Hybrid Mimo

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

10 IEEE COMMUNICATIONS LETTERS, VOL. 13, NO.

1, JANUARY 2009

Hybrid Zero-Forcing Beamforming/Orthogonal Beamforming with


User Selection for MIMO Broadcast Channels
Caihua Zhang, Wei Xu, Student Member, IEEE, and Ming Chen, Member, IEEE

Abstract—This letter proposes a hybrid transmission scheme introduced in [7] and later modified in [8]. Generally, it is
by combining zero-forcing beamforming (ZFBF) and orthogo- known that, for systems with low to moderate number of users,
nal beamforming (OBF) for the multiple-input multiple-output OBF outperforms ZFBF at low SNR whereas ZFBF performs
broadcast channel. This method first utilizes ZFBF for only a part
of the scheduled users with approximately orthogonal channels. better when SNR becomes large [8].
By further constructing orthogonal beamformers in the null This letter proposes a hybrid ZFBF/OBF (referred to as
space of the ZFBF user channels, it then employs OBF for the HBF) scheme to achieve promising performance in various
rest of users to fully exploit the spatial dimension. Theoretical scenarios. The essential idea behind our HBF scheme is
analysis, as well as simulation results, verify that our scheme to dynamically choose ZFBF or OBF for designing the
outperforms the conventional ZFBF scheme and OBF scheme
with user selection. beamformers of the scheduled users. The HBF scheme first
exploits ZFBF for only a part of the scheduled users with
Index Terms—Broadcast channel, multiple-input multiple- near-orthogonal channels. Then, to fully exploit the spatial
output, zero-forcing beamforming, orthogonal beamforming.
dimension, it selects several more users and assigns them
orthogonal beamformers which are chosen in the null space
I. I NTRODUCTION of the ZFBF user channels. The optimal HBF scheme selects
users and their beamforming strategies by exhaustive search.
R ECENTLY, multiple-input multiple-output (MIMO)
broadcast systems have attracted significant attention as
they provide both spatial multiplexing gain and multi-user
To reduce the complexity, we further propose an efficient sub-
optimal scheme based on greedy search.
Notation: We use uppercase and lowercase boldface letters
diversity gain [1]. Dirty paper coding (DPC) is a capacity-
to denote matrices and vectors, respectively. A† denotes the
achieving scheme for the MIMO broadcast channel (BC) [2].
conjugate transpose of matrix A. [A]i,j denotes the (i, j)th
For practical applications, however, DPC is difficult to imple-
element of matrix A. x denotes the Euclidean norm of
ment due to its high complexity. Therefore, linear beamform-
vector x. IM is the M × M identity matrix. |A| is the
ing is investigated as an alternative [3]–[9]. In particular, [3]
cardinality of set A.
proposes an iterative algorithm to find the optimal beamform-
ers that maximize the sum rate (RMBF). Due to the iterative
II. S YSTEM M ODEL
nature, RMBF is still of high complexity. Hence, a series of
more efficient beamforming strategies, including zero-forcing Consider a MIMO BC with an M -antenna base station
beamforming (ZFBF) [4]–[6], orthogonal beamforming (OBF) (BS) and K (K ≥ M ) single-antenna users. Denote the
[7], [8], and orthogonal random beamforming (ORBF) [9], are flat fading channel of user k as hk ∈ C1×M . Its entries
developed. are independent and identically complex Gaussian distributed
When there are many users in the system, user selection is with zero mean and unit variance. Assume that perfect chan-
an essential technique. It selects a subset of users for simul- nel state information of all users is known at the BS. Let
taneous transmission. In [5], [6], two efficient user selection A = {k1 , · · · , kM } ⊆ {1, · · · , K} be the set of scheduled
methods, namely semi-orthogonal user selection (SUS) and users. The received signal at user k ∈ A is
  
greedy user selection (GUS), are proposed for MIMO BCs yk = Pk hk wk sk + hk Pj wj sj + nk (1)
with ZFBF, respectively. Works in [5] show that ZFBF with j∈A,j=k
SUS (ZFBF-SUS) is able to achieve the optimal sum rate
asymptotically when the number of users approaches infinity. where Pk is the transmit power for user k, wk ∈ CM×1 is the
However, for systems with low to moderate numbers of users, corresponding unit-norm beamformer, sk is the normalized
its performance degrades, especially in the low SNR regime data symbol, and nk is the complex Gaussian noise with
[5], [7]. To improve the performance in this case, OBF is zero
mean and unit variance. The transmit power constraint
is k∈A Pk = P . For simplicity, equal power allocation
P
Manuscript received September 12, 2008. The associate editor coordinating is assumed, i.e. Pk = M , ∀k ∈ A. Thus, the signal-to-
the review of this letter and approving it for publication was F. Granelli. interference-plus-noise ratio (SINR) of user k is
This work was supported by the National 863 High Technology Development
P 2
Project (No. 2007AA01Z207 and No. 2007AA01Z268), Program for New
M |hk wk |
Century Excellent Talents in University, and Research Fund of National SINRk = P
 2 . (2)
Mobile Communications Research Laboratory, Southeast University (No. M j∈A,j=k |hk wj | + 1
2008A06).
The authors are with the National Mobile Communications Research Given A, the sum rate of our system is calculated as
Laboratory, Southeast University, Nanjing 210096, China (e-mail: {caihuaz, 
weiwell hsu, chenming}@seu.edu.cn). R (A) = log2 (1 + SINRk ) . (3)
Digital Object Identifier 10.1109/LCOMM.2009.081466 k∈A
1089-7798/09$25.00 
c 2009 IEEE
ZHANG et al.: HYBRID ZERO-FORCING BEAMFORMING/ORTHOGONAL BEAMFORMING WITH USER SELECTION FOR MIMO BROADCAST CHANNELS 11

The objective of our work is to select M users (A) for Accordingly, we can calculate the sum rate of HBF with
transmission and design their beamformers (wk , k ∈ A) so respect to L and A by
as to maximize the sum rate [R (A)]. M

RHBF (L, A) = log2 (1 + SINRki ) . (9)
i=1
III. H YBRID ZFBF/OBF S CHEME
Therefore, the achievable sum rate of HBF can be maximized
As mentioned earlier, ZFBF and OBF are two commonly by considering every possible choice of L and A
used beamforming strategies, each of which has its own
advantage. ZFBF performs quite well for a large user system, RHBF = max RHBF (L, A) . (10)
1≤L≤M,A⊆{1,··· ,K}:|A|=M
but suffers performance degradation for systems with low to
moderate number of users, especially at low SNR [5], [7]. Finding the optimal solution to problem (10) requires
OBF overcomes the drawback of ZFBF while sacrificing the exhaustive search, which is computationally prohibitive for
performance for a large user system or under high SNR [8]. large K. To reduce the complexity, we propose an efficient
Therefore, this study presents a HBF scheme which provides sub-optimal scheme. In the scheme, we iteratively select a
satisfactory performance in various scenarios. user from the unselected users and determine its beamforming
The essential idea behind our HBF scheme is to combine strategy by maximizing the sum rate together with the already
ZFBF and OBF for designing the beamformers of the sched- selected users. Note that the first user is selected with the
uled users. The proposed scheme first exploits ZFBF only for largest channel magnitude and its beamforming strategy may
users with near-orthogonal channels. The number of these be set as either ZFBF or OBF. To further simplify the scheme,
users, denoted as L, may be less than M , the maximum we set the beamforming strategy of the first user as ZFBF
number of users allowed for simultaneous transmission. Sub- and introduce the following constraint: once the beamforming
sequently, it selects (M − L) additional users to fully exploit strategy of a user is selected as OBF, then for the subsequently
the spatial dimension, and assigns orthogonal beamformers to selected users, their beamforming strategies will always be
them. It is important to note that these orthogonal beamform- OBF. The details of our proposed scheme are described in
ers are chosen in the null space of the ZFBF user channels Algorithm 1. It returns S as the set of scheduled users. In the
in order to introduce no inter-user interference to users using algorithm, RZFBF (A) denotes the sum rate of ZFBF for a given
ZFBF. set A, and BF indicator indicates the beamforming strategy
Mathematically, given the set of scheduled users A, we of the newly selected user such that BF indicator = 0
assume the first L (1 ≤ L ≤ M ) users employ ZFBF represents ZFBF and BF indicator = 1 means OBF.
 †
and others use OBF. Let H (L, A) = h†k1 , · · · , h†kL be Algorithm 1 HBF with User Selection
the concatenated channel vectors of the first L users. The Initialization: B = {1, · · · , K}, S = ∅, and L = M
beamforming vectors for them are r1 = arg maxl∈B hl h†l , BF indicator = 0, S = S∪{r1 }
for n = 2 to M do
wki = qi /qi , i = 1, · · · , L (4)
if BF indicator = 0 then
where qi is the ith column of rZF = arg maxl∈B\S RZFBF (S ∪ {l})
rOBF = arg maxl∈B\S RHBF (n − 1, S ∪ {l}) 1
 −1

F (L, A) = H (L, A) H (L, A) H (L, A)

. (5) if RZFBF (S ∪ {rZF }) > RHBF (n − 1, S ∪ {rOBF })
then
Subsequently for users ki (L + 1 ≤ i ≤ M ), their beamform- rn = rZF
ers, which are chosen in the null space of the subspace spanned else
rn = rOBF , BF indicator = 1, L = n − 1
by h†k1 , · · · , h†kL , wkL+1 , · · · , wki−1 , can be given by
end if
else
wki = gi /gi , i = L + 1, · · · , M (6)
rn = arg maxl∈B\S RHBF (L, S ∪ {l})
where end if
⎛ ⎞ S = S ∪ {rn }
L
 h†kj hkj i−1
 end for
gi = ⎝IM − − wkj wk† j ⎠ h†ki . (7)
j=1
hkj 2
j=L+1
Subsequently, we briefly discuss the computational com-
When i = L + 1, this implies gi = plexity of Algorithm 1. For each n ≤ M , Algorithm 1 evalu-
 
L h†kj hkj † ates (K −n+1) rates RZFBF (S ∪ {l}) and (K −n+1) SINRs
IM − j=1 hk 2 hki . Substituting (4) and (6) into (2), in the worst case. From [6], we know that the evaluation of
j

we obtain the SINR of the scheduled users as each RZFBF (S ∪ {l}) has complexity O(M 2 ). Moreover, from
⎧ (8), we can easily see that the evaluation of each SINR also
⎨ M (H(L,A)H(L,A)† )−1  , 1 ≤ i ≤ L
P

1 Note that in HBF, a selected OBF user does not cause any interference
SINRki = gi 2
i,i


⎩ i−1 |h w |2 +M/P , L + 1 ≤ i ≤ M. to the previously selected users. Thus, instead of selecting the user with the
j=1
ki kj largest sum rate, we can choose the user with the largest SINR to reduce the
(8) complexity.
12 IEEE COMMUNICATIONS LETTERS, VOL. 13, NO. 1, JANUARY 2009

ZFBF-SUS [5], ZFBF-GUS [6], OBF-GUS [8], and ORBF [9].


For all the above schemes, we assume equal power allocation
over the selected users. For reference, we also include the
performance of RMBF [3] with GUS (RMBF-GUS).
Fig. 1 shows the sum rate versus the number of users K
for M = 8, P = 5 dB. We can see that HBF-US outperforms
all the other schemes except for RMBF-GUS in all ranges
of K. Moreover, for growing K, the sum rate of HBF-US
approaches that of ZFBF-GUS. This is due to the fact that the
transmitter can always select M users with semi-orthogonal
channels when the number of users is sufficiently large. In
this case, HBF-US reduces to ZFBF-GUS. Although there is a
noticeable gap between HBF-US and RMBF-GUS, we should
Fig. 1. Sum rate versus the number of users K for M = 8, P = 5 dB. mention that HBF-US has similar complexity to ZFBF-GUS,
which is much simpler than RMBF-GUS.
In Fig. 2, the sum rate versus average SNR P for M =
8, K = 50 is given. It can be observed that HBF-US achieves
higher sum rate than all the other schemes except RMBF-GUS
in all ranges of P . It can also be seen that HBF-US achieves
almost the same sum rate as ZFBF-GUS at high SNR and
similar sum rate to OBF-GUS at low SNR. The reason is that
eliminating the inter-user interference is of crucial importance
at high SNR and is not necessary at low SNR. As a result, in
HBF-US, all the selected users employ ZFBF as P → ∞ and
use OBF as P → 0.

V. C ONCLUSIONS
Fig. 2. Sum rate versus average SNR P for M = 8, K = 50.
In this letter, we propose a hybrid transmission scheme for
the MIMO BC aiming to achieve satisfactory performance
has complexity O(M 2 ). Thus, the complexity of Algorithm 1 in various scenarios. Both theoretic analysis and simulation
is O(M 3 K), while the complexity of the GUS algorithm for results show that the proposed HBF scheme outperforms the
ZFBF is also O(M 3 K) [6]. conventional ZFBF scheme and OBF scheme in all ranges of
Finally, we compare the maximum achievable sum rate of SNR and the number of users.
HBF with that of ZFBF and OBF in a theoretical way. For
the sake of comparison, the maximum achievable sum rate of
ZFBF and OBF are formulated by R EFERENCES

RZFBF = max RZFBF (A) (11) [1] D. Gesbert, M. Kountouris, R. W. Heath, C. B. Chae, and T. Salzer,
A⊆{1,··· ,K}:|A|=M “Shifting the MIMO paradigm,” IEEE Signal Processing Mag., vol. 24,
no. 5, pp. 36–46, Sept. 2007.
ROBF = max ROBF (A) (12) [2] H. Weingarten, Y. Steinberg, and S. Shamai, “The capacity region of
A⊆{1,··· ,K}:|A|=M
the Gaussian MIMO broadcast channel,” in Proc. IEEE ISIT, Chicago,
where ROBF (A) represents the sum rate of OBF for a given IL, June/July 2004, p. 174.
set A. The following theorem shows that HBF outperforms [3] M. Stojnic, H. Vikalo, and B. Hassibi, “Rate maximization in multi-
antenna broadcast channels with linear preprocessing,” IEEE Trans.
ZFBF and OBF, all with optimal user scheduling. Wireless Commun., vol. 5, no. 9, pp. 2338–2342, Sept. 2006.
Theorem 1: The maximum achievable sum rate of HBF [4] G. Caire and S. Shamai, “On the achievable throughput of a multi-
is larger than that of ZFBF and OBF, i.e., RHBF ≥ antenna Gaussian broadcast channel,” IEEE Trans. Inform. Theory, vol.
49, no. 7, pp. 1691–1706, July 2003.
max {RZFBF , ROBF }. [5] T. Yoo and A. Goldsmith, “On the optimality of multianntenna broadcast
Proof: Note that for a given user set A with |A| = M , scheduling using zero-forcing beamforming,” IEEE J. Select. Areas
HBF reduces to OBF when L = 1 and to ZFBF when Commun., vol. 24, no. 3, pp. 528–541, Mar. 2006.
[6] G. Dimic and N. Sidiropoulos, “On downlink beamforming with greedy
L = M . Hence, we have RHBF (1, A) = ROBF (A) and user selection: performance analysis and a simple new algorithm,” IEEE
RHBF (M, A) = RZFBF (A). Obviously, the maximum achiev- Trans. Signal Processing, vol. 53, no. 10, pp. 3857–3868, Oct. 2005.
able sum rate of HBF for a given set A can be lower bounded [7] R. de Francisco, M. Kountouris, D. Slock, and D. Gesbert, “Orthogonal
as max1≤L≤M RHBF (L, A) ≥ max {RZFBF (A) , ROBF (A)}. linear beamforming in MIMO broadcast channels,” in Proc. IEEE
WCNC, Hong Kong, China, Mar. 2007, pp. 1210–1215.
By considering every possible choice of A, we conclude that [8] J. Duplicy, D. P. Palomar, and L. Vandendorpe, “Adaptive orthogonal
RHBF ≥ max {RZFBF , ROBF }. beamforming for the MIMO broadcast channel,” in Proc. IEEE CAM-
SAP, Virgin Islands, USA, Dec. 2007, pp. 77–80.
IV. S IMULATION R ESULTS [9] M. Sharif and B. Hassibi, “On the capacity of MIMO broadcast channels
In this section, we compare the performance of our HBF with partial side informaion,” IEEE Trans. Inform. Theory, vol. 51, no.
2, pp. 506–522, Feb. 2005.
scheme in Algorithm 1 (HBF-US) with the following schemes:

You might also like