Professional Documents
Culture Documents
Rupul Safaya
Organization
Introduction
Theoretical Background
Channel Estimation Algorithm
Conclusions
Future Work
2
Introduction
3
Definitions:
Channel: In its most General sense can describe everything from
the source to the sink of the radio signal. Including the physical
medium.
In this work “Channel” refers to the physical medium.
Channel Model: Is a mathematical representation of the transfer
characteristics of the physical medium.
Channel models are formulated by observing the characteristics of the received
signal.
The one that best explains the received signal behavior is used to model the
channel.
Channel Estimation: The process of characterizing the effect of the
physical medium on the input sequence.
4
General Channel Estimation
Procedure
T r a n s m it t e d s e q u e n c e A c tu a l
R e c e iv e d S ig n a l
x(n)
Channel
Y (n)
E rro r
+ S ig n a l
e (n )
+
-
E s t im a t e d
Channel
M odel E s t im a te d
S ig n a l
Yˆ (n)
E s tim a tio n A lg o r it h m
5
Aim of any channel estimation procedure:
Minimize some sort of criteria, e.g. MSE.
Utilize as little computational resources as possible allowing
easier implementation.
A channel estimate is only a mathematical estimation of
what is truly happening in nature.
Why Channel Estimation?
Allows the receiver to approximate the effect of the channel on
the signal.
The channel estimate is essential for removing inter symbol
interference, noise rejection techniques etc.
Also used in diversity combining, ML detection, angle of arrival
estimation etc.
6
Training Sequences vs.
Blind Methods
There Are two Basic types of Channel Estimation Methods:
7
Algorithm Overview
Consider a radio communications system using training sequences to
do channel estimation.
This thesis presents a method of improving on the training sequence
based estimate without anymore bandwidth wastage.
Jakes Model: Under certain assumptions we can adopt the Jakes
model for the channel.
This allows us to have a second estimate independent of the data based
(training sequence) estimate.
The Kalman estimation algorithm uses these two independent
estimates of the channel to produce a LMMSE estimate.
Performance improvement: As a result of using the Jakes model in
conjunction with the data based estimates there is a significant gain in
the channel estimate.
8
Theoretical Background
9
Signal Multipath
P a th 2
B u ild in g
D ir e c t P a th
M o b ile T e r m in a l
P a th 3
B a s e S t a t io n
T a ll B u ild in g s
10
Multipath
Signal multipath occurs when the transmitted signal arrives at the
receiver via multiple propagation paths.
Each path can have a separate phase, attenuation, delay and
doppler shift associated with it.
Due to signal multipath the received signal has certain undesirable
properties like Signal Fading, Inter-Symbol-Interference, distortion
etc.
Two types of Multipath:
Discrete: When the signal arrives at the receiver from a limited number
of paths.
Diffuse: The received signal is better modeled as being received from a
very large number of scatterers.
11
Diffuse Multipath
The Signal arrives via a continuumof multipaths.
Thus the received signal is given by:
~ ~(τ;t)~
y(t) α ~(,t)
s(tτ)dτwhere α is the complex
M
2
A tte n u a to r M
M1 M 1 A tte n u a to r
2
2 2
A tte n u a to r A tte n u a to r
1 0
2 2
A tte n u a to r A tte n u a to r
~
y(t)
13
Tap-Delay Line Model
T h e re c e iv e d s ig n a l c a n b e w ritte n a s :
~y ( t ) m
g~ ( t ) ~s ( t ) w h e re :
m W
m
1 ~ m
g~ ( t ) h ( ;t) is th e s a m p le d (in th e d o m a in ) c o m p le x
m W W
lo w -p a s s e q u iv a le n t im p u ls e re s p o n s e . W : is th e B a n d w id th o f
th e B a n d p a s s S ig n a l
:
W S S U S (W id e S e n s e S ta tio n a ry U n c o rre la te d
S c a tte r in g ) : A s s u m in g W S S U S th e d e la y p ro file a n d s c a tte rin g
fu n c tio n a re a s fo llo w s :
1 ~ * ~
M u ltip a th In te n s ity P ro file : R ( ) E h ( , t ) h ( , t ) T h is
C 2
d e fin e s th e v a ria tio n o f a v e ra g e re c e iv e d p o w e r w ith d e la y .
D e la y s p re a d is th e ra n g e (in d e la y ) fo r w h ic h th e a v e ra g e
p o w e r is n o n -z e ro .
S c a t t e r i n g f u n c t i o n : S ( ; v ) F [ R C ( )] T h i s d e s c r i b e s t h e p o w e r
s p e c tra l d e n s ity a s a fu n c tio n o f D o p p le r fre q u e n c y (fo r fix e d
d e la y )
14
Tap gain functions
Complex Gaussian process: Assuming infinite
scatterers, as a consequence of the Central Limit
Theorem, we can model the impulse response as a
complex Gaussian process.
Rayleigh: If there is no one single dominant path, then the
process is zero mean and the channel is Rayleigh fading.
Ricean: If there is asingle dominant path then the process is non-
zero-mean and the channel is Ricean.
The tap gain functions are then sampled complex
gaussian processes.
15
Model Parameters
The tap-delay model requires the following
information.
Number of taps are TMW+1. Where TM is the delay spread and W
is the information bandwidth.
Tap Spacing is: 1/W.
Tap gain functions are discrete time complex Gaussian processes
with variance given by the Multipth spread and PSD given by the
scattering function.
Tap gain functions as key: Once having specified the
tap spacing and the number, it only remains to track
the time varying tap gain functions in order to
characterize the channel as modeled by the tap-delay
line.
Jakes model: The Jakes model (under certain
assumptions) assigns the spectrum and autocorrelation
to the tap-gain processes.
16
Jakes Model
P r o p a g a tio n
B a s e S ta tio n
P a th
M o b ile
V e h ic le V e lo c it y
w
av
el
eng
t
hof
th
ec
arr
ie
ra
n
di
sth
ea
ng
l
eof
ar
ri
val
)
a
nd
is
thu
sbo
u
nde
d.
U n ifo rm a n g le o f a rriv a l: A s s u m e th a t th e re c e iv e d
p o w e r is u n ifo rm ly d is trib u te d o v e r th e a n g le o f a rriv a l.
C o n s ta n t v e h ic le v e lo c ity : F in a lly th e a s s u m p tio n is
m a d e th a t th e v e h ic le is n o t a c c e le ra tin g .
P o w e r s p e c tru m e x p re s s io n :
2 1 /2
f f
S ( f ) 1 C w h e re f is th e c a rrie r fre q u e n c y a n d f is
f C m
m
th e d o p p le r s p re a d .
S h a p in g F ilte r: T h e J a k e s s p e c tru m c a n b e s y n th e s iz e d
b y a s h a p in g filte r w ith th e fo llo w in g im p u ls e re s p o n s e :
h j (t) 21 / 4 ( 3 ) fm (2 fm t) 1 / 4 J1 / 4 (2 fm t)
4
19
Jakes Spectrum at a Doppler spread of 530 Hz.
20
The Channel Estimation
Algorithm
21
Introduction
Aim: To improve on the data-only estimate.
Jakes model: We have adopted the Jakes model for the radio channel.
Tap-gains as auto-regressive processes: The Jakes power spectrum is
used to represent the tap-gains as AR processes.
State-Space Representation: We have two independent estimates of
the process from the data-based estimate and the Jakes model.
These are used to formulate a State-Space representation for the tap-gain
processes.
An appropriate Kalman filter is derived from the state-space
representation.
Derivation: The algorithm is developed first for a Gauss-Markov
Channel and then for the Jakes Multipath channel
22
AR representation of the
Tap-gains
S (n)
w(n) S (n 1)
.
.
.
U n it U n it U n it
D e la y D e la y D e la y
S (n N )
h1
. h2
.
. hN
23
D iffe re n c e E q u a tio n fo rm :
N
S (n )
i1
iS (n i) w (n ) W h e re S (n ): is th e c o m p le x g a u s s ia n
p ro c e ss, i a re th e p a ra m e te rs o f th e m o d e l. T h e m o d e l is d riv e n
by w (n ): A seq uence o f id e n tic a lly d is trib u te d z e ro -m e a n
C o m p le x G a u s s ia n ra n d o m v a ria b le s .
S o lv in g fo r th e A R p a ra m e te rs : T h is c a n b e d o n e in tw o w a y s
Y u le -W a lk e r e q u a tio n s o lu tio n : T h e a u to c o rre la tio n c o e ffic ie n t
o f th e p ro c e s s , is a n N (n u m b e r o f ta p s in th e A R m o d e l) o rd e r
d iffe re n c e e q u a tio n th a t c a n b e s o lv e d .
P S D o f th e p ro c e s s : T h is m e th o d is m o re c o m p le x a n d in v o lv e s
fin d in g th e fo rm o f th e s h a p in g filte r fo r th e p ro c e s s P S D .
24
Data based estimator
A s s u m p t io n : T h e c h a n n e l r e m a in s c o n s t a n t o v e r
t h e s p a n o f t h e t r a in in g s e q u e n c e .
S p e c if ic s :
T r a in in g s e q u e n c e : L e t t h e ‘M ’ le n g t h t r a in in g s e q u e n c e b e :
x x 0 x 1 ..... .... x M 1 T
C h a n n e l I m p u ls e R e s p o n s e : L e t th e ‘L ’ le n g th im p u ls e
r e s p o n s e b e h h 0 h 1 h 2 . .. . h L 1
~ ~ ~ ~ T
R e c e iv e d s ig n a l: F o r c h a n n e l n o is e n C o f v a ria n c e C
2
th e
re c e iv e d s ig n a l in v e c to r fo rm is g iv e n b y : Y X h nc. W h e re X is
a n L N 1 L T o e p litz m a trix c o n ta in in g d e la y e d v e rs io n s o f th e
tra in in g s e q u e n c e s e n t.
25
Channel estimate: The data based estimate is given by
correlating the received signal with the training sequence:
hˆ ( X T X ) 1 ( X T Y )
26
Tracking a Gauss-Markov
Channel
27
A Gauss-Markov tap-gain process has an exponential autocorrelation.
28
Kalman Filter Derivation
System M odel: Since the auto-correlation is
exponential, the tap gain is a simple first order A R
process:
S ( n ) 1 S ( n 1) w ( n ) W here
S (n ) : is the complex gaussian tap-gain process.
1 is the parameter of the AR model assum ed.
O bservation M odel: Since the data based estimator
produces “noisy” estim ates of the process. The
follow ing model emerges:
X (n) S (n) v (n) . W here
X(n) is the data based estim ate of S(n)
v(n) is the error of the data based estimate and
2
σ
σ 2
v
M
is the error variance
c
29
Kalman filter equations
S c a la r K a lm a n filte r: G iv e n th e s ta te s p a c e
re p re s e n ta tio n , a s ta n d a rd s c a la r K a lm a n filte r is u s e d
to tra c k th e p ro c e s s .
E q u a tio n s :
T h e in itia l c o n d itio n s a re :
S ˆ ( 0 )=E S(n) 0
w
P ( 1 ) 2 and 2
v
T h e K a lm a n g a in is g iv e n b y :
P (n )
k (n )
P ( n ) v2 ( n )
T h e c u rre n t e s tim a te o f th e p ro c e s s , a fte r re c e iv in g th e d a ta e s tim a te is
g iv e n b y :
S ˆ cur (n)= S ˆ (n) k(n) [ X(n)- S ˆ (n )]
T h e p re d ic te d e s tim a te o f th e p ro c e s s is g iv e n b y :
S ˆ (n 1 )= 1 S ˆ cu r (n)
T h e c u rre n t e rro r c o v a ria n c e is g iv e n b y :
P cu r ( n ) 1 k ( n ) P ( n )
T h e p re d ic tio n e rro r c o v a ria n c e (M S E in th is c a s e ) is g iv e n b y :
P (n 1) 1
2
P cur (n ) w
2
30
Simulation Parameters
S y s t e m e q u a t i o n : S(n) . 9 S(n 1 ) w(n)
O b s e rv a tio n E q u a tio n : X (n ) S (n ) v(n )
T h e f r a m e r a t e i s : R F 5 10 4 Frames/ sec . T h e s i m u l a t i o n i s r u n a t t h e
fra m e ra te .
M 8 : T h e le n g th o f th e tra in in g s e q u e n c e . T h e c h a n n e l is a s s u m e d
to b e in v a ria n t fo r th e s e M b its .
T h e s ig n a l to n o is e ra tio o f th e c h a n n e l, fo r E B 1 is :
E E
B
B
6 dB
N O 2 c
2
T h us c
2
. 1256
2
T h e d a ta e s tim a to r v a ria n c e σ 2
V c
0.0157
M
T h e v a ria n c e o f th e ta p g a in p la n t n o is e is σ 2
w 2 V
2
0 . 0314
31
Results:
T h e p re d ic tio n M S E is :
S te a d y S ta te :
S im u la tio n R e s u lts :
P e r f o r m a n c e I m p r o v e m e n t: W e c a n s e e th a t c o m p a re d to th e d a ta o n ly
e s tim a to r, th e re is a s ig n ific a n t im p ro v e m e n t:
V
2
P curr
SS
28 %
V
2
33
Tracking a single Jakes
Tap-Gain Process
34
Single ray Jakes Channel
Consider a single ray line of sight radio channel.
More Complex Channel: The underlying channel model is no longer
a single tap AR process.
AR representation: The tap-gain process with the Jakes spectrum is
a stationary process. We can represent it as an AR process.
Parameters: We derive the co-efficients for the process from the closed
form expression of the Jakes channel-shaping filter.
State-Space representation: Using the AR model and the data
based estimator, a state-space representation is derived.
Kalman tracking filter: Similar to the Gauss-Markov case, a Kalman
filter to track the process is derived from the State-Space
representation.
35
AR representation of the
Jakes Process
J a k e s S h a p in g F ilte r: T h e c lo s e d fo rm e x p r e s s io n o f th e
J a k e s filte r is g iv e n b y :
3
h (t) 21 / 4 ( ) f (2 f t) 1 / 4 J
j
(2 f t) . W h e re is th e G a m m a
4 m m 1/4 m
fu n c tio n a n d J
1/4
is th e fra c tio n a l B e s s e l fu n c tio n .
F IR filte r: T h is e x p r e s s io n is s a m p le d to p ro d u c e th e F IR
J a k e s c h a n n e l s h a p in g filte r.
O u tp u t: If th e in p u t to th is filte r is g a u s s ia n w h ite n o is e ,
th e n th e o u tp u t is s im p ly th e c o n v o lu tio n s u m :
M 1
S (n )
m 0
h j(m )w (n m ) . W h e re M is th e lie n g th o f th e F IR filte r.
36
Convolution as a MA sum: The convolution is nothing but a
weighted moving average of the white noise inputs.
w(n 1) w(n 2)
w(n M 1)
w(n) U n it
.
.
.
U n it U n it
D e la y D e la y D e la y
.
. h hn- MN 1
.
h2
S (n)
h1
h0
T r u n c a te d A R m o d e l: T h e a c c u r a c y o f th e tr u n c a te d m o d e ls in
r e p r e s e n tin g th e J a k e s p r o c e s s is s tu d ie d .
P a r a m e te r s :
f m 50 Hz . T h is i s t h e D o p p l e r b a n d w id t h o f th e c h a n n e l.
F S 16 f m . T h is i s t h e J a k e s - s h a p in g - f ilte r s a m p lin g r a t e .
F S PSD 5 10 3 samples / Hz . T h is is t h e J a k e s s p e c tr u m s a m p l in g f r e q u e n c y .
N MA 64 . T h i s i s th e F I R s h a p in g f il te r l e n g th .
N is th e l e n g th o f t h e tr u n c a te d A R m o d e l.
P S D a n d a u to c o r r e la tio n :
P S D is g iv e n b y th e a n a ly tic a l e x p r e s s io n : 2
S SS ( f ) W
2
N
1
i 1
i e j 2 fi
39
Jakes autocorrelation for truncated processes
40
Kalman filter Derivation
S ta te S p a c e re p re s e n ta tio n : A s in th e p re v io u s c a s e ,
w e a re g o in g to re p re s e n t th e s y s te m u s in g th e S ta te -
S p a c e fo rm a n d d e riv e th e a p p ro p ria te K a lm a n filte r.
S y s te m m o d e l: T h e A R fo rm o f th e J a k e s p ro c e s s is
N
S (n ) S (n i) w (n )
i
. W h e re i a re th e A R c o e ffic ie n ts
i 1
c a lc u la te d a n d N is th e o rd e r o f th e m o d e l u s e d .
Its tw o e q u iv a le n t fo rm s a re :
S(n-
N
S(n) . . 1 ) w(n)
S(n- 1 ) 11 .
2
. 0 0 S(n- 2 ) 0
. 0 1 . 0 0 . . in m a trix fo rm
. 0 . . . 0 . .
S(n-N 1 ) 0 0 0 1 0 S(n-N) 0
an d S (n ) A S (n 1) W (n ) in v e c to r fo rm .
41
T h e s y s te m m a trix is d e fin e d a s :
. .
11 .
2
. 0
N
0
A 0 1 . 0 0
0 . . . 0
0 0 0 1 0
T h e P la n t n o is e c o v a ria n c e m a trix is
σ w
2
0 . .0
0
_ H
_ 0 0 .... . 0
Q E W (n) W (n)
0 0 0 ... 0
0 0 ... 0
O b s e rv a tio n E q u a tio n :
X (n ) S (n ) v(n ) W h e re v (n ) is th e e rro r o f th e d a ta b a s e d
2
e s tim a te w ith a v a ria n c e o f 2
σ c
v M
E x p re s s e d in m a trix a n d v e c to r fo rm :
X(n) S(n) v(n)
X(n- 1 ) S(n- 1 ) v(n- 1 )
. . . o r X (n ) H S (n ) v (n )
. . .
X(n-N 1 ) S(n-N 1 ) v(n-N 1 )
w h e re H is th e id e n tity m a trix .
42
O b s e rv a tio n n o is e c o v a ria n c e m a trix :
1 0 0 . .. 0
_ H
_
v (n )
2 0 1 .... .. 0
R E v (n )
σ
v
0 0 1 ... . 0
0 0 ... . 1
K a lm a n filte r E q u a tio n s : G iv e n th e a b o v e S ta te S p a c e
fo rm u la tio n , w e c a n u s e a v e c to r K a lm a n filte r to tra c k
th e ta p g a in p ro c e s s .
T h e in itia l c o n d itio n s a re :
S ˆ ( 0 )=E S (n ) = z e r o m a t r i x o f l e n g t h (N L)
P ( 1 ) 2 I a nd 2 I
w v
T h e K a lm a n g a in is g iv e n b y :
K (n ) P (n )H T
[ HP (n )H T
R ]1
T h e c u rre n t e s tim a te o f th e p ro c e s s , g iv e n th e d a ta e s tim a te is g iv e n
b y :
S ˆ c u r ( n )= S ˆ (n ) K(n ) [ X ( n )- H S ˆ (n ) ]
T h e p re d ic te d e s tim a te o f th e p ro c e s s , is g iv e n b y :
S ˆ (n 1 )=A S ˆ cu r (n )
T h e c u rre n t e rro r c o v a ria n c e is g iv e n b y :
P cu r ( n ) I K ( n ) H P ( n )
T h e p re d ic te d e rro r c o v a ria n c e is g iv e n b y :
P ( n 1 ) A P cu r ( n ) A T Q
43
Simulation Parameters
S y s te m e q u a tio n :
S (n ) A S (n 1) W (n )
N = 5 . T h is is th e n u m b e r o f ta p s in th e A R m o d e l.
O b s e rv a tio n E q u a tio n :
X (n ) H S (n ) v (n )
T h e D o p p le r b a n d w id th is f m 50 Hz
T h e fra m e ra te is : R
F
5 10 4 Frames/ s ec . T h e s im u la tio n is ru n a t th e fra m e
44
ra te .
M 8 : T h e le n g th o f th e tr a in in g s e q u e n c e .
T h e s ig n a l to n o is e ra tio o f th e c h a n n e l, fo r E
B
1 is :
E EB
B
6 dB
N O 2 c2
T h u s c2 . 1256
T h e v a ria n c e o f th e ta p g a in p la n t n o is e is σ 2 2 2 0 . 0314
w V
1 0 0 ...... 0
0 ........ ... 0
Q 0 . 0314
0 0 ........ . 0
0 0 ......... 0
2
T h e d a ta e s tim a to r v a ria n c e 2
σ c 0.0157
V M
1 0 0 ... 0
0 1 .......... . 0
R . 0157 .
0 0 1 ........ 0
0 0 ......... 1
45
Results
46
Error covariance:
T h e e r r o r c o v a r ia n c e is d e f in e d a s :
^ ^
H
P E S ( n 1 ) S (n 1) S ( n 1 ) S (n 1)
W e c a n th e n in te r p r e t th e d ia g o n a l e le m e n ts a s f o l lo w s :
MSE ( n 1 / n ) * * ....... *
* MSE * * ....... *
(n / n )
P * * MSE * * ....... *
(n 1 / n )
.
. * * .......... .......... ... MSE
S (n N 2 / n )
W h e re MSE
(n 1 / n )
is th e M S E o f th e p r e d ic t io n .
MSE
(n / n )
is t h e M S E o f th e c u r r e n t s ta te s e s ti m a te
47
T h e s te a d y s ta te a n d s im u la te d e rro r c o v a ria n c e
m a tric e s a re :
1.0921 1.0777
0.1117 0.1125
diag ( P )
SS
0.0565 diag ( P
Sim
) 0.1081
0.0388 0.1118
0.0296 0.1144
P e r f o r m a n c e I m p r o v e m e n t : T h e re is a s ig n if ic a n t
p e rfo rm a n c e g a in c o m p a re d to th e d a ta o n ly e s tim a to r
2 diag ( P )( 2 ,1 )
V Sim 29 %
2
V
48
Multipath Channel
Estimation
49
Multipath Jakes Channel
Consider a multipath radio channel.
Assume the Jakes model on each path.
AR representation: For the Multipath case, a modification of the
single ray AR system model is presented.
State-Space representation: Using the AR model and the data
based estimator, a state-space representation is derived.
Kalman tracking filter:Once again a vector Kalman filter is used to
track the tap-gain functions.
50
System model
A s s u m p tio n s : T h e ta p g a in p ro c e s s e s a re in d e p e n d e n t
b u t h a v e th e s a m e J a k e s s p e c tru m .
A R R e p r e s e n ta tio n :
N
S (n ) S (n i) w (n )
1 i 1 1
i1
N
S (n ) S (n i) w (n )
2 i 2 2
i1
.
.
N
S (n ) S (n i) w (n )
L i L L
i1
w h e re
S ( n ) is th e l th p ro c e s s to b e tra c k e d
l
i
a re th e A R m o d e l p a ra m e te rs . T h e s e a re th e s a m e fo r e a c h
p ro c e s s .
w l ( n ) is th e p la n t n o is e d riv in g th e ta p g a in fu n c tio n . T h e re la tiv e
v a ria n c e is d e te rm in e d b y th e p o w e r d e la y p ro file o f th e c h a n n e l.
L is th e n u m b e r o f p ro c e s s e s b e in g tra c k e d
51
T h e m a trix f o r m o f th e s y s te m e q u a tio n f o r ‘ L ’ p r o c e s s e s is :
S 1 (n)....... .......S L (n) ........ S 1 (n - 1)........ S L (n - 1) w (n)....... ..w (n)
1 2 N 1 L
S
1 (n - 1)........ ..S (n - 1) 1 0 0......0 S 1 (n - 2)........ S L (n - 2) 0......... .......... 0
L
. . 0 1......... . 0 . . . .
. .......... ...... 0 . .
S 1 (n - N 1) ...S L (n - N 1) 0 0 .......1 0 S 1 (n - N) ....S L (n - N) 0 .......... ......... 0
In v e c to r f o r m : S ( n ) A S ( n 1) W ( n ) , w h e re
1 2 ........ N
1 0 0 ...... 0
A 0 1 .......... 0 is th e s y s te m m a tr ix .
.......... ...... 0
0 0 ....... 1 0
L 2
σ w (l) ........ 0
l 1
_ H 0 ......... .......... 0
_
Q E W (n) W (n) 0 ......... ......... . 0 is th e p la n t n o is e c o v a r ia n c e m a tr ix .
.
0 ......... ........ .. 0
52
Observation Model
Assuming that the data based estimates are path-wise
independent, we have the following model:
X (n) S (n) v (n)
1 1 1
X (n) S (n) v (n)
2 2 2
.
.
X (n) S (n) v (n)
L L L
Where,
Sl (n) : The l’th process at time n.
th
X l (n) : The l’ data based estimate of S(n)
th
vl (n) : Error of the l’ data based estimate.
c2
σv2 is assumed to be the same for all paths
M
53
The Observation Equation can be written in matrix and vector form as
follows:
X (n) X (n) S (n) S (n) v (n) v (n)
1 . . L 1 . . L 1 . . L
X1 (n - 1) . . X
L
(n - 1) S1 (n - 1) . . L (n - 1)
S v1 (n - 1) . . L (n - 1)
v
. . . . . . . . . . . .
. . . . . . . . . . . .
X1 (n - N 1) . . X (n - N 1) S (n - N 1)
L
. . S (n - N 1) v (n - N 1) . . v (n - N 1)
1 L 1 L
_ _ _
X (n) H S (n) v(n) where H is an identity matrix.
The observation noise covariance matrix is given by:
R E[v (n) v (n) H ] ( L 2 ) [ I ] where [I ] is an ( N N ) identity matrix.
v
54
Simulation parameters
N=5. This is the number of taps in the AR model.
frame rate.
M 8: The length of the training sequence. The channel is assumed to
be invariant for these M bits.
55
T h e s ig n a l to n o is e ra tio o f th e c h a n n e l, fo r E B 1 is :
E E
B
B
6 dB
N O 2 c
2
T h u s c . 1256
2
T h e p o w e r d e la y p ro file σ 2 ( l ) 0 . 0314 [ 1 , 0 . 9 , 0 . 81 ]
w
T h e c o v a ria n c e o f th e p la n t n o is e is :
0.0 851 . 0
Q . . .
0 . 0
2
T h e d a ta e s tim a to r v a ria n c e σ 2
V c
0.0157 . H e re th e a s s u m p tio n is
M
m a d e th a t th e tra in in g s e q u e n c e h a s a n id e a l a u to c o rre la tio n .
R 3 0 . 0157 . I
56
Results
58
Channel estimation for the third process.
59
Error Covariance
T h e E r r o r c o v a r ia n c e is d e f i n e d a s :
H
P E S(n 1 ) Sˆ (n 1 ) S(n 1 ) Sˆ (n 1 )
W e c a n t h e n in t e r p r e t t h e d i a g o n a l e le m e n t s a s f o l l o w s :
MSE (l) * * ...... .*
l (n 1 /n)
* MSE (l) * * ...... .*
l (n/n)
P * * MSE (l) * * ...... .* w h e re
l (n 1 /n)
.
.* *...... .......... ....... MSE (l)
l (n N 2 /n)
MSE (l ) : t h e s u m o f t h e M S E o f t h e p r e d i c te d s t a te e s t i m a t e
l (n 1 / n )
o n a ll p r o c e s s e s .
MSE (l) : t h e M S E o f t h e c u r r e n t s t a t e s e s ti m a t e o n a l l
l (n/n)
p ro c e sse s.
60
T h e s im u la te d e rro r c o v a ria n c e d ia g o n a l is g iv e n b y :
Pr ocess # (l )
diag (P Sim (l) 1 2 3
diag ( P Sim ) diag ( P SS )
0 .0 3 8 4 0 .0 3 5 1 0 .0 3 2 1 0 .1 0 5 6 0 .1 1 1 1
0 .0 1 1 2 0 .0 1 0 9 0 .0 1 0 6 0 .0 3 2 7 0 .0 3 2 1
0 .0 0 9 8 0 .0 0 9 7 0 .0 0 9 7 0 .0 2 9 2 0 .0 1 6 9
0 .0 1 0 5 0 .0 1 0 5 0 .0 1 0 5 0 .0 3 1 5 0 .0 1 2 2
0 .0 1 1 2 0 .0 1 1 1 0 .0 1 1 2 0 .0 3 3 5 0 .0 0 9 6
0 .0 1 1 2 0 .0 1 0 9 0 .0 1 0 6 0 .0 3 2 7 0 .0 3 2 1
0 .0 0 9 8 0 .0 0 9 7 0 .0 0 9 7 0 .0 2 9 2 0 .0 1 6 9
0 .0 1 0 5 0 .0 1 0 5 0 .0 1 0 5 0 .0 3 1 5 0 .0 1 2 2
0 .0 1 1 2 0 .0 1 1 1 0 .0 1 1 2 0 .0 3 3 5 0 .0 0 9 6
0 .0 1 1 7 0 .0 1 1 7 0 .0 1 1 7 0 .0 3 5 1 0 .0 0 8
P e r fo r m a n c e im p r o v e m e n t o n e a c h p a th :
T h e d a ta b a s e d e s tim a te h a s a M S E o f σ 2
V 0 . 0157 o n e a c h p a th .
V
2
diag ( P curr )( 1 , 1 ) . 0157 . 0112
P a th 1 im p ro v e m e n t: sim
100 28 . 66 %
V
2
. 0157
. 0157 . 0109
P a th 2 im p ro v e m e n t: 100 30 . 57 %
. 0157
. 0157 . 0106
P a th 3 im p ro v e m e n t: 100 32 . 48 %
. 0157
A lm o s t a 3 0 % im p ro v e m e n t o n e a c h p a th
61
Conclusions
Developed a Kalman filter based channel estimation
algorithm for the Multipath radio channel.
Significant gain in performance over a training sequence
based estimator.
This improvement is obtained without wasting any more
bandwidth.
Also allows us to predict the channel state without
having to wait for data.
62
Future Work
Use Of Multiple Sampling Rates:
Instead of waiting for the data to arrive at the end of every frame we can run
the Kalman filter at a higher rate than the frame rate.
In the absence of a data based estimate perform the time-update portion of the
algorithm and do a measurement update when data is received.
Allows estimates to be available as required.
Different process models on each path:
In case the process model varies with path, we can still use the Kalman filter but
with some modifications to the system matrix.
Correlated paths:
For correlated paths the Kalman filter needs to be modified.
63