See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3382828

Tapped delay line model of linear randomly time-variant WSSUS channel

Article  in  Electronics Letters · October 2000

DOI: 10.1049/el:20001168 · Source: IEEE Xplore

CITATIONS READS

17 1,618

1 author:

Jan Sykora
Czech Technical University in Prague
89 PUBLICATIONS   356 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Special Issue "Advanced Coding and Stochastic Signal Processing in Dense Communication Networks" View project

All content following this page was uploaded by Jan Sykora on 09 August 2016.

The user has requested enhancement of the downloaded file.

 On the TDL Model of the Linear Randomly

Time-Variant WSSUS Channel


Jan Sykora

June 14, 2000

Abstract
Tapped Delay Line (TDL) model of the WSSUS channel is widely used
as a propagation model in digital communications. It is usually assumed
to have uncorrelated channel taps coecients. However this assumption
proves to be only an approximation. A violation of the approximation
conditions leads to substantial modeling error especially for the channels
with short rectangular-like power delay proles. A correct model with
mutually correlated coecients is presented.

1 Introduction
Wide Sense Stationary Uncorrelated Scattering (WSSUS) Linear Randomly

Time-Variant (LRTV) channel model is widely used for modeling of the sig-

nal propagation in mobile communications environment. The theoretical back-

ground of the stochastic description of such channel is treated in large number

of works, e.g. [1], [2]. However most of these sources can be back-tracked to

the thorough treatment of the topic done by Bello in [3]. The stochastic LRTV

model is generally valid for arbitrary channel excitation signals. An important

 Czech Technical University of Prague, faculty of electrical engineering, K337 dept. of
radioelectronics, Technicka 2, 166 27 Praha 6, Czech Republic
E-mail: [email protected], http://radio.feld.cvut.cz/~sykora

1
simplication of the LRTV model can be obtained for a special but practically

very important case of bandwidth limited channel excitation. In that case the

cannel can be modeled as a Tapped Delay Line (TDL) with randomly time-

variant channel taps. So far the treatment of the topic in the publications is

very consistent and rigorous. The major confusion and departure from rigorous

treatment appears at the point where the particular stochastic properties of the

TDL coecients are evaluated. A clarication of this problem and rectication

of various misinterpretations spread in the literature is the main goal of this

letter.

2 System model
Here, the LRTV channel model will be briey dened. Only the characteristics

needed for the development of the topic are dened. The details can be found

in [3] and [1].The whole treatment is performed on signal complex envelopes .

Channel input-output relations are

Z 1
( )=
y t (
x t )( )
 g t;  d (1)
1

where () ()
x t ;y t are channel excitation and response respectively and time-

variant impulse response is ( ). Time-variant frequency response is G(t; f ) =

g t; 

F f [g(t;  )]. A zero mean channel (Rayleigh) E[g(t;  )] = 0 is assumed. Gener-

( )

alization for nonzero mean can be easily done on the superposition principle.

Channel correlation function is dened as follows. It is assumed that the

channel response ( ) is WSS in t-time and with US in  -time which means

g t; 

that the correlation E[g (t ;  )g (t ;  )] is zero for any  6=  . Spaced-time

1
 1 2 2 1 2

Delay Correlation function is

E [g(t +
t;  +
 )g (t;  )] WSSUS

= Rg (
t;  )Æ(
 ) (2)

2
3 Tapped delay line model
In the next treatment it will be assumed that the channel excitation is bandwidth
limited jf j  1=(2Tp)
to and can be correctly sampled with Tp sampling period.

Using generalized sampling/interpolation theorem ( sinc(t) = sin(t)=(t))

X
()=
x t (
x kTp + t ) sinc ((t (kTp + t ))=Tp)
0 0 (3)
k

which holds for arbitrary t0 the following can be easily get

Z 1
( )=
y t (
x t )( )
 g t;  d
1
X Z 1
= (
x t mTp ) ( ) sinc ( =Tp
g t;  )
m d: (4)
m 1

The substitution (kTp + t ) = mTp where m is integer number was used here.
t 0

It is important to stress that no assumption on the speed of g (t;  ) changes in

neither t nor  variable was needed to get the result. Then the TDL coecients

are the projection of ( ) into the fsinc((

g t;  ) )gm basis
mTp =Tp

Z 1
am t ()= ( ) sinc ( =Tp
g t;  )
m d (5)
1

and TDL model is

X
( )=
y t () (
am t x t )
mTp : (6)
m

Several possible approaches to this result follows.

1. Uncorrelated TDL coecients g(t; mTp). The result (5) is generally not

recognized in the literature. A widely spread error is the one stating that

am t ( ) = Tpg(t; mTp) (7)

which is implied by the fact that the ( ) can be correctly sampled in

g t; 

3
  -time with Tp sampling period. E.g. in the [1, chapter 14-5-1] Eq. 14-5-4

would hold only for nowhere stated case of ( ) = 0 for jf j > 1=(2Tp).
G t; f

However this assumption would clearly violate the US condition in (2) as

can be seen from application of interpolation theorem in  -time

X
E [g(t ;  )g(t ;  )] = E[g(t ; kTp)g(t ; kTp)]
1 1 2 2 1 2
k
 sinc( =Tp k) sinc ( =Tp k) :
1 2 (8)

The reference [2, chapter 11.3] also assumes the bandwidth limited ( ).
G t; f

A closer investigation can be found in [3, section VI.A.1.a] however it also

carries errors and does not treat the topic to the end. As a consequence of

the (7) the TDL coecients are widely (but erroneously) assumed to be

uncorrelated. However this is not correct. The consequences of this faulty

conclusion will be discussed later. Surprisingly to the fact implied by the

(7) the TDL coecient variance is usually considered to be E jam(t)j  =

2

R (0; mTp ) which requires WSSUS channel and holds on condition of

Tp g

slowly changing Rg (0;  ) (see also (10)).

2. Mutually correlated TDL coecients. Correct investigation of the TDL

coecient correlation properties with WSSUS assumption reveals that

Z 1
E[ak (t1)a (t2)] =
i Rg (t 1 ; )
t2 
1
 sinc( =Tp k ) sinc( =Tp i) d: (9)

The coecients are generally mutually correlated.

3. Slowly changing Rg (
t;  ). It is not dicult to see that the special case

of Rg (t1 ; ) changing very slowly in  -time with respect to the Tp

t2 

allows an approximation

E[ak (t )ai (t )]  TpRg (t

1 2 1 ;
t2 kTp Æ k)[ i ] (10)

4
 where ( ) and ai(t) are uncorrelated
ak t for k 6= i (Æ[0] = 1, Æ[k] = 0 for
k 6= 0). The approximation actually corresponds in the consequences

to the case 1 however this is obtained from entirely dierent principle.

Here the US assumption holds and the condition reects the properties

of Rg (
t;  ) rather then properties of G(t; f ). This approximation can

thus be fullled only if the channel delay spread d  Tp . Unfortunately

a large number of practically used channels do not fall into this case. This

is especially true for channel models with the d equal to small multiple

of Tp and with rectangular-like power delay prole.

4 TDL coecient properties impact on the chan-

nel model
Here it will be shown what consequences can be expected on the channel char-

acteristics for dierent TDL coecients properties. There will be investigated

three cases. 1. TDL channel model with correctly obtained correlated coe-

cients (5). 2. TDL model with mutually uncorrelated coecients (7) get by

sampling of (
g t; mTp ) with E jam(t)j  = TpRg (0; mTp). 3.
2
As a reference

channel model the original input-output relation (1) will be used. The corre-

sponding signals and characteristics for 1. and 2. cases will be denoted by

subscript (:) ; (:)

1 2 respectively. In all cases the channel excitation will be ar-

bitrary bandwidth limited ( jf j < 1=(2Tp)) signal xp (t) (in order to satisfy the
conditions for TDL derivation). The reference behavior of the channel is based

on the physical interpretation of the Spaced-time Delay Correlation function

which can be easily obtained from the above stated denitions with WSSUS

assumption

Z 1
Ry (t; t) = E[jy(t)j ] =
2
Rg (0;  ) E[jx(t  )j ] d:
2
(11)
1

This holds for arbitrary signal.

5
 Case 1. Now an arbitrary bandwidth limited signal xp t ( ) = Pk x(kTp ) sinc((t ) )
kTp =Tp

is applied into the TDL model (6) with (9) assumed. Then the correlation

Ry1 (t; t) = E[jy (t)j ] is evaluated. As expected, the same result is obtained as
1
2

for the reference case

Z 1
Ry1 (t; t) = Rg (0;  ) E[jxp (t  )j ] d:
2
(12)
1

Case 2. Now performing the same as in previous case but with uncorrelated

coecients gives

Z 1 X
Ry2 (t; t) = Tp Rg (0; mTp )Æ( mTp )
1 m
 E[jxp (t  )j ] d:
2
(13)

In the next step we investigate an actual impact of the two presented cases of

the coecients properties on the channel model they represent. We use an band-

limited excitation xp1 t ( ) = sinc(t=Tp) and compare the response correlation

properties of the channel (12) and (13). A physical interpretation of these

two is similar to the power delay prole of the channel with the modication

such that the channel is excited not by Dirac delta impulse but by the band-

limited impulse ( ).
xp1 t Example of the above given results for particular case

of relatively short rectangular Rg (0;  ) is given on Figure 1.

5 Conclusions
We can clearly observe that the response of the channel model with uncorrelated

coecients does not match that of the correctly obtained ones. The channel

model with uncorrelated coecients in fact prolongs the channel response and

introduces clearly visible side-lobes. Such channel model when used e.g. for

the system simulation can substantially depart from the intended channel prop-

erties. This usually demonstrates itself as an unintended self-noise caused by

erroneous modeling. This phenomenon is mainly aecting channels with short

6
or suddenly changing (rectangular-like) power proles such that the approxi-

mation (10) does not hold. The only correct solution in such a case is to use

mutually correlated coecients given by (9).

6 Acknowledgment
This work was supported by the program #J04/98:212300014 sponsored by the

Ministry of education, youth and sports of the Czech Republic.

References
[1] J. G. Proakis, Digital Communications . McGraw-Hill, 1995.

[2] H. Meyr, M. Moeneclaey, and S. A. Fechtel, Digital Communication Re-

ceivers, Synchronization Channel Estimation and Signal Processing . John

Wiley & Sons, 1998.

[3] P. A. Bello, Characterization of randomly time-variant linear channels,

IEEE Trans. Commun. Syst. , vol. CS-11, pp. 360393, Dec. 1963.

7
 1.2

0.8

0.6

0.4

0.2

−0.2
−4 −3 −2 −1 0 1 2 3 4 5 6 7
t/Tp

Figure 1: Response R ( )
y1 t; t (dashed), R ( ) (solid) to the sinc(t=Tp ) excita-
y2 t; t
tion. Actual delay proleR (0; )
g t (dotted).

View publication stats