Design of mmWave massiveMIMO cellular systems

Abbas Kazerouni and Mainak ChowdhuryFaculty mentor: Andrea Goldsmith

Wireless Systems Lab, Stanford University

March 23, 2015

Future cellular networks

• Higher data rates

• More users

• Greater energy efficiency

Enablers for next generationcellular

• More spectrum(mmWave)

• Efficient frequencyreuse/coexistence(massive MIMO)

Future cellular networks

• Higher data rates

• More users

• Greater energy efficiency

Enablers for next generationcellular

• More spectrum(mmWave)

• Efficient frequencyreuse/coexistence(massive MIMO)

Properties of mmWave and massive MIMO

mmWave

• Massive amount of spectrum ⇒ Higher data rates

• Large attenuation from path loss and shadowing

• Stringent ADC requirements

• High speed baseband processor requirements

massive MIMO

• High directivity

• Large antenna array size

• Channel estimation challenges

mmWave massive MIMO in cellular

Benefits

• Directivity of massive MIMO compensates for high mmWaveattenuation, reduces multipath and multiuser interference

• mmWave frequencies reduce the size required for massiveMIMO antenna arrays

Some key challenges in mmWave massive MIMO

• Channel estimation

• ADC requirements and baseband complexity

Massive MIMO testbed, Lund Uni-versity, Globecom 2014

Hybrid beamforming architecture for mmwave, Samsung, Globecom2014

mmWave massive MIMO in cellular

Benefits

• Directivity of massive MIMO compensates for high mmWaveattenuation, reduces multipath and multiuser interference

• mmWave frequencies reduce the size required for massiveMIMO antenna arrays

Some key challenges in mmWave massive MIMO

• Channel estimation

• ADC requirements and baseband complexity

Massive MIMO testbed, Lund Uni-versity, Globecom 2014

Hybrid beamforming architecture for mmwave, Samsung, Globecom2014

Our proposed work addressesthese key challenges

Channel estimation

Pilot contamination

• Reusing the pilots in other cells results in interference

• In the infinite antenna regime, this is the only limiting factoron the capacity1

1Marzetta, Thomas L. ”Noncooperative cellular wireless with unlimited numbers of base station antennas.”

IEEE Trans. on Wireless Commun. 9.11 (2010): 3590-3600.

Pilot contamination in small cells

• Cell size reduction mitigates pilot contamination 2,SIR =(

1R2ρ

)α

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 20

10

20

30

40

50

60

User Density ρ (person/m2)

Wo

rst−

Ca

se

Ca

pa

city (

bits/s

ec/H

z)

α = 2

α = 4

α = 4

α = 8

• In the infinite antenna regime, cell size reduction increases theuser density and the capacity of each user

2A. Kazerouni, F. J. Lopez-Martinez, A. Goldsmith, “Increasing capacity in massive MIMO cellular networks

via small cells”, IEEE Globecom workshop, 2014

Pilot contamination for moderate array sizes

• With non-infinite antenna arrays, the effect of noise andintra-cell interference is not negligible

101 102 103 104 105 106

10−3

10−2

10−1

number of antennas

cros

sco

rrel

atio

nb

etw

een

colu

mn

sof

H

y = Hx + ν,H ∈ Rn×4

• How do these additional factors affect the capacity ?• How many antennas do we need to have a certain

performance?

Pilot contamination for moderate array sizes

• With non-infinite antenna arrays, the effect of noise andintra-cell interference is not negligible

101 102 103 104 105 106

10−3

10−2

10−1

number of antennas

cros

sco

rrel

atio

nb

etw

een

colu

mn

sof

H

y = Hx + ν,H ∈ Rn×4

• How do these additional factors affect the capacity ?• How many antennas do we need to have a certain

performance?

Do we need channel estimation at all?

Short answerNot always, and not estimating the channel may also address thesecond key challenge about baseband processing complexity

Noncoherent architecture for a multiuser SIMO system

• Coherent phase reference may not be necessary 3

Tx Rx to energy based baseband processor

‖·‖2n

• Analog or hybrid analog/digital architectures reduce demandson phase recovery circuits, ADCs, etc. 4 ⇒ Energy efficiency

3M. Chowdhury, A. Manolakos, A. Goldsmith, “Coherent versus noncoherent massive SIMO systems: which

has better performance?”, IEEE ICC 20154M. Chowdhury, A. Manolakos, A. Goldsmith, “Noncoherent energy-based communications for the massive

SIMO MAC”, under revision, IEEE Trans. Info. Theory, 2015

Do we need channel estimation at all?

Short answerNot always, and not estimating the channel may also address thesecond key challenge about baseband processing complexity

Noncoherent architecture for a multiuser SIMO system

• Coherent phase reference may not be necessary 3

Tx Rx to energy based baseband processor

‖·‖2n

• Analog or hybrid analog/digital architectures reduce demandson phase recovery circuits, ADCs, etc. 4

⇒ Energy efficiency

3M. Chowdhury, A. Manolakos, A. Goldsmith, “Coherent versus noncoherent massive SIMO systems: which

has better performance?”, IEEE ICC 20154M. Chowdhury, A. Manolakos, A. Goldsmith, “Noncoherent energy-based communications for the massive

SIMO MAC”, under revision, IEEE Trans. Info. Theory, 2015

Do we need channel estimation at all?

Short answerNot always, and not estimating the channel may also address thesecond key challenge about baseband processing complexity

Noncoherent architecture for a multiuser SIMO system

• Coherent phase reference may not be necessary 3

Tx Rx to energy based baseband processor

‖·‖2n

• Analog or hybrid analog/digital architectures reduce demandson phase recovery circuits, ADCs, etc. 4 ⇒ Energy efficiency

3M. Chowdhury, A. Manolakos, A. Goldsmith, “Coherent versus noncoherent massive SIMO systems: which

has better performance?”, IEEE ICC 20154M. Chowdhury, A. Manolakos, A. Goldsmith, “Noncoherent energy-based communications for the massive

SIMO MAC”, under revision, IEEE Trans. Info. Theory, 2015

Noncoherent architectures for different channel models

MISO, MIMO channels

• Can massive antenna arrays at transmitter be exploitedwithout channel state information (CSI)?

Wideband

• Number of channel coefficients to be estimated is large

• Multi-carrier energy based system achieves same capacityscaling behavior as coherent systems 5

• How do single carrier implementations compare with amulti-carrier implementation?

5M. Chowdhury, A. Manolakos, F. Gomez-Cuba, E. Erkip, A. Goldsmith, “ Capacity scaling in noncoherent

wideband massive SIMO systems”, IEEE ITW, Jerusalem, 2015

Validation of theory on empirical channel models

Empirical channel models for mmWave

• mmWave channel models still being developed

• Gains depend on models used

• Correlation in Rx antenna reduces capacity due to loss indiversity

101 101.5 102 102.5

4

6

8

number of antennas

un

corr

elat

eder

god

icca

pac

ity

101 101.5 102 102.5

4

6

8

number of antennas

corr

elat

eder

god

icca

pac

ity

• How do results change for real life mmWave channels?

• How does correlation affect general MIMO transmission?

Validation of theory on empirical channel models

Empirical channel models for mmWave

• mmWave channel models still being developed

• Gains depend on models used

• Correlation in Rx antenna reduces capacity due to loss indiversity

101 101.5 102 102.5

4

6

8

number of antennas

un

corr

elat

eder

god

icca

pac

ity

101 101.5 102 102.5

4

6

8

number of antennas

corr

elat

eder

god

icca

pac

ity

• How do results change for real life mmWave channels?

• How does correlation affect general MIMO transmission?

Thanks for your attention

Questions ?

Thanks for your attention

Questions ?