the effects of time delay spread
TRANSCRIPT
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 1/11
IEEE JOUR NAL ON SELECTED AREAS IN CO MMUNICATIONS, VOL. SAC-5, NO. 5 , JUNE
1987 879
The Effects of Time Delay Spread on Portable Radio
Communications Channels with Digital Modulation
Abstract-Frequency-selective fading caused by multipatb time de-
lay spread degrades digital commu nication channels by cau sing inter-
symbol interference, thus resulting in an irreducible
BER
and impos-
ing a upper limit on the data symbol rate. In this paper, a frequency-
selective, slowly fading channel is studiedy computer simulation. The
unfiltered BPSK, QPSK, OQPS K, and MSK modu lations are consid-
ered first to illustrate the physical insights and the error mechanisms.
Two classes of modu lation with spectral-shaping filtering are studied
next to assess the tradeoff between spectral occupancy and the perfor-
mance under the influence f time delay spread. The simulation is very
flexible so that different channel parameters can be studied and opti-
mized either ndividually or collectively. The rreducible BER aver-
aged over fading samples with a given delay rofile is used to compare
different modulation/detection methods, while the cumu lative distri-
bution of short-term BER s employed to show allowable data symbol
rates for given values f delay spread. It is found that both MSK and
QPSK withaised-cosineyquistulsereuitableor
TDM /TDMA digital portable commun ications channel.
I. INTRODUCTION
T
E emergence of a demand for personal portable
communications, along with a need for integrated
voice and data in our mobile society, has made universal
digital portable communications
[l]
highly desirable. A
candidate for providing a flexible, high-capacity digital
service is the TDM / TDMA (time-division multiplexing/
time-division multiple-access) channel allocation scheme
[l],
[2].
TDM architecture has several advantages over
the more conventional FDM (frequency-division multi-
plexing) such as simpler radio hardware and variable-rate
voice communications
[3];
however, the signaling rate re-
quired for such a system is higher. As a result, the influ-
ence of frequency-selective fading caused by delay spread
becomes a crucial issue
[4]-[6].
A flexible computer simulation is presented in this pa-
per to assess the effects of delay spread and to optimize
system parameters such as modulation/detection scheme
and data symbol rate. Adaptive equalization of the fading
channel is not assumed for this study.
Section I1 illustrates the channel model, and Section 111
describes the simulation. Numerical results for some un-
filtered modulations are shown in Section IV. In Section
V modulations employing spectral-shaping filters, as
would be used in a practical system are studied. Section
VI summarizes the results of the paper.
Manu script received January 14, 1987; revised February 17, 1987.
The author is with Bell Comm unications Research Incorporated, Red
IEEE
Log Number 8714459.
Bank, NJ 07701.
JUSTIN C-I CHUANG
11. MODEL
A multipath radio propagation channel can be described
mathematically by its impulse response
h
(
t ) .
A baseband
model of a multipath radio channel is shown in Fig.
1 .
A
quasi-static channel is considered in this study, that is,
the channel is assumed to be time-invariant over many
symbol periods. This is a reasonable assumption fora
portable communications system with around 1 GHz car-
rier frequency and several hundred kbits/s signaling rate,
since the maximum Doppler frequency is only on the or-
der of lop5 imes the bit rate.
Let t )be the baseband representation of the modu-
lated waveform. The total received waveform is then
z ( t ) = r ( t ) + n ( t ) ( 1 )
r ( ’ t )= u t ) *
h t ) . (2)
where n t ) s the additive Gaussian noise and
Intersymbol interference (ISI) caused by the delay spread
of h
( t ) results in the frequency-selective fading. With in-
creasing signal-to-noise ratio (SNR), an irreducible
“floor” of bit error rate (BER) is approached because IS1
increases in proportion to the signal level.
In this paper, the short-term, small-scale (“micro-
scopic”) signal variations due to multipath fading are
studied; therefore, the “macroscopic” effects such as
shadowing and distance-related attenuation are normal-
ized as unity. When an overall system design is consid-
ered, proper scaling factors should be incorporated in a
power budget to account for these long-term, large-scale
variations [ 1
The impulse response of the channel can be expressed
as
[71
h t ) = C
Arnej m6(t T m ) .
( 3 )
A general form of h t ) an be expressed as a continuous-
time function. Both
A ,
and 4rn re slowly varying random
quantities that introduce a small Doppler shift for the
communications channel.A commonly accepted model
[8], [9] suggests that
A m
be a random variable with a Ray-
leigh distribution and 4 be a random variable uniformly
distributed from 0 to
27r;
therefore,
h
( t ) s a zero-mean
complex Gaussian random variable. Physically, at a spe-
cific time delay, the received signal approaches a complex
Gaussian distribution according
to
the central limit theo-
m
0733-8716/87/0600-0879 01OO
O
1987 IEEE
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 2/11
880
IEEE JOURNAL ON
SELECT ED AREAS IN COMMUNICATIONS, VOL. SAC-5, NO.5.,JUNE
1987
Digital
Source
a k = ’
Bits
Encoderodulator
Wave‘orm
Transmitter
I
Receiver
Demodulator
Decoder
n(t)
Fig.
1 .
A baseband model for a portable radio communications channel.
rem because it is a combination of signals from a large
number of unresolved paths.
For t # t ’ ; it is reasonable to assume that h ( t ) and
h ( ’ ) are uncorrelated since they are composedof signals
from ndependent sets of paths with different Doppler
spectra (‘ ‘uncorrelated scattering” [9] . Consequently,
( h * ( t ) ( t ’ ) )
=
p ( t ) ( t ‘ ) (4)
where
(
) denotes the ensemble average, and
P O ) = ( l h ( t , l ’ > . 5 )
The function p ( t ) s a measurable profile [101 called the
“power delay profile.” Some measured results for p ( t )
in the portable radio environment are documented by De-
vasirvatham [ l], [121.
A measure of the width of p ( t ) s theroot mean-square
(rms) delay spread
T
defined as the square root of the sec-
ond
central moment
[9].
That
is
1
/*
( t D f p ( t ) t
T = [‘
j P ( t ) d t
]
( 6 )
where the average delay D , i.e., the centroid of p ( t ) , s
p ( t ) , d t
O ) d l *
D = 7)
The error rates for transmission through a channel with a
small delay spread are most strongly dependent on the
normalized rms delay spread [4],
[5],
[13]. Normalized
delay spread is defined as
d = T
where
T is
the symbol period. In this study,
d 0.2 is
considered.
111.
SIMULATION
Fig. 2 is the flowchart for coherent detection of the
modulations considered except GMSK. The simulations
rMP
NUMBER
Processes
Gausslan
A hin, hqn n=1,2
,...,
Impulse
Carrier
Recovery
1 (t)e-j
Receiver
Filterlng
fm(t)
INPUT
BITS
kO(k)<
0
I
Svmbol
t Error
BERER
Compu-istri-
tationsution
Fig. 2. The flowchart of a computer simulation
of
a frequency-selective
fading channel.
for GMSK and differential detection involve minor mod-
ifications to what is shown.
A . Sampling p ( t )
The powerdelay profilep ( ) s sampled at = tn where
n = 1 , 2 ,
,
N. The spacing between ime instants is
chosen according to the accuracy needed.
In addition to a measured profile [111 as’shown in Fig.
3 ,
three idealized profiles are considered:
a) Gaussian profile
b) One-sided exponential profile
1
exp { - t / 7 } for t 2 o
for t < 0 .
P ( t ) =
r;
(9b)
c) Equal amplitude two-ray profile
p ( t )
=
; [ a c t
T +
6 ( t + T ] . ( 9 4
Both
the Gaussian and he two-ray profiles are represented
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 3/11
CHUANG: EFFECTSFIME DELAYPREAD
ON
PORTABLE RADIO COMMUNICATIONS
881
V Y / L
Z .4 .6
B 1
1.2 1.4
1 E
Time
In
I ic rowwonds
Fig.
3 .
A
power delay profile obtained from measurements done in an
of-
fice building; the
rms
delay spread
is
approximately
250
ns.
by noncausal functions because the average delays are re-
moved for mathematical convenience.
The channel impulse response at
t,,
denoted by
h ( t , )
= hi,
+
jhqn , s a complex zero-mean Gaussian random
process with variance
p ( t , ) ;
therefore,
hi,
and
h,,
are un-
correlated Gaussian processes each with variance
P ( t , ) / 2 .
B. Generating Gaussian Random Processes and h ( t )
From (4), the covariance matrix for the 2N-dimension
Gaussian random “vector”
{ hi
,
hql
,
*
,
hiN hqN
can be expressed as
M =
P(tl)
2
P(t l )
2
0
0
A subroutine is called to generate the 2 -deviate Gauss-
ian random vectors with a covariance matrix
M .
A sample
of the channel impulse response can then be constructed
by the following formula:
N
h ( t ) = C (hi, + j h , , ) ( t , ) . (11)
n = l
C . Convolving the Modulated Waveform with h ( t )
All the modulations treated here except GMSK can be
simulated as quadrature amplitude modulation (QAM)
with a signaling waveform g ( ) of symbol period T.
GMSK is simulated as a form of frequency modulation
(FM).
1 UnJiltered BPSK , QP SK, O QPSK : Signaling with a
rectangular pulse,
Note that 1) BPSK is simulated by one input bit sequence;
2) QPSK and OQPSK are simulated by two independent
input bit sequences; and
3)
a time offset of T/2 offset is
introduced between the two sequences for OQPSK.
2) MSK: Signaling with a sinusoidal pulse,
g ( t > = s m 0
a t
={;F -T/2 /2
.
(12b)
otherwise.
Note that MSK is simulated in the same way as OQPSK
except with a different signaling waveform.
3) QPSK with raised-cosine Nyquist signaling pulse
7]
(RC-QPSK):
Signaling with a pulse with the following
form.
g ( t )
=
g c ( t )
( 1 2 4
where
g ,
( t ) can be realized by a filter that has the square
root of a raised-cosine spectrum with a roll-off factor CY.
This signaling waveform and its matched filter form an
ISI-free Nyquist pulse in the absence of delay spread.
4) GMSK: Signaling with a pulse with the following
form.
s ( t )= g p ( t ) g g ( t ) (12d)
where
g p
t ) s a rectangular pulse expressed by (12a) and
g g
t )
s the impulse response of a Gaussian filter whose
width is controlled by the BT, ( 3 dB bandwidth normal-
ized by bit rate) product
[
141.
The linear combination of binary pulses is used to fre-
quency-modulate the carrier, with the total phase change
for each pulse being a 1 2 .
With the exception of GMSK, the received waveform’
is a linear combination of g
( t )
*
h ( t ) .
D. C arrier Recovery for Coherent Detection
For a portable communications system using around 1
GHz for the carrier and several hundred Kbits/s for the
signaling rate, the following assumptions are reasonable.
a) The bandwidth of the carrier recovery circuit is as-
sumed to be much higher than the channel fading rate;
therefore, the carrier phase can be tracked as if h ( t ) s
time-invariant.
b) The bandwidth of the carrier recovery loop is much
lower than the symbol rate; herefore, for an idealized
carrier recovery circuit that removes the modulation, the
carrier phase is extracted from the average of h ( t ) over
several symbol periods.
Phase jitter on the recovered carrier caused by Gaussian
noise is neglected as we focus on the effects of delay
spread; herefore, he low SNR performance should be
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 4/11
882 IEEEOURNALNELECTED AREASN COMMUNICATIONS,
VOL.
SAC-5,O. 5 , J U N E
1987
viewed as the idealized case, which is reasonable because where
E b / N o
s average signal-to-noise ratio per bit at the
of b), above.
input of the receiver and “erfc” is the complementary
For
d 0.2,
he average of h
( t )
over several symbols
error function.
can be approximated by the average over the whole time
b) If O k ) d k
<
0, errors occur if AWGN causes no
axis; as a result, the recovered carrier phase is then errors; therefore,
h ( t ) dt = phasef H ( f ) l f = o (13) 1
P , ( k )
=
1 erfc
2
where H ( f is the Fourier transform of h
( t ) .
The recovered carrier can be represented by For differential detection of BPSK, sampled values of
cible BER. To compute the short-term BER when AWGN
(14) is present, the BER formula, ( 1 2 )
PEbINo,
s not exact
because differential detection is not a inear operation.
Carrier recovery is then treated
h
( t , However, we can get a reasonable approximation forhe
h ( t ) - j in theimulation. conditions assumed in this paper by multiplying the sig-
nal-to-noise ratio by I
k ) O k
1) I and then using
E. Receiveriltering the above BER formula. -The. easons ares follows:
For all the modulations considered except GMSK, the
a) This fOrmula is exact for Eb
/ N o
-
O
i.e. when
optimal receiver filter in the absence of delay spread is AWGN is not present; therefore, the irreducible BER is
the matched filter with impulse response
g*
( - t ) ; this is
curate and the short-term BER at high SNR is a good
the receiver filter assumed in the simulation. For GMSK,
approximation.
we simulate the type of receiver described by Murota and
b) This fOrmula is exact if k ) = I k 1 I be-
Hirada [14]; this is a parallel implementation of the MSK
cause the noise Processes are effectively changed by the
receiver, which is a suboptimal receiver for GMSK in the
Same ratio for the two bits. It Will be Seen from the next
absence of delay spread [151. section that IS1 is much smaller than the signal component
unless the signal is in a deep fade as aresult of multipath
F. Timing Rec ove rynd Sampling cancellation forhe range .0.02 .2 simulated;
therefore,
I
O(
k ) = I
O k 1
)
I in most cases. When
( k ) k )
is
very
1 r m( ) r z ( t T ) are computed to determine the irredu-
H (O )
= r ‘
A squaring timing loop
[
161 is used in the simulation to the signal is
in
a deep fade, I
recover
the
timing The timing
td caused by a Small, the BER is very large, the channel is unusable, and
multipath channel was analyzed in an earlier paper [17]. the exact value
of
the BER is of less interest.
The timing jitter caused by Gaussian noise is neglected;
Since
the of the delay-spread
effects
on
the
tion is a lower bound.
cible BER) performance, the low SNR region where the
therefore, the BER computed under the low SNR condi- two detection methods is based on the high SNR (irredu-
For small
d ,
the recovered timing tracks the centroid of approximation is less accurate does not effect the conclu-
I h
( t ) 2
Twocases Of timing are considered:
If
sions reached in the paper. The approximate BER calcu-
timing recovery loop,
p ( t )
is
the fading rate is much higher than
timing; whereas the short-term Ih
( t )
is used when the large.
fading rate is much lower than the bandwidth of the tim-
For
the
ing recovery loop. It will be seen later that the BER per-
is
computed.
fonnances are about the same forboth
the bandwidth of the lation for differential detection
should
be
used
with tau-
t’ generate detection tion, especially when SNR is low and the value of
d
is
of
GMSK, only the irreducible BER
cases f
d
is small.
For
all
the modulations treated except GMSK, ( k ) s
Once td
is
computed, the waveform
rm t ) a weighted sum of the samples off,
( t )
(coherent detec-
o ( k ) ,
tion) or
f m ( t )f t T )
(differential detection) where
s sampled at t = kT
+
td for the
kth
bit; the
is used for symbol detection.
f , ( t )
=
g ( t )
*
h ( t )
*
g * (
- t ) .
For example, O ( k ) can
G .
Symbol
Detection and BER Computation
be expressed in the following form for coherent QPSK,
The quantity k ) s normalized in such a way that it
has a 0 dB rms value in theabsence of delay spread.The o ( k ) = Re [
( d n l + j d n 2 ) f m [ ( k
n ) T +
t d ] ]
effect of AWGN is then included in the following ways
for the coherent detection: 16 )
a) If
O k ) d k 0,
errors occur if AWGN causes er-
.rors; the resulting short-term BER, P,
) ,
is
where .Re denotes the real part and
d n l , d n 2
= 1. It is
clear that
dklfm
td
) is the desired signal term in the kth
interval, while other terms cause intersymbol interfer-
ence. Interference between
I
and Q rails in general exists
m
n =
- m
P, )
=
erfc
i . . x i ) .
(I sa ) (“cross-rail nterference’’).
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 5/11
CHUANG:EFFECT S OF TIME DELAYSPREADON PORTABLE RADIOCOMMUNICATIONS
883
0.10
10-5
0.01 0.02 0.04 0.06 , 0.08
0.10
d
Fig. 4. Comparison
of
results
of
the simulationwith those obtainedby
analysis and experiment.
IV. RESULTS
OR
THE UNFILTERED MODULATIONS
In this section, results for the unfiltered modulations are
presented. Major parameters in the simulation are
1)
type
of modulation: BPSK, QPSK, OQPSK, and MSK; 2) type
of detection: coherent and differential;
3)
delay profile and
its normalized
rms
delay spread; 4) ratio of the bandwidth
B
of the timing recovery loop to the fading rate
F ,
and
5
the signal-to-noise ratio: Eb N o .
Simulations using different profiles indicate that the
BER performance is not sensitive to the shape of the delay
profile, for the range of d simulated. Examples involving
different profiles are shown in the following; however, the
results apply regardless of the profile used.
A . Verijication of the Simula tion with Results in the
Existing Literature
Bello and Nelin [4] calculated the irreducible BER av-
eraged over themultipath fading samples for differentially
detected BPSK in a channel with a Gaussian delay profile.
Y shida
et al. [
181 performed microscopic BER measure-
ments for coherent BPSK with a two-ray profile. Fig. 4
indicates that the results of the present simulations com-
pare well with the available analytical and experimental
results.
B .
The
Error
Mechanisms and the Efects
of
the Timing
Recovery Circuit
In Figs.
5
and
6 ,
two examples are shown to illustrate
the error mechanisms for small d ; a rectangular signaling
pulse with a matched receiver filter is considered. Both
figures indicate how three consecutive symbols combine
and interfere with one another to form the output
0
k ) at
the kth detection timing. The channel impulse response in
each example has two equal-amplitude “rays” which are
in-phase for Fig. 5 and out-of-phase for Fig. 6.
O k)=dkfm O)+dk.lfm T)+dk+lfm -T)= l-d)dk+0.5d dk+l+dk-l)
O(k)dk>O NO IRREDUCIBLE ERROR
Fig. 5 . Illustration
of
the
error
mechanism by using a two-ray model (equal-
amplitude and in-phase rays).
\
)LO.5d T
\
-1
O
1
T
+-I
k -4
2d
2dd
I
0.5d I
t
T
\
O(k)=0.5d(dk+l-dk-1)
P,(O(k)dk <
1 = 0.5
Fig. 6 . Illustration of the error mechanism by using a two-ray model (equal-
amplitude and out-of-phase rays).
The “.irreducible” errors (i.e., those made at very high
SNR)
cannot occur unless IS1 outweighs the signal com-
ponent at the sampling instant; therefore, there are three
error mechanisms: 1) A faded signal component caused
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 6/11
IEEE JOURNAL ON
SELECTED AREAS
IN
COMMUNICATIONS,
VOL.
SAC-5, NO. 5 , J U N E 1987
-40
-35
71J
-25 -20 -15 -10 -5 0 5
10
Signal
Level, dB
Fig. 7 . Amplitude distributions of the detected waveform w hen irreducible
symbol errors occur for coherent BPSK in a channel with a two-ray pro-
file. T he Rayleigh distribution results for the entire set
of h t ) .
by multipath cancellation; 2) IS1 caused by nonzero d;
and
3)
shift of sampling timing as a result of delay spread.
It
is
clear from
both figures
that,
for
small d,
1) IS1
causes
a very small perturbation at the tail end; and
2)
a small
shift of sampling timing has a negligible effect or t h e ir-
reducible BER. Therefore, the major error mechanism is
signal fading in this case. For example, the signal com-
ponent is much stronger than IS1 in Fig. 5 ; as a result,
irreducible errors will not occur. On the contrary, the sig-
nal component for Fig.
6
is in a deep fade and 0 k ) de-
pends only one two adjacent symbols; a BER of about 0.5
is expected because the output bit is uncorrelated with the
corresponding input bit.
It is indicated by,the simulation that the occurrence of
irreducible symbol errors is very bursty. For example, for
a channel with a Gaussian profile and d =
0.05,
a simu-
lation for differential detection of BPSK yields a
1.5
X
average irreducible BER. 1n.this simulation,only 10
out of
2000
samples of h ( t ) result in irreducible symbol
errors; in these
10,
the BER is very high.
To determine whether the burst of errors is a result of
envelope fading, cumulative distributions of the ampli-
tude of O
k )
when the irreducible symbol error occurs
are computed. The distributions shown in Fig. 7 corre-
spond to two-ray channels with d =
0.05,
0.1 and 0.2 for
coherent detection of BPSK, along with a Rayleigh dis-
tribution, which is the amplitude distribution of the entire
1
I
I I
I l l 1
NRZ, Two-Ray Profile
U/D
=
1 OO
Delay
d =
2T
10-1
O T
10-5
I I I
I 1 I I I
0.01
0.02
0.04 0 06 0.08 0.10
d
Fig. 8. The irreducible
BER
performance for BPSK with two kinds of de-
tection and two B / F ratios. (two-ray profile)
sample space of
h (
t ) with or without symbol errors). The
rms signal level for the Rayleigh-faded signal is 0 dB. All
cumulative distributions are computed as probabilities in
the
entire sample space. It is indicated that
O k s
al-
ways in a fade when irreducible symbol errors occur; for
example, the depth of fading is at east 15dB with respect.
to the
rms
signal level for d .1.
To determine the influence of timing error on the irre-
ducible BER, more simulations were performed. In the
case of a two-ray propagation environment, a comparison
of the average irreducible BER for a system using a very
fast squaring timing recovery circuit (i.e., loop band-
width
B
is much higher than the fading rate F ) with one
using a very slow timing circuit (i.e., B << F ) or BPSK
is shown in Fig. 8 . It shows that if delay spread is not
severe, a very fast timing loop improves the irreducible
BER performance only slightly. This suggests that timing
error is not the major mechanism for the bursty irreduci-
ble symbol errors. Simulations using other profiles indi-
cate the same result. In the following, only results for B
<<
Fa re shown.
In summary: 1) For small delay spread, envelope fad-
ing is the most important mechanism causing error bursts;
and
2
for severe delay spread, extrapolation
o f
Fig. 8
suggests that timing error could be a significant factor if
the timing recovery circuit is not fast enough.
The most significant implication o f this result is that
diversity selection can be effective in this case of small
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 7/11
CHUANG: EFFECTS OF TIMEELAYPREAD ON PORTABLE RADIO COMMUNICATIONS
885
+
BPSK
Coherent Detection
OQPSK
* Q p SK M o d u la s
]
X
MSK
10-4I
I
I I 1 1 1 1 I I I I I l l
10-2
d
Fig. 9. The irreducible BER performance for different modu lations with
coherent detection for a channel with a Gaussian profile. Th e parameter
d is the rms delay spread normalized by symbol period .
delay spread because an irreducible symbol error rarely
occurs fora diversity branch that has a high received
power.
C. Comparison
of
Modulation and Dete ction Methods
Fig. 8 indicates that coherent detection performs better
than differential detection; we shall focus the discussion
on coherent detection.
Fig.
9
shows the average irreducible BER as functions
of
d
for different unfiltered modulation methods with co-
herent detection; the multipath channel is simulated by
using a Gaussian delay profile. This figure indicates that
the delay spread performance of various unfiltered mod-
ulations when normalized to the same symbol period is
ranked in the following order: 1) BPSK, 2 ) QPSK, 3)
OQPSK, 4) MSK. The performance of BPSK is the best
because cross-rail interference does not exist. Both
OQPSK and MSK have a
T / 2
iming offset between two
bit sequences, hence the cross-rail IS1 is more severe;
therefore, their performances are inferior to that of QPSK.
In Fig. 9, the normalization factor for parameter d is
the symbol period T , during which two bits are transmit-
ted over the channel for QPSK, OQPSK, and MSK, while
only one bit is sent for BPSK. A fairer comparison of
performance for the same information capacity should be
based on
d’ =
r
/ Tb
where
r
is the rms delay spread and
Tb is the bit period. Fig. 1 0 is the same set of functions
as those in Fig. 9 plotted against
d ’
. When this normali-
10-1
I I
I I
I
I l l
I
I I
I I
I l l
Coherent Detection
I
PSK 1
OQPSK
* QPSKodulation
X MSK
d’ = rms delay spread
bitper iod
Fig. 10. Thesame set of curves as in Fig. 9, plottedagainstrmsdelay
spread normalized by bit period.
zation is applied, it is clear that 4-level modulations
(QPSK, OQPSK, and MSK) are more resistant to delay
spread than BPSK for constant information throughput.
Higher level modulations were also considered. For ex-
ample, Fig.
11
indicates that the performance of 8-PSK
as SNR approaches infinity is not superior to that of QPSK
even though it derives 3 bits per symbol. Since higher
level modulations are less efficient than 4-level modula-
tions at low SNR, we shall concentrate on 4-level modu-
lations in the next section.
It is also interesting to note that all the curves shown in
Figs. 9,
10,
and
11
are nearly parallel to a straight line
o f
slope
2 ;
that is, an order of magnitude increase in delay
spread results in about two orders of magnitude increase
in the irreducible BER within the range of
d
simulated.
By showing that the group delay is a Student’s
t
distri-
bution with two degrees of freedom, Andersen
et al. [13]
have proved that the irreducible BER caused by fre-
quency-selective fading is proportional to
d 2
when d is
small and that the proportionality constant depends on the
method of modulation and detection. The results of the
simulation are consistent with this earlier result.
D . Cumulative Distribution of BER
Because symbol errors are very bursty, it is important
to predict the cumulative distribution functions (cdf) of
BER, namely, the probability that short-term BER per-
formance is worse than a certain value, say,
lop3.
Fig.
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 8/11
10.4
I I I I
I I l l
I I I I 1 1 1 1
10-2
10-1 1 0
d ’= ms delay spread
bit eriod
Fig . 11. The irreducible BER performance
for
QPSK and 8-PSK . The pa-
rameter
d ‘
is the rms delay spread normalized by bit period.
12 is aypical cdf plot which includes a set of distribution
curves for BPSK with two detections (coherent and dif-
ferential) simulated by using a Gaussian profile with d
=
0.08.
In
the
example
set
of
curves,
the
average
& , / N o
(sometimes called the ‘‘local mean”) is varied from 0 dB
to 40 dB in 5 dB steps so that a change of signal level due
to shadow fading and other large-scale variations can be
accounted for.
For each cdf plot, the abscissa indicates a better BER
performance in the left-hand part of the figure while the
ordinate indicates a more reliable coverage in the lower
part of the figure; therefore, the curve appearing in the
left-most and lowest position represents the “best” com-
bination. Both axes are expressed in a log scale. It is easy
to see from Fig. 12 that, as expected, 1) coherent detec-
tion
is
better than differential detection and 2) as SNR
increases, the performance gets better; however, dimin-
ishing returns are observed as the “irreducible BER” is
approached.
V. RESULTSFOR MODULATIONS
ITH
SPECTRAL-
SHAPING FILTERS
Two classes of spectrally efficient modulation methods
that have been considered for practical applications in
mobile and satellite communications are considered: 1)
Gaussian-filtered minimum shift keying (GMSK)
[
141 and
2) QPSK with a raised-cosine Nyquist pulse (RC-QPSK)
1191.
102
C: Coherent Detection
D Differential D
1o- I l I 1 1 1 1 1 1 1 1 1 I 1 1 1 1 1 1 1 “ 1 1 1 1 1 1 1 l l f i l l l l l
I
‘ l u L L J
10.5
10.4 10-3 10-2
10-1
100
BER
Fig.
12.
BER distr ibutions
for
BPSK in a channel with a Gaussian profile
d
=
0 . 0 8 .
A .
General
Similar to the unfiltered modulations in Section IV, we
find that: 1 ) Timing jitter caused by small delay spread is
not crucial; 2) coherent detection is he more desired
choice;
3
the
BER
performance
is
relatively
insensitive
to the shape of the delay profiles, but this sensitivity in-
creases with
d ;
and 4) the irreducible BER increases about
two orders of magnitude as
d
increases from 0.02 to 0.2.
A measured power delay profile with about 250 ns rms
delay spread as shown in Fig. 3 was used to simulate a
real-world portable communications channel. Only re-
sults for coherent detection will be shown.
B. RC-QPSK
Fig.13 shows the average irreducible BER perfor-
mance of RC-QPSK as a function
of
roll-off factor
a ,
along with that for unfiltered QPSK, for different values
of
mis
delay spread.
As
a
increases, the irreducible BER
for a given value of d decreases monotonically due to de-
creasing ISI;however, he spectral occupancy also in-
creases. It is interesting to note that RC-QPSK with a 2
0.75 is more resistant to delay spread than the unfiltered
NRZ-QPSK.
C. GMSK
Fig. 14 shows the BER performance as a function of
the BTb product of the GMSK premodulation filter, along
with that for unfiltered MSK modulation. It is found that
the best BER performance is achieved by choosing BTb =
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 9/11
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 10/11
Coherent
a
=
0.25
RC-QPSK
EdNo
=
30
dB
-e d’ =
0.40
o d’
=
0.20
+
d‘=0.10
A
d ‘=0 .04
10-1 1
J
1 0 - 5 10-4 10-3 10-2
10-1
1
BER
Fig. 15. D etermining the maximum al lowable s ignal ing rate by using a cdf
plot . The parameter d’ s the rms delay spread normalized by bit period.
higher signaling rate is possible with even better perfor-
mance if diversity is implemented.
VI.CONCLUSION
A flexible simulation for evaluating the BER perfor-
mance of a frequency-selective, slowly fading digital ra-
dio channel has been described. Results from the simu-
lation for normalized delay spread in the range of 0.02
d 0.2 are consistent with the following performance
characteristics:
a) Only a small fraction of. the multipath channel im-
pulse responses encountered will exhibit “irreducible” bit
errors; however, once channel conditions which cause er-
rors occur, the resulting short-term BER is very high.
b) The major error mechanism is envelope fading; the
degradation due to timing error caused by a small delay
spread is not significant.
c) Since the irreducible errors occur only when the sig-
nal is in a deep fade, diversity will be an effective way to
lower the irreducible BER or to permit higher transmis-
sion rates for a given delay spread.
d) The BER performance is more sensitive to the rms
value of the delay spread than to the shape of the delay
profile. By measuring the rms delay spread, both symbol
rate and the modulation/detection scheme can be chosen
using the simulation. A power-of-two dependence on the
rms delay spread for the irreducible BER is found.
e) The BER averaged over fading samples of a given
delay profile provides a comparison among different mod-
ulation/detection schemes; cumulative distributions can
be used to determine the allowable signaling rate.
f ) Coherent detection is more resistant to delay spread
than is differential detection.
g
4-level modulation yields greater information rates
for a given delay spread than does 2-level modulation.
h) GMSK with BT,
=
0.25 is near optimum for resis-
tance to delay spread.
i)RC-QPSK performance degrades monotonically as
the roll-off factor is reduced from 1 to 0.
j) It is possible to transmit a few hundred kbits/s using
a TDM /TDMA architecture in a typical portable, radio
environment without diversity or equalization.
ACKNOWLEDGMENTS
The author would like to thank D. C. Cox,
P.
T. Por-
ter, and H . W . Arnold, for theirdiscussions and guidance
during the course of this work. Special thanks are due.to
D. Devasirvatham for providing a delay profile measured
by him to be used in the simulation. Finally, the author
is grateful to theanonymous reviewers for their comments
that improve the quality of this paper.
REFERENCES
[l]
D. C. Cox, “U niversal por table radio com municat ions ,” ZEEE Trans.
Veh. Techno l . , vol . VT-34, pp. 117-121, Aug. 1985.
121 K. Raith, J Stjernvall, and
J
Uddenfeldt , “Mult ipath eq ual izat ion
for digital cellular radio o perating at 3 00 K bit/s,” n Proc. 19861EEE
VT Conf . , May 20-22, 1986.
tablecommunicat ions:Anappl ied esearchperspect ive,” n Pro c .
ICC’8 6 , June 22-25, 1986.
141 P.A .Bel loand B. D .Nelin, “The effects of frequency elective
fading on the binary
rror
probabilities of incoherent and differentially
coherent ma tched filter receivers,” IEEE Trans. Commun. Syst . , vol .
CS-11, pp. 170-186,
June
1963.
151 B . Glance and L. J Greenstein, “Frequency-select ive fading effects
in digi tal mobi le radio with divers i ty combining,” EEE Trans. Com-
m u n . , vol . COM-31, pp. 1085-1094, Sept . 1983.
161
J H .
Winters and Y . S Yeh, “On the performance of w ideband dig-
ital adio ransmissionwithinbuildingsusingdiversity,” n Pro c .
GLOBECOM’8 5 Co n , Dec. 1985.
[7]
J
G .Proakis , DigitalCommunications. NewYork:McGraw-Hil l ,
1983.
[SI W. C . Jakes , Ed. , Microwave Mobile C ommunications. New
York:
Wiley,1974.
[9] P. A. Bello, “Characte rization of random ly time-va riant linear chan-
nels,” IEEE Trans. Commun. Syst . , vol . CS-11, pp. 360-393, Dec.
1963.
[lo] D. C. Cox and R. P. eck, “Correlat ion bandwidth and delay spread
multipathpropagation tatistics for910MHzurbanmobile adio
channels ,”
IEEE Trans.
Cornmun., vol .COM-23,pp.1271-1280,
Nov.1975.
11 11 D . M. J Devasirvatham, “Time delay spread measurements f wide-
band radio s ignals within a bui lding,”
Electron.
Let t . , vol . 20, pp.
950-951, Nov. 1984.
1121 D . M.
J
Devasirvatham , “Time delay spread and s ignal level mea-
surements of 850 MHz radio wavesn bui lding environments ,” IEEE
Trans. Antennas Propagat. , vol . 34, pp. 1300-1305, Nov. 1986.
[13] J
B .
Andersen, S. L. Lauri tzen, and C. Th ommesen, “Stat is t ics of
[3]
D. C .
Cox,
H .
W .
A r n o l d ,
and
P. T.
Porter
“ U ni v e r s a l d i g i t a l
por-
8/10/2019 The Effects of Time Delay Spread
http://slidepdf.com/reader/full/the-effects-of-time-delay-spread 11/11