1

Millimeter-Wave MIMO: Basic Theory, Architectures, and Technology Development

Akbar M. SayeedWireless Communications and Sensing Laboratory

Electrical and Computer Engineering

University of Wisconsin-Madison

http://dune.ece.wisc.edu

Huawei University DaysAugust 4-5, 2016

Chicago, IL

TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAAAAAAAAAAA

Supported by the NSF and the Wisconsin Alumni Research Foundation

• mmW is mature!

• How we we estimate the channel if we don’t even know the channel characteristics?

• Non-coherent vs coherent (with training)?

• Physically accurate channel modeling vs mathematically tractable statistical models

• Dumb antennas or remote radio heads!

Some Recent Comments

AMS mmW MIMO 1

KL, VS, AS, ‘03

iid fading

reality

2

Explosive Growth in Wireless Traffic

AMS mmW MIMO 2

Current Industry Approach: Small Cells & Het Nets

Denser spatial reuse of spectrum

New Spectrum: mmW and cmW

3kHz

300kHz

3MHz

30MHz

3GHz

30GHz

300MHz

300kHz

3MHz

30MHz

300MHz

3GHz

30GHz

300GHz

mmW: 30-300 GHz

Short range mmW: 60 GHz; Longer range mmW: 30-40GHz, 70/80/90GHz

Current cellular wireless: 300MHz - 5GHz

AMS mmW MIMO 3

cmW: 6-30GHz

3

• A key component of 5G

– Multi-Gigabits/s speeds

– millisecond latency

• Key Gigabit use cases

– Wireless backhaul

– Wireless fiber-to-home (last mile)

– Small cell access

• New FCC mmW allocations

– Licensed (3.85 GHz): 28, 37, 39 GHz

– Unlicensed (7 GHZ): 64-71 GHz

• New NSF-led Advanced Wireless Initiative– mmW Research Coordination Network

AMS mmW MIMO 4

Exciting Times for mmW Research

mmW Wireless: 30-300 GHz

A unique opportunity for addressing the wireless traffic growth

– Large bandwidths (GHz)

– Short wavelength (1-100mm) compact high-dimensional antenna arrays

15dBi @ 3GHz 35dBi @ 30GHz

Highly directive narrow beams

(low interference/higher security)

6” x 6” antenna @ 80GHz: 6400-element antenna array

AMS mmW MIMO 5

Beamwidth:

4 deg @ 30 GHz35 deg @ 3 GHz

Large antenna gain:

4

AMS mmW MIMO 6

mmW MIMO Potential: Dense Beamspace Multiplexing

100X antenna gain

100X spec.eff.gain

6” x 6” access point (AP) antenna array: 6000 elements @80GHz vs. 9 vs. elements @3GHz

Potential power & spectral efficiency gains over 4G

>100X increase in capacity due to beamspace multiplexing

>10X increase in capacity due to extra bandwidth

(1GHz vs 100MHz)

Small Cell Access Comparison: 4G LTE vs 5G mmW

(AS & JB GCOM’13; JB & AS SPAWC ’14)

Slice

Array

Multi User MIMO

w/ Beamforming

K users/

sector

Nb beams/sector

Small Cell

mmWave Backhaul

>1000X increase in overall small cell throughput

Idealized upper bound (non-interfering users)

Key Opportunities and Challenges

• Hardware complexity: spatial analog-digital interface

• Computational complexity: high-dimensional DSP

Electronic multi-beam steering & MIMO data multiplexing

Conceptual Framework: Beamspace MIMO

Key Challenges:

Key Operational Functionality:

AMS mmW MIMO(AS & NB Allerton’10; JB, NB, & AS TAPS ‘13)

7

5

Beamspace MIMO

Antenna space multiplexing

Beamspacemultiplexing

Discrete Fourier Transform (DFT)

n dimensional signal space

n-element array( spacing)

n orthogonal beams

Related: comm. modes in optics (Gabor ‘61, Miller ‘00, Friberg ‘07)

n spatial channels

(AS TSP ’02; AS & NB Allerton ’10; JB, NB & AS TAPS ‘13)

AMS mmW MIMO 8

Multiplexing data into multiple highly-directional (high-gain) beams

n-Element (Phased) Antenna Array

Spatial angle

TX: steering vectoror

RX: response vector

AMS mmW MIMO

Spatial frequency:

n-dimensional spatial sinusoid

9

6

Orthogonal Spatial Beams

n orthogonal spatial beams

DFT spatial modulation matrix:

n = 40

Spatial resolution/beamwidth:

(DFT)

Unitary

AMS mmW MIMO

(n-dimensional orthogonal basis)

10

Antenna vs Beamspace MIMO Representation

AMS mmW MIMO 11(AS TSP ’02)

(DFT)(DFT)

Multipath Channel

(correlated)

(decorrelated)

7

AMS mmW MIMO 12

Multipath Channel

TX RX

Antenna Domain Beamspace

Beamspace channel coefficients

Dirichlet sinc

DFT

Beamwidth:

mmW MIMO Channel: Beamspace Sparsity

Point-to-point LoS Link

Point-to-multipointmultiuser link

Communication occurs in a low-dimensional (p) subspaceof the high-dimensional (n) spatial signal space

How to optimally access the p active beams with the lowest – O(p) - transceiver complexity?

(DFT)(DFT)

TX Beam Dir.

RX

Bea

m D

ir.

AMS mmW MIMO

mmW propagation X-tics• Directional, quasi-optical• Predominantly line-of-sight• Single-bounce multipath• Beamspace sparsity

13(AS & NB Allerton ’10; Pi & Khan ‘11; Rappaport et. al, ‘13)

8

Continuous Aperture Phased (CAP) MIMO

Lens computes analog spatial DFT

Data multiplexing through

O(p)active beams

Scalable performance-complexity optimization

p digital data streams

O(p) transceivercomplexity Beam Selection

O(p) << nactive beams

AMS mmW MIMO

Hybrid Analog-Digital Beamspace MIMO Transceiver Architecture

(AS & NB Allerton ‘10, APS ‘11; JB, NB & AS TAPS ‘13) 14

Focal surface feed antennas:direct access to beamspace

Lens Array for Analog Multi-Beamforming

CAP-MIMO vs Conventional MIMO: Spatial Analog-Digital Interface

CAP MIMO:Analog Multi-Beamforming

Conventional MIMO:Digital Beamforming

Beam Selectionp << n

active beams

O(p) transceiver complexityO(n) transceiver complexity

n: # of conventional MIMO array elements (1000-10,000)

p: # spatial channels/data streams (10-1000)AMS mmW MIMO 15

p data streams p data

streams

9

CAP-MIMO vs Phased-Array-Based Architecture

Multi-beam forming mechanismO(p) transceiver

complexity

Lens+

Beamspace Array+

mmW Beam Selector Network

Phase ShifterNetwork (np)

+Combiner Network

CAP-MIMO:

Phased-Array-Based:n phase shifters per data stream

AMS mmW MIMO(e.g., Samsung; Qualcomm; Ayach et. al ‘12) 16

• O(np) complexity of the phase shifting network limits state-of-the-art to p=1

• All existing industry prototypes are based on single-beam phased arrays of modest size (<256 elements)

• Multiplexing is not possible with single-beam phased arrays

• Multiple apertures for multiple beams

AMS mmW MIMO 17

State-of-the-Art: Single-Beam Phased Arrays

10

AMS mmW MIMO 18

Multi-beam CAP-MIMO vs Single-beam Phased Arrays

14-16 Gbps100% BW/user

63 pJ/bit

1-2 Gbps1-7% BW/user

476 pJ/bit

CAP-MIMO

PHASED ARRAY

20dB

4, 25-beam CAP-MIMO Arrays(100 total beams)

(1 user/beam)

16, single-beamPhased Arrays

(16 total beams)(7 users/beam)

Beamspace MIMO framework enables optimization of both architectures

CAP-MIMO has 8X higher energy and spectral efficiency over phased arrays

Small cell design for supporting 100 users

AMS mmW MIMO 19

CAP-MIMO vs Multi-beam Phased Array

Spectral Efficiency Energy Efficiency

(X. Gao, L. Dai & AS ’16)

N=256, K=16 users, p = K =16 RF chains N=256, p = K RF chains

11

2D Arrays for 3D Beamforming

AMS mmW MIMO 20

k-th user channel:

BeamspaceTransformation Matrix:

(2D DFT)

(JB & AS SPAWC ’14)

2D steering vector:

Beamspace Linear Precoding

AMS mmW MIMO

n x K Beamspace Precoder Matrix

n x K Multiuser Channel Matrix

21

Downlink K x n system

Slice

Array

Multi User MIMO

w/ Beamforming

K users/

sector

Nb beams/sector

Small Cell

mmWave Backhaul

12

Small-Cell AP Design: 2D Beam Footprints

AMS mmW MIMO(JB & AS SPAWC ’14)

0.5” x 3” antenna @ 80GHz 2.3” x 12” antenna @ 80GHz

22

1.1” x 6” ant. @ 80GHz

Cell coverage (200m x 100m)

AMS mmW MIMO 23

Beam Selection: Sparsity Mask

Lower-dimensional K x p system

Full dimensional K x n system

Power-based thresholding to determine the set of dominant active beams

Beam sparsity mask

(AS & JB GCOM ‘13, JB & AS SPAWC ’14, JG & AS CISS ‘16)

13

Sparse Precoding: Low Computational Complexity

AMS mmW MIMO 24

(MMSE precoder: Joham, Utschick, & Nossek 2005)

(AS & JB GCOM ‘13, JB & AS SPAWC ‘14)

Capacity:

MMSEPrecoder:

BeamspaceDownlink:

O(n) O(p) matrix operations

Performance vs Complexity: Perfect CSI

AMS mmW MIMO

Upperbound vs MMSE precoder performance

25

p=16K=1600beams

2.3” x 12” ant. @ 80GHz

p=K=100beams

0.5” x 3” ant. @ 80GHz

p=4K=400beams

1.1” x 6” ant. @ 80GHz

p=K=100 (0.5” x 3”)

p=4K=400(1.1” x 6”)

p=16K=1600(2.3” x 12”)

x 3

x 10

4 beam mask/uservs

Full dimension

400 maxactive beams

p=16K=1600

Full dim. vs low-dim.# RF chains: p=K, 2K, 4K

(JB & AS SPAWC ‘14)

14

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz)

10-1

100

101

102

103

Full Dimensional Perfect CSIPerfect CSI, p=KNoisy BS, p=KNoisy CE, p=KNoisy BS+CE, p=K

Performance-vs-Complexity: with Beam Selection and Channel Estimation

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz)

10-1

100

101

102

Full Dimensional Perfect CSIPerfect CSI, p=KNoisy BS, p=KNoisy CE, p=KNoisy BS+CE, p=K

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz)

10-1

100

101

102

Full Dimensional Perfect CSIPerfect CSI, p=2KNoisy BS, p=2KNoisy CE, p=2KNoisy BS+CE, p=2K

K=10

K=50

p=K p=2K

(Path loss: 75-101 dB)

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz)

10-1

100

101

102

103

Full Dimensional Perfect CSIPerfect CSI, p=2KNoisy BS, p=2KNoisy CE, p=2KNoisy BS+CE, p=2K

Full dim: ñ = 158

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz/

use

r)

10-1

100

101

102

103

Full Dimensional Perfect CSIp=K Perfect CSIp=2K Perfect CSIFull Dimensional Noisy BS+CEp=K Noisy BS+CEp=2K Noisy BS+CE

Transmit SNR (dB)60 70 80 90 100

Cap

acit

y (

b/s

/Hz/

use

r)

10-1

100

101

102

103

Full Dimensional Perfect CSIp=K Perfect CSIp=2K Perfect CSIFull Dimensional Noisy BS+CEp=K Noisy BS+CEp=2K Noisy BS+CE

Full vs low-dim

Channel estimation (not Beam Sel.) is the limiting factorAMS mmW MIMO 26(JH & AS CISS’16)

Wideband mmW MIMO: Beam Squint Problem & Multi-beam Solution

AMS mmW MIMO 27

f/fc

-0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1

|Hb

,i(f

)|2/M

(d

B)

-18-15-12-9-6-30

i = io=25

i=24i=23i=26i=27

f/fc

-0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1

Σ|H

b,i(f

)|2/M

(d

B)

-18-15-12-9-6-30

3 beam5 beam

Phased array 3-beam CAP-MIMO 5-beam CAP-MIMO

(JB & AS ICC ’15)

15

10GHz CAP-MIMO Prototype: Proof of Concept

40cm x 40cm Lens array

n=676, p=4 (channels), R=10ft

10 GHz LoS prototype theoretical performance:100 Gigabits/sec (1 GHz BW) at 20dB SNR

Compelling gains over state-of-the-art

AMS mmW MIMO 28

DLA 1DLA 2

(JB, NB & AS TAPS ’13; JB, PT, DV & AS SPAWC ’14)

FPGA DAC IQ

VCO

BPF PA

FPGAADC

LNA BPF IQ

LO

Features: • RF Bandwidth: 1GHz• Symbol rate: 125MS/s• 4x4 CAP-MIMO• Real-time comm. • Channel meas.

P2P Link: Spatial Filtering + Diff. Detection

AMS mmW MIMO 29(AS and JB ICC ’15)

DLA 1DLA 2

RF Bandwidth: 1GHzSymbol rate: 125MS/s4x4 P2P CAP-MIMOReal-time comm.

16

P2MP Link: Space-Time Filtering & Coherent Detection

AMS mmW MIMO 30

Raw RX samples Spatial filtering Spatial+temporalfiltering

TX Frame RX Frame (MS2)

P2MP Channel Measurements @ 10GHz

AMS mmW MIMO 31

MS1MS2

Wide MS spacing

Narrow MS spacing

17

28 GHz Advanced CAP-MIMO Prototype

AMS mmW MIMO 32

6” Lens with 16-feed Array

Use cases• Real-time testing of PHY protocols• Multi-beam channel measurements• Scaled-up testbed network

Features• RF Bandwidth: 1 GHz, Symbol rate: 370 MS/s • 4 simultaneous beams/data streams• Altera Arria 10 FPGA board for backed DSP• AP – 4 MS bi-directional P2MP link• Electronic multi-beam steering & data mux.

P2MP Link

Equivalent to 600-element

conventional array!Beamwidth=4 deg

5G Outlook and Impact: Multi-scale mmW Multi-beam MIMO Networks

Key Features:• Multi-Gbps Speeds• millisecond latency• Electronic multibeam steering & data multiplexing

AMS mmW MIMO 33

Key Use Cases• Backhaul• Mobile access• Last mile fiber-to-home• Vehicular networks (sensing & comm.)• M2M comm. (Gigabit)

mmWSmall Cell

mmWBackhaul

Timeline• Fixed wireless (1-5 yr.) • Mobile access (5-10 yr.)

18

AMS mmW MIMO 34

CAP-MIMO Team: 2010-present

Gen 1: Hua Deng, Paul Thomas, David Virgilio & John Brady

Gen 3: John Brady, Luke Li, Sabareesh Ganapathy, Anuj Gadiyar

Gen 2: John Brady, Justin Cummings, Wasim Sheikh & John Hogan

• Innovations in basic theory & technology development

• Gen 2 prototype: 28 GHz, advanced functionality

• Channel measurements: massive, beamspace, and multi-beam

• Lens array and beam selector network optimization

• Spatial analog-digital interface design– Gigabit-rate DSP power hungry; more analog processing?

• Wideband high-dimensional MIMO– New insights into the “beam-squint” problem

– OFDM, SC, SC-OFDM? Short-Time Fourier Signaling

• Scaled up CAP-MIMO testbed & commercialization

Ongoing Work

AMS mmW MIMO 35

19

Conclusion

• Beamspace MIMO: Versatile theory & design framework for mmW

• CAP-MIMO: practical modular architecture

– Scalable performance-complexity optimization

• Compelling advantages over state-of-the-art

– Capacity/SNR gains

– Operational functionality

– Electronic multi-beam steering & data multiplexing

• Timely applications (multi-Gigabits/s speeds & ms latency)

– Wireless backhaul networks: fixed point-to-multipoint links

– Smart 5G Basestations: dynamic beamspace multiplexing

– Last-mile connectivity, vehicular comm, M2M, satcom. etc.

• Prototyping & technology development

– Multi-beam CAP-MIMO vs Phased arrays?

AMS mmW MIMO 36

Relevant Publications(http://dune.ece.wisc.edu)

• A. Sayeed, Deconstructing Multi-antenna Fading Channels, IEEE Trans. Signal Proc., Oct 2002.• A. Sayeed and N. Behdad, Continuous Aperture Phased MIMO: Basic Theory and Applications, Allerton

Conference, Sep. 2010.• J. Brady, N. Behdad, and A. Sayeed, Beamspace MIMO for Millimeter-Wave Communications: System

Architecture, Modeling, Analysis, and Measurements, IEEE Trans. Antennas & Propagation, July 2013.• G.-H Song, J. Brady, and A. Sayeed, Beamspace MIMO Transceivers for Low-Complexity and Near-Optimal

Communication at mm-wave Frequencies, IEEE ICASSP 2013.• A. Sayeed and J. Brady, Beamspace MIMO for High-Dimensional Multiuser Communication at Millimeter-

Wave Frequencies, IEEE Globecom, Dec. 2013.• J. Brady and A. Sayeed, Beamspace MU-MIMO for High Density Small Cell Access at Millimeter-Wave

Frequencies, IEEE SPAWC, June 2014.• J. Brady, P. Thomas, D. Virgilio, A. Sayeed, Beamspace MIMO Prototype for Low-Complexity Gigabit/s

Wireless Communication, IEEE SPAWC, June 2014.• J. Hogan and A. Sayeed, Beam Selection for Performance-Complexity Optimization in High-Dimensional

MIMO Systems, 2016 Conference on Information Sciences and Systems (CISS), March 2016.• J. Brady and A. Sayeed, Differential Beamspace MIMO for High-Dimensional Multiuser Communication,

IEEE GlobalSIP, December 2015.• A. Sayeed and J. Brady, High Frequency Differential MIMO: Basic Theory and Transceiver Architectures,

IEEE ICC, June 2015.• J. Brady and A. Sayeed, Wideband Communication with High-Dimensional Arrays: New Results and

Transceiver Architectures, IEEE ICC, Workshop on 5G and Beyond, June 2015.• A. Sayeed and T. Sivanadyan, Wireless Communication and Sensing in Multipath Environments Using

Multiantenna Transceivers, Handbook on Array Processing and Sensor Networks, S. Haykin & K.J.R. Liu Eds, 2010.

• K. Liu, V. Raghavan, A. Sayeed, Capacity Scaling and Spectrum Efficiency in Wideband Correlated MIMO Channels, iEEE Trans. Inf. Th., Oct. 2003.

AMS mmW MIMO 37Thank You!