Vous êtes sur la page 1sur 5

On the cooperation between 3G and 2G systems

L. Sartori, S-E. Elayoubi, F. Morlot and B. Fouresti e Orange Labs 38-40 rue du G n ral Leclerc - 92130 Issy Les Moulineaux - France e e Phone +33 1 45 29 88 67 - Fax +33 1 45 29 63 07 {luca.sartori, salaheddine.elayoubi, frederic.morlot, benoit.fourestie}@orange-ftgroup.com

AbstractIn this work, we study cooperation between UMTS and GSM systems. The aim is to nd the best Joint Radio Resource Management (JRRM) strategies that achieve lowest blocking rates. We rst present analytical models for capacity in GERAN and UTRAN. We then consider different strategies of Joint Radio Resource Management (JRRM) with or without inter-system (vertical) handovers and show how to calculate the steady-state probabilities and the performance measures (blocking probabilities, loads). Our numerical results compare the different JRRM strategies and point out the advantages and drawbacks for each of them. We show that the best performance is obtained when authorising vertical handovers. However, a simple strategy without vertical handover may, in some cases, achieve comparable performance with less signaling overhead.

I. I NTRODUCTION The presence of two or more radio network systems in the same area is becoming more and more frequent as fourth generation systems will be composed of heterogeneous networks [1]. Various interworking and handover strategies have been presented in the literature for the interworking between 3G and WLAN systems [4][2][3]. The reason for this large interest in the combination of UMTS and WLAN systems is that they are viewed as complementary since the former provides wide-area coverage with high mobility and medium rates, while the latter provides local coverage and low mobility associated with high data rates. However, less attention has been given to the cooperation between 2G and 3G systems, although these networks already coexist. In fact, the majority of operators deploying 3G networks have legacy 2G networks and will still exploit these two networks for the next years. It is then interesting for the operators to integrate 2G systems with 3G ones to provide users with a better service. In this work, we model and analyse a network composed of the cooperation of UTRAN and GERAN systems. We consider co-localised base stations as it is likely to reduce deployment costs. Before modeling the cooperative system, we must rst have reliable models for the individual systems. To our knowledge, most of the 3G models consider a simplistic model where the inter-cell interference is proportional to the (instantaneous) intra-cell interference, and where no distinction is made between cell-centre users (where intra-cell interference is preponderant) and cell-edge ones (where inter-cell interference is very high). We then develop a model that models accurately the inter-cell interference, and differentiate between users following their positions in the cell. We also present an

algorithm that calculates accurately the load in the cell taking into account feedback between cells due to interference. Having modeled the individual systems, we are able to study the cooperative system in order to compare different strategies of Joint Radio Resource Management (JRRM) including joint admission control and vertical handover. We rst consider a strategy based on the system at the user arrival and which depends on the state of each system load. In this case, once the selection is made, the user cannot be transfered to the other system even if the network conditions in the chosen system become worse. Secondly, we consider a more complete JRRM policy where the rst choice depends on the system, but vertical (inter-system) handovers are allowed at call departures or arrivals. Note that we consider several regions in the cell, as the propagation conditions are different for the two systems, leading to different behaviors for cell-edge and cell-centre users in both systems. For each of the considered JRRM strategies we construct the corresponding Markovian model. We then nd several performance measures, namely the global blocking probability and the loads in each system. The remainder of this paper is organised as follows. Section II presents the UTRAN model while section III presents the GSM model. Several JRRM strategies for the cooperative network are analysed in section IV. These strategies are compared by the numerical results in section V. Finally section VI concludes the paper. II. UMTS MODEL In the downlink of WCDMA systems, the SINR achieved for user k situated at distance r0 from the target cell base station is given by:
U SIN Rk (r0 ) = Sk
U Iinter,k

Pk,0 /qk,0 , + Iintra,k (r0 ) + N0

(1)

where Sk is the spreading factor (normally greater than 16), Pk,0 is the power received by user k from the target cell base station, Iinter,k = l=0 Ptot,l /qk,l is the inter-cell interference due to the base station k, with Ptot,l the total power transmitted by the base station l and qk,l the pathloss between user k and the base station l, Iintra,k (r0 ) = (Ptot,0 Pk,0 )/qk,0 is the intra-cell interference originated from the common channels and from other users, with the orthogonality factor, and N0 is the thermal noise. Differently from other works (e.g. see [6]) in which the powers in the interfering cells equal the instantaneous power

1-4244-0264-6/07/$25.00 2007 IEEE

in the target cell, here we consider that such an equality holds only for the mean power and it is independent of the instantaneous uctuations of Ptot,0 . Here the total power of the base station l can be written as Ptot,l = U Pmax , where U is the average load in a typical cell of the system and Pmax is the maximal transmission power. Let us notice that the 3G mean load is dened as the ratio between the used and the total power. As the inter-cell interference can be expressed in terms of the well-known F-factor Fk (see [5] and [7]), eqn. (1) can be rewritten in the following way:
U SIN Rk = Sk

III. GSM MODEL Radio resources are allocated in GERAN on both time and frequency bases. Given a total set of Ntot frequencies, we can partition them amongst a number of cells to cover a certain area, in a frequency reuse larger than 1. In general, we provide each base station with NT RX transmitters (TRXs) which, in the case of synthetised frequency hopping, is smaller than the number of frequencies in each cell. The allocation of TRXs and frequencies to the cells is done so that inter-cell interference is minimised. A frequency reuse factor of 3 or even 7 is then used in order to increase SINR at cells edge. We now assume that each mobile uses randomly all the frequencies in its cell. During a call the frequency hopping consists of changing frequency while passing from one time slot (TS) to another(7 TSs out of 8 TSs carry trafc). Hereafter we will denote with NGSM the maximal number of TSs reserved for 2G voice calls. The latter corresponds also to the maximal number of GSM calls in the system. In order to keep the same model as for 3G, we divide the cell into n regions and assume that in each of them there are MiG GSM calls, for i = 1, . . . , n. Thus, the number of TSs occupied by all the GSM calls is
n

Pk,0 , (2) (Ptot,0 Pk,0 ) + U Pmax Fk + N0 qk,0


U SIN Rk , Sk + SIN RU k

whence, after dening k :=

(3)

one obtains k = Pk,0 . Ptot,0 + U Pmax Fk + N0 qk,0


(4)

If there are M U UMTS voice users in the cell, the total power used from the base station in the target cell is Ptot,0 = PCCH +
M k=1
U

k (U Pmax Fk + N0 qk,0 ) , 1 M U k

N T S (MG ) =

MiG ,
i=1

(5)
G G G (M1 , . . . , Mn )

where the required SINR for UMTS voice calls is equal to 11 dB. Here PCCH denotes the power associated with the Common Channels. Differentiating n regions in the cell (for instance, two regions regrouping cell-edge users in region 1 and cell-centre users in region 2), and considering MiU UMTS users in each region i yields: Ptot,0 = PCCH + k

where M = and the maximal number of GSM calls in region i, knowing the number of calls in the other regions, is:
G G G G MiG,max (M1 , ..., Mi1 , Mi+1 , ..., Mn ) = NGSM k=i G Mk

n U i=1 ( Pmax Fi + n k i=1 MiU

N0 qi )MiU

(6)

(9) The load of the cell G is dened as the ratio between the mean number of allocated TSs per frame and the number of all the available TSs: G (MG ) =
TS N T S (MG ) + Nlog (NT RX ) , 8NT RX

q where (F1 , ..., Fn ) and (1 , ..., qn ) are respectively the average F-factors and path-losses over each different region. As the total used power will be never greater than the maximal power, the maximal number of UMTS calls is obtained from Ptot,0 = Pmax using eqn. (6):
n

(10)

TS where Nlog is the number of TSs allocated to the logical channel and depends on the number of transmitters.

IV. C OOPERATIVE MODEL We analyse now the performance of a cooperative network composed of GSM and UMTS, based on the previous models for these systems. We will consider four different possible strategies of JRRM: 1) a rst strategy that allocates calls with priority to GSM. Here a new arrival in the network is rstly accepted by GSM; then, once the latter is fully loaded, the calls are processed by UMTS. However, we aim also at minimising the number of cell-edge calls in 3G, as these calls are more subject to interference. We then allow vertical handovers in order to redirect, from 3G to 2G, a cell-edge user that had been accepted in 3G due to a lack of resources in 2G. This can happen either when a 2G call terminates or when a cell-centre 3G call departs and is replaced by a cell-centre 2G call, making place

k
i=1

(U Pmax Fi + N0 qi + Pmax )MiU Pmax PCCH , U

that is, the maximal number of UMTS users that can be accepted in the region i, knowing the number of UMTS calls in the other regions, is given by eqn. (7), at the top of next page, where x is the greatest integer less than x. Load As 3G system is interference-limited, the load is dened as the ratio between the emitted and the maximal power. Thus, using eqn. (6) and knowing the number of calls in the cell, one nds the following expression of the 3G load: U (MU ) =
n PCCH + k i=1 (U Pmax Fi + N0 qi )MiU n U) Pmax (1 k i=1 Mi

(8)

U U where MU = (M1 , ..., Mn ).

U U U U MiU,max (M1 , ..., Mi1 , Mi+1 , ..., Mn ) =

Pmax PCCH v
v (U P

U + N0 qk + Pmax )Mk 2 + N0 q2 + Pmax ) max F

U k=i ( Pmax Fk

(7)

for the cell-edge call in 2G. Note that this policy allows a large number of inter-system vertical handovers. 2) A second strategy analogous to the rst one except for giving priority to UMTS and deploying GSM only when the rst is full.Vertical handovers are also allowed to maximise the number of cell-centre users in 3G as they experience higher inter-cell interference. 3) A third strategy resembles the rst one but it does not allow vertical handovers: a call starting in GSM terminates there and only new arriving users encountering the 2G system full will be served by UMTS. 4) Finally, the fourth strategy describes a scenario similar to that in the third one with priority given to 3G and where no vertical handover is admitted. Let us notice that vertical handover aims at using the available space in one system due to call departures from the other one. As vertical handover is allowed to keep the best possible conditions over the whole communication duration, one can then expect better performance from the systems deploying the rst and the second strategy. A. System-priority strategies with vertical handover 1) Markovian model: we suppose that call durations are exponentially distributed, and calls arrive to the cell according to a Poisson process of intensity . When the strategies with vertical handovers are used, the state of the system can be fully described by the number of users in each region, M = (M1 , . . . , Mn ) with Mi = MiG + MiU in each region i, whilst the distribution of users in each system (2G or 3G) can be derived for each conguration, using the dened JRRM strategies. For example, by applying the rst selection strategy one obtains MiG := Mi MiG,max if Mi MiG,max otherwise.

load via the maximal number of UMTS calls in each region. The steady-state distribution is calculated by solving Q = 0 such that e = 1, (12)

where the latter is a normalisation condition with a vector of ones e of the same dimension as which is dened on A. 2) Performance measures: rst, the blocking probability in region i is given by
block i = MBi

(M)

with

Bi := {M A|M = (M1 , . . . , Mimax , . . . , Mn )}


Mimax = MiU,max + MiG,max where MiU,max and MiG,max are obtained using eqns. (7) and (9). The blocking probability over the whole cell is
n

block =
i=1

block p i i .

(13)

Moreover, we can now calculate the mean loads. The mean 2G load is readily obtained using the steady-state distribution as G =
MA G G G M1 (M), . . . , Mn (M) (M),

(14)

and MiU = Mi MiG . Let A be the set of all the possible states of the cooperative network. The system can be described by a Markov chain and the transition matrix Q associated with it has transition rates described in eqn. (11) on the top of the next page, where is the mean call duration, pi is the proportion of new arrivals at the region i and is the arrival rate. Let us notice that this model does not account for mobility within the cell: a user starting its communication in region i terminates it there. Mobility can be simply introduced in the model, by adding additional transitions between states, corresponding to call migration between regions, as it has been done for 3G systems in [8]. The generator-matrix associated with the aforementioned transition probabilities Q is used to nd the steady-state probabilities. Note that this matrix depends on the mean 3G

where G is given by eqn. (10). As for the 3G mean load, its calculation is more complicated, because the load in the target cell depends on the interference from adjacent cells, which, in turn, is a function of the load in the adjacent cells. For a homogeneous network, i.e. where the mean load is the same in each cell, we propose to calculate the steady-state probabilities solving eqn. (12) with the following iterative algorithm: 1) set the initial value for the mean 3G load iterations, e.g. taking the mean 3G load equal to 0.5; 2) calculate (M) deploying the aforementioned initial value of the mean load to be used along with eqn. (8) (right-hand side) to deduce a new mean load (left-hand side of eqn. (8)); 3) use this new values of the mean loads in the generatormatrix and repeat the iterations until the sequence of the mean load converges. The convergence of this algorithm is insured by the Fixed Point theorem. B. System-priority strategies without vertical handover For the sake of comparison we analyse the performance of the two strategies based on 2G and 3G system priority when no inter-system handover is allowed. In these strategies, knowing the number of calls in each region does not lead automatically to the distribution of calls within the two systems (2G and 3G),

P (M1 , . . . , Mi , . . . , Mn M1 , . . . , Mi 1, . . . , Mn ) = Mi P (M1 , . . . , Mi , . . . , Mn M1 , . . . , Mi + 1, . . . , Mn ) = pi

(11)

since vertical handovers are not allowed to redistribute calls as above. We thus need to analyse the system with 2 n variables, corresponding to the numbers of 2G or 3G calls in each of the n regions. This leads to a large transition matrix and a huge number of balance equations to solve. However, the fact that no vertical handover is possible simplies the analysis, the moment the two systems can be analysed as two queuing systems where the entry of the second is equal to the calls blocked from the rst. This is illustrated in g. 1 when calls are rst accepted in 2G. In each of these policies, the call dynamics can then be modeled as a two-Markov chain sequence. Note that this is only an approximation: the blocking rate of the rst queue is not Poissonian, thus the arrivals at the second queue consist of a bursty process rather than a Poisson one. However, this still represents an acceptable approximation of the average performance. In the case of 2G-

H=

MU A

(bG pi /)Mi MiU ! U i=1

The total blocking is reached when the maximal number of UMTS is attained. As for the 3G-priority strategy, we adopt the same model exchanging GSM for UMTS and using the GSM blocking probability as the total blocking probability. More explicitly: (MU ) =
H=

1 H

(pi /)Mi , MiU ! i=1


n
U

(17)

MU A

(pi /)Mi , MiU ! U i=1

with AU the set of all the possible 3G states, and, after calculating the blocking probabilities bU , ..., bU in the n regions, n 1 we obtain: (MG ) =
K=
n i=1 G Mi NGSM

1 K

(bU /)Mi i , MiG ! i=1


(bU /)Mi i MiG ! i=1
n
G

(18)

The mean 2G and 3G loads are calculated knowing the steady-state distributions as in the previous two policies. V. N UMERICAL APPLICATIONS In order to evaluate the user QoS, we show here some numerical results in which the blocking probabilities and the loads are used in order to compare and explain the behaviour of the different JRRM strategies. Here we assume that there are only two regions in the cell: an inner one around to the cell centre with a proportion of new arrivals equal to 0.3 and an outer one limited by the cell edge with a proportion of new arrivals equal to 0.7. We consider 4 TRXs for the GSM system while for UMTS we x the Node B maximal power at 20 W, the power of the Common Channels at 4.2 W, the chip rate at 3.84 Mc/sec and the thermal noise at 1210.68 dB. With this set of parameters the time of convergence for the algorithm presented in section IV-A2 does not exceed 5 min. The blocking rates for the four strategies are plotted in g. 2. Let us rst focus on the strategies with vertical handover. The rst observation is that blocking probability is lower when priority is given to 2G. This is due to the fact that, when 2G is lled rst, 2G load is high (g. 3), and 3G load is low (g. 4). And as the 3G system is interference-limited, lower loaded 3G cells generate less interference, leading to a better 3G network capacity. However, when no vertical handover is possible, the trend is inverted: lling 3G system rst enhances overall capacity. The explanation is simple: when calls are originally directed to

Fig. 1. Model of the system with priority to 2G without vertical handovers. Here, only calls that are blocked in 2G are redirected to 3G, with intensity bG , bG being the blocking in 2G.

priority, we denote with M G a typical GSM state. The steadystate probabilities in GSM are then calculated as solutions of a classical Erlang-B formula with arrival rate and departure rate as follows: (M G ) =
K=
M G NGSM

1 (/)M , K M G!
(/)M M G!
G

(15)

Once the GSM blocking bG is obtained by the following equation: bG = (NGSM ), the steady-state probabilities for UMTS MU are obtained solving another Birth-and-Death model in the following way: (MU ) = 1 H
(bG pi /)Mi , MiU ! i=1
n
U

(16)

3G, a natural selection is performed, as cell-edge users, more subject to interference, will be blocked more often than cellcenter ones, and do not thus generate excessive interference in adjacent cells. However, when 2G system is lled rst, all calls will be treated alike as 2G is less sensitive to interference; 3G receives then equal arrival rates in the two regions (redirected from 2G), and cell-edge users will be blocked leading to degraded capacity. Let us now compare the strategies with and without vertical handover. We can see that, when 3G is lled rst, the strategies with or without handover have almost the same performance (with a small advantage to the one without vertical handover). However, the strategy that gives priority to 2G with vertical handovers outperforms the others, as explained above. To conclude, we suggest the 2G system is given priority when vertical handovers are possible in order to continuously limit the number of cell-edge users in 3G. If vertical handover algorithms are not yet implemented in the network, we recommend to always begin by lling the 3G system and redirect blocked calls to 2G.

0.9

0.8

3G Load

0.7

0.6

0.5
3G priority handover 2G priority handover 3G priority no handover 2G priority no handover

0.4

0.22

0.24

0.26 0.28 0.32 0.3 Global arrival rate (calls/sec)

0.34

0.36

Fig. 4.

3G cell load in the four strategies.

VI. C ONCLUSION In this paper, we developed models for the capacity of a mobile network where 2G and 3G systems cooperate. Different system selection rules and JRRM strategies were presented, with and without vertical handovers between systems. We began by developing an analytical model for capacity and load in 3G, and then developed Markovian models that describe the evolution of the network for each of the studied policies. Our numerical results show that the strategy that gives priority to 2G, with vertical handover in order to redirect cell-edge users to 3G gives the best performance (lower blocking rates). These results were obtained considering that the handovers were ideal: no delays nor imperfections were introduced. The obtained performance can then be viewed as an upper limit of the performance of the system. When vertical handover algorithms are not implemented, or have many imperfections (delay and loss), we advocate using another policy where we begin by lling 3G and redirect blocked calls to 2G. R EFERENCES
[1] Y. Raivio, 4G - Hype or Reality, IEEE 3G Mobile Communication Technologies, March 2002. [2] Q. Song, A. Jamalipour, Network Selection in an Integrated Wireless LAN and UMTS Environment Using Mathematical Modeling and Computing Techniques, IEEE Wireless Communications, Volume 12, Issue 3, June 2005 Page(s):42-48. [3] K. Ahmavaara, H. Haverinen and R. Pichna, Interworking architecture between 3GPP and WLAN systems, IEEE Communications Magazine 41 (11) (2003) 74-81. [4] D. Chen, X. Wang, A.K. Elhakeem, Performance Analysis of UMTS Handover with the Help of WLAN, Second International Conference on Quality of Service in Heterogeneous Wired/Wireless Networks (QShine05), 22-24 Aug. 2005. [5] H. Holma and A. Toskala, WCDMA for UMTS: Radio access for third generation mobile communications, John Wiley & Sons, 3rd Ed. (2004). [6] I. Koukoutsidis, E. Altman, J.M. Kelif, A non-homogeneous QBD approach for the admission and GoS control in a multiservice WCDMA system, INRIA Research Report No. RR-5358 (2004). [7] K. Sipila, Z. Honkasalo, J. Laiho-Steffens and A. Wackr, Estimation of Capacity and Required Transmission Power of WCDMA Downlink Based on a Downlink Pole Equation, IEEE VTC 2000. [8] S.E. Elayoubi and T. Chahed, Admission Control in the downlink of UMTS, Lecture notes on computer science LNCS 3427, Springer-Verlag, April 2005.

0.2

0.18

0.16
Blocking probability (total)

3G priority handover 2G priority handover 3G priority no handover 2G priority no handover

0.14

0.12

0.1

0.08

0.06

0.04
0.02

0 0.2

0.22

0.24 0.28 0.26 Global arrival rate (calls/sec)

0.3

0.32

Fig. 2.

Total blocking probability in the four strategies.

0.95

0.9

2G Load

0.85

0.8
3G priority handover 2G priority handover 3G priority no handover 2G priority no handover

0.75

0.7

0.65 0.22

0.24

0.28 0.26 0.3 0.32 Global arrival rate (calls/sec)

0.34

0.36

Fig. 3.

2G cell load in the four strategies.

Vous aimerez peut-être aussi