Modelling and Simulation of MIMO Channels

Modelling and Simulation of MIMO Channels
Matthias Patzold
Agder University College
Mobile Communications Group
Grooseveien 36, NO-4876 Grimstad, Norway
E-mail: matthias.paetzold@hia.no
Homepage: http://ikt.hia.no/mobilecommunications/
BEATS/CUBAN Workshop 2004
1/37
Contents
I. Introduction
II. Principles of Fading Channel Modelling
III. Modelling of MIMO Channels
IV. Simulation of MIMO Channels
V. Conclusion
2/37
I. Introduction
Mobile Communications Using Smart Antennas
Transmitter
Channel
) x , t , ( h J
Receiver
Mobile station Base station
1
2
M
R
01 1 01 0 01 1 01 0
Space-time signal processing
Demodulation
Decoding
Space-time coding
Modulation
1
2
M
T
Realistic spatio-temporal channel models are required for the design, test, and optimization
of mobile communication systems employing smart antenna technologies.
3/37
Meaning of Channel Modelling and Simulation for Mobile Communications
Reference system:
+
Channel
model
Noise
Data source Data sink
Transmitter
Receiver
d(i) x(t) y(t)

d(i)
Performance measure: Bit error probability P
b
= f
_
E
b
N
0
_
Simulation system:
+
Channel
Noise
Data source Data sink
Transmitter
Receiver
simulator
d(i) x(t) y(t)

d(i)
Performance measure: Bit error probability

P
b
= f
_
E
b
N
0
,
_
P
b
Model error
The model error describes the essential error which is introduced by the channel simulator.
Channel simulators are important for the test, the parameter optimization, and the perfor-
mance analysis of mobile communication systems.
4/37
The Two Main Categories of Fading Channel Simulators
Simulators based on reference models: Simulators based on measurements:
Simulation model
Reference model
Measurement
Real-world
Snapshot
measurements
campaigns
channel
channel
Real-world
Simulation model
Problem: Reference models are not
close to real-world channels.
Problem: Finding of typical scenarios.
The dilemma of mobile fading channel modelling.
5/37
Some Requirements for Fading Channel Simulators
High precision w.r.t. a given reference model or w.r.t. measured channels
High eciency
The underlying stochastic channel simulator must be ergodic to reduce the number of trials
Easy to determine the model parameters
Easy to understand and easy to implement
Reproducibility for enabling fair performance comparisons
Easy to extend (frequency selectivity, spatial selectivity, multi-cluster propagation scenarios, number
of antenna elements, etc.)
Low complexity
6/37
II. Principles of Fading Channel Modelling
Two Fundamental Methods for Modelling of Coloured Gaussian Noise Processes
1. Filter method:
Stochastic process (t)
Reference model
2. Sum-of-sinusoids (SOS) method:
cos(2f
1
t +
1
)
cos(2f
2
t +
2
)
cos(2f
N
t +
N
)
c
1
c
2
c
N
.
.
.
.
.
.
.
.
.
.
.
.

e
-
-
e
-
-
e
-
-
+
-
(t)
Reference model
7/37
For a nite number of harmonic functions N, we obtain:
Parameters: ensemble
n
= Random variable (RV)
f
n
= const.
c
n
= const.
cos(2f
1
t +
1
)
cos(2f
2
t +
2
)
cos(2f
N
t +
N
)
c
1
c
2
c
N
.
.
.
.
.
.
e
-
-
e
-
-
e
-
-
+
-
(t)
Stochastic simulation model
Parameters: single realization
n
= const.
f
n
= const.
c
n
= const.
cos(2f
1
t +
1
)
cos(2f
2
t +
2
)
cos(2f
N
t +
N
)
c
1
c
2
c
N
.
.
.
.
.
.
e
-
-
e
-
-
e
-
-
+
-
(t)
Deterministic process (t)
Deterministic simulation model
8/37
Relationships Between Stochastic Processes, Random Variables,
Sample Functions, and Real-Valued Numbers
Gaussian process: (t) = lim
N
N
n=1
c
n
cos(2f
n
t +
n
) (c
n
= const., f
n
= const.,
n
= RV)
Stochastic process: (t) =
N
n=1
c
n
cos(2f
n
t +
n
) (c
n
= const., f
n
= const.,
n
= RV)
simulation
model
Deterministic
Stochastic
simulation
model
model
Reference
Gaussian process
Stochastic process
Random
variable

Sample
function
Real number
t = t
0
n
= const. t = t
0
(t
0
) = (t
0
,
n
)
(t
0
) = (t
0
)
(t) = (t,
n
)
N < N
(t) = (t,
n
)
(t) = (t,
n
)
n
= const.
n
= RV
9/37
Principle of Deterministic Channel Modelling
Step 1: Starting point is a reference model derived from one, two or several Gaussian
processes.
Step 2: Derive a stochastic simulation model from the reference model by replacing each
Gaussian process by a sum-of-sinusoids with xed gains, xed frequencies, and
random phases.
Step 3: Determine the deterministic simulation model by xing all model parameters of the
stochastic simulation model, including the phases.
Step 4: Compute the model parameters of the simulation model by using a proper param-
eter computation method.
Step 5: Simulate one (or some few) sample functions (deterministic processes).
10/37
Example: Derivation of a channel simulator for Rice processes
Step 1: Reference model Steps 2 & 3: Simulation model
(t)
(t)
2
1
(t)
f t+ m (t)=

)
2
sin(2
cos(2f t+ m (t)=
1
)
H (f)
H (f)
(t)
2
v
(t)
1
v
(t)
(t)
(t)
WGN
WGN
2
1
Stochastic Rice process (t) Deterministic Rice process

(t)
Step 4: Compute the model parameters by tting the statistics of the deterministic simulation
model to those of the stochastic reference model.
Step 5: Simulate the deterministic Rice process

(t).
11/37
Methods for the Calculation of the Model Parameters
Deterministic process:
i
(t) =
N
i
n=1
c
i,n
cos(2f
i,n
t +
i,n
)

Model parameters: gains frequencies phases
Historical overview
Methods Disadvantages
Rice method (Rice, 1944) Small period
Jakes method (Jakes, 1974) Special case only
Monte Carlo method (Schulze, 1988) Non-ergodic
Method of equal distances (Patzold, 1994) Small period
Method of equal areas (Patzold, 1994) Slow convergency
Harmonic decomposition technique (Crespo, 1995) Small period
Mean-square-error method (Patzold, 1996) Small period
Method of exact Doppler spread (Patzold, 1996) Special case only
L
p
-norm method (Patzold, 1996)
Method porposed by Zheng and Xiao (2002) Non-ergodic
12/37
Classes of SOS Channel Simulators and Their Statistical Properties
Deterministic process:
i
(t) =
N
i
n=1
c
i,n
cos(2f
i,n
t +
i,n
)
Class Gains Frequencies Phases First-order Wide-sense Mean- Autocor.-
c
i,n
f
i,n

i,n
stationary stationary ergodic ergodic
I const. const. const.
II const. const. RV yes yes yes yes
III const. RV const. no/yes
a
no/yes
a
no/ yes
a
no
IV const. RV RV yes yes yes no
V RV const. const. no no no/yes
a
no
VI RV const. RV yes yes yes no
VII RV RV const. no/yes
a
no/yes
a
no/yes
a
no
VIII RV RV RV yes yes yes no
a
If certain boundary conditions are fullled.
13/37
Application of the Scheme to Parameter Computation Methods
Stochastic process:
i
(t) =
N
i
n=1
c
i,n
cos(2f
i,n
t +
i,n
)

Model parameters: gains frequencies phases (RVs)
Parameter computation methods Class FOS WSS
Mean-
ergodic
Autocor.-
ergodic
Rice method II yes yes yes yes
Monte Carlo method IV yes yes yes no
Jakes method II yes yes yes yes
Harmonic decomposition technique II yes yes yes yes
Method of equal distances II yes yes yes yes
Method of equal areas II yes yes yes yes
Mean-square-error method II yes yes yes yes
Method of exact Doppler spread II yes yes yes yes
L
p
-norm method II yes yes yes yes
Method proposed by Zheng and Xiao IV yes yes yes no
14/37
Applications of the Principle of Deterministic Channel Modelling
Frequency-nonselective channels (Rayleigh, Rice, ext. Suzuki, Nakagami, etc.),
Frequency-selective channels (COST 207, CODIT),
Space-time wideband channels (COST 259),
Multiple cross-correlated Rayleigh fading channels,
Multiple uncorrelated Rayleigh fading channels,
Perfect modelling and simulation of measured 2-D and 3-D channels,
Frequency-hopping fading channels,
Design of ultra fast channel simulators (tables system),
Design of burst error models,
Development of MIMO channels.
15/37
III. Modelling of MIMO Channels
Block Diagram of a MIMO Mobile Communication System
Transmitter Receiver
x
1
x
2
x
M
T
r
1
r
2
r
M
R
h
11
h
1
2
h
2
1
h
M
R
1
h
22
h
M
R
2
h
1
M
T
h
M
R
M
T
h
2
M
T
_
_
_
_
_
_
-
-
-
H
H
H
H
H
H
H
H
H
H
H
Hj
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
Sw
*
H
H
H
H
H
H
H
H
H
H
H
Hj
3
Channel coecients: The channel coecient h
ij
(t) describes the transmission behavior of the
channel from the jth transmit antenna to the ith receive antenna.
Channel matrix: H(t) = [h
ij
(t)] C
M
R
M
T
Channel capacity: C(t) = log
2
det
_
I
M
T
+
S
BS, total
M
T
N
noise
H(t)H
H
(t)
_
[bits/s/Hz]
(M
R
M
T
)
16/37
Generalized Principle of Deterministic Channel Modelling
Step 1: Starting point is a geometrical model with an innite number of scatterers, e.g.,
(a) one-ring model (b) two-ring model (c) elliptical model
BS MS
BS MS
BS MS
Step 2: Derive a stochastic reference model from the geometrical model.
Step 3: Derive an ergodic stochastic simulation model from the reference model by using
only a nite number of N scatterers.
Step 4: Determine the deterministic simulation model by xing all model parameters of the
stochastic simulation model.
Step 5: Compute the parameters of the simulation model by using a proper parameter
computation method, e.g., the L
p
-norm method (LPNM).
Step 6: Generate one (or some few) sample functions by using the deterministic simulation
model with xed parameters.
17/37
Illustration of the Generalized Principle of Deterministic Channel Modelling
3 4 6
5
2 1
Innite complexity
(N )

sample functions sample functions
Non-realizable Realizable
sample functions
Non-realizable
Reference model
E E < >
Statistical properties
Deterministic SoS
simulation model
computation
Parameter
Simulation of
sample functions
Fixed
parameters
Stochastic SoS
simulation model
Finite complexity
One (or some few) Innite number of Innite number of
Finite complexity
(N 25) (N 25)
Lp-norm method (LPNM)
Geometrical model
18/37
Application of the Generalized Principle of Deterministic Channel Modelling
A Geometrical Model for a MIMO Channel
D
n2
D
n1
A
BS
1
BS
A
2
MS
MS
MS
0
MS
BS
n
BS
A
MS
1
D
2n
BS
max
S
n
v
y
0
BS
D
2
A
BS
MS
D
n
1n
x
R
v
The geometrical model is known as the one-ring model for a 2 2 channel.

The local scatterers are laying on a ring around the MS.
If the number of scatterers N , then the discrete AOA
MS
n
tends to a continuous RV
MS
with
a given distribution p(
MS
), e.g., the uniform distribution, the von Mises distribution, the Laplacian
distribution, etc.
19/37
The Reference Model
Channel coecients: h
11
(t) = lim
N
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
h
12
(t) = lim
N
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
h
21
(t) = lim
N
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
h
22
(t) = lim
N
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
where the phases
n
are i.i.d. RVs and
a
n
= e
j(
BS
/)
[
cos(
BS
)+
BS
max
sin(
BS
) sin(
MS
n
)
]
b
n
= e
j(
MS
/) cos(
MS
n

MS
)
f
n
= f
max
cos(
MS
n

v
)
Channel matrix: H(t) = [h
ij
(t)] =
_
h
11
(t) h
12
(t)
h
21
(t) h
22
(t)
_
Channel capacity: C(t) = log
2
det
_
I
2
+
S
BS, total
2N
noise
H(t)H
H
(t)
_
[bits/s/Hz]
20/37
Statistical Properties of the Reference Model
Space-time CCF:
11,22
(
BS
,
MS
, ) := Eh
11
(t)h
22
(t + )
= lim
N
1
N
N
n=1
a
2
n
b
2
n
e
j2f
n
a
2
n
(
BS
)b
2
n
(
MS
)e
j2f
n
p(
MS
)d
MS
Time ACF: r
h
11
() := Eh
11
(t)h
11
(t + )
=
e
j2f
max
cos(
MS
v)
p(
MS
) d
MS
Relationships: r
h
11
() = r
h
12
() = r
h
21
() = r
h
22
() =
11,22
(0, 0, )
2D space CCF: (
BS
,
MS
) :=
11,22
(
BS
,
MS
, 0)
=
a
2
n
(
BS
)b
2
n
(
MS
)p(
MS
)d
MS
21/37
The Stochastic Simulation Model
Channel coecients:

h
11
(t) =
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
The phases
n
are i.i.d. RVs
Channel matrix:

H(t) =
_

h
11
(t)

h
12
(t)
h
21
(t)

h
22
(t)
_
Stochastic channel matrix
Channel capacity:

C(t) = log
2
det
_
I
2
+
S
BS, total
2N
noise
H(t)

H
H
(t)
_
[bits/s/Hz]
Stochastic process
The analysis of

C(t) has to be performed by using statistical aver-
ages, e.g., C
s
= E
C(t).
22/37
Statistical Properties of the Stochastic Simulation Model
Space-time CCF:
11,22
(
BS
,
MS
, ) := E
h
11
(t)
22
(t + )
=
1
N
N
n=1
a
2
n
(
BS
)b
2
n
(
MS
)e
j2f
n
Time ACF: r
h
11
() := E
h
11
(t)
11
(t + )
=
1
N
N
n=1
e
j2f
max
cos(
MS
n

v
)
Relationships: r
h
11
() = r
h
12
() = r
h
21
() = r
h
22
() =
11,22
(0, 0, )
2D space CCF: (
BS
,
MS
) :=
11,22
(
BS
,
MS
, 0)
=
1
N
N
n=1
a
2
n
(
BS
)b
2
n
(
MS
)
=
1
N
N
n=1
e
j2(
BS
/)[cos(
BS
)+
BS
max
sin(
BS
) sin(
MS
n
)]
e
j2(
MS
/) cos(
MS
n

MS
)
23/37
The Deterministic Simulation Model
Channel coecients:

h
11
(t) =
1
N
N
n=1
a
n
b
n
e
j(2f
n
t+
n
)
The phases
n
are constant quantities
Channel matrix:

H(t) =
_

h
11
(t)

h
12
(t)
h
21
(t)

h
22
(t)
_
Deterministic channel matrix
Channel capacity:

C(t) = log
2
det
_
I
2
+
S
BS, total
2N
noise
H(t)

H
H
(t)
_
[bits/s/Hz]
Deterministic process
The analysis of

C(t) has to be performed by using time averages,
e.g., C
t
=<

C(t) >.
24/37
Statistical Properties of the Deterministic Simulation Model
Space-time CCF:
11,22
(
BS
,
MS
, ) := <

h
11
(t)
22
(t + ) >
=
1
N
N
n=1
a
2
n
(
BS
)b
2
n
(
MS
)e
j2f
n
Time ACF: r
h
11
() := <

h
11
(t)
11
(t + ) >
=
1
N
N
n=1
e
j2f
max
cos(
MS
n

v
)
Relationships: r
h
11
() = r
h
12
() = r
h
21
() = r
h
22
() =
11,22
(0, 0, )
2D space CCF: (
BS
,
MS
) :=
11,22
(
BS
,
MS
, 0)
=
1
N
N
n=1
a
2
n
(
BS
)b
2
n
(
MS
)
=
1
N
N
n=1
e
j2(
BS
/)[cos(
BS
)+
BS
max
sin(
BS
) sin(
MS
n
)]
e
j2(
MS
/) cos(
MS
n

MS
)
25/37
Parameter Computation Method
Parameters: The model parameters to be determined are the discrete AOAs
MS
n
(n = 1, 2, . . . , N).
Aim: Determine the model parameters
MS
n
such that
11,22
(
BS
,
MS
, )
11,22
(
BS
,
MS
, ) .
Solution: L
p
-norm method
E
(p)
1
:=
_
1
max
max
_
0
[r
h
11
() r
h
11
()[
p
d
_
1/p
E
(p)
2
:=
_
1
BS
max
MS
max
BS
max
_
0
MS
max
_
0
[(
BS
,
MS
) (
BS
,
MS
)[
p
d
MS
d
BS
_
1/p
Two alternatives: (i) Joint optimization: E
(p)
= w
1
E
(p)
1
+ w
2
E
(p)
2
w
1
, w
2
: weighting factors
(ii) Using two independent sets
/
MS
n
and
MS
n

in r
h
11
() and (
BS
,
MS
), respectively.
Orthogonalization of the optimization problem.
26/37
Structure of the Simulation Model
+
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . . . . .
. . .
.

.

.
.

.

.
+
.

.

.
.

.

.
1/
N
1/
N
1/
N
s
2
(t)
r
1
(t)
r
2
(t)
b
N
2
b
1
b
N
b
2
b
1
a
1
a
2
a
N
s
1
(t)
a
N
a
2
a
1
e
j(2f
N
t +
N
)
e
j(2f
2
t +
2
)
e
j(2f
1
t +
1
)
27/37
Performance Evaluation
Comparison of the time ACFs r
h
11
() (reference model) and r
h
11
() (simulation model)
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
0.5
0
0.5
1
Time separation, (s)
T
i
m
e

a
u
t
o
c
o
r
r
e
l
a
t
i
o
n

f
u
n
c
t
i
o
n

Reference model
Simulation model
Simulation
max
Reference model: Based on the one-ring model with uniformly distributed AOAs
MS
.
Simulation model: Designed by using the L
p
-norm method (N = 25 , p = 2,
max
= 0.08 s, f
max
= 91 Hz).
28/37
Reference model Simulation model
2-D space CCF (
BS
,
MS
) 2-D space CCF (
BS
,
MS
)
0
5
10
15
20
25
30
0
0.5
1
1.5
2
2.5
3
0.5
0
0.5
1
2
D

s
p
a
c
e

c
r
o
s
s
c
o
r
r
e
l
a
t
i
o
n

f
u
n
c
t
i
o
n

0
5
10
15
20
25
30
0
0.5
1
1.5
2
2.5
3
0.5
0
0.5
1
2
D

s
p
a
c
e

c
r
o
s
s
c
o
r
r
e
l
a
t
i
o
n

f
u
n
c
t
i
o
n

MS
/
BS
/
MS
/
BS
/
Reference model: Based on the one-ring model with uniformly distributed AOAs
MS
.
Simulation model: Designed by using the Lp-norm method with N = 25
(p = 2,
BS
max
= 30,
MS
max
= 3,
BS
=
MS
= 90
,
BS
max
= 2
).
29/37
IV. Simulation of MIMO Channels
Channel capacity:

C(t) := log
2
det
_
I
2
+
S
BS, total
2N
noise
H(t)

H
H
(t)
_
[bits/s/Hz]
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
1
2
3
4
5
6
7
8
9
10
11
Time, t (s)
S
i
m
u
l
a
t
e
d

c
h
a
n
n
e
l

c
a
p
a
c
i
t
y
,

(
b
i
t
s
/
s
/
H
z
)
Capacity
Mean Capacity
Parameters: N = 25, M
BS
= M
MS
= 2,
BS
=
MS
= /2, SNR = 17 dB
30/37
Scattering Scenario
Geometrical one-ring model:
scatterers
Cluster of
BS
y
x
MS
BS
max
v
Parameters:
M
BS
= M
MS
= 2
BS
=
MS
= /2
BS
max
= 2
v
=
MS
0
= 180
f
max
= 91 Hz
Reference model: assuming the von Mises distribution for the AOA
MS
:
p(
MS
) =
1
2I
0
()
e
cos(
MS
MS
0
)
= 0 : p(
MS
) = 1/(2) (isotropic scattering)
: p(
MS
) (
MS
MS
0
) (extremely nonisotropic scattering)
Simulation model: designed by using the LPNM with N = 25.
31/37
PDF of the Channel Capacity

C(t)
0 2 4 6 8 10
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
P
D
F

o
f

t
h
e

c
a
p
a
c
i
t
y

C
(
t
)
Level, r
= 0
= 10
= 100
/
Observations: inuences the PDF p
C
(r) of the capacity

C(t)
If ; angular spread of the AOA
MS
; Var
C(t)
32/37
Normalized LCR of the Channel Capacity

C(t)
0 2 4 6 8 10
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Level, r
N
C
(
r
)
/
f
m
a
x
= 0
= 3
= 5
= 10
= 100
/
Observations: has a strong inuence on the LCR

N
C
(r) of the capacity

C(t)
MS
;

N
C
(r)
33/37
Normalized ADF of the Channel Capacity

C(t)
0 2 4 6 8 10
0
50
100
150
200
250
300
350
400
450
500
Level, r
T
C
(
r
)

f
m
a
x
= 0
= 10
= 100
/
Observations: has a strong inuence on the ADF

T
C
(r) of the capacity

C(t)
MS
;

T
C
(r) .
34/37
The Eect of on the BER Performance
4 5 6 7 8 9 10
10
4
10
3
10
2
10
1
10
0
E
b
/N
0
(dB)
B
i
t

e
r
r
o
r

r
a
t
e
= 0
= 20
= 30
= 40
Coding system: 4-D 16QAM linear 32-state space-time trellis code with 2 transmit and 2 receive
antennas (
BS
= 30 and
MS
= 3).
Observations:
; angular spread ; BER
If decreases from 40 to 0, then a gain of ca. 1.5 dB can be achieved at a BER
of 2 10
3
.
35/37
The Eect of the Antenna Spacing on the BER Performance
4 5 6 7 8 9 10
10
4
10
3
10
2
10
1
10
0
E
b
/N
0
(dB)
B
i
t

e
r
r
o
r

r
a
t
e
BS
/ = 1/2,
MS
/ = 1/2
BS
/ = 1 ,
MS
/ = 1
BS
/ = 30 ,
MS
/ = 3
BS
/ = 30 ,
MS
/ = 1/2
BS
/ = 1/2,
MS
/ = 3
BS
/ = 3 ,
MS
/ = 1
Coding system: 4-D 16QAM linear 32-state space-time trellis code with 2 transmit and 2 receive
antennas.
Observations:

BS
,
MS
; spatial correlation ; BER
A large spacing among the antennas at the BS provides a better performance
than a large antenna spacing at the MS.
36/37
VI. Conclusion
a
The principle of deterministic channel modelling has been generalized.
a
The generalized concept has been applied on the geometrical one-ring scattering model.
a
It has been shown how the parameters of the simulation model can be determined by using the
L
p
-norm method (LPNM).
a
The designed deterministic MIMO channel simulator with N = 25 scatterers has nearly the
same statistics as the underlying stochastic reference model with N = scatterers.
a
The new MIMO space-time channel simulator is very useful in the investigation of the PDF,
LCR, and ADF of the channel capacity C(t) from time-domain simulations.
a
Finally, it has been demonstrated that the system performance decreases if the spatial correlation
increases.
37/37

Modelling and Simulation of MIMO Channels

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Modelling and Simulation of MIMO Channels

Transféré par

Droits d'auteur :

Formats disponibles

Modelling and Simulation of MIMO Channels

Stochastic Rice process (t) Deterministic Rice process

The geometrical model is known as the one-ring model for a 2 2 channel.

Vous aimerez peut-être aussi