16 (t, t \up0 \expndtw0\charscalex115 + At], \ul0\nosupersub\cf3\f4\fs16 a ce rtain number of noncalling \up0 \expndtw0\charscalex114 vehicles in location \up 0 \expndtw0\charscalex117 (0, \ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosupersub\cf3 \f4\fs16 move forward and pass beyond \up0 \expndtw0\charscalex122 location \ul0 \nosupersub\cf8\f9\fs16 x. \ul0\nosupersub\cf3\f4\fs16 In addition, some new arr ivals of noncalling \line \up0 \expndtw0\charscalex108 vehicles enter into the s ystem, while some leave from location \up0 \expndtw0\charscalex113 (0, \ul0\nosu persub\cf8\f9\fs16 x] \ul0\nosupersub\cf3\f4\fs16 at the time interval. We thus have\par\pard\ql \li241\sb119\sl-184\slmult0 \up0 \expndtw0\charscalex128 \ul0\n osupersub\cf8\f9\fs16 N(x, t + \ul0\nosupersub\cf3\f4\fs16 At) - \ul0\nosupersub \cf8\f9\fs16 N(x, t) =\par\pard\ql \li644\sb85\sl-184\slmult0 \up0 \expndtw0\cha rscalex129 - n(x, t)v(x, t)At + [C; (x, t + \ul0\nosupersub\cf3\f4\fs16 At) \ul0 \nosupersub\cf8\f9\fs16 - C;(x,t)]\par\pard\ql \li639\sb90\sl-184\slmult0\tx4129

\up0 \expndtw0\charscalex139 - [C;(x,t + At) - C;(x,t)1+ o(At)\tab \up0 \expndt w0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 (2.3)\par\pard\qj \li25\ri1524\sb12 3\sl-208\slmult0\tx1528\tx2372 \up0 \expndtw0\charscalex113 where \ul0\nosupersu b\cf8\f9\fs16 o(At)1,6d\tab \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf3\f4\ fs16 0 as At\tab \up0 \expndtw0\charscalex112 0, assuming that \ul0\nosupersub\c f8\f9\fs16 II \ul0\nosupersub\cf3\f4\fs16 and v are \up0 \expndtw0\charscalex111 continuous in the neighborhood of \ul0\nosupersub\cf8\f9\fs16 (x, t). \ul0\nosu persub\cf3\f4\fs16 Dividing both sides \up0 \expndtw0\charscalex121 of (2.3) by At, letting At \up0 \expndtw0\charscalex120 -4 0, and diferentiating w.r.t. \ul0 \nosupersub\cf8\f9\fs16 x\par\pard\ql \li20\sb23\sl-184\slmult0\tx4326 \up0 \exp ndtw0\charscalex108 \ul0\nosupersub\cf3\f4\fs16 yields (2.1). Equation \up0 \exp ndtw0\charscalex110 (2.2) is obtained in the same way.\tab \up0 \expndtw0\charsc alex105 \u10065?\par\pard\qj \li20\ri1524\sb11\sl-210\slmult0\fi177 \up0 \expndt w0\charscalex106 We remark that (2.2) corresponds to the fundamental conser-\lin e \up0 \expndtw0\charscalex110 vation equation in (2.7) of [7]. Before showing h ow (2.1) and \line \up0 \expndtw0\charscalex113 (2.2) are coupled due to calling activity, we need additional \line \up0 \expndtw0\charscalex118 notation. Let \ ul0\nosupersub\cf8\f9\fs16 E+ (x,1) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosuper sub\cf8\f9\fs16 E- (x, t) \ul0\nosupersub\cf3\f4\fs16 be the number of non-\line \up0 \expndtw0\charscalex109 calling vehicles arriving to and departing from th e highway in \line \up0 \expndtw0\charscalex108 location (0, xl during time inte rval ( -oc, \ul0\nosupersub\cf8\f9\fs16 t], \ul0\nosupersub\cf3\f4\fs16 respecti vely (Case\par\pard\ql \li25\sb24\sl-184\slmult0 \up0 \expndtw0\charscalex129 \u l0\nosupersub\cf8\f9\fs16 (a) \ul0\nosupersub\cf3\f4\fs16 for c; and \ul0\nosupe rsub\cf5\f6\fs14 c7, \ul0\nosupersub\cf3\f4\fs16 above). We use \ul0\nosupersub\ cf8\f9\fs16 E+(x,t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 Ei \up0 \expndtw0\charscalex129 (x,t)\par\pard\qj \li25\ri1524\sb8\sl-208\slmult 0 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs16 to denote the respect ive number of calling vehicles arriving \up0 \expndtw0\charscalex113 to and depa rting from the highway in location (0. \ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosup ersub\cf3\f4\fs16 in time \up0 \expndtw0\charscalex114 (-oo, \ul0\nosupersub\cf8 \f9\fs16 t]. \ul0\nosupersub\cf3\f4\fs16 The associated rate densities are\par\p ard\ql \li1306\sb52\sl-184\slmult0\tx3126 \up0 \expndtw0\charscalex104 a2E-y (x7 0\tab \up0 \expndtw-4\charscalex100 \ul0\nosupersub\cf14\f15\fs16\ul a2E7 \up0 \expndtw-2\charscalex100 , \up0 \expndtw0\charscalex133 (x,t)\par\pard\ql \li471 \sb46\sl-128\slmult0\tx2070 \up0 \expndtw0\charscalex149 \ul0\nosupersub\cf8\f9\ fs16 e;(x,t)=\tab \up0 \expndtw0\charscalex116 ' , \ul0\nosupersub\cf3\f4\fs16 e n (x, t)\par\pard\ql \li1532\sb0\sl-128\slmult0\tx3352 \up0 \expndtw0\charscalex 129 \ul0\nosupersub\cf8\f9\fs16 axat\tab \up0 \expndtw0\charscalex134 \ul0\nosup ersub\cf3\f4\fs16 axat\par\pard\ql \li1425\sb0\sl-184\slmult0 \par\pard\ql \li11 91\sb109\sl-184\slmult0\tx1643\tx3246 \up0 \expndtw0\charscalex120 \ul0\nosupers ub\cf8\f9\fs16 a2E\tab \up0 \expndtw0\charscalex133 (x t)\tab \up0 \expndtw0\cha rscalex120 82E\ul0\nosupersub\cf14\f15\fs16\ul q\ul0\nosupersub\cf8\f9\fs16 -(x t)\par\pard\sect\sectd\sbknone\cols5\colno1\colw6257\colsr160\colno2\colw161\col sr60\colno3\colw388\colsr40\colno4\colw954\colsr160\colno5\colw3950\colsr160\ql \li1440\sb0\sl-128\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\f4\f s16 well deined.\par\pard\column \ql \li20\sb2\sl-128\slmult0 \up0 \expndtw-2\ch arscalex100 \ul0\nosupersub\cf8\f9\fs16 e\par\pard\column \ql \li20\sb0\sl-144\s lmult0 \up0 \expndtw0\charscalex129 (x t)\par\pard\column \ql \li20\sb0\sl-144\s lmult0 \up0 \expndtw0\charscalex406 - \ul0\super\cf16\f17\fs4\ul0\super\cf16\f1 7\fs3 q\par\pard\column \ql \li20\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex11 6 \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 e- (x, \ul0\nosupe rsub\cf3\f4\fs16 t)\par\pard\sect\sectd\sbknone\cols2\colno1\colw7151\colsr20\co lno2\colw4929\colsr160\ql \li1440\ri0\sb0\sl-189\slmult0\fi168\tx6494\tx6078 \up 0 \expndtw0\charscalex115 Furthermore, let C(x, t) and \ul0\nosupersub\cf8\f9\fs1 6 C7.:(x, \ul0\nosupersub\cf3\f4\fs16 t) be the number of\tab \up0 \expndtw0\cha rscalex80 \ul0\nosupersub\cf8\f9\fs16 + \line \up0 \expndtw0\charscalex106 \ul0\ nosupersub\cf3\f4\fs16 noncalling vehicles \ul0\nosupersub\cf8\f9\fs16 entering \ul0\nosupersub\cf3\f4\fs16 or \ul0\nosupersub\cf8\f9\fs16 leaving \ul0\nosupers ub\cf3\f4\fs16 in location (0, \ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosupersub\cf

3\f4\fs16 during \tab \up0 \expndtw0\charscalex117 Further, let \up0 \expndtw-3\ charscalex100 \ul0\nosupersub\cf8\f9\fs16 0\par\pard\column \ql \li20\ri1524\sb0 \sl-198\slmult0\fi307\tx2382 \up0 \expndtw-6\charscalex100 \ul0\nosupersub\cf17\ f18\fs22 axat\tab \up0 \expndtw0\charscalex132 \ul0\nosupersub\cf8\f9\fs16 oast \line \up0 \expndtw0\charscalex136 (x,t)n(x.t) \ul0\nosupersub\cf3\f4\fs16 and \ up0 \expndtw0\charscalex129 \ul0\nosupersub\cf8\f9\fs16 ^y(x,t)q(x,t) \ul0\nosup ersub\cf3\f4\fs16 be the rates\par\pard\sect\sectd\sbknone\cols2\colno1\colw5921 \colsr160\colno2\colw6019\colsr160\qj \li1425\ri0\sb0\sl-207\slmult0\fi9 \up0 \e xpndtw0\charscalex108 time interval (-oo, t], respectively. Similarly, we use \u l0\nosupersub\cf8\f9\fs16 C + (x, t) \line \up0 \expndtw0\charscalex121 \ul0\nos upersub\cf3\f4\fs16 and Cg- \ul0\nosupersub\cf8\f9\fs16 (x, t) \ul0\nosupersub\c f3\f4\fs16 to denote the respective number of calling \line \up0 \expndtw0\chars calex110 vehicles entering or leaving in location (0, \ul0\nosupersub\cf8\f9\fs1 6 x] \ul0\nosupersub\cf3\f4\fs16 in time (-oo, \ul0\nosupersub\cf8\f9\fs16 t]. \ line \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 A noncalling \up0 \expndtw0\charscalex116 (calling) vehicle is considered to be entering \line \up 0 \expndtw0\charscalex113 the system, if either: a) it is an actual arrival of a noncalling \line \up0 \expndtw0\charscalex126 (calling) vehicle to the highway, or b) it was a calling \line \up0 \expndtw0\charscalex119 (noncalling) vehicle existing on the highway but with its \line \up0 \expndtw0\charscalex110 call jus t terminated (started). Likewise, a noncalling (calling) \line \up0 \expndtw0\ch arscalex116 vehicle leaves if it departs from the highway or becomes a \line \up 0 \expndtw0\charscalex128 \ul0\nosupersub\cf5\f6\fs14 calling (noncalling) \ul0\ nosupersub\cf4\f5\fs14 vehicle \ul0\nosupersub\cf5\f6\fs14 by initiating (termin ating) a call.\par\pard\column \qj \li20\ri1510\sb0\sl-206\slmult0\fi4 \up0 \exp ndtw0\charscalex122 \ul0\nosupersub\cf3\f4\fs16 at which noncalling and calling vehicles actually depart \up0 \expndtw0\charscalex130 from the highway at locati on \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 at time \ul0\nosupe rsub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 In addition, \line \up0 \expndtw 0\charscalex123 let \ul0\nosupersub\cf8\f9\fs16 A(x,t)n(x,t) \ul0\nosupersub\cf3 \f4\fs16 be the call-origination rate of noncalling \line \up0 \expndtw0\charsca lex125 vehicles and \ul0\nosupersub\cf8\f9\fs16 ii(x,t)q(x,t) \ul0\nosupersub\cf 3\f4\fs16 be the call-termination rate of \line \up0 \expndtw0\charscalex118 cal ling vehicles at location \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\f s16 and time \ul0\nosupersub\cf8\f9\fs16 t. \up0 \expndtw0\charscalex113 \ul0\no supersub\cf3\f4\fs16 (In the stochastic \up0 \expndtw0\charscalex110 model, thes e are stochastic intensities for individual vehicles; \up0 \expndtw0\charscalex1 13 here these are actual deterministic low rates.)\par\pard\qj \li25\ri1514\sb1\ sl-204\slmult0\fi172 \up0 \expndtw0\charscalex111 We note that the rate densitie s, \ul0\nosupersub\cf8\f9\fs16 c17(s, \ul0\nosupersub\cf3\f4\fs16 t), c- \ul0\no supersub\cf8\f9\fs16 (x,1 \ul0\nosupersub\cf3\f4\fs16 ), \ul0\nosupersub\cf8\f9\ fs16 cc/(x, \ul0\nosupersub\cf3\f4\fs16 t), \up0 \expndtw0\charscalex111 and c9 \ul0\nosupersub\cf8\f9\fs16 (x, t) \ul0\nosupersub\cf3\f4\fs16 can be expressed in terms of these parameters as \up0 \expndtw0\charscalex131 \ul0\nosupersub\cf5 \f6\fs14 follows and we omit the proof. \par\pard\sect\sectd\fs24\paperw12240\pa perh15840{\bkmkstart Pg4}{\bkmkend Pg4}\par\pard\li1675\sb0\sl-184\slmult0\par\p ard\li1675\sb0\sl-184\slmult0\par\pard\li1675\sb0\sl-184\slmult0\par\pard\li1675 \sb67\sl-184\slmult0\fi0\tx5289 \up0 \expndtw-7\charscalex83 \ul0\nosupersub\cf3 \f4\fs16 1356\tab \up0 \expndtw-6\charscalex84 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\sect\sectd\sbknone\cols2\ colno1\colw6152\colsr160\colno2\colw5788\colsr160\ql \li1838\ri2\sb226\sl-316\sl mult0\tx2294\tx5761 \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf8\f9\fs16 Lem ma \ul0\nosupersub\cf3\f4\fs16 2.2: The four rate densities can be expressed as \line\tab \up0 \expndtw0\charscalex122 a) \ul0\nosupersub\cf8\f9\fs16 c\u355?(x, \ul0\nosupersub\cf3\f4\fs16 t) = en \ul0\nosupersub\cf8\f9\fs16 (x. t) + kc(x,t )q(x,t) \tab \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 (2.4)\par\ pard\ql \li2294\sb62\sl-184\slmult0\tx5764 \up0 \expndtw0\charscalex127 b) \ul0\ nosupersub\cf8\f9\fs16 c; (x, \ul0\nosupersub\cf3\f4\fs16 t) = \ul0\nosupersub\c f8\f9\fs16 0(x, t)n(x,t) + A(x,t)n(x,t)\tab \up0 \expndtw0\charscalex105 \ul0\no supersub\cf3\f4\fs16 (2.5)\par\pard\ql \li2294\sb85\sl-184\slmult0\tx3503\tx5764

\up0 \expndtw0\charscalex114 c) c4 \ul0\nosupersub\cf8\f9\fs16 (x, t) =\tab \up 0 \expndtw0\charscalex122 (x, t) + A(x,t)n,(x,t)\tab \up0 \expndtw0\charscalex10 5 \ul0\nosupersub\cf3\f4\fs16 (2.6)\par\pard\ql \li2294\sb90\sl-184\slmult0\tx57 64 \up0 \expndtw0\charscalex118 d) cq \ul0\nosupersub\cf8\f9\fs16 (x, \ul0\nosup ersub\cf3\f4\fs16 t) = \ul0\nosupersub\cf8\f9\fs16 y(x,t)q(x,t) + 1.1(x, t)q(x, t).\tab \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 (2.7)\par\pard\ qj \li1670\ri2\sb143\sl-207\slmult0\fi172 \up0 \expndtw0\charscalex116 We now co mbine Lemmas 2.1 and 2.2 to obtain the fol-\line \up0 \expndtw0\charscalex111 lo wing coupled PDE's characterizing the densities \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \up0 \expndtw0\charscalex120 \ul0\nosupersub \cf8\f9\fs16 q(x ,t) in \ul0\nosupersub\cf3\f4\fs16 our model. These PDE's can b e regarded as the \up0 \expndtw0\charscalex113 deterministic luid model.\par\par d\qj \li1675\ri2\sb2\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex110 \ul0\nosu persub\cf8\f9\fs16 Theorem 2.1: \ul0\nosupersub\cf3\f4\fs16 The densities of non calling and calling vehi-\line \up0 \expndtw0\charscalex115 cles, \ul0\nosupersu b\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs1 6 q(x, t), \ul0\nosupersub\cf3\f4\fs16 satisfy the coupled PDE's:\par\pard\ql \l i1761\sb100\sl-195\slmult0\tx2649\tx2836 \up0 \expndtw0\charscalex148 \ul0\nosup ersub\cf18\f19\fs16\ul and,\ul0\tab \up0 \expndtw-2\charscalex100 \ul0\nosupersu b\cf8\f9\fs16 a\tab \up0 \expndtw-8\charscalex85 \ul0\nosupersub\cf17\f18\fs22 [ n(x\par\pard\column \qj \li1680\sb0\sl-211\slmult0 \par\pard\qj \li20\ri1274\sb9 2\sl-211\slmult0\fi4 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\f4\fs16 a function of \ul0\nosupersub\cf8\f9\fs16 q(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 n(x, t). \ul0\nosupersub\cf3\f4\fs16 It is also wort h noting that, \line \up0 \expndtw0\charscalex118 by choosing some initial time r with \up0 \expndtw0\charscalex141 \ul0\nosupersub\cf5\f6\fs14 \u8212?oo < r < oo \ul0\nosupersub\cf3\f4\fs16 such \line \up0 \expndtw0\charscalex125 that x(r) \up0 \expndtw0\charscalex123 = 0 as the initial condition for (2.10), \ul0\nosu persub\cf8\f9\fs16 n(x(t),t) \line \up0 \expndtw0\charscalex130 \ul0\nosupersub\ cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x(t),t) \ul0\nosupersub\cf3\f4\fs1 6 can be solved for all T \up0 \expndtw0\charscalex274 \ul0\nosupersub\cf1\f2\fs 10 < t \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf3\f4\fs16 < \up0 \expndtw0 \charscalex117 00 from \line \up0 \expndtw0\charscalex106 (2.10)\u8212?(2.12), e .g., by a Runge\u8212?Kutta method. Of course, the \line \up0 \expndtw0\charscal ex112 solution depends on the initial conditions n(0, r) and q(0, \ul0\nosupersu b\cf8\f9\fs16 r). \line \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf3\f4\fs16 Thus, if T is selected properly, one can obtain \ul0\nosupersub\cf8\f9\fs16 n(x , t) \ul0\nosupersub\cf3\f4\fs16 and \line \up0 \expndtw0\charscalex110 \ul0\nos upersub\cf8\f9\fs16 q(x, t) \ul0\nosupersub\cf3\f4\fs16 for the location region and the time interval of interest.\par\pard\qj \li24\ri1284\sb3\sl-208\slmult0\f i177\tx1937 \up0 \expndtw0\charscalex118 We now suppose that the highway is divi ded into cells, \line \up0 \expndtw0\charscalex107 indexed by i = 1, 2, 3, -\tab \up0 \expndtw0\charscalex109 For i > 1, let the boundary between \line \up0 \ex pndtw0\charscalex113 cell i \u8212? 1 and cell i be located at \ul0\nosupersub\c f1\f2\fs10 xi \u8212? \ul0\nosupersub\cf3\f4\fs16 1 and xo =- 0. Further, \line \up0 \expndtw0\charscalex115 let \ul0\nosupersub\cf8\f9\fs16 WO \ul0\nosupersub\ cf3\f4\fs16 be the instantaneous offered load in cell i at time \ul0\nosupersub\ cf8\f9\fs16 t; \line \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf3\f4\fs16 th at is, the total number of calls in progress in cell i at time\par\pard\ql \li29 \sb23\sl-184\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 Let hi(t) denote the rate of ongoing calls handed o f from\par\pard\sect\sectd\sbknone\cols4\colno1\colw2566\colsr40\colno2\colw407\ colsr160\colno3\colw3003\colsr160\colno4\colw5784\colsr160\ql \li2409\sb0\sl-144 \slmult0 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf8\f9\fs16 +\par\pard\ql \li1680\sb0\sl-249\slmult0 \par\pard\ql \li1680\sb0\sl-249\slmult0 \par\pard\ql \li1680\ri78\sb159\sl-249\slmult0\tx1771 \up0 \expndtw0\charscalex105 \ul0\nosup ersub\cf3\f4\fs16 and \line\tab \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf1 4\f15\fs16\ul 0q(x, \ul0\nosupersub\cf18\f19\fs16\ul t)\par\pard\column \ql \li2 0\sb52\sl-184\slmult0 \up0 \expndtw0\charscalex124 \ul0\nosupersub\cf8\f9\fs16 a x\par\pard\ql \li1680\sb0\sl-184\slmult0 \par\pard\ql \li1680\sb0\sl-184\slmult0

\par\pard\ql \li1680\sb0\sl-184\slmult0 \par\pard\ql \li1680\sb0\sl-184\slmult0 \par\pard\ql \li58\sb131\sl-184\slmult0 \up0 \expndtw-2\charscalex100 a\par\par d\column \ql \li20\sb0\sl-198\slmult0 \up0 \expndtw-4\charscalex100 \ul0\nosuper sub\cf17\f18\fs22 , \ul0\nosupersub\cf8\f9\fs16 t)v (x , t)] \up0 \expndtw0\char scalex114 =e \up0 \expndtw-2\charscalex100 , \up0 \expndtw0\charscalex108 \ul0\n osupersub\cf17\f18\fs22 (x, \ul0\nosupersub\cf8\f9\fs16 t) + kt(x ,t)q(x ,t)\par \pard\ql \li975\sb42\sl-184\slmult0 \up0 \expndtw0\charscalex128 \u8212?[0(x,t)+ A(x,t)]n(x,t)\par\pard\ql \li2636\sb80\sl-184\slmult0 \up0 \expndtw0\charscalex 102 \ul0\nosupersub\cf3\f4\fs16 (2.8)\par\pard\column \ql \li20\sb0\sl-144\slmul t0 \up0 \expndtw0\charscalex117 cell i \up0 \expndtw0\charscalex118 \u8212? 1 to cell i at time t.\par\pard\qj \li30\ri1289\sb0\sl-206\slmult0\fi177 \up0 \expnd tw0\charscalex114 \ul0\nosupersub\cf8\f9\fs16 Theorem 2.2: \ul0\nosupersub\cf3\f 4\fs16 For cell i > 1, its instantaneous ofered load \up0 \expndtw0\charscalex11 9 and call handof rate at time \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3 \f4\fs16 are\par\pard\ql \li1680\sb0\sl-184\slmult0 \par\pard\ql \li4038\sb15\sl -184\slmult0 \up0 \expndtw0\charscalex104 (2.13)\par\pard\sect\sectd\sbknone\col s3\colno1\colw2240\colsr160\colno2\colw3771\colsr160\colno3\colw5779\colsr160\ql \li1972\sb56\sl-184\slmult0 \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf8\f9 \fs16 at\par\pard\column \ql \li20\sb0\sl-144\slmult0\tx457 \up0 \expndtw0\chars calex119 + \u8222? \tab \up0 \expndtw0\charscalex122 [q(x,t)v(x,t)] \up0 \expndt w0\charscalex113 =eq \ul0\nosupersub\cf17\f18\fs22 (x, t) \ul0\nosupersub\cf8\f9 \fs16 + A(x,t)n(x,t)\par\pard\ql \li212\sb0\sl-128\slmult0 \up0 \expndtw0\charsc alex127 ax\par\pard\ql \li1704\sb1\sl-144\slmult0 \up0 \expndtw0\charscalex113 \ u8212? [-y(x,t) + pc(x,t)]q(x,t).\par\pard\ql \li3394\sb87\sl-184\slmult0 \up0 \ expndtw0\charscalex103 \ul0\nosupersub\cf3\f4\fs16 (2.9)\par\pard\column \ql \li 25\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex103 and\par\pard\ql \li1680\sb0\s l-184\slmult0 \par\pard\ql \li1680\sb0\sl-184\slmult0 \par\pard\ql \li20\sb126\s l-184\slmult0 \up0 \expndtw0\charscalex106 respectively.\par\pard\sect\sectd\sbk none\cols2\colno1\colw6161\colsr160\colno2\colw5779\colsr160\qj \li1680\ri6\sb10 \sl-211\slmult0\fi172 \up0 \expndtw0\charscalex119 With an additional assumption , the PDE's in \up0 \expndtw0\charscalex115 (2.8) and \line \up0 \expndtw0\char scalex111 (2.9) can be converted into a set of three ordinary diferential \line \up0 \expndtw0\charscalex113 equations (ODE' \ul0\nosupersub\cf5\f6\fs14 s), \ul 0\nosupersub\cf3\f4\fs16 which are easier to solve in some cases. \line \up0 \ex pndtw0\charscalex120 (This is the classical method of characteristics.) For this \line \up0 \expndtw0\charscalex109 purpose, let the location \ul0\nosupersub\cf 8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 as a time function, \ul0\nosupersub\cf8\ f9\fs16 x(t), \ul0\nosupersub\cf3\f4\fs16 be given by\par\pard\ql \li3225\sb104\ sl-184\slmult0 \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 dx(t)\pa r\pard\column \qj \li20\ri1284\sb5\sl-211\slmult0\fi345 \up0 \expndtw0\charscale x113 Proof \ul0\nosupersub\cf3\f4\fs16 Equation (2.13) is immediately given by t he def-\line \up0 \expndtw0\charscalex120 inition of \ul0\nosupersub\cf8\f9\fs16 q(x, t). \ul0\nosupersub\cf3\f4\fs16 To consider the rate of handof calls, let \line \up0 \expndtw0\charscalex107 \ul0\nosupersub\cf8\f9\fs16 HZ (t) \ul0\nosup ersub\cf3\f4\fs16 be the number of calls handed off from cell i \u8212?1 to cell \line \up0 \expndtw0\charscalex118 i before time \ul0\nosupersub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 Since \ul0\nosupersub\cf8\f9\fs16 Hi(t + \ul0\nosup ersub\cf3\f4\fs16 At) is equal to Hi (t) plus the \line \up0 \expndtw0\charscale x114 calls handed of during the time interval (t, \ul0\nosupersub\cf8\f9\fs16 t + At], \ul0\nosupersub\cf3\f4\fs16 we have\par\pard\ql \li150\sb162\sl-184\slmul t0 \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf8\f9\fs16 Hi(t + At) \u8212? H i(t) =\par\pard\sect\sectd\sbknone\cols3\colno1\colw3531\colsr160\colno2\colw331 1\colsr160\colno3\colw4948\colsr160\ql \li3345\sb3\sl-184\slmult0 \up0 \expndtw0 \charscalex113 dt\par\pard\column \ql \li20\sb0\sl-144\slmult0\tx2026 \up0 \expn dtw0\charscalex129 = v(x(t),t).\tab \up0 \expndtw0\charscalex103 \ul0\nosupersub \cf3\f4\fs16 (2.10)\par\pard\column \ql \li20\sb114\sl-184\slmult0 \up0 \expndtw 0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 q(xj_i, t)[1 \u8212? p,(xj_i,t)At]v( xi_i, \ul0\super\cf3\f4\fs24 t)At\par\pard\sect\sectd\sbknone\cols2\colno1\colw6 166\colsr160\colno2\colw5774\colsr160\ql \li1684\sb0\sl-144\slmult0 \up0 \expndt w0\charscalex113 Equation (2.10) is one of the three ODE's.\par\pard\qj \li1684\

ri7\sb6\sl-211\slmult0\fi168 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf8\f9 \fs16 Lemma 2.3: \ul0\nosupersub\cf3\f4\fs16 Given (2.10), the PDE's (2.8) and \up0 \expndtw0\charscalex113 (2.9) are \up0 \expndtw0\charscalex111 equivalent t o\par\pard\ql \li1728\sb96\sl-195\slmult0\tx3575 \up0 \expndtw0\charscalex121 \u l0\nosupersub\cf8\f9\fs16 dn(x(t), \ul0\nosupersub\cf14\f15\fs16\ul t) \up0 \ex pndtw-8\charscalex81 \ul0\nosupersub\cf17\f18\fs22 =\tab \up0 \expndtw0\charscal ex134 \ul0\nosupersub\cf8\f9\fs16 + 4(x(0,0q(x(0,\par\pard\column \qj \li889\ri1 289\sb0\sl-248\slmult0\tx4048 \up0 \expndtw0\charscalex125 \ul0\nosupersub\cf3\f 4\fs16 + n(xi-1,t)A(xi-1,t)Atv(xi_1, \ul0\nosupersub\cf8\f9\fs16 t)At + 0(6,0. \ line \tab \up0 \expndtw0\charscalex103 \ul0\nosupersub\cf3\f4\fs16 (2.15)\par\pa rd\ql \li20\sb165\sl-184\slmult0 \up0 \expndtw0\charscalex119 By deinition, \ul0 \nosupersub\cf8\f9\fs16 hi(t) \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf3\ f4\fs16 = \ul0\nosupersub\cf8\f9\fs16 dHi(t)/dt. \ul0\nosupersub\cf3\f4\fs16 Thu s, dividing by At and\par\pard\sect\sectd\sbknone\cols3\colno1\colw2523\colsr40\ colno2\colw3613\colsr160\colno3\colw5774\colsr160\ql \li2040\sb56\sl-184\slmult0 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf8\f9\fs16 dt\par\pard\ql \li1684 \sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684 \sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684 \sb88\sl-184\slmult0 \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 an d\par\pard\ql \li1737\sb99\sl-184\slmult0 \up0 \expndtw0\charscalex117 \ul0\nosu persub\cf14\f15\fs16\ul dq(x(t), t)\par\pard\column \ql \li423\sb0\sl-144\slmult 0 \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf8\f9\fs16 (x(t), t)\par\pard\ql \li298\sb105\sl-184\slmult0 \up0 \expndtw0\charscalex130 \ul0\nosupersub\cf14\f 15\fs16\ul rav(x,t)\par\pard\qj \li303\ri40\sb0\sl-120\slmult0\fi710\tx570 \up0 \expndtw0\charscalex127 \ul0\nosupersub\cf3\f4\fs16 + \ul0\nosupersub\cf8\f9\fs1 6 0(x(t), t) \ul0\nosupersub\cf3\f4\fs16 + \ul0\nosupersub\cf8\f9\fs16 A(x(t), t )]n(x(t),t) \line \up0 \expndtw-6\charscalex80 \ul0\nosupersub\cf4\f5\fs14 L \ta b \up0 \expndtw0\charscalex124 \ul0\nosupersub\cf8\f9\fs16 ax\par\pard\ql \li314 9\sb38\sl-184\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 ( 2.11)\par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slm ult0 \par\pard\ql \li20\sb110\sl-184\slmult0\tx1033 \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf8\f9\fs16 = e+ (x(t),\tab \up0 \expndtw0\charscalex104 +\par\ pard\column \ql \li20\sb13\sl-184\slmult0\tx4321 \up0 \expndtw0\charscalex108 \u l0\nosupersub\cf3\f4\fs16 taking the limit as At \u8212?\u9632? 0 in (2.15) yiel ds (2.14).\tab \up0 \expndtw0\charscalex101 \u10065?\par\pard\qj \li20\ri1279\sb 5\sl-211\slmult0\fi172 \up0 \expndtw0\charscalex118 We remark that \up0 \expndt w0\charscalex117 (2.14) corresponds to the conservation \up0 \expndtw0\charscale x112 equation (2.6) in \up0 \expndtw0\charscalex120 [7]. It gives a low rate at a point, which \up0 \expndtw0\charscalex115 does not actually require that cells be deined: i.e., \up0 \expndtw0\charscalex110 (2.14) is \up0 \expndtw0\charscal ex114 valid for arbitrary \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\f s16 as well as xi \u8212? 1.\par\pard\qj \li20\ri1279\sb10\sl-206\slmult0\fi172 \up0 \expndtw0\charscalex116 We conclude this subsection by commenting on the ra te \line \up0 \expndtw0\charscalex112 densities of vehicle entering and leaving the highway for the\par\pard\sect\sectd\sbknone\cols4\colno1\colw2652\colsr160\c olno2\colw685\colsr40\colno3\colw2649\colsr160\colno4\colw5774\colsr160\ql \li20 40\sb0\sl-184\slmult0 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf8\f9\fs16 d t\par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684\sb0\sl-184\slmult0 \par\pard\ql \li1684\sb74\sl-184\slmult0 \up0 \expndtw0\charscalex106 \ul0\nosu persub\cf3\f4\fs16 respectively.\par\pard\column \ql \li20\sb1\sl-410\slmult0 \u p0 \expndtw0\charscalex47 \ul0\nosupersub\cf19\f20\fs42\ul rav(x,t)\par\pard\ql \li20\sb31\sl-161\slmult0\tx289 \up0 \expndtw-6\charscalex80 \ul0\nosupersub\cf4 \f5\fs14 L\tab \up0 \expndtw0\charscalex124 \ul0\nosupersub\cf8\f9\fs16 ax\par\p ard\column \ql \li342\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex123 (x(t), t)n (x(t), t)\par\pard\ql \li20\sb169\sl-184\slmult0 \up0 \expndtw0\charscalex130 + y(x(t),t) + ti(x(t),t)]q(x(t),t),\par\pard\ql \li2190\sb142\sl-184\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 (2.12)\par\pard\column \qj \ li20\ri1274\sb0\sl-196\slmult0\tx334\tx1443 \up0 \expndtw0\charscalex110 case wh ere vehicles can enter or leave only at entrances/exits \line \up0 \expndtw0\cha

rscalex115 at ixed locations, as in real highway systems. Suppose that \line \up 0 \expndtw-2\charscalex100 \{ \tab \up0 \expndtw0\charscalex120 : i \up0 \expndt w0\charscalex112 = 1, 2, 3, \tab \up0 \expndtw0\charscalex115 -\} is the locatio n of the ith entrance/exit \line \up0 \expndtw0\charscalex120 on the highway. Le t us use \ul0\nosupersub\cf8\f9\fs16 Gl'i (t) \ul0\nosupersub\cf3\f4\fs16 and \u l0\nosupersub\cf8\f9\fs16 q(t) \ul0\nosupersub\cf3\f4\fs16 to denote the \line \ up0 \expndtw0\charscalex112 external arrival rate of noncalling and calling vehi cles at the \line \up0 \expndtw0\charscalex113 ith entrance at time t, respectiv ely. Then, we have\par\pard\sect\sectd\sbknone\cols2\colno1\colw6175\colsr160\co lno2\colw5765\colsr160\ql \li2035\ri740\sb0\sl-257\slmult0\tx2361\tx4806 \up0 \e xpndtw0\charscalex116 \ul0\nosupersub\cf8\f9\fs16 Proof. \ul0\nosupersub\cf3\f4\ fs16 By the chain rule, we have \line\tab \up0 \expndtw0\charscalex129 \ul0\nosu persub\cf14\f15\fs16\ul dn(x(t),t) \up0 \expndtw0\charscalex138 \ul0\nosupersub\ cf8\f9\fs16 \u8212? on(x,t) \ul0\nosupersub\cf14\f15\fs16\ul dx(t) \ul0\tab \up0 \expndtw0\charscalex132 \ul0\nosupersub\cf14\f15\fs16\ul an(x,t)\par\pard\ql \l i2668\sb39\sl-184\slmult0\tx3595\tx4286\tx5025\tx5428 \up0 \expndtw0\charscalex1 13 \ul0\nosupersub\cf8\f9\fs16 dt\tab \up0 \expndtw0\charscalex127 ax\tab \up0 \ expndtw0\charscalex109 dt\tab \up0 \expndtw0\charscalex120 at\tab \up0 \expndtw2\charscalex100 \u8226?\par\pard\qj \li1684\ri11\sb53\sl-211\slmult0\fi4\tx5977 \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 Substitute \up0 \expnd tw0\charscalex115 (2.8) and \up0 \expndtw0\charscalex119 (2.10) into this equat ion. After some \line \up0 \expndtw0\charscalex116 algebraic manipulation, we ge t \up0 \expndtw0\charscalex112 (2.11). Equation \up0 \expndtw0\charscalex117 (2. 12) is \line \up0 \expndtw0\charscalex110 obtained in the same way. \tab \up0 \e xpndtw-2\charscalex100 \u10065?\par\pard\qj \li1684\ri6\sb1\sl-206\slmult0\fi172 \up0 \expndtw0\charscalex113 Due to the partial derivative of v(x, \ul0\nosuper sub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 w.r.t. \ul0\nosupersub\cf8\f9\fs1 6 x \ul0\nosupersub\cf3\f4\fs16 on the r.h.s. \line \up0 \expndtw0\charscalex108 of (2.11) and (2.12), they are ODE's if and only if \ul0\nosupersub\cf8\f9\fs16 v(x, \ul0\nosupersub\cf3\f4\fs16 t) is not\par\pard\column \ql \li1158\sb56\sl368\slmult0\tx4038 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf8\f9\fs16 en ( x ,t) = \ul0\nosupersub\cf20\f21\fs32 En\ul0\nosupersub\cf8\f9\fs16 (t)8(x \u821 2? yi)\tab \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf5\f6\fs14 (2.16)\par\p ard\ql \li6335\sb0\sl-184\slmult0 \par\pard\ql \li20\sb56\sl-184\slmult0 \up0 \e xpndtw0\charscalex103 \ul0\nosupersub\cf3\f4\fs16 and\par\pard\ql \li1379\sb28\s l-345\slmult0\tx1811\tx4048 \up0 \expndtw0\charscalex55 \ul0\nosupersub\cf21\f22 \fs32 (x,\tab \up0 \expndtw0\charscalex106 \ul0\nosupersub\cf20\f21\fs32 = E \ul 0\nosupersub\cf8\f9\fs16 4(t)8(x \u8212? yi)\tab \up0 \expndtw0\charscalex103 \u l0\nosupersub\cf3\f4\fs16 (2.17)\par\pard\ql \li1302\sb211\sl-230\slmult0 \up0 \ expndtw-9\charscalex81 \ul0\nosupersub\cf22\f23\fs20 +E\par\pard\ql \li20\sb0\sl -120\slmult0\tx1547 \up0 \expndtw0\charscalex107 \ul0\nosupersub\cf3\f4\fs16 whe re limE_,0 \up0 \expndtw-6\charscalex90 \ul0\nosupersub\cf22\f23\fs20 f\ul0\nosu persub\cf23\f24\fs10 x\ul0\nosupersub\cf22\f23\fs20 x\tab \up0 \expndtw-2\charsc alex100 \ul0\nosupersub\cf1\f2\fs10 6 \up0 \expndtw0\charscalex134 (y)dy = \ul0\ nosupersub\cf3\f4\fs16 1 if \ul0\nosupersub\cf8\f9\fs16 x = \ul0\nosupersub\cf3\ f4\fs16 0 and \ul0\nosupersub\cf5\f6\fs14 0 otherwise.\par\pard\ql \li1273\sb0\s l-90\slmult0 \up0 \expndtw0\charscalex216 \ul0\nosupersub\cf23\f24\fs10 -5 \par\ pard\sect\sectd\fs24\paperw12240\paperh15840{\bkmkstart Pg5}{\bkmkend Pg5}\par\p ard\li1507\sb0\sl-115\slmult0\par\pard\li1507\sb0\sl-115\slmult0\par\pard\li1507 \sb0\sl-115\slmult0\par\pard\li1507\sb0\sl-115\slmult0\par\pard\li1507\sb0\sl-11 5\slmult0\par\pard\li1507\sb0\sl-115\slmult0\par\pard\li1507\sb1\sl-115\slmult0\ fi0\tx10406 \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf1\f2\fs10 LEUNG \ul0\ nosupersub\cf23\f24\fs10 et al.: \ul0\nosupersub\cf1\f2\fs10 TRAFFIC MODELS FOR WIRELESS COMMUNICATION NETWORKS\tab \dn2 \expndtw0\charscalex122 1357\par\pard\s ect\sectd\sbknone\cols2\colno1\colw6053\colsr110\colno2\colw5937\colsr160\qj \li 1502\sb0\sl-211\slmult0 \par\pard\qj \li1502\ri0\sb108\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf3\f4\fs16 As for vehicles leaving the highway, we use \ul0\nosupersub\cf8\f9\fs16 pn(t) \ul0\nosupersub\cf3\f4\fs16 a nd \line \up0 \expndtw-3\charscalex100 \ul0\nosupersub\cf8\f9\fs16 24\par\pard\q j \li1511\ri0\sb0\sl-159\slmult0\fi172 \up0 \expndtw0\charscalex111 (t) \ul0\nos

upersub\cf3\f4\fs16 to denote the fraction of noncalling and calling vehicles \u p0 \expndtw0\charscalex120 departing when they pass by the ith entrance/exit at time \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf8\f9\fs16 t, \ul0\nosupersub \cf3\f4\fs16 respectively. If these departing vehicles leave at the same \up0 \e xpndtw0\charscalex111 velocity as they move forward along the highway, then\par\ pard\ql \li2443\sb26\sl-369\slmult0\tx5524 \up0 \expndtw0\charscalex114 \ul0\nos upersub\cf8\f9\fs16 #(x, t) = v(x,t) \ul0\nosupersub\cf24\f25\fs34 E \ul0\nosupe rsub\cf8\f9\fs16 24.(t)8(x \u8212? yi)\tab \up0 \expndtw0\charscalex106 \ul0\nos upersub\cf3\f4\fs16 (2.18)\par\pard\ql \li1511\sb0\sl-184\slmult0 \par\pard\ql \ li1511\sb104\sl-184\slmult0 \up0 \expndtw0\charscalex105 and\par\pard\ql \li2428 \sb30\sl-355\slmult0\tx5524 \up0 \expndtw-3\charscalex100 \ul0\nosupersub\cf25\f 26\fs34 7\up0 \expndtw0\charscalex107 (x, \ul0\nosupersub\cf8\f9\fs16 t) = v(x, \ul0\nosupersub\cf24\f25\fs34 E \ul0\nosupersub\cf8\f9\fs16 pg(t)6(x \u8212? yz) .\tab \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 (2.19)\par\pard\q l \li1511\sb0\sl-184\slmult0 \par\pard\ql \li1511\sb0\sl-184\slmult0 \par\pard\q l \li1511\sb5\sl-184\slmult0 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf8\f9 \fs16 B. The Time-Homogeneous Deterministic Fluid Model\par\pard\qj \li1516\ri2\ sb71\sl-212\slmult0\fi172 \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf3\f4\fs 16 We now cease to focus on time dynamics, and assume \up0 \expndtw0\charscalex1 19 that the system has reached a steady state with respect to \up0 \expndtw0\cha rscalex110 time. As a result, all system variables and parameters become \up0 \e xpndtw0\charscalex119 independent of time. For this reason, we simply drop the \ up0 \expndtw0\charscalex108 variable t from our previously deined notation, and use primes \up0 \expndtw0\charscalex114 to denote derivatives w.r.t. \ul0\nosupe rsub\cf8\f9\fs16 x.\par\pard\ql \li1699\sb23\sl-184\slmult0 \up0 \expndtw0\chars calex119 Theorem 2.3: \ul0\nosupersub\cf3\f4\fs16 At temporal equilibrium, the d ensities of\par\pard\column \qj \li6167\sb0\sl-210\slmult0 \par\pard\qj \li24\ri 1419\sb125\sl-210\slmult0 \up0 \expndtw0\charscalex116 have a clear physical mea ning and are natural for the time-\line \up0 \expndtw0\charscalex115 homogeneous model under consideration. When \ul0\nosupersub\cf8\f9\fs16 A(x) \up0 \expndtw0 \charscalex119 = \ul0\nosupersub\cf3\f4\fs16 A \line \up0 \expndtw0\charscalex12 2 and \ul0\nosupersub\cf8\f9\fs16 p(x) \up0 \expndtw0\charscalex114 = 13, \ul0\n osupersub\cf3\f4\fs16 vehicles initiate and terminate calls at rates \line \up0 \expndtw0\charscalex114 independent of their locations. The condition \up0 \expn dtw0\charscalex126 \ul0\nosupersub\cf8\f9\fs16 #(x) \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf3\f4\fs16 = \ul0\nosupersub\cf8\f9\fs16 -y(x) \line \up0 \exp ndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 indicates that a vehicle departs from the highway at the same \line \up0 \expndtw0\charscalex108 rate, regardles s of whether it is a calling or noncalling vehicle. \line \up0 \expndtw0\charsca lex130 The ratio \ul0\nosupersub\cf8\f9\fs16 eVx)le \up0 \expndtw0\charscalex111 , (x) \up0 \expndtw0\charscalex126 = q(xo)/n(xo) \up0 \expndtw0\charscalex137 = p, \ul0\nosupersub\cf3\f4\fs16 means that \line \up0 \expndtw0\charscalex111 th e proportion of vehicles arriving to the highway at location \line \up0 \expndtw 0\charscalex118 \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 which are calling vehicles is identical to that of existing \line \up0 \expndtw0\chars calex110 vehicles at location \ul0\nosupersub\cf8\f9\fs16 x0, \ul0\nosupersub\cf 3\f4\fs16 which in turn is equal to the ratio \ul0\nosupersub\cf8\f9\fs16 Ala.\p ar\pard\qj \li24\ri1414\sb2\sl-209\slmult0\fi177 \up0 \expndtw0\charscalex109 \u l0\nosupersub\cf3\f4\fs16 If the conditions for the proportionality result are s atisied, it \line \up0 \expndtw0\charscalex110 is unnecessary to use two ODE's t o describe the movement of \line \up0 \expndtw0\charscalex108 noncalling and cal ling vehicles. Instead, one ODE is suficient. \line \up0 \expndtw0\charscalex120 For this purpose, let \ul0\nosupersub\cf8\f9\fs16 L(x) \ul0\nosupersub\cf3\f4\f s16 be the total number of vehicles \line \up0 \expndtw0\charscalex111 in locati on \up0 \expndtw0\charscalex118 (0, \ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosupers ub\cf3\f4\fs16 at steady state and l(x) \up0 \expndtw0\charscalex132 = \ul0\nosu persub\cf8\f9\fs16 dL(x)/dx. \ul0\nosupersub\cf3\f4\fs16 By \line \up0 \expndtw0 \charscalex122 deinition, \ul0\nosupersub\cf8\f9\fs16 L(x) = N(x) + Q(x) \ul0\no supersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 1(x) = n(x) + q(x).\par\par d\ql \li207\sb28\sl-184\slmult0 \up0 \expndtw0\charscalex115 Theorem 2.4: \ul0\n

osupersub\cf3\f4\fs16 If the conditions in Lemma 2.4 with xo \up0 \expndtw0\cha rscalex120 =\par\pard\qj \li20\ri1433\sb9\sl-206\slmult0\fi4 \up0 \expndtw0\char scalex122 0 are satisied, then the vehicular density is given by the \line \up0 \expndtw0\charscalex110 following ODE:\par\pard\ql \li135\sb130\sl-184\slmult0\t x817\tx1542 \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf14\f15\fs16\ul dl(x)\ ul0\tab \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf8\f9\fs16 1\tab \up0 \ex pndtw0\charscalex120 \ul0\nosupersub\cf14\f15\fs16\ul dv(\ul0\nosupersub\cf8\f9\ fs16 x)\par\pard\sect\sectd\sbknone\cols3\colno1\colw6084\colsr160\colno2\colw21 98\colsr20\colno3\colw3648\colsr160\qj \li1511\ri83\sb1\sl-216\slmult0 \up0 \exp ndtw0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 noncalling and calling vehicles, \ul0\nosupersub\cf8\f9\fs16 n(x) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersu b\cf8\f9\fs16 q(x), \ul0\nosupersub\cf3\f4\fs16 at any location \up0 \expndtw0\c harscalex117 \ul0\nosupersub\cf8\f9\fs16 x > \ul0\nosupersub\cf3\f4\fs16 0 satis fy the following ODE's:\par\pard\ql \li1516\sb161\sl-184\slmult0\tx2678 \up0 \ex pndtw0\charscalex127 \ul0\nosupersub\cf8\f9\fs16 v(x)n'(x) =\tab \up0 \expndtw0\ charscalex114 (x) + (x)q(x) \u8212? [fi(x) + A(x) + v/ (x)1n(x)\par\pard\ql \li55 29\sb23\sl-184\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 (2.20)\par\pard\column \ql \li2108\sb0\sl-144\slmult0 \up0 \expndtw-3\charscalex 100 \ul0\nosupersub\cf8\f9\fs16 1\par\pard\ql \li150\sb0\sl-136\slmult0\tx520\tx 721\tx1571 \up0 \expndtw0\charscalex118 dx\tab \up0 \expndtw0\charscalex108 =\ta b \up0 \expndtw0\charscalex139 v\u8216?x \ul0\nosupersub\cf26\f27\fs28 L\tab \up 0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 d\ul0\nosupersub\cf14\f15\ fs16\ul x\par\pard\ql \li20\sb80\sl-184\slmult0 \up0 \expndtw0\charscalex113 \ul 0\nosupersub\cf3\f4\fs16 whose solution is\par\pard\ql \li1507\sb0\sl-92\slmult0 \par\pard\ql \li673\sb69\sl-92\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosup ersub\cf15\f16\fs8 e \up0 \expndtw0\charscalex225 \u8212?1(x) \ul0\nosupersub\cf 13\f14\fs8 \{ [1+ p\par\pard\ql \li121\sb12\sl-153\slmult0\tx1691 \up0 \expndtw2\charscalex100 \ul0\nosupersub\cf8\f9\fs16 1 \up0 \expndtw0\charscalex116 (x) \ up0 \expndtw0\charscalex100 =\tab \up0 \expndtw-7\charscalex91 \u8212?]\par\pard \column \ql \li20\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex117 (x)] 1(x) + e" -(x)[1 + Al\} ,\par\pard\ql \li1758\sb4\sl-184\slmult0 \up0 \expndtw0\charscalex 105 \ul0\nosupersub\cf3\f4\fs16 (2.25)\par\pard\ql \li1507\sb0\sl-184\slmult0 \p ar\pard\ql \li1507\sb0\sl-184\slmult0 \par\pard\ql \li25\sb82\sl-184\slmult0 \up 0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 e+(u)el(u)dn +1(0)v(0)),\p ar\pard\sect\sectd\sbknone\cols2\colno1\colw6003\colsr160\colno2\colw5937\colsr1 60\ql \li1516\sb27\sl-184\slmult0 \up0 \expndtw0\charscalex103 \ul0\nosupersub\c f3\f4\fs16 and\par\pard\ql \li1507\sb0\sl-184\slmult0 \par\pard\ql \li1516\sb16\ sl-184\slmult0 \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf8\f9\fs16 v(x)q/(x ) = ey (x) A(x)n(x) \u8212? [7(x) + p(x) + v'(x)]q(x)\u8226?\par\pard\ql \li5529 \sb0\sl-177\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 (2. 21)\par\pard\qj \li1511\ri12\sb16\sl-206\slmult0\fi350 \up0 \expndtw0\charscalex 114 \ul0\nosupersub\cf8\f9\fs16 Proof Set \ul0\nosupersub\cf3\f4\fs16 all variab les in (2.8) and (2.9) to be indepen-\line \up0 \expndtw0\charscalex120 dent of \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 and note that\par\pard \ql \li2347\sb82\sl-184\slmult0 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf 8\f9\fs16 d\par\pard\column \ql \li817\sb0\sl-144\slmult0\tx1786\tx2117\tx2401 \ up0 \expndtw0\charscalex123 v(x)\tab \up0 \expndtw-7\charscalex80 \ul0\nosupersu b\cf3\f4\fs16 A\tab \up0 \expndtw-2\charscalex100 o\tab \up0 \expndtw-2\charscal ex100 q\par\pard\ql \li4042\sb0\sl-134\slmult0 \up0 \expndtw0\charscalex117 ( .2 6)\par\pard\ql \li24\sb20\sl-184\slmult0 \up0 \expndtw0\charscalex101 where \ul0 \nosupersub\cf8\f9\fs16 .1-(x) \up0 \expndtw0\charscalex122 = fox Oda. \ul0\nos upersub\cf3\f4\fs16 Furthermore,\par\pard\ql \li1507\sb0\sl-184\slmult0 \par\par d\ql \li1575\sb73\sl-184\slmult0\tx2338\tx2626\tx4047 \up0 \expndtw0\charscalex1 25 \ul0\nosupersub\cf8\f9\fs16 n(x) =\tab \up0 \expndtw0\charscalex60 11\tab \up 0 \expndtw-2\charscalex100 1 \up0 \expndtw0\charscalex116 (x)\tab \up0 \expndtw0 \charscalex105 \ul0\nosupersub\cf3\f4\fs16 (2.27)\par\pard\ql \li2324\sb0\sl-134 \slmult0 \up0 \expndtw0\charscalex120 + \up0 \expndtw-2\charscalex100 \ul0\nosup ersub\cf8\f9\fs16 p\par\pard\ql \li20\sb120\sl-184\slmult0 \up0 \expndtw0\charsc alex107 \ul0\nosupersub\cf3\f4\fs16 and\par\pard\sect\sectd\sbknone\cols4\colno1 \colw2504\colsr20\colno2\colw5735\colsr40\colno3\colw432\colsr20\colno4\colw3369

\colsr160\qj \li2294\ri0\sb0\sl-128\slmult0 \up0 \expndtw0\charscalex120 \ul0\no supersub\cf8\f9\fs16 \u8212? \line \up0 \expndtw0\charscalex115 dx\par\pard\colu mn \ql \li20\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex113 [n(x)v(x)] \up0 \ex pndtw0\charscalex140 = v(x)n (x) n(x)v (x)\par\pard\ql \li5204\sb0\sl-144\slmult 0 \up0 \expndtw0\charscalex118 q(x) =\par\pard\column \ql \li1507\sb0\sl-134\slm ult0 \par\pard\ql \li1507\sb0\sl-134\slmult0 \par\pard\ql \li20\sb12\sl-134\slmu lt0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 A + \ul0\nosupersub \cf8\f9\fs16 \par\pard\column \ql \li20\sb104\sl-184\slmult0\tx1479 \up0 \expndtw -2\charscalex100 1 \up0 \expndtw0\charscalex118 (x).\tab \up0 \expndtw0\charscal ex105 \ul0\nosupersub\cf3\f4\fs16 (2.28) \par\pard\sect\sectd\sbknone \li1507\sb 0\sl-144\slmult0\fi4\tx5817 \up0 \expndtw0\charscalex113 and similarly for \ul0\ nosupersub\cf8\f9\fs16 q(x)v(x).\tab \up0 \expndtw0\charscalex113 \u10065?\par\p ard\sect\sectd\sbknone\cols2\colno1\colw6003\colsr160\colno2\colw5937\colsr160\q j \li1511\ri7\sb0\sl-207\slmult0\fi177 \up0 \expndtw0\charscalex115 \ul0\nosuper sub\cf3\f4\fs16 In general, \ul0\nosupersub\cf8\f9\fs16 n(x) \ul0\nosupersub\cf3 \f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x) \ul0\nosupersub\cf3\f4\fs16 can be solved from the coupled \line \up0 \expndtw0\charscalex112 ODE's in (2.20) and (2.21) plus initial conditions. However, \line \up0 \expndtw0\charscalex115 unde r reasonable assumptions, the two ODE's can be com-\line \up0 \expndtw0\charscal ex113 bined into one, and \ul0\nosupersub\cf8\f9\fs16 n(x) \ul0\nosupersub\cf3\f 4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x) \ul0\nosupersub\cf3\f4\fs16 can be o btained explicitly. \line \up0 \expndtw0\charscalex109 To prove this, we present the following proportionality result.\par\pard\ql \li1684\ri238\sb0\sl-269\slmu lt0\tx4915\tx5627 \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9\fs16 Lemma 2.4: \ul0\nosupersub\cf3\f4\fs16 For some xo > 0, \line\tab \up0 \expndtw0\char scalex134 \ul0\nosupersub\cf8\f9\fs16 e-(x) \tab \up0 \expndtw-7\charscalex80 \u l0\nosupersub\cf3\f4\fs16 A\par\pard\ql \li4915\sb59\sl-184\slmult0 \up0 \expndt w0\charscalex109 \ul0\nosupersub\cf8\f9\fs16 en (x)\par\pard\qj \li1689\ri12\sb2 4\sl-240\slmult0\fi2193\tx5524 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf3\ f4\fs16 for all \ul0\nosupersub\cf8\f9\fs16 x > xo.\tab \up0 \expndtw0\charscale x104 \ul0\nosupersub\cf3\f4\fs16 (2.22) \line \up0 \expndtw0\charscalex105 and\p ar\pard\column \qj \li20\ri1433\sb0\sl-206\slmult0\fi355\tx4332 \up0 \expndtw0\c harscalex114 \ul0\nosupersub\cf8\f9\fs16 Proof: \ul0\nosupersub\cf3\f4\fs16 Base d on the fact that \ul0\nosupersub\cf8\f9\fs16 1(x) = n(x) + q(x), \ul0\nosupers ub\cf3\f4\fs16 (2.25) \line \up0 \expndtw0\charscalex110 is obtained by substitu ting conditions a) and b) of Lemma 2.4 \line \up0 \expndtw0\charscalex111 into ( 2.20) and (2.21) and adding them together. Since (2.25) \line \up0 \expndtw0\cha rscalex116 is a linear, irst-order ODE, the solution is readily obtained \line \ up0 \expndtw0\charscalex116 and given by \up0 \expndtw0\charscalex118 (2.26). Th e proof is completed by using the \line \up0 \expndtw0\charscalex109 proportiona lity result in Lemma 2.4. \tab \up0 \expndtw0\charscalex101 \u10065?\par\pard\qj \li24\ri1429\sb3\sl-211\slmult0\fi91 \up0 \expndtw0\charscalex113 - For example , we can apply Theorem 2.4 to a special case \line \up0 \expndtw0\charscalex115 where vehicles arrive only at location \ul0\nosupersub\cf8\f9\fs16 x = \ul0\nosu persub\cf3\f4\fs16 0 and they never \line \up0 \expndtw0\charscalex118 leave the highway. Then, we have q(0)/n(0) = A/p, \ul0\nosupersub\cf8\f9\fs16 I(x) = \lin e \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf3\f4\fs16 0, \ul0\nosupersub\cf 8\f9\fs16 e \up0 \expndtw0\charscalex113 , (x) \up0 \expndtw0\charscalex118 = \u l0\nosupersub\cf3\f4\fs16 0 and \ul0\nosupersub\cf8\f9\fs16 e-\up0 \expndtw0\cha rscalex119 (x) \up0 \expndtw0\charscalex119 = \ul0\nosupersub\cf3\f4\fs16 0 for all \ul0\nosupersub\cf8\f9\fs16 x > \ul0\nosupersub\cf3\f4\fs16 0. The solution in \line \up0 \expndtw0\charscalex110 (2.27) and \up0 \expndtw0\charscalex109 (2 .28) becomes\par\pard\ql \li797\sb124\sl-184\slmult0\tx2430 \up0 \expndtw0\chars calex116 \ul0\nosupersub\cf18\f19\fs16\ul n(0)v(0)\ul0\tab \up0 \expndtw0\charsc alex111 \ul0\nosupersub\cf18\f19\fs16\ul q(0)v(0)\par\pard\sect\sectd\sbknone\co ls3\colno1\colw3905\colsr40\colno2\colw2068\colsr160\colno3\colw5937\colsr160\qj \li1507\sb0\sl-134\slmult0 \par\pard\qj \li1852\ri0\sb7\sl-134\slmult0\tx3446 \ up0 \expndtw0\charscalex118 \ul0\nosupersub\cf3\f4\fs16 b) \ul0\nosupersub\cf8\f 9\fs16 q(xo) \ul0\nosupersub\cf3\f4\fs16 is inite and \ul0\nosupersub\cf18\f19\f s16\ul q(xo) \line\ul0\tab \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9\f

s16 n(x0)\par\pard\ql \li2111\sb41\sl-184\slmult0\tx2716 \up0 \expndtw0\charscal ex114 q(x)\tab \up0 \expndtw-7\charscalex82 \ul0\nosupersub\cf3\f4\fs16 A\par\pa rd\column \ql \li20\sb1\sl-170\slmult0\tx236 \up0 \expndtw0\charscalex120 =\tab \up0 \expndtw-7\charscalex80 A\par\pard\ql \li1604\sb0\sl-128\slmult0 \up0 \expn dtw0\charscalex104 (2.23)\par\pard\column \ql \li197\sb0\sl-144\slmult0\tx1532\t x3130\tx4047 \up0 \expndtw0\charscalex125 \ul0\nosupersub\cf8\f9\fs16 n(x) =\tab \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf 8\f9\fs16 q(x) =\tab \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\f4\fs16 fo r \ul0\nosupersub\cf8\f9\fs16 x > \ul0\nosupersub\cf3\f4\fs16 0;\tab \up0 \expnd tw0\charscalex105 (2.29)\par\pard\ql \li961\sb0\sl-128\slmult0\tx2574 \up0 \expn dtw0\charscalex112 v(x)\tab \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf8\f9\ fs16 v(x)\par\pard\ql \li20\sb108\sl-184\slmult0 \up0 \expndtw0\charscalex120 \u l0\nosupersub\cf3\f4\fs16 i.e., the vehicular density is inversely proportional to the\par\pard\sect\sectd\sbknone\cols4\colno1\colw2062\colsr40\colno2\colw388\ colsr40\colno3\colw3497\colsr160\colno4\colw5933\colsr160\ql \li1713\sb0\sl-144\ slmult0 \up0 \expndtw0\charscalex126 hen\par\pard\column \ql \li20\sb108\sl-128\ slmult0 \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf8\f9\fs16 n(x)\par\pard\c olumn \ql \li20\sb0\sl-144\slmult0\tx3034 \up0 \expndtw0\charscalex140 \ul0\nosu persub\cf3\f4\fs16 = for all \ul0\nosupersub\cf8\f9\fs16 x > xo.\tab \up0 \expnd tw0\charscalex104 \ul0\nosupersub\cf3\f4\fs16 (2.24)\par\pard\ql \li221\sb0\sl-1 29\slmult0 \up0 \expndtw-2\charscalex100 p\par\pard\column \qj \li20\ri1433\sb0\ sl-178\slmult0 \up0 \expndtw0\charscalex113 velocity. It is noteworthy that the arrival rates of noncalling \up0 \expndtw0\charscalex117 and calling vehicles at location \up0 \expndtw0\charscalex123 0 in this special case are\par\pard\sect\ sectd\sbknone\cols2\colno1\colw6007\colsr160\colno2\colw5933\colsr160\qj \li1507 \ri1\sb6\sl-209\slmult0\fi182 \up0 \expndtw0\charscalex115 We actually obtain a stronger proportionality result for a \line \up0 \expndtw0\charscalex114 time-de pendent setting. The result given as Lemma 2.5 and \line \up0 \expndtw0\charscal ex115 its proof are presented in the Appendix. Lemma 2.4 can be \line \up0 \expn dtw0\charscalex108 viewed as a consequence of the time-dependent result since al l \line \up0 \expndtw0\charscalex110 quantities here are independent of time. We choose to present \line \up0 \expndtw0\charscalex113 Lemma 2.4 here because, as explained below, its conditions\par\pard\column \ql \li20\ri1419\sb0\sl-210\slm ult0\tx203 \up0 \expndtw0\charscalex118 e+ (0) = q(0)v(0) and e:(0) = n(0)v(0), respectively. \line\tab \up0 \expndtw0\charscalex115 It should be clear that the conditions in Theorem \up0 \expndtw0\charscalex113 2.4 are \line \up0 \expndtw0 \charscalex116 quite natural, so that we often will have the single ODE in \up0 \expndtw0\charscalex108 (2.25). On the other hand, the interest in explicitly co nsidering \up0 \expndtw0\charscalex118 two kinds of vehicles is primarily for th e situations where \up0 \expndtw0\charscalex113 these conditions do \ul0\nosuper sub\cf8\f9\fs16 not \ul0\nosupersub\cf3\f4\fs16 hold. \par\pard\sect\sectd\fs24\ paperw12240\paperh15840{\bkmkstart Pg6}{\bkmkend Pg6}\par\pard\li1622\sb0\sl-115 \slmult0\par\pard\li1622\sb0\sl-115\slmult0\par\pard\li1622\sb0\sl-115\slmult0\p ar\pard\li1622\sb0\sl-115\slmult0\par\pard\li1622\sb0\sl-115\slmult0\par\pard\li 1622\sb111\sl-115\slmult0\fi0\tx5241 \dn2 \expndtw0\charscalex125 \ul0\nosupersu b\cf1\f2\fs10 1358\tab \up0 \expndtw0\charscalex126 IEEE JOURNAL ON SELECTED ARE AS IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\ql \li1699\sb0\sl-11 5\slmult0 \par\pard\ql\li1699\sb0\sl-115\slmult0 \par\pard\ql\li1699\sb0\sl-115\ slmult0 \par\pard\ql\li1699\sb38\sl-115\slmult0 \up0 \expndtw0\charscalex115 v(x ,t) \par\pard\ql \li7204\sb26\sl-138\slmult0 \up0 \expndtw0\charscalex108 \ul0\n osupersub\cf6\f7\fs12 t = 30 min. \par\pard\ql \li8673\sb0\sl-138\slmult0 \par\p ard\ql\li8673\sb0\sl-138\slmult0 \par\pard\ql\li8673\sb0\sl-138\slmult0 \par\par d\ql\li8673\sb0\sl-138\slmult0 \par\pard\ql\li8673\sb130\sl-138\slmult0 \up0 \ex pndtw0\charscalex105 Total \par\pard\ql \li6767\sb14\sl-172\slmult0 \up0 \expndt w-2\charscalex100 \ul0\nosupersub\cf3\f4\fs16 8 \par\pard\li6767\sb0\sl-207\slmu lt0\par\pard\li6767\sb13\sl-207\slmult0\fi0\tx8491 \dn3 \expndtw-2\charscalex100 0\tab \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf6\f7\fs12 Non-Calling\par\ pard\ql \li1579\sb0\sl-115\slmult0 \par\pard\ql\li1579\sb14\sl-115\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf1\f2\fs10 0.3 \par\pard\ql \li6767\sb0 \sl-138\slmult0 \par\pard\ql\li6767\sb48\sl-138\slmult0 \up0 \expndtw-1\charscal

ex100 \ul0\nosupersub\cf6\f7\fs12 0 \par\pard\ql \li7828\sb22\sl-138\slmult0 \up 0 \expndtw0\charscalex109 Calling & Handoff Rate (Calls/Min) \par\pard\li1809\sb 111\sl-115\slmult0\fi0\tx2736\tx3043\tx3662\tx3969\tx5966 \up0 \expndtw-1\charsc alex100 \ul0\nosupersub\cf1\f2\fs10 0\tab \up0 \expndtw-1\charscalex100 3\tab \u p0 \expndtw-1\charscalex100 4\tab \up0 \expndtw-1\charscalex100 6\tab \up0 \expn dtw-1\charscalex100 7\tab \up0 \expndtw0\charscalex100 \ul0\nosupersub\cf28\f29\ fs6 X\par\pard\ql \li6767\sb36\sl-138\slmult0 \up0 \expndtw-1\charscalex100 \ul0 \nosupersub\cf6\f7\fs12 0 \par\pard\li1622\sb13\sl-115\slmult0\fi4\tx1910\tx2140 \up0 \expndtw0\charscalex130 \ul0\nosupersub\cf1\f2\fs10 Fig.\tab \up0 \expndtw -1\charscalex100 1.\tab \up0 \expndtw0\charscalex129 Vehicle velocity as \ul0\no supersub\cf6\f7\fs12 a function of location for 35 < \ul0\nosupersub\cf27\f28\fs 12 t \ul0\nosupersub\cf6\f7\fs12 < 55 min.\par\pard\li1622\sb92\sl-138\slmult0\f i5510\tx7464\tx7790\tx8116\tx8447\tx8745\tx9072\tx9398\tx9724\tx10055\tx10372 \u p0 \expndtw-1\charscalex100 0\tab \up0 \expndtw-1\charscalex100 2\tab \dn2 \expn dtw-1\charscalex100 4\tab \dn2 \expndtw-1\charscalex100 6\tab \dn2 \expndtw-1\ch arscalex100 8\tab \dn2 \expndtw0\charscalex101 10\tab \dn2 \expndtw0\charscalex1 01 12\tab \dn2 \expndtw0\charscalex101 14\tab \dn2 \expndtw0\charscalex101 16\ta b \dn2 \expndtw-1\charscalex100 18\tab \dn2 \expndtw0\charscalex109 20\par\pard\ sect\sectd\sbknone\cols2\colno1\colw6163\colsr110\colno2\colw5827\colsr160\qj \l i1622\ri6\sb95\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex115 \ul0\nosupersub \cf3\f4\fs16 Once \ul0\nosupersub\cf8\f9\fs16 n(x) \ul0\nosupersub\cf3\f4\fs16 a nd \ul0\nosupersub\cf8\f9\fs16 q(x) \ul0\nosupersub\cf3\f4\fs16 are computed fro m (2.20) and (2.21) \line \up0 \expndtw0\charscalex115 or (2.27) and \up0 \expnd tw0\charscalex118 (2.28), the offered load and call handof rate \line \up0 \expn dtw0\charscalex118 for cell i can be obtained from \up0 \expndtw0\charscalex112 (2.13) and \up0 \expndtw0\charscalex114 (2.14), with the \line \up0 \expndtw0\ch arscalex113 variable \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 o mitted.\par\pard\ql \li1627\sb0\sl-184\slmult0 \par\pard\ql \li2827\sb79\sl-184\ slmult0 \up0 \expndtw-4\charscalex100 III. NUMERICAL EXAMPLES\par\pard\qj \li162 2\ri2\sb77\sl-212\slmult0\fi177 \up0 \expndtw0\charscalex110 In this section, we present numerical examples to illustrate \up0 \expndtw0\charscalex115 the time and space dynamics captured by the deterministic \up0 \expndtw0\charscalex118 fl uid model in Section II. As indicated in Section I, these \up0 \expndtw0\charsca lex109 examples also apply to the stochastic models to be introduced \up0 \expnd tw0\charscalex114 later, but with a diferent interpretation.\par\pard\qj \li1627 \ri1\sb0\sl-209\slmult0\fi177 \up0 \expndtw0\charscalex117 The examples here ass ume no constraint on the number \line \up0 \expndtw0\charscalex113 of available channels. Furthermore, unless stated otherwise, \line \up0 \expndtw0\charscalex1 11 the average think time (time before initiating a call) and call-\line \up0 \e xpndtw0\charscalex109 holding time are 10 and 2 min, respectively. That is, A(x, \ul0\nosupersub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 =\par\pard\qj \li162 7\ri0\sb8\sl-211\slmult0\fi4 \up0 \expndtw0\charscalex111 0.1 and p,(x, \ul0\nos upersub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf3\f4\fs16 = 0.5 for all \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 > 0 and \u8212?oo < \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 < oo. The \line \up0 \expndtw0\charscalex116 highway has a single entrance at location \u p0 \expndtw0\charscalex119 0 at which only \line \up0 \expndtw0\charscalex113 no ncalling vehicle arrive at a constant rate (denoted by a) of\par\pard\qj \li1627 \ri1\sb10\sl-206\slmult0\fi4 \up0 \expndtw0\charscalex111 30 cars/min, and vehic les are assumed not to depart from the \up0 \expndtw0\charscalex105 highway.\par \pard\qj \li1631\ri2\sb1\sl-212\slmult0\fi172 \up0 \expndtw0\charscalex111 We ir st consider a time-dependent case where the velocity \line \up0 \expndtw0\charsc alex115 ield v(x, \ul0\nosupersub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 is a function of location and time. It is assumed \line \up0 \expndtw0\charscalex11 2 that v(x, \ul0\nosupersub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex119 \ul0\nos upersub\cf3\f4\fs16 = 1 km/min for all \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupe rsub\cf3\f4\fs16 > 0 when \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\f s16 < 35 or \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 > 55 \line \up0 \expndtw0\charscalex116 min. However, for 35 < \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 < 55 min, the velocity ield is\par\pard\ql \li298

5\ri875\sb77\sl-211\slmult0\fi9\tx4276 \up0 \expndtw-2\charscalex100 1\tab \up0 \expndtw0\charscalex122 if \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\ fs16 < 3 \line \up0 \expndtw0\charscalex100 1\u8212? \up0 \expndtw0\charscalex10 7 0.7 \up0 \expndtw0\charscalex130 (x \u8212? 3) if 3 < \ul0\nosupersub\cf8\f9\f s16 x \ul0\nosupersub\cf3\f4\fs16 < 4\par\pard\ql \li2121\sb23\sl-184\slmult0\tx 4276\tx5730 \up0 \expndtw0\charscalex114 v(x, \ul0\nosupersub\cf8\f9\fs16 t) \up 0 \expndtw0\charscalex120 \ul0\nosupersub\cf3\f4\fs16 =\tab \up0 \expndtw0\chars calex125 if 4 < \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 < 6\ta b \up0 \expndtw0\charscalex103 (3.1)\par\pard\qj \li2985\ri871\sb1\sl-216\slmult 0\tx4273 \up0 \expndtw0\charscalex118 0.3 + 0.7(x \u8212? 6) if 6 < \ul0\nosuper sub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 < 7 \line \up0 \expndtw-2\charscal ex100 1 \tab \up0 \expndtw0\charscalex122 if \ul0\nosupersub\cf8\f9\fs16 x \ul0\ nosupersub\cf3\f4\fs16 > 7.\par\pard\qj \li1636\ri0\sb97\sl-211\slmult0\fi172 \u p0 \expndtw0\charscalex115 This velocity ield is depicted in Fig. \up0 \expndtw 0\charscalex113 1. It is U-shaped as \line \up0 \expndtw0\charscalex115 a functi on of location on the highway for 35 \up0 \expndtw0\charscalex134 < \ul0\nosuper sub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 < 55, so \line \up0 \expndtw0\char scalex114 that it can be used to simulate the slowing down of trafic in \line \u p0 \expndtw0\charscalex109 the time interval due to an accident. For this exampl e, Lemma\par\pard\qj \li1636\ri0\sb0\sl-211\slmult0 \up0 \expndtw0\charscalex115 2.3 can be applied to convert the PDE's (2.8) and (2.9) into \up0 \expndtw0\cha rscalex111 the ODE's (2.10)\u8212?(2.12). By using the extrapolated Gragg's \up0 \expndtw0\charscalex110 modiied mid-point method [l], contained in the PORT lib rary \up0 \expndtw0\charscalex111 developed at AT&T Bell Laboratories, we numeri cally solve \up0 \expndtw0\charscalex114 the ODE's with the initial condition of n(0, t) = a/v(0, t) for \up0 \expndtw0\charscalex119 all \ul0\nosupersub\cf8\f9 \fs16 t \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf3\f4\fs16 > 0 in order to obtain \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0 \nosupersub\cf8\f9\fs16 q(x, t).\par\pard\qj \li1631\ri0\sb0\sl-209\slmult0\fi18 2 \up0 \expndtw0\charscalex100 \ul0\nosupersub\cf3\f4\fs16 Figs. \up0 \expndtw0 \charscalex113 2-5 show the total vehicular density, the densities of \line \up0 \expndtw0\charscalex111 noncalling and calling vehicles, and the call handoff r ate as a\par\pard\column \ql \li6273\sb0\sl-138\slmult0 \par\pard\ql \li1724\sb3 1\sl-138\slmult0 \up0 \expndtw0\charscalex106 \ul0\nosupersub\cf6\f7\fs12 Locati on on the Highway (Km)\par\pard\ql \li6273\sb0\sl-138\slmult0 \par\pard\ql \li20 \sb41\sl-138\slmult0\tx544 \up0 \expndtw0\charscalex108 Fig. \up0 \expndtw-1\cha rscalex100 2.\tab \up0 \expndtw0\charscalex118 Vehicle density in the highway sy stem with \ul0\nosupersub\cf1\f2\fs10 a \ul0\nosupersub\cf6\f7\fs12 traffic acci dent.\par\pard\qj \li6777\sb0\sl-211\slmult0 \par\pard\qj \li524\ri4251\sb210\sl -211\slmult0\fi436 \up0 \expndtw0\charscalex106 t = 40 min. \line \up0 \expndtw1\charscalex100 n\par\pard\ql \li6777\sb0\sl-115\slmult0 \par\pard\ql \li524\sb1 8\sl-115\slmult0\tx740 \up0 \expndtw0\charscalex100 \ul0\nosupersub\cf1\f2\fs10 co\tab \up0 \expndtw-1\charscalex100 -\par\pard\ql \li524\sb63\sl-138\slmult0 \u p0 \expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 0\par\pard\ql \li519\sb1 31\sl-138\slmult0 \up0 \expndtw0\charscalex104 0\ul0\nosupersub\cf28\f29\fs6 0\p ar\pard\ql \li6772\sb0\sl-92\slmult0 \par\pard\ql \li6772\sb0\sl-92\slmult0 \par \pard\ql \li6772\sb0\sl-92\slmult0 \par\pard\ql \li6772\sb0\sl-92\slmult0 \par\p ard\ql \li6772\sb0\sl-92\slmult0 \par\pard\ql \li6772\sb0\sl-92\slmult0 \par\par d\ql \li6772\sb0\sl-92\slmult0 \par\pard\ql \li519\sb58\sl-92\slmult0\tx3404 \up 0 \expndtw-1\charscalex100 0\tab \up0 \expndtw0\charscalex103 \ul0\nosupersub\cf 6\f7\fs12 Total\par\pard\ql \li6772\sb0\sl-103\slmult0 \par\pard\ql \li519\sb64\ sl-103\slmult0\tx1480\tx3217 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf28\ f29\fs6 0\tab \up0 \expndtw0\charscalex102 \ul0\nosupersub\cf6\f7\fs12 Calling\t ab \up0 \expndtw0\charscalex104 Non-Calling\par\pard\ql \li519\sb0\sl-57\slmult0 \up0 \expndtw0\charscalex106 \ul0\nosupersub\cf28\f29\fs6 CV\par\pard\ql \li677 2\sb0\sl-80\slmult0 \par\pard\ql \li519\sb45\sl-80\slmult0\tx1566 \up0 \expndtw1\charscalex100 0\tab \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf23\f24\fs1 0 r\par\pard\ql \li6772\sb0\sl-115\slmult0 \par\pard\ql \li519\sb84\sl-115\slmul t0\tx942 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf28\f29\fs6 0\tab \up0 \ expndtw0\charscalex127 \ul0\nosupersub\cf6\f7\fs12 --\up0 \expndtw-4\charscalex1

00 \u8212?a-ndof Rate ( \up0 \expndtw0\charscalex239 a in\par\pard\ql \li6278\sb 0\sl-138\slmult0 \par\pard\ql \li889\sb74\sl-138\slmult0\tx1225\tx1547\tx1878\tx 2204\tx2507\tx2833\tx3160\tx3491\tx3812\tx4129 \up0 \expndtw-1\charscalex100 0\t ab \up0 \expndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 4\tab \up0 \expndtw-1\charscalex100 6\tab \up0 \expndtw-1\charscalex100 8\tab \up0 \expndtw 0\charscalex101 10\tab \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf1\f2\fs10 12\tab \up0 \expndtw0\charscalex117 14\tab \up0 \expndtw0\charscalex112 16\tab \ up0 \expndtw0\charscalex122 18\tab \up0 \expndtw0\charscalex127 20\par\pard\ql \ li6278\sb0\sl-138\slmult0 \par\pard\ql \li1734\sb26\sl-138\slmult0 \up0 \expndtw 0\charscalex106 \ul0\nosupersub\cf6\f7\fs12 Location on the Highway (Km)\par\par d\ql \li6278\sb0\sl-138\slmult0 \par\pard\ql \li25\sb22\sl-138\slmult0\tx548 \up 0 \expndtw0\charscalex111 Fig. \up0 \expndtw-1\charscalex100 3.\tab \up0 \expndt w0\charscalex120 Vehicle density in the highway system with \ul0\nosupersub\cf1\ f2\fs10 a \ul0\nosupersub\cf6\f7\fs12 trafic accident.\par\pard\qj \li6278\sb0\s l-217\slmult0 \par\pard\qj \li25\ri1313\sb108\sl-217\slmult0 \up0 \expndtw0\char scalex113 \ul0\nosupersub\cf3\f4\fs16 function of location at time prior (Fig. 2 ), during (Fig. \up0 \expndtw0\charscalex111 3-4), \line \up0 \expndtw0\charsca lex114 and after \up0 \expndtw0\charscalex102 (Fig. \up0 \expndtw0\charscalex123 5) the trafic accident. For t \up0 \expndtw0\charscalex127 = 30 in Fig. \line \ up0 \expndtw0\charscalex121 2, \ul0\nosupersub\cf8\f9\fs16 n(x,t) \ul0\nosupersu b\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x,t) \ul0\nosupersub\cf3\f4\fs16 reach their "equilibrium" solution for \line \up0 \expndtw0\charscalex108 sufic iently large \ul0\nosupersub\cf8\f9\fs16 x. \ul0\nosupersub\cf3\f4\fs16 The reas on for this is explained as follows. \line \up0 \expndtw0\charscalex122 Recall t hat all vehicles arriving at \ul0\nosupersub\cf8\f9\fs16 x \up0 \expndtw0\charsc alex125 \ul0\nosupersub\cf3\f4\fs16 = 0 are noncalling \line \up0 \expndtw0\char scalex115 vehicles. They start to make calls as they move forward on \line \up0 \expndtw0\charscalex109 the highway. As a result, the density of noncalling and calling \line \up0 \expndtw0\charscalex110 vehicles decreases and increases, res pectively, as \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 increase s. \line \up0 \expndtw0\charscalex115 Since vehicles initiate and terminate call s independently at \line \up0 \expndtw0\charscalex111 constant rates of A(x, \ul 0\nosupersub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 = 0.1 and \ul0\nosupersu b\cf8\f9\fs16 ic(x. t) \ul0\nosupersub\cf3\f4\fs16 = 0.5, respectively, \line \u p0 \expndtw0\charscalex109 such decrease and increase of vehicular densities app roach an \line \up0 \expndtw0\charscalex109 equilibrium at locations farther dow n the highway. In fact, the \line \up0 \expndtw0\charscalex121 ratio \ul0\nosupe rsub\cf8\f9\fs16 n(x)/q(x) \ul0\nosupersub\cf3\f4\fs16 in this case tends to equ al to \ul0\nosupersub\cf8\f9\fs16 p,(x,t)IA(x,t) \line \up0 \expndtw0\charscalex 115 \ul0\nosupersub\cf3\f4\fs16 for suficiently large \ul0\nosupersub\cf8\f9\fs1 6 x. \ul0\nosupersub\cf3\f4\fs16 Since \ul0\nosupersub\cf8\f9\fs16 v(x, t) \ul0\ nosupersub\cf3\f4\fs16 = 1 at \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\ f4\fs16 = 30, according \line \up0 \expndtw0\charscalex114 to (2.14), the densit y of calling vehicles equals handof rate, \line \up0 \expndtw0\charscalex114 so that their curves shown in the igure coincide.\par\pard\qj \li34\ri1304\sb0\sl-2 19\slmult0\fi177 \up0 \expndtw0\charscalex120 At \ul0\nosupersub\cf8\f9\fs16 t \ ul0\nosupersub\cf3\f4\fs16 = 40 in Fig. 3, vehicles start to build up sharply at \up0 \expndtw0\charscalex106 location \up0 \expndtw0\charscalex111 (3, 7] where velocity is relatively low. Accordingly, \par\pard\sect\sectd\fs24\paperw12240\ paperh15840{\bkmkstart Pg7}{\bkmkend Pg7}\ql \li2793\sb106\sl-138\slmult0 \up0 \ expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 \u9642?\par\pard\sect\sectd\ sbknone\cols2\colno1\colw6272\colsr160\colno2\colw5668\colsr160\ql \li1416\sb0\s l-138\slmult0 \par\pard\ql \li1416\sb0\sl-138\slmult0 \par\pard\ql \li1416\sb0\s l-138\slmult0 \par\pard\ql \li1416\sb0\sl-138\slmult0 \par\pard\ql \li1416\sb126 \sl-138\slmult0 \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf1\f2\fs10 LEUNG \ ul0\nosupersub\cf27\f28\fs12 et at.: \ul0\nosupersub\cf1\f2\fs10 TRAFFIC MODELS FOR WIRELESS COMMUNICATION NETWORKS\par\pard\ql \li2299\sb0\sl-138\slmult0 \par\ pard\ql \li2299\sb0\sl-138\slmult0 \par\pard\ql \li2299\sb0\sl-138\slmult0 \par\ pard\ql \li2299\sb111\sl-138\slmult0 \up0 \expndtw0\charscalex145 \ul0\nosupersu b\cf6\f7\fs12 t. 50\par\pard\ql \li1857\sb39\sl-92\slmult0 \up0 \expndtw-1\chars

calex100 \ul0\nosupersub\cf15\f16\fs8 0\par\pard\column \ql \li10305\sb0\sl-138\ slmult0 \par\pard\ql \li10305\sb0\sl-138\slmult0 \par\pard\ql \li10305\sb0\sl-13 8\slmult0 \par\pard\ql \li10305\sb0\sl-138\slmult0 \par\pard\ql \li3893\sb131\sl -138\slmult0 \up0 \expndtw-3\charscalex100 \ul0\nosupersub\cf6\f7\fs12 1359\par\ pard\ql \li8025\sb0\sl-138\slmult0 \par\pard\ql \li8025\sb0\sl-138\slmult0 \par\ pard\ql \li1613\sb100\sl-138\slmult0 \up0 \expndtw0\charscalex119 TABLE I\par\pa rd\ql \li20\sb30\sl-138\slmult0 \up0 \expndtw-4\charscalex100 APPROXIMATE BLOCKI NG PROBABILITIES FOR THE FIXED CELL SIZE\par\pard\ql \li1857\sb0\sl-138\slmult0 \par\pard\ql \li1508\sb50\sl-138\slmult0\tx2612\tx3428 \up0 \expndtw0\charscalex 115 Offered Load\tab \up0 \expndtw0\charscalex121 No. of\tab \up0 \expndtw0\char scalex114 Blocking\par\pard\sect\sectd\sbknone\cols5\colno1\colw6147\colsr160\co lno2\colw550\colsr160\colno3\colw916\colsr160\colno4\colw714\colsr160\colno5\col w3163\colsr160\ql \li1862\sb20\sl-92\slmult0 \up0 \expndtw-3\charscalex86 \ul0\n osupersub\cf15\f16\fs8 CO\par\pard\ql \li1857\sb0\sl-92\slmult0 \par\pard\ql \li 1862\sb13\sl-92\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf13\f14\f s8 0\par\pard\ql \li1857\sb0\sl-92\slmult0 \par\pard\ql \li1857\sb0\sl-92\slmult 0 \par\pard\ql \li1857\sb0\sl-92\slmult0 \par\pard\ql \li1857\sb0\sl-92\slmult0 \par\pard\ql \li1857\sb77\sl-92\slmult0 \up0 \expndtw-1\charscalex100 0\par\pard \ql \li1857\sb0\sl-92\slmult0 \par\pard\ql \li1857\sb90\sl-92\slmult0 \up0 \expn dtw-1\charscalex100 \ul0\nosupersub\cf15\f16\fs8 0\par\pard\ql \li3753\sb100\sl115\slmult0 \up0 \expndtw0\charscalex129 \ul0\nosupersub\cf1\f2\fs10 Total\par\p ard\ql \li1857\sb0\sl-144\slmult0 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\ cf29\f30\fs18 8\par\pard\ql \li1857\sb1\sl-192\slmult0\tx2889\tx3571 \up0 \expnd tw0\charscalex45 0\ul0\nosupersub\cf13\f14\fs8 0.1\tab \up0 \expndtw0\charscalex 103 \ul0\nosupersub\cf6\f7\fs12 Calling\tab \up0 \expndtw0\charscalex104 Non-Cal ling\par\pard\ql \li1862\sb0\sl-92\slmult0 \par\pard\ql \li1862\sb58\sl-92\slmul t0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf15\f16\fs8 0\par\pard\ql \li1 862\sb0\sl-115\slmult0 \par\pard\ql \li1862\sb81\sl-115\slmult0\tx2255 \up0 \exp ndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 0\tab \up0 \expndtw-5\charscale x78 \ul0\nosupersub\cf6\f7\fs12 1--" \up0 \expndtw0\charscalex107 H"\u8212?aidof Rate (CallMin)\par\pard\column \ql \li77\sb0\sl-108\slmult0 \up0 \expndtw0\chars calex111 Time\par\pard\ql \li6307\sb0\sl-138\slmult0 \par\pard\ql \li20\sb25\sl138\slmult0 \up0 \expndtw0\charscalex139 \ul0\nosupersub\cf27\f28\fs12 t \ul0\no supersub\cf6\f7\fs12 = 30\par\pard\ql \li6312\sb0\sl-138\slmult0 \par\pard\ql \l i6312\sb0\sl-138\slmult0 \par\pard\ql \li6312\sb0\sl-138\slmult0 \par\pard\ql \l i6312\sb0\sl-138\slmult0 \par\pard\ql \li25\sb16\sl-138\slmult0\tx130 \up0 \expn dtw-1\charscalex100 \ul0\nosupersub\cf27\f28\fs12 t\tab \up0 \expndtw0\charscale x132 \ul0\nosupersub\cf6\f7\fs12 = 50\par\pard\column \ql \li20\sb0\sl-108\slmul t0 \up0 \expndtw0\charscalex116 Cell Location\par\pard\qj \li280\ri337\sb128\sl180\slmult0 \up0 \expndtw0\charscalex110 (2,4] \line \up0 \expndtw0\charscalex11 0 (4,6] \line \up0 \expndtw0\charscalex108 (6,8]\par\pard\ql \li241\sb33\sl-138\ slmult0 \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf30\f31\fs12\ul (8,10]\par \pard\qj \li241\ri314\sb4\sl-175\slmult0\fi38\tx290\tx290\tx290\tx290 \up0 \expn dtw0\charscalex108 \ul0\nosupersub\cf6\f7\fs12 (2,4] \line\tab \up0 \expndtw0\ch arscalex108 (4,6] \line \tab \up0 \expndtw0\charscalex108 (4,6] \line \tab \up0 \expndtw0\charscalex108 (6,8] \line \tab \up0 \expndtw0\charscalex108 (6,8] \lin e \up0 \expndtw0\charscalex108 (8,10]\par\pard\column \qj \li20\ri194\sb0\sl-216 \slmult0\tx69 \up0 \expndtw0\charscalex112 (Erlang) \line\tab \up0 \expndtw0\cha rscalex110 8.2455\par\pard\ql \li59\sb21\sl-138\slmult0 \up0 \expndtw0\charscale x112 9.4716\par\pard\ql \li59\sb44\sl-138\slmult0 \up0 \expndtw0\charscalex112 9 .8408\par\pard\ql \li59\sb35\sl-138\slmult0 \up0 \expndtw0\charscalex113 \ul0\no supersub\cf30\f31\fs12\ul 9.9520\par\pard\ql \li39\sb39\sl-138\slmult0 \up0 \exp ndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 1 \up0 \expndtw0\charscalex109 1.2459\par\pard\ql \li20\sb40\sl-138\slmult0 \up0 \expndtw0\charscalex113 29.497 7\par\pard\ql \li20\sb35\sl-138\slmult0 \up0 \expndtw0\charscalex113 29.4977\par \pard\ql \li34\sb39\sl-138\slmult0 \up0 \expndtw0\charscalex109 13.2327\par\pard \ql \li34\sb35\sl-138\slmult0 \up0 \expndtw0\charscalex109 13.2327\par\pard\ql \ li87\sb40\sl-138\slmult0 \up0 \expndtw0\charscalex115 9.9980\par\pard\column \ql \li20\ri1808\sb0\sl-108\slmult0\tx1033 \up0 \expndtw0\charscalex114 Channels \l ine\tab \up0 \expndtw0\charscalex111 Prob.\par\pard\ql \li49\sb0\sl-108\slmult0

\up0 \expndtw0\charscalex115 \ul0\nosupersub\cf30\f31\fs12\ul Per Cell\par\pard\ ql \li217\sb1\sl-118\slmult0\tx947 \up0 \expndtw0\charscalex109 \ul0\nosupersub\ cf6\f7\fs12 20\tab \up0 \expndtw0\charscalex113 0.00023\par\pard\ql \li217\sb39\ sl-138\slmult0\tx947 \up0 \expndtw0\charscalex105 20\tab \up0 \expndtw0\charscal ex114 0.00107\par\pard\ql \li212\sb49\sl-138\slmult0\tx947 \up0 \expndtw0\charsc alex109 20\tab \up0 \expndtw0\charscalex114 0.00159\par\pard\ql \li212\sb40\sl-1 38\slmult0\tx947 \up0 \expndtw0\charscalex109 \ul0\nosupersub\cf30\f31\fs12\ul 2 0\ul0\tab \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf30\f31\fs12\ul 0.00178\ par\pard\ql \li217\sb34\sl-138\slmult0\tx947 \up0 \expndtw0\charscalex105 \ul0\n osupersub\cf6\f7\fs12 20\tab \up0 \expndtw0\charscalex113 0.00565\par\pard\ql \l i212\sb40\sl-138\slmult0\tx947 \up0 \expndtw0\charscalex109 20\tab \up0 \expndtw 0\charscalex113 0.37105\par\pard\ql \li207\sb39\sl-138\slmult0\tx947 \up0 \expnd tw0\charscalex109 45\tab \up0 \expndtw0\charscalex114 0.00179\par\pard\ql \li212 \sb35\sl-138\slmult0\tx942 \up0 \expndtw0\charscalex109 20\tab \up0 \expndtw0\ch arscalex113 0.02055\par\pard\qj \li203\ri1730\sb12\sl-172\slmult0\fi9\tx942\tx93 6 \up0 \expndtw0\charscalex101 25\tab \up0 \expndtw0\charscalex114 0.00127 \line \up0 \expndtw0\charscalex113 20 \tab \up0 \expndtw0\charscalex113 0.00187\par\p ard\sect\sectd\sbknone\cols2\colno1\colw5907\colsr160\colno2\colw6033\colsr160\q l \li1420\sb0\sl-138\slmult0 \par\pard\ql \li2227\sb65\sl-138\slmult0\tx2558\tx2 875\tx3211\tx3527\tx3835\tx4161\tx4487\tx4814\tx5140\tx5457 \up0 \expndtw-5\char scalex81 O\tab \up0 \expndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 4\tab \up0 \expndtw-1\charscalex100 6\tab \up0 \expndtw-1\charscalex100 8\tab \ up0 \expndtw0\charscalex101 10\tab \up0 \expndtw0\charscalex101 12\tab \up0 \exp ndtw0\charscalex101 14\tab \up0 \expndtw0\charscalex101 16\tab \up0 \expndtw0\ch arscalex101 18\tab \up0 \expndtw0\charscalex109 20\par\pard\ql \li1420\sb0\sl-13 8\slmult0 \par\pard\ql \li3067\sb26\sl-138\slmult0 \up0 \expndtw0\charscalex106 Location on the Highway (Km)\par\pard\ql \li1420\sb0\sl-138\slmult0 \par\pard\ql \li1420\sb41\sl-138\slmult0\tx1929 \up0 \expndtw0\charscalex108 Fig. \up0 \expn dtw-1\charscalex100 4.\tab \up0 \expndtw0\charscalex119 Vehicle density in the h ighway system with a trafic accident.\par\pard\ql \li2308\sb0\sl-138\slmult0 \pa r\pard\ql \li2308\sb0\sl-138\slmult0 \par\pard\ql \li2308\sb0\sl-138\slmult0 \pa r\pard\ql \li2308\sb77\sl-138\slmult0 \up0 \expndtw0\charscalex106 t = 60 min.\p ar\pard\qj \li1876\sb0\sl-95\slmult0 \par\pard\qj \li1876\sb0\sl-95\slmult0 \par \pard\qj \li1876\sb0\sl-95\slmult0 \par\pard\qj \li1876\ri3890\sb21\sl-95\slmult 0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 o \line \up0 \expndt w-2\charscalex100 co\par\pard\ql \li1876\sb0\sl-92\slmult0 \par\pard\ql \li1876\ sb13\sl-92\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf13\f14\fs8 0\ par\pard\ql \li1876\sb0\sl-92\slmult0 \par\pard\ql \li1876\sb89\sl-92\slmult0 \u p0 \expndtw-3\charscalex80 0<0\par\pard\ql \li1876\sb0\sl-92\slmult0 \par\pard\q l \li1876\sb80\sl-92\slmult0 \up0 \expndtw-1\charscalex100 0\par\pard\ql \li1876 \sb0\sl-92\slmult0 \par\pard\ql \li1876\sb85\sl-92\slmult0 \up0 \expndtw-1\chars calex100 0\par\pard\ql \li2880\sb77\sl-138\slmult0 \up0 \expndtw0\charscalex103 \ul0\nosupersub\cf6\f7\fs12 Total\par\pard\ql \li1876\sb0\sl-72\slmult0 \up0 \ex pndtw-1\charscalex100 \ul0\nosupersub\cf13\f14\fs8 0\par\pard\ql \li2692\sb0\sl138\slmult0 \par\pard\ql \li2692\sb8\sl-138\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf6\f7\fs12 Non-Calling\par\pard\column \qj \li25\ri1524\sb60\s l-211\slmult0\fi4 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf3\f4\fs16 not c losely relect the vehicle movement following a trafic \line \up0 \expndtw0\chars calex109 accident because the model allows vehicles to resume moving \line \up0 \expndtw0\charscalex112 at a speciic velocity regardless of the high vehicular d ensity. \line \up0 \expndtw0\charscalex117 In real situation, with the vehicles close to location \up0 \expndtw0\charscalex132 7 irst \line \up0 \expndtw0\chars calex112 resuming the normal velocity after the accident, the vehicles \line \up 0 \expndtw0\charscalex107 accumulated in location (3, 7] will slowly "difuse" to the right. \line \up0 \expndtw0\charscalex118 Such difusion type of vehicle mov ement can be captured, \line \up0 \expndtw0\charscalex116 if the model is augmen ted with an appropriate relationship \line \up0 \expndtw0\charscalex107 between velocity and vehicular density, as discussed in Section \line \up0 \expndtw-1\ch arscalex100 VI.\par\pard\qj \li20\ri1524\sb5\sl-211\slmult0\fi187 \up0 \expndtw0 \charscalex114 As pointed out earlier, the call density in a region should \line

\up0 \expndtw0\charscalex120 be treated as its ofered load. To illustrate how t he ofered \line \up0 \expndtw0\charscalex111 load results can be used for system engineering and planning \line \up0 \expndtw0\charscalex112 purposes, let the h ighway be served by nonoverlapping cells \line \up0 \expndtw0\charscalex119 of i xed size where each cell covers \up0 \expndtw0\charscalex116 2 km of the highway . \line \up0 \expndtw0\charscalex119 By (2.13), the offered load at a given time is obtained for \line \up0 \expndtw0\charscalex115 each cell. Given the number of channels available at a cell, \par\pard\sect\sectd\sbknone \ql \li3456\sb0\sl -108\slmult0 \up0 \expndtw0\charscalex106 \ul0\nosupersub\cf6\f7\fs12 Calling\ul 0\nosupersub\cf30\f31\fs12\ul & \ul0\nosupersub\cf6\f7\fs12 Handof Rate (Calls/ Min)\par\pard\sect\sectd\sbknone\cols2\colno1\colw5912\colsr160\colno2\colw6028\ colsr160\ql \li1876\sb1\sl-111\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupe rsub\cf1\f2\fs10 0\par\pard\ql \li1411\sb0\sl-138\slmult0 \par\pard\ql \li1411\s b0\sl-138\slmult0 \par\pard\ql \li1411\sb0\sl-138\slmult0 \par\pard\ql \li2251\s b100\sl-138\slmult0\tx2572\tx2894\tx3225\tx3551\tx3854\tx4185\tx4507\tx4833\tx51 59\tx5476 \up0 \expndtw-5\charscalex80 \ul0\nosupersub\cf6\f7\fs12 O\tab \up0 \e xpndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 4\tab \up0 \expndtw-1 \charscalex100 6\tab \up0 \expndtw-1\charscalex100 8\tab \up0 \expndtw0\charscal ex101 10\tab \up0 \expndtw-3\charscalex100 12\tab \up0 \expndtw0\charscalex101 1 4\tab \up0 \expndtw0\charscalex101 16\tab \up0 \expndtw0\charscalex101 18\tab \u p0 \expndtw0\charscalex109 20\par\pard\ql \li1411\sb0\sl-138\slmult0 \par\pard\q l \li3086\sb22\sl-138\slmult0 \up0 \expndtw0\charscalex107 Location on the Highw ay (Km)\par\pard\ql \li1411\sb0\sl-138\slmult0 \par\pard\ql \li1420\sb17\sl-138\ slmult0\tx1929 \up0 \expndtw0\charscalex108 Fig. \up0 \expndtw-1\charscalex100 5 .\tab \up0 \expndtw0\charscalex119 Vehicle density in the highway system with a trafic accident.\par\pard\qj \li1411\sb0\sl-218\slmult0 \par\pard\qj \li1416\ri6 \sb101\sl-218\slmult0\fi4 \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs 16 as indicated in the igure, the call density in this region is also \line \up0 \expndtw0\charscalex108 higher than elsewhere because of the velocity reduction began \line \up0 \expndtw0\charscalex112 at \ul0\nosupersub\cf8\f9\fs16 t \ul0\ nosupersub\cf3\f4\fs16 = 35. Note that the vehicular density in location (6,12] is \line \up0 \expndtw0\charscalex114 lower than that beyond location x > \up0 \ expndtw0\charscalex112 12. This is because the \line \up0 \expndtw0\charscalex11 5 vehicles that would have been at this location if there were \line \up0 \expnd tw0\charscalex113 no reduction in velocity starting at \ul0\nosupersub\cf8\f9\fs 16 t \ul0\nosupersub\cf3\f4\fs16 = 35 have been trapped \line \up0 \expndtw0\cha rscalex109 in location \up0 \expndtw0\charscalex119 (1 7] due to the low velocit y. At \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 = 50 in Fig. \li ne \up0 \expndtw0\charscalex116 4, the vehicular trafic and call density continu e to build up \line \up0 \expndtw0\charscalex109 in location \up0 \expndtw0\char scalex113 (3, 7]. In addition, the "dip" of vehicular density \line \up0 \expndt w0\charscalex113 has shifted to the right from the position shown in Fig. 3, as \line \up0 \expndtw0\charscalex111 vehicles continue to move forward on the high way.\par\pard\qj \li1411\ri0\sb0\sl-216\slmult0\fi182 \up0 \expndtw0\charscalex1 08 Finally, Fig. 5 shows that the whole density curves continue \line \up0 \expn dtw0\charscalex127 to propagate to the right at \ul0\nosupersub\cf8\f9\fs16 t \u p0 \expndtw0\charscalex125 \ul0\nosupersub\cf3\f4\fs16 = 60, as vehicles resume \line \up0 \expndtw0\charscalex117 their original velocity of v(x, \ul0\nosupers ub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex125 \ul0\nosupersub\cf3\f4\fs16 = 1 a t all locations after \line \up0 \expndtw0\charscalex116 the accident has been c leared at \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 = 55. In par ticular, those \line \up0 \expndtw0\charscalex109 vehicles that were located at location (3, 7] at \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 = 5 5 now have \line \up0 \expndtw0\charscalex111 moved into location \up0 \expndtw0 \charscalex114 (8, 12] at \ul0\nosupersub\cf8\f9\fs16 t \up0 \expndtw0\charscale x117 \ul0\nosupersub\cf3\f4\fs16 = 60 at a constant velocity \line \up0 \expndtw 0\charscalex124 of \ul0\nosupersub\cf8\f9\fs16 v(x, t) \up0 \expndtw0\charscalex 137 \ul0\nosupersub\cf3\f4\fs16 = 1 after \ul0\nosupersub\cf8\f9\fs16 t \up0 \ex pndtw0\charscalex122 \ul0\nosupersub\cf3\f4\fs16 > 55. Note that these results m ay\par\pard\column \qj \li20\ri1524\sb0\sl-197\slmult0\fi4 \up0 \expndtw0\charsc

alex110 the blocking probability can be approximated by applying the \up0 \expnd tw0\charscalex113 offered load to the Erlang-B formula. This approximation is \u p0 \expndtw0\charscalex114 naturally supported by the stochastic model, indeed t he full \up0 \expndtw0\charscalex119 highway PALM in Section V; also see Section s \up0 \expndtw0\charscalex121 5 and \up0 \expndtw-2\charscalex100 7 \up0 \expnd tw0\charscalex113 of [7] for further discussion. \up0 \expndtw0\charscalex112 (W ith the highway PALM, the \up0 \expndtw0\charscalex115 offered load in each cell has a Poisson distribution. For the \up0 \expndtw0\charscalex111 inite capacity , a natural approximation is a truncated Poisson \up0 \expndtw0\charscalex111 di stribution, which is the Erlang-B formula.)\par\pard\qj \li20\ri1524\sb4\sl-211\ slmult0\fi172 \up0 \expndtw0\charscalex115 Table I presents the blocking probabi lities at \ul0\nosupersub\cf8\f9\fs16 t \up0 \expndtw0\charscalex124 \ul0\nosup ersub\cf3\f4\fs16 = 30 and \line \up0 \expndtw0\charscalex117 50, for which the call density has been depicted in Figs. \up0 \expndtw-2\charscalex100 2 \line \u p0 \expndtw0\charscalex109 and 4, respectively. As shown in the table, assuming that each \line \up0 \expndtw0\charscalex114 cell has 20 channels, the blocking probabilities at \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 = 30 are \line \up0 \expndtw0\charscalex112 a fraction of a percent, which are satisf actory. However, due \line \up0 \expndtw0\charscalex113 to the trafic congestion caused by the accident, the blocking \line \up0 \expndtw0\charscalex111 probabi lity in the cell at location \up0 \expndtw0\charscalex110 (4, 6] at t \up0 \expn dtw0\charscalex113 = 50 increases to\par\pard\qj \li20\ri1519\sb0\sl-210\slmult0 \fi4 \up0 \expndtw0\charscalex115 37.1%! In fact, it is found that for the surge of ofered load, \up0 \expndtw0\charscalex113 the cell has to be equipped with 4 5 channels to maintain the \up0 \expndtw0\charscalex113 blocking probability sat isfactorily low. Similarly, the cell at \up0 \expndtw0\charscalex104 location \u p0 \expndtw0\charscalex111 (6, 8] also requires ive additional channels to handl e \up0 \expndtw0\charscalex111 the offered load adequately. These results show t he potential \up0 \expndtw0\charscalex113 beneit of dynamic channel assignment. They also show that \up0 \expndtw0\charscalex116 the proposed trafic models can serve as a valuable tool for \up0 \expndtw0\charscalex110 system engineering and planning.\par\pard\qj \li24\ri1510\sb4\sl-209\slmult0\fi172 \up0 \expndtw0\char scalex116 The second set of examples uses the time-homogeneous \line \up0 \expnd tw0\charscalex110 luid model to consider only the space dynamics. The velocity \ line \up0 \expndtw0\charscalex122 ield \ul0\nosupersub\cf8\f9\fs16 v(x) \ul0\nos upersub\cf3\f4\fs16 in Fig. \up0 \expndtw0\charscalex115 1 is used at all time \ ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 and all other system \l ine \up0 \expndtw0\charscalex116 parameters remain unchanged. Then, solving the ODE's in \par\pard\sect\sectd\fs24\paperw12240\paperh15840{\bkmkstart Pg8}{\bkmk end Pg8}\par\pard\li1699\sb0\sl-149\slmult0\par\pard\li1699\sb0\sl-149\slmult0\p ar\pard\li1699\sb0\sl-149\slmult0\par\pard\li1699\sb149\sl-149\slmult0\fi0\tx531 3 \dn2 \expndtw-3\charscalex100 \ul0\nosupersub\cf6\f7\fs12 1360\tab \up0 \expnd tw0\charscalex126 \ul0\nosupersub\cf1\f2\fs10 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\li1699\sb0\sl-138\slmult0\ par\pard\li1699\sb0\sl-138\slmult0\par\pard\li1699\sb0\sl-138\slmult0\par\pard\l i1699\sb0\sl-138\slmult0\par\pard\li1699\sb120\sl-138\slmult0\fi5169 \up0 \expnd tw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 0\t \dn2 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 n\par\pard\li1699\sb116\sl-138\slmult0\fi456\tx6868 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 0\tab \up0 \expndtw-1 \charscalex100 \ul0\nosupersub\cf15\f16\fs8 0\par\pard\li1699\sb0\sl-72\slmult0\ fi5169 \up0 \expndtw-1\charscalex100 CO\par\pard\li1699\sb55\sl-138\slmult0\fi46 0\tx6868 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 0\t \dn2 \exp ndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 -\tab \up0 \expndtw-1\charscale x100 0\par\pard\li1699\sb0\sl-72\slmult0\fi5169 \up0 \expndtw-1\charscalex100 \u l0\nosupersub\cf15\f16\fs8 N-\par\pard\li1699\sb74\sl-115\slmult0\fi460\tx6868 \ up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 0\tab \dn2 \expndtw-1\c harscalex100 0\par\pard\li1699\sb0\sl-115\slmult0\par\pard\li1699\sb49\sl-115\sl mult0\fi5169 \up2 \expndtw-1\charscalex100 0\t \dn2 \expndtw-1\charscalex100 In\ par\pard\li1699\sb0\sl-115\slmult0\par\pard\li1699\sb38\sl-115\slmult0\fi5169 \u p0 \expndtw-1\charscalex100 0\par\pard\li1699\sb1\sl-145\slmult0\fi1368\tx7780\t

x8712 \dn2 \expndtw0\charscalex107 \ul0\nosupersub\cf6\f7\fs12 Total\tab \dn2 \e xpndtw0\charscalex105 Total\tab \up2 \expndtw0\charscalex103 Calling\par\pard\li 1699\sb141\sl-230\slmult0\fi1862\tx8054\tx8990 \up0 \expndtw0\charscalex100 \ul0 \nosupersub\cf31\f32\fs4 \u8226?\tab \dn4 \expndtw0\charscalex41 \ul0\nosupersub \cf22\f23\fs20 ........ \u8226?\tab \up0 \expndtw0\charscalex136 \ul0\nosupersub \cf6\f7\fs12 - -A\par\pard\li1699\sb0\sl-144\slmult0\fi456\tx6859 \up0 \expndtw2\charscalex100 \ul0\nosupersub\cf3\f4\fs16 0\tab \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 0\par\pard\li1699\sb0\sl-108\slmult0\fi2304 \up0 \ex pndtw0\charscalex103 \ul0\nosupersub\cf6\f7\fs12 Calling\par\pard\sect\sectd\sbk none\cols2\colno1\colw6240\colsr110\colno2\colw5750\colsr160\ql \li4852\sb0\sl-1 08\slmult0 \up0 \expndtw0\charscalex104 Non-Calling\par\pard\ql \li2155\sb0\sl-1 44\slmult0\tx3700\tx4463 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf3\f4\fs 16 0\tab \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf13\f14\fs8 \u8226? \tab \up0 \expndtw-5\charscalex83 \ul0\nosupersub\cf6\f7\fs12 -N.\par\pard\ql \li215 5\sb0\sl-108\slmult0\tx3508 \up0 \expndtw0\charscalex204 o -\tab \up0 \expndtw0\ charscalex110 Handof Rate (Calls/Min)\par\pard\ql \li3859\sb37\sl-69\slmult0\tx4 507\tx4828\tx5155\tx5486\tx5812 \up0 \expndtw0\charscalex100 \ul0\nosupersub\cf2 8\f29\fs6 1\tab \up0 \expndtw0\charscalex100 1\tab \up0 \expndtw0\charscalex100 1\tab \up0 \expndtw0\charscalex100 1\tab \up0 \expndtw0\charscalex100 1\tab \up0 \expndtw0\charscalex100 1\par\pard\ql \li2524\sb80\sl-138\slmult0\tx2855\tx3167 \tx3494\tx3830\tx4156\tx4478\tx4804\tx5130\tx5462\tx5759 \up0 \expndtw-1\charsca lex100 \ul0\nosupersub\cf6\f7\fs12 0\tab \up0 \expndtw-1\charscalex100 1\tab \up 0 \expndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 3\tab \up0 \expnd tw-1\charscalex100 4\tab \up0 \expndtw-1\charscalex100 5\tab \up0 \expndtw-1\cha rscalex100 6\tab \up0 \expndtw-1\charscalex100 7\tab \up0 \expndtw-1\charscalex1 00 8\tab \up0 \expndtw-1\charscalex100 9\tab \up0 \expndtw0\charscalex101 10\par \pard\ql \li1699\sb0\sl-138\slmult0 \par\pard\ql \li3360\sb31\sl-138\slmult0 \up 0 \expndtw0\charscalex112 \ul0\nosupersub\cf1\f2\fs10 Location \ul0\nosupersub\c f6\f7\fs12 on the Highway (Km)\par\pard\ql \li1699\sb0\sl-138\slmult0 \par\pard\ ql \li1703\sb70\sl-138\slmult0\tx2217 \up0 \expndtw0\charscalex106 Fig. \up0 \ex pndtw-1\charscalex100 6.\tab \up0 \expndtw0\charscalex118 Vehicle density in tem poral steady state (ixed call-origination rate).\par\pard\qj \li1699\sb0\sl-212\ slmult0 \par\pard\qj \li1699\ri0\sb122\sl-212\slmult0\fi4 \up0 \expndtw0\charsca lex112 \ul0\nosupersub\cf3\f4\fs16 (2.20) and (2.21) yields \ul0\nosupersub\cf8\ f9\fs16 n(x) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x) \u l0\nosupersub\cf3\f4\fs16 and they are portrayed \line \up0 \expndtw0\charscalex 112 in Fig. 6. It is interesting to note that this igure is practically \line \u p0 \expndtw0\charscalex111 identical to Fig. \up0 \expndtw0\charscalex114 4 in l ocation \up0 \expndtw0\charscalex113 (0, 10]. This is so because, in \line \up0 \expndtw0\charscalex115 the time-dependent case, the system has almost reached i ts \line \up0 \expndtw0\charscalex116 steady state at \ul0\nosupersub\cf8\f9\fs1 6 t \ul0\nosupersub\cf3\f4\fs16 = 50 with the accident started at \ul0\nosupersu b\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 = 35. Of \line \up0 \expndtw0\charsc alex109 course, this time-homogeneous example does not have the dip \line \up0 \ expndtw0\charscalex110 of vehicular densities after location \up0 \expndtw0\char scalex112 12 in Fig. 4.\par\pard\qj \li1703\ri0\sb4\sl-212\slmult0\fi177 \up0 \e xpndtw0\charscalex107 Fig. 7 considers a case that uses all system parameters of the \line \up0 \expndtw0\charscalex109 case shown in Fig. 6, except that the ca ll-origination rate a \ul0\nosupersub\cf8\f9\fs16 (x) \line \up0 \expndtw0\chars calex116 \ul0\nosupersub\cf3\f4\fs16 is inversely proportional to \ul0\nosupersu b\cf8\f9\fs16 v(z). \ul0\nosupersub\cf3\f4\fs16 In particular, we stipulate \lin e \up0 \expndtw0\charscalex114 that \ul0\nosupersub\cf8\f9\fs16 A(x)v(x) = \ul0\ nosupersub\cf3\f4\fs16 0.1 for all \ul0\nosupersub\cf8\f9\fs16 x > \ul0\nosupers ub\cf3\f4\fs16 0. This setting can be applied \line \up0 \expndtw0\charscalex108 to study the phenomenon that people are likely to make phone \line \up0 \expndt w0\charscalex109 calls when the vehicle velocity is reduced, e.g., when they are \line \up0 \expndtw0\charscalex109 trapped in a trafic jam. Because of an incre ased value of A(x) \line \up0 \expndtw0\charscalex109 at location \up0 \expndtw0 \charscalex118 3 < \ul0\nosupersub\cf8\f9\fs16 x < 7, \ul0\nosupersub\cf3\f4\fs1 6 the call density in this region in Fig.\par\pard\qj \li1708\ri0\sb4\sl-212\slm

ult0 \up0 \expndtw0\charscalex112 7 can be more than two times of that shown in Fig. 6, where \line \up0 \expndtw0\charscalex112 A(x) is identical at all locati ons. These results clearly reveal \line \up0 \expndtw0\charscalex113 the importa nce of space dynamics (i.e., vehicle velocity and \line \up0 \expndtw0\charscale x107 density) and the customers' calling behavior in determining the \line \up0 \expndtw0\charscalex112 ofered load and thus the performance of wireless systems .\par\pard\qj \li1708\ri0\sb0\sl-212\slmult0\fi182 \up0 \expndtw0\charscalex112 Finally, we consider a location-dependent call-origination \line \up0 \expndtw0\ charscalex110 rate that decays exponentially in location. Speciically, \ul0\nosu persub\cf8\f9\fs16 A(x) =\par\pard\qj \li1708\ri0\sb4\sl-211\slmult0\fi14 \up0 \ expndtw0\charscalex118 1 + \ul0\nosupersub\cf3\f4\fs16 0.9e-42. Such a form of c all-origination rate function \line \up0 \expndtw0\charscalex113 can be used to approximate the calling activity of spectators \line \up0 \expndtw0\charscalex12 2 departing from a stadium after a sporting event. In such \line \up0 \expndtw0\ charscalex110 situations, vehicles are likely to initiate calls when they begin \line \up0 \expndtw0\charscalex111 to leave the stadium (i.e., close to location 0). Fig. 8 assumes \line \up0 \expndtw0\charscalex114 v(x) \up0 \expndtw0\chars calex127 = 0.77 for all \ul0\nosupersub\cf8\f9\fs16 x \up0 \expndtw0\charscalex1 16 > \ul0\nosupersub\cf3\f4\fs16 0, which is the velocity averaged \line \up0 \e xpndtw0\charscalex111 over the location \up0 \expndtw0\charscalex115 (0, 10] for the velocity ield shown in Fig. \line \up0 \expndtw-7\charscalex83 1. \up0 \exp ndtw0\charscalex116 Although the settings for Figs. \up0 \expndtw0\charscalex123 7 and 8 have the same \line \up0 \expndtw0\charscalex108 average velocity, thei r call densities and handoff rates are very \line \up0 \expndtw0\charscalex113 d iferent. Hence, the space dynamics and calling patterns are \line \up0 \expndtw0 \charscalex117 shown again to have signiicant impacts on the trafic loads \line \up0 \expndtw0\charscalex112 even for systems in temporal steady state.\par\pard \qj \li1713\ri0\sb9\sl-207\slmult0\fi182 \up0 \expndtw0\charscalex129 We have al so considered other examples for time-\line \up0 \expndtw0\charscalex108 homogen eous cases where the highway has multiple entrances \line \up0 \expndtw0\charsca lex109 and exits, although they are not given here. In these cases, the \line \u p0 \expndtw0\charscalex113 ODE's are solved for segments of the highway between two\par\pard\column \ql \li529\sb0\sl-117\slmult0\tx3006\tx3476 \up0 \expndtw0\c harscalex50 \ul0\nosupersub\cf1\f2\fs10 CS1\tab \up0 \expndtw-2\charscalex100 \u l0\nosupersub\cf22\f23\fs20 ,\tab \up0 \expndtw0\charscalex105 \ul0\nosupersub\c f6\f7\fs12 Non-Calling\par\pard\ql \li529\sb0\sl-90\slmult0 \up0 \expndtw-1\char scalex100 \ul0\nosupersub\cf1\f2\fs10 0\par\pard\ql \li1777\sb0\sl-108\slmult0 \ up0 \expndtw0\charscalex112 \ul0\nosupersub\cf6\f7\fs12 \u8212?'''''T-landof Rat e (Calls/Min)\par\pard\ql \li529\sb0\sl-90\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 0\par\pard\ql \li6350\sb0\sl-138\slmult0 \par\pard\ ql \li898\sb14\sl-138\slmult0\tx1239\tx1556\tx1887\tx2214\tx2540\tx2871\tx3193\t x3524\tx3850\tx4153 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 0\ tab \up0 \expndtw-1\charscalex100 1\tab \up0 \expndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 3\tab \up0 \expndtw-1\charscalex100 4\tab \up0 \expndt w-1\charscalex100 5\tab \up0 \expndtw-1\charscalex100 6\tab \up0 \expndtw-1\char scalex100 7\tab \up0 \expndtw-1\charscalex100 8\tab \up0 \expndtw-1\charscalex10 0 9\tab \up0 \expndtw0\charscalex101 10\par\pard\ql \li6350\sb0\sl-138\slmult0 \ par\pard\ql \li1738\sb36\sl-138\slmult0 \up0 \expndtw0\charscalex106 Location on the Highway (Km)\par\pard\qj \li6350\sb0\sl-158\slmult0 \par\pard\qj \li20\ri12 70\sb14\sl-158\slmult0\tx567 \up0 \expndtw0\charscalex111 Fig. \up0 \expndtw0\c harscalex104 7.\tab \up0 \expndtw0\charscalex124 Vehicle density in temporal ste ady state \up0 \expndtw0\charscalex123 (call-origination rate x \line \up0 \exp ndtw0\charscalex112 velocity \up0 \expndtw0\charscalex118 = constant).\par\pard\ ql \li6849\sb0\sl-184\slmult0 \par\pard\ql \li6849\sb0\sl-184\slmult0 \par\pard\ ql \li519\sb24\sl-184\slmult0 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf3\ f4\fs16 )\par\pard\ql \li6849\sb0\sl-138\slmult0 \par\pard\ql \li6849\sb0\sl-138 \slmult0 \par\pard\ql \li2415\sb29\sl-138\slmult0 \up0 \expndtw0\charscalex107 \ ul0\nosupersub\cf6\f7\fs12 Total\par\pard\ql \li519\sb184\sl-184\slmult0 \up0 \e xpndtw-2\charscalex100 \ul0\nosupersub\cf3\f4\fs16 2\par\pard\ql \li519\sb85\sl184\slmult0\tx2238\tldot\tx3352 \up0 \expndtw-2\charscalex100 8\tab \up0 \expndt

w0\charscalex104 \ul0\nosupersub\cf6\f7\fs12 Non-Calling\expndtw0\charscalex100\ tab\par\pard\ql \li519\sb85\sl-184\slmult0 \up0 \expndtw0\charscalex51 \ul0\nosu persub\cf3\f4\fs16 Cs\par\pard\ql \li519\sb65\sl-115\slmult0 \up0 \expndtw-1\cha rscalex100 \ul0\nosupersub\cf1\f2\fs10 0\par\pard\ql \li519\sb134\sl-138\slmult0 \tx1110\tx2372 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf6\f7\fs12 )\tab \ up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf1\f2\fs10 /\tab \up0 \expndtw0\ch arscalex102 \ul0\nosupersub\cf6\f7\fs12 Calling\par\pard\ql \li1023\sb45\sl-138\ slmult0\tx1134 \up0 \expndtw-1\charscalex100 /\tab \up0 \expndtw-1\charscalex100 (\par\pard\ql \li519\sb0\sl-108\slmult0\tx1009 \up0 \expndtw-1\charscalex100 \u l0\nosupersub\cf1\f2\fs10 0\tab \up0 \expndtw-5\charscalex100 \ul0\nosupersub\cf 8\f9\fs16 1/\par\pard\ql \li3245\sb0\sl-108\slmult0 \up0 \expndtw-1\charscalex10 0 \ul0\nosupersub\cf6\f7\fs12 -\par\pard\ql \li2525\sb112\sl-138\slmult0 \up0 \e xpndtw0\charscalex110 Handof Rate (Calls/Min)\par\pard\ql \li519\sb92\sl-92\slmu lt0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf13\f14\fs8 0\par\pard\ql \li 7214\sb0\sl-138\slmult0 \par\pard\ql \li884\sb83\sl-138\slmult0\tx1225\tx1542\tx 1873\tx2195\tx2530\tx2852\tx3174\tx3500\tx3831\tx4129 \up0 \expndtw-1\charscalex 100 \ul0\nosupersub\cf6\f7\fs12 0\tab \up0 \expndtw-1\charscalex100 1\tab \up0 \ expndtw-1\charscalex100 2\tab \up0 \expndtw-1\charscalex100 3\tab \up0 \expndtw1\charscalex100 4\tab \up0 \expndtw-1\charscalex100 5\tab \up0 \expndtw-1\charsc alex100 6\tab \up0 \expndtw-1\charscalex100 7\tab \up0 \expndtw-1\charscalex100 8\tab \up0 \expndtw-1\charscalex100 9\tab \up0 \expndtw0\charscalex105 10\par\pa rd\ql \li8054\sb0\sl-138\slmult0 \par\pard\ql \li1724\sb31\sl-138\slmult0 \up0 \ expndtw0\charscalex106 Location on the Highway (Km)\par\pard\ql \li30\sb131\sl-1 38\slmult0 \up0 \expndtw0\charscalex119 Fig. 8. Vehicle density in temporal stea dy state (exponential call-origination\par\pard\ql \li30\sb44\sl-138\slmult0 \up 0 \expndtw0\charscalex112 rate).\par\pard\ql \li6360\sb0\sl-184\slmult0 \par\par d\ql \li34\sb77\sl-184\slmult0 \up0 \expndtw0\charscalex109 \ul0\nosupersub\cf3\ f4\fs16 successive entrances/exits. Based on the vehicular densities at\par\pard \ql \li30\sb32\sl-184\slmult0 \up0 \expndtw0\charscalex110 the end of one segmen t (i.e., just before an entrance/exit), the\par\pard\ql \li30\sb22\sl-184\slmult 0 \up0 \expndtw0\charscalex112 probability of a vehicle leaving from the exit, a nd the low of\par\pard\ql \li30\sb32\sl-184\slmult0 \up0 \expndtw0\charscalex112 vehicles entering from the entrance, we can obtain the initial\par\pard\ql \li3 0\sb23\sl-184\slmult0 \up0 \expndtw0\charscalex118 conditions for the next segme nt. Then, solving the ODE's\par\pard\ql \li34\sb32\sl-184\slmult0 \up0 \expndtw0 \charscalex113 with these initial conditions yields the vehicular densities in\p ar\pard\ql \li34\sb27\sl-184\slmult0 \up0 \expndtw0\charscalex113 the next segme nt of the highway.\par\pard\ql \li6369\sb0\sl-184\slmult0 \par\pard\ql \li826\sb 59\sl-184\slmult0 \up0 \expndtw0\charscalex113 IV. THE \ul0\nosupersub\cf6\f7\fs 12 STOCHASTIC TRAFFIC \ul0\nosupersub\cf3\f4\fs16 MODEL\par\pard\ql \li217\sb99\ sl-184\slmult0 \up0 \expndtw0\charscalex122 In contrast to the deterministic lui d model introduced\par\pard\ql \li39\sb27\sl-184\slmult0 \up0 \expndtw0\charscal ex121 above, the stochastic traic model captures the stochastic\par\pard\ql \li3 9\sb23\sl-184\slmult0 \up0 \expndtw0\charscalex120 luctuations. Speciically, the model considers the random\par\pard\ql \li39\sb22\sl-184\slmult0 \up0 \expndtw0 \charscalex111 calling status of each individual vehicle as it moves along on\pa r\pard\ql \li39\sb27\sl-184\slmult0 \up0 \expndtw0\charscalex112 the highway. Ho wever, it tuns out that the PDE's and ODE's\par\pard\ql \li39\sb28\sl-184\slmult 0 \up0 \expndtw0\charscalex110 which govern the expected values are identical to those of the\par\pard\ql \li44\sb27\sl-184\slmult0 \up0 \expndtw0\charscalex110 deterministic luid model. Hence, the numerical examples that\par\pard\ql \li44\ sb22\sl-184\slmult0 \up0 \expndtw0\charscalex113 we have just considered apply e qually well to the stochastic \par\pard\sect\sectd\fs24\paperw12240\paperh15840{ \bkmkstart Pg9}{\bkmkend Pg9}\par\pard\sect\sectd\sbknone\cols2\colno1\colw6100\ colsr20\colno2\colw5980\colsr160\ql \li1454\sb0\sl-138\slmult0 \par\pard\ql \li1 454\sb0\sl-138\slmult0 \par\pard\ql \li1454\sb0\sl-138\slmult0 \par\pard\ql \li1 454\sb0\sl-138\slmult0 \par\pard\ql \li1454\sb91\sl-138\slmult0 \up0 \expndtw0\c harscalex119 \ul0\nosupersub\cf1\f2\fs10 LEUNG \ul0\nosupersub\cf27\f28\fs12 et al.: \ul0\nosupersub\cf1\f2\fs10 TRAFFIC MODELS FOR WIRELESS COMMUNICATION NETWO RKS\par\pard\qj \li1449\sb0\sl-211\slmult0 \par\pard\qj \li1449\ri7\sb100\sl-211

\slmult0 \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 model. In Figs . 2-8, we must simply replace the actual values \up0 \expndtw0\charscalex113 on the y-axis by expected values.\par\pard\qj \li1449\ri7\sb5\sl-211\slmult0\fi177\ tx5487 \up0 \expndtw0\charscalex110 The stochastic model has the same highway se tting. Unless \line \up0 \expndtw0\charscalex108 stated otherwise, the same nota tion is used as for the determin-\line \up0 \expndtw0\charscalex115 istic luid m odel. Using the same deinitions, \ul0\nosupersub\cf8\f9\fs16 N(x, t), Q(x, t), \ line \up0 \expndtw0\charscalex116 C+ (x, t), C \up0 \expndtw0\charscalex114 (x, t), C: \up0 \expndtw0\charscalex109 24- (x, t), CQ (x, t), En (x, t), \tab \up0 \expndtw0\charscalex113 (x, t), \line \up0 \expndtw0\charscalex117 E+(x, \ul0\no supersub\cf3\f4\fs16 t), and \ul0\nosupersub\cf8\f9\fs16 E+(x,t) \ul0\nosupersub \cf3\f4\fs16 become integer-valued random vari-\line \up0 \expndtw0\charscalex11 4 ables in the stochastic model. Now, the densities \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \line \up0 \expndtw0\charscalex109 \ul0\ nosupersub\cf8\f9\fs16 q(x, t) \ul0\nosupersub\cf3\f4\fs16 are deined as the par tial derivatives of expected values; \line \up0 \expndtw0\charscalex103 i.e.,\pa r\pard\ql \li2587\sb86\sl-184\slmult0\tx4578 \up0 \expndtw0\charscalex119 \ul0\n osupersub\cf14\f15\fs16\ul OE[N(x,t)]\ul0\tab \up0 \expndtw0\charscalex103 \ul0\ nosupersub\cf14\f15\fs16\ul a E[Q (x,\par\pard\column \ql \li10358\sb0\sl-115\sl mult0 \par\pard\ql \li10358\sb0\sl-115\slmult0 \par\pard\ql \li10358\sb0\sl-115\ slmult0 \par\pard\ql \li10358\sb0\sl-115\slmult0 \par\pard\ql \li10358\sb0\sl-11 5\slmult0 \par\pard\ql \li4258\sb101\sl-115\slmult0 \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf1\f2\fs10 1361\par\pard\qj \li1444\sb0\sl-206\slmult0 \par\pa rd\qj \li202\ri1457\sb118\sl-206\slmult0\tx466 \up0 \expndtw0\charscalex113 \ul0 \nosupersub\cf3\f4\fs16 4) There are no capacity constraints. That is, each cell has \line\tab \up0 \expndtw0\charscalex109 an ininite number of channels such t hat no call blocking\par\pard\ql \li452\sb15\sl-184\slmult0 \up0 \expndtw0\chars calex105 occurs.\par\pard\qj \li20\ri1461\sb61\sl-207\slmult0\fi177 \up0 \expndt w0\charscalex118 As in [6] and [7], we can construct \ul0\nosupersub\cf8\f9\fs16 Q(x, t) \ul0\nosupersub\cf3\f4\fs16 by stochastic \up0 \expndtw0\charscalex114 integration as follows. For \ul0\nosupersub\cf8\f9\fs16 j \ul0\nosupersub\cf3\f4 \fs16 > 1, let \ul0\nosupersub\cf8\f9\fs16 Ts+ (j) \ul0\nosupersub\cf3\f4\fs16 a nd \ul0\nosupersub\cf8\f9\fs16 T; (j) \ul0\nosupersub\cf3\f4\fs16 be the \up0 \e xpndtw0\charscalex110 time when a vehicle arriving to the highway at time \ul0\n osupersub\cf8\f9\fs16 s \ul0\nosupersub\cf3\f4\fs16 initiates \up0 \expndtw0\cha rscalex113 and terminates its jth call, respectively. Then, we have\par\pard\ql \li1444\sb0\sl-184\slmult0 \par\pard\ql \li1444\sb0\sl-184\slmult0 \par\pard\ql \li1444\sb0\sl-184\slmult0 \par\pard\ql \li20\sb134\sl-184\slmult0 \up0 \expndtw 0\charscalex103 and\par\pard\sect\sectd\sbknone\cols2\colno1\colw4755\colsr160\c olno2\colw7185\colsr160\ql \li1843\sb0\sl-128\slmult0\tx2918\tx3542 \up0 \expndt w0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 n(x, t) =\tab \up0 \expndtw0\charsc alex124 ax\tab \up0 \expndtw0\charscalex114 and q(x, t) =\par\pard\column \ql \l i20\sb103\sl-128\slmult0\tx2866 \up0 \expndtw0\charscalex124 ax\tab \up0 \expndt w-2\charscalex100 \ul0\nosupersub\cf3\f4\fs16 t\par\pard\sect\sectd\sbknone\cols 2\colno1\colw5955\colsr160\colno2\colw5985\colsr160\qj \li1449\ri12\sb70\sl-213\ slmult0 \up0 \expndtw0\charscalex116 respectively, where E[Y] denotes the expect ed value of Y. \line \up0 \expndtw0\charscalex108 Correspondingly, we let the ra te densities be the second partial \line \up0 \expndtw0\charscalex111 derivative s of expected \up0 \expndtw0\charscalex109 values; i.e.,\par\pard\ql \li3115\sb5 5\sl-184\slmult0 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf8\f9\fs16 /\par \pard\column \ql \li730\sb99\sl-184\slmult0\tx1887 \up0 \expndtw0\charscalex129 Q (x \u8226? t) =\ul0\super\cf1\f2\fs15 f\tab \up0 \expndtw0\charscalex114 \ul0\ nosupersub\cf8\f9\fs16 1\{L.(t)E(-00,x1x undA(s)\par\pard\ql \li1585\sb21\sl-128 \slmult0 \up0 \expndtw-7\charscalex86 \u8212?cc\par\pard\ql \li1444\sb0\sl-184\s lmult0 \par\pard\ql \li20\sb46\sl-184\slmult0 \up0 \expndtw0\charscalex121 \ul0\ nosupersub\cf3\f4\fs16 where \ul0\nosupersub\cf8\f9\fs16 A(t) \ul0\nosupersub\cf 3\f4\fs16 counts the number of vehicles arriving to the\par\pard\sect\sectd\sbkn one\cols3\colno1\colw2412\colsr40\colno2\colw988\colsr160\colno3\colw8510\colsr1 60\ql \li1627\sb0\sl-216\slmult0 \up0 \expndtw-5\charscalex100 \ul0\nosupersub\c f12\f13\fs24 ct(x,t) \up0 \expndtw-11\charscalex80 =\par\pard\ql \li1444\sb0\sl-

276\slmult0 \par\pard\ql \li1627\sb195\sl-276\slmult0 \up0 \expndtw-11\charscale x76 c,1-(x,t)\par\pard\column \ql \li20\sb0\sl-216\slmult0 \up0 \expndtw0\charsc alex100 a2 \ul0\nosupersub\cf8\f9\fs16 E[C: \up0 \expndtw-8\charscalex93 \ul0\n osupersub\cf14\f15\fs16\ul x \ul0\nosupersub\cf8\f9\fs16 ,\par\pard\ql \li385\sb 0\sl-216\slmult0 \up0 \expndtw-11\charscalex91 \ul0\nosupersub\cf12\f13\fs24 axa t\par\pard\ql \li25\sb120\sl-276\slmult0 \up0 \expndtw-11\charscalex80 \ul0\nosu persub\cf11\f12\fs24\ul a2\ul0\nosupersub\cf12\f13\fs24 \ul0\nosupersub\cf11\f1 2\fs24\ul E[c\up0 \expndtw0\charscalex63 H,-\ul0\nosupersub\cf12\f13\fs24 \ul0\ nosupersub\cf11\f12\fs24\ul (x,\par\pard\column \ql \li980\sb0\sl-216\slmult0\tx 1652\tx2550 \up0 \expndtw-8\charscalex75 a2\ul0\nosupersub\cf12\f13\fs24 \ul0\n osupersub\cf11\f12\fs24\ul E\ul0\tab \up0 \expndtw0\charscalex114 \ul0\nosupersu b\cf14\f15\fs16\ul (x,\ul0\tab \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\ f4\fs16 highway up to time \ul0\nosupersub\cf8\f9\fs16 t, \up0 \expndtw0\charsc alex115 1B \ul0\nosupersub\cf3\f4\fs16 is an indicator function such that\par\pa rd\ql \li20\sb0\sl-192\slmult0 \up0 \expndtw-3\charscalex100 \ul0\nosupersub\cf1 2\f13\fs24 , \up0 \expndtw-11\charscalex77 c,T.(x,t) \up0 \expndtw-11\charscalex 80 =\par\pard\ql \li1354\sb0\sl-160\slmult0\tx2559 \up0 \expndtw-9\charscalex90 axat\tab \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs16 1B = 1 if \ul0 \nosupersub\cf8\f9\fs16 B \ul0\nosupersub\cf3\f4\fs16 is true and 0 otherwise, a nd the location process\par\pard\ql \li2545\sb0\sl-144\slmult0 \up0 \expndtw0\ch arscalex115 \ul0\nosupersub\cf8\f9\fs16 L3(t) \ul0\nosupersub\cf3\f4\fs16 specif ies the position and calling status of the vehicle\par\pard\ql \li2540\sb0\sl-14 4\slmult0 \up0 \expndtw0\charscalex118 that arrived at time \ul0\nosupersub\cf8\ f9\fs16 s. \ul0\nosupersub\cf3\f4\fs16 That is, \ul0\nosupersub\cf8\f9\fs16 L3(t ) = (x, k) \ul0\nosupersub\cf3\f4\fs16 where \ul0\nosupersub\cf8\f9\fs16 x \ul0\ nosupersub\cf3\f4\fs16 is the\par\pard\ql \li778\sb0\sl-192\slmult0 \up0 \expndt w-11\charscalex84 \ul0\nosupersub\cf12\f13\fs24 _ \ul0\nosupersub\cf11\f12\fs24\ ul a2\ul0\nosupersub\cf12\f13\fs24 \ul0\nosupersub\cf11\f12\fs24\ul E [c7\ul0\n osupersub\cf12\f13\fs24 \ul0\nosupersub\cf11\f12\fs24\ul (x,\par\pard\ql \li254 0\sb0\sl-134\slmult0 \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf3\f4\fs16 po sition on the highway and\par\pard\sect\sectd\sbknone\cols4\colno1\colw3646\cols r160\colno2\colw3095\colsr160\colno3\colw451\colsr20\colno4\colw4588\colsr160\ql \li2251\sb0\sl-160\slmult0 \up0 \expndtw0\charscalex167 \ul0\super\cf8\f9\fs24 - \ul0\super\cf14\f15\fs24\ul \ul0\nosupersub\cf12\f13\fs24 axat\par\pard\ql \l i1617\sb97\sl-276\slmult0 \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9\fs 16 e:(x,t) = \ul0\super\cf14\f15\fs24\ul a2 E\ul0\super\cf12\f13\fs36 axat\ul0\s uper\cf14\f15\fs24\ul x,\par\pard\column \ql \li135\sb0\sl-216\slmult0\tx1138\tx 1906 \up0 \expndtw-11\charscalex95 \ul0\nosupersub\cf12\f13\fs24 (x t)\tab \up0 \expndtw-11\charscalex92 axat\tab \up0 \expndtw-3\charscalex100 '\par\pard\ql \l i20\sb16\sl-207\slmult0\tx380 \up0 \expndtw0\charscalex80 \ul0\nosupersub\cf8\f9 \fs16 \u8212? \tab \up0 \expndtw-1\charscalex100 \u8222? _ \ul0\nosupersub\cf11\ f12\fs24\ul a2E[E;(x,t)]\par\pard\ql \li1138\sb0\sl-201\slmult0 \up0 \expndtw-11 \charscalex92 \ul0\nosupersub\cf12\f13\fs24 axat\par\pard\column \ql \li20\sb1\s l-515\slmult0 \up0 \expndtw0\charscalex40 \ul0\nosupersub\cf32\f33\fs56 k = \ul0 \nosupersub\cf33\f34\fs56 1\par\pard\column \ql \li34\sb5\sl-184\slmult0\tx1805 \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf3\f4\fs16 1 if \ul0\nosupersub\cf 8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 E Ur' \up0 \expndtw-4\charscalex100 \ul 0\nosupersub\cf8\f9\fs16 [Ts+ (j),\tab \up0 \expndtw0\charscalex126 (j))\par\par d\ql \li20\sb27\sl-184\slmult0\tx2741 \up0 \expndtw0\charscalex109 \ul0\nosupers ub\cf3\f4\fs16 0 otherwise.\tab \up0 \expndtw0\charscalex102 (4.3) \par\pard\sec t\sectd\sbknone \li1444\sb12\sl-184\slmult0\fi4843 \up0 \expndtw0\charscalex113 In this context, an analog of Lemma 2.1 holds, which is a\par\pard\sect\sectd\sb knone\cols4\colno1\colw2292\colsr40\colno2\colw1036\colsr160\colno3\colw2442\col sr160\colno4\colw5990\colsr160\ql \li1502\sb177\sl-276\slmult0 \up0 \expndtw-10\ charscalex85 \ul0\nosupersub\cf12\f13\fs24 e+ \ul0\nosupersub\cf8\f9\fs16 (x, \u l0\nosupersub\cf12\f13\fs24 t) \up0 \expndtw-10\charscalex80 =\par\pard\column \ ql \li20\sb38\sl-276\slmult0 \up0 \expndtw-11\charscalex77 \ul0\nosupersub\cf11\ f12\fs24\ul a2\ul0\nosupersub\cf12\f13\fs24 \ul0\nosupersub\cf11\f12\fs24\ul E[ E+\ul0\nosupersub\cf12\f13\fs24 \ul0\nosupersub\cf11\f12\fs24\ul (x,\par\pard\q l \li390\sb64\sl-184\slmult0 \up0 \expndtw0\charscalex128 \ul0\nosupersub\cf8\f9

\fs16 axat\par\pard\column \ql \li1172\sb43\sl-276\slmult0 \up0 \expndtw-11\char scalex74 \ul0\nosupersub\cf11\f12\fs24\ul a2\ul0\nosupersub\cf12\f13\fs24 \ul0\ nosupersub\cf11\f12\fs24\ul E[.E \up0 \expndtw-11\charscalex78 (x,\ul0\nosupers ub\cf12\f13\fs24 \ul0\nosupersub\cf11\f12\fs24\ul t)]\par\pard\ql \li20\sb0\sl139\slmult0\tx548 \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 and\t ab \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf8\f9\fs16 (x, t) \up0 \expndtw 0\charscalex108 =\par\pard\ql \li1542\sb0\sl-128\slmult0 \up0 \expndtw0\charscal ex129 axat\par\pard\column \ql \li20\ri1490\sb11\sl-211\slmult0\tx198 \up0 \expn dtw0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 natural extension of the conserva tion equation (2.7) of [7]. \line\tab \up0 \expndtw0\charscalex118 \ul0\nosupers ub\cf8\f9\fs16 Lemma 4.1: \ul0\nosupersub\cf3\f4\fs16 In the stochastic trafic m odel, the densities \line \up0 \expndtw0\charscalex113 of noncalling and calling vehicles, \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ ul0\nosupersub\cf8\f9\fs16 q(x, t), \ul0\nosupersub\cf3\f4\fs16 satisfy\par\pard \sect\sectd\sbknone\cols2\colno1\colw5985\colsr110\colno2\colw6005\colsr160\qj \ li1444\ri0\sb0\sl-200\slmult0 \up0 \expndtw0\charscalex116 For the stochastic mo del, we inroduce additional notation. \up0 \expndtw0\charscalex119 Let \ul0\nosu persub\cf8\f9\fs16 a(t) \ul0\nosupersub\cf3\f4\fs16 be the total arrival rate of vehicles arriving to the \up0 \expndtw0\charscalex117 highway at time \ul0\nosu persub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 By deinition,\par\pard\ql \li1 444\sb0\sl-184\slmult0 \par\pard\ql \li1891\sb25\sl-184\slmult0\tx2447\tx3787\tx 5553 \up0 \expndtw0\charscalex124 \ul0\nosupersub\cf8\f9\fs16 a(t)\tab \up0 \exp ndtw0\charscalex121 1\{E [E17(oc,\tab \up0 \expndtw0\charscalex108 + E[Eq4 (cc, 0] \ul0\nosupersub\cf3\f4\fs16 1.\tab \up0 \expndtw0\charscalex103 (4.1)\par\par d\qj \li1444\ri1\sb182\sl-206\slmult0\fi177 \up0 \expndtw0\charscalex120 We make the following assumptions for the stochastic \line \up0 \expndtw0\charscalex104 model:\par\pard\qj \li1641\ri0\sb39\sl-211\slmult0\tx1871 \up0 \expndtw0\charsc alex117 1) Vehicles arrive to the highway according to a pair of \line\tab \up0 \expndtw0\charscalex108 two-dimensional stochastic jump processes \ul0\nosupersu b\cf8\f9\fs16 E+ (x, t) \ul0\nosupersub\cf3\f4\fs16 and\par\pard\qj \li1871\ri6\ sb6\sl-211\slmult0\fi153 \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf8\f9\fs1 6 + (x ,t) \ul0\nosupersub\cf3\f4\fs16 with nondecreasing sample paths having on ly \line \up0 \expndtw0\charscalex108 unit jumps and deterministic intensity fun ctions \ul0\nosupersub\cf8\f9\fs16 e, \up0 \expndtw0\charscalex115 (x, t) \line \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8 \f9\fs16 e+ (x t) \ul0\nosupersub\cf3\f4\fs16 where the total arrival rate \ul0\ nosupersub\cf8\f9\fs16 a(t) \ul0\nosupersub\cf3\f4\fs16 in (4.1) is \line \up0 \ expndtw0\charscalex110 integrable over all \up0 \expndtw0\charscalex113 \u8212?o o < \ul0\nosupersub\cf8\f9\fs16 t < Do.\par\pard\qj \li1622\ri6\sb0\sl-211\slmul t0\tx1871 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf3\f4\fs16 2) Vehicles m ove forward on the highway at a deterministic \line\tab \up0 \expndtw0\charscale x116 velocity speciied by the velocity ield v(x, t).\par\pard\ql \li1627\sb23\sl -184\slmult0 \up0 \expndtw0\charscalex118 3) The state of each vehicle after it arrives evolves as a\par\pard\qj \li1862\ri1\sb7\sl-209\slmult0\fi9\tx4528 \up0 \expndtw0\charscalex113 nonstationary continuous-time Markov chain, while it \li ne \up0 \expndtw0\charscalex107 moves deterministically down the highway. The Ma rkov \line \up0 \expndtw0\charscalex111 chains of different vehicles are conditi onally stochasti-\line \up0 \expndtw0\charscalex107 cally independent given thei r arrival times. (The Markov \line \up0 \expndtw0\charscalex106 chains are not u nconditionally independent due to depen-\line \up0 \expndtw0\charscalex115 dence induced through the arrival times, but once we \line \up0 \expndtw0\charscalex1 09 condition upon the arrival times, there is no dependence \line \up0 \expndtw0 \charscalex116 left.) A calling vehicle becomes a noncalling vehicle \line \up0 \expndtw0\charscalex115 and vice versa (due to call termination and initiation) \line \up0 \expndtw0\charscalex109 randomly with intensity \ul0\nosupersub\cf8\f 9\fs16 (x, t) \ul0\nosupersub\cf3\f4\fs16 and \tab \up0 \expndtw0\charscalex109 \ ul0\nosupersub\cf8\f9\fs16 (x, t), \ul0\nosupersub\cf3\f4\fs16 respectively. \li ne \up0 \expndtw0\charscalex118 In addition, a calling \up0 \expndtw0\charscalex 115 (noncalling) vehicle leaves the \line \up0 \expndtw0\charscalex113 highway r andomly with intensity \ul0\nosupersub\cf8\f9\fs16 \u8226?-y(x,t)(0(x,t)).\par\p

ard\column \ql \li35\sb13\sl-184\slmult0 \up0 \expndtw0\charscalex113 \ul0\nosup ersub\cf3\f4\fs16 (2.1) and \up0 \expndtw0\charscalex104 (2.2).\par\pard\qj \li 30\ri1486\sb28\sl-196\slmult0\fi355 \up0 \expndtw0\charscalex110 \ul0\nosupersub \cf8\f9\fs16 Proof: \ul0\nosupersub\cf3\f4\fs16 To consider the expected values, (2.3) is replaced \line \up0 \expndtw0\charscalex105 by\par\pard\ql \li6095\sb0 \sl-184\slmult0 \par\pard\ql \li20\sb77\sl-184\slmult0 \up0 \expndtw0\charscalex 118 \ul0\nosupersub\cf8\f9\fs16 E[N(x, t + At)] \u8212? E[N(x, t)] =\par\pard\qj \li241\ri1491\sb20\sl-268\slmult0\tx4146 \up0 \expndtw0\charscalex124 \u8212? n (x, t)v(x, \ul0\nosupersub\cf3\f4\fs16 t)At + \ul0\nosupersub\cf8\f9\fs16 \{E[C, -(x, t + At)] \ul0\nosupersub\cf3\f4\fs16 \u8212? \ul0\nosupersub\cf8\f9\fs16 E[ C:(x,t)]\} \line \up0 \expndtw0\charscalex123 \u8212? \{E[C;(x,t + At)] \u8212? E [C \up0 \expndtw0\charscalex107 , (x, t)] \} \up0 \expndtw0\charscalex108 + o( .6,t). \tab \up0 \expndtw0\charscalex103 \ul0\nosupersub\cf3\f4\fs16 (4.4)\par\p ard\qj \li6100\sb0\sl-206\slmult0 \par\pard\qj \li25\ri1491\sb29\sl-206\slmult0\ fi4\tx4347 \up0 \expndtw0\charscalex108 The proof is completed by following the same argument in the \line \up0 \expndtw0\charscalex110 proof of Lemma 2.1. \tab \up0 \expndtw-2\charscalex100 \u10065?\par\pard\qj \li25\ri1491\sb19\sl-201\slm ult0\fi182 \up0 \expndtw0\charscalex116 Similarly, we have the following, for wh ich we omit the \line \up0 \expndtw0\charscalex105 proof.\par\pard\qj \li35\ri14 90\sb11\sl-206\slmult0\fi163 \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9 \fs16 Lemma 4.2: \ul0\nosupersub\cf3\f4\fs16 The results in Lemma 2.2, namely \ up0 \expndtw0\charscalex116 (2.4) to \up0 \expndtw0\charscalex114 (2.7), hold fo r the stochastic trafic model.\par\pard\qj \li25\ri1491\sb10\sl-206\slmult0\fi17 2 \up0 \expndtw0\charscalex113 We can combine Lemmas 4.1 and 4.2 to obtain an an alog \up0 \expndtw0\charscalex113 of Theorem 2.1.\par\pard\qj \li25\ri1496\sb14\ sl-208\slmult0\fi187 \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf8\f9\fs16 Th eorem 4.1: \ul0\nosupersub\cf3\f4\fs16 In the stochastic trafic model, the densi ties \line \up0 \expndtw0\charscalex114 of noncalling and calling vehicles, \ul0 \nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\ cf8\f9\fs16 q(x, t), at \ul0\nosupersub\cf3\f4\fs16 any \line \up0 \expndtw0\cha rscalex114 location \ul0\nosupersub\cf8\f9\fs16 x > \ul0\nosupersub\cf3\f4\fs16 0 and time \u8212?co < \ul0\nosupersub\cf8\f9\fs16 t < t \ul0\nosupersub\cf3\f4\ fs16 satisfy (2.8) and (2.9).\par\pard\qj \li25\ri1491\sb1\sl-211\slmult0\fi177 \up0 \expndtw0\charscalex116 Since the PDE's for the stochastic model are identi cal to \up0 \expndtw0\charscalex115 those for the deterministic luid model, Lemm a 2.3 remains \up0 \expndtw0\charscalex111 applicable to convert them into ODE' s.\par\pard\qj \li25\ri1481\sb5\sl-206\slmult0\fi172 \up0 \expndtw0\charscalex12 2 Now suppose that the highway is divided into cells at \line \up0 \expndtw0\cha rscalex110 some locations \up0 \expndtw0\charscalex111 \{xo (a \up0 \expndtw0\ch arscalex114 0), xi , x2, \u8226? \u8226? .\}, \ul0\nosupersub\cf8\f9\fs16 Qi(t) \ul0\nosupersub\cf3\f4\fs16 is the number \up0 \expndtw0\charscalex113 of ongoin g calls \up0 \expndtw0\charscalex121 (i.e., offered load) in cell i at time t, a nd \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf8\f9\fs16 Hi(t) \ul0\nosupersu b\cf3\f4\fs16 is the number of calls handed off from cell \ul0\nosupersub\cf8\f9 \fs16 i \ul0\nosupersub\cf3\f4\fs16 \u8212?1 to cell \up0 \expndtw0\charscalex11 5 i before time \ul0\nosupersub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 The f ollowing is an analog of Theorem 2.2; \up0 \expndtw0\charscalex114 it follows fr om Theorem 3.1 of [7]. \par\pard\sect\sectd\fs24\paperw12240\paperh15840{\bkmkst art Pg10}{\bkmkend Pg10}\par\pard\li1660\sb0\sl-115\slmult0\par\pard\li1660\sb0\ sl-115\slmult0\par\pard\li1660\sb0\sl-115\slmult0\par\pard\li1660\sb0\sl-115\slm ult0\par\pard\li1660\sb0\sl-115\slmult0\par\pard\li1660\sb63\sl-115\slmult0\fi0\ tx5284 \up0 \expndtw0\charscalex125 \ul0\nosupersub\cf1\f2\fs10 1362\tab \dn2 \e xpndtw0\charscalex126 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\li1646\sb0\sl-184\slmult0\par\pard\li1646\sb167\sl -184\slmult0\fi182\tx6312\tx7137\tx7363\tx8025 \up0 \expndtw0\charscalex121 \ul0 \nosupersub\cf8\f9\fs16 Theorem 4.2: \ul0\nosupersub\cf3\f4\fs16 In the stochast ic traic model,\tab \up0 \expndtw0\charscalex121 By (4.1), \tab \up0 \expndtw0\c harscalex120 '\tab \dn2 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 ft(x )dx\tab \dn2 \expndtw0\charscalex121 \ul0\nosupersub\cf3\f4\fs16 = 1 for all \ul 0\nosupersub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 Given that this arrival\

par\pard\sect\sectd\sbknone\cols2\colno1\colw6187\colsr110\colno2\colw5803\colsr 160\qj \li1651\ri0\sb5\sl-211\slmult0\fi168 \up0 \expndtw0\charscalex113 a) For each cell i > 1 and at any given time \u8212?oo < t < oo, \up0 \expndtw0\charsca lex118 \ul0\nosupersub\cf8\f9\fs16 WO is \ul0\nosupersub\cf3\f4\fs16 a stochasti c process with mean\par\pard\ql \li3921\sb13\sl-184\slmult0 \up0 \expndtw-2\char scalex100 \ul0\nosupersub\cf8\f9\fs16 x,\par\pard\ql \li2870\sb8\sl-184\slmult0\ tx4171\tx5759 \up0 \expndtw0\charscalex110 E[Q1(t)] =\tab \up0 \expndtw0\charsca lex127 q(x,t)dx.\tab \up0 \expndtw0\charscalex105 \ul0\nosupersub\cf3\f4\fs16 (4 .5)\par\pard\ql \li3840\sb5\sl-177\slmult0 \up0 \expndtw-7\charscalex94 x, -1\pa r\pard\qj \li1646\ri0\sb59\sl-211\slmult0\fi172 \up0 \expndtw0\charscalex121 b) For each cell \ul0\nosupersub\cf8\f9\fs16 i \ul0\nosupersub\cf3\f4\fs16 > \up0 \expndtw-7\charscalex92 1, \up0 \expndtw-7\charscalex81 \ul0\nosupersub\cf8\f9\ fs16 111 \up0 \expndtw0\charscalex118 (t) \ul0\nosupersub\cf3\f4\fs16 is a stoch astic process as a \up0 \expndtw0\charscalex117 function of t with rate\par\pard \ql \li2961\sb42\sl-184\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf1 4\f15\fs16\ul dE[Ili(t)]\par\pard\ql \li2342\sb0\sl-128\slmult0\tx3244\tx3762\tx 5231\tx5759 \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 hi(t)\tab \ up0 \expndtw0\charscalex113 dt\tab \up0 \expndtw0\charscalex127 = q(x,\ul0\sub\c f23\f24\fs15 -\ul0\nosupersub\cf8\f9\fs16 1, t)v(x\tab \up0 \expndtw0\charscalex 115 t).\tab \up0 \expndtw0\charscalex103 (4.6)\par\pard\qj \li1646\ri0\sb149\sl214\slmult0\fi355 \up0 \expndtw0\charscalex122 Proof: \ul0\nosupersub\cf3\f4\fs1 6 Part a) is elementary, so we only discuss b). \line \up0 \expndtw0\charscalex1 19 To consider handof calls, let \ul0\nosupersub\cf8\f9\fs16 r(s, x) \ul0\nosupe rsub\cf3\f4\fs16 denote the time when \line \up0 \expndtw0\charscalex118 a vehic le that arrives to the highway at time \ul0\nosupersub\cf8\f9\fs16 s \ul0\nosupe rsub\cf3\f4\fs16 reaches the \line \up0 \expndtw0\charscalex115 location \ul0\no supersub\cf8\f9\fs16 x. \ul0\nosupersub\cf3\f4\fs16 We assume \ul0\nosupersub\cf 8\f9\fs16 r(s, x) = oo \ul0\nosupersub\cf3\f4\fs16 if the vehicle will never \li ne \up0 \expndtw0\charscalex127 reach location \ul0\nosupersub\cf8\f9\fs16 x. Ha (t) \ul0\nosupersub\cf3\f4\fs16 is deined as the number of jth \line \up0 \expn dtw0\charscalex112 ongoing calls handed of from cell i \u8212? 1 to cell i befor e time\par\pard\ql \li1651\sb18\sl-184\slmult0 \up0 \expndtw0\charscalex116 \ul0 \nosupersub\cf8\f9\fs16 t. \ul0\nosupersub\cf3\f4\fs16 Then,\par\pard\ql \li1651 \sb0\sl-184\slmult0 \par\pard\ql \li2179\sb35\sl-184\slmult0\tx4818 \up0 \expndt w0\charscalex128 \ul0\nosupersub\cf8\f9\fs16 H,? (t) = fool\tab \up0 \expndtw0\c harscalex122 cindA(s).\par\pard\ql \li3331\sb0\sl-90\slmult0 \up0 \expndtw0\char scalex130 \ul0\nosupersub\cf23\f24\fs10 -(T; \up0 \expndtw0\charscalex128 \ul0\n osupersub\cf27\f28\fs12 (j)<r(.9,x \up0 \expndtw0\charscalex132 -1)<T.,-\par\par d\qj \li1651\ri0\sb164\sl-213\slmult0 \up0 \expndtw0\charscalex104 \ul0\nosupers ub\cf3\f4\fs16 Clearly, \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs1 6 11,(t) is \ul0\nosupersub\cf3\f4\fs16 the sum of \up0 \expndtw-7\charscalex93 \ul0\nosupersub\cf8\f9\fs16 11, \up0 \expndtw0\charscalex126 (t) \ul0\nosupersub \cf3\f4\fs16 over all j. The rate of \line \up0 \expndtw0\charscalex111 \ul0\nos upersub\cf8\f9\fs16 E[Hi(t)] \ul0\nosupersub\cf3\f4\fs16 as given in (4.6) is ob tained by considering only the \line \up0 \expndtw0\charscalex114 expected value s and by following the approach in Theorem\par\pard\ql \li1651\sb12\sl-184\slmul t0\tx5951 \up0 \expndtw0\charscalex101 2.2.\tab \up0 \expndtw0\charscalex101 \u1 0065?\par\pard\ql \li2443\sb0\sl-184\slmult0 \par\pard\ql \li2443\sb74\sl-184\sl mult0 \up0 \expndtw0\charscalex124 \ul0\nosupersub\cf6\f7\fs12 V. THE MARKOVIAN HIGHWAY \ul0\nosupersub\cf3\f4\fs16 PALM\par\pard\qj \li1641\ri0\sb75\sl-214\slm ult0\fi182\tx3299 \up0 \expndtw0\charscalex114 We obtain the full Markovian high way PALM simply by \line \up0 \expndtw0\charscalex113 assuming, in addition to t he assumptions of Section IV, that \line \up0 \expndtw0\charscalex114 \ul0\nosup ersub\cf8\f9\fs16 E (x t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9 \fs16 E+ (x, t) \ul0\nosupersub\cf3\f4\fs16 are independent two-dimensional Pois -\line \up0 \expndtw0\charscalex119 son counting processes. Let .E+(B) and \ul0\ nosupersub\cf8\f9\fs16 E\u355?(B) \ul0\nosupersub\cf3\f4\fs16 be random \line \u p0 \expndtw0\charscalex115 measures associated with the stochastic counting proc esses \line \up0 \expndtw0\charscalex130 \ul0\nosupersub\cf8\f9\fs16 E9(x\ul0\no supersub\cf3\f4\fs16 ,t) and \ul0\nosupersub\cf8\f9\fs16 E \up0 \expndtw0\chars

calex118 +,(x, \ul0\nosupersub\cf3\f4\fs16 t); i.e., \ul0\nosupersub\cf8\f9\fs16 E\ul0\nosupersub\cf27\f28\fs12 qI \up0 \expndtw-2\charscalex100 n \up0 \expndtw 0\charscalex118 \ul0\nosupersub\cf8\f9\fs16 (B) \ul0\nosupersub\cf3\f4\fs16 coun ts the number of \line \up0 \expndtw0\charscalex131 arrivals in the set \ul0\nos upersub\cf8\f9\fs16 B \ul0\nosupersub\cf3\f4\fs16 where \ul0\nosupersub\cf8\f9\f s16 B is a \ul0\nosupersub\cf3\f4\fs16 set of \up0 \expndtw0\charscalex124 \ul0\ nosupersub\cf8\f9\fs16 (x, t) \ul0\nosupersub\cf3\f4\fs16 pairs in \line \up0 \e xpndtw0\charscalex111 [0, oo) x \ul0\nosupersub\cf8\f9\fs16 R. \ul0\nosupersub\c f3\f4\fs16 The Poisson assumption means that the numbers \line \up0 \expndtw0\ch arscalex122 of arrivals \ul0\nosupersub\cf8\f9\fs16 E+ (Bi) \ul0\nosupersub\cf3\ f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 .E\u355?(B,) \ul0\nosupersub\cf3\f4\fs16 of calling and noncalling \line \up0 \expndtw0\charscalex114 vehicles in disjoi nts subsets \ul0\nosupersub\cf8\f9\fs16 B1,1 \up0 \expndtw0\charscalex130 < i < n, \ul0\nosupersub\cf3\f4\fs16 of [0, oo) x \ul0\nosupersub\cf8\f9\fs16 R \line \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 are mutually independen t random variables with Poisson dis-\line \up0 \expndtw0\charscalex111 tribution s determined by the deterministic intensity functions \line \up0 \expndtw0\chars calex108 \ul0\nosupersub\cf8\f9\fs16 eq+ (x t) \ul0\nosupersub\cf3\f4\fs16 and \ ul0\nosupersub\cf8\f9\fs16 e \up0 \expndtw-2\charscalex100 , \up0 \expndtw0\char scalex117 (x, \tab \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf3\f4\fs16 resp ectively; e.g.,\par\pard\ql \li1636\sb0\sl-184\slmult0 \par\pard\ql \li5755\sb98 \sl-184\slmult0 \up0 \expndtw0\charscalex103 (5.1)\par\pard\ql \li1641\sb171\sl184\slmult0 \up0 \expndtw0\charscalex106 where\par\pard\ql \li2755\sb161\sl-184\ slmult0\tx3940\tx5745 \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf8\f9\fs16 y+ (B) = f\tab \up0 \expndtw0\charscalex128 e:(u,v)dudv.\tab \up0 \expndtw0\char scalex105 \ul0\nosupersub\cf3\f4\fs16 (5.2)\par\pard\qj \li1636\sb0\sl-206\slmul t0 \par\pard\qj \li1641\ri16\sb19\sl-206\slmult0 \up0 \expndtw0\charscalex111 Fo r example, for \ul0\nosupersub\cf8\f9\fs16 B = \ul0\nosupersub\cf3\f4\fs16 [0, \ ul0\nosupersub\cf8\f9\fs16 x] \ul0\nosupersub\cf3\f4\fs16 x (\u8212?co, t] and \ ul0\nosupersub\cf8\f9\fs16 ^y+ (x , t) '7+ (B) \up0 \expndtw0\charscalex123 \ul0 \nosupersub\cf3\f4\fs16 for this \ul0\nosupersub\cf8\f9\fs16 B,\par\pard\ql \li3 600\sb15\sl-184\slmult0 \up0 \expndtw0\charscalex216 x t\par\pard\ql \li2587\sb1 2\sl-184\slmult0\tx5745 \up0 \expndtw0\charscalex106 -y+ (x,t) \up0 \expndtw0\ch arscalex170 = f f e:(u,v)dudv.\tab \up0 \expndtw0\charscalex103 \ul0\nosupersub\ cf3\f4\fs16 (5.3)\par\pard\qj \li1636\sb0\sl-211\slmult0 \par\pard\qj \li1636\ri 11\sb6\sl-211\slmult0\fi177 \up0 \expndtw0\charscalex107 Alternatively (equivale ntly), we can have a one-dimensional \line \up0 \expndtw0\charscalex108 nonhomog eneous Poisson arrival process A(t) with arrival rate\par\pard\column \qj \li35\ ri1265\sb23\sl-208\slmult0\tx2494 \up0 \expndtw0\charscalex127 occurs at locatio n \ul0\nosupersub\cf8\f9\fs16 x, \ul0\nosupersub\cf3\f4\fs16 we make it a callin g vehicle with \line \up0 \expndtw0\charscalex111 probability \ul0\nosupersub\cf 8\f9\fs16 e-1 (x, t) 1[4 (x, t) \tab \up0 \expndtw0\charscalex127 e:(x,t)] \ul0\ nosupersub\cf3\f4\fs16 and a noncalling \line \up0 \expndtw0\charscalex110 vehic le otherwise.\par\pard\qj \li49\ri1270\sb11\sl-211\slmult0\fi167 \up0 \expndtw0\ charscalex113 With this extra Poisson assumptions, the PALM results in \up0 \exp ndtw-1\charscalex100 [6], \up0 \expndtw0\charscalex110 171 imply the following. (See Theorem 3.1 of [7].)\par\pard\qj \li35\ri1269\sb0\sl-208\slmult0\fi187 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf8\f9\fs16 Theorem \ul0\nosupersub\cf3 \f4\fs16 5.1: If, in addition to the assumptions in Section \line \up0 \expndtw0 \charscalex114 IV, \ul0\nosupersub\cf8\f9\fs16 E+(x, t) \ul0\nosupersub\cf3\f4\f s16 and \ul0\nosupersub\cf8\f9\fs16 E+ (x, t) \ul0\nosupersub\cf3\f4\fs16 are in dependent two-dimensional \line \up0 \expndtw0\charscalex118 Poisson counting pr ocesses, then the stochastic processes\par\pard\qj \li35\ri1279\sb0\sl-213\slmul t0\fi14 \up0 \expndtw0\charscalex127 \ul0\nosupersub\cf8\f9\fs16 \{Q(x,t) \up0 \expndtw0\charscalex114 : x \up0 \expndtw0\charscalex120 > \ul0\nosupersub\cf3\ f4\fs16 0\} and \ul0\nosupersub\cf8\f9\fs16 \{H,(t) \ul0\nosupersub\cf3\f4\fs16 : \u8212?oo < \ul0\nosupersub\cf8\f9\fs16 t < oo\} \ul0\nosupersub\cf3\f4\fs16 a re Poisson \line \up0 \expndtw0\charscalex110 processes. Moreover, \ul0\nosupers ub\cf8\f9\fs16 Qi(t) \ul0\nosupersub\cf3\f4\fs16 for \ul0\nosupersub\cf8\f9\fs16 i \ul0\nosupersub\cf3\f4\fs16 > 1 are mutually independent \up0 \expndtw0\chars

calex111 Poisson random variables with means in (4.5).\par\pard\qj \li30\ri1269\ sb0\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex118 Unlike Theorem 3.1 of \up 0 \expndtw0\charscalex121 [7], the processes \ul0\nosupersub\cf8\f9\fs16 C+(x, t ) \ul0\nosupersub\cf3\f4\fs16 and \line \up0 \expndtw0\charscalex109 \ul0\nosupe rsub\cf8\f9\fs16 C\u8212? (x, t) \ul0\nosupersub\cf3\f4\fs16 are not two-dimensi onal Poisson processes, because \line \up0 \expndtw0\charscalex114 calls can ent er and leave more than once. We also point out \line \up0 \expndtw0\charscalex11 8 that the Poisson property for \ul0\nosupersub\cf8\f9\fs16 Q1(t) \ul0\nosupersu b\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 WO \ul0\nosupersub\cf3\f4\fs16 hol ds without \line \up0 \expndtw0\charscalex111 the Markov assumption for each veh icle made in Section IV. \line \up0 \expndtw0\charscalex115 However, without the se Markov assumptions, the PDE's in \line \up0 \expndtw0\charscalex111 (2.8) and (2.9) are no longer valid. In this case, \ul0\nosupersub\cf8\f9\fs16 q(x, t) \u l0\nosupersub\cf3\f4\fs16 cannot \line \up0 \expndtw0\charscalex116 be easily co mputed because the model has to keep track of \line \up0 \expndtw0\charscalex109 the residual think time and call-holding time for each vehicle; \line \up0 \exp ndtw0\charscalex100 see \up0 \expndtw-1\charscalex100 [7].\par\pard\qj \li25\ri1 289\sb19\sl-210\slmult0\fi182 \up0 \expndtw0\charscalex114 If only the location dynamics at temporal steady state are \up0 \expndtw0\charscalex116 of interest, the arrival process can be a time-homogeneous \up0 \expndtw0\charscalex117 Poiss on process. All results of the Markovian PALM pre-\line \up0 \expndtw0\charscale x112 sented above remain valid simply because a stationary Pois-\line \up0 \expn dtw0\charscalex115 son process is a special case of time-homogeneous Poisson \up 0 \expndtw0\charscalex116 process. In this case, as for the deterministic luid m odel in \up0 \expndtw0\charscalex109 Section II-B, the densities for noncalling and calling vehicles, \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf8\f9\fs16 n (x) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 q(x), \ul0\nosup ersub\cf3\f4\fs16 can be solved from (2.20) and (2.21), or given \up0 \expndtw0\ charscalex112 by (2.27) and (2.28), respectively.\par\pard\ql \li1636\sb0\sl-138 \slmult0 \par\pard\ql \li1636\sb0\sl-138\slmult0 \par\pard\ql \li1138\sb1\sl-138 \slmult0 \up0 \expndtw0\charscalex126 \ul0\nosupersub\cf6\f7\fs12 VI. APPROXIMAT ING VEHICLE\par\pard\ql \li635\sb83\sl-138\slmult0 \up0 \expndtw0\charscalex122 INTERACTIONS IN THE STOCHASTIC MODELS\par\pard\qj \li25\ri1294\sb81\sl-210\slmul t0\fi177 \up0 \expndtw0\charscalex123 \ul0\nosupersub\cf3\f4\fs16 A key to tract ability in the stochastic trafic models in \line \up0 \expndtw0\charscalex109 Se ctions IV and V is the assumption that vehicles (customers) \line \up0 \expndtw0 \charscalex114 do not interact; i.e., the nonstationary Markov location pro-\lin e \up0 \expndtw0\charscalex117 cesses for diferent vehicles are conditionally in dependent, \line \up0 \expndtw0\charscalex116 given their arrival times. This me ans that diferent vehicles \line \up0 \expndtw0\charscalex110 cannot interact.\p ar\pard\qj \li20\ri1284\sb11\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex107 H owever, there is no such interaction restriction in the deter-\line \up0 \expndt w0\charscalex110 ministic luid model. For example, vehicle velocity commonly \li ne \up0 \expndtw0\charscalex111 depends on vehicular density, with velocity decr easing as the \line \up0 \expndtw0\charscalex112 density increases. This feature is easily incorporated into the \line \up0 \expndtw0\charscalex110 deterministi c fluid model via the PDE's in (2.8) and (2.9). For \line \up0 \expndtw0\charsca lex114 example, one possible model has velocity decrease linearly \line \up0 \ex pndtw0\charscalex110 with density \up0 \expndtw-1\charscalex100 [2], \up0 \expnd tw-1\charscalex100 [9], \up0 \expndtw0\charscalex113 [10]. Velocity can reach it s maximum \line \up0 \expndtw0\charscalex113 value \ul0\nosupersub\cf8\f9\fs16 v m(x, \ul0\nosupersub\cf3\f4\fs16 t), when the vehicular density is close to zero at \line \up0 \expndtw0\charscalex111 location \ul0\nosupersub\cf8\f9\fs16 x \u l0\nosupersub\cf3\f4\fs16 and time \ul0\nosupersub\cf8\f9\fs16 t. \ul0\nosupersu b\cf3\f4\fs16 On the other hand, the velocity reduces \line \up0 \expndtw0\chars calex121 to zero if the density is increased to a critical value \up0 \expndtw0\ charscalex103 (i.e., \line \up0 \expndtw0\charscalex111 bumper-to-bumper density ). That is, for some critical density\par\pard\ql \li25\sb18\sl-184\slmult0\tx25 5 \up0 \expndtw-7\charscalex91 \ul0\nosupersub\cf8\f9\fs16 lc\tab \up0 \expndtw0 \charscalex122 > \ul0\nosupersub\cf3\f4\fs16 0,\par\pard\ql \li2449\sb80\sl-184\

slmult0\tx3160 \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf14\f15\fs16\ul n(x \ul0\nosupersub\cf8\f9\fs16 ,t)\tab \up0 \expndtw0\charscalex130 \ul0\nosupersu b\cf14\f15\fs16\ul q(x,t)]\par\pard\sect\sectd\sbknone\cols3\colno1\colw9089\col sr160\colno2\colw1021\colsr160\colno3\colw1680\colsr160\qj \li1636\ri568\sb0\sl180\slmult0\tx6988 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\f4\fs16 func tion \ul0\nosupersub\cf8\f9\fs16 a(t) \ul0\nosupersub\cf3\f4\fs16 as in (4.1) an d let an arrival at time t occur in\tab \up0 \expndtw0\charscalex114 v(x, \ul0\n osupersub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs1 6 = v,n(x, \ul0\nosupersub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 [1 \up0 \e xpndtw0\charscalex114 the spatial location \ul0\nosupersub\cf8\f9\fs16 x \ul0\no supersub\cf3\f4\fs16 with density\par\pard\column \ql \li20\sb56\sl-184\slmult0 \up0 \expndtw0\charscalex122 /,\par\pard\column \ql \li20\sb0\sl-144\slmult0 \up 0 \expndtw0\charscalex103 (6.1)\par\pard\sect\sectd\sbknone\cols3\colno1\colw308 0\colsr160\colno2\colw2360\colsr160\colno3\colw6350\colsr160\ql \li2395\sb212\sl -230\slmult0 \up0 \expndtw0\charscalex107 \ul0\nosupersub\cf34\f35\fs20 h(x) =\p ar\pard\column \ql \li20\sb96\sl-184\slmult0\tx615 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf8\f9\fs16 (x, t)\tab \up0 \expndtw0\charscalex116 e- (x, t)\pa r\pard\qj \li346\ri222\sb0\sl-144\slmult0\fi1291 \up0 \expndtw0\charscalex101 \u l0\nosupersub\cf34\f35\fs20 x > \ul0\nosupersub\cf3\f4\fs16 0. \line \up0 \expnd tw0\charscalex129 a(t)\par\pard\column \ql \li567\sb1\sl-160\slmult0 \up0 \expnd tw0\charscalex112 This formula can easily be inserted into (2.8) and (2.9).\par\ pard\ql \li20\sb70\sl-184\slmult0\tx750 \up0 \expndtw0\charscalex103 (5.4)\tab \ up0 \expndtw0\charscalex114 In a full stochastic analog of this phenomenon, the actual\par\pard\ql \li567\sb1\sl-164\slmult0 \up0 \expndtw0\charscalex112 distri bution of vehicles at any time would be random, so that \par\pard\sect\sectd\fs2 4\paperw12240\paperh15840{\bkmkstart Pg11}{\bkmkend Pg11}\par\pard\sect\sectd\sb knone\cols2\colno1\colw6143\colsr20\colno2\colw5937\colsr160\ql \li1502\sb0\sl-1 15\slmult0 \par\pard\ql \li1502\sb0\sl-115\slmult0 \par\pard\ql \li1502\sb0\sl-1 15\slmult0 \par\pard\ql \li1502\sb0\sl-115\slmult0 \par\pard\ql \li1502\sb0\sl-1 15\slmult0 \par\pard\ql \li1502\sb106\sl-115\slmult0 \up0 \expndtw0\charscalex12 1 \ul0\nosupersub\cf1\f2\fs10 LEUNG \ul0\nosupersub\cf23\f24\fs10 et al.: \ul0\n osupersub\cf1\f2\fs10 TRAFFIC MODELS FOR WIRELESS COMMUNICATION NETWORKS\par\par d\qj \li1497\sb0\sl-212\slmult0 \par\pard\qj \li1497\ri12\sb107\sl-212\slmult0 \ up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 both the vehicular densi ty, however deined, and the resulting \line \up0 \expndtw0\charscalex117 vehicle velocity deined by \up0 \expndtw0\charscalex115 (6.1) also become random. We \l ine \up0 \expndtw0\charscalex121 would need some sort of stochastic diferential equation, \line \up0 \expndtw0\charscalex110 corresponding to a complicated inte racting stochastic particle\par\pard\column \ql \li10391\sb0\sl-115\slmult0 \par \pard\ql \li10391\sb0\sl-115\slmult0 \par\pard\ql \li10391\sb0\sl-115\slmult0 \p ar\pard\ql \li10391\sb0\sl-115\slmult0 \par\pard\ql \li10391\sb0\sl-115\slmult0 \par\pard\ql \li10391\sb0\sl-115\slmult0 \par\pard\ql \li4248\sb5\sl-115\slmult0 \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf1\f2\fs10 1363\par\pard\qj \li14 88\sb0\sl-209\slmult0 \par\pard\qj \li20\ri1423\sb118\sl-209\slmult0 \up0 \expnd tw0\charscalex112 \ul0\nosupersub\cf3\f4\fs16 whether or not they have calls in progress. The deterministic \line \up0 \expndtw0\charscalex112 model treats each type of vehicle as a continuous luid, while \line \up0 \expndtw0\charscalex109 the stochastic models consider the behavior of each individual \line \up0 \expnd tw0\charscalex108 vehicle. All models use the same two coupled PDE's or ODE's\pa r\pard\sect\sectd\sbknone\cols2\colno1\colw6034\colsr110\colno2\colw5956\colsr16 0\ql \li1502\sb1\sl-182\slmult0 \up0 \expndtw0\charscalex103 system.\par\pard\qj \li1502\ri0\sb11\sl-211\slmult0\fi172\tx3774 \up0 \expndtw0\charscalex108 To be more concrete, we might count the random number of \line \up0 \expndtw0\charsca lex112 vehicles in the interval \ul0\nosupersub\cf8\f9\fs16 (x, x \tab \up0 \exp ndtw0\charscalex110 1] \ul0\nosupersub\cf3\f4\fs16 and let this be the "random \ up0 \expndtw0\charscalex123 density" at \ul0\nosupersub\cf8\f9\fs16 x, \ul0\nosu persub\cf3\f4\fs16 say \ul0\nosupersub\cf8\f9\fs16 D(x, t). \ul0\nosupersub\cf3\ f4\fs16 We could then let the random \line \up0 \expndtw0\charscalex117 velocity at location \ul0\nosupersub\cf8\f9\fs16 x \ul0\nosupersub\cf3\f4\fs16 and time \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 be\par\pard\ql \li1488

\sb0\sl-184\slmult0 \par\pard\ql \li2491\sb88\sl-184\slmult0\tx4257\tx5606 \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9\fs16 V(x, t) \ul0\nosupersub\cf3\ f4\fs16 = vm(x, \ul0\nosupersub\cf8\f9\fs16 t) \ul0\nosupersub\cf3\f4\fs16 [1\ta b \up0 \expndtw0\charscalex142 I(i \up0 \expndtw0\charscalex138 zc't)].\tab \up0 \expndtw0\charscalex103 (6.2)\par\pard\ql \li1488\sb0\sl-211\slmult0 \par\pard\ ql \li1497\ri0\sb63\sl-211\slmult0\fi4\tx1675 \up0 \expndtw0\charscalex112 Obvio usly, (6.2) is a stochastic analog of (6.1). \line\tab \up0 \expndtw0\charscalex 115 Unfortunately, however, the stochastic model with \up0 \expndtw0\charscalex1 03 (6.2) \line \up0 \expndtw0\charscalex115 seems very diicult to analyze. Our i dea is to approximately \line \up0 \expndtw0\charscalex107 capture the behavior of complicated stochastic dynamics given \line \up0 \expndtw0\charscalex108 by ( 6.2) by substituting appropriate relations among determin-\line \up0 \expndtw0\c harscalex113 istic quantities in the deterministic analog \up0 \expndtw0\charsca lex112 (6.1). Indeed, we \line \up0 \expndtw0\charscalex109 can apply (6.1) dire ctly to the stochastic model if we interpret \line \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf8\f9\fs16 n(x , t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupe rsub\cf8\f9\fs16 q(x , t) \ul0\nosupersub\cf3\f4\fs16 as the densities of expect ed values. Then \line \up0 \expndtw0\charscalex107 the deterministic velocity is allowed to depend on the expected \line \up0 \expndtw0\charscalex112 values via \ul0\nosupersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupe rsub\cf8\f9\fs16 q(x, t). \ul0\nosupersub\cf3\f4\fs16 As for the deterministic m odel, \line \up0 \expndtw0\charscalex108 (6.1) can be inserted into (2.8) and (2 .9) and solved. Just as for \line \up0 \expndtw0\charscalex114 the deterministic model, this can have a signiicant inluence \line \up0 \expndtw0\charscalex120 u pon v(x, \ul0\nosupersub\cf8\f9\fs16 t), n(x, t), \ul0\nosupersub\cf3\f4\fs16 an d \ul0\nosupersub\cf8\f9\fs16 q(x, t).\par\pard\qj \li1492\ri0\sb0\sl-211\slmult 0\fi177 \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf3\f4\fs16 Now we can cons truct a stochastic model according to the \line \up0 \expndtw0\charscalex113 ass umptions of Section IV using this resulting deterministic \line \up0 \expndtw0\c harscalex118 velocity ield v(x, t). By Theorem \up0 \expndtw0\charscalex117 4.1, the densities of the \line \up0 \expndtw0\charscalex113 mean values, \ul0\nosup ersub\cf8\f9\fs16 n(x, t) \ul0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9 \fs16 q(x, t), \ul0\nosupersub\cf3\f4\fs16 in this stochastic model will \line \ up0 \expndtw0\charscalex117 be of the appropriate form. Moreover, if we let the arrival \line \up0 \expndtw0\charscalex113 process have the Poisson property as assumed in Section V, \line \up0 \expndtw0\charscalex120 then we will be able to deduce the Poisson distributional \line \up0 \expndtw0\charscalex106 results as well. The resulting Markovian highway PALM based \line \up0 \expndtw0\charscale x112 on (6.1) then is a candidate approximation for the stochastic \line \up0 \e xpndtw0\charscalex116 model based on (6.2). Of course, it remains to evaluate th e \line \up0 \expndtw0\charscalex111 quality of such approximations.\par\pard\qj \li1492\ri0\sb0\sl-211\slmult0\fi172 \up0 \expndtw0\charscalex106 As emphasized in [7], even though the PALM models cannot \line \up0 \expndtw0\charscalex116 d irectly model vehicle interactions, they often can capture \line \up0 \expndtw0\ charscalex107 essential features. As illustrated by the examples in Section III, \line \up0 \expndtw0\charscalex107 the stochastic model cannot directly represe nt vehicles slowing \line \up0 \expndtw0\charscalex113 down in response to other vehicles in front of them slowing \line \up0 \expndtw0\charscalex115 down, but we can represent the vehicles slowing down at a \line \up0 \expndtw0\charscalex1 11 certain time and space under the assumption that an accident \line \up0 \expn dtw0\charscalex110 occurs there.\par\pard\ql \li1670\sb14\sl-184\slmult0 \up0 \e xpndtw0\charscalex119 Similarly, we can indirectly capture the dependence of\par \pard\column \qj \li29\ri1428\sb3\sl-208\slmult0\fi4 \up0 \expndtw0\charscalex12 2 to describe the evolution of the system. The call density \up0 \expndtw0\chars calex117 and call handof rate are readily computable by solving the \up0 \expndt w0\charscalex113 equations. The two kinds of models are complementary, be-\line \up0 \expndtw0\charscalex108 cause additional features such as the interdependen cy between \up0 \expndtw0\charscalex115 velocity and vehicular density can be ea sily included in the \up0 \expndtw0\charscalex117 luid model, whereas the stocha stic trafic models also give \up0 \expndtw0\charscalex109 probability distributi

ons.\par\pard\qj \li29\ri1443\sb6\sl-211\slmult0\fi182 \up0 \expndtw0\charscalex 112 Numerical examples were presented to illustrate the com-\line \up0 \expndtw0 \charscalex110 putability of our results and investigate various aspects of the \line \up0 \expndtw0\charscalex109 time and space dynamics of wireless networks. The numerical \line \up0 \expndtw0\charscalex109 results indicate that both the time-dependent behavior and the \line \up0 \expndtw0\charscalex113 mobility of vehicles play important roles in determining the \line \up0 \expndtw0\charscalex 112 system performance. Our results show that even for systems \line \up0 \expnd tw0\charscalex115 in temporal steady state, the movement of vehicles and the \li ne \up0 \expndtw0\charscalex118 calling patterns can signiicantly affect the ofe red loads in \line \up0 \expndtw0\charscalex121 a given region of the system. Th erefore, the models will \line \up0 \expndtw0\charscalex117 be useful tools to e xamine various phenomena in wireless \line \up0 \expndtw0\charscalex115 networks . Furthermore, our numerical examples also show \line \up0 \expndtw0\charscalex1 07 how the proposed models can be used to compute approximate \line \up0 \expndt w0\charscalex112 blocking probabilities, as illustrated in Section III. Thus, th e \line \up0 \expndtw0\charscalex110 models have the potential for evolving into tools for planning \line \up0 \expndtw0\charscalex110 and engineering wireless networks.\par\pard\qj \li20\ri1447\sb1\sl-211\slmult0\fi182 \up0 \expndtw0\chars calex107 This work can be extended in several areas. First, the quality \line \u p0 \expndtw0\charscalex114 of the approximation approach to capturing the interd epen-\line \up0 \expndtw0\charscalex115 dence between velocity and vehicular den sity via expected \line \up0 \expndtw0\charscalex110 values in Section VI should be evaluated. Second, the models \line \up0 \expndtw0\charscalex111 assume no c apacity constraints. As shown in Section III, it is \line \up0 \expndtw0\charsca lex111 desirable to apply approximation techniques with the models \line \up0 \e xpndtw0\charscalex108 to develop suitable approximate approaches to quantifying the \line \up0 \expndtw0\charscalex113 blocking probability for new call attempt s and handof calls. \line \up0 \expndtw0\charscalex106 (A start was also made in [7].) These approximation techniques \line \up0 \expndtw0\charscalex107 should be further enhanced to consider retries of blocked calls. \line \up0 \expndtw0\c harscalex115 Lastly, the call blocking is actually afected by the channel-\line \up0 \expndtw0\charscalex112 assignment strategy in use. These new models could be used \line \up0 \expndtw0\charscalex110 to study the tradeoffs of various ass ignment strategies.\par\pard\ql \li1488\sb0\sl-184\slmult0 \par\pard\ql \li1887\ sb161\sl-184\slmult0 \up0 \expndtw-7\charscalex95 APPENDIX\par\pard\qj \li197\ri 1937\sb0\sl-259\slmult0\fi312 \up0 \expndtw-6\charscalex100 A TIME-DEPENDENT PRO PORTIONALITY RESULT \up0 \expndtw0\charscalex118 \ul0\nosupersub\cf8\f9\fs16 Lem ma 2.5: \ul0\nosupersub\cf3\f4\fs16 If for all \ul0\nosupersub\cf8\f9\fs16 x \ul 0\nosupersub\cf3\f4\fs16 and \ul0\nosupersub\cf8\f9\fs16 t, \ul0\nosupersub\cf3\ f4\fs16 we have\par\pard\ql \li3317\sb158\sl-184\slmult0\tx4176 \up0 \expndtw0\c harscalex141 \ul0\nosupersub\cf8\f9\fs16 e:(x,t)\tab \up0 \expndtw-7\charscalex8 7 \ul0\nosupersub\cf3\f4\fs16 A\par\pard\sect\sectd\sbknone\cols3\colno1\colw598 8\colsr160\colno2\colw3796\colsr160\colno3\colw2006\colsr160\qj \li1488\ri11\sb0 \sl-206\slmult0\fi4 \up0 \expndtw0\charscalex121 velocity upon vehicular density as in \up0 \expndtw0\charscalex124 (6.1), if we make \line \up0 \expndtw0\char scalex115 (6.1) apply to expected values. From the perspective of the \up0 \expn dtw0\charscalex122 deterministic luid model, we can think of the stochastic \lin e \up0 \expndtw0\charscalex114 model as an enhancement to be able to deduce appr oximate \up0 \expndtw0\charscalex108 distributional conclusions.\par\pard\column \ql \li198\sb0\sl-144\slmult0\tx2008 \up0 \expndtw0\charscalex105 A(x, \ul0\nos upersub\cf8\f9\fs16 t) \up0 \expndtw0\charscalex154 \ul0\nosupersub\cf3\f4\fs16 = \ul0\nosupersub\cf8\f9\fs16 ti(x, \ul0\nosupersub\cf3\f4\fs16 t) =\tab \up0 \e xpndtw-2\charscalex100 \ul0\nosupersub\cf8\f9\fs16 0 \up0 \expndtw0\charscalex11 3 (x, \ul0\nosupersub\cf3\f4\fs16 t) \up0 \expndtw0\charscalex124 = \ul0\nosuper sub\cf8\f9\fs16 1(x, \ul0\nosupersub\cf3\f4\fs16 t), et (x,\par\pard\ql \li20\ri 396\sb144\sl-393\slmult0\tx88\tx1198\tx3066 \up0 \expndtw0\charscalex127 and \ul 0\nosupersub\cf8\f9\fs16 tx(t)lt \up0 \expndtw0\charscalex113 > \ul0\nosupersub \cf3\f4\fs16 to\} satisfying (2.10), then \line \tab \up0 \expndtw0\charscalex11 3 \ul0\nosupersub\cf14\f15\fs16\ul q(x(to), \ul0\nosupersub\cf18\f19\fs16\ul to)

\ul0\tab \up0 \expndtw-7\charscalex80 \ul0\nosupersub\cf3\f4\fs16 A \tab \up0 \ expndtw-2\charscalex100 _ \up0 \expndtw-7\charscalex80 A\par\pard\column \ql \li 20\sb1\sl-158\slmult0 \up0 \expndtw-7\charscalex80 \u8212?\par\pard\sect\sectd\s bknone\cols2\colno1\colw7073\colsr40\colno2\colw4987\colsr160\ql \li6192\sb61\sl -184\slmult0 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf8\f9\fs16 n(x(to), t o)\par\pard\ql \li2236\sb76\sl-184\slmult0 \up0 \expndtw-3\charscalex100 \ul0\no supersub\cf3\f4\fs16 VII. CONCLUSIONS AND FUTURE WORK\par\pard\column \ql \li20\ sb9\sl-184\slmult0\tx419 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf8\f9\fs1 6 =\tab \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf3\f4\fs16 implies that \u l0\super\cf8\f9\fs24 n(x(t),\ul0\super\cf18\f19\fs24\ul t) \up0 \expndtw0\chars calex118 \ul0\nosupersub\cf5\f6\fs14 \u8212? \ul0\nosupersub\cf3\f4\fs16 it for all \ul0\super\cf8\f9\fs24 t > to.\par\pard\sect\sectd\sbknone\cols2\colno1\colw 5984\colsr160\colno2\colw5956\colsr160\qj \li1488\ri0\sb75\sl-208\slmult0\fi177 \up0 \expndtw0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 We have presented a det erministic luid model, a stochastic \line \up0 \expndtw0\charscalex114 traic mod el, and a Markovian highway PALM for a portion \line \up0 \expndtw0\charscalex11 6 of a wireless network along a highway. Vehicles can enter\par\pard\column \qj \li20\ri1452\sb0\sl-158\slmult0\fi355 \up0 \expndtw0\charscalex113 \ul0\nosupers ub\cf8\f9\fs16 Proof: \ul0\nosupersub\cf3\f4\fs16 Let (I)(x, \ul0\nosupersub\cf8 \f9\fs16 t) \up0 \expndtw0\charscalex125 \ul0\nosupersub\cf3\f4\fs16 =- \ul0\no supersub\cf8\f9\fs16 An(x,t) \up0 \expndtw0\charscalex116 \u8212? pq(x.t). \ul0 \nosupersub\cf3\f4\fs16 Substituting \line \up0 \expndtw0\charscalex112 this and our hypotheses in (2.8) and (2.9) yields\par\pard\ql \li43\sb134\sl-230\slmult0 \tx941 \up0 \expndtw0\charscalex121 \ul0\nosupersub\cf8\f9\fs16 an(\ul0\nosupers ub\cf14\f15\fs16\ul x, t)\ul0\tab \up0 \expndtw-3\charscalex100 \ul0\nosupersub\ cf35\f36\fs26 a\par\pard\sect\sectd\sbknone\cols2\colno1\colw6857\colsr160\colno 2\colw5083\colsr160\qj \li1483\ri333\sb6\sl-201\slmult0\fi4\tx6374 \up0 \expndtw 0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 and leave the system at multiple ent rances and exits, and they\tab \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf8 \f9\fs16 a \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf3\f4\fs16 are classiie d as noncalling and calling \ul0\nosupersub\cf8\f9\fs16 vehicles, \ul0\nosupersu b\cf3\f4\fs16 depending on\par\pard\column \qj \li20\ri3569\sb0\sl-128\slmult0\f i235\tx579 \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf8\f9\fs16 [n(x \up0 \ expndtw0\charscalex116 t)v(x ,t)] \up0 \expndtw0\charscalex108 = \line \up0 \exp ndtw0\charscalex124 ax \tab \up0 \expndtw-2\charscalex100 " \par\pard\sect\sectd \fs24\paperw12240\paperh15840{\bkmkstart Pg12}{\bkmkend Pg12}\par\pard\sect\sect d\sbknone\cols2\colno1\colw2663\colsr20\colno2\colw9417\colsr160\ql \li1617\sb0\ sl-115\slmult0 \par\pard\ql \li1617\sb0\sl-115\slmult0 \par\pard\ql \li1617\sb0\ sl-115\slmult0 \par\pard\ql \li1617\sb0\sl-115\slmult0 \par\pard\ql \li1617\sb0\ sl-115\slmult0 \par\pard\ql \li1617\sb82\sl-115\slmult0 \up0 \expndtw0\charscale x110 \ul0\nosupersub\cf1\f2\fs10 1364\par\pard\ql \li1603\sb0\sl-184\slmult0 \pa r\pard\ql \li1603\sb162\sl-184\slmult0 \up0 \expndtw0\charscalex103 \ul0\nosuper sub\cf3\f4\fs16 and\par\pard\ql \li1622\sb48\sl-218\slmult0\tx2495 \up0 \expndtw 0\charscalex116 \ul0\nosupersub\cf18\f19\fs16\ul aq(x, t)\ul0\tab \up0 \expndtw2\charscalex100 \ul0\nosupersub\cf12\f13\fs24 a\par\pard\ql \li2256\sb0\sl-129\s lmult0 \up0 \expndtw0\charscalex104 \ul0\nosupersub\cf8\f9\fs16 +\par\pard\ql \l i1823\sb0\sl-192\slmult0\tx2398 \up0 \expndtw-11\charscalex89 \ul0\nosupersub\cf 12\f13\fs24 at \tab \up0 \expndtw-11\charscalex89 ax\par\pard\column \ql \li5232 \sb0\sl-115\slmult0 \par\pard\ql \li5232\sb0\sl-115\slmult0 \par\pard\ql \li5232 \sb0\sl-115\slmult0 \par\pard\ql \li5232\sb0\sl-115\slmult0 \par\pard\ql \li5232 \sb0\sl-115\slmult0 \par\pard\ql \li2569\sb77\sl-115\slmult0 \up0 \expndtw0\char scalex126 \ul0\nosupersub\cf1\f2\fs10 IEEE JOURNAL ON SELECTED AREAS IN COMMUNIC ATIONS, VOL. 12, NO. 8, OCTOBER 1994\par\pard\qj \li1603\sb0\sl-153\slmult0 \par \pard\qj \li1603\sb0\sl-153\slmult0 \par\pard\qj \li3927\ri1332\sb71\sl-153\slmu lt0 \up0 \expndtw0\charscalex128 \ul0\nosupersub\cf6\f7\fs12 G. Montenegro, M. S engoku, Y. Yamaguchi, and T. Abe, "Time-\line \up0 \expndtw0\charscalex123 depen dent analysis of mobile communication trafic in a ring-shaped\par\pard\ql \li393 2\sb23\sl-138\slmult0 \up0 \expndtw0\charscalex120 service area with nonuniform vehicle distribution," \ul0\nosupersub\cf27\f28\fs12 IEEE Trans. \ul0\nosupersub \cf6\f7\fs12 Veh.\par\pard\ql \li20\sb0\sl-128\slmult0 \up0 \expndtw0\charscalex

124 \ul0\nosupersub\cf8\f9\fs16 [q(x t)v(s t)] \up0 \expndtw0\charscalex104 =\pa r\pard\ql \li3932\sb0\sl-108\slmult0 \up0 \expndtw0\charscalex114 \ul0\nosupersu b\cf27\f28\fs12 Technol., \ul0\nosupersub\cf6\f7\fs12 vol. \up0 \expndtw0\charsc alex120 41 pp. \up0 \expndtw0\charscalex119 243-254, \up0 \expndtw0\charscalex10 5 1992.\par\pard\ql \li3932\sb0\sl-108\slmult0 \up0 \expndtw0\charscalex115 I. S eskar, S. Maric, J. M. Holtzman, and J. Wasserman, "Rate of location\par\pard\se ct\sectd\sbknone\cols2\colno1\colw6430\colsr160\colno2\colw5510\colsr160\qj \li1 603\ri347\sb0\sl-217\slmult0\fi1900 \up0 \expndtw0\charscalex108 \ul0\nosupersub \cf4\f5\fs14 eq \ul0\nosupersub\cf8\f9\fs16 (x ,t) \u8212? /3(x, t)n(x , t) + \u l0\nosupersub\cf3\f4\fs16 4D(x, \ul0\nosupersub\cf8\f9\fs16 t). \up0 \expndtw0\c harscalex111 \ul0\nosupersub\cf3\f4\fs16 Combining these equations gives us\par\ pard\ql \li2073\sb90\sl-184\slmult0\tx2922\tx3412\tx3647 \up0 \expndtw0\charscal ex113 81,(x,\ul0\nosupersub\cf18\f19\fs16\ul \ul0\nosupersub\cf8\f9\fs16 t)\tab \up0 \expndtw0\charscalex122 v(x\tab \up0 \expndtw-2\charscalex100 \ul0\nosuper sub\cf3\f4\fs16 8\tab \up0 \expndtw0\charscalex111 (x, \ul0\nosupersub\cf8\f9\fs 16 t)\par\pard\column \qj \li20\ri1337\sb0\sl-140\slmult0 \up0 \expndtw0\charsca lex117 \ul0\nosupersub\cf6\f7\fs12 area updates in cellular systems," in \ul0\no supersub\cf27\f28\fs12 Proc. Third WINLAB Workshop an \line \up0 \expndtw0\chars calex116 Third Generation Wireless Networks, \ul0\nosupersub\cf6\f7\fs12 Apr. \u p0 \expndtw0\charscalex105 1992.\par\pard\qj \li20\ri1337\sb0\sl-154\slmult0 \up 0 \expndtw0\charscalex125 P. E. Wright, "A vehicular-trafic based model of cellu lar systems," \line \up0 \expndtw0\charscalex120 Tech. Memo., AT&T Bel Labs., Mu rray Hill, NJ, 1993.\par\pard\sect\sectd\sbknone\cols3\colno1\colw4303\colsr40\c olno2\colw586\colsr20\colno3\colw7161\colsr160\ql \li3081\sb0\sl-144\slmult0\tx3 518\tx4094 \up0 \expndtw0\charscalex115 \ul0\nosupersub\cf8\f9\fs16 x, \ul0\nosu persub\cf3\f4\fs16 t)\tab \up0 \expndtw0\charscalex49 11)\tab \up0 \expndtw0\cha rscalex120 =\par\pard\ql \li2299\sb0\sl-192\slmult0\tx3609 \up0 \expndtw-11\char scalex86 \ul0\nosupersub\cf12\f13\fs24 at\tab \up0 \expndtw-11\charscalex91 ax\p ar\pard\ql \li2721\sb44\sl-184\slmult0 \up0 \expndtw0\charscalex135 \ul0\nosuper sub\cf3\f4\fs16 \u8212? \ul0\nosupersub\cf8\f9\fs16 [A+A+0(x,t)+\par\pard\column \qj \li20\ri0\sb207\sl-240\slmult0\tx223 \up0 \expndtw0\charscalex118 ov(\ul0\n osupersub\cf14\f15\fs16\ul x, t) \line\ul0\tab \up0 \expndtw0\charscalex121 \ul0 \nosupersub\cf8\f9\fs16 ax\par\pard\column \ql \li1612\sb0\sl-184\slmult0 \par\p ard\ql \li1612\sb0\sl-184\slmult0 \par\pard\ql \li20\sb6\sl-184\slmult0 \up0 \ex pndtw0\charscalex101 \ul0\nosupersub\cf14\f15\fs16\ul ]\ul0\nosupersub\cf8\f9\fs 16 \ul0\nosupersub\cf3\f4\fs16 43(x, \ul0\nosupersub\cf8\f9\fs16 t). \up0 \exp ndtw0\charscalex119 \ul0\nosupersub\cf3\f4\fs16 (Al)\par\pard\qj \li2823\ri1342\ sb5\sl-153\slmult0 \up0 \expndtw0\charscalex129 \ul0\nosupersub\cf6\f7\fs12 Kin K. Leung (S'78-M'86-SM'93) received the \line \up0 \expndtw0\charscalex127 B.S. degree with irst class honors in electronics\par\pard\sect\sectd\sbknone\cols2\c olno1\colw4789\colsr20\colno2\colw7291\colsr160\qj \li1612\ri0\sb0\sl-170\slmult 0 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf3\f4\fs16 By the chain rule and the fact that \ul0\nosupersub\cf8\f9\fs16 \{x(t)lt \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs16 (2.10), we have\par\pard\ql \li1996\sb109\sl-184\sl mult0\tx3062 \up0 \expndtw0\charscalex59 \ul0\nosupersub\cf14\f15\fs16\ul (.143 \up0 \expndtw0\charscalex119 (x(t), t)\ul0\tab \up0 \expndtw0\charscalex101 \ul0 \nosupersub\cf14\f15\fs16\ul 04) \up0 \expndtw0\charscalex118 (x(t),\ul0\nosuper sub\cf8\f9\fs16 t)\par\pard\column \ql \li63\sb0\sl-144\slmult0\tx2958 \up0 \ex pndtw0\charscalex118 \ul0\nosupersub\cf3\f4\fs16 > \ul0\nosupersub\cf8\f9\fs16 t o\} \ul0\nosupersub\cf3\f4\fs16 satisfying\tab \up0 \expndtw0\charscalex119 \ul0 \nosupersub\cf6\f7\fs12 from the Chinese University of Hong Kong, Hong\par\pard\ qj \li2948\ri1347\sb17\sl-158\slmult0\fi4 \up0 \expndtw0\charscalex121 Kong, in 1980, and the M.S. and Ph.D. degrees in \line \up0 \expndtw0\charscalex116 compu ter science from the University of California,\par\pard\ql \li20\sb0\sl-148\slmu lt0\tx323\tx2953 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf14\f15\fs16\ul a\ul0\tab \up0 \expndtw0\charscalex119 \ul0\nosupersub\cf14\f15\fs16\ul x(t), t) \ul0\tab \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf6\f7\fs12 Los Angeles, i n \up0 \expndtw0\charscalex112 1982 and \up0 \expndtw0\charscalex113 1985, resp ectively.\par\pard\sect\sectd\sbknone\cols2\colno1\colw3780\colsr160\colno2\colw 8160\colsr160\ql \li2313\sb50\sl-195\slmult0\tx3383 \up0 \expndtw0\charscalex113

\ul0\nosupersub\cf8\f9\fs16 dt\tab \up0 \expndtw-3\charscalex100 \ul0\nosupersu b\cf34\f35\fs20 at\par\pard\column \ql \li20\sb0\sl-144\slmult0\tx1000\tx3961 \u p0 \expndtw0\charscalex120 \ul0\nosupersub\cf8\f9\fs16 + v(x(t), \ul0\nosupersub \cf14\f15\fs16\ul t)\ul0\tab \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf14\f 15\fs16\ul iN\ul0\tab \up0 \expndtw0\charscalex120 \ul0\nosupersub\cf6\f7\fs12 H e attended UCLA under an exchange program\par\pard\ql \li1192\sb0\sl-160\slmult0 \up0 \expndtw-1\charscalex100 \ul0\nosupersub\cf34\f35\fs20 ax\par\pard\ql \li3 817\sb0\sl-108\slmult0 \up0 \expndtw0\charscalex117 \ul0\nosupersub\cf6\f7\fs12 between the Chinese University of Hong Kong and\par\pard\sect\sectd\sbknone\cols 2\colno1\colw7577\colsr160\colno2\colw4363\colsr160\ql \li1612\sb0\sl-144\slmult 0 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf3\f4\fs16 Putting (Al) into the above equation now yields\par\pard\ql \li1896\sb75\sl-184\slmult0\tx4540 \up0 \ expndtw0\charscalex121 \ul0\nosupersub\cf14\f15\fs16\ul d4)(x(t),t)\ul0\tab \up0 \expndtw0\charscalex132 \ul0\nosupersub\cf14\f15\fs16\ul av(x,t)\par\pard\colum n \qj \li20\ri1347\sb0\sl-129\slmult0 \up0 \expndtw0\charscalex121 \ul0\nosupers ub\cf6\f7\fs12 the University of California. At UCLA, he served \line \up0 \expn dtw0\charscalex123 as a teaching assistant from \up0 \expndtw0\charscalex117 198 1 to \up0 \expndtw0\charscalex122 1983 and a \line \up0 \expndtw0\charscalex117 postgraduate research engineer from \up0 \expndtw0\charscalex112 1983 to \up0 \e xpndtw0\charscalex105 1985.\par\pard\sect\sectd\sbknone\cols3\colno1\colw2724\co lsr40\colno2\colw3345\colsr160\colno3\colw5841\colsr160\ql \li2217\sb71\sl-184\s lmult0 \up0 \expndtw0\charscalex109 \ul0\nosupersub\cf8\f9\fs16 dt\par\pard\ql \ li1612\sb123\sl-184\slmult0 \up0 \expndtw0\charscalex114 \ul0\nosupersub\cf3\f4\ fs16 Hence for all t\par\pard\column \ql \li20\sb0\sl-144\slmult0\tx381\tx904\tx 2396 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf8\f9\fs16 =\tab \up0 \expndt w0\charscalex101 [A +\tab \up0 \expndtw0\charscalex118 + 0(x, t) +\tab \up0 \exp ndtw0\charscalex107 \ul0\nosupersub\cf14\f15\fs16\ul ](1.(x,\ul0\nosupersub\cf8\ f9\fs16 t).\par\pard\ql \li1983\sb0\sl-128\slmult0 \up0 \expndtw-1\charscalex10 0 a x\par\pard\ql \li59\sb112\sl-184\slmult0 \up0 \expndtw0\charscalex120 \ul0\n osupersub\cf3\f4\fs16 > to, we have\par\pard\column \qj \li20\ri1346\sb0\sl-156\ slmult0 \up0 \expndtw0\charscalex112 \ul0\nosupersub\cf6\f7\fs12 Since \up0 \ex pndtw0\charscalex129 1986, he has been working in the Teletrafic Theory and Syst em \line \up0 \expndtw0\charscalex126 Performance Department at AT&T Bell Labora tories, Holmdel, NJ. His \line \up0 \expndtw0\charscalex124 research interests i nclude stochastic modeling, distributed processing and \line \up0 \expndtw0\char scalex117 databases, and wireless and computer communication networks.\par\pard\ sect\sectd\sbknone\cols4\colno1\colw2412\colsr40\colno2\colw1417\colsr110\colno3 \colw1093\colsr160\colno4\colw6888\colsr160\ql \li1699\sb0\sl-184\slmult0 \par\p ard\ql \li1699\sb25\sl-184\slmult0 \up0 \expndtw0\charscalex114 \ul0\nosupersub\ cf8\f9\fs16 (D(x(t),t)\par\pard\column \ql \li2452\sb0\sl-184\slmult0 \par\pard\ ql \li20\sb25\sl-184\slmult0 \up0 \expndtw0\charscalex110 \ul0\nosupersub\cf3\f4 \fs16 = (I)(x(to), to) exp\par\pard\column \ql \li3969\sb0\sl-184\slmult0 \par\p ard\ql \li20\sb25\sl-184\slmult0 \up0 \expndtw0\charscalex152 \ul0\nosupersub\cf 5\f6\fs14 \u8212? \ul0\nosupersub\cf8\f9\fs16 f \ul0\nosupersub\cf3\f4\fs16 [\pa r\pard\column \ql \li4223\sb0\sl-161\slmult0 \par\pard\ql \li20\sb67\sl-161\slmu lt0 \up0 \expndtw0\charscalex131 \ul0\nosupersub\cf4\f5\fs14 (x(s), a)\par\pard\ sect\sectd\sbknone\cols3\colno1\colw4399\colsr60\colno2\colw3128\colsr160\colno3 \colw4363\colsr160\ql \li4223\sb2\sl-161\slmult0 \up0 \expndtw-6\charscalex100 t o\par\pard\column \ql \li20\sb0\sl-198\slmult0 \up0 \expndtw-9\charscalex81 \ul0 \nosupersub\cf17\f18\fs22 A +>L \ul0\nosupersub\cf4\f5\fs14 +\par\pard\ql \li217 \sb16\sl-161\slmult0\tx418 \up0 \expndtw-2\charscalex100 \ul0\nosupersub\cf36\f3 7\fs14\ul a\ul0\tab \up0 \expndtw0\charscalex140 \ul0\nosupersub\cf36\f37\fs14\u l (x(s),\par\pard\ql \li58\sb1\sl-128\slmult0\tx322\tx884 \up0 \expndtw0\charsca lex120 \ul0\nosupersub\cf4\f5\fs14 +\tab \up0 \expndtw-2\charscalex100 \ul0\nosu persub\cf36\f37\fs14\ul v\ul0\tab \up0 \expndtw0\charscalex193 \ul0\nosupersub\c f4\f5\fs14 slds\}\par\pard\column \ql \li4958\sb0\sl-138\slmult0 \par\pard\ql \l i4958\sb0\sl-138\slmult0 \par\pard\ql \li20\sb23\sl-138\slmult0 \up0 \expndtw0\c harscalex127 \ul0\nosupersub\cf6\f7\fs12 William A. Massey received the A.B. deg ree in\par\pard\sect\sectd\sbknone\cols2\colno1\colw6099\colsr160\colno2\colw584 1\colsr160\ql \li4958\sb0\sl-144\slmult0 \up0 \expndtw0\charscalex124 \ul0\nosup

ersub\cf8\f9\fs16 ax\par\pard\ql \li1612\sb113\sl-184\slmult0\tx4962\tx5294 \up0 \expndtw0\charscalex132 \ul0\nosupersub\cf3\f4\fs16 From this, it follows that \up0 \expndtw0\charscalex105 (I)(x(to), to)\tab \up0 \expndtw0\charscalex120 =\ tab \up0 \expndtw0\charscalex122 0 implies\par\pard\ql \li1617\sb28\sl-184\slmul t0 \up0 \expndtw0\charscalex56 \ul0\nosupersub\cf8\f9\fs16 (1) \up0 \expndtw0\ch arscalex131 (x(t),t) \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf3\f4\fs16 = 0 for all \ul0\nosupersub\cf8\f9\fs16 t \ul0\nosupersub\cf3\f4\fs16 > \ul0\nosup ersub\cf8\f9\fs16 to, \ul0\nosupersub\cf3\f4\fs16 which proves the lemma.\par\pa rd\ql \li3110\sb0\sl-138\slmult0 \par\pard\ql \li3110\sb0\sl-138\slmult0 \par\pa rd\ql \li3110\sb34\sl-138\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\c f6\f7\fs12 ACKNOWLEDGMENT\par\pard\ql \li1785\sb112\sl-184\slmult0 \up0 \expndtw 0\charscalex111 \ul0\nosupersub\cf3\f4\fs16 The authors thank Y. Levy for his he lpful comments.\par\pard\ql \li1699\sb0\sl-138\slmult0 \par\pard\ql \li3374\sb11 4\sl-138\slmult0 \up0 \expndtw0\charscalex116 \ul0\nosupersub\cf6\f7\fs12 REFERE NCES\par\pard\ql \li1699\sb0\sl-138\slmult0 \par\pard\ql \li1699\sb65\sl-138\slm ult0\tx1938 \up0 \expndtw0\charscalex117 [I]\tab \up0 \expndtw0\charscalex119 R. Bulirsch and J. Stoer, "Numerical treatment of ordinary differential\par\pard\q l \li1943\sb1\sl-133\slmult0 \up0 \expndtw0\charscalex118 equations by extrapola tion methods," \ul0\nosupersub\cf27\f28\fs12 Numerische Mathematik, \ul0\nosuper sub\cf6\f7\fs12 vol. \up0 \expndtw0\charscalex104 8,\par\pard\ql \li1943\sb22\sl -138\slmult0\tx2203 \up0 \expndtw0\charscalex109 pp.\tab \up0 \expndtw0\charscal ex112 1-13, \up0 \expndtw0\charscalex105 1966.\par\pard\qj \li1699\ri11\sb0\sl161\slmult0\tx1948 \up0 \expndtw0\charscalex116 [2] R. Haberman, \ul0\nosupersub \cf27\f28\fs12 Mathematical Models: Mechanical Vibrations, Population \line\tab \up0 \expndtw0\charscalex125 Dynamics, and Traffic Flow. \ul0\nosupersub\cf6\f7\ fs12 Englewood Cliffs, NJ: Prentice-Hall,\par\pard\ql \li1963\sb9\sl-138\slmult0 \up0 \expndtw0\charscalex104 1977.\par\pard\qj \li1699\ri16\sb0\sl-153\slmult0\ tx1953 \up0 \expndtw0\charscalex117 13] D. L. Jagerman, "Nonstationaty blocking in telephone \ul0\nosupersub\cf27\f28\fs12 trafic," Bell Syst. \line\tab \up0 \e xpndtw0\charscalex115 Tech. J., \ul0\nosupersub\cf6\f7\fs12 vol. \up0 \expndtw0\ charscalex118 54, pp. \up0 \expndtw0\charscalex118 625-661, \up0 \expndtw0\chars calex105 1975.\par\pard\qj \li1699\ri6\sb0\sl-154\slmult0\tx1948 \up0 \expndtw0\ charscalex126 [4] W. C. Y. Lee, \ul0\nosupersub\cf27\f28\fs12 Mobile Cellular Te lecommunications Systems. \ul0\nosupersub\cf6\f7\fs12 New \line \tab \up0 \expnd tw0\charscalex114 York: McGraw-Hill, \up0 \expndtw0\charscalex102 1989.\par\pard \ql \li1699\sb8\sl-138\slmult0\tx2433 \up0 \expndtw0\charscalex100 [5]\tab \up0 \expndtw0\charscalex122 \ul0\nosupersub\cf27\f28\fs12 Mobile Communications Desi gn Fundamentals. \ul0\nosupersub\cf6\f7\fs12 New York:\par\pard\ql \li1943\sb20\ sl-138\slmult0 \up0 \expndtw0\charscalex111 Wiley, \up0 \expndtw0\charscalex104 1993.\par\pard\qj \li1699\ri6\sb0\sl-156\slmult0\tx1943 \up0 \expndtw0\charscal ex121 [6] W. A. Massey and W. Whitt, "Networks of ininite server queues with \li ne\tab \up0 \expndtw0\charscalex122 nonstationary Poisson input,- \ul0\nosupersu b\cf27\f28\fs12 Queueing Syst., \ul0\nosupersub\cf6\f7\fs12 vol. \up0 \expndtw0\ charscalex118 13, pp. \up0 \expndtw0\charscalex116 183-250,\par\pard\ql \li1963\ sb17\sl-138\slmult0 \up0 \expndtw0\charscalex104 1993.\par\pard\ql \li1699\sb5\s l-138\slmult0\tx2423 \up0 \expndtw0\charscalex100 [7]\tab \up0 \expndtw0\charsca lex125 "A stochastic model to capture space and time dynamics in\par\pard\ql \li 1703\ri7\sb22\sl-144\slmult0\fi244 \up0 \expndtw0\charscalex118 wireless communi cation systems," Prob. Eng. Inf. Sci., to appear. \line \up0 \expndtw0\charscale x124 [8] K. S. Meier-Hellsten, E. Alonso, and D. R. O'Neil, "The use of SS7\par\ pard\ql \li1953\sb19\sl-138\slmult0 \up0 \expndtw0\charscalex117 and GSM to supp ort high density personal communication systems," in\par\pard\ql \li1953\sb20\sl -138\slmult0 \up0 \expndtw0\charscalex113 \ul0\nosupersub\cf27\f28\fs12 Proc. Th ird WINLAB Workshop on Third Generation Wireless Networks,\par\pard\ql \li1948\s b21\sl-138\slmult0\tx2284 \up0 \expndtw0\charscalex108 \ul0\nosupersub\cf6\f7\fs 12 Apr.\tab \up0 \expndtw0\charscalex102 1992.\par\pard\column \qj \li1498\ri134 2\sb0\sl-143\slmult0\fi14 \up0 \expndtw0\charscalex126 1977 from Princeton Unive rsity in mathematics \line \up0 \expndtw0\charscalex123 (Magna Cum Laude, Phi Be ta Kappa, and Sigma \line \up0 \expndtw0\charscalex123 Xi). In \up0 \expndtw0\ch arscalex124 1981, he received the Ph.D. degree from \line \up0 \expndtw0\charsca

lex116 Stanford University in mathematics.\par\pard\qj \li1498\ri1342\sb0\sl-158 \slmult0\fi143 \up0 \expndtw0\charscalex110 Since \up0 \expndtw0\charscalex121 1981, he has been a member of technical \line \up0 \expndtw0\charscalex117 staff in the Mathematical Sciences Research Center \line \up0 \expndtw0\charscalex119 at AT&T Bell Laboratories. He has written papers \line \up0 \expndtw0\charscale x114 on nonstationary queues, stochastic ordering, queue-\line \up0 \expndtw0\ch arscalex114 ing networks, database theory, and wireless commu-\line \up0 \expndt w0\charscalex121 nications. His research interests include queueing\par\pard\qj \li20\ri1346\sb2\sl-153\slmult0 \up0 \expndtw0\charscalex118 theory, applied pro bability, performance modeling, telecommunications, and \up0 \expndtw0\charscale x116 special functions.\par\pard\qj \li20\ri1346\sb5\sl-153\slmult0\fi139 \up0 \ expndtw0\charscalex122 Dr. Massey is a member of the American Mathematical Socie ty and the \up0 \expndtw0\charscalex118 Operations Research Society of America.\ par\pard\qj \li6259\sb0\sl-158\slmult0 \par\pard\qj \li6259\sb0\sl-158\slmult0 \ par\pard\qj \li6259\sb0\sl-158\slmult0 \par\pard\qj \li6259\sb0\sl-158\slmult0 \ par\pard\qj \li1503\ri1346\sb46\sl-158\slmult0 \up0 \expndtw0\charscalex127 Ward Whitt received the A.B. degree in mathe-\line \up0 \expndtw0\charscalex123 mati cs from Dartmouth College in \up0 \expndtw0\charscalex122 1964 and the \line \up 0 \expndtw0\charscalex124 Ph.D. degree in operations research from Conell \line \up0 \expndtw0\charscalex117 University in \up0 \expndtw0\charscalex105 1969.\pa r\pard\qj \li1503\ri1342\sb0\sl-158\slmult0\fi139 \up0 \expndtw0\charscalex122 H e taught in the Department of Operations Re-\line \up0 \expndtw0\charscalex124 s earch at Stanford University in \up0 \expndtw0\charscalex119 1968-1969, and \lin e \up0 \expndtw0\charscalex125 in the Department of Administrative Sciences at \ line \up0 \expndtw0\charscalex120 Yale University from \up0 \expndtw0\charscalex 119 1969 to \up0 \expndtw0\charscalex117 1977. Since \up0 \expndtw0\charscalex10 4 1977, \line \up0 \expndtw0\charscalex117 he has been employed by AT&T Bell Lab oratories. \line \up0 \expndtw0\charscalex126 He currently works in the Mathemat ical \ul0\nosupersub\cf1\f2\fs10 Sciences \line \up0 \expndtw0\charscalex127 \ul 0\nosupersub\cf6\f7\fs12 Research Center, Murray Hill, NJ. His research\par\pard \qj \li20\ri1346\sb0\sl-156\slmult0 \up0 \expndtw0\charscalex120 interests inclu de queueing theory, stochastic processes, stochastic models in \up0 \expndtw0\ch arscalex117 telecommunication and numerical inversion of transforms.\par\pard\qj \li20\ri1346\sb10\sl-153\slmult0\fi139 \up0 \expndtw0\charscalex118 Dr. Whitt i s a member of the Operations Research Society of America and \up0 \expndtw0\char scalex119 the Institute of Mathematical Statistics. \par\pard\sect\sectd\fs24}