Lecture 6

Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH

Physical Modeling Lecture 6: Modeling of MIMO Channels

Modeling of MIMO Theoretical Foundations of Wireless Communications1
Fading Channels

Lars Kildehøj
CommTh/EES/KTH

Thursday, May 3, 2018

9:30-12:00, Conference Room SIP

1
Textbook: D. Tse and P. Viswanath, Fundamentals of Wireless Communication
1 / 23
 Overview
Lecture 5: Spatial Diversity, MIMO Capacity
Lecture 6 • SIMO, MISO, MIMO
Modeling of MIMO
Channels • Degrees of freedom
Lars Kildehøj
CommTh/EES/KTH • MIMO capacity

Physical Modeling
Lecture 6: MIMO Channel Modeling
Modeling of MIMO
Fading Channels
1 Physical Modeling
Line-of-Sight Channels
Geographically Separated Antennas
LOS Plus One Reflected Path
Summary
2 Modeling of MIMO Fading Channels
General Concept
Angular Domain Representation
ADR of MIMO Channels
Statistical Modeling in the Angular Domain
Degrees of Freedom and Diversity
Antenna Spacing

2 / 23
 Overview

Lecture 6
Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH Motivation:
• How does the multiplexing capability of MIMO channels depend on
Physical Modeling

Modeling of MIMO
the physical environment?
Fading Channels
• When can we gain (much) from MIMO?
• How do we have to design the system?

3 / 23
 Physical Modeling
– Line-of-Sight Channels: SIMO
• Free space without scattering
and reflections.
Lecture 6
Modeling of MIMO • Antenna separation ∆r λc , with
Channels

Lars Kildehøj
carrier wavelength λc and the
CommTh/EES/KTH normalized antenna separation
Physical Modeling
∆r ; nr receive antennas.
Line-of-Sight (D. Tse and P. Viswanath, Fundamentals of Wireless Communi-
Channels • Distance between transmitter
cations.)
Geographically
Separated Antennas and i-th receive antenna: di
LOS Plus One
Reflected Path
Summary
• Continuous-time impulse between transmitter and i-th receive
Modeling of MIMO antenna:
Fading Channels
hi (τ ) = a · δ(τ − di /c)
• Base-band model (assuming di /c  1/W , signal BW W ):
   
2πfc di 2πdi
hi = a · exp −j = a · exp −j
c λc
• SIMO model: y = h · x + w, with w ∼ CN (0, N0 I)
→ h: signal direction, spatial signature.

4 / 23
 Physical Modeling
– Line-of-Sight Channels: SIMO
• Paths are approx. parallel, i.e.,
Lecture 6
Modeling of MIMO di ≈ d + (i − 1)∆r λc cos(φ)
Channels

Lars Kildehøj
CommTh/EES/KTH
• Directional cosine

Physical Modeling Ω = cos(φ) (D. Tse and P. Viswanath, Fundamentals of Wireless Communi-
Line-of-Sight cations.)
Channels
Geographically • Spatial signature can be expressed as
Separated Antennas
LOS Plus One
 
Reflected Path 1
Summary
  exp(−j2π∆r Ω) 
Modeling of MIMO 2πd 
 
Fading Channels h = a · exp −j exp(−j2π2∆r Ω) 
λc
 
 .. 
 . 
exp(−j2π(nr − 1)∆r Ω)
→ Phased-array antenna.
• SIMO capacity (with MRC)

Pkhk2 Pa2 nr
   
C = log 1 + = log 1 +
N0 N0
→ Only power gain, no degree-of-freedom gain.
5 / 23
 Physical Modeling
– Line-of-Sight Channels: MISO
• Similar to the SIMO case:
Lecture 6
∆t , λc , di , φ, Ω,...
Modeling of MIMO
Channels • MISO channel model:
Lars Kildehøj
CommTh/EES/KTH y = h∗ x + w ,
Physical Modeling
Line-of-Sight
with w ∼ CN (0, N0 ). (D. Tse and P. Viswanath, Fundamentals of Wireless Communi-

Channels cations.)
Geographically
Separated Antennas • Channel vector
LOS Plus One
Reflected Path  
Summary 1
Modeling of MIMO
Fading Channels   exp(−j2π∆t Ω) 
2πd
 
h = a · exp j
 exp(−j2π2∆t Ω) 
λc
 
 .. 
 . 
exp(−j2π(nt − 1)∆t Ω)

• Unit spatial signature in the directional cosine Ω:

√
e(Ω) = 1/ n · [1, exp(−j2π∆Ω), . . . , exp(−j2π(n − 1)∆Ω)]T

→ et (Ωt ) and er (Ωr ) with nt , ∆t and nr , ∆r , respectively.

6 / 23
 Physical Modeling
– Line-of-Sight Channels: MIMO

• Linear transmit and receive array with nt , ∆t and nr , ∆r .

Lecture 6
Modeling of MIMO • Gain between transmit antenna k and receive antenna i
Channels

Lars Kildehøj
CommTh/EES/KTH hik = a · exp (−j2πdik /λc )

Physical Modeling • Distance between transmit antenna k and receive antenna i

Line-of-Sight
Channels
Geographically dik = d + (i − 1)∆r λc cos(φr ) − (k − 1)∆t λc cos(φt )
Separated Antennas
LOS Plus One
Reflected Path • MIMO channel matrix (with Ωt = cos(φt ) and Ωr = cos(φr ))
Summary
Modeling of MIMO
√
 
2πd
Fading Channels
H = a nt nr exp −j er (Ωr )et (Ωt )∗
λc
√
→ H is a rank-1 matrix with singular value λ1 = a nt nr
→ Compare with SVD decomposition in Lecture 5:
k
X
H= λi ui vi∗
i=1

7 / 23
 Physical Modeling
– Line-of-Sight Channels: MIMO

Lecture 6
Modeling of MIMO
Channels • MIMO capacity
Pa2 nr nt
Lars Kildehøj
 
CommTh/EES/KTH
C = log 1 +
N0
Physical Modeling
Line-of-Sight
Channels
→ Only power gain, no degree-of-freedom gain.
Geographically
Separated Antennas
• nt = 1: power gain equals nr → receive beamforming.
LOS Plus One
Reflected Path • nr = 1: power gain equals nt → transmit beamforming.
Summary
• General nt , nr : power gain equals nr · nt
Modeling of MIMO
Fading Channels
→ Transmit and receive beamforming.
• Conclusion: In LOS environment, MIMO provides only a power gain
but no degree-of-freedom gain.

8 / 23
 Physical Modeling
– Geographically Separated Antennas at the Transmitter

Example/special case
Lecture 6
Modeling of MIMO
• 2 distributed transmit antennas,
Channels
attenuations a1 , a2 , angles of incidence
Lars Kildehøj
CommTh/EES/KTH φr 1 , φr 2 , negligible delay spread.
Physical Modeling
• Spatial signature (nr receive antennas)
Line-of-Sight
√
Channels
 
2πd1k
Geographically
Separated Antennas
hk = ak nr exp −j er (Ωrk )
LOS Plus One (D. Tse and P. Viswanath, Fundamentals of
λc
Reflected Path
Summary Wireless Communications.)
• Channel matrix H = [h1 , h2 ]
Modeling of MIMO
Fading Channels

• H has independent columns as long as (Ωr i = cos(φr i ))

1
Ωr = Ωr 2 − Ωr 1 6= 0 mod
∆r
→ Two non-zero singular values λ21 , λ22 ; i.e., two degrees of freedom.
→ But H can still be ill-conditioned!

9 / 23
 Physical Modeling
– Geographically Separated Antennas at the Transmitter

• Conditioning of H is determined by how the spatial signatures are

Lecture 6 aligned (with Lr = nr ∆r ):
Modeling of MIMO
Channels
sin(πLr Ωr )
Lars Kildehøj
CommTh/EES/KTH | cos(θ)| = |fr (Ωr 2 − Ωr 1 )| = | er (Ωr 1 )∗ er (Ωr 2 ) | =
| {z } nr sin(πLr Ωr /nr )
Physical Modeling =fr (Ωr 2 −Ωr 1 )
Line-of-Sight
Channels
Geographically
• Example (a1 = a2 = a)
Separated Antennas s
LOS Plus One
Reflected Path λ21 = a2 nr (1 + | cos(θ)|) λ1 1 + | cos(θ)|
⇒ =
Summary λ22 = a2 nr (1 − | cos(θ)|) λ2 1 − | cos(θ)|
Modeling of MIMO
Fading Channels
• fr (Ωr ) is periodic with nr /Lr .
• Maximum at Ωr = 0; fr (0) = 1.
• fr (Ωr ) = 0 at Ωr = k/Lr with
k = 1, . . . , nr − 1.
• Resolvability 1/Lr ,
if Ωr  1/Lr , then the signals
(D. Tse and P. Viswanath, Fundamentals of Wireless Commu- from the two antennas cannot
nications.) be resolved.
10 / 23
 Physical Modeling
– Geographically Separated Antennas at the Transmitter

Beamforming pattern
Lecture 6
Modeling of MIMO
Channels • Assumption: signal arrives with
Lars Kildehøj angle φ0 ; receive beamforming
CommTh/EES/KTH
vector er (cos(φ0 )).
Physical Modeling
• A signal form any other direction φ
Line-of-Sight
Channels will be attenuated by a factor
Geographically
Separated Antennas
LOS Plus One
Reflected Path |er (cos(φ0 ))∗ er (cos(φ))|
Summary
Modeling of MIMO = |fr (cos(φ) − cos(φ0 ))|
Fading Channels

• Beamforming pattern
(D. Tse and P. Viswanath, Fundamentals of Wireless

Communications.)
( φ, |fr (cos(φ) − cos(φ0 ))| )

• Main lobes around φ0 and any angle φ for which cos(φ) = cos(φ0 ).
→ In a similar way, separated receive antennas can be treated.

11 / 23
 Physical Modeling
– LOS Plus One Reflected Path

Lecture 6
• Direct path:
Modeling of MIMO
Channels
φt1 , Ωr 1 , d (1) , and a1 .
Lars Kildehøj
CommTh/EES/KTH • Reflected path:
Physical Modeling
φt2 , Ωr 2 , d (2) , and a2 .
Line-of-Sight
Channels
Geographically (D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)
Separated Antennas
LOS Plus One • Channel model follows from signal superposition
Reflected Path
Summary
Modeling of MIMO
H = a1b er (Ωr 1 )et (Ωt1 )∗ + a2b er (Ωr 2 )et (Ωt2 )∗ ,
Fading Channels
with
√ 2πd (i)
 
aib = ai nt nr exp −j .
λc
→ H has rank 2 as long as
1 1
Ωt1 6= Ωt2 mod and Ωr 1 6= Ωr 2 mod .
∆t ∆r
→ H is well conditioned if the angular separations |Ωt |, |Ωr | at the
transmit/receive array are of the same order or larger than 1/Lt,r .
12 / 23
 Physical Modeling
– LOS Plus One Reflected Path

Lecture 6
• Direct path:
Modeling of MIMO
Channels
φt1 , Ωr 1 , d (1) , and a1 .
Lars Kildehøj
CommTh/EES/KTH • Reflected path:
Physical Modeling
φt2 , Ωr 2 , d (2) , and a2 .
Line-of-Sight
Channels
Geographically (D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)
Separated Antennas
LOS Plus One
Reflected Path
• H can be rewritten as H = H00 H0 , with
Summary
et ∗ (Ωt1 )
 
Modeling of MIMO
H00 = [a1b er (Ωr 1 ), a2b er (Ωr 2 )] and H0 =
Fading Channels
et ∗ (Ωt2 )

→ Two imaginary receivers at points A and B (virtual relays).

• Since the points A and B are geographically widely separated, H0
and H00 have rank 2 and hence H has rank 2 as well.
• Furthermore, if H0 and H00 are well-conditioned, H will be
well-conditioned as well.
→ Multipath fading can be viewed as an advantage which can be
exploited!
13 / 23
 Physical Modeling
– LOS Plus One Reflected Path

Lecture 6
Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH

Physical Modeling
Line-of-Sight
Channels
Geographically
Separated Antennas
LOS Plus One
Reflected Path
Summary
Modeling of MIMO
Fading Channels (D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

• Significant angular separation is required at both the transmitter

and the receiver to obtain a well-conditioned matrix H.
• If the reflectors are close to the receiver (downlink), we have a small
angular separation ⇒ not very well-conditioned matrix H.
• Similar, if the reflectors are close to the transmitter (uplink).
→ Size of an antenna array at a base station will have to be many
wavelengths to be able to exploit the spatial multiplexing effect.
14 / 23
 Physical Modeling
– Summary

Lecture 6
Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH

Physical Modeling
Line-of-Sight
Channels
Geographically
Separated Antennas
LOS Plus One
Reflected Path
Summary
Modeling of MIMO
Fading Channels

(D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

15 / 23
 Modeling of MIMO Fading Channels
– General Concept

• Antenna lengths Lt , Lr limit the resolvability2 of the transmit and

Lecture 6 receive antenna in the angular domain.
Modeling of MIMO
Channels
→ Sample the angular domain at fixed angular spacings of 1/Lt at the
Lars Kildehøj
CommTh/EES/KTH transmitter and 1/Lr at the receiver.
Physical Modeling
→ Represent the channel (the multiple paths) in terms of these input
Modeling of MIMO
and output coordinates.
Fading Channels
General Concept
Angular Domain
Representation
ADR of MIMO
Channels
Statistical Modeling • The (k, l)-th channel gain
in the Angular
Domain follows as the aggregation of all
Degrees of Freedom
and Diversity paths whose transmit and
Antenna Spacing
receive directional cosines lie in
a (1/Lt × 1/Lr ) bin around the
point (l/Lt , k/Lr ).

(D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

2
Note, if Ωr ,t  1/Lr ,t , the paths cannot be separated.
16 / 23
 Modeling of MIMO Fading Channels
– Angular Domain Representation (ADR)

Lecture 6
Modeling of MIMO
Channels • Orthonormal basis for the received signal space (nr basis vectors)
Lars Kildehøj  
CommTh/EES/KTH 1 nr − 1
Sr = er (0), er ( ), . . . , er ( )
Physical Modeling Lr Lr
Modeling of MIMO
Fading Channels → Orthogonality follows directly from the properties of fr (Ω).
General Concept
Angular Domain • Orthonormal basis for the transmitted signal space (nt basis vectors)
Representation
ADR of MIMO  
Channels 1 nt − 1
Statistical Modeling St = et (0), et ( ), . . . , et ( )
in the Angular
Domain
Lt Lt
Degrees of Freedom
and Diversity
Antenna Spacing
→ Orthogonality follows directly from the properties of ft (Ω).
• Orthonormal bases provide a very simple (but approximate)
decomposition of the total received/transmitted signal up to a
resolution 1/Lr , 1/Lt .

17 / 23
 Modeling of MIMO Fading Channels
– Angular Domain Representation

Examples: Receive beamform patterns of the angular basis vectors in Sr

Lecture 6
Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH
• (a) Critically spaced (∆r = 1/2),
Physical Modeling each basis vector has a single pair
Modeling of MIMO of main lobes.
Fading Channels
General Concept
Angular Domain
Representation • (b) Sparsely spaced (∆r > 1/2),
ADR of MIMO
Channels some of the basis vectors have
Statistical Modeling
in the Angular more than one pair of main lobes.
Domain
Degrees of Freedom
and Diversity
Antenna Spacing
• (c) Densely spaced (∆r < 1/2),
some of the basis vectors have no
pair of main lobes.

(D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

18 / 23
 Modeling of MIMO Fading Channels
– ADR of MIMO Channels
(Assumption: critically spaced antennas)
Lecture 6
Modeling of MIMO
• Observation: The vectors in St and Sr form unitary matrices Ut and
Channels Ur with dimensions (nt × nt ) and (nr × nr ), respectively. (→ IDFT
Lars Kildehøj
CommTh/EES/KTH
matrices!)
• With3 xa = U∗t x and ya = U∗r y we get
Physical Modeling

Modeling of MIMO
Fading Channels
ya = U∗r HUt xa + U∗r w = Ha xa + wa , with wa ∼ CN (0, N0 Inr ).
General Concept P b ∗
Angular Domain
Representation
• Furthermore, with H = i ai er (Ωri )et (Ωti ) , we get
ADR of MIMO
Channels
a
hkl = er (k/Lr )∗ H et (l/Lt )
Statistical Modeling X
in the Angular
Domain
= aib [er (k/Lr )∗ er (Ωri )] · [et (Ωti )∗ et (l/Lt )]
i
Degrees of Freedom
| {z } | {z }
and Diversity (1) (2)
Antenna Spacing

• The terms (1) and (2) are significant for the i − th path if
k 1 k 1
Ωri − < and Ωti − < .
Lr Lr Lt Lt
(→ Projections on the basis vectors in Sr , St .)
3
The superscript “a” denotes angular domain quantities.
19 / 23
 Modeling of MIMO Fading Channels
– Statistical Modeling in the Angular Domain

• Let Tl and Rk be the sets of physical paths which have most energy
Lecture 6
Modeling of MIMO
in directions of et (l/Lt ) and er (k/Lr ).
Channels
a
Lars Kildehøj • hkl corresponds to the aggregated gains aib of paths which lie in
CommTh/EES/KTH
Rk ∩ Tl .
Physical Modeling
• Independence and time variation
Modeling of MIMO
Fading Channels • Gains of the physical paths aib [m] are independent.
General Concept a [m] are independent across m.
Angular Domain
⇒ The path gains hkl
Representation
ADR of MIMO • The angles {φri [m]}m and {φti [m]}m evolve slower than aib [m].
Channels
Statistical Modeling ⇒ The physical paths do not move from one angular bin to another.
in the Angular
Domain ⇒ The path gains hkl [m] are independent across k and l.
Degrees of Freedom
and Diversity
Antenna Spacing • If there are many paths in an angular bin ⇒ Central Limit Theorem
⇒ hkla [m] can be modeled as complex circular symmetric Gaussian.
a
• If there are no paths in an angular bin ⇒ hkl [m] ≈ 0.

• Since Ut and Ur are unitary matrices, the matrix H has the same
i.i.d. Gaussian distribution as Ha .

20 / 23
 Modeling of MIMO Fading Channels
– Statistical Modeling in the Angular Domain
Example for Ha

Lecture 6
Modeling of MIMO
Channels

Lars Kildehøj
CommTh/EES/KTH

Physical Modeling

Modeling of MIMO
Fading Channels
General Concept
Angular Domain
Representation
ADR of MIMO
Channels
Statistical Modeling
in the Angular
Domain
Degrees of Freedom
and Diversity
Antenna Spacing

(D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

• (a) Small angular spread at the transmitter.

• (b) Small angular spread at the receiver.
• (c) Small angular spread at both transmitter and receiver.
• (d) Full angular spread at both transmitter and receiver.
21 / 23
 Modeling of MIMO Fading Channels
– Degrees of Freedom and Diversity
Degrees of freedom
Lecture 6 • Based on the derived statistical model, we get the following result:
Modeling of MIMO
Channels with probability 1, the rank of the random matrix Ha is given by
Lars Kildehøj
CommTh/EES/KTH rank(Ha ) = min{ number of non-zero rows,
Physical Modeling number of non-zero columns }.
Modeling of MIMO
Fading Channels • The number of non-zero rows and columns depends on two factors:
General Concept
• Amount of scattering and reflection; the more scattering and
Angular Domain
Representation reflection, the larger the number of non-zero entries in Ha .
ADR of MIMO
Channels • Lengths Lr and Lt ; for small Lr , Lt many physical paths are mapped
Statistical Modeling
in the Angular into the same angular bin; with higher resolution, more paths can be
Domain
Degrees of Freedom
represented.
and Diversity
Antenna Spacing Diversity: The diversity is given by the number of non-zero entries in Ha .
Example: Same number of degrees of freedom but different diversity.

(D. Tse and P. Viswanath, Fundamentals of Wireless Communications.)

22 / 23
 Modeling of MIMO Fading Channels
– Antenna Spacing
So far: critically spaced antennas with ∆r = 1/2.
• One-to-one correspondence between the angular windows and the
Lecture 6
Modeling of MIMO resolvable bins.
Channels

Lars Kildehøj
Setup 1: vary the number of antennas for a fixed array length Lr ,t .
CommTh/EES/KTH • Sparsely spaced case (∆r > 0.5)
Physical Modeling
• Beamforming patterns of some basis vectors have multiple main
lobes.
Modeling of MIMO
Fading Channels • Different paths with different directions are mapped onto the same
General Concept basis vector.
Angular Domain
Representation → Resolution of the antenna array, number of degrees of freedom, and
ADR of MIMO
Channels diversity are reduced.
Statistical Modeling
in the Angular • Densely spaced case (∆r < 0.5)
Domain
Degrees of Freedom • There are basis vectors with no main lobes which do not contribute
and Diversity
Antenna Spacing
to the resolvability.
• Adds zero rows and columns to Ha and creates correlation in H.
Setup 2: vary the antenna separation for a fixed number of antennas.
• Rich scattering: number of non-zero rows in Ha is already nr ; i.e.,
no improvement possible.
• Clustered scattering: scattered signal can be received in more bins;
i.e., increasing number of degrees of freedom.
23 / 23