WHAT STARTS HERE CHANGES THE WORLD

(c) 2014 Robert W Heath Jr.

Millimeter Wave MIMO Precoding/Combining:

Challenges and Potential Solutions

Robert W. Heath Jr., Ph.D., P.E.

Joint work with Ahmed Alkhateeb, Jianhua Mo, and Nuria González-Prelcic

Wireless Networking and Communications Group

Department of Electrical and Computer Engineering
The University of Texas at Austin

www.profheath.org
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Heath Group in the WNCG @ UT Austin

11 PhD students

mmWave communication
mmWave beamforming and radar for car-to-car

mmWave wearables
mmWave licensed
mmWave for tactical ad shared access for 5G
hoc networks
next generation
mmWave LAN

mmWave for infrastructure-to-car mmWave 5G performance

2
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

MIMO precoding
RF RF
DAC ADC
Chain Chain

MIMO
Precoding
dnabesaB
gnidocerP
DAC
RF
Chain
H RF
Chain ADC MIMO
Baseband
Combining
Precoding
and
Equalization

RF RF
DAC ADC
Chain Chain

Precoding is a staple of modern MIMO cuisine

Widely used in commercial wireless systems especially WLAN and cellular
MIMO is a key feature of mmWave systems
Shu Sun, T. Rappapport, R. W. Heath, Jr., A. Nix, and S. Rangan, `` MIMO for Millimeter Wave Wireless Communications: Beamforming, Spatial
Multiplexing, or Both?,'' IEEE Communications Magazine, December 2014.

How will MIMO precoding work in mmWave 5G? 3

 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

mmWave Precoding is Different

 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Different hardware constraints

Analog RF ADC
processing Chain

Analog RF
processing ADC
Chain Joint Baseband
Baseband
processing Processing
Precoding

Phase Analog RF
shifters
ADC
processing Chain

Cost, power, and complexity limit the # of RF chains (high-resolution ADCs)

Precoding and combining can not be done entirely in the baseband
Analog beamforming usually uses a network of phase shifters
Additional constraints: Constant gain and quantized angles

Precoding and channel estimation algorithms should account for these constraints
5
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Different antenna scales

Large  antenna  arrays  at  

Tx  and  Rx

Base station Mobile Stations

Large antenna arrays result in

Large-dimensional precoding/combining matrices
High channel estimation, training, and feedback overheads

Need to design low-complexity precoding and channel estimation algorithms

6
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Different channel characteristics

microwave mmWave mmWave
Wifi or Cellular Wifi 5G (???)
bandwidth 1.4 MHz toMi 160 MHz 2.16 GHz 100 MHz to 2 GHz
# antennas @ BS or AP 1 to 8 16 to 32 64 to 256
# antennas at MS 1 or 2 16 to 32 4 to 32
delay spread 100 ns to 10 us 5 to 47 ns 12 to 40 ns
angle spread 1° to 60° 60° to 100° up to 50°
# clusters 4 to 9 <4 <4
orientation sensitivity low medium high
small-scale fading Rayleigh Nakagami non-fading or Nakagami
distant dependent + distant dependent + distant dependent +
large-scale fading
shadowing shadowing blockage
path loss exponent 2-4 2 LOS, 2.5 to 5 NLOS 2 LOS, 3.5 to 4.5 NLOS
penetration loss some varies possibly high
channel sparsity less more more
spatial correlation less more more

Some channel characteristics can be leveraged in the precoding design

7
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Different sensitivity to blockages

X
line-of-sight non-line-of-sight

blockage due to people

blockage due to buildings
User
Base station
Handset X
Blocked by users’ body

hand blocking self-body blocking

Need models for these forms of blockage
8
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Different communication channel bandwidth

mmWave noise bandwidth
Analog Baseband
processing processing

Receiver
UHF noise bandwidth How to implement equalization?

Large channel bandwidth (high noise power, low SNR before beamforming)
Implementing random access, channel training and estimation functions is challenging
Broadband channels coupled with delay spread
Equalization would likely be required at the receiver
Hardware constraints may make it difficult to perform equalization entirely in baseband

Need new algorithms and architectures for mmWave broadband communication

9
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

mmWave Suitable Precoding/Combining

 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Analog beamforming De-­‐facto  approach  in  

IEEE  802.11ad  /  WiGig    
and  Wireless  HD

Baseband RF RF Baseband
DAC
RFain ADC
RFain

Chain Chain

Phase shifters

Motivated by ADCs power consumption and implementation complexity

Suitable for single-stream trans. (complicated for multi-stream or multi-user)
Joint search for optimal beamforming/combining vectors with codebooks
* J.Wang, Z. Lan, C. Pyo, T. Baykas, C. Sum, M. Rahman, J. Gao, R. Funada, F. Kojima, H. Harada et al., “Beam codebook based beamforming protocol for multi-Gbps millimeterwave
WPAN systems,” IEEE Journal on Selected Areas in Communications, vol. 27, no. 8, pp. 1390–1399, 2009.
** S. Hur, T. Kim, D. Love, J. Krogmeier, T. Thomas, and A. Ghosh, “Millimeter wave beamforming for wireless backhaul and access in small cell networks,” IEEE Transactions on
Communications, vol. 61, no. 10, pp. 4391–4403, 2013.

11
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Hybrid analog-digital precoding/combining

tib-1 RF RF 1-bit
DAC RF + RF ADC
CDA Chain Chain ADC
Beamforming Combining

+
Baseband Baseband
dnabesaB Baseband
Precoding Combining
gnidocerP Precoding

RF RF
tib-1 RF Beamforming + Combining RF 1-bit
DAC ADC
CDA Chain Chain ADC

Makes compromise between hardware complexity and system performance

Hybrid precoding enables multi-stream and multi-user transmissions
* **

Digital can correct for analog limitations

Approaches the performance of unconstrained digital solutions
* O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath Jr., “Spatially sparse precoding in millimeter wave mimo systems,” IEEE Transactions on Wireless Communications, vol.
13, no. 3, pp. 1499–1513, March 2014.
** A. Alkhateeb, G. Leus, and R. W. Heath Jr, “Limited feedback hybrid precoding for multi-user millimeter wave systems,” arXiv:1409.5162, 2014. 12
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Design challenges: low-complexity precoding schemes

RF RF
Hybrid precoding design is non-trivial
chain chain
Coupled analog and digital precoding matrices
Baseband RF RF Baseband
RF phase shifters have constant modulus,
abesaB
Precoding Beamforming Combining Combining

quant. angles
RF RF
chain chain
Non-convex feasibility constraints

Sparse precoding solutions

Joint analog/digital precoder design w/ matching pursuit*
Approaches performance of unconstrained solutions mmWave channels are sparse in the angular
domain
Leverage lens antenna array structure** (only a few paths exist)
Extension to multiuser interference channels ***
* O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath Jr., “Spatially sparse precoding in millimeter wave mimo systems,” IEEE Transactions on Wireless Communications, vol.
13, no. 3, pp. 1499–1513, March 2014.
** J. Brady, N. Behdad, and A. Sayeed, “Beamspace MIMO for millimeter-wave communications: System architecture, modeling, analysis, and measurements,” IEEE Trans. on Ant. and
Propag., vol. 61, no. 7, pp. 3814–3827, July 2013.
*** M. Kim and Y.H. Lee, “MSE-based Hybrid RF/Baseband Processing for Millimeter Wave Communication Systems in MIMO Interference Channels”, IEEE TVT, to appear. 13
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Design challenges: channel estimation

14 with hybrid precoding
120
90 5
60 120
90 2
60 120
90 2
60 mmWave channel estimation is challenging
2.5 30 150 1 30 1 30

Large channel matrices -> training/feedback overhead

150 150

180 0 180 0 180 0

210 330 210 330 210 330 Low SNR before beamforming design
240 300 240 300 240 300

In hybrid architecture, channel is seen through RF BF lens

270 270 270
5 RF Chains 10 RF Chains 15 RF Chains
Beams generated using hybrid precoders
with different numbers of RF chains
mation of one of the beamforming vectors in the second codebook level with different numbers

3 paths Adaptive compressed sensing solution*

Sparse nature of mmWave channel can be leveraged
2 paths
= Cs (ABS,D AH BS,D ) 1A mmWave
BS,D G(s,k) . Further, given the available Channel estimation -> parameter estimation
system model
Lowbeamforming
g matrix F(s,k) is defined as F(s,k) = FRF,(s,k) FBS,(s,k) . As each training overhead with compressed sensing (CS) tools
used in a certain time instant, we will design each of them Adaptive CS estimation
independently in of multi-path mmWave channels
g/digitl precoders. Consequently, the CS and
design of the hybrid analog
1 path andhybrid
digitalprecoders lead to efficient training codebooks
s is accomplished by solving
* A. Alkhateeb, O. E. Ayach, G. Leus, and R. W. Heath Jr, “Channel estimation and hybrid precoding for millimeter wave cellular systems.” IEEE J. Selected Topics in Signal Processing (JSTSP), vol. 8, no. 5,
May 2014, pp. 831-846 14
⇥ ⇤ ⇥ ⇤
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Combining with 1-bit ADCs

Different transmit
architectures possible,
analog, hybrid, or
otherwise

Use 1-bit ADCs (pair) for each RF chain

Perform digital combining for all the highly quantized received signals
Ultra low power solution - only 1 comparator for each ADC, no need for AGC
Limitation: Capacity is bounded by 2 Nr bps/Hz (important at high SNR)
* J. Mo and R.W. Heath, Jr., “Capacity Analysis of One-Bit Quantized MIMO Systems with Transmitter Channel State Information” arxiv 1410.7353
See also extensive work by research groups led by U. Madhow, J. Nossek, G. Fettweis, G. Kramer, and O. Dabeer and others 15
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Design challenges: capacity analysis with 1-bit ADCs

Finding the exact capacity is challenging
Quantization is a nonlinear operation
Optimal input has discrete distribution
Special case: Rotated QPSK (optimal for SISO channel)*

Initial contributions **
MISO optimal strategy is MRT + QPSK signaling
Derived high SNR capacity for SIMO and MIMO
Use numerical methods to find optimal inputs***
Assumption: Known CSI at transmitter

*J. Singh, O. Dabeer, and U. Madhow, “On the limits of communication with low-precision analog-to-digital conversion at the receiver,” TCOM 2009
**J. Mo and R.W. Heath, Jr., “Capacity Analysis of One-Bit Quantized MIMO Systems with Transmitter Channel State Information” arxiv 1410.7353
16
***J. Huang and S. Meyn, “Characterization and computation of optimal distributions for channel coding,” TIT 2005
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Design challenges: channel estimation with 1-bit ADCs

Channel estimation is hard with 1-bit ADCs
Amplitude information is lost in the quantization
Conventional sparse reconstruction algorithms like LASSO
do not work with 1-bit quantization
Stochastic resonance appears when using GAMP: estimation
error may increase w/ SNR

mmWave with 2 paths, and 128-length Dimensionality reduction

training sequence

Possible approaches Φ ⊕ Q( )

Expectation-maximization algorithm*
Dithered quantization: Quantization threshold is adaptive** Adaptive threshold
Exploit sparse nature of mmWave channels with GAMP ***
*A. Mezghani, F. Antreich, and J. Nossek, "Multiple parameter estimation with quantized channel output," ITG 2010
**O. Dabeer and U. Madhow, “Channel estimation with low precision ADC”, ICC, 2010
***J. Mo, P. Schniter, N. G. Prelcic and R. W. Heath, Jr. “Channel Estimation in Millimeter Wave MIMO Systems with One-Bit Quantization”, Asilomar 2014 17
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Design challenges: broadband channels with 1-bit ADCs

Down Remove Estimate De- De-

converter ADC guard
S/P DFT P/S Equalize Mapper Interleaver
Decoder

conventional OFDM receiver

ADC
Down Remove Estimate De- De-
converter guard S/P DFT P/S Equalize Mapper Interleaver
Decoder

ADC

OFDM receiver with analog DFT

mmWave has broadband channels Analog DFT

10-40 ns delay spread in 2.16GHz BW in 11ad Orthogonalization: No inter-carrier
Equalization after quantization is challenging interference
Lower PAPR: Low-resolution ADCs
Possibly lower power vs digital DFT

*S. Suh, A. Basu, C. Schlottmann, P. Hasler, and J. Barry, “Low-power discrete Fourier transform for OFDM: A programmable analog approach,” IEEE Transactions on
18
Circuits and Systems I: Regular Papers, 58.2, 2011
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Future Research Directions

 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Future research directions (1/4)

Beams are assigned for each user, while multi-user W1 RF combiner
interference is handled in the baseband*
W2
FBB FRF
MS

W3
BS Limited Feedback

Multi-user mmWave systems with hybrid precoding

Enables different beams to be assigned to different users
Better interference management capability in digital domain
Initial work proposes two-stage hybrid precoding algorithm*
Considering out-of-cell interference is also interesting (extension to multi-layer precoding**)
* A. Alkhateeb, G. Leus, and R. W. Heath Jr., “Limited Feedback Hybrid Precoding for Multi-User Millimeter Wave Systems,” submitted to IEEE Trans. Wireless Commun., arXiv
preprint arXiv:1409.5162, 2014.
** Ahmed Alkhateeb, Geert Leus, and Rober W. Heath Jr, "Multi-Layer Precoding for Full-Dimensional Massive MIMO Systems ," in Proc. of Asilomar Conference on Signals, 20
Systems and Computers , Pacific Grove, CA, November 2014.
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Future research directions (2/4)

Compressed sensing tools leverage the sparse nature

of mmWave channels in the angular domain

Compressed sensing (CS) mmWave channel estimation

CS can leverage mmWave channel sparsity for efficient channel training/estimation
Designing CS-based pilot signals* for mmWave systems is an interesting open problem
Challenges are mainly due to the different hardware constraints (e.g., w/ hybrid precoding)

Body, hand and self-body blockages

Consider blockage model into the channel matrix
Precoding and channel estimation with array diversity** on the handset

* D. Ramasamy, S.Venkateswaran, and U. Madhow, “Compressive adaptation of large steerable arrays,” in Proc. of 2012 Information Theory and ApplicationsWorkshop (ITA),
CA, 2012, pp. 234–239.
** W. Roh et al., “Millimeter-Wave Beamforming as an Enabling Technology for 5G Cellular Communications: Theoretical Feasibility and Prototype Results”, IEEE 21
Communications Magazine, Feb. 2014
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Future research directions (3/4)

BB RF
MIMO with limited feedback
Feedback help establishing forward link Limited Feedback

Feedback has to be limited due to large channel dimensions and low rate during initial access
Need to design efficient precoding codebooks* (for hybrid architectures, 1-bit ADCs, …)
Channel sparsity may be leveraged for low-complexity solutions**
Initial hybrid beamforming codebooks based on adaptive refining***

MIMO over broadband channels RF

chain

Narrowband analog and broadband digital equalization

Narrowband Broadband
Exploiting channel sparsity, analog beams can be designed per cluster Beamforming Equalization

Adjusting analog / beam switching in OFDM, SC-FDMA**** RF

chain

* J. Singh, and R. Sudhir, "On the feasibility of beamforming in millimeter wave communication systems with multiple antenna arrays." arXiv preprint arXiv:1410.5509, 2014.
** A. Alkhateeb, G. Leus, and R. W. Heath Jr., “Limited Feedback Hybrid Precoding for Multi-User Millimeter Wave Systems,” submitted to IEEE TWC, arXiv preprint arXiv:1409.5162, 2014.
*** A. Alkhateeb, O. E. Ayach, G. Leus, and R. W. Heath Jr, “Channel estimation and hybrid precoding for millimeter wave cellular systems.” IEEE JSTSP, vol. 8, no. 5, May 2014, pp. 831-846
**** A. Ghosh et. al. “Millimeter-wave Enhanced Local Area Systems: A high-data-rate approach for future wireless networks”, IEEE JSAC vol. 32, no. 6, pp. 1152-1163, June 2014. 22
 -10 0.045 0.067 0.061 0.053 0.068 -10 0.067 0.056 0.069 0.069
-5 0.135 0.185 0.179 0.166 0.195 -5 0.193 0.177 0.197 0.198
0 0.369 0.424 0.455 0.440 0.479 0 0.482 0.471 0.494 0.500
3 0.602 0.610 0.693 0.687 0.736 3 0.759 0.744 0.777 WHAT
0.791 STARTS HERE CHANGES THE WORLD
5
7
0.769
0.903
0.733
0.843
0.889
1.098
0.869
1.064
0.931
1.133
5
7
0.975
1.215
0.955
1.180
1.002
1.248
1.029
1.294
(c) 2014 Robert W Heath Jr.
10 0.991 0.972 1.473 1.409 1.417 10 1.584 1.533 1.634 1.730
12 0.992 1.032 1.703 1.655 1.579 12 1.846 1.766 1.886 2.037

Future research directions (4/4)

15
17
20
1.000
1.000
1.000
1.091
1.115
1.136
1.930
1.987
1.999
1.921
1.987
1.999
1.765
1.853
1.938
15
17
20
2.253
2.508
2.837
2.138
2.423
2.808
2.232
2.427
2.655
2.514
2.838
3.329

4
Training signal design for systems with 1-bit ADCs
at most 𝐾+1 points. Of course, Theorem 1 is applicable to any
quantizer choice (and not just optimal symmetric quantizers).
3.5 Unquantized
Discrete input discrete output
Thus, it is possible that there might exist a 𝐾-bin quantizer
3−bit ADC for which the capacity is indeed achieved by exactly 𝐾 + 1
3 (optimal and PAM / ML)
Need not to estimate the exact channel state
points. We leave open, therefore, the question of whether or
Capacity (bits/channel use)

not the result in Theorem 1 can be tightened to guarantee

2.5 2−bit ADC
(optimal and PAM / ML) Estimate the channel response to certain training symbols
the achievability of capacity with at most 𝐾 points for the
AWGN-QO channel.
2

C. Comparison with Unquantized Observations

1−bit ADC
1.5 (optimal)
Performance analysis with >1-bit ADCs
We now compare the capacity results for different quantizer
precisions against the capacity with unquantized observations.
Again, the plots are shown in Figure 6 and the data values are
1
Tradeoff between achievable rate and power consumption
given in Table II. We observe that at low SNR, the performance
degradation due to low-precision quantization is small. For in-
0.5
Achievable rates of quant. MIMO channels are unknown**
stance, at -5 dB SNR, 1-bit receiver quantization achieves 68%
of the capacity achievable without any quantization, while
0
0 5 10 15 20 Uniform quantization is near-optimal**
with 2-bit quantization, we can get as much as 90% of the
SNR (dB) unquantized capacity. Even at moderately high SNRs, the loss
Fig. 6. Capacity plots for different ADC precisions. For 2 and 3-bit ADC, due to low-precision quantization remains quite acceptable.
Capacity
solid plotstofor
curves correspond different
optimal ADC
solutions, while precisions
dashed curves show the For example, 2-bit quantization achieves 85% of the capacity

Other channel state assumptions

performance of the benchmark scheme (PAM input with ML quantization).
in SISO channel (from *) attained using unquantized observations at 10 dB SNR, while
3-bit quantization achieves 85% of the unquantized capacity at
solutions (again verified by comparing with brute force 20 dB SNR. For the specific case of binary antipodal signaling,
Connections with non-coherent MIMO techniques
optimization results). [7] has earlier shown that a large fraction of the capacity can
While our results show that the iterative procedure (with be obtained by 2-bit quantization.
*J. Singh, O. Dabeer, and U. Madhow, “On the limits of communication On the other with
hand,low-precision analog-to-digital
if we fix the spectral efficiency conversion
to that at the receiver,” TCOM 2009
benchmark initialization) has provided (near) optimal solutions
**Q. Bai, J. A. Nossek, “Energy efficiency maximization for 5G multi-antenna
attained by an receivers”,
unquantized systemETTat 2014
10 dB (which is 1.73
at different SNRs, we leave the question of whether it will 23
converge to an optimal solution in general as an open problem. bits/channel use), then 2-bit quantization incurs a loss of 2.30
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Conclusions
mmWave precoding/combining is different than traditional UHF solutions
Different hardware constraints, antenna scales, channel characteristics, channel bandwidth

New transceiver architectures, precoding/combining solutions are needed

Promising solutions: Hybrid precoding/combining and combining with low-resolution ADCs
Design challenges with these solutions need to be addressed
Many research directions (multi-user extensions, new architectures, ….)

Submit your work to the forthcoming IEEE JSTSP special issue on Millimeter
Wave Communication - Manuscripts are due May15

Ahmed Alkhateeb, Jianhua Mo, Nuria González Prelcic and Robert W. Heath, Jr., ``MIMO Precoding and Combining
Solutions for Millimeter Wave Systems,'' IEEE Communications Magazine, December 2014.

http://www.profheath.org/research/millimeter-wave-cellular-systems/ 24
 WHAT STARTS HERE CHANGES THE WORLD
(c) 2014 Robert W Heath Jr.

Questions?
Robert W. Heath Jr.
The University of Texas at Austin

www.profheath.org 25