User-deployed and Capacity Evaluation of Multihop

Wireless Networks: A Case Study for Nicaragua

Marvin Sánchez G.
Dept of Communication Systems,
Royal Inst of Technology (KTH),
Electrum 418, 164 40 Kista, Sweden.
Email: marvins@radio.kth.se
Oscar Somarriba J.
Dept of Communication Systems,
Royal Inst of Technology (KTH),
Electrum 418, 164 40 Kista, Sweden.
Email: oscars@radio.kth.se
Jens Zander
Dept of Communication Systems,
Royal Inst of Technology (KTH),
Electrum 418, 164 40 Kista, Sweden.
Email: jens.zander@radio.kth.se

Abstract—In this paper we present a methodology for capac- II. M ETHOD OF A NALYSIS
ity evaluation of user-deployed multihop wireless networks in
rural areas. The capacity is described by the maximum end- The general research question of interest to be answered is:
to-end transmission rate (throughput) provided to each node • What is the capacity that a user-deployed MWN can provide
that comprises the network utilizing interference-based link subject to low-cost deployment constraint for Internet access
scheduling. A user-deployed scenario for Internet access is studied for the 13 community sites (telecenters)?
utilizing 13 sites in rural communities located in the north center
highland region of Nicaragua (telecenter candidates). To reduce For low-cost user-deployed scenario we consider the follow-
the Internet service cost, shared common access points are used ing:
in mesh configuration under asymmetric traffic demands. The • Share Internet access: for low-cost Internet service (net-
radio propagation environment is estimated utilizing the digital
map GTOPO30 and the upper-bound capacity resulting from
work service recurrent cost) nodes share access provider.
the user-deployed approach is reckoned by founding the link • Economy of scale: we assume the utilization of radio
transmission schedule using nonlinear optimization. The results equipment parameters on the 2.4GHz frequency band. In
provide useful figure-of-merits and show that MWN is a feasible Nicaragua, that frequency band is part of the allocated
alternative to provide rural communication to those communities. unlicensed radio spectrum subject to rules in AA001-2006
[2]. These rules are similar to the FCC part 15 in the USA.
To share common Internet access points in a mesh configu-
I. I NTRODUCTION ration we assume asymmetric traffic demand from each node
to a gateway node connected to the Internet and viceversa.
The user-deployed scenario is analyzed estimating the radio
Nicaragua is a developing country located in Central Amer-
propagation environment of located nodes. The path-losses
ica, composed by several urban and wide rural areas. In order
in the network are derived utilizing the digital GTOPO30
to provide telecommunication services to the poorly covered
(W100N40) [3] with the Longley-Rice model [4] as imple-
areas in the countryside, the Nicaraguan government has social
mented by the Radio Mobile freeware program by VE2DBE
programs like FITEL, National Fund for Telecommunication
[1].
Investment (by its acronym in Spanish), to provide access to
telecommunication services in low-income and rural areas of
the country.
The rural communities of interest in this study have been
provided by the regulatory body of Nicaragua, TELCOR,
through the Project Coordination Unit (UCP). The UCP is an
special unit that coordinates, proposes and evaluates project
carried out through FITEL.
The rural communities are located in the north central
highlands region of the country, in the departments of Estelı́,
Madriz and Nueva Segovia. Fig. 1 shows the geographical
locations for the communities utilizing a digital map of the
terrain.
The paper is organized as follows: in section II we present
the system models and methodology for the user-deployed
multihop wireless network and the capacity evaluation method.
The user-deployed approach and its results for the rural
communities are presented in section III. The conclusions are Fig. 1. Rural communities of Nicaragua for case study (120km × 80km).
presented in section IV. Source: Map was derived using the freeware software ”Radio Mobile” [1].
 The (upper-bound) capacity resulting from a user-deployed We assume that the average traffic from nodes to the Internet
approach is evaluated by finding the link transmission schedule is 10% of the traffic from the Internet to the nodes. We think
applying nonlinear optimization [5]. this is a reasonable assumption if for instance the gateway
is connected via an ADSL (Asymmetric Digital Subscriber
A. Analysis and User-deployed Methodology: Line) service and we would like to share this connection
providing similar service to all telecenters connected to the
We can summarize the general methodology applied in the
same gateway. In ADSL over Plain Old Telephone Service
following sections by the following steps:
(POTS), ITU G.992.1 Annex A standard, the downstream rate
I) Network Deployment: is 12Mbps and the upstream rate is 1.3Mbps (which is 10.8%
i. Determine the radio network parameters that will be of the downstream rate). With this assumption the total traffic
used. load can be rewritten by:
ii. Define the Internet gateway to be used (e.g. closest
available land-based infrastructure).
λgi + λig
P P
iii. Determine the link-budget between stations includ- λ =
∀i6=g ∀i6=g (1)
ing the gateway. = (N − 1)(λ (UL)
+ λ(DL) ) = 1.1(N − 1)λ(DL) ,
iv. Determine the current network topology.
v. If there are nodes that cannot reach the gateway, where λ(UL) is the uplink traffic (from node to the gateway),
add digital-repeater nodes (digipeaters) to extend the λ(DL) is the downlink traffic (from the Internet to node) and
connectivity by multihopping and go to step Iiii (N − 1) the number of nodes (telecenters) interconnected to
II) Capacity analysis: the Internet gateway.
i. Define the external traffic load demand from each In the following, to better explain the application of the
node to the Internet gateway and viceversa. method we utilize a simple network example.
ii. Determine the routing matrix.
C. Example of the method with a 4-nodes network
iii. From steps IIi and IIii estimate the link-traffic load.
iv. Compute the equivalent link-path routing matrix To illustrate the methodology we utilize a simple 4-nodes
Rmesh . ring network. The distance between nodes is set to 15km. For
v. Determine the network capacity by interference- this illustrative example we use a distance dependent radio
based scheduling (STDMA) propagation model to find the path gain matrix. We use the
Recreation of the user-deployment is related to step Ii to Iv parameters from table I with 6dBi omnidirectional antennas.
where users try to connect their communities to the Internet With this parameters the G matrix is given by:
gateway. At the early stage, the WMN start comprised only  
∞ - 129.8 - 134.3 - 129.8
by the nodes located in the communities. Then, progressively  - 129.8 ∞ - 129.8 - 134.3 
from the gateway towards the communities, the users add G=  [dB]
 - 134.3 - 129.8 ∞ - 129.8 
digital-repeaters to connect the network.
- 129.8 - 134.3 - 129.8 ∞
To determine the network topology we assume that the
hardware utilized has the physical layer parameters shown in
table I. Those parameters are based on utilizing equipment TABLE I
with similar physical layer to IEEE 802.11g operating on the P HYSICAL LAYER PARAMETERS
2.4GHz frequency band [6] but under FCC part 15 rules. Note
Frequency [GHz] 2.4 2.4 2.4 2.4 2.4 2.4 2.4 2.4 2.4 2.4
that the receiver sensitivity utilized is conservative as compared Data rate 1 2 6 9 12 18 24 36 48 54
(R_set)[Mbps] *
for instance with the specifications for the equipment in [7]. Modulation 'CCK' 'CCK' 'BPSK' 'BPSK' 'QPSK' 'QPSK' '16QAM' '16QAM' '64QAM' '64QAM'
Code rate 1 1 1/2 3/4 1/2 3/4 1/2 3/4 2/3 3/4
To find the path losses, we assume utilization of the first Peak output
power** [dBm]
30 30 30 30 30 30 30 30 30 30

channel on IEEE802.11g that corresponds to operate on the Transmitter

antenna gain
6 6 6 6 6 6 6 6 6 6

frequency range: 2401 MHz- 2423MHz. To determine the [dBi]

EIRP [dBm] 36 36 36 36 36 36 36 36 36 36
Receiver antenna 0 0 0 0 0 0 0 0 0 0
radio propagation path losses we have used the “Radio Mobile” gain [dBi]
Bandwidth 22 22 22 22 22 22 22 22 22 22
freeware software with nodes using 20-meter antenna heights. [MHz]
Thermal Noise -174 -174 -174 -174 -174 -174 -174 -174 -174 -174
[dBm/Hz]
Receiver Noise 10 10 10 10 10 10 10 10 10 10
B. Traffic model Figure [dB]
Noise Power -90.6 -90.6 -90.6 -90.6 -90.6 -90.6 -90.6 -90.6 -90.6 -90.6
[dBm]
We assume asymmetric traffic demand from each node to the Interference
Margin [dB]
2 2 2 2 2 2 2 2 2 2

gateway node connected to the Internet. Under this assumption Minimum Es/N0
[dB]
0.6 3.6 4 5 7 9 12 16 20 21

Sensitivity -88 -85 -84.6 -83.6 -81.6 -79.6 -76.6 -72.6 -68.6 -67.6
the average traffic from node i to the Internet gateway, node g, (minimum Rx.

is defined as λgi and from the gateway to node i as λig . Hence,

Power)
[dBm]***
Maximum 124 121 120.6 119.6 117.6 115.6 112.6 108.6 104.6 103.6
the total network traffic load is given by: allowed path
loss[dB]
*Rates 5.5 and 11 are omitted assuming that all nodes use parameters similar to the one utilize at the physical layer of equipment
based on IEEE802.11g.
X X ** According to FCC part 15.247 (b)(1) and (b)(3)(i): the power must be reduced by 1dB for every 3dB that the directional gain of the

λ= λgi + λig antenna exceeds 6dBi. For omnidirectional antennas FCC part 15.247(b)(3) and (b)(3)(iii) apply, i.e. the maximum EIRP must be
36dBm (4W).
*** By increasing the interference margin the minimum required receive power is increased and the network topology could be
∀i6=g ∀i6=g modified.
 Hence, using the parameters in the aforementioned table, carrying traffic. This equivalent matrix is defined as Rmesh =
the received node power matrix P r is: (R ∗ diag(Λ|λ(DL) =1 ) and removing links carrying no traffic)
  which results in:
∞ −87.8 −92.3 −87.8  
 −87.8 ∞ −87.8 −92.3  0 0 1 0 0 0
Pr = 
 −92.3 −87.8
 [dBm].  0.1 0 0 0 0 0 
∞ −87.8  
 0

−87.8 −92.3 −87.8 ∞ 0 1 1 0 0 
Rmesh =   0.1 0.1 0 0 0 0  .

Since the received sensitivity is −88dBm, the feasible
 
 0 0 0 0 0 0.1 
links are those for which Pr ≥ −88dBm (sensitivity at rate 0 0 0 0 1 0
R0 = 1Mbps). We use the node-arc incident matrix, A, to
represent the network topology [8] and is given by: Note that for this redefined matrix, the mesh-link
labels are according to the order list of the set
{(2, 1), (1, 2), (3, 2), (2, 3), (4, 3), (3, 4)}.
 
−1 −1 1 0 0 0 1 0
 1 0 −1 −1 1 0 0 0 
A=  0

STDMA Scheduling for mesh
0 0 1 −1 −1 0 1 
0 1 0 0 0 1 −1 −1
To find the interference-based scheduling for constant trans-
Gateway, routing, and link traffic mission power and variable rate systems we follow a procedure
In this example we arbitrarily define node 3 as the gateway similar to the one described in [9].
for Internet access. Therefore the external traffic load in We find the sets of cliques containing links having the
matrix form can be written as: property that all links in the same clique can transmit si-
  multaneously selecting one of the available data rates in
0 0 0.1 0 table I according to their SINR. We represent by S the the
 0 0 0.1 0  (DL) compatible matrix used for interference-based schedule. The
Λ=  1 1 0 1 λ

column vectors of S correspond to the set of cliques and they
0 0 0.1 0 can be linearly combined to create the STDMA schedule. We
T
To express the traffic load in vector form we label only the define the vector of wights α = [α1 . . . αK ] corresponding
(S,D) pairs with traffic load higher than zero as p = 1 . . . 6 cor- to the fraction of the time that each column vector of S is
responding to the indexes of the ordered source-to-destination activated within a STDMA frame. Hence, for a given α the
set {(1, 3), (2, 3), (3, 1), (3, 2), (3, 4), (4, 3)}. Hence the traffic allocated link capacity, c = [c1 . . . cL ]T is given by
load in vector form is given by: K
T (DL) X

Λ = 0.1 0.1 1 1 1 0.1 λ . c = Sα ; αk = 1. (2)
k=1
Using the source-to-destination pairs in Λ, and the Dijkstra The capacity allocation can be done through a schedul-
algorithm with equal cost for all links (minimum hop routing) ing algorithm. The scheduling algorithm allocates slots and
we get the the link-path routing matrix R [5]: transmission rates depending on the amount of traffic passing
 
0 0 1 0 0 0 though each link and the objective function to be maximized.
 0 0 0 0 0 0 
  In order to find the interference-based schedule for max-min
 1 0 0 0 0 0 
  fair allocation we utilize the technique refer to as “column
 0 0 1 1 0 0  generation” [5], [10], [11] for constant transmission power and
R=  1 1 0 0 0 0  .

  variable transmission rate as formulated in [9] but with the
 0 0 0 0 0 1 
  redefined mesh equations derived in this paper. That is, we
 0 0 0 0 0 0  find S, α and Λmesh = [λ1 . . . λp ] that solve the following
0 0 0 0 1 0 optimization problem:
Here we can see that for source-to-destination labeled 1, maximize λmin
links 3 and 5 are utilized for routing (i.e. these link labels subject to λi ≥ λmin ∀i;
correspond to links (1,2) and (2,3)). Source-to-destination Rmesh Λmesh ≤ Sα ;
labeled 2 utilizes only link 5, etc. P
αk = 1 ;
From the routing and external traffic we can find the traffic k
load demand for each link by Π = RΛ. Note that in this case λmin ≥ 0; Λmesh  0 ; 0  α  1 .
links 2 and 7 do not have traffic demand since links 1 and 3 are III. U SER - DEPLOYED AND C APACITY E VALUATION FOR
the ones selected for the downlink and uplink traffic for node THE RURAL C OMMUNITIES IN N ICARAGUA
1. Since they are carrying no traffic they can be removed from
been scheduled for transmission. In fact, in our implementation A. Description of User-deployed approach
we redefine an equivalent link-path routing matrix accounting We assume a simple user-deployed behavior. The com-
for the relative traffic load and removing the links that are not munity starts by locating their own nodes at their locations
(telecenters) and the shared Internet access node (Gateway) Stage 1
located at the closest land-based infrastructure. At the commu- 7

nities under consideration there are two municipalities head-

end (towns) where it is possible to have access to land-based 6
2

Internet services: San Juan de Limay and San Juan del Rio 1
5
Coco. The geographical coordinates are shown in table III-A.
In San Juan de Limay the operator ENITEL has a 63-meter
high tower, and we assume that a similar tower could be 4
utilized (or is available) in San Juan del Rio Coco. 8

By utilizing these gateways we can group the communities 3

into two potential user-deployed subnetworks as shown in
Fig.2. We call the set of communities that are geographically Stage 2
closer to ENITEL subnetwork 1 in San Juan de Limay, and 7
the set of communities closer to San Juan del Rio Coco
subnetwork 2. 6
2
The user-deployed approach corresponding to each subnet- 1
5
work is then applied. The stages for connecting the commu- 9

nities in subnetwork 1 are illustrated in Fig.3 when utilizing

6dBi omnidirectional antennas (transmitter and receiver) and 4
8
the parameters in table I.
After stage 1 and its resulting connectivity, the users add a 3

relaying node to connect node 5 (El Carrizo) and node 6 (El

Angel 1) to ENITEL(San Juan de Limay). For that, they select Stage 3
an intermediate (high) point for the relaying node. We refer to 7
10
relaying nodes that don’t generate external traffic as Digipeater
(Digi). On the next stage (stage 3), node 7 (El Mojon) has to be 6
2
connected by also adding another digipeater. One high point in 1

the region can be located that connect to node 5. The positions 5

9

for the digipeaters are summarized in the table III.

B. Capacity evaluation for subnetwork 1 8

4

We apply the described method before together with the

3
above user-deployed network and also utilizing beam-steering
antenna systems. The capacity evaluations are summarized in
Fig. 3. User-deployed stages for subnetwork 1. The gateway is indicated by
a square (node 8), digipeaters by rhombus (nodes 9 and 10), and the nodes at
TABLE II the communities are represented by a circle.
G EOGRAPHICAL COORDINATES FOR I NTERNET G ATEWAYS

Node name Latitude Longitude Altitude

ENITEL (Sn Juan de Limay) 13 10’23”N 08636’23” W 948.6ft
San Juan del Rio Coco 1333’00”N 086 10’ 00”W 2908.5ft

Fig. 2. Subnetwork 1 and subnetwork 2. Fig. 4. User-deployed (subnetwork 1).

Source: Map was derived using Google EarthTM . c
Google Inc. Source: Map was derived using Google EarthTM . c
Google Inc.
 TABLE III
G EOGRAPHICAL COORDINATES FOR DIGIPEATERS IN SUBNETWORK 1

Node name Latitude Longitude Altitude

Digi-1(node 9) 1301’56”N 08638’23”W 3218.5ft
Digi-2(node 10) 1318’15”N 08640’24”W 3802.5ft

the following table:

TABLE IV
C APACITY EVALUATION FOR SUBNETWORK 1(*)
Antenna Interference Antenna Uplink Downlink Total
System Margin [dB] Gain [dBi] (rate/node) (rate/node) rate
Omni 2 6 115kbps 1.2Mbps 8.8Mbps
Beam-steering 2 12 233kbps 2.3Mbps 18Mbps
(HPBW=30)
Beam-steering 5 12 517kbps 5.2Mbps 40Mbps
(HPBW=30)
Fig. 5. User-deployed (subnetwork 2).
* Communities and digipeaters: San Luis, La Fraternidad, San Lorenzo, La Grecia, El Carrizo, El Angel 1, El Mojon, Source: Map was derived using Google EarthTM . c
Google Inc.
ENITEL (San Juan de Limay), Digi-1, Digi-2.

Pt=28dBm, Sensitivity = −88 dBm

We can note that utilizing an interference margin of 5dB and
Beam-steering antennas produces higher end-to-end data rate. 5

This is because when utilizing IM=2dB with beam-steering 4

antennas, the network topology changes with respect to the
omnidirectional case making link (3,5) feasible. Link (3,5) is a
weak link utilized for routing traffic towards the gateway (node
8) and to satisfy its traffic demand its activation consumes 3 9
much network capacity. 7 6
In summary, by the simple user-deployed approach for 8 2

the subnetwork 1, the (upper bound) allocated end-to-end 1

downlink transmission rate is up to about 5.2Mbps per node
Fig. 6. Network topology (subnetwork 2). Node 7 (San Juan del Rio Coco)
(telecenter) and for the uplink is up to about 517kbps per node. is the Internet gateway and nodes 8 and 9 are the digipeaters.

C. User-deployed subnetwork 2
To connect the remaining communities we assume that an interference margin to 5dB we found that the community called
Internet gateway can be installed at the municipality head- El Jobo gets disconnected. Fig.6 shows the network topology
end called San Juan del Rio Coco. Next, following the user- when utiling beam-steering antennas with 2dB interference
deployed approach described before we add two digipeaters in margin.
the locations summarized in the following table: Note that there is a long range connection of 36.2km from
For the gateway at San Juan del Rio Coco we assume El Ojoche (node 1) to San Miguel (node 4). This is because
63m tower height. As before, all nodes in the network are there is line of sight between these two points as illustrated in
assumed to be installed utilizing 20m antenna height. Figure Fig.7.
5 shows the locations of the communities and the locations of
the digipeaters needed for connectivity to the Internet gateway D. The capacity evaluations for subnetwork 2
to be located in San Juan del Rio Coco.
We apply the described method before together with the
when utilizing omnidirectional antennas with the selected
above user-deployed network and also utilizing beam-steering
points there is not fully connectivity for all nodes that com-
antenna systems. The capacity evaluations are summarized in
poses the network. The utilization of beam-steering antennas
table VI.
with an Interference margin of 2dB make possible to connect
all the communities to the Internet gateway. The increment TABLE VI
on the interference margin is done in order to allowed more C APACITY EVALUATION FOR SUBNETWORK 2(**)
resistance to interference and by this way we remove links with Antenna
System
Interference
Margin [dB]
Antenna
Gain [dBi]
Uplink
(rate/node)
Downlink
(rate/node)
Total
rate
low transmission rate (we avoid them to be used for routing Omni
Beam-steering
2
2
6
12
-
390kbps
-
3.9Mbps
-
26Mbps
traffic and to be scheduled). However, when increasing the (HPBW=30)
Beam-steering 5 12 - - -
(HPBW=30)

** Communities and digipeaters: : El Ojoche, San Marcanda, El Varrillal, San Miguel, Cristo Rey, El Jobo , San Juan del
TABLE V Rio Coco, Digi-3, Digi-4.
L OCATIONS FOR DIGIPEATERS IN SUBNETWORK 2

Node name Latitude Longitude Altitude Hence, for this user-deployed subnetwork, the resulting
Digi-3 1331’45.1”N 08609’27.4”W 3438.3ft (upper bound) end-to-end downlink transmission rate is up to
Digi-4 1334’59.6”N 08659’02.8”W 3676.2ft about 3.9Mbps per node (community) and 388.5kbps per node

Fig. 7. Terrain profile between El Ojoche and San Miguel.
for the uplink.
IV. C ONCLUDING R EMARKS
In this work we have introduced a methodology for capacity
evaluation of user-deployed multihop wireless networks in
rural areas. We have applied the methodology to evaluate a
user-deployed scenario for Internet access in thirteen rural
communities of Nicaragua. These communities are currently
candidate sites to install telecenter by the regulatory body of
Nicaragua.
In order to reduce the Internet service cost we have uti-
lized shared common access points under asymmetric traffic
demands. In our practical example, due to the terrain roughness
and long distance, the communities were divided into two
geographical sets conforming disjoint community subnetworks
connected by two independent gateways. In order to overcome
the high path-losses due to the long distance and mountains,
digital repeaters (relaying nodes) were assumed to be installed
to progressively connect each community from the gateways.
To study the systems we have considered the utilization of
6dBi omnidirectional antennas and also beam-steering circular
array antenna systems with half power beamwidth of 30o (12
dBi).
The theoretical maximum end-to-end transmission rate
(throughput) provided to each node by the user-deployed
approach was found by link transmission scheduling apply-
ing nonlinear optimization and considering asymmetric traffic
demands.
The utilization of beam-steering circular array antenna sys-
tems provided the benefit of radio range extension and reduc-
tion of multiple access interference yielding the need of fewer
number of digipeaters and therefore faster deployment. With
the underlying assumptions and the radio parameters in table
I, the aggregated end-to-end data rate per user was higher than
the typical ADSL connection when utilizing beam-steering
antennas and approximated to the typical ADSL connection
when connectivity was possible using omnidirectional antennas
(subnetwork 1). The maximum end-to-end rate per node i.e.
from the gateway to the community telecenters (downlink)
was about 1.2 Mbps when connectivity was possible using
omnidirectional antennas and up to around 5 Mbps per node
with beam-steering circular antennas. The results provide
useful figure-of-merits under the hypothetical system and con-
ditions assumed in our evaluations,and show that MWN is a
feasible alternative to provide rural communication to those
communities.
R EFERENCES
[1] (2008) Radio Mobile Freeware by VE2DBE. [Online]. Available:
http://www.cplus.org/rmw/english1.html
[2] (2006, Jan.) AA 001-2006, Wireless Access System operation on
the frequency bands: 900 MHZ, 2.4 GHZ, and 5 GHZ. Instituto
Nicaragüense de Telecomunicaciones y Correos (TELCOR). In
Spanish. [Online]. Available: {http://www.telcor.gob.ni/Descargar.asp?
DOC\ ID=42024}
[3] (2007) GTOPO30 - tile W100N40. [Online]. Available: http://edc.usgs.
gov/products/elevation/gtopo30/w100n40.html
[4] G.A. Hufford, A.G. Longley, W.A. Kissick, “A Guide to the Use of
the ITS Irregular Terrain Model in the Area Prediction Mode,” National
Telecommunications and Information Administration (NTIA) Report 82-
100, Apr. 1982.
[5] M. Johansson and L. Xiao, “Cross-layer optimization of wireless
networks using nonlinear column generation,” IEEE Trans. Wireless
Commun., vol. 5, no. 2, pp. 435–444, Feb. 2006.
[6] J. E. Håkegård, B. Myhre, P. H. Lehne, T. Ormhaug,
V. Bjugan, M. Mondin, M. Elkotob, and F. Steuer, “Open
Broadband Access Network (OBAN): D8 scenarios and wireless
performance and covegare,” OBAN consortium, Mar. 2005.
[Online]. Available: http://oban.prz.tu-berlin.de/D8 Scenarios and
wireless performance and coverage.pdf
[7] “Cisco Aironet 1522 Lightweight Outdoor Mesh Access Point,
Data Sheet,” Cisco Systems Inc., California, USA. [Online].
Available: http://www.cisco.com/en/US/prod/collateral/wireless/ps5679/
ps8368/product data sheet0900aecd8066a16c.pdf
[8] R. K. AHuja, T. L. Magnanti, and J. B. Orlin, Network Flows: Theory,
Algorithms, and Applications. Prentice Hall, Inc., 1993.
[9] M. Sánchez, J. Zander, and B. Hagerman, “On the performance of power
and rate control in STDMA multihop networks with advanced antennas,”
in Proc. International Conference on Networking (ICN2008). Cancun,
Mexico: IEEE, Apr. 2008.
[10] P. V. P. Bjorklund and D. Yuan, “Resource optimization of Spatial TDMA
in ad hoc radio networks: a column generation approach,” in Proc. IEEE
INFOCOM 2003, San Francisco, USA, Apr. 2003, pp. 818–824.
[11] S. G. Nash and A. Sofer, Linear and Nonlinear Programming, ser.
Industrial Engineering. McGraw-Hill, 1996.