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: