1Massive MIMO for 5G: From Theory to Practice
Massive MIMO for 5G: From Theory to Practice
Linglong Dai (戴凌龙)
Department of Electronic EngineeringTsinghua University
Dec. 2015
2Massive MIMO for 5G: From Theory to Practice
Content
5G and Massive MIMO1
Massive MIMO: Theoretical Performance2
Massive MIMO: Practical Solutions3
4 Future Research
5 Summary
3Massive MIMO for 5G: From Theory to Practice
How to realize 5G? Key requirement of 5G: 1000-fold capacity How to realize this goal from Shannon capacity? Three technical directions for 5G
C = D * W * M * log (1+SINR)
No. of APs Bandwidth
No. of antennas Interference mitigation
4Massive MIMO for 5G: From Theory to Practice
Use hundreds of BS antennas to simultaneously serve multiple users
Conventional MIMOM:2~8, K:1~4 (LTE-A)Conventional MIMOM:2~8, K:1~4 (LTE-A)
Massive MIMOM: ~100~1000, K: 16~64
Massive MIMOM: ~100~1000, K: 16~64
T. L. Marzetta, “Non-cooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas,” IEEE Transactions on Wireless Communications, vol. 9, no. 11, pp. 3590-3599, Nov. 2010. (2013 IEEE Marconi prize)
What is massive MIMO?
5Massive MIMO for 5G: From Theory to Practice
Content
5G and Massive MIMO1
Massive MIMO: Theoretical Performance2
Massive MIMO: Practical Solutions3
4 Future Research
5 Summary
6Massive MIMO for 5G: From Theory to Practice
Spatial multiplexing‒ Rate: min , log 1 SNR
Reliability‒ ∼ SNR
Array gain (beamforming)‒ Several antennas can be used to transmit signals
Why Massive MIMO ?
No. of antennas Error Probability ( ) Capacity ( ), bps/Hz
1, 1 (SISO) ∼ SNR log 1 SNR
1, 1 (SIMO) ∼ SNR log 1 SNR
1, 1 (MIMO) ∼ SNR min , log 1 SNR
:Diversity Gain min , :Multplexing Gain
Massive , : significantly increased spectral efficiency
7Massive MIMO for 5G: From Theory to Practice
Theoretical Capacity Analysis MIMO link, channel ∈ with
log 1SNR
, SNR
If ∼ 1, then we have ∑ , so‒ Rank-1 channel (LoS): , ⋯ 0
‒ Full rank channel: ⋯ favorable propagation
If is i.i.d. and ≫ , then we have favorable propagation
8Massive MIMO for 5G: From Theory to Practice
Theoretical Capacity Analysis In massive MIMO, the channel matrix is decomposed
into two parts‒ Small-scale fading: of size , elements are i.i.d.‒ Large-scale fading: a diagonal matrix /
⋯ 0⋮ ⋱ ⋮0 ⋯
⋯ 0⋮ ⋱ ⋮0 ⋯
, ≫
Ideal massive MIMO channel has favorable propagation
Asymptotical orthogonality
9Massive MIMO for 5G: From Theory to Practice
Ideal Channels Massive MIMO has much larger ordered singular values
than conventional MIMO
F. Rusek, D. Persson, B. Lau, E. Larsson, T. Marzetta, O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities and challenges with very large arrays,” IEEE Signal Processing Magazine, vol. 30, no. 1, pp. 40-60, Jan 2013.
10Massive MIMO for 5G: From Theory to Practice
Ideal Detection Optimal (coherent) uplink detector has complexity ~exp
min ∥ ∥ With favorable propagation in massive MIMO ( ≫ )
1
we can have
min ∥1
∥ ⇔ min
In massive MIMO, we can use simple (linear) detectors like MF, ZF with good enough performance and low complexity ∼
Similarly, simple (linear) precoders can be also used
11Massive MIMO for 5G: From Theory to Practice
Feb. 2012, Rice university & Bell labs, Argos, 64 antennas, 15 users, 85 bit/s/Hz, 1/64 power consumption
Sep. 2013, Rice university & Bell labs, ArgosV2, 96 antennas, 32 users
July 2013, Linköping & Lund University, 128 antennas, 36 users
Recent Advances of Massive MIMO
12Massive MIMO for 5G: From Theory to Practice
World’s First Massive MIMO PrototypeSamsung, 2014
13Massive MIMO for 5G: From Theory to Practice
Opportunities and challenges Advantages
‒ Improve the spectrum efficiency by orders of magnitude ‒ Improve the energy efficiency by orders of magnitude
Vision‒ Considered as a promising key technology for 5G
Challenges‒ Theoretical analysis with practical constraints‒ Reduce the power consumption of RF chains‒ Pilot contamination in the uplink‒ Efficient pilot design and channel estimation algorithm ‒ Efficient channel feedback mechanism‒ Low-complexity near-optimal signal detection algorithm
14Massive MIMO for 5G: From Theory to Practice
Content
5G in The World1
Massive MIMO: Theoretical Performance2
Massive MIMO: Practical Solutions3
4 Future Research
5 Summary
15Massive MIMO for 5G: From Theory to Practice
Work 1: Performance Analysis of Massive MIMO with Practical Constraints Work 2: Pilot Decontamination Based on Graph Coloring Work 3: Efficient Pilot Design and Channel Estimation Based on Compressive Sensing Work 4: Low-Complexity Multi-User Detection for Uplink Massive SM-MIMO Work 5: Energy-Efficient SIC-Based Hybrid Precoding for Massive MIMO Work 6: Beamspace Massive MIMO
Practical Solutions for Massive MIMO
16Massive MIMO for 5G: From Theory to Practice
Practical Solutions: Work 1
Performance Analysis of Massive MIMO with Practical Constraints
Jiayi Zhang, Linglong Dai, Xinlin Zhang, Emil Björnson, and Zhaocheng Wang, “Ergodic Capacity of Massive MIMO Systems with Transceiver Hardware Impairments over Rician Fading Channels,” to appear in IEEE Transactions on Vehicular Technology.
17Massive MIMO for 5G: From Theory to Practice
Motivation The performance of massive MIMO systems is usually
limited by practical constraints‒ Hardware impairments
• Phase noise• I/Q imbalance• Amplifier non-linearities• Quantization errors
‒ Space constrained‒ Low-resolution ADC‒ Channel aging‒ Imperfect CSI‒ Inter-carrier interference‒ Co-channel interference‒ …
18Massive MIMO for 5G: From Theory to Practice
Motivation Huge spatial degrees-of-freedom of massive MIMO
systems are achieved by coherent processing over these massive arrays, which provide ‒ strong signal gains‒ resilience to imperfect channel knowledge ‒ and low interference
However, the hardware cost and circuit power consumption scale linearly/exponentially with the number of BS antennas, and practical constraints cannot be removed completely
Whether massive MIMO systems can provide robustnessto practical constraints?
19Massive MIMO for 5G: From Theory to Practice
The system model can be written as [1]
The additive distortion noise terms can be analytically approximated by the central limit theorem as
In LTE, the error vector magnitude (EVM) are in the range
Hardware Impairments
t ry = H x + η + η + n
21
2
0, diag , ,
0, tr
t
r
t N
r r N
CN q q
CN
tη
η Q I
0.08,0.175t
[1] T. Schenk, RF Imperfections in High-Rate Wireless Systems: Impact and Digital Compensation. Springer, 2008.
20Massive MIMO for 5G: From Theory to Practice
Normalized noise variance
The ergodic achievable rate R can be expressed as
Hardware Impairments
22
22
1 ,
1 ,
t
r
Htr N t r
t
Htr N t r
t
N NN
N NN
H H IΦ
HH I
12
12
log det ,
log det ,
t
r
HN t r
t
HN t r
t
E N NN
R
E N NN
I H HΦ
I HH Φ
21Massive MIMO for 5G: From Theory to Practice
Proposition 1: The exact achievable rate of MIMO systems with residual hardware impairments over Ricianfading channels can be expressed as
,1 1 0
1 / 1 /b1 1
1
2 21
2 2
ln 2 1 1
1 1
min( , ), max( , )
1, ,
1 1 !
i
kq qn
n mn m k
p q m kK a K
p q m k t p q m k tt
t r t r
qt t i
t t t t
p q m kqGR Dk p q k
K Ke E e Ea b
q N N p N N
ea b G
N N p q
1
qj i
i j q
Hardware Impairments
22Massive MIMO for 5G: From Theory to Practice
For → ∞, the achievable rate reduces to
For → ∞, the achievable rate reduces to
For and → ∞, the achievable rate reduces to
Hardware Impairments
2 2 2log 11tN r
t r
R N
2 2
1log 1rN t
t
R N
2 2
2 22 2
1log det log det
1 1r r
t H HtN N
t r t r
R EN N
HH I HH I
2 2log det log det
1/ , 1/ ,
r r
r r
H HN N
N N
E a b
J a J b
HH I HH I
I I
23Massive MIMO for 5G: From Theory to Practice
A finite rate ceiling An increase in SNR tends
to increase achievable rate of both systems
The relative difference between the curves gets steadily larger
Higher K values yield lowerachievable rate
The gap decreases as K increases
Hardware Impairments
0.15, 2t r t rN N
SNR [dB]-5 0 5 10 15 20 25 30 35 40
Ach
ieva
ble
Rat
e [b
its/s
/Hz]
0
5
10
15
20
25
30
Non-ideal (Analytical)Non-ideal (Simulations)Ideal capacity
K = 0, 10, 100
Ceiling
24Massive MIMO for 5G: From Theory to Practice
The ceiling disappears for large numbers of transmit and receive antennas
Larger values of K will decrease the rank of correlation matrix and the system’s achievable rate
Utilize ideal hardware at massive MIMO systems when operating over strong LoS environment
Hardware Impairments
0.15, , 10t r t rN N
Number of Transmit/Receive Antennas (Nt=Nr)0 10 20 30 40 50 60
0
20
40
60
80
100
120
140
160
180Non-ideal (Analytical)Non-ideal (Simulations)Ideal Achievable Rate
K=0
K=100
K=10
25Massive MIMO for 5G: From Theory to Practice
A critical issue pertaining to practical massive MIMO systems is the dense deployment for a large number of antennas in a limited physical space
The channel vectors for different UEs will be asymptotically non-orthogonal
Therefore, a space-constrained massive MIMO architecture will suffer from increased spatial correlation, whose impact needs to be rigorously quantified and analyzed
Space Constrained
26Massive MIMO for 5G: From Theory to Practice
The received vector y at the BS is given by
Channel matrix / , where is the transmit steering matrix and is given by
With receiver matrix T, the achievable uplink rate is
Space Constrained
up y Gx n
1 2
2 2sin 1 sin
, , ,
1 1, , ,i i
P
Td dj j M
i e eP
A a a a
a
2
2 2 2log 1
Hu k k
k KH
u k l kl k
pR E
p
t g
t g t
27Massive MIMO for 5G: From Theory to Practice
Proposition 1: For space-constrained massive MIMO systems with MRC receivers, the approximated sum achievable rate is given by
where is the ith eigenvalue of A, and denotes the lth element of
Space Constrained
2 2
1MRC2
2
1
log 1
P
u i ki
K P
u l i kl k i
p MR
p M
28Massive MIMO for 5G: From Theory to Practice
denotes the normalized total antenna array space
The sum rate saturates with an increasing number of BS antennas
MRC suffers a substantial rate degradation when spatial correlation is high
the gap decreases as increases, which implies that the effect of becomes less pronounced
Space Constrained
0 , 12, 6d dM P K
50 100 150 200 250 300 350 400 450 5007
7.5
8
8.5
9
9.5
10
Monte-Carlo simulationAnalytical approximation
29Massive MIMO for 5G: From Theory to Practice
Proposition 2: For space-constrained massive MIMO systems with ZF receivers, the sum achievable rate is lower bounded by
Space Constrained
1 22
1log 1 exp
P
nK KP KZF n P K
L u k n P Pk n k j i j ii j i j
R p K n
YY
1
1,
,ln ,
qp
k qp qp p
q nq n
Y
30Massive MIMO for 5G: From Theory to Practice
Proposition 3: For space-constrained massive MIMO systems with ZF receivers, the sum achievable rate is upper bounded by
Space Constrained
2 12 1 1
1 1
12
1
K log1
ln 2
ZFU uK K K K
j i j ii i j i i j
P
nKn P K
Pn j ii j
R pK i K i
K n
Δ Δ
Y
12 p,q
2 p,q
, 1, , 1
, 2, ,
qp
qp
q P K
q P K q P K P
Ξ
Φ
1 1 1 2 2 21
1 p,q
1 p,q
,
,q 1, ,
1 ,q 1, ,
qp
qp
P K
q P K P K P
Δ ΞΦ Δ Ξ Φ
Ξ
Φ
31Massive MIMO for 5G: From Theory to Practice
The lower bounds can explicitly predict the exact sum rate Large antennas can improves the sum rate of the massive
MIMO ZF by suppressing thermal noise, even in the space constrained scenario
Space Constrained
0 4, 12d P
10 20 30 40 50 60 70 80 90 1004
6
8
10
12
14
16
18
ZF Lower BoundZF Upper BoundMonte-Carlo Simulation
32Massive MIMO for 5G: From Theory to Practice
The massive MIMO systems is more prone to practical constraints in strong LoS fading channel
For space-constrained massive MIMO systems, we derive ‒ approximated sum rate expression of MRC receivers‒ new lower and upper bounds on the sum rate of ZF
receivers The performance of ZF receivers is superior to the one of
MRC receivers for space-constrained massive MIMO systems
Conclusions
33Massive MIMO for 5G: From Theory to Practice
Graph Coloring Based Pilot Decontamination for Massive MIMO
Xudong Zhu, Linglong Dai, and Zhaocheng Wang, “Graph Coloring Based Pilot Allocation to Mitigate Pilot Contamination for Multi-Cell Massive MIMO Systems,” IEEE Communications Letters, vol. 19, no. 10, pp. 1842-1845, Oct. 2015.
Practical Solutions: Work 2
34Massive MIMO for 5G: From Theory to Practice
Massive MIMO in Mobile Network A mobile cellular network with cells, each of singe-
antenna users and ( ≫ ) antennas BS– The system works in TDD protocol– Channel of -th user in -th BS to -th BS: , , , , , ,
– Large-scale fading coefficient: , , 0 (channel strength).– Small-scale fading vector: , , ∈ ,
35Massive MIMO for 5G: From Theory to Practice
What is Pilot Contamination (PC)? Channel estimation in uplink transmission
– User ⟨ , ⟩ utilizes pilot sequence ∈ for channel estimation ⋯ ∈ ,
– The channel estimation for user ⟨ , ⟩ will be contaminated by users in other cells with the same pilot
, , , , , , , , .
Uplink SINR limitation– By adopting MF detector, uplink SINR of user ⟨ , ⟩ is limited by PC
SINR ,, , , ,
∑ | , , , , | ,
→ , ,
∑ , ,
– , denotes the power of uncorrelated interference and noise
36Massive MIMO for 5G: From Theory to Practice
Example of PC Pilot reuse in adjacent cells
– Due to limited pilot resource, pilot reuse is unavoidable– PC to users in cell center is usually light– PC to users in cell edge is usually severe
– Total 3 pilots are utilized for 3 cells– User ⟨3,1⟩ suffers from slight PC– User 1,1 and 2,1 are
contaminated to each other
37Massive MIMO for 5G: From Theory to Practice
Existing Technology Frame structure design
– Using time-shifted pilots for asynchronous transmission among adjacent cells is able to mitigate pilot contamination
Exploiting channel properties– AoA (angle-of-arrival) based methods– Subspace partitioning based blind methods– Coordinated multi-cell precoding– ……
38Massive MIMO for 5G: From Theory to Practice
Motivation Potential PC among users in different cells
– Potential PC , ,⟨ , ⟩ is utilized to measure PC severity between two users when they are assigned with the same pilot
, ,⟨ , ⟩, ,
, ,
, ,
, ,.
– , ,⟨ , ⟩ is actually the ratio of the interference channel strength and the effective channel strength
– Larger , ,⟨ , ⟩ indicates more severe PC will be introduced between user ⟨ , ⟩ and ⟨ , ⟩ when they are assigned with the same pilot
– Key idea: Assign different pilots to users with a large
39Massive MIMO for 5G: From Theory to Practice
PC Graph Construction Potential PC threshold
– Based on a potential PC threshold , a binary potential PC matrix , , , can be generated as
, ,⟨ , ⟩
1, , ,1, , , , , ,0, otherwise.
– A PC graph can be constructed based on
, , , , , , , 1
Pilot reuse among connected users should be avoided.
40Massive MIMO for 5G: From Theory to Practice
Conventional Graph Coloring (GC) Minimizing the total number of colors
– Conventional GC algorithms aim to minimize the total number of colors to assign different colors to connected vertices
– To find the minimal number of colors for a certain is a NP problem– The total number of required colors is defined as– For various , is different and uncertain
Total 4 pilots are required
41Massive MIMO for 5G: From Theory to Practice
GC based Pilot Allocation (GC-PA) Under limited pilot resource
– For users within each cell, usually only pilots are available– Users are sorted according to their degrees in , and assign pilots to
these users in a sequential way– Assign different pilots to connected users as much as possible– For various , only pilots are utilized for pilot assignment
Total 3 pilots are utilized
42Massive MIMO for 5G: From Theory to Practice
Analysis of Threshold How to obtain
– The potential PC graph is constructed based on – The initial interval of can be easily obtained as
∈ ,
– min , ,⟨ , ⟩ and max , ,⟨ , ⟩
– By setting , only users within one cell are connected– By setting , all users are connected to each other– To obtain a near-optimal threshold , an iterative grid search (IGS)
algorithm is adopted
, , , .– denotes the number of grids in each search step– denotes the number of iterations
43Massive MIMO for 5G: From Theory to Practice
Iterative Grid Search (IGS) of IGS Algorithm
– The interval , is uniformly sampled by points in the first iteration
: 1 , ,
Δ , Δ 1 .
– By denoting as one out of that can achieve the best performance, a sub interval can be obtained after the first iteration, i.e.,
Δ2 ,
Δ2
– The sub interval will be further sampled, and this procedure is carried out in an iterative way for times
– Finally, a near-optimal threshold can be obtained
44Massive MIMO for 5G: From Theory to Practice
IGS Algorithm– System parameters: (1) 7; (2) 8; (3) 128– IGS parameters: (1) 20; (2) 2
Simulation result (1)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 110
12
14
16
18
(th-min)/max
Ave
rage
upl
ink
SIN
R (d
B)
First iteration of the IGS
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 117
17.05
17.1
17.15
(th-max(1) +(1)/2)/(max
(1) +(1)/2)
Ave
rage
upl
ink
SIN
R (d
B)
Second iteration of the IGS
The sub-interval for the nextiteration of the IGS process
The final threshold th=max(2)
45Massive MIMO for 5G: From Theory to Practice
CDF curve of users’ uplink achievable rate– System parameters: (1) 4; (2) 4; (3) 128– The optimal solution is obtained by exhaustive search
Simulation result (2)
1 2 3 4 5 6 70
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
User UL achievalable rate (bps/Hz)
CD
F
Classical scheme [1]Conventional GCAs [5]Proposed GC-PA schemeOptimal PA
The performance of the userssuffering from severe PC hasbeen significantly improved
The proposed GC-PA schemecan approach the optimal pilotallocation
46Massive MIMO for 5G: From Theory to Practice
Average uplink achievable rate against antenna number– System parameters: (1) 7; (2) 84; (3) 10 10000
Simulation result (3)
101 102 103 1041.5
2
2.5
3
3.5
4
4.5
5
Number of BS antennas M
Ave
rage
UL
capa
city
per
use
r (bp
s/H
z)
Classical scheme [1]Conventional GCAs [5]Proposed GC-PA scheme
Gain of the significantly reducedPC by conventional GCA and theproposed GC-PA scheme
The performance of theclassical random schemeis limited by PC
Gain of the restricted pilotresource.
47Massive MIMO for 5G: From Theory to Practice
A graph coloring based pilot allocation (GC-PA) scheme isproposed to mitigate pilot contamination for massiveMIMO
Basic ideas– Construct potential PC graph for multi-cell multi-user network
– GC-PA: Assign different pilots to connected users in PC graph to mitigate severe PC as much as possible
– An iterative grid search (IGS) algorithm is proposed to obtain a near-optimal threshold for PC graph construction
Simulation result (2) has verified the near-optimal (0.1bps/Hz) performance of our method compared withoptimal solution through exhaustive search
Conclusions
48Massive MIMO for 5G: From Theory to Practice
Practical Solutions: Work 3
Compressive Sensing Based Efficient Pilot Design and Channel Estimation
Zhen Gao, Linglong Dai, Zhaocheng Wang, Sheng Chen, “Spatially common sparsity based adaptive channel estimation and feedback for FDD massive MIMO,” IEEE Transactions on Signal Processing, vol. 63, no. 23, pp. 6169-6183, Dec. 2015.
49Massive MIMO for 5G: From Theory to Practice
Motivation Orthogonal pilots for LTE/LTE-A
Different channels are distinguished by orthogonal pilots 100% pilot !
50Massive MIMO for 5G: From Theory to Practice
Angle-Domain Massive MIMO Channels
Massive MIIMO channels is sparse in the angle domain due to angle spread is small (e.g., 10o) at the BS with limited scattersX. Rao and V. K. N. Lau, “Distributed compressive CSIT estimation and feedback for FDD multi-user massive MIMOsystems,” IEEE Trans. Signal Process., vol. 62, no. 12, pp. 3261–3271, Jun. 2014.
T Tn n n n n B n ny w w x h x A h
Angle-domain channel
a DFT matrix for ULA with
180 / 1.406sf M
10 10 1.406 8
128M
suppn n aS M h
/ 2d
51Massive MIMO for 5G: From Theory to Practice
Spatially common sparsity holds due to spatial propagation characteristics of the channels within the system bandwidth are almost unchanged
X. Rao and V. K. N. Lau, “Distributed compressive CSIT estimation and feedback for FDD multi-user massive MIMOsystems,” IEEE Trans. Signal Process., vol. 62, no. 12, pp. 3261–3271, Jun. 2014.
K M
K M
K M
Spatially Common Sparsity
1 2supp supp supp N h h h
52Massive MIMO for 5G: From Theory to Practice
Compressive Sensing (CS)‒ (1949) Shannon-Nyquist sampling theory: sufficient condition for
perfect reconstruction of a bandwidth limited signals
‒ (2006) Compressive sensing: Acquire and reconstruct a sparse signal by a sampling rate much lower than the Nyquist rate
fs 2B
M N
Shannon
DonohoD. L. Donoho, “Compressed sensing”, IEEE Trans. Info. Theory, vol. 52, no. 4, pp. 1289–1306, Apr. 2006. (cited by 14106 times)
Background of Compressive Sensing
53Massive MIMO for 5G: From Theory to Practice
Key idea of CS‒ Conventional way: sampling the signal at a high rate first, and then compress the signal
to remove the redundancy‒ CS: directly sampling the inherent information of signals, and then reconstruct the high-
dimensional signal from low-dimensional measurement via optimization‒ Three steps of CS: 1) Spare representation; 2) Compression; 3) Recovery‒ Applications of CS: Image/Vedio processing, MRI, Radar, Wireless communications …
Compressive sensing is a breakthrough theory for signal processing, which has great potential impacts in many applied fields including wireless communications
Background of Compressive Sensing
54Massive MIMO for 5G: From Theory to Practice
Conventional orthogonal pilot proposed non-orthogonal pilots
Proposed Non-Orthogonal Pilots
55Massive MIMO for 5G: From Theory to Practice
Non-orthogonal pilots based channel estimation byexploiting CS theory
CS-Based Channel Estimation
N subc
arrier
s
K M
K M
K M
M BS antennas
1 2supp supp supp N h h h Spatially common sparsity:
Received pilot signal in G time slots: ( ) ( )[ , ] [ , ] * [ , ] [ , ] [ , ]T q qq G q G q G q G q G
p pp p B p p p r S A h v Φ h v
56Massive MIMO for 5G: From Theory to Practice
Proposed Distributed Sparsity AdapativeMatching Pursuit (DSAMP) Algorithm
Joint Processing
57Massive MIMO for 5G: From Theory to Practice
2
0, 1min .
l
L
ll lf
f Problem 1
D
Pilot design
Performance Analysis
i i i i i i i r S h S Ah Θ h
s.t. ,supp ,l l l l l d Φ f f
, ,[ , ]
,1 ,1t m pjq G
p t me t G m M S
58Massive MIMO for 5G: From Theory to Practice
Performance approaches the CRLB with substantially reduced pilot overhead !
Simulation parameters:
1. Carrier frequency 2 GHz
2. System bandwidth 10 MHz
3. FFT size 2048
4. BS 128 Tx
5. 15°angle spread
10 15 20 25 3010-3
10-2
10-1
100
SNR (dB)
MS
E
J-OMP, Fixed Time Overhead, T=18DSAMP, Fixed Time Overhead, T=18CRLB of Conventional Linear Algorithms
Simulation Results
59Massive MIMO for 5G: From Theory to Practice
Conclusions This paper focuses on the downlink pilot design and
channel estimation for massive MIMO systems At the transmitter, compared with standardized
orthogonal pilots (pilot overhead ∝ No. of Tx), weproposed the non-orthogonal pilots design based onstructured CS can effectively solve this issue (pilotoverhead ∝ small angle spread)
At the receiver, the proposed DSAMP algorithm canexploit the spatially common sparsity of massive MIMOchannels for reliable channel estimation
Moreover, the proposed scheme can be applied in theuplink to solve the issue of pilot contamination
60Massive MIMO for 5G: From Theory to Practice
Practical Solutions: Work 4
Low-Complexity Multi-User Detection for Uplink Massive SM-MIMO
Zhen Gao, Linglong Dai, Zhaocheng Wang, Sheng Chen, and Lajos Hanzo, “Compressive Sensing Based Multi-User Detector for Large-Scale SM-MIMO Uplink,” to appear in IEEE Transactions on Vehicular Technology.
61Massive MIMO for 5G: From Theory to Practice
Motivation and Background
Key requirements of 5G Spectrum Efficiency (SE) Energy Efficiency (EE)
Key techniques Massive MIMO improves SE at cost of low EE
62Massive MIMO for 5G: From Theory to Practice
Motivation and Background
Key requirements of 5G Spectrum Efficiency (SE) Energy Efficiency (EE)
Key techniques Massive MIMO improves SE at cost of low EE Spatial modulation (SM) MIMO improves EE at cost of low SE
63Massive MIMO for 5G: From Theory to Practice
System Model
Spatial Modulation (SM) MIMO No. of RF chains << No. of antennas
64Massive MIMO for 5G: From Theory to Practice
System Model
Spatial Modulation (SM) MIMO No. of RF chains << No. of antennas 3-D constellation set Spatial and signal constellation symbols
65Massive MIMO for 5G: From Theory to Practice
System Model
Massive SM-MIMO High SE
Large No. of antennasLow cost antennas
High EESmall number of RF chainsLow hardware cost Low power consumption
66Massive MIMO for 5G: From Theory to Practice
System Model
Challenges of Massive SM-MIMO Support multi-user transmission in the uplink
Only consider single-user scenario
Optimal for multi-user detection Large No. of users Large No. of antennas Optimal maximum like-hood (ML): High complexity Sphere decoding: High complexity
Low complexity signal detector LMMSE-based detector [Renzo’14]: poor performance CS-based detector [Liu’14]: poor performance
[Liu’14] W. Liu, N. Wang, M. Jin, and H. Xu, “Denoising detection for the generalized spatial modulation system using sparseproperty,” IEEE Commun. Lett., vol. 18, no. 1, pp. 22-25, Jan. 2014.
[Renzo’14] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, “Spatial modulation for generalized MIMO: Challenges,opportunities and implementation,” Proc. IEEE, vol. 102, no. 1, pp. 56-103, Jan. 2014.
67Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
How to support multi-user transmission in uplink?
One RF chain but multiple antennas for each user !
68Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
How to reduce the complexity of signal detector?
Sparsity of SM signals can be exploited!
1x
2x
Kx
69Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
Sparsity of SM signals can be exploited!
k k ksx eSignal constellation symbol (M-PSK, M-QAM)
Spatial constellation symbol
0 2supp , 1, 1.k k k e Q e e
70Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
1 1
K K
k k kk k
y y w H x w
The kth user's MIMO channel matrix
The kth user's SM signal
AWGN
Sparsity of SM signals can be exploited!
Kx
1x
2xx
71Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
1 1
K K
k k kk k
y y w H x w
y Hx w
Compressive sensing problem !
Sparsity of SM signals can be exploited!
Kx
1x
2xx
72Massive MIMO for 5G: From Theory to Practice
Compressive Sensing (CS)‒ (1949) Shannon-Nyquist sampling theory: sufficient condition for
perfect reconstruction of a bandwidth limited signals
‒ (2006) Compressive sensing: Acquire and reconstruct a sparse signal by a sampling rate much lower than the Nyquist rate
fs 2B
M N
Shannon
DonohoD. L. Donoho, “Compressed sensing”, IEEE Trans. Info. Theory, vol. 52, no. 4, pp. 1289–1306, Apr. 2006. (cited by 14106 times)
Background of Compressive Sensing
73Massive MIMO for 5G: From Theory to Practice
Key idea of CS‒ Conventional way: sampling the signal at a high rate first, and then compress the signal
to remove the redundancy‒ CS: directly sampling the inherent information of signals, and then reconstruct the high-
dimensional signal from low-dimensional measurement via optimization‒ Three steps of CS: 1) Spare representation; 2) Compression; 3) Recovery‒ Applications of CS: Image/Video processing, MRI, Radar, Wireless communications …
Compressive sensing is a breakthrough theory for signal processing, which has great potential impacts in many applied fields including wireless communications
Background of Compressive Sensing
74Massive MIMO for 5G: From Theory to Practice
Sparsity of SM signals can be exploited!
Proposed Solutions
y Hx w
Standard compressive sensing
Uniquely sparsity
Block sparsity
Reduced complexity and improved performance
Kx
1x
2xx
75Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
How to enhance the reliability of signal detector? Introduce the structured sparsity
1 2supp supp supp Jk k k x x x
,1 .j j j j j J y H x w
( ) ( )
1 1
22( ) ( )( ) ( ) ( ) ( )
21 1 1 2
min minJ Jj j
j j
J J Kj jj j j jkk
j j k
x x
y H x y H x
2( )
0s.t. 1,1 ,1 .
jk j J k K x 1 2supp supp supp J
k k k x x x
Kx
1x
2xx
3J
76Massive MIMO for 5G: From Theory to Practice
Proposed Solutions
How to enhance the reliablity of signal detector? Introduce the channel diversity by interleaving
1 2 J H H H
Temoral correlationof channels
Improved performance
Cyclic shiftInterleaving
Substantially improvedperformance
,1 .j j j j j J y H x w
1 2 J H H H
77Massive MIMO for 5G: From Theory to Practice
Example for Interleaving (J=2)
78Massive MIMO for 5G: From Theory to Practice
Proposed Group SP (GSP) algorithm
79Massive MIMO for 5G: From Theory to Practice
Proposed Solutions Complexity comparison
‒ The optimal ML detector suffers from high complexity‒ MMSE and CS detectors have low complexity but poor performance‒ Proposed GSP algorithm enjoys low complexity with good performance
Algorithm ML MMSE CS ProposedGSP
Complexity ( )KtO Ln 2 3
RF ( ) ( )( )t tO M n K n K 2 3RF2O M K K 2 3
RF2( )O M K K
80Massive MIMO for 5G: From Theory to Practice
Simulations
Obvious performance gain from interleaving! Point-to-point SM-MIMO:1. No. Tx 642. No. Rx 163. Correlation coff 0.4 for Tx/Rx4. One Tx RF chain5. 6 bit spatial constel. syb.6. 8-PSKsignal constel. syb.7. J=2
[Liu’14] W. Liu, N. Wang, M. Jin, and H. Xu, “Denoising detection for the generalized spatial modulation system using sparseproperty,” IEEE Commun. Lett., vol. 18, no. 1, pp. 22-25, Jan. 2014.
[Renzo’14] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, “Spatial modulation for generalized MIMO: Challenges,opportunities and implementation,” Proc. IEEE, vol. 102, no. 1, pp. 56-103, Jan. 2014.
81Massive MIMO for 5G: From Theory to Practice
Simulations
Higher throughput than massive MIMO ! Near-optimal signal detection performance !
Multi-user SM-MIMO uplink:1. BS 64 Tx but 18 receive RF chains2. 8 Users each 4 Tx and 1 transmit RF chain3. Correlation coff 0.5 for Tx/Rx
[Liu’14] W. Liu, N. Wang, M. Jin, and H. Xu, “Denoising detection for the generalized spatial modulation system using sparseproperty,” IEEE Commun. Lett., vol. 18, no. 1, pp. 22-25, Jan. 2014.
[Renzo’14] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, “Spatial modulation for generalized MIMO: Challenges,opportunities and implementation,” Proc. IEEE, vol. 102, no. 1, pp. 56-103, Jan. 2014.
82Massive MIMO for 5G: From Theory to Practice
Conclusions
This paper focuses on the multi-user detection for uplinkmassive SM-MIMO
A reliable and low-complexity multi-user signal detectoris proposed
Proposed signal detector can fully exploit the blocksparsity of equivalent SM signal for the reducedcomplexity
By introducing the SM signal interleaving, the signaldetection performance can be further improved
Simulation results have demonstrated the goodperformance of the proposed scheme
83Massive MIMO for 5G: From Theory to Practice
Practical Solutions: Work 5
SIC-Based Energy-Efficient Hybrid Precoding for Massive MIMO
Xinyu Gao, Linglong Dai, Shuangfeng Han, Chih-Lin I, and Robert Heath, “Energy-Efficient Hybrid Analog and Digital Precoding formmWave MIMO Systems with Large Antenna Arrays”, to appear in IEEE Journal on Selected Areas in Communications.
84Massive MIMO for 5G: From Theory to Practice
MmWave massive MIMO Why mmWave?
Why mmWave + massive MIMO?– Short wavelength enables large antenna array in massive MIMO– Massive MIMO provides sufficient gains to compensate the serious
path-loss by using precoding
mmWave
High frequencies Short wavelength Serious path-loss
Spectrum extension Massive MIMO Small cell
1000x capacity increase!
85Massive MIMO for 5G: From Theory to Practice
Challenges– Traditional MIMO: one dedicated RF chain for one antenna Enormous number of RF chains due to large antenna array High complexity in signal processing Unaffordable energy consumption (250 mW per RF chain) 64 antennas → 64 RF chains → 16 W !
How to reduce the requirednumber of RF chains?How to reduce the requirednumber of RF chains?
Challenges of mmWave massive MIMO
86Massive MIMO for 5G: From Theory to Practice
Precoding for mmWave massive MIMO Traditional precoding
– Preformed in digital domain with optimized performance – One RF chain is required to support one transmit antenna– Impractical in energy consumption for mmWave massive MIMO
250mW per RF chain, and 16W for 64 antennas [Amadori’15] !
Hybrid analog and digital precoding– Actual degree of freedom (i.e., #users) is much smaller than #antennas– Divide digital precoding with large size into:
Digital precoding with small size Analog precoding with large size (realized by phase shifter, PS)
– Significantly reduced number of RF chains – Power-efficient, low complexity, without obvious performance loss
[Amadori’15] P. Amadori and C. Masouros, “Low RF-complexity millimeter-wave beamspace-MIMO systems by beam selection,” IEEE Trans. Commun., vol. 63, no. 6, pp. 2212-2222, Jun. 2015.
87Massive MIMO for 5G: From Theory to Practice
Existing hybrid precoding architectures Fully-connected architecture
– RF chain is fully connected to all antennas Large number of PSs (N2M) Near-optimal but energy-intensive
– Spatially sparse precoding [Ayach’14]– Codebook-based hybrid precoding [Roh’14]
Sub-connected architecture– RF chain is partially connected to a subset
of antennas Smaller number of PSs (NM) More energy-efficient
– The optimal solution is unavailable Challenge: changed constraints
[Ayach’14] O. El Ayach S. Rajagopal, S. Abu-Surra,Z. Pi, and R.W. Heath, “Spatially sparse precoding in millimeter wave MIMO systems,”IEEE Trans. Wireless Commun., vol. 13, no. 3,pp. 1536-1276, Mar. 2014.
[Roh’14] W. Roh, et al., “Millimeter-wave beamforming as an enabling technology for 5G cellular communications: Theoretical feasibility andprototype results,” IEEE Commun. Mag., vol. 52, no. 2, pp. 0163-6804, Feb. 2014.
88Massive MIMO for 5G: From Theory to Practice
Problem formulation System model
Total achievable rate
2 2log .H HNR
N
I HPP H
, y HADs n HPs n
Three non-convex constraints– Structure constraint:– Amplitude constraint: All elements of have fixed amplitude – Power constraint:
Target– Jointly design A and D to maximize the achievable rate
FNP
1 1diag , , diag , ,N Nd d P AD a a ia 1/ M
89Massive MIMO for 5G: From Theory to Practice
SIC-based hybrid precoding Successive interference cancelation (SIC) for multi-user
signal detection4
( ) ( )1
i ii
y h x
90Massive MIMO for 5G: From Theory to Practice
1p 2p Np
1T 1NT0 NT I
where is the nth column of P, , and
SIC-based hybrid precoding Proposition 1: The total rate R can be decomposed as
SIC-based hybrid precoding– Total rate sub-rate of sub-antenna array– Optimize the sub-rate of each sub-antenna array one by one by exploiting the
concept of SIC for multi-user detection
12 12
1
log 1N
H Hn n n
n
RN
p H T Hp
2H H
n N n nN
T I HP P H 0 NT Inp
91Massive MIMO for 5G: From Theory to Practice
Find sufficiently close to to maximize the achievable sub-rate Find sufficiently close to to maximize the achievable sub-rate
Solution to the sub-rate optimization problem
We prove that it is equivalent to a simplified problem
Target– Optimize achievable rate of the nth sub-antenna array
– Consider non-zero elements
opt2 12arg max log 1
n
Hn n n nN
p
p p G p ,
2opt1 2
arg minn
n n
p
p v p ,
where 11 1
Hn n
G H T H
opt2 12arg max log 1
n
Hn n n n
F N
p
p p G p ,
where , 1 1H
n n G RG R 1 MM M n M M N n R 0 I 0
where is the first right singular vector of 1v 1nG
np 1v
Proposition 2. The optimization problem is equivalent to the following problem
92Massive MIMO for 5G: From Theory to Practice
Design of analog and digital precoder Problem
– As we have , equals to
Solution– Analog precoder: – Digital precoder:– Hybrid precoder:– Easy to check all the three constraint conditions are satisfied
n n ndp a 21 2nv p
1angle( )opt 1 / jn M e va
Summary of our method– SVD of to obtain – Compute for the nth sub-antenna array– Update for the (n+1)th sub-antenna array
2 221 12 1Re 1 Ren n
H Hn nd v p v va a
opt1 1 1
/Re Hn nd M v a v
1nG 1v 1angle( )opt
1 11 / j
n M e vp v
nG
1angle( )opt1 1
1 / jn M e vp v
93Massive MIMO for 5G: From Theory to Practice
Computation of – Only the first right singular vector of is required– Realized by power iteration algorithm with complexity
1v1nG
2M
Acquire the optimal precoder– The complexity is only to obtain 1angle( )opt
1 11 / j
n M e vp v
Update
– Corresponding complexity is
nG M
12
1 1 1 1 12 2 ,1 Hn n N N
G G v v
2M
Total complexity – Only 10% of [11] !
2 ( )M NS K
is largest singular value of 1 1nG
Complexity analysis
Proposition 2. The matrix can be simplified as nG
94Massive MIMO for 5G: From Theory to Practice
-30 -25 -20 -15 -10 -5 00
5
10
15
20
25
SNR (dB)
Ach
ieva
ble
rate
(bps
/Hz)
Optimal unconstrained precoding (full-connected)Spatially sparse precoding (full-connected) [6]Optimal unconstrained precoding (sub-connected)Proposed SIC-based precoding (sub-connected)Conventional analog precoding (sub-connected) [21]
-30 -25 -20 -15 -10 -5 00
5
10
15
20
25
30
SNR (dB)
Ach
ieva
ble
rate
(bps
/Hz)
Optimal unconstrained precoding (fully-connected)Spatially sparse precoding (fully-connected) [6]Optimal unconstrained precoding (sub-connected)Proposed SIC-based precoding (sub-connected)Conventional analog precoding (sub-connected) [21]
Simulation setup– Antennas: (1) (2)– RF chains: (1) (2) – Channel: Geometric Saleh-Valenzuela model
64 16NM K 128 32NM K 8N 16N
87%
SIC-based hybrid precoding is near-optimal!SIC-based hybrid precoding is near-optimal!
Simulation results
95Massive MIMO for 5G: From Theory to Practice
We proposed a hybrid precoding scheme with sub-connectedarchitecture for mmWave massive MIMO systems
Basic ideas– Decompose the total achievable rate into the sum of sub-rates
– Optimize the sub-rate of each sub-antenna array one by one by exploiting the concept of SIC for multi-user detection
The computational complexity of our method is ,only 10% of that of conventional scheme
Simulation results have verified the near-optimal (87%)performance of our method
2 ( )M NS K
Conclusions
96Massive MIMO for 5G: From Theory to Practice
Practical Solutions: Work 6
Beamspace Massive MIMO
Xinyu Gao, Linglong Dai, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang, “Channel Estimation and Beam Selection in Beamspace for Millimeter-Wave Massive MIMO,” to be submitted to IEEE Transactions on Signal Processing.
97Massive MIMO for 5G: From Theory to Practice
Advantages of mmWave massive MIMO
Advantages– Larger bandwidth: 50MHz → 1GHz More users and higher capacity
– Larger antenna array: 1~8 → 256~1024 Larger antenna gain to compensate serious path loss More data streams to improve spectral efficiency
mmWave
High frequencies Short wavelength Serious path-loss
Spectrum expansion Large antenna array Small cell
1000x data rates increase!
98Massive MIMO for 5G: From Theory to Practice
Challenges– Traditional MIMO: one dedicated RF chain for one antenna Enormous number of RF chains due to large antenna array High complexity in signal processing Unaffordable energy consumption (250 mW per RF chain) 64 antennas → 64 RF chains → 16 W !
How to reduce the requirednumber of RF chains?How to reduce the requirednumber of RF chains?
Challenges of mmWave massive MIMO
99Massive MIMO for 5G: From Theory to Practice
Category 1: Hybrid beamforming
Basic idea [Ayach’14,Gao’15]– Decompose fully digital beamformer of large size: Digital beamformer with small size (realized by RF chains) Analog beamformer with large size (realized by phase shifters)
Performance– Reduce RF chains by signal
processing– Not obvious performance loss– Require complicated design– High computational complexity
[Ayach’14] O. El Ayach, et al., “Spatially sparse precoding in millimeter wave MIMO systems,” IEEE Trans. Wireless Commun., 2014.
[Gao’15] X. Gao, et al., “Energy-efficient hybrid analog and digital precoding for mmwave MIMO systems with large antenna arrays,”IEEE J. Sel. Areas Commun., 2015.
100Massive MIMO for 5G: From Theory to Practice
Category 2: Beamspace MIMO
Basic idea [Brady’14]– Transform spatial channel into beamspace channel (realized by lens) Limited scatters → beamspace channel is approximately sparse
– Select beams to reduce dimension (realized by switching network) – Digital beamformer with small size (realized by RF chains)
[Brady’14] J. Brady, et al., “Beamspace MIMO for millimeterwave communications: System architecture, modeling,analysis, and measurements,” IEEE Trans. Ant. and Propag., vol. 61, no. 7, pp. 3814–3827, Jul. 2013.
Performance– Reduce RF chains by discrete
lens array (DLA)– Classical beamformers can be
directly employed– Low computational complexity
A different but promising thought to reduce RF chainsA different but promising thought to reduce RF chains
101Massive MIMO for 5G: From Theory to Practice
System model– single-antenna users, BS with antennas, RF chains
– Saleh-Valenzuela channel model [Ayach’14]
where : spatial direction and : physical direction
– Transform the spatial channel into beamspace
where
Principle of beamspace MIMO
,H H y H x n H Ps n 1 2, , , N KK
H h h h NK
0 0
1
,L
i ik k k k k
i
h a a
21 ,j m
m Ne
N
a
LoS path NLoS paths ULA steering vector
sind
,H H H y H U Ps n H Ps n
DFT matrix realized by DLA
1 / 2, 0,1, , 1 ,N l N l N Beamspace channel
RFN K
1 2, , , ,H
N U a a a
1 / 2 / , 1,2, ,n n N N n N
[Ayach’14] O. El Ayach, et al., “Spatially sparse precoding in millimeter wave MIMO systems,” IEEE Trans. Wireless Commun., 2014.
102Massive MIMO for 5G: From Theory to Practice
Sparsity
– only has small number of dominantelements
– Approximately sparse
User index
Bea
m in
dex
2 4 6 8 10 12 14 16
10
20
30
40
50
60
1
2
3
4
5
6
7
8
9
10
11
12
Beamspace channel
1 2 1 2, , , , , ,K K H h h h UH Uh Uh Uh
Beam selection– Select a small number of dominant beams
– is the dimension-reduced precoder
– Only a small number of RF chains
kh
r r ,H y H P s n r ,:l
l
H H
rP
103Massive MIMO for 5G: From Theory to Practice
Challenges
1st : Channel estimation– should be estimated with only RF chains– Sparsity of should be fully utilized
2nd : Beam selection– Different users may select the same beam Severe interference Number of RF chains is uncertain and unfixed
– Near-optimal performance should be achieved with low complexity
HH
We propose an interference-aware (IA) beam selection scheme We propose an interference-aware (IA) beam selection scheme
HH
We propose a support detection (SD) based channel estimation algorithmWe propose a support detection (SD) based channel estimation algorithm
RFN K
104Massive MIMO for 5G: From Theory to Practice
Outline of SD-based channel estimation Technological process
– Consider TDD system– All the users transmit pilots to the BS– The BS employs analog combining to obtain measurement vectors– OMP algorithm is utilized to estimate the channel with low overhead– Channel reciprocity is utilized to obtain the downlink channel matrix
K
105Massive MIMO for 5G: From Theory to Practice
Measurements– All the users transmit orthogonal pilots over time slots – During the mth time block
– The BS employs a combining matrix of size
Consider the kth column of the measurement matrix
Measurements
Pilot:
mW K N
K M QK
UL , 1,2, , ,m m m m m m M Y UHΨ N HΨ N H Hm m m m KΨ Ψ Ψ Ψ I
ULm m m m m m m R W Y W HΨ W N effH
m m m m m Z R Ψ W H N
eff1, 1 1,
eff2, 2 2,
eff, ,
,
k k
k kk k k
M k M M k
k
z W nz W n
z
hz h W n
W n
A typical sparse signal recovery problem !A typical sparse signal recovery problem !
mZ
106Massive MIMO for 5G: From Theory to Practice
How to realize the combining matrix?
Beam selection is realized by switching network
Switching network can not be used for combing!
107Massive MIMO for 5G: From Theory to Practice
Proposed adaptive phase shifter network
We propose to replace switching network by adaptive phase shifter
network
The bridge to connect hybrid precoding and beamspace MIMO
108Massive MIMO for 5G: From Theory to Practice
How to realize beam selection by phase shifter network?– Based on , turn off some phase shifters (i.e., 0) and set some
phase shifters to shift the phase 0 degree (i.e., 1)
Beam selection via phase shifter network
opt
111 12
221 22
1 2
N
N
KNK K
jj j
jj j
jj j
e e ee e e
e e e
Channel estimation and beam selection can share the same moduleChannel estimation and beam selection can share the same module
Combining matrix (realized by phase shifter network) for
channel estimation
Switching matrix (realized by adaptively reconfiguring the phase shifter network) for beam selection
0
0
0
0 00 0
0 0
j
j
j
ee
e
109Massive MIMO for 5G: From Theory to Practice
Design of combinging matrix– i.i.d. Bernoulli random matrix enjoys satisfying estimate performance Each element of belongs to with equal probability
– Realized by phase shifters Only 1-bit phase shifters is required, low energy consumption
How to design combining matrix
W
Observation
Sparse beamspace channel vectorPerforms like the sensing matrix in CS
1/ 1, 1Q
Estimate the channel– Classical compressed sensing algorithms can be used– Poor performance when SNR is low
[Ayach’14] O. El Ayach, et al., “Spatially sparse precoding in millimeter wave MIMO systems,” IEEE Trans. Wireless Commun., 2014.
,kk k hz W n
W
More efficient algorithm is requiredMore efficient algorithm is required
110Massive MIMO for 5G: From Theory to Practice
SD-based channel estimation Problem: poor performance at low SNR Solution: exploiting asymptotical orthogonality and
structural characteristics of the beamspace channel
Insight of Proposition 1– The total channel estimation problem can be decomposed into a series
of independent sub-problems
Proposition 1. Represent the beamspace channel as ,where is the ith channel component of in the beamspace. Then, anytwo channel components and in the beamspace are asymptoticallyorthogonal when the number of antennas N in beamspace mmWave massiveMIMO systems tends to infinity, i.e.,
kh0
/ Lk ii
N L
h c i ic Uc kh
ic jc
lim 0, , 0,1, , , .Hi jN
i j L i j
c c
111Massive MIMO for 5G: From Theory to Practice
SD-based channel estimation Problem: poor performance at low SNR Solution: exploiting asymptotical orthogonality and
structural characteristics of the beamspace channel
Proposition 2. Consider the beamspace channel of the kth user. The ratiobetween the power of V strongest elements of and the total power of thechannel can be lower-bounded by
Once the position of the strongest element of is determined, other V-1strongest elements will uniformly locate around it.
khVP kh TP
/2
21 2
2 1 .2 1
sin2
VV
iT
PP N i
N
khn
112Massive MIMO for 5G: From Theory to Practice
Best caseWorst case
1 / N 1 / N
k1
1
sin2
NN
13sin2
NN
1 / 2N
Insight of Proposition 2– can be well-approximated by a sparse vector
– The support (positions of nonzero elements) of is determined by
kh
256, 8, / 95%V TN V P P
kh
We can directly obtain the support of at time according to We can directly obtain the support of at time according to kh
CS-based channel estimation
n
n
113Massive MIMO for 5G: From Theory to Practice
SD-based channel estimation
Insight
114Massive MIMO for 5G: From Theory to Practice
Problem– Only retain power without considering multi-user interferences – The number of required RF chains is uncertain and unfixed
Existing beam selection method
[Sayeed’14] A. Sayeed and J. Brady, “Beamspace MIMO for high-dimensional multiuser communication at millimeter-wave frequencies,” in Proc. IEEE GLOBECOM’13, Dec. 2013, pp. 3679–3684.
Magnitude maximization (MM) beam selection [Sayeed’14]– Select rows (beams) of with the largest magnitude– The corresponding beam indices set– The selected beams for all K users
khV 1 2= , , , V
k k k ks s s
1,2,= k
k K
Difficult to be realized in practical system!Difficult to be realized in practical system!
115Massive MIMO for 5G: From Theory to Practice
Interference-aware (IA) beam selection
Stage 1: Identify IUs and NIUs– Classify all K users into two user groups, i.e., interference-users (IUs)
and noninterference-users (NIUs), according to their strongest beams.
Stage 2: Search the best unshared beam– Propose an incremental algorithm to search the appropriate beam for
each IU
Motivation– Select the best beam for each user without repeat– The required number of RF chains is fixed as the number of users
116Massive MIMO for 5G: From Theory to Practice
Inspiration– The strongest beam of each user contains most of the total power– will also lead to unobvious multi-user interferences– Can we directly choose ?
Stage 1: Identify IUs and NIUs
kb
* * *1 2, , , Kb b b
kb
Lemma 5. Assume that spatial directions for follow the i.i.d.uniform distribution within . The probability Pr2 that there exist userssharing the same strongest beam is
!Pr 2 1 .
!K
NN N K
0k 1,2, ,k K
0.5,0.5
Definitions– NIUs: one user k is NIU if its strongest beam is different from any
other strongest beams, i.e., – IUs: any two users k1 and k2 are IUs if
kb
* * * * *1 1 1, , , , ,k k k Kb b b b b
* *1 2k kb b
117Massive MIMO for 5G: From Theory to Practice
Stage 2: Search the best unshared beam
Step 2: Search the optimal beam set of IUs– Select the beams for NIUs– Select beams from as – Combine and to form the set – Based on , select beams of beamspace channel– The dimension-reduced MIMO system
– Search the optimal by maximizing the achievable sum-rate
– Form the optimal set of selected beams for all K users
optI
IUCard IU
K
r r r, ,: ,H
ll
y H P s n H H
H
Dimension-reduced digital precoderoptI R
IU
optIU arg max ,R
2
1
log 1 ,K
kk
R
2
, ,2 2
, ,
Hr k r k
k Hr k r mm k
h p
h p
opt opt optIU NIU
opt *NIU NIU= |kb k *
NIU1,2, , \ |kN b k
IUoptNIU
118Massive MIMO for 5G: From Theory to Practice
Interference-aware (IA) beam selection
Bea
m in
dex
119Massive MIMO for 5G: From Theory to Practice
Interference-aware (IA) beam selection
120Massive MIMO for 5G: From Theory to Practice
Simulation setup System parameters
– Frequency: 28 GHz– MIMO configuration:– Total time slots: ( time slots for per user)– Select candidate beams in the first stage of IA beam selection– Digital beamformer: Zero forcing (ZF)
Channel parameters– Channel model: Saleh-Valenzuela model– Antenna array: ULA at BS, with antenna spacing– Multiple paths: One LoS component and two NLoS components– LoS component Amplitude: Spatial direction:
– NLoS components Amplitude: Spatial direction:
2L
256 16,N K RF 16N K
/ 2d
0 0,1k 0 1 1,2 2k
2~ 0,10ik
1 1,2 2
ik
1 i L
3V 128M QK 8Q
121Massive MIMO for 5G: From Theory to Practice
NMSE of channel estimation Observations
– CS-based channel estimation can achieve satisfying accuracy– The required number of RF chains is only instead of 256– The overhead can be reduced by 62.5% (96 instead of 256 time slots)
RF 16N
2
22
2
ˆNMSE=
k k
k
k
h h
h
0 5 10 15 20 25 3010-2
10-1
100
101
Uplink SNR (dB)
NM
SE
(dB
)
Conventional OMP-based channel estimationProposed SD-based channel estimation
122Massive MIMO for 5G: From Theory to Practice
Energy efficiency & sum-rate
0 5 10 15 20 25 300
10
20
30
40
50
60
70
80
90
100
Downlink SNR (dB)
Ach
ieva
ble
sum
-rate
(bits
/s/H
z)
Fully digital systemIA beam selction with perfect CSISD-based channel estimation (uplink SNR = 0 dB)OMP-based channel estimation (uplink SNR = 0 dB)SD-based channel estimation (uplink SNR = 10 dB)OMP-based channel estimation (uplink SNR = 10 dB)SD-based channel estimation (uplink SNR = 20 dB)OMP-based channel estimation (uplink SNR = 20 dB)
8 10 20 30 40 50 60 640
2
4
6
8
10
12
14
16
18
Number of users K
Ene
rgy
effic
ient
(bps
/Hz/
W)
Fully digital systemConventional MM beam selection (2 beams per user)Proposed IA beam selection (1 beam per user)
Observations– IA beam selection can achieve much higher energy efficiency – IA beam selection with SD-based channel estimation is near-optimal
123Massive MIMO for 5G: From Theory to Practice
Conclusions We solve two challenging problems in beamspace MIMO The proposed CS-based channel estimation scheme may
be the first work to address the challenging channel estimation problem
The proposed CS-based channel estimation is realized by exploiting asymptotical orthogonality and structural characteristics of the beamspace channel
We design an adaptive selecting network to adaptively realize channel estimation and beam selection for beamspace MIMO systems
The proposed IA beam selection requires fixed number of RF chains, and achieves the performance quite close to the fully digital system
124Massive MIMO for 5G: From Theory to Practice
Content
5G in The World1
Massive MIMO: Theoretical Performance2
Massive MIMO: Practical Solutions3
4 Future Research
5 Summary
125Massive MIMO for 5G: From Theory to Practice
Advantages– Larger bandwidth: 50MHz → 1GHz More users and more traffic
– Larger antenna array: 1~8 → 64~256 Larger antenna gain to compensate serious path loss More data streams to improve spectral efficiency
mmWave
High frequencies Short wavelength Serious path-loss
Spectrum expansion Large antenna array Small cell
1000x data rates increase!
Millimeter-Wave Massive MIMO
Xinyu Gao, Linglong Dai, Shuangfeng Han, Chih-Lin I, and Robert Heath, “Energy-Efficient Hybrid Analog and Digital Precoding for mmWave MIMO Systems with Large Antenna Arrays”, to appear in IEEE Journal on Selected Areas in Communications, available at: http://arxiv.org/abs/1507.04592
126Massive MIMO for 5G: From Theory to Practice
Basic idea [Brady’14]– Transform (realized by lens) into beamspace channel Limited scatters → beamspace channel is sparse
– Select beams (realized by switching network) to reduce dimension– Digital beamformer with small size (realized by RF chains)
Xinyu Gao, Linglong Dai, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang, “Channel Estimation and Beam Selection forBeamspace Millimeter-Wave Massive MIMO Systems,” submitted to IEEE Transactions on Signal Processing.
Performance– Reduce RF chains by hardware
architecture– Classical beamformers can be
directly employed– Low computational complexity
A different but promising thought to reduce RF chainsA different but promising thought to reduce RF chains
Beamspace Massive MIMO
127Massive MIMO for 5G: From Theory to Practice
Spatial Modulation (SM)– Exploit antenna selection pattern to transmit extra data– Energy efficient but spectrum inefficient
Massive SM-MIMO– Exploit more antennas to increase the spectrum efficiency– Key challenges: signal detection and channel estimation
Massive MIMO with Spatial Modulation
Zhen Gao, Linglong Dai, Zhaocheng Wang, Sheng Chen, and Lajos Hanzo, “Compressive Sensing Based Multi-User Detector for Large-Scale SM-MIMO Uplink,” to appear in IEEE Transactions on Vehicular Technology.
128Massive MIMO for 5G: From Theory to Practice
Massive MIMO with 1-Bit ADC Performance analysis
‒ MmWave channel model‒ Different antenna array (ULA, UPA, UCA)‒ Hardware impairment‒ Imperfect CSI
Signal detector‒ 1-bit detection performance‒ Design optimization
Jiayi Zhang, Linglong Dai, Shengyang Sun, and Zhaocheng Wang, “On The Spectral Efficiency of Massive MIMO Systems with Low-Resolution ADCs,” submitted to IEEE Communications Letters.
129Massive MIMO for 5G: From Theory to Practice
Massive MIMO for THz
Massive MIMO and NOMA
Massive MIMO for UDN
Massive MIMO for Energy Harvesting
Massive MIMO for Wireless Power Transfer
Extension of Massive MIMO
Linglong Dai, Bichai Wang, Yifei Yuan, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang, “Non-Orthogonal Multiple Access for 5G: Solutions, Challenges, Opportunities, and Future Research Trends,” IEEE Communications Magazine, vol. 53, no. 9, pp. 74-81, Sep. 2015.
130Massive MIMO for 5G: From Theory to Practice
Content
Introduction1
5G and Massive MIMO2
Seven Proposals for Massive MIMO3
4 Future Research Plan
5 Summary
131Massive MIMO for 5G: From Theory to Practice
Summary Massive MIMO is very promising technology for 5G wireless
communications Theoretically, massive MIMO can increase the spectrum and
energy efficiency by orders of magnitude We have proposed several practical solutions to address
challenging problems to realize massive MIMO– Performance analysis with practical constraints – Pilot decontamination based on graph coloring– Efficient pilot design and channel estimation based on CS– Low-complexity multi-user detection for uplink massive SM-MIMO– Energy-efficient SIC-based hybrid precoding– Beamspace massive MIMO
Future research directions– mmWave massive MIMO, massive SM-MIMO, beamspace massive
MIMO– Massive MIMO for NOMA, UDN, THz, energy harvesting, wireless
power transfer…
132Massive MIMO for 5G: From Theory to Practice
Reference and Recent Publications1. Xinyu Gao, Linglong Dai, Shuangfeng Han, Chih-Lin I, and Robert Heath, “Energy-Efficient Hybrid Analog and
Digital Precoding for mmWave MIMO Systems with Large Antenna Arrays”, to appear in IEEE Journal on Selected Areas in Communications, 2015. (IF: 4.138)
2. Zhen Gao, Linglong Dai, Zhaocheng Wang, Sheng Chen, and Lajos Hanzo, “Compressive Sensing Based Multi-User Detector for Uplink Large-Scale SM-MIMO,” to appear in IEEE Transactions on Vehicular Technology, 2015.
3. Xinyu Gao, Linglong Dai, Chau Yuen, and Zhaocheng Wang, “Turbo-Like Beamforming Based on Tabu Search Algorithm for Millimeter-Wave Massive MIMO Systems,” to appear in IEEE Transactions on Vehicular Technology, 2015.
4. Wenqian Shen, Linglong Dai, Byonghyo Shim, Shahid Mumtaz, and Zhaocheng Wang, “Joint CSIT Acquisition Based on Low-Rank Matrix Completion for FDD Massive MIMO Systems,” to appear in IEEE Communications Letters, 2015.
5. Zhen Gao, Linglong Dai, Zhaocheng Wang, and Sheng Chen, “Spatially Common Sparsity Based Adaptive Channel Estimation and Feedback for FDD Massive MIMO”, IEEE Transactions on Signal Processing, vol. 63, no. 23, pp. 6169-6183,. Dec. 2015. (IF: 3.198)
6. Linglong Dai, Xinyu Gao, Xin Su, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang, “Low-Complexity Soft-Output Signal Detection Based on Gauss-Seidel Method for Uplink Multi-User Large-Scale MIMO Systems,” IEEE Transactions on Vehicular Technology, vol. 64, no. 10, pp. 4839-4845, Oct. 2015.
7. Zhen Gao, Linglong Dai, De Mi, Zhaocheng Wang, Muhammad Ali Imran, and Muhammad Zeeshan Shakir, “MmWave Massive MIMO Based Wireless Backhaul for 5G Ultra-Dense Network,” IEEE Wireless Communications, vol. 22, no. 5, pp. 13-21, Oct. 2015. (IF: 6.524)
133Massive MIMO for 5G: From Theory to Practice
Reference and Recent Publications8. Xudong Zhu, Linglong Dai, and Zhaocheng Wang, “Graph Coloring Based Pilot Allocation to Mitigate Pilot
Contamination for Multi-Cell Massive MIMO Systems,” IEEE Communications Letters, vol. 19, no. 10, pp. 1842-1845, Oct. 2015.
9. Wenqian Shen, Linglong Dai, Xudong Zhu, and Zhaocheng Wang, “Compressive Sensing Based Differential Channel Feedback for Massive MIMO,” Electronics Letters, vol. 51, no. 22, pp. 1824-1826, Oct. 2015.
10. Zhen Gao, Linglong Dai, Chau Yuen, and Zhaocheng Wang, “Asymptotic Orthogonality Analysis of Time-Domain Sparse Massive MIMO Channels,” IEEE Communications Letters, vol. 19, no. 10, pp. 1826-1829, Oct. 2015.
11. Linglong Dai, Bichai Wang, Yifei Yuan, Shuangfeng Han, Chih-Lin I, and Zhaocheng Wang, “Non-Orthogonal Multiple Access for 5G: Solutions, Challenges, Opportunities, and Future Research Trends,” IEEE Communications Magazine, vol. 53, no. 9, pp. 74-81, Sep. 2015. . (IF: 4.460)
12. Jiayi Zhang, Linglong Dai, Wolfgang H. Gerstacker, and Zhaocheng Wang, “Effective capacity of communication systems over κ-µ shadowed fading channels,” Electronics Letters, vol. 51, no. 19, pp. 1540-1542, Sep. 2015. (1.068)
13. Xinyu Gao, Linglong Dai, Yuting Hu, Yu Zhang, and Zhaocheng Wang, “A Low-Complexity Signal Detection Algorithm for Large-Scale MIMO in Optical Wireless Communications,” IEEE Journal on Selected Areas in Communications, vol. 33, no. 9, pp. 1903-1912, Sep. 2015. (IF: 4.138)
14. Jiayi Zhang, Linglong Dai, Yanjun Han, Yu Zhang, and Zhaocheng Wang, “On the Ergodic Capacity of MIMO Free-Space Optical Systems over Turbulence Channels,” IEEE Journal on Selected Areas in Communications, vol. 33, no. 9, pp. 1925-1934, Sep. 2015. (IF: 4.138)
134Massive MIMO for 5G: From Theory to Practice
Reference and Recent Publications15. Jiayi Zhang, Linglong Dai, Yu Zhang, and Zhaocheng Wang, “Unified Performance Analysis of Mixed Radio
Frequency/Free-Space Optical Dual-Hop Transmission Systems,” IEEE/OSA Journal of Lightwave Technology, vol. 33, no. 11, pp. 2286-2293, June 2015.
16. Wenqian Shen, Linglong Dai, Zhen Gao, and Zhaocheng Wang, “Spatially Correlated Channel Estimation Based on Block Iterative Support Detection for Massive MIMO,” Electronics Letters, vol. 51, no.7, pp. 587-588, Apr. 2015.
17. Xinyu Gao, Linglong Dai, Yongkui Ma, and Zhaocheng Wang, “Low-Complexity Near-Optimal Signal Detection for Uplink Large-Scale MIMO Systems,” Electronics Letters, vol. 50, no. 18, pp. 1326-1328, Aug. 2014.
18. Zhen Gao, Linglong Dai, Zhaohua Lu, Chau Yuen, and Zhaocheng Wang, “Super-Resolution Sparse MIMO-OFDM Channel Estimation Based on Spatial and Temporal Correlations,” IEEE Communications Letters, vol. 18, no. 7, pp. 1266-1269, Jul. 2014.
19. Zhen Gao, Linglong Dai, and Zhaocheng Wang, “Structured Compressive Sensing Based Superimposed Pilot Design in Downlink Large-Scale MIMO Systems,” Electronics Letters, vol. 50, no. 12, pp. 896-898, Jun. 2014.
20. Linglong Dai, Zhengyuan Xu, and Zhaocheng Wang, “Flexible Multi-Block OFDM Transmission for High-Speed Fiber-Wireless Networks,” IEEE Journal on Selected Areas in Communications, vol. 31, no. 12, pp. 788-796, Dec. 2013. (IF: 4.138)
21. Linglong Dai, Jintao Wang, Zhaocheng Wang, Paschalis Tsiaflakis, and Marc Moonen, “Spectrum- and Energy-Efficient OFDM Based on Simultaneous Multi-Channel Reconstruction,” IEEE Transactions on Signal Processing, vol. 61, no. 23, pp. 6047-6059, Dec. 2013. (IF: 3.198)
22. Linglong Dai, Zhaocheng Wang, and Zhixing Yang, “Compressive Sensing Based Time Domain Synchronous OFDM Transmission for Vehicular Communications,” IEEE Journal on Selected Areas in Communications, vol. 31, no. 9, pp. no. 460-469, Sep. 2013. (IF: 4.138)
135Massive MIMO for 5G: From Theory to Practice
Reference and Recent Publications23. Linglong Dai, Zhaocheng Wang, Jun Wang, and Zhixing Yang, “Joint Time-Frequency Channel Estimation for Time
Domain Synchronous OFDM Systems,” IEEE Transactions on Broadcasting, vol. 59, no. 1, pp. 168-173, Mar. 2013.
24. Linglong Dai, Zhaocheng Wang, and Zhixing Yang, “Spectrally Efficient Time-Frequency Training OFDM for Mobile Large-Scale MIMO Systems,” IEEE Journal on Selected Areas in Communications, vol. 31, no. 2, pp. 251-263, Feb. 2013. (IF: 4.138)
25. Linglong Dai, Chao Zhang, Zhengyuan Xu, and Zhaocheng Wang, “Spectrum-Efficient Coherent Optical OFDM for Transport Networks,” IEEE Journal on Selected Areas in Communications, vol. 31, no. 1, pp. 62-74, Jan. 2013. (IF: 4.138)
26. Linglong Dai, Zhaocheng Wang, and Zhixing Yang, “Time-Frequency Training OFDM with High Spectral Efficiency and Reliable Performance in High Speed Environments,” IEEE Journal on Selected Areas in Communications, vol. 30, no. 4, pp. 695-707, May 2012. (IF: 4.138)
27. Linglong Dai, Zhaocheng Wang, and Zhixing Yang, “Next-Generation Digital Television Terrestrial Broadcasting Systems: Key Technologies and Research Trends,” IEEE Communications Magazine, vol. 50, no. 6, pp. 150-158, Jun. 2012. (IF: 4.460)
28. Linglong Dai, Zhaocheng Wang, Changyong Pan, and Sheng Chen, “Wireless Positioning Using TDS-OFDM Signals in Single-Frequency Networks,” IEEE Transactions on Broadcasting, vol. 58, no. 2, pp. 236-246, Jun. 2012.
29. Linglong Dai, Zhaocheng Wang, and Jian Song, “TDS-OFDMA: A Novel Multiple Access System Based on TDS-OFDM,” IEEE Transactions on Consumer Electronics, vol. 57, no. 4, pp. 1528-1534, Nov. 2011.
30. Linglong Dai, Zhaocheng Wang, and Cheng Shen, “A Novel Uplink Multiple Access Scheme Based on TDS-FDMA,” IEEE Transactions on Wireless Communications, vol. 10, no. 3, pp. 757-761, Mar. 2011.
136Massive MIMO for 5G: From Theory to Practice
Jiayi Zhang, Zhen Gao, Xudong Zhu, Wenqian Shen, Xinyu Gao
MOST (973, 863), NSFC