This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the WCNC 2010 proceedings.

Optimization of Common Air Interface in Cellular Multihop Wireless Networks in the Presence of Traffic Variation
Beatriz Lorenzo, Student Member, IEEE
CWC, University of Oulu Oulu, Finland

Savo Glisic, Senior Member, IEEE

Telecom Laboratory, University of Oulu Oulu, Finland

Abstract - In this paper we define the jointly optimum topology for the duplex transmission (uplink/downlink) in multihop cellular networks which is aware of the intercell interference and a protocol that reconfigures the optimum topology based on the observation of the temporal traffic in the network. In addition we also consider the application of network coding in cellular networks to combine the uplink and downlink transmissions and incorporate it into the optimum bidirectional relaying with intercell interference awareness resulting in a comprehensive solution for 4G common air interface. Keywords-cellular network, common air interface, intercell interference, network coding, relay, topology control.

I. INTRODUCTION Recently, relaying has been studied intensively for applications in multihop cellular networks [1-5]. In [6] relaying techniques that increase the throughput in multihop wireless networks are analyzed by applying network coding over bidirectional traffic flows. This technique has been included in our approach with the focus on mitigating the intercell interference and adapting the relaying topology to traffic variations across the network. Two basic coding strategies for the one-relay case were proposed by Cover and El Gamal in [7]: decode-and-forward (DF) and compress-and-forward (CF). Furthermore, [7, Th. 7] provides a general lower bound on the capacity of one-relay networks which can be achieved using a combination of DF and CF. The relaying concept is the basis of cooperative and virtual antenna transmission too [8-12]. The bounds of the information theoretic capacity of a discrete memoryless channel are given by [13] based on a timesharing approach. The capacity analysis for the special case of degraded relay channels by the use of superposition block Markov encoding is presented in [14]. For other type of channels the capacity is upper-bounded (max-flow-min-cut theorem [15]) by the minimum of mutual information obtained by the broadcast channel (transmission from the source to relay and destination) and the multiple access channel (independent and simultaneous transmission from the source and relay to the destination). Capacity bounds and power allocation for wireless relay channels are presented in [16] for halfduplex relay and single-antenna terminals. Algorithms for finding the capacity bounds for the multi-antenna terminals are given in [17], [18]. Currently considered relays are assumed to work under half-duplex mode, by using an orthogonal duplexing (in time or frequency) between the relay-receive phase and the relay-

transmit phase. This phase separation allows defining several half-duplex relay protocols. The number of options leads to the four protocol definition [19]-[20] referred to as protocol 1, 2 and 3, and forwarding. In protocol 1 the source communicates with the relay and destination during the relayreceive phase and in the relay-transmit phase, the relay terminal communicates with the destination. In protocol 2 during the relay-receive phase the source only transmits to the relay. It is assumed that the destination is not able of receiving the message from the source in that phase. In the relaytransmit phase source and relay transmit simultaneously to the destination. Hence in the relay-transmit phase the channel becomes a multiple access channel (MAC). Protocol 3 is a combination of protocols 1 and 2. The source transmits to the relay and the destination in the relay-receive phase and in the relay-transmit phase, the source and the relay transmit to the destination. Notice that the relay is transmitting during the second phase, so that it cannot be aware of the signal transmitted by the source in the second phase. This protocol can achieve a better spectral efficiency than previous ones. Finally, the traditional forwarding protocol consists of a transmission from the source to the relay during the relayreceive phase and a transmission from the relay to the destination in the relay-transmit phase. Having in mind the above results, in this paper we present the design of a relaying protocol jointly optimizes relaying topology, routing and scheduling in the presence of intercell interference. By using newly developed TSL algorithm for the search of the optimum topology we show that the optimum choice of the relaying topology can provide significant performance improvements. We apply network coding to bidirectional links (uplink/downlink) and combine it with the optimum relaying to define a new cognitive common air interface for 4G cellular networks. We also demonstrate that a reconfigurable relaying topology provides better network utility and presents the framework for quantifying these improvements for spatially and temporally varying traffic. Numerical examples show that a combination of these components provides a flexible optimal solution for future 4G common air interface in cellular networks. II. SYSTEM MODEL 2.1 Network and Intercell Interference Model A. Uplink We consider a cellular network with a set of I={i} base stations. Let us assume that in a reference cell with index i=r, there is a user m(r) connected to the access point AP(r) (base

978-1-4244-6398-5/10/$26.00 2010 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings.

( station) with channel gain g mr()r ) . At the same time a cochannel

interfering user m(i) is connected to access point AP(i) in cell i, ( i I-r = {i r} , with channel gain g mi()i ) . Assuming that all the users are expected to reach their respective access points with ( the same required received power S mi()i ) , we have for the

AP ( i ) and SINR((rm )

(r)

is SINR at m(r). The optimum radio

resource management is again defined by (4)-(6).

( transmitted powers Pmi( i)) of the useful and interfering signal

AP(r) m(r)= 0 1 2

3 2 1

AP(i)

0=m(i)

( ( ( S mi()i ) = g mi()i ) Pmi( i)) ; i I

(1)
-a-

We denote by I

position of the reference receiver AP(r) due to the interfering cochannel signal transmitted in cell i. This can be presented as
( ( ( I mr( )i ) = g mr()i ) Pmi( i)) ; i I-r

(r) m( i )

the interference power level at the

AP(r) m(r)= 0 -buplink

1 2 3

3 1
3 3 3 1 2 3

AP(i) 2 1
1 2

(2)

2
3

0 = m(i)

( where gmr()i ) is the gain of the channel between the interfering

user m(i) and AP(r). The signal to interference plus noise ratio ( SINRmr( )r ) at AP(r) in the presence of all interfering users m(i) is

downlink
1

SINR
where N r = S

(r ) m( r )

( nr + I mr( )i ) ir

( S mr( )r )

g ( r()i ) = N r1 + mi ) ( i r g m( i )

(3)

Fig. 1. Modeling interfering users positions for 2-cells. 2.2 Relaying and scheduling
( ( We will use notation r ( m2r ) , m1( r ) , m (2i ) , m1i ) ) , to denote

(r ) m( r )

/ nr = SNR , and the channel capacity per

( ( cmr()r ) = log(1 + SINRmr( )r ) )

unit spectra can be represented as (4) (5) The network capacity is then given by
( C = cmr()r ) r

simultaneous transmission (relaying) on reference cell r I

( ( from user m1( r ) to m2r ) and interfering users from m1( i ) to m2i )

If the radio resource management is defined as channel assignment function A(m(r)) responsible to allocate to each user m(r) proper channel then the optimum assignment is defined as

( i I-r) position in all interfering cells. Under these conditions the corresponding link capacity will be denoted as ( ( c ( r ) ( m2r ) , m1( r ) , m (2i ) , m1i ) ) . This capacity can be calculated by the following set of equations
( ( ( ( ( ( S mmr2) ) = Pm(m2 ) g mmr2) ) ; S mmi2) ) = Pm(m)2 ) g mmi2) r) i ( ( ( (
1 1 1 1 1 1 (r ) (r ) (r) (i ) (i) (i)

A(m( r ) ) = max A( m( r ) ) C; r I

(6)

( I mmi2) (
1

(r)

( ( = Pm(m)2 ) g mmi2) i (
1 1

(i)

(r )

( ( ( = S mmr2) ) g mmi2) ) / g mmi2) ( ( (

1 1 1

(i )

(r)

(i)

B. Downlink In this scenario the reference access point AP(r) is providing power S(m) r
(r )

to the reference user m(r). At the same time,

(i)

SINR

( ( m2r ) ) ( m1 r )

(m

(i ) 2

,m

(i ) 1

)= n

( Smmr2) (
1

(r )

)
(r )

( + I mmi2) ( i r
1

=
(8)
1

interfering AP

providing the same signal level S

m( i ) (i )

to the

user m(i), is producing interference to the useful signal of user m(r). So, we have

( (r) ( (i ) = N r1 + g mmi2) ) / g mmi2) ) ( ( 1 1 ir

m( i ) (i )
(r)

=g
)

( m( i ) ) ( m( i ) ) (i ) (i )

,i I
)

( ( (1) (2) m1i ) = (m1(1) , m1(2) ,..., m1( Nc ) ); m(2i ) = (m2 , m2 ,..., m2Nc ) ); i I-r

I ((im )

= g((im ) P((i )m )
(r )

(r )

(i)

= S(m) g((im i )

(i )

(r)

/ g((im ) , i I r )
1

(i )

(7)

m ( ( ( c( r ) (m2r ) , m1( r ) , m(2i ) , m1i ) ) = log 1 + SINRm(2r ) (m(2i ) , m1i ) ) ; i I-r

1

(r)

SINR((rm )
where g
(i )

(r ) (i ) = N r1 + g((im ) / g((im ) ) ) i r

where Nc is the number of cells. We define now the multihop (H hops) route as a series of relaying transmissions
( ( ) ( i i ( ( r ) (mHr ) , mHr1 ,..., m2r ) , m1( r ) , m(H) , m(H)1 ,..., m(2i ) , m1i ) )

(9) (10)

(m (i )

(i )

is the channel gain between AP

(i )

(i )

and user m ,

(i)

The capacity of the route is defined as

(m (i )

is the power needed at AP

(i)
(r )

to provide power P for

(r)

c(

(r)

( ( ( ) = min h cr()r ) (mhr ) , mhr)1 , m(hi ) , m(hi1 ); h = 2,..., H

user m , I ((im )

is the interference power at m

produced by

m ( ( ( ) cr()r ) = log 1 + SINRm(2r ) ( mhr ) , mhr)1 , m (hi ) , m (hi1 )

1

(r )

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings.

which is equal to the minimum link capacity on the route. The optimum set of relaying routes is defined as

{ } = max
(r )

H ,( r )

C () ; where C () = c( ) ; r I (11)
(r )

In order to reduce the interference produced by the concurrent transmission of the different relaying segments, a scheduling in different time slots is introduced in the sequel. This scheduling will also solve the constraint that nodes in a wireless network can not receive and transmit the signal simultaneously on the same channel. 2.3. Two Dimensional Relaying Topology The cochannel interference can be further reduced by scheduling different transmissions in different channels (time slots). All necessary transmissions between all users and their respective access points should be completed in B slots (scheduling cycle) in both directions: uplink and downlink. As an illustration, for the two cells scenario and notation shown in Fig.1, a possible (feasible) topology is shown in Fig.1b. For a systematic presentation of the problem the cell area is divided into concentric rings (three rings for each cell in Fig.1). It is assumed that one cochannel user from each ring has unidirectional connection with the corresponding access point. For the uplink, the topology consists of four partial topologies representing transmissions in four consecutive time slots (B=4). In the first time slot (the first partial topology) there are two simultaneous transmissions: packet originating from ring 3 in cell #r is transmitted from ring 3 to ring 2 and at the same time packet originating from ring 1 of cell #i is transmitted from ring 1 to access point AP(i). In the second time slot (the second partial topology), packet originating from ring 3 in cell #r is transmitted from ring 2 to AP(r) and at the same time packet originating from ring 2 of cell #i is transmitted from ring 2 to access point AP(i). Similarly the same notation is then used for transmission in time slot 3 and 4. Similar topology for the downlink is presented in the lower part of Fig.1b. These seven partial topologies together are referred to as a possible or feasible two dimensional (time and space) topology and will be represented in the sequel by a given topology index t. For this concept (11) becomes ( 2) ( 2,r ) ( ( 2 ) ) (2,r ) = max ; where C ( ) = c( ) (12) ( 2,r ) C

and = ( up ) ( down ) . To elaborate this concept in more detail an example of possible topology that includes network coding is shown in Fig.3 for two cell scenario from Fig.1a. The traffic between users and access point is bidirectional, so given a schedule that alternates the transmissions between the different rings, after certain number of time slots all intermediate users (m(i), i I) have information frames buffered for transmission in both directions. Whenever an opportunity arises, the intermediate users combine two information frames, one for each direction, with a simple XOR operation and send it to its neighbors in a single omnidirectional transmission. Both receiving nodes already know one of the frames combined, while the other frame is new. Thus, one transmission allows two users to decode a new packet, effectively doubling the capacity of the path, reducing the power consumption of the transmitter node and reducing the number of time slots required to complete the transmission. If we denote by n(i) the number of rings in cell i and

N = n ( i ) the total number of rings in the network, the

i =1

Nc

vector (i ) = (1(i ) ,..., n((ii)) ), (i ) R n

(i )

defines the amount of

generated source traffic by the users situated in the different rings in cell i to be transmitted to the access point AP(i) on the (i ) uplink, and (i ) = (1(i ) ,..., n((ii)) ), (i ) R n the traffic that the

access point AP(i) is transmitting to the users on the downlink. (i) (i) For the same traffic vectors , the base station can schedule the transmission through different channels (time slots) resulting in temporal and spatial MAC protocol. The network traffic on the uplink and downlink is defined as = ( (1) , (2) ,..., ( Nc ) ) = (1 ,..., N ) and = ((1) , (2) ,...,

( Nc ) ) = (1 ,..., N ) respectively.
The base stations jointly assign an access vector

a = (a (1) , a(2) ,..., a( Nc ) ) = ( a1 ,..., a N ) , where each component

an (0,1) , to the different rings to give them permission to

transmit. With an=1 the users from ring n are allowed to transmit otherwise not. In the two cell case a=(a(1),a(2)), the first half of the coefficients represents the permissions to transmit for the rings in cell #r and the second half for rings in cell #i, i I-r. We consider symmetric bidirectional transmission (duplex connection) in the sense that the access point will only transmit to the users situated in the rings activated by a. The transmission schedule presented in Fig. 2 defines a possible topology for two cell scenario and access vector a=1. In this case all rings have duplex connection and the topology consists of eight partial topologies representing transmissions in eight consecutive time slots. In the first time slot (the first partial topology) there are two simultaneous transmissions; packet originating from the access point AP(r) (addressed to user in ring 3) is transmitted to ring 2 in cell #r and at the same time packet originating from ring 2 (addressed to access point AP(i)) of cell i is transmitted from ring 2 to ring 1. In the

B , H ,

r I and (2) is the two dimensional relaying topology to be

elaborated in more detail in the next section. 2.4 Bidirectional Relaying and Network Coding In this section we additionally introduce network coding and combine it with the previous results on optimum relaying to achieve further improvements in system performance. Let us assume that the hops are indexed in increasing order for uplink as h(up) and for downlink as h(down). By combining the uplink and downlink traffic from the previous hop at hop h ( ( ( as yhdown ,up ) = yhdown ) yhup ) the number of overall time slots 1 1 needed for transmission in cycle B can be reduced. The optimization process defined by (12) now becomes ( 2) ( 2, r ) ( ( 2) ) (2,r ) = max ; where C ( ) = c ( ) (13) ( 2,r ) C

B , H , ,

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings.

second time slot (the second partial topology), packet originating from access point AP(r) (addressed to user in ring 1) is transmitted to ring 1, at the same time packet originating from ring 3 (addressed to access point AP(r) ) is transmitted from ring 3 to 2 and, packet originating at AP(i) (addressed to user in ring 2) is transmitted to ring 1 in the adjacent cell. Similarly the same notation is then used for transmission in time slot from 3 to 8. As already discussed earlier these eight partial topologies together are referred to as a possible two dimensional (time and space) topology and will be represented in the sequel by a given topology index (t). So there will be limited interference transmission for 3 users per cell in 8 channels (8 slots in Fig. 2) giving the intercell throughput 6/8=3/4, as opposed to the 6/12=1/2 in a conventional TDMA system where each cell uses a half of available channels (slots). Although scheduling in Fig.1b requires 7 time slots it also assumes transmissions over three rings which requires higher power.
AP(r) -cslot: 1
2 3 4 5 6 7 8

For simplicity, in the sequel we will describe the protocol for only one direction of the traffic and then at the end of the section make comments on how we extend the protocol for bidirectional case. - The routing matrix (or relaying matrix) R=[rln] has entries rln=1 if source n (n=1,2,..,N) is using link l (l=1,2,..,L) and 0 otherwise. Recall that, parameter N is the number of overall rings in the network. Parameters rln = r ( i ) are calculated as

rlm ( i ) =
1

( L = T (m2i ) , m1( i ) ) .
( m1 i )

( m 2i )

( rlm ( i ) T ( m 2i ) , m1( i ) )
2

lm1

and,

3 1 2
3( r ) 3(r ) 3( r ) 2( r )

AP(i) 2 1
2(i )

m(r)=0
1( r )

0=m(i)

3( r )

- The scheduling set will be combined with the routing matrix R resulting into two dimensional routing protocol characterized by extended routing matrix R (2) . By assuming that the scheduling cycle within the maximum clique has B steps, the optimization process will include: a) Utility function (15) U = (1/ B ) an log( xn ) / Pn

3( r )

(r ) 2

1( r )

2( r ) 2( r )

2(i ) 2(i ) 2(i ) 3(i ) 1(i ) 3(i ) 3(i ) 3(i ) 3(i ) 1(i )

Fig. 2. Possible schedule by using network coding. III. PROTOCOL DESCRIPTION The examples of topologies presented in the previous section are based on intuition and we need a systematic approach to the system optimization. For these purposes we introduce the following definitions. For i I,
( - We define topology matrix T ( i ) = T ( m2i ) , m1( i ) ) with ( ( T (m2i ) , m1(i) ) = 1, if user m1( i ) is transmitting to m2i ) and 0,
( ( otherwise with indexes m1( i ) , m2i ) = 0,1, 2,..., n ( i ) , m2i ) < m1( i ) .
( Each ( m2i ) , m1( i ) ) pair is represented by a specific link index l as shown in Fig. 3 for the case of two cells. With this notation the vector of equivalent (source + relayed) rates in cell i is ( (i ( ( (14) x mi2) = x sm)2 + T ( m 2i ) , m1( i ) ) x mi1)
( m1 i )

where Pn is the aggregate power needed for transmission of information from the source n to the access point on the uplink or viceversa on the downlink, an is the access parameter as defined in the previous section. b) Constraint R ( 2) x( 2) c( 2) (R ( 2) ) with the following definitions of extended system parameters x(2)T = ( xT (1), xT (2)..xT ( B) ) ; c(2)T = ( cT (1), cT (2)..cT ( B) ) (15a)
R(2) = diag R(b) , b = 1, 2,.., B; R R, x R(2)

( where x( i ) = xmi ) for each direction of the traffic. The 2 overall topology matrix will be formally defined as T=diag[T(i)] and x = x(i ) is the vector of the overall

where c are the logical link capacities calculated as discussed in Section II which capture the functional dependency of power control and interference level in the network. c) Each component of the set of feasible routes in should provide directional connection for each terminal to the corresponding access point. This means that the sequence of links generated in a clique cycle must provide connection for all terminals to the corresponding access point. To define this constraint explicitly we introduce the link hopping distance hl and the vector h = {hl}. hl represents the number of rings that link l is hoping over, from its transmitter/receiver to the corresponding receiver/transmitter. Similarly the source hopping distance is denoted as d = {hn}. The sum of link hopping distances on the route from source n to the access point should be equal to the source hopping distance T (15b) R (b)h(b) = d
( ) (i ( ( ) ( I T ) x = x n xmi 2 = x sm) 2 + T ( m2i ) , m1( i ) ) xmi1
( m1 i )

(15c)

aggregate rate.
l3 l2 l1
(r) AP

l9 l5 l11 l6
2 2 33 3

l8 l10 l7
(i) 1

The formulation of the problem obtained by equations (15a)(15c) can be summarized as: P: maximize U
T ,x

l4
1 1
(r) m(r) ,m1 2

l12
2

m ,m

(i) 2

14

AP

(i)

subject to R (2) x ( 2) c( 2) ( R (2) ); R (b)h (b) = d

T b

(16)

l31

Fig.3. Link notation

(I T)x = x n ; R R, x R (2)

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings.

In the case of bidirectional traffic an independent set of equations (15) should be written for both directions and (16) should be modified to include overall utility function U=U(up)+U(down) with separate set of constrains for both directions. IV. SIMULATION RESULTS

In this section, we provide several numerical examples to illustrate the performance of the proposed protocol. We ( ( calculate the link capacities c( r ) (m2r ) , m1( r ) , m(2i ) , m1i ) ) as specified in Section II. While the analysis is general, for simplicity in this section, the channel gains used to calculate

both cells have permission to transmit. As the number of topologies obtained for this access vector is very high, we plot the segment of topologies close to the optimum topology. The maximum utility is u=0.6991 by using network coding, and with no coding the maximum utility is u=0.5826. Both are smaller than in the previous case due to higher interference level. As several topologies provide the maximum utility, in Fig. 6 we show the transmission pattern for one of the optimum topologies (t=7) in the case with no coding for the previous access vector defined by the set of links

T7 = {{l1,down},{l10,up},{l4,down , l7,up},{l4,up},{l1,up},{l7,down},{l10,down}} .

SIRm( r )
1

( ( m2 r ) )

(m

(i ) 2

,m
2 1

(i ) 1

are

2 1

( g mmi2) (
1

(r)

1 / d m( r )m(i)
2 1

and

( ( m2i ) ) ( m1 i )

1/ d m( i ) m( i ) , where d m( r ) m( i ) is the distance between the

We can see that isolated short range transmissions are favored which can simultaneously reduce the intercell interference and power consumption. In Fig. 7 we plot the transmission pattern for topology

reference receiver in ring m2 and interfering transmitter in ring

m1(i ) , analogous for d m( i ) m( i ) and, is the propagation constant.
2 1

T5621 = {{l1,down},{l4,up},{l7,down},{l1,up l4,down , l10,up},{l7,up l10,down}}

In the simulations we use =4, and SNR=10. The calculation of the distances is straightforward from the geometry presented in Fig.1a. In the sequel we present the utility for different access vectors a versus the topology index (t) for the scenario presented in Fig. 1a. The resulting topologies, indexed by t, represent a certain combination of the active in B slots and will be represented formally as (2) = T (b ) = L( b ) (l ) .
b b

that corresponds to one of the topologies with coding that provides the maximum utility for a=[010010]. We can see an improvement in the number of slots needed with coding (5 slots) compared to 7 slots in the case with no coding. So for the same type of isolated and short range transmissions the utility function is improved by reducing the number of slots.
a=[010010] 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 no coding coding

In Fig. 4, the utility function is shown for a=[010100]. With this access vector user from ring 2 in cell #r and user from ring 1 in cell #i have permission to transmit. The maximum utility is obtained for topology index t=478 (u478=0.8739) by using network coding. We see a significant improvement compared with the maximum utility with no coding obtained for topology index t=215 (u215=0.6640). The optimum topologies for both cases are given by

T478 = {{l1,down },{l7,down },{l4,up },{l1,up l4,down , l7,up }}

a=[010100] 0.9 0.8 0.7 no coding coding

and

Utility Function

1000

2000

3000 4000 Topology Index (t)

5000

6000

T215 = {{l1, down },{l4,up },{l4, down , l7,up },{l7, down },{l1,up }}

Fig. 5. Utility function for access vector a=[010010].

l1

AP(r)

1 1

(r) m(r) ,m1 2

2 2

(i) m(i) ,m1 2

1 4

AP(i)

Time slot 1

l10

AP(r)
0.6 Utility Function 0.5

AP(i)
l7

Time slot 2

l4

AP(r)
0.4 0.3 0.2 0.1 0

1 1

2 2

33

Time slot 3 AP(i) Time slot 4 Time slot 5 Time slot 6 AP(i) Time slot 7 AP(i)

l4

AP(r)
l1

1 1

2 2

33

AP(i) AP(i)
l7

AP(r) AP(r)

11

2 2

33

-0.1

100

200

300 400 Topology Index (t)

500

600

l10

Fig. 4. Utility function for access vector a=[010100]. In Fig. 5, the utility function versus the topology index for a=[010010] is shown. With this access vector user in ring 2, in

AP

(r)

252

4 1

Fig. 6. Representation of the transmission pattern defined by the topology index t=7

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings.

l1

AP(r) AP(r) AP(r)

l1

1 1

2 2

(r) m(r) ,m1 2

33

14 (i) (i) m(i) ,m1 AP 2 1

Time slot 1 Time slot 2 Time slot 3

l4
1 1 2 2 33 3 2

u=0.5826. Both values are smaller than in the previous case due to higher interference level. REFERENCES
[1] H. Wu, C. Qiao, S. De, O. Tonguz, Integrated Cellular and Ad Hoc Relaying Systems: iCAR, IEEE Journal on Selected Areas in Communications, , vol. 19, no. 10, Oct. 2001, pp. 210515. R. Pabst et al., Relay-Based Deployment Concepts for Wireless and Mobile Broadband Radio, IEEE Commun. Mag., vol. 42, no. 9, Sept. 2004, pp. 8089. Y. Liu, R. Hoshyar, X. Yang, R. Tafazolli, Integrated Radio Resource Allocation for Multihop Cellular Networks With Fixed Relay Stations, IEEE Journal on Selected Areas in Communications, vol. 24, issue 11, Nov. 2006, pp. 2137 2146. L. Long, E. Hossain, Multihop Cellular Networks: Potential Gains, Research Challenges, and a Resource Allocation Framework, IEEE Commun. Magazine, vol. 45, issue 9, Sept. 2007, pp. 66 73. X. Shen, Z. Ma, W. Wang, K. Zheng, G. Liu, B. Fan, Multihop cellular networks toward LTE-advanced, IEEE Vehicular Technology Magazine, vol. 4, issue 3, Sept. 2009, pp. 40-47. P. Popovski, H. Yomo, Wireless Network Coding by Amplify-andForward for Bi-Directional Traffic Flows, IEEE Communications Letters, vol. 11, no.1, January 2007. T. Cover and A. E. Gamal, Capacity theorems for the relay channel, IEEE Trans. Inf. Theory, vol. 25, no. 5, pp. 572584, Sept. 1979. A. Sendonaris, E. Erkip, B. Aazhang, User cooperation diversity-part I: System description, IEEE Trans. on Communications, vol. 51, no. 11, pp. 1927-1938, Nov. 2003. J. Laneman, D.N.C. Tse, G.W. Wornell, Cooperative diversity in wireless networks: Efficient protocols and outage behavior, IEEE Trans. Information Theory, vol. 50 no. 12, pp. 3062 3080. Dec. 2004. A. Wittneben, B. Rankov, Impact of cooperative relays on the capacity of rank deficient MIMO channels, in Proc. IST Mobile & Wireless Communications Summit (IST-2003), Aveiro (Portugal), June 2003. M. Dohler, Virtual Antenna Arrays, PhD Thesis, Kings College London, London, UK, 2003. M. Dohler, E. Lefranc, A.H. Aghvami, "Virtual Antenna Arrays for Future Wireless Mobile Communication Systems", ICT 2002, Conference CD-ROM, Beijing, China, June 2002. E.C. van der Meulen, Three-terminal communication channels, Adv. Appl. Prob., vol. 3, pp. 120-154, 1971. T.M. Cover, A.A. El Gamal, Capacity theorems for the relay channel, IEEE Trans. on Information Theory, vol. 25, no. 5, pp. 474-584, Sept. 1979. T.M. Cover, J.A. Thomas. Elements of Information Theory. John Wiley & Sons, 1991. A. Host-Madsen, J. Zhang, Capacity bounds and power allocation for wireless relay channels, IEEE Trans. on Information Theory, vol. 51, no. 6, pp. 2020-2040, June 2005. B. Wang, J. Zhang, A. Host-Madsen, On the capacity of MIMO relay channels, IEEE Trans. on Information Theory, vol. 51, no. 1, pp 29-43, Jan. 2005. A. Host-Madsen, Capacity bounds for cooperative diversity, IEEE Trans. on Information Theory, vol 52, no. 4, pp. 1522-1544, April 2006. R. Nabar, H. Blcskei, F. Kneubhler, Fading relay channels: performance limits and spacetime signal design, IEEE Journal Selected Areas Communications (JSAC), vol. 22, no. 6, pp. 1099-1109, Aug. 2004. H. Ochiani, P. Mitran, V. Tarokh, Variable rate two phase collaborative communications protocols for wireless networks, IEEE Trans. on Information Theory, vol. 52, no. 9, pp.4299-4313, Sep. 2006.

AP(i)
l7

14

AP(i)

l4
1 2 3 3 25

l10
4 1

AP(r)

AP(i)
l7

Time slot 4

[2]
Time slot 5

l10

AP(r)

25

4 1

AP(i)

[3]

Fig. 7. Representation of the transmission pattern defined by the topology index t=5621
[4]

In Fig. 8 we represent the overall capacity for the previous access vector a. For the optimum topologies with coding we can see that the overall capacity of the system improves by a factor 4 compared with the case without coding.
a=[010010] 16 no coding coding

[5] [6] [7] [8] [9]

14

12 Overall Capacity

10

[10]
8

[11] [12]

1000

2000

3000 4000 Topology Index (t)

5000

6000

Fig. 8. Overall capacity for a=[010010]. V. CONCLUSION In this paper we present some solutions on intercell interference aware optimum relaying topology that includes bidirectional links and network coding. The utility function used in the optimization process drives the solution towards the topology favoring simultaneously isolated and short range transmissions. As expected, within these solutions further improvements are obtained by using network coding to reduce the number of slots needed for transmission. For example, for access vector a=[010100], the maximum utility obtained is u=0.8739 by using network coding, which is a significant improvement compared with the maximum utility with no coding u=0.6640. For a=[010010], the maximum utility is u=0.6991 by using network coding, and with no coding

[13] [14] [15] [16] [17] [18] [19]

[20]