Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
6, JUNE 2008
Abstract—Collaborative relays achieve the benefits of spatial diversity without requiring physical antenna arrays at the end devices.
Although many studies have demonstrated its effectiveness in an isolated single source-destination system, applying cooperative
relays to a large-scale wireless network remains challenging. We show that a large wireless system with cooperative relays can be
penalized by the elevated level of interference it produces. By examining the interdependency between interference management
and cooperative relay strategies, the penalty is modeled by an increase in spectrum resource usage and translated into a penalty
on single-link throughput. This throughput penalty serves as a reference for designing collaborative relays in isolated scenarios.
To mitigate the impact of interference, we present two channel allocation mechanisms for collaborative relay systems with different
fairness and utilization trade-offs. Simulation results confirm our analytical findings and demonstrate the effectiveness of our proposed
channel allocation mechanisms.
1 INTRODUCTION
TABLE 1
Summary of Relay Strategies
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
726 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
where
xy
fðx; yÞ :¼ :
xþyþ1
Outage capacity at an outage probability of
CO;DT ¼ log 2s;d þ 1 ;
pffiffiffiffiffiffi ð2Þ
CO;AF RAKE ¼ logð þ 1Þ;
2s;r þ2r;d
where 1
:¼ 212 2s;r 2r;d
, and 2i;j :¼ Eðjhij ðtÞj2 Þ. hij ðtÞ repre-
s;d
sents the channel response from node i to j at time t that
captures the effects of path loss, shadowing, and fading.
Table 1 provides a summary of various relay strategies.
Note that the capacities of AF-TDD are half of those of
AF-RAKE since AF-TDD splits the transmissions into
two time slots. The throughput of AF-FDD is the same as
that of AF-RAKE, assuming that in AF-RAKE, nodes S, R, Fig. 2. Throughput of different cooperative strategies for W ¼ 1.
(a) Outage ( ¼ 10 percent) throughtput. (b) Ergodic throughout.
and D all operate on a single channel of bandwidth W ,
whereas in AF-FDD, they operate on two channels, each
of bandwidth W . Therefore, the per-band throughput of interference models, the Physical Model and the Protocol
AF-FDD is half of that of AF-RAKE. Fig. 2 plots the Model, to characterize radio transmissions in the presence of
throughput of different strategies as a function of , interference. The physical model was introduced in [18] to
assuming that W ¼ 1 and all channels follow Rayleigh fading integrate the impact of channel fading, whereas the protocol
with 2i;j ¼ 1. We see that AF-RAKE and AF-FDD signifi- model is a simplified alternative [19], [20].
cantly outperform DT, and AF-TDD is slightly beneficial in Physical model. Let SS denote the set of wireless nodes
terms of outage throughput for a small signal-to-noise ratio. that communicate on the same channel.1 When node i0
Overall, AF schemes achieve a higher performance improve- communicates to node i, the received signal-to-interference-
ment in terms of outage capacity. This is because transmitting noise ratio at node i is defined as
via two independent paths can significantly improve the
packet delivery rate. For more details about ergodic and Pi0 jhii0 ðtÞj2 d
ii0
outage capacities, please see [17, Chapter 5] and [2]. SINRi ðtÞ ¼ ;
2 þ Ii ðtÞ
X 2 ð3Þ
Ii ðtÞ ¼ Pj hij ðtÞ d
ij ;
3 MODELING COLLABORATIVE RELAY IN AD HOC j2S
S;j6¼i0
NETWORKS
Although the results in Fig. 2 demonstrate the effectiveness where Pj is the transmit power at neighboring node j,
of collaborative relays in a network of three nodes, 2 represents the thermal noise variance, hij ðtÞ represents the
deploying them in a large wireless network must consider small-scale channel fading from transmitter j to receiver i
the impact of elevated interference from relaying. In this at time t, and dij is the distance between i and j. The
section, we present an analytical model on the network- large-scale fading (path loss) is approximated by d ij , where
wide interference. We start with describing two general ¼ 2 for free space propagation, and > 2 for general
interference models and extend them to collaborative-relay- environments [18], [21]. Each link’s transmission can be
based networks. successfully decoded only if its received SINR is above a
3.1 General Interference Model given threshold SINR0 for a given transmission rate.
We consider a large-scale network with static nodes and 1. We assume that each source only communicates with one destination
single-hop transmissions. We use two receiver-driven node and vice versa.
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
ZHU AND ZHENG: UNDERSTANDING THE IMPACT OF INTERFERENCE ON COLLABORATIVE RELAYS 727
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
728 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
ZHU AND ZHENG: UNDERSTANDING THE IMPACT OF INTERFERENCE ON COLLABORATIVE RELAYS 729
channels is small and hence some links are blocked, we need this graph is to associate each node with a disk that is
to choose the links intelligently to maximize system utility. centered on the node and with radius d=2. Two vertices
Next, we present two different allocation strategies that connect if and only if the two corresponding disks overlap. A
assign different priorities to direct and relay links: simple geometric argument can justify the equivalency
between these two constructions. Additional information
1. Collaborative allocation (CA). This allocation on unit disk graphs can be found in [28]. Next, we show that
scheme prioritizes direct links over relay links by one can approximate the interference condition of a
performing allocation sequentially starting from collaborative relay network by a unit disk graph.
direct links. The system first allows direct links to
negotiate for channel usage via the schemes in [25] 5.2 A Simplified Model of Large-Scale Random
and [26]. Only direct links with channel allocated Networks
can set up their relay links. Using AF-RAKE, a relay In Section 4, we show that each collaborative relay or DT
is set up only if the channel assigned to the direct has an interference-sensitive region. However, the size of
link is available at the relay link, that is, it does not the disk depends on the distance between source, relay, and
conflict with neighboring links with channel as- destination nodes. To make the analysis tractable, we
signed. Using AF-FDD, a relay is set up only if a consider the worst case scenario: each transmission pair is
channel different from that of the direct link is of the maximum distance apart. This produces the largest
available. Overall, using such sequential assignment, interference-sensitive region. Again, we assume that each
the CA scheme is fairness driven, that is, it attempts node has the same interference-sensitive range rI .
to minimize the impact of relay links on direct links. Under these assumptions, we show that one can model
Although minimizing the probability of blocking the DT and the collaborative relay transmission using a unit
DTs, this strategy may not be the optimal strategy disk graph. To do so, we first choose the center of the link as
for maximizing system throughput. the disk center. We approximate the interference-sensitive
2. Noncollaborative allocation (NCA). This scheme region with a large disk of radius X, and the final unit
performs channel allocation at the transmission set disk has a radius of X=2, which follows from the
level. A collaborative relay set consists of both direct second construction of a unit disk graph described in
and relay links. For AF-RAKE, each set is assigned Section 5.1. Next, we illustrate the process for both DT and
with one channel. If there is no channel available, collaborative relay transmission.
both links are disabled. For AF-FDD, each set is In a DT, source node S and destination node D are of
distance rC apart, where rC is the communication range. For
assigned with two channels. If there is no channel
any two links, if their medians are rI þ rC apart, then they
available, both links are disabled. If there is only
will not conflict. Therefore, we can approximate the
one channel available, only the direct link will be
interference-sensitive region of link S D by an open disk
set up. This scheme attempts to maximize system
with radius dd :
throughput at the cost of blocking some direct links.
In addition to using a conflict-free channel allocation, we dd ðrC Þ ¼ rI þ rC ¼ ð þ 1ÞrC ; ð9Þ
can also implement CA and NCA strategies using channel
where ¼ rI =rC , hereby referred to as the interference-
contention protocols. In multichannel MAC protocols, links
sensitive ratio. Thus, the disk associated with the S D link
contend for channels. CA can assign a higher priority in has a radius of dd =2.
terms of a shorter contention window, whereas NCA In the corresponding conflict graph, each link ðS; DÞ is
requires that both direct and relay links contend as a set. represented by a disk of radius dd =2 centered at the median
location between S and D. dd represents the maximum
5 ASYMPTOTIC ANALYSIS OF separation between the centers of two disks where each
transmission will not be disrupted by any transmitter of the
COLLABORATIVE RELAYS
other disk. Any two links could conflict if their correspond-
In this section, we analyze the effectiveness of collaborative ing disks overlap, that is, the centers of both disks are no
relays in large-scale wireless networks. We explore the more than dd apart.
asymptotic trend, assuming a large-scale random wireless Similarly, when S and D are assisted by a relay node R,
network where each node communicates with one of its where dðS; RÞ ¼ r1 and dðR; DÞ ¼ r2 , we can approximate
farthest neighbors at the edge of communication range. We the interference-sensitive region as a disk of radius3
reduce the network into a unit disk graph, model the penalty
of expanded interference by an increase in channel resource dcoop ðrC ; r1 ; r2 Þ ¼
8
usage, and translate it into a penalty of single-link capacity. >
> r þ rC : r21 þ r22 r2C ;
This penalty can be used to design a cooperative relay < I sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
2
strategy that takes into account the impact of interference. > 1 ðr21 þr22 r2C Þ
>
: r I þ 2 r C 1 þ 2 þ 1 : otherwise:
4r21 r2C ðr2C þr21 r22 Þ
5.1 Unit Disk Graph Model
ð10Þ
We start with a brief discussion on the unit disk graph model
used for our analysis. Assuming a network of n randomly
deployed nodes, we can build a unit disk graph as follows: 3. We note that in practice, relaying leads to an increase in the transmit
power density, which directly affects rI . For simplicity, we assume that rI
each node maps to a vertex; two vertices connect if and only remains the same and focus on the expansion of the overall interference-
if they are within a distance of d. An alternative way to build sensitive region.
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
730 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
ð þ 1Þ2
: 2=2 :
approximate the interference-sensitive region of the relay
link with an open disk of radius dcoop : ð13Þ
8 pffiffiffi One observation p from
<ð þ 1ÞrC : 1=2 < < 2=2; ffiffiffi (13) is that when the relay distance
dcoop ¼ 2
pffiffiffi ð11Þ factor is less than 2=2, collaborative relay does not result
: þ pffiffiffiffiffiffiffiffiffiffiffi
2
þ 12 rC : 2=2 : in extra interference. However, the trade-off is that
2 1=4
restricting the relay selection to such a small area will limit
As a result, the disk associated with the cooperative relay the possibility of finding a relay. Finally, we recognize that
link is centered at the median of S, D, and R and with
the model simplification in Figs. 5 and pffiffiffi 6 can heavily impact
radius dcoop =2.
this conclusion, at least the value of 2=2. Nevertheless, this
Using the above disk model, we can characterize a large
random network using a random disk graph. Each vertex trend indicates the advantage of using relay nodes in close
represents a randomly deployed link (direct or collaborative distance.
relay), located at the median. Each vertex is associated with 5.4 Effectiveness of Cooperative Relay
a disk of radius dd =2 or dcoop =2, where if two disks overlap,
the corresponding vertices conflict, and an edge exists When the amount of spectrum resource is fixed, using an
between the two. When a network contains only direct links extra number of channels implies reduced per-channel
or only collaborative relay links, the corresponding random resource. We use this metric to evaluate the penalty of
disk graph becomes a random unit disk graph. elevated interference. Next, we integrate this penalty
measure with the per-link relay gain to compare the overall
5.3 Asymptotic Resource Usage of a Random system throughput with and without collaborative relays.
Network Let the system throughput S be the sum of individual
Using existing results of random unit disk graphs [29], we user throughput. Because each transmission is of identical
can derive the asymptotic number of channels required to length, the overall system throughput is the product of the
achieve a conflict-free transmission.
throughput per-resource and the number of transmissions:
Theorem 1 [29]. Suppose n disks with diameter dðnÞ are
uniformly distributed in a unit area. Let d ¼ dðnÞ satisfy B
SDT ðnL Þ ¼ nL CDT ;
d ! 0 and d2 n= lnðnÞ ! 1 as n ! 1. Let
ðnÞ represent the
D ðnL Þ
minimum number of channels required to assign each disk B
SAF RAKE ðnL Þ ¼ nL CAF RAKE ; ð14Þ
with a channel without any conflict. Let
C ðnL Þ
2 B
kðnÞ ¼
d n: ð12Þ SAF F DD ðnL Þ ¼ nL CAF RAKE ;
4 2
C ðnL Þ
pffiffiffi
Then, as n ! 1,
ðnÞ=kðnÞ ! 2 3= almost surely.5 where B represents the total amount of available resource,
and CDT , CAF RAKE , and CAF F DD represent the capacity of
We define
D ðnL Þ as the minimum number of channels DT, AF-RAKE, and AF-FDD defined in Section 2, respec-
required when there are nL DT links only and
C ðnL Þ as tively. As mentioned before, AF-FDD requires 2 channels
the minimum number of channels for nL collaborative compared to AF-RAKE.
relay links only. Because
ðnÞ is a random variable, so are For AF-RAKE-based collaborative relays to be effective,
D ðnL Þ and
C ðnL Þ. Hence, in the following discussions, the that is, having a higher system throughput compared to that
4. We assume that relay nodes must be fairly close to its source and
of DT, we require
destination nodes, . On the other hand, must be large enough to
allow a sufficient probability of finding a relay. SAF RAKE ðnL Þ
5. “Almost surely” means that as n goes to infinity, the event happens
1: ð15Þ
SDT ðnL Þ
with a probability of 1 [30, Chapter 6].
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
ZHU AND ZHENG: UNDERSTANDING THE IMPACT OF INTERFERENCE ON COLLABORATIVE RELAYS 731
TABLE 2
Numerical Values of Critical Ratio ð ¼ 1Þ
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
732 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
TABLE 3
The Number of Channels Required to
Achieve a Conflict-Free Allocation
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
ZHU AND ZHENG: UNDERSTANDING THE IMPACT OF INTERFERENCE ON COLLABORATIVE RELAYS 733
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
734 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
Fig. 10. Cooperation usage and link success statistics for systems with Fig. 11. System ergodic throughput and outage throughput as a
two channels. (a) Probability of direct link failure. (b) Probability of relay function of the number of channels assuming 160 active direct links.
usage. (a) Ergodic throughput. (b) Outage throughput.
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
ZHU AND ZHENG: UNDERSTANDING THE IMPACT OF INTERFERENCE ON COLLABORATIVE RELAYS 735
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.
736 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 7, NO. 6, JUNE 2008
Yan Zhu received the BE and MS degrees from Haitao Zheng received the BS degree (with
the Electronic Engineering Department, Tsin- highest honor) from Xian Jiaotong University in
ghua University, in July 2002 and July 2005, July 1995, and the MSEE and PhD degrees in
respectively. Since September 2005, he has electrical and computer engineering from the
been pursuing the PhD degree in the Depart- University of Maryland, College Park, in May
ment of Electrical Engineering and Computer 1998 and July 1999, respectively. Since Sep-
Science, Northwestern University, Evanston, tember 2005, she has been an assistant
Illinois. His research interests include wireless professor in the Computer Science Department,
communications, information theory, communi- University of California at Santa Barbara. Her
cation network, and signal processing. His research interests include wireless systems and
current research focuses on the interference problems in wireless networking and multimedia computing. She was recently named as the
systems, including interference management and fundamental limita- 2005 MIT Technology Review top 35 innovators under the age of 35 for
tions of interference channels. He is a student member of the IEEE. her work on cognitive radios. Her work was selected by the MIT
Technology Review as one of the 10 emerging technologies in 2006.
She also received the best student paper award in the 2007 IEEE
Symposium on New Frontiers in Dynamic Spectrum Access Networks
(DySPAN), the 2002 Bell Laboratories President’s Gold Award from
Lucent Bell-Labs, and the 1998-1999 George Harhalakis Outstanding
Graduate Student Award from the Institute of System Research,
University of Maryland. She is a member of the IEEE.
Authorized licensed use limited to: TU Delft Library. Downloaded on June 22,2010 at 09:39:10 UTC from IEEE Xplore. Restrictions apply.