Академический Документы
Профессиональный Документы
Культура Документы
Abstract—Massive multiple-input multiple-output (MIMO) is in frequency division duplexing (FDD) systems. Since FDD
widely recognized as a promising technology for future 5G systems are more widely deployed and show some advantages
arXiv:1512.03230v1 [cs.IT] 10 Dec 2015
wireless communication systems. To achieve the theoretical per- over TDD systems in delay-sensitive and traffic-symmetric
formance gains in massive MIMO systems, accurate channel state
information at the transmitter (CSIT) is crucial. Due to the applications [2], it is of great importance to study downlink
overwhelming pilot signaling and channel feedback overhead, channel training and uplink channel feedback in FDD massive
however, conventional downlink channel estimation and uplink MIMO systems.
channel feedback schemes might not be suitable for frequency- For conventional channel training and feedback scheme, the
division duplexing (FDD) massive MIMO systems. In addition, BS firstly transmits orthogonal pilots, the number of which
these two topics are usually separately considered in the liter-
ature. In this paper, we propose a joint channel training and scales linearly with the number of BS antennas. Then, users
feedback scheme for FDD massive MIMO systems. Specifically, estimate the CSI using the conventional technique such as
we firstly exploit the temporal correlation of time-varying chan- least squares (LS) or minimum mean square error (MMSE)
nels to propose a differential channel training and feedback algorithm [3]. The estimated CSI is then fed back to the BS
scheme, which simultaneously reduces the overhead for downlink via dedicated uplink resources. In massive MIMO systems,
training and uplink feedback. We next propose a structured
compressive sampling matching pursuit (S-CoSaMP) algorithm the number of BS antennas might be an order of magnitude
to acquire a reliable CSIT by exploiting the structured sparsity larger than that of state-of-the-art LTE-Advanced system, so
of wireless MIMO channels. Simulation results demonstrate that the training overhead is a serious concern [4]. Furthermore,
the proposed scheme can achieve substantial reduction in the the massively expanded MIMO channel matrix also renders
training and feedback overhead. precise CSI feedback an extremely challenging problem [2].
Index Terms—Massive MIMO; channel estimation; channel To alleviate the overhead, several downlink pilot design and
feedback; temporal correlation; structured sparsity. uplink CSI feedback schemes have been proposed [5], [6],
[7], [8], [9], [10], [11], [12]. Among these, compressive
I. I NTRODUCTION sensing (CS) has merged as a promising technique in recent
Massive multiple-input multiple-output (MIMO) is widely years. The fundamental principle of CS-based schemes relies
recognized as a key technology for future 5G wireless commu- on the fact that broadband wireless channels have sparse
nications due to its substantial gains in spectrum and energy channel impulse response (CIR) due to the limited number of
efficiency. In massive MIMO systems, the base station (BS) significant paths [4], [9], [10]. To date, number of CS-based
is equipped with a large number of antennas to provide channel estimation schemes have been proposed where the
high spatial degrees of freedom (DoF). To fully capitalize sparse channel impulse response (CIR) can be recovered from
the DoF gain provided by massive MIMO systems, channel a reduced number of received pilots [9], [10]. Nevertheless, the
state information at the transmitter (CSIT) is essential. Recall reduction of training overhead is insignificant when CIR is not
that CSIT is crucial in channel adaptive techniques such as sparse enough. On the other hand, several CS-based channel
beamforming, power allocation, and interference alignment feedback schemes have also been proposed. The underlying
[1]. In time-division duplexing (TDD) systems, channel reci- assumption on this work is that users firstly achieve perfect
procity can be used to obtain the downlink CSI via uplink CSI estimation, and then compressing (through projection)
channel estimation. However, this feature cannot be exploited the sparse (or compressible in some transform domains) CSI
into low-dimensional measurements for feedback [11], [12].
Copyright (c) 2015 IEEE. Personal use of this material is permitted. However, the separate treatments of the two coupled com-
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. munication procedures (downlink channel training and uplink
W. Shen, L. Dai, and Z. Wang are with the Department of Elec- channel feedback) result in some performance loss, and induce
tronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: unnecessary computational overhead especially for power-
swq13@mails.tsinghua.edu.cn; {daill, zcwang}@tsinghua.edu.cn).
Y. Shi is with Huawei Technologies, Beijing 100095, China (e-mail: limited users. In [13], an approach based on compressive
wn.shiyi@gmail.com). sensing (CS) has been proposed to reduce both the downlink
B. Shim is with Institute of New Media and Communications, School of training and uplink CSI feedback overhead. Although this
Electrical and Computer Engineering, Seoul National University, Seoul 151-
742, Korea (e-mail: bshim@snu.ac.kr). approach is promising when the channel matirces of different
This work was supported by the National Key Basic Research Program users are sparse and partially share common support, it is not
of China (Grant No. 2013CB329203), the National Natural Science Foun- effective when these assumptions are violated.
dation of China (Grant Nos. 61571270 and 61201185), the Beijing Natural
Science Foundation (Grant No. 4142027), and the Foundation of Shenzhen In this paper, we propose a structured-CS based differential
government. joint channel training and feedback scheme for massive MIMO
2
systems, where downlink training and uplink feedback are [4], [9], [10], it is assumed that the number of non-zero
considered in a joint manner. We design a system such that elements K in CIR hi is much smaller than the maximal
users will directly feed back the received pilots to the BS channel length L. This vector is often referred to as K-
without the channel estimation. BS estimates the CSI using sparse vector. The parameter nΩ = [n1 , · · · , nP ]T represents
CS-based algorithm. Such “closed-loop” operation streamlines the independent and identically distributed (i.i.d.) additive
the CSIT acquisition process by removing channel estima- white complex Gaussian noise (AWGN) with zero mean and
tion and the compression procedures at the user side. In variance σn2 imposed on pilots. For notation simplicity, the
addition, our proposed scheme also features a differential equation (1) can also be written as
operation, which exploits the temporal correlation between
yΩ = Θh + nΩ , (2)
two consecutive CIRs to reduce the number of required pilots.
Finally, due to the close antenna spacing at the BS and where Θ = [C1 (FL )Ω , C2 (FL )Ω , · · · , CM (FL )Ω ], and h =
resulting similar path delays, CIRs associated with different [hT1 , hT2 , · · · , hTM ]T denotes the aggregate CIR from M BS
BS antennas usually have a common support (i.e. the locations antennas. In the sequel, we refer to the aggregate CIR as CIR
of non-zero elements) [14]. By exploiting this feature, we unless it causes ambiguity.
propose a structured compressive sampling matching pursuit
(S-CoSaMP) algorithm to further reduce the training as well
B. Temporal Correlation of Time-Varying MIMO Channels
as feedback overhead.
The remainder of this paper is organized as follows. Sec- We consider block-fading MIMO channels, where the CIR
tion II briefly introduces the massive MIMO system model, h changes from slot to slot but remains unchanged during
−1
where the temporal and spatial correlations of channels are one time slot. The CIR series {h(t) }Tt=0 in T consecutive
emphasized. Section III addresses the proposed differential time slots usually exhibits temporal correlation, and thus is
joint channel training and feedback scheme, together with compressible in the time domain [11], [15]. The dynamic
the proposed S-CoSaMP algorithm. Section IV presents the channel can be modeled by the variation of CIR’s support
simulation results. Finally, conclusions are drawn in Section and the evolution of the non-zero elements’ amplitudes as [9],
V. [10]
Notations: Lower-case and upper-case boldface letters de- h(t) = s(t) ◦ a(t) , (3)
note vectors and matrices, respectively; (·)T , (·)H and (·)−1 where s(t) (l) ∈ {0, 1}, a(t) (l) ∈ C, and ◦ denotes the
denote the transpose, conjugate transpose and inverse of a Hadamard product. The variation of {s(t) (l)}Tt=0 −1
over time
matrix, respectively; Θ† = (ΘH Θ)−1 ΘH is the Moore- can be modeled as a first-order Markov process [16], which
Penrose pseudoinverse of Θ; ΘS denotes the sub-matrix can be fully characterized by two transition probabilities
consisted of columns of Θ according to the index set S. k·kp is p10 = Pr{s(t+1) (l) = 1|s(t) (l) = 0} and p01 = Pr{s(t+1) (l) =
the lp -norm; S c denotes the complementary set of S; T (x, K) 0|s(t) (l) = 1}, and a distribution µ(0) = Pr{s(0) (l) = 1}
denotes a prune operator on x that sets all but K elements with in the initial time slot t = 0. For the steady-state Markov
the largest amplitudes to zero; Γx denotes the support of x, process, where Pr{s(t) (l) = 1} = µ, for all t and l, only
i.e., Γx = {i, x(i) 6= 0}. two parameters p01 and µ are sufficient to characterize the
process since p10 = µp01 /(1 − µ). The evolution of amplitude
II. S YSTEM M ODEL −1
{a(t) (l)}Tt=0 over time can be modeled by the first-order
A. System Model autoregressive model as [17]
We consider a massive MIMO system operating in FDD p
a(t) (l) = ρa(t−1) (l) + 1 − ρ2 w(t) (l), (4)
mode with the ubiquitous orthogonal frequency division mul-
tiplexing (OFDM) modulation. There are M antennas at the where the correlation coefficient ρ = J0 (2πfd τ ) is given
BS and U scheduled single-antenna users. The length of one by the zero-order Bessel function of the first kind with fd
OFDM symbol is N . The BS transmits pilots ci ∈ C P ×1 at being the maximal Doppler frequency and τ being the time
the i-th transmit antenna, where i = 1, 2 · · · , M , and P is the slot duration, the parameter w(t) (l) ∼ CN (0, σω2 ) is the i.i.d.
number of pilots. At a certain user, the received pilots yΩ in complex Gaussian variables.
the frequency domain can be expressed as
M
X C. Spatial Correlation of Massive MIMO Channels
yΩ = Ci (FL )Ω hi + nΩ , (1) Due to the close antenna spacing at the BS, CIRs {hi }M
(t)
i=1
i=1
between BS antennas and the single receive antenna of a user
where Ci = diag{ci }, FL ∈ C N ×L is a sub-matrix consisting have similar path delays. Thus, they share a common support
of the first L columns of the discrete fourier transform (DFT) [14], i.e.,
matrix of size N × N , (FL )Ω is the sub-matrix consisting of Γh(t) = Γh(t) = · · · = Γh(t) , (5)
1 2 M
the rows of FL with indices from the index set Ω of subcarriers
(t)
assigned to pilots, which can be randomly selected from the This property of h is referred to as structured sparsity. Thus,
(t)
subcarrier set {1, 2, · · · , N }, hi = [hi (1), hi (2), · · · , hi (L)]T {hi }M i=1 can be generated with the same support vectors
(t) M (t)
is the CIR between the i-th BS antenna and the user. Due to {si }i=1 , where {si }M i=1 can be generated at random without
the physical propagation characteristics of multi-path channels loss of generality.
3
3
CS-based CSI CSI CS-based CSI
estimation projection recovery 2
Training 1
Received Estimated
pilots Downlink pilots CSI Measurements Uplink Measurements CSIT 0
channel channel 10 20 30 40 50 60
(a)
0
(a) 10 20 30 40 50 60
CS-based CSI (b)
recovery
3
Training Received Received 2
pilots Downlink pilots Uplink pilots CSIT
channel channel 1
0
10 20 30 40 50 60
(c)
Downlink Uplink channel
channel training (b) Feedback
Fig. 2. A snapshot of the previous CIR, the current CIR, and the differential
Fig. 1. Comparison between conventional and proposed joint scheme: (a) CIR: (a) the previous CIR; (b) the current CIR; (c) the differential CIR.
Conventional CS-based channel training and feedback scheme. (b) Proposed
CS-based joint channel training and feedback scheme.
AWGN channel2 with the same SNR of downlink as in [18],
the received pilots at the BS can still be modeled by (2) except
III. D IFFERENTIAL J OINT C HANNEL T RAINING AND
that in this case, the noise parameter σn2 denotes the overall
F EEDBACK BASED ON S-C O S A MP
noise power both in the downlink and uplink [18].
In this section, we present the proposed differential joint To exploit the temporal correlation of time-varying MIMO
channel training and feedback scheme. By exploiting the channels, we consider the differential CIR between two CIRs
structured sparsity of massive MIMO channels, we propose in adjacent time slots, which can be expressed as
the S-CoSaMP algorithm to recover CIR h(t) from received
(t)
pilots yΩ at the BS. ∆h(t) = h(t) − h(t−1) (6)
(t) (t) (t−1) (t) (t−1) (t−1)
=s ◦(a −a )+(s −s )◦a
A. Differential Joint Channel Training and Feedback p
= s(t) ◦[ 1−ρ2w−(1−ρ)a(t−1)]+(s(t) −s(t−1) )◦a(t−1) .
Conventional channel training and feedback schemes usu-
ally consists of two steps: downlink CSI estimation at user side When the movement velocity of users is not very high, e.g.,
and CSI feedback in the uplink. Under the framework of CS, v = 12km/h, and the carrier frequency is 900MHz, the
channel estimation algorithm [9], [10] and channel feedback resulting Doppler frequency fd is 10Hz. For the typical time
scheme [11], [12] exploring the sparsity of the CSI have been slot duration τ = 0.5ms, the correlation coefficient ρ is 0.9911
proposed to achieve overhead reduction. In these approaches, [11]. Thus, the first term on the last line of (6) is close to zero.
these two coupled communication procedures are optimized On the other hand, since the delay indices of non-zero taps
separately as depicted in Fig. 1 (a). are varying slowly [4], i.e., the CIR’s support s(t) changes
To address the drawbacks of the conventional schemes, we slowly, the number of non-zero elements of the second term
propose a novel CS-based joint scheme that considers channel is also small. Fig. 2 presents the snapshot of the previous CIR
training and feedback in a joint manner. The corresponding h(t−1) , the current CIR h(t) , and the differential CIR ∆h(t)
schematics is illustrated in Fig.1 (b), where users directly feed of the channels described by (3), (4) and (6). We can observe
the received training pilots back to the BS without channel es- that the differential CIR enjoys a much stronger sparsity3 than
timation, and then the BS recovers the CIR via CS algorithms. the original CIR due to the temporal correlation of channels.
Note that the proposed CS-based joint channel training and To utilize this observation, we modify (2) and express the
feedback scheme avoids the complex computations at power- received pilots at the BS as
limited users for channel estimation1 and CSI projection,
which not only relieves the computational burden for the (t)
yΩ = Θh(t) + nΩ .
(t)
(7)
user devices, but also brings channel feedback performance
improvement, as will be verified by simulations in Section We try to exploit the temporal correlation of channels by
IV. As the dedicated uplink channel can be modeled as an computing the difference between received pilots at the BS
1 The precondition is that CSI is not required at users, which is normal
2 In this paper, we discuss the analog feedback [7], [8], [18], [21], thus, the
in massive MIMO systems with precoding [1]. Otherwise, CSI can also
be obtained by using the proposed differential operation and S-CoSaMP quantization noise of digital feedback is not considered.
algorithm at users, which will achieve better estimation performance than 3 The sparsity level K ′ of the differential CIR is mainly dependent on the
conventional channel estimation scheme as will be shown in Section IV. transition probability p01 and p10 .
4
10
1
10 Perfect CSIT
9 CoSaMP−based conventional feedback (η=40%)
CoSaMP−based conventional feedback (η=70%)
8
6
NMSE
−1 5
10
Fig. 3. NMSE performance comparison between the proposed joint scheme Fig. 4. Downlink data rate performance comparison with two different
and the traditional scheme. training and feedback overhead: η = 40% and η = 70%.
sparsity level is K = µL = 6, the transition probability p01 scheme in both cases. Note that, S-CoSaMP based differential
is set as 0.16, the maximal Doppler frequency is fd = 10Hz, joint scheme with η = 40% and CoSaMP-based conventional
2 scheme with η = 70% can achieve similar data rate. This
the time slot duration is τ = 0.5ms, σw = 1 and the initial
(0) means that the proposed scheme achieves around 30% re-
amplitudes a ∼ CN (0, 1); 3) The proposed scheme will
be initiated every R = 3 time slots; 4) The coefficient ξ in duction in channel training and feedback overhead over the
Algorithm 1 is set as 10−3 . From equation (2), the signal-to- conventional scheme in achieving the same data rate.
2
noise ratio (SNR) can be defined as SN R = ||Θh|| 2
σn , where
the SNR is an overval SNR when σn2 denotes the sum of noise V. C ONCLUSION
power in the downlink and uplink. For simplicity, we call it as In this paper, we have investigated the challenging problem
SNR unless it causes confusion. The same training overhead of channel training and feedback for FDD massive MIMO
η = P/N is considered to ensure fair comparision. For the systems. By exploiting the temporal correlation of MIMO
proposed differential scheme, the average pilot overhead is channels, we have proposed the differential joint channel
0
η = P +PRN ∗(R−1)
. training and feedback scheme, where users directly feed back
Firstly, we compare the proposed joint channel training the received pilots and then the explicit CSI can be obtained
and feedback scheme with its conventional counterpart in by CS algorithms at the BS. By exploiting the structured
Fig. 3. Note that the NMSE of joint orthogonal matching sparsity of MIMO channels, we have proposed the S-CoSaMP
pursuit (J-OMP) algorithm [13] and Kalman filter [5], [6] are algorithm to further reduce the overhead. In the future, we will
also presented for comparison. The dotted lines denote the consider the spatial correlation of CSI from different users that
conventional scheme. Note that the conventional scheme treats is available at the BS to further reduce the overhead.
channel training and feedback separately. We observe that the
proposed joint scheme outperforms the conventional scheme, R EFERENCES
which means the joint consideration of channel training and [1] F. Rusek, D. Persson, B. Lau, E. Larsson, T. Marzetta, O. Edfors, and F.
feedback improves the CIR recovery performance at the BS. In Tufvesson, “Scaling up MIMO: Opportunities and challenges with very
large arrays,” IEEE Signal Process. Mag., vol. 30, no. 1, pp. 40-60, Jan.
addition, S-CoSaMP based differential joint scheme achieves 2013.
the best NMSE performance due to the exploiting of the [2] B. Lee, J. Choi, J. Seol, D. J. Love, and B. Shim, “Antenna grouping
temporal correlation and structured sparsity. The performance based feedback compression for FDD-based massive MIMO systems,”
IEEE Trans. Commun., vol. 63, no. 9, pp. 3261-3274, Sep. 2015.
of J-OMP and CoSaMP are similar because the channel [3] M. Ozdemir and H. Arslan, “Channel estimation for wireless OFDM
assumption of common support is not satisfied and J-OMP systems,” IEEE Commun. Surv. Tutor., vol. 9, no. 2, pp. 18-48, 2nd
degrades as OMP, which has similar recovery performance Quarter 2007.
[4] L. Dai, J. Wang, Z. Wang, P. Tsiaflakis, and M. Moonen, “Spectrum- and
with CoSaMP. The performance of Kalman filter is not good energy-efficient OFDM based on simultaneous multi-channel reconstruc-
when the channel autocorrelation information is unknown [5]. tion,” IEEE Trans. Signal Process., vol. 61, no. 23, pp. 6047-6059, Dec.
Fig. 4 shows the data rate comparison when two differ- 2013.
[5] S. Noh, M. Zoltowski, Y. Sung, and D. Love, “Pilot beam pattern design
ent channel training and feedback overheads are considered: for channel estimation in massive MIMO systems,” IEEE J. Sel. Topic
η = 40% and η = 70%. The ideal case where the CSIT is Signal Proess., vol. 8, no. 5, pp. 781-801, Oct. 2014.
perfectly known at the BS is also presented as a benchmark [6] J. Choi, D. Love, and P. Bidigare, “Downlink training techniques for
FDD massive MIMO systems: Open-loop and closed-loop training with
for comparison. As shown in Fig. 4, S-CoSaMP based differ- memory,” IEEE J. Sel. Topic Signal Proess., vol. 8, no. 5, pp. 802-814,
ential joint scheme outperforms CoSaMP-based conventional Oct. 2014.
6