Sie sind auf Seite 1von 6

2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

Multi-Pair Full-Duplex Amplify-and-Forward


Relaying with Very Large Antenna Arrays
Xiaochen Xia, Wei Xie, Dongmei Zhang, Kui Xu, and Youyun Xu
College of Communication Engineering, PLA University of Science and Technology, Nanjing, China
Email: tjuxiaochen@gmail.com

AbstractThis paper considers the multi-pair full-duplex mas- HEI


sive MIMO relaying, where multiple source-destination pairs
communicate simultaneously with a common full-duplex amplify-
HSR HRD
and-forward relay. The low-complexity transceiver design at the
relay based on zero-forcing (ZF) processing is presented. The R
asymptotic expression of the end-to-end signal-to-interference- S1 D1
plus-noise power (SINR) is derived under the general power
scaling scheme. Through the theoretical results, we show that the
deployment of very large antenna arrays at the relay has potential
D2
to eliminate the effect of echo interference due to the full-duplex S2
operation, if the power scaling scheme is properly selected.
Moreover, it is shown that the multi-pair full-duplex relaying
outperforms the multi-pair half-duplex relaying in the spectral
and energy efficiencies when the number of relay receive/transmit SK DK
antennas is large. HSD
Index TermsMulti-pair full-duplex relaying, amplify-and-
forward, very large antenna arrays, asymptotic analysis. Fig. 1: The FDR network with source-destination pairs.

has received considerable attentions recently. In FDR, the relay


I. I NTRODUCTION
transmits and receives simultaneously at the same frequency
In multi-user MIMO systems, one main challenge is the and time resource, but at the cost of a strong echo interference
increased complexity and energy consumption of the signal due to signal leakage between the relay output and input
processing to mitigate the interference between multiple co- [5]. Inspired by the recent echo interference cancellation
channel users. To deal with these problems, the multi-user techniques [5], several works have dedicated to the study of
MIMO system with very large antenna arrays (also known as FDR protocol. For example, the distributed beamforming and
massive MIMO system) has been advocated recently in [1]- power allocation scheme for FDR was proposed in [6], [7].
[2]. The key result is that, with very large antenna arrays at The joint beamforming and antenna subset selection scheme
each base station, both the intracell and intercell interferences for MIMO FDR was considered in [8]. It was shown in [6]-[8]
can be substantially reduced with simple linear processing. that the FDR, although cannot double the spectral efficiency,
Moreover, the transmit powers of the reverse (uplink) and does provide considerable performance gain when compared
forward (downlink) links can be made arbitrarily small as the with the traditional HDR. As a result, it is not surprising
number of base station antennas approaches to infinity, while that the FDR combined with massive MIMO is a promising
a deterministic signal-to-interference-plus-noise (SINR) which candidate for the development of future energy and spectral
depends only on the large-scale fading of channels is achieved. efficient cellular system.
On the parallel avenue, relaying has been accepted by In this paper, we consider the multi-pair full-duplex massive
several standards such as IEEE 802.11s, IEEE 802.16j and MIMO relaying (multi-pair FDR for short) where multiple
LTE-Advanced as a powerful technique to provide spatial source-destination pairs communicate simultaneously with the
diversity and extend the coverage of the wireless networks. help of a common full-duplex amplify-and-forward (AF) relay.
A large number of existing works on cooperative communi- The low-complexity transceiver design at the relay based on
cations considered the half-duplex relaying (HDR), where the zero-forcing (ZF) processing is presented. The asymptotic
relay transmits and receives using the orthogonal frequency expression of the end-to-end signal-to-interference-plus-noise
or time resources [3]-[4]. However, one main drawback of power (SINR) is derived. Different from the majority of the
such system is the inherent spectral efficiency loss of the previous works [2], [9], the general power scaling scheme is
half-duplex operation, since the required channels for data considered. We show that the effect of echo interference can
transmission from source to destination are doubled due to be eliminated through the large antenna arrays of the relay, if
the half-duplex constraint of the relay. the power scaling scheme is properly chosen. Moreover, the
Full-duplex relaying (FDR) is a promising approach to considered multi-pair FDR outperforms the multi-pair HDR
improve the spectral efficiency of cooperative network, which [9] when the number of relay receive/transmit antennas is

978-1-4799-8406-0/15/$31.00 2015 IEEE 304


2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

large. The rest of this paper is organized as follow. In section where indicates processing delay at . Substituting (2) into
II, we will describe the system model and some preliminary (1), the received signal at the relay can be rewritten as
knowledge. The relay transceiver design is considered in (
)

section III. Simulation results are presented in section IV and y [] = W H W
some conclusions will be drawn in the last section. (
=0 ) (3)

Notations: () denotes the expectation. means propor- W H s [ ] + W n [ ]
tional to. I is used for identity matrix. () , () and Meanwhile, the received signal of the th destination at time
() are used for conjugate, transpose and conjugate-transpose, instant can be expressed as
respectively. (A), (A) and Tr{A} denote the th eigen-
value, spectral radius and trace of matrix A, respectively. [] = h, x [] + h, x [] + []

= h, W W H x [ ]
II. S YSTEM M ODEL AND P RELIMINARIES
+ h, W W n [ ] + h, x [] + []

( )
A. Multi-Pair Full-Duplex AF Relaying Model

Consider the network with source-destination pairs and + h, W W
H W
a single full-duplex AF relay, as illustrated in Fig. 1, where ( =1 )

the source wishes to communicate with the destination W H x [ ( + 1) ] + W n [ ( + 1) ]
( {1, , }) with the help of the relay . It is assumed (4)
that the sources and destinations are equipped with single where [] denotes the AWGN at and has distribution
antenna due to the size limitation, while the relay is equipped (0, 1). Herein, the signal from the direct channel is treated
with 2 antennas ( for reception and for transmission). as interference due to the lack of the CSI of direct channel.
Let H = [h,1 , , h, ] denote the chan- Note that the received signals at the relay and destinations
nel matrix between all sources and the receive antenna array involve the infinitely repeating echo terms due to the full-
of the relay and let H = [h,1 , , h, ] duplex operation. Therefore, gain control is needed [10] at
denote the channel matrix between all destinations and the to prevent the relay and destinations oscillation (i.e., the
transmit antenna array of the relay. H ( {, }) is received signal powers at the and grow without bound).
1/2
modeled as H = G D , where G charac- This problem will be considered in section III.
terizes the small-scale fading of the channel whose entries B. Review of Some Limit Results
are independent identically distributed (i.i.d) with distribution
Lemma 1. Let p = [1 , , ] and q = [1 , , ]
(0, 1), and D is a diagonal matrix with
denote two mutually independent 1 random vectors, whose
[D ], = , representing the large-scale fading coefficients.
elements are i.i.d. zero-mean random variables with variances
Moreover, H denotes the echo interference 2 2
[ ] = and [ ] = , then according to the law of
channel matrix between the transmit and receive arrays of the
large numbers, we have
relay, whose entries are i.i.d with distribution (0, ).
1 .. 1 ..
H = [h,1 , , h, ] denotes the chan- p p and p q 0, as (5)
nel matrix between all the sources and destinations with
..
[H ], (0, ,, ) representing the fading channel where means the almost sure convergence. Moreover, from
between the th source and th destination. the Lindeberg-levy central limit theorem, we have
At time instant , all sources transmit their unit-power 1
p q (0, ) , as (6)
symbols , [] ( = 1, 2, , ) to the relay simultaneously
and the relay forwards the amplified signal to the destinations.
where means convergence in distribution.
The received signal at the relay can be expressed as
( ) Lemma 2. [11, Lemma 1] If S denotes a deterministic
y [] = W H x []+ H x []+n [] (1) matrix with uniformly bounded spectral radius for all . Let

where and denote the transmit powers of sources and p = 1 [1 , , ] and q = 1 [1 , , ] denote two

relay, respectively. x [] = [,1 [] , , , []] . n [] mutually independent 1 random vectors, whose elements
(0, I ) denotes the additive white Gaussian noise (AWGN) are i.i.d. zero-mean random variables with unit variance and
vectors at the relay. Moreover, x [] denotes the transmit finite eighth moment. Then,
vector of relay with unit-power, i.e., [Tr{x [] x []}] = 1, .. 1 ..
at time instant . Here the expectation is over the transmit p Sp Tr {S} and p Sq as (7)

vector x [] and AWGN. W = [w,1 , , w, ] denotes III. T RANSCEIVER D ESIGN AND A SYMPTOTIC A NALYSIS
the receive beamforming matrix at the relay. Then the relay
retransmits the received signal using the transmit beamforming This section first presents the beamforming scheme at the re-
matrix W = [w,1 , , w, ]. As a result, the transmit lay under the assumption that the relay has perfect knowledge
vector of the relay x [] can be expressed as of useful channels (i.e., H and H ) and no knowledge
of echo interference channel. Then the asymptotic end-to-
x [] = W y [ ] (2) end SINR of multi-pair FDR with the proposed beamforming

305
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

scheme is analyzed under the general power scaling scheme at TABLE I: Power of transmit vector at the relay
the sources and relay, i.e., = / and = / , Number of antennas ( ) 100 200 300 400
where and are fixed and , 0. = = = 0 1.13 1.06 1.03 1.02
= = = 0.5 1.14 1.06 1.04 1.02
A. Relay Transceiver Design = = = 1 1.21 1.09 1.06 1.04

In beamforming scheme considered in [7], a ZF constraint, TABLE II: Violation probability of sufficient condition

i.e., W H W = 0, is added at the relay to eliminate the Number of antennas ( ) 50 60 70 80 90
echo interference completely by exploiting the spacial degree = = = 0 0.46 0.088 6.4e-3 6.4e-4 1.5e-5
= = = 0.5 0.54 0.10 1.7e-2 8.7e-4 3.9e-5
of freedom provided by the relay antenna arrays. However, a = = = 1 0.52 0.16 3.5e-2 5e-4 6.3e-4
main drawback of such scheme is the need of the knowledge
for echo interference channel H to compute the transmit In Constraint 1 and Constraint 2, the convergence of re-
beamforming matrix W . Since H is an matrix, ceived signal powers at the destinations, i.e, [ []
[]],

the learning of H requires the training sequence at with is not considered. In fact, a sufficient condition for the conver-
length greater than or equal to [2]. This task becomes gence of [ []
[]] is [Tr{y [ ] y [ ]}] <

impossible for the relay with very large antenna arrays since
. To see this, we rewrite [ [] []] using the property
the duration of channel coherent time is fixed. As a result, Tr{AB} = Tr{BA} as
in this paper we assume the relay has perfect knowledge of [ ]

the useful channels H and H , but no knowledge of [] []
H . We modify the ZFR/ZFT scheme considered in [9] to {
[ ]}
eliminate the inter-pair interference. The receive and transmit = Tr W h, h, W y [ ] y
[ ]
2
beamforming matrices at the relay can be expressed as1 + h, + 1
( )1 ( )1 (12)
W = H H H , W = H H H Since both W h, h, W and [y [ ] y
[ ]]
(8) are positive semi-definite matrices, it follows from (22) in the
where is chosen to satisfy the following two constraints: Appendix that
[ ] { }
Constraint 1. must be chosen to guarantee the unit-power
[] [] Tr W h , h , W
transmit signal constraint at the relay, i.e., { [ ]} (13)
2
[ { }] Tr y [ ] y [ ] + h, + 1
Tr x [] x [] = 1 (9)
From (13), it is sufficient to determine that satisfies the
Constraint 2. must be chosen to prevent relay oscillation Constraint 1 and Constraint 2.
(i.e., the received signal power at the relay goes into infinity)
[ { }] Lemma 5. As approaches to infinity, that satisfies

Tr y [ ] y [ ] < (10) Constraint 1 and Constraint 2 with power scaling scheme
= / and = / can be expressed as
The expectations in (9) and (10) are over the transmit vector
..
x [] and AWGN. Although it is difficult to determine the

exact that satisfies the Constraint 1 and Constraint 2, we 1 ( ) (14)
present the following lemmas to give some insights. 1
Tr{ D +D D }+
1 1 1 1 1
, ,
=1
Lemma 3. For arbitrary , the necessary and sufficient con-
dition for the convergence of (10) is [Tr{x [] x []}] < . Proof: See Appendix.
In practice, the number of relay receive/transmit antennas
Proof: See Appendix. cannot goes into infinity. For the practical implementation
From Lemma 3, any that results in finite transmit signal with finite , the expression of should be determined
power at the relay also ensures Constraint 2. This result under Constraint 1 and Constraint 2. Unfortunately, this task is
facilitates the asymptotic analysis in the following. challenging since the knowledge of echo interference channel
Lemma 4. For arbitrary , the sufficient condition that is not available. To deal with this problem, we advocate the
ensures the convergence of (10) can be expressed as scheme in which the asymptotic expression of given by
Lemma 5 is employed in the computation of beamforming
1
< ( 1 2 1
)
(11) matrix. This scheme is reasonable since the scenario of interest
(H H ) Z (H H ) Z(H H )
in this paper is the case with very large relay receive/transmit

where Z is defined as Z = H H H . antenna arrays. The power of x [] and the violation probabil-
ity of the sufficient condition in Lemma 4 when the asymptotic
Proof: See Appendix. expression of is used are depicted in Table I and II, where
1 The presented scheme differs from the traditional ZFR/ZFT scheme [9]
= 10 and the simulation setup is the same with that in
since the derivation of the normalized factor in the multi-pair FDR is very section IV. We can see that the power of x [] is only slightly
different from that in [9], which will be seen in the below. larger than 1 and it approaches to 1 as increasing. Moreover,

306
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

the violation probability of the sufficient condition is extremely Since (19) is in the similar form with the expression of
small even for the moderate case1 (e.g. = 90). [Tr{x [] x []}] given by (30), the asymptotic expression
2
B. Asymptotic Expression of End-to-end SINR of [ ] can be derived using the similar approach as in
the derivation of (35), which can be expressed as in (18). The
Substituting (8) into (4), the received signal at can be
detailed derivation is omitted for brevity.
rewritten as
, [] = , [ ] + Finally, substituting the asymptotic expression of the ampli-
( )1 fied AWGN (which can be readily derived by using Lemma
+ e H H H n [ ] (15) 1 on the fourth term of right-hand side of (15)) and (18) into
(15), the asymptotic expression of SINR is obtained.
+ h, x [] + []
An interesting observation from Theorem 1 is that, as a
where e is the all-zero vector except for the th component
merit of using very large antenna arrays at the relay, the effect
which is equal to 1. represents the infinitely repeating
of echo interference due to the full-duplex operation can be
echo interference term, which can be expressed as
( )1 eliminated, if the power scaling scheme is properly selected.
= h, H H H To see this, we first note from Theorem 1 that the powers of
( )1 )
( )1 (

echo interference and amplified AWGN can be expressed as
H H Z H H [ 2 ]
(=1 )1 (16) (12 )max{1 ,0}
H H H (20)
2 1
( ) 1 (/ ) , (1 )max{1 ,0}

H x [ ( + 1) ] + n [ ( + 1) ] respectively. (20) shows that, with very large antenna array
With (15), (16) and Lemma 5, the asymptotic expression of at the relay, the effect of echo interference on the end-to-end
the end-to-end SINR at is given by the following theorem. SINR becomes negligible as long as > 0. The result is
quite attractive since the effect of echo interference is simply
Theorem 1. With perfect knowledge of useful channels, the eliminated without the knowledge echo interference channel
end-to-end SINR at with power scaling scheme = and other active echo interference cancellation schemes (e.g.,
/ and = / can be expressed asymptotically digital and analog cancellation schemes [5]).
(in ) as
= IV. S IMULATION R ESULTS
2 In this section, we examine the spectral and energy effi-
2 (/ )
[ 2 ]
+ 1 (/ ) ,
2 1 2
+ h, +1
ciencies
of the multi-pair FDR which are
1
defined as =
=1 log2 (1 + ) and = + =1 log2 (1 + ),
(17) respectively [2]. The spectral and energy efficiencies of the
Wherein, is given by Lemma 5 and the echo interference
power can be expressed asymptotically (in ) as multi-pair HDR considered in [9] are also simulated for
[ 2 ] .. ( )4 2 12 1 comparison. Since the sources and relay transmit only half

, of the time in HDR, the transmit powers of and are

{ 1 } set to 2 and 2 to make a fair comparison. Moreover, it
Tr D1 1 1
+ D D (18) is assumed that in multi-pair HDR, the relay uses the same
1
( )2 numbers of transmit and receive antennas as that in multi-
1 1
1 =1 , , pair FDR. Similar to [2], [9], D and D are set to
where the expectation is over the transmit vector x [] and D = D = I for convenience.
AWGN. Fig. 2 shows the spectral efficiencies of multi-pair FDR and
HDR as a function of . We can see from the figure that
Proof: Using the property Tr {AB} = Tr {BA} and (5)
the asymptotic expression derived in section III can predict
in Lemma 1, the instantaneous echo interference power can
the performance of multi-pair FDR precisely as > 150.
be expressed as
[ 2 ] Moreover, it is seen that the multi-pair FDR outperforms multi-
pair HDR in spectral efficiency for the large scenario and
{( ) the converse is true as is small. As and increasing, the
.. ( )2 1
1 1
Tr I + D required number of relay receive/transmit antennas for multi-

( 1
=1
) ( )}
1 pair FDR to achieve a performance gain compared with multi-
D1 Z D e e
D 1
ZD pair HDR decreases. This is because that the effect of echo
(19) interference in multi-pair FDR becomes negligible in this case
1 Thanks to this result, to eliminate the relay oscillation in practi- as suggested by (20).
cal massive MIMO relaying network, a hard threshold can be added to Fig. 3 depicts the energy efficiencies of multi-pair FDR and
cut down the amplitude of x [] as it becomes very large, and then
HDR as a function of , where = 10 source-destination
[Tr{y [ ] y [ ]}] remains bounded from Lemma 3. The effect
of this operation on the system performance is negligible since the probability pairs are adopted in the simulation. Again, we can see that
of such event is extremely small even as = 90 (See Table II). the multi-pair FDR outperforms multi-pair HDR in energy

307
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

(
)2
{ } 2
Tr A2 = 2 (A) (A) = (Tr {A}) (21)
=1 =1
and similar result exists for matrix B, we have

Tr {AB} Tr {A2 } Tr {B2 } Tr {A} Tr {B} (22)
With (22), we are now ready to prove Lemma 3. First, if

[Tr{y [ ] y [ ]}] < , according to (2) we have
[ { }]

Tr x [] x []
[ { }]

= Tr W y [ ] y [ ] W (23)
{ } { [ ]}
Tr W W Tr y [ ] y
[ ] <
where the second step follows from Tr{AB} = Tr{BA} and
Fig. 2: The spectral efficiencies of multi-pair FDR and HDR (22). This completes the proof of necessity.
v.s. the number of relay receive/transmit antennas, where If [Tr{x [] x []}] < , substituting (3) into (10),

= 1, ,, = 0.1, = 10, = 5dB, = . [Tr{y [ ] y [ ]}] can be rewritten as
[ { }]
{ ( )1

Tr y [] y [] = H H H H
=0
( )1 )
( )1 (
H H Z H H
( ( )1 )
I + H H
( }
( )1 ( )1 )

H H Z H H
() { } [ { }]
Tr H H Tr x [] x []
(24)
where the step (a) follows from (22) and the property
Tr{AB} = Tr{BA}. This completes the proof of sufficiency.
Fig. 3: The energy efficiencies of multi-pair FDR and HDR v.s. B. Proof of Lemma 4
the number of relay receive/transmit antennas, where = 1, Substituting (3) into (10), we can obtain
,, = 0.1, = 10, = 5dB, = . [ { }]
{ ( ) }

Tr y [] y [] = Tr X1 X2 X1
2

efficiency for the large scenario and the multi-pair HDR { ( ) =0 }



2
achieves better performance as is small. Moreover, the = Tr X1 X1 X2
figure shows that the power scaling scheme which results in =0
(25)
the best spectral efficiency is not necessary to be optimal for
where X1 = (H H )1 Z(H H )1 and X2 =
the energy efficiency. Specifically, the power scaling scheme 1
with = = 0 (which is optimal in Fig. 2) achieves the I + (H H ) . Note that both (X1 ) X1 and X2 are
worst energy efficiency as shown in Fig. 3. positive semi-definite. Applying (22) on (25) results in
[ { }] {( ) }


V. C ONCLUSION Tr y [] y [] Tr {X2 } Tr X1 X1
2
This paper investigates the multi-pair full-duplex massive =0
MIMO relaying. The transceiver design at the relay is pre- (26)
sented and the asymptotic end-to-end SINR is analyzed for Moreover, since X1 X1 is positive semi-definite,

the limit of large . Our result shows that, when the transmit Tr{(X1 ) X1 } is upper bounded by
powers of sources and relay are scaled by 1/ and 1/ , {
} { 1
}
the sufficient condition that the effect of echo interference is Tr (X1 ) X1 = Tr (X1 ) X1 X1 X1
{ }1 (27)
eliminated as approaches to infinity is > 0. Moreover, 1 1
(X1 X1 )Tr (X1 ) X1
it is shown that the multi-pair FDR outperforms the multi-pair
HDR for the large scenario. Using (27) on (26) repeatedly, we can obtain
[ { }]

A. Proof of Lemma 3
A PPENDIX Tr y [ ] y [ ]
( ( )) (28)
In the space of positive semi-definite matrices, it is easy
Tr {X2 } 2 X1 X1
to show that trace is a proper inner-product. As a result, =0
obeys the Cauchy-Schwarz inequality, i.e., Tr {AB}
it From (28), we can see that the sufficient condition that
Tr {A2 } Tr {B2 }, where the matrices A and B are
ensures the convergence of [Tr{y [ ] y [ ]}] can
positive semi-definite. Moreover, since be expressed as in (11).

308
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals

( 1 1 1 ) ( )1
Z D D D Z
C. Proof of Lemma 5 3 D1

1
2 D ZD
1
Substituting (2) and (3) into (9), and using the property ( )
1 1
Tr{AB} = Tr{BA}, [Tr{x [] x []}] can be rewritten as .. 2
1+ 1
, ,
[ { }] {( ( )1 ) =1
( )1 ( )1
Tr x [] x [] =2 Tr I+ H H
2 D1 1
D1
1 1
Z D 2 D ZD
=0 ( ( ))
( )1 )
( )1 (
.. 2 1 1
H H
Z H H
1+
1
, , D1


( ) } =1
( )1 ( )1 ( )1 (34)
H H H H Z H H where the last step is obtained by using the second step of (34)
repeatedly. Substituting (34) into (30) and using the equation
(29) = 1 , it can be shown that (30) can be rewritten as
=0 1
Using (5) in Lemma 1 on (29), we have [ { }]
[ { }] .. ( )2 { }
.. Tr x [] x [] Tr 1 D1 1 1
+D D
Tr x [] x []
1
( )2
{( 1 ) 2

1 1( ) ( =1 , , )
1 1

Tr
I +D
[ { }] (35)
( =0 ) ( )
1 1 1 1 1 Setting Tr x [] x [] = 1, one can obtain the asymp-
2 D Z D D 2 D ZD
totic expression of in Lemma 5. Finally, according to Lemma
(30)
3, the resulting also satisfies Constraint 2.
Recall that Z = H H H , it can be shown that
[ 1 1 1 ] ACKNOWLEDGMENT
1
3 Z D D
D Z ,
This work is supported by Major Research Plan of National
2 1
Natural Science Foundation of China (No. 91438115), Na-
= 13 , , h, H h, h, H h, tional Natural Science Foundation of China (No. 61371123,
=1 No. 61301165), Jiangsu Province Natural Science Founda-
()

2 1 h,
2

= ,
, 3 h , H U diag (e 1 ) UH h
,
tion (BK2011002, BK2012055), China Postdoctoral Science
=1 Foundation (2014M552612) and Jiangsu Postdoctoral Science

2 1 h, 2
= , , 3 h , H diag (e 1 ) H h
,
Foundation (No.1401178C).
=1
R EFERENCES
, , h, 2 h ,1
2

[1] T. L. Marzetta, Noncooperative cellular wireless with unlimited numbers
= 2 2 h, h ,1 h ,1 h, of BS antennas, IEEE Trans. Wireless Commun., vol. 9, no. 11, pp. 3590-
, ,
=1 3600, Nov. 2010.
( ) (31) [2] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, Energy and spectral effi-
where h, = h, / , , U is an unitary ciency of very large multiuser MIMO systems, IEEE Trans. Commun.,
matrix, H = H U , h ,1 is the first column of H vol. 61, no. 4, pp. 1436-1449, 2013.
[3] Y. Xu, X. Xia, K. Xu, and D. Zhang, On the hybrid relaying protocol for
and h ,1 = h ,1 / h ,1 . The step (a) is based on the time division broadcasting, to appear in Trans. Emerging Telecommun.
eigenvalue decomposition and the rank-1 property of matrix Tech., 2013.

(h, /h, ) (h, /h, ) . [4] X. Xia, D. Zhang, K. Xu, and Y. Xu, Interference-limited two-way DF
relaying: symbol-error-rate analysis and comparison, to appear in IEEE
Note that the spectral radius of h ,1 h ,1 is Trans. Veh. Tech., 2013.
[5] D. Bharadia, E. McMilin, and S. Katti, Full duplex radios, ACM
(h ,1 h ,1 ) = Tr{h ,1 h ,1 } = 1 for arbitrary Sigcomm 2013, Hong Kong, China, Aug. 2013
, it follows from Lemma 2 that [6] L. Jong-Ho and S. Oh-Soon, Full-Duplex Relay Based on Distributed
{ Beamforming in Multiuser MIMO Systems, IEEE Trans. Veh. Tech., vol.
1

.. , = 62, no. 4, pp. 1855-1860, 2013.
h, h ,1 h ,1 h, (32) [7] H. A. Suraweera, I. Krikidis, Z. Gan, Y. Chau, and P. J. Smith, Low-
0, = Complexity End-to-End Performance Optimization in MIMO Full-Duplex
Relay Systems, IEEE Trans. Wireless Commun., vol. 13, no. 2, pp. 913-
Substituting (32) into (31) and using (5) in Lemma 1, we can 927, 2014.
obtain [8] X. Xia, K. Xu, D. Zhang, and Y. Xu, Low-complexity transceiver design
1
[ 1 1 1 ] and antenna subset selection for sooperative half- and full-duplex relaying
3 Z D D D Z , systems To appear in IEEE Globecom 2014.

1

1 1
[9] H. A. Suraweera, N. Hien Quoc, T. Q. Duong, Y. Chau, and E. G. Larsson,
, , , , =
.. (33) Multi-pair amplify-and-forward relaying with very large antenna arrays,
=1 in IEEE ICC 2013, pp. 4635-4640, Jun. 2013.

0, = [10] T. Riihonen, S. Werner, and R. Wichman, Optimized gain control for
single-frequency relaying with loop interference, IEEE Trans. Wireless
According
[ to ] (33), we have Commun., vol. 8, no. 6, pp. 2801-2806, 2009.
1 1 1 1 .. 1 1 1
3 Z D D
D Z (
=1 , , )D
[11] J. Evans and D. N. C. Tse, Large system performance of linear
multiuser receivers in multipath fading channels, IEEE Trans. Info.
and Theory, vol. 46, no. 6, pp. 2059-2078, 2000.
( ) ( )
1 1 1 1 1
2 D Z D D 2 D ZD
( )1
2 1 1
= 1+ 2 D Z D D1

309

Das könnte Ihnen auch gefallen