webinar smart signal processing - linköping universityebjornson/webinar_smart... · interference...
TRANSCRIPT
Smart Signal Processing for Massive MIMO in 5G and Beyond
Associate Professor Emil Björnson
Department of Electrical Engineering (ISY)Linköping UniversitySweden
Formula for data throughput [bit/s/km2]:
Area Throughput bit/s/km2 = Spectrum Hz × Spectral ef;iciency bit/s/Hz/cellCell size [km2/cell]
Raising the Efficiency of Cellular Communications
2
BS in coverage tier BS in hotspot tier UE in any tier
• Two-tier networks• Hotspot tier
• High cell density, short range per cell
• Wide bandwidths in mm-wave bands
• Spectral efficiency less important
• Coverage tier (focus today)• Provide coverage, elevated base stations
• Outdoor-to-indoor coverage: Operate <6 GHz
• High spectral efficiency is desired
3.4-3.8 GHz primary 5G band in Europe
and elsewhere
3
• dffd
100001000
2
Position [m]
4
500
Spec
tral E
ffici
ency
[bit/
s/H
z]
Position [m]
500
6
8
00
100001000
2
Position [m]
4
500
Spec
tral E
ffici
ency
[bit/
s/H
z]
Position [m]
500
6
8
00
Non-uniform Spectral Efficiency is the Issue!How do we get here?
Desired: Stronger signal, same interference levels
Stronger Signals for Unlucky Users
4
Fixed beamwith 16 dBi gain
Lucky usersSNR improved by 16 dB
Unlucky usersNo benefit from antenna gain
We need adaptive beamforming!
Shadow by buildings
Evolution of Adaptive Beamforming in LTE
5
Sector antenna8 elements (7 dBi each)1 transceiver chainFixed beam 16 dBi
Classical antenna array64 elements (32 per polarization)8 transceiver chains (2 per column)Up to 8 horizontal beams
Massive antenna array64 elements64 transceiver chainsUp to 64 3D beams
1 antenna 8 antennas64 antennas
Number of antennas equals number of transceiver chains!
Using Multiple Beams for Spatial Multiplexing
• Space-division multiple access (SDMA)• Uplink in 1980s: Use ! antennas to resolve " signals
• Downlink in 1990s: Transmit beamformed signals to multiple users• Typical: ! ≈ "
6
DownlinkUplink
Massive MIMO approach: Massive number of antennas, ! → ∞T. Marzetta, “Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas,” IEEE Trans. Wireless Communications, 2010.
Interference-Free Communication with Many Antennas
7
!"#" !$#$
%→'
%→'
%→'
“Law of large numbers”
Conclusion: Noise and interference free communication
• Example: Uplink• Two users, send signals #( for ) = 1,2• Channels: !( ∈ ℂ%, random, zero mean, unit variance• Noise: 0 ~ 23(5, 6%)• Received: 8 = !"#" + !$#$ + 0Detection of #" using matched filter :" = "
% !":
:";8 =1<!";!"s" +
1<!";!$s$ +
1<!";0
s" + 0 + 0
Canonical Form of Massive MIMO
• Cellular network• Many antennas ! per base station ! ≥ 64• Spatial multiplexing of many users % % ≥ 8• Antenna-user ratio: !/% > 1• Linear transmit precoding and receive combining
8
Massive MIMO in TDD Operation
• Frame structure = Coherence block
• Coherence time: !" s• Coherence bandwidth: #" Hz• Block length: $" = !"#" symbols• Typically: $" ∈ [100,10000]
9
Bc
Time
Frequency
Tc
DownlinkUplink
Frame structure
Pilots to estimate channels:, symbols
Uplink and downlink payload data$" − , symbols
Many Signal Processing Schemes to Choose Between
10
Channel estimation
Precoding and combining
LS (least squares)MMSE (minimum-mean squared error)
Element-wise MMSE…
Matched filteringZero-forcing (ZF)
Regularized ZF
MMSE processing… Some choices are smart, some are just bad…
10
20
30
4050 60
70
80
[bit/s/Hz]
90
100 10
20
30
4050 60
70
80
[bit/s/Hz]
90
100
or
Matched Filtering is Not Optimal
Uplink example• 10 users, SNR = 15 dB, i.i.d. Rayleigh fading• Inter-cell interference is 30 dB weaker
11
0 500 1000 1500 2000Number of Antennas
0
20
40
60
80
100
120
Sum
Spe
ctra
l Effi
cien
cy [b
it/s/
Hz]
Zero forcingMatched filtering
0 500 1000 1500 2000Number of Antennas
0
20
40
60
80
100
Sum
Spe
ctra
l Effi
cien
cy [b
it/s/
Hz]
Asymptotic LimitZero forcingMatched filtering
Constant gap
Single-cell setup Multi-cell setup
More antennas
Bad processing:Matched filtering
101 102 1030
0.5
1
1.5
2
Number of Antennas
Spec
tral E
ffici
ency
[bit/
s/H
z/us
er]
MMSE processingZero forcing, Regularized ZFMatched filtering
Zero-Forcing is Not Optimal Either
• Uplink example• Rayleigh fading with a little spatial correlation• SNR −6 dB in the cell, similar to other cells
12
Growing gap
Bad processing:Matched filtering
Zero forcing
BSUE 1UE 2
300 m
300
m Cell edge
E. Björnson, J. Hoydis, L. Sanguinetti,“Massive MIMO has Unlimited Capacity,”
IEEE Trans. Wireless Communications, 2018
Interference from Other Cells is the Bottleneck
• Text
13
Desired signal Interference:UEs with same pilot
Interference:UEs with different pilot
0
5
10
15
20
25
Rec
eive
d po
wer
ove
r noi
se p
ower
[dB] Matched filter
Zero forcing, Regularized ZFMMSE processing
Matched filtering:Ignores interference
Zero forcing:Suppress interference
from own cell
MMSE processing:Suppress interference
from anywhere
Price to pay: Loss in desired signal power
! = 400
What Makes MMSE Processing Smart?
• Example: Uplink with ! cells
• "#$ = &#'$ … &#)$ ∈ ℂ,×) Channels to BS . from UEs in cell /
14
Regularized zero forcing:
0$ = "$$ "$$1+ 3
4'"$$
Matched filter:0$ =
16"$$
MMSE processing: 0$ ="'$ … "7$ "'$ … "7$
1 + 34'"$$
Desireduser
Same-celluser
Other-cell user
“…all known capacity lower bounds approach a finite limit when ! → ∞” Marzetta et al., Fundamentals of Massive MIMO (p. 158)
• We only know an estimate $%&' of %&'
• Same ( pilots used in every cell – pilot contamination
• With i.i.d. Rayleigh fading
$%&' ∝ $%'', + = 1, … , /
Hence: $%0' … $%1' $%0' … $%1'2 ∝ $%'' $%''
2
15
Destroys the gain of MMSE processing!
Causes the upper limit
This is an artifact of using a simplistic channel model!
A Little Spatial Channel Correlation Changes Everything
• Natural power variations – easy to estimate
• 4-7 dB in ULA measurements (Lund University)
16
X. Gao, O. Edfors, F. Tufvesson, E. G. Larsson, “Massive MIMO in Real Propagation Environments: Do All Antennas Contribute Equally?,” IEEE Trans. Comm., 2015
A line-of-sight user
A non-line-of-sight user
0 1 2 3 4 50
0.5
1
1.5
2
2.5
3
3.5
4
Standard deviation of power variations
Spec
tral e
ffici
ency
[bit/
s/H
z/us
er]
MMSE processingRegularized ZFMatched filtering
Random
power
variations
Channel
estimator uses
statistics to get
!"#$ ∝ !"$$
9 10 11 12
13 14 15 16
1 2
5 6
9 10
13 14
9 10
13 14
3 4
7 8
11 12
15 16
11 12
15 16
1 2
5 6
1 2 3 4
5 6 7 8
3 4
7 8
An arbitrary cell
0.25 km
1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
Which Channel Estimation Scheme to Use?
• Compare different channel estimators• 16 cells, ! = 100, % = 10
17
MMSE estimator Element-wise MMSE Least squares 0
10
20
30
40
50
60
Sum
Spe
ctra
l Effi
cien
cy [b
it/s/
Hz/
cell]
MMSE processingRegularized ZFMatched filtering
Least squaresOK if “bad” processing is used
Smart estimation:MMSE or Element-wise MMSE
Full channelstatistics
Known averagepower per element
No statisticalknowledge
MMSE estimator Element-wise MMSE Least squares 0
10
20
30
40
50
60
Sum
Spe
ctra
l Effi
cien
cy [b
it/s/
Hz/
cell]
MMSE processingRegularized ZFMatched filtering
Conclusion: Dangerous to Extrapolate Results
• From single-cell to multi-cell• Zero-forcing only good in single-cell scenarios
• Smart signal processing: MMSE processing for precoding/combining
• Observations based on simplistic channel models • Tractable with i.i.d. fading – lead to overemphasize on pilot contamination
• Matched filtering is not asymptotically optimal
• Smart signal processing: Uses spatial correlation
18
Is any of this new insights?Interference rejection combining known for decades
Yet, bad signal processing schemes dominates the Massive MIMO literature!
Definition: Massive MIMO 2.0
• Massive MIMO system• At least 64 antennas, at least 8 users
• Smart channel estimation• TDD operation, utilizing channel reciprocity• Exploit spatial correlation for better channel estimation
• Smart precoding and combining• MMSE processing to suppress all kinds of interference
19
Questions? $99 for hardback bookFree PDF at massivemimobook.com
Emil Björnson, Jakob Hoydis and Luca Sanguinetti (2017), “Massive MIMO Networks:
Spectral, Energy, and Hardware Efficiency”
Papers: ebjornson.com/researchCode: github.com/emilbjornson
Massive MIMO blog Youtube channel: