ISSN 1392-2114 ULTRAGARSAS (ULTRASOUND), Vol. 63, No. 4, 2008.

30
Spread spectrum signal performance investigation in a bandlimited channel
L. Svilainis, G. Motiejunas
Signal processing department, Kaunas University o f Technology,
Studentu str. 50, LT-51368 Kaunas, Lithuania, tel. +370 37 300532, E-mail.:[email protected]

Abstract
The paper investigates performance of spread spectrum signals performance in a bandlimited channel. Available spectrum
spreading techniques are discussed. The performance is evaluated as the time of flight (ToF) accuracy estimation using the direct
correlation technique. Numerical simulations have been carried out. Ultrasonic transducer was approximated as a bandlimited channel
for simulation. The influence of ToF estimation with a constant chirp center frequency spreading and continuously rising chirp center
frequency where analyzed. Random errors of ToF and how effective bandwidth of chirp signal is affected by a bandlimiting channel
where evaluated. It has been shown that when chirp sweep frequencies are around the constant center frequency then the ToF deviation
is insignificant. When the upper chirp sweep frequency is increasing, the random errors are reduced. The results showed that the
optimum chirp signal bandwidth can be beyond the channel bandwidth.
Keywords: Spread spectrum signals, time-of-flight estimation, ultrasonic signal processing.

Introduction
Positioning and navigation systems are used in a
variety of applications. Location estimation is essential in
order to implement autonomous robot navigation [1].
There are several location estimation methods. The
location estimation techniques used in navigation can be
assigned to triangulation, trilateration and the combination
of the two. The trilateration technique is using the time-of-
flight (ToF) for lateral distances estimation [2, 3].
From a radar theory the time standard deviation
estimation is defined using the Cramer Rao lower bound
[4,5]. This relation suggests that there are two possible
ways to minimize the uncertainty of the ToF estimation
[6]: i) by maximizing the signal energy; ii) by maximizing
the effective signal bandwidth. Many authors investigate
the case when the ultrasonic signal spectrum is matched to
the passband shape of the ultrasonic transducer. The aim of
this paper was to investigate the behavior of spread
spectrum signals when their spectrum is not matched to the
ultrasonic transducer passband shape.
The ToF estimation precision
As suggested in [4] and [5] the ToF estimation can be
done by the direct correlation maximization to find and
estimate the true position of the signal arrival. The
variance of the ToF standard deviation is given by [7]:
( )
0
2
2
1
N
E
F
TOF std
e
CRLB

, (1)
where E is the signal s
T
(t) energy, F
e
is the effective
bandwidth of the signal. The effective signal bandwidth
can be calculated as:

2
0
2 2
f F
e
+ = , (2)
where is the envelope bandwidth and f
0
is the center
frequency:
( ) ( )
E
df f S f f

=
2 2
0
2
,
( )
2
2
2
2
0
E
df f S f
f

. (3)
Eq. 1 suggests that increasing effective signal
bandwidth the ToF variation decreases linearly while
increasing the signal energy is influencing the ToF
variation by the square root. Therefore an assumption can
be made that in a bandlimited channel it is more efficient
to increase the effective signal bandwidth despite energy
losses.
The spectrum spreading
Reduction of random errors is possible by maximizing
the signal energy, reducing the noise level and increasing
the effective signal bandwidth as Eq. 1 suggests. Since a
noise level has some physical limits it is more feasible to
increase only the energy and the bandwidth. The increase
of both is possible only if spread spectrum signals are used.
The spread spectrum technique allows having a long
duration pulse and the desired bandwidth simultaneously.
In order to have the pulse compression at the reception
end, the signal must be passed through a matched filter,
which corresponds to a direct correlation processing. In
addition, the pulse compression allows reducing the non-
singularity which is present in long signals. The example
in Fig.1 demonstrates the peak area of the correlation
function of the continuous wave (CW) burst.

-40.0 -20.0 0.0 20.0 40.0
-2000
-1000
0
1000
2000
X
c
o
r
r
,

A
.
U
time,s

Fig. 1. The matched filter output of CW burst
ISSN 1392-2114 ULTRAGARSAS (ULTRASOUND), Vol. 63, No. 4, 2008.
31
Though the output looks quite steep, the tip zoom in
showed on the top right corner is very flat. Even the
moderate noise level will cause the peak detection errors
which are not accounted in Eq. 1 and random errors will
increase significantly. The Fig. 2 is presenting the same
temporal window of the same length signal with the only
difference that the linear frequency modulation is applied
which gives a significant spread of signal spectrum.

-40.0 -20.0 0.0 20.0 40.0
-600
-400
-200
0
200
400
600
800
1000
X
c
o
r
r
,

A
.
U
time,s

Fig. 2. The matched filter output of spread-spectrum burst

It can be seen that peak steepness is increased. Several
techniques can be applied for to spread the signal
spectrum: phase manipulated sequences, chirp and
arbitrary waveform excitation.
The phase manipulated sequences (Fig.3) make the
excitation task easer since square waves can be used for
the signal generation. Square waves make the excitation
hardware less complex because a power amplifier may be
replaced by a high speed switch. Application of the
orthogonal coded sequences allows easy separate the
probing channels [1] so simultaneous surrounding area
scanning with several ranging channels can be done.

chip

Fig. 3. Phase manipulated sequence
In [8] authors were trying to fit the Golay sequences
spectrum into the bandwidth of an ultrasonic transducer:
the signal bandwidth was either narrower or wider than
that of the ultrasonic transducer. The problem is that the
resulting signal bandwidth is predicted by the smallest chip
(Fig. 3) duration. If the integer number of half-periods
must be used for chip this is a disadvantage since then the
bandwidth adjustment step is discrete. Another
disadvantage is that there is no possibility to shape the
signal spectrum in the arbitrary way.
Application of the arbitrary waveform produces any
shape of amplitude and phasing spectrum and correlation
function. But this technique is used rarely as this technique
requires complicated excitation hardware. There are
attempts to use some approximation using only limited
number or excitation levels. Publication [9] is reporting a
quinary excitation method for linear frequency modulation
windowing implementation.
Another frequency spreading technique is using a
carrier frequency modulation [10, 11] and frequently is
addressed as a chirp signal (Fig. 4).

0.0 5.0 10.0 15.0 20.0 25.0 30.0
-1.0
-0.5
0.0
0.5
1.0
A
m
p
l
i
t
u
d
e
,

V
Time, s

Fig. 4. Chirp signal
Chirp signal excitation offers any spectral shape and
easy excitation circuit and looks an attractive solution. In
[10] an application of chirp signals for air-coupled non-
destructive testing is reported. The combination of chirp
signals and wideband micromachined capacitance (CMUT)
transducers allowed compensation of the signal to noise
ratio (SNR) losses present in air-coupled ultrasound
applications. Authors did not care nor accounted for
transducer and signal bandwidth matching. The signal
bandwidth was chosen to be wider than the transducer
passband. In publication [11] authors are using a nonlinear
frequency modulation of a square wave signal in order to
produce the signal spectrum which corresponds to
ultrasonic transducer bandwidth shape. But there are no
publications which investigate the case when the excitation
signal bandwidth is wider than the ultrasonic transducer
pass band. As it was indicated earlier, Eq. 1 suggests that
using such signal should increase the effective signal
bandwidth. Of course, the penalty will be the reduction of
a signal energy passed through the filter. Since the ToF
variation decreases linearly with the effective signal
bandwidth increase and the ToF variation increase due to
energy losses is by the square root, one could expect that
there still should be a ToF variation reduction gain.
Therefore an assumption was made that in a bandlimited
channel it is more efficient to increase the effective signal
bandwidth despite energy losses.
The numerical experiment setup
In order to check the aforementioned assumption it
was decided to make a numerical experiment in order to
evaluate how the effective bandwidth (F
e
) of the chirp
signal is affected by a bandlimiting channel. The main two
questions where risen:
i) How effective bandwidth of the chirp signal is
affected by bandlimiting?
ii) How random errors of the ToF will behave in a
bandlimited channel?
The numerical simulations have been carried out in
order to evaluate the influence of a bandlimited channel on
a TOF estimation performance. The signal has been
simulated as the linear frequency modulation burst.
ISSN 1392-2114 ULTRAGARSAS (ULTRASOUND), Vol. 63, No. 4, 2008.
32
( ) t t f f rect t s
d pulse T
) ( 2 cos ) ( ) (
1
+ = , (6)
where
pulse
is the chirp signal duration, f
1
is the start
frequency, f
d
is the deviation frequency. It should be noted
that the upper frequency of the resulting frequency
spectrum is twice higher the deviation because of the
frequency derivative effects. Therefore a corresponding
normalization of the deviation frequency was used. The
ultrasonic transducer simulated had the passband of
100 kHz to 200 kHz. The simulation has been carried out
in MATLAB. The noise has been simulated using randn
function, which assumes the additive white Gaussian noise.
Random errors of the ToF have been extracted by taking a
large number of runs (10,000) and calculating the standard
deviation of the ToF value estimated.
Two cases were investigated: a) with sweep
frequencies centered around constant center frequency, b)
when upper chirp sweep frequency was increasing.
Three types of experiments in MATLAB for the ToF
estimation have been carried out:
1) unfiltered measurement signal and reference;
2) unfiltered reference, but filtered measurement
signal;
3) filtered measurement signal and reference.

Fig. 5. Numerical experiment setup 1: unfiltered signal and reference

Fig. 6. Numerical experiment setup 1: filtered signal

Fig. 7. Numerical experiment setup 1: filtered signal and reference
The goal of the numerical simulation was to reveal the
influence of the channel bandwidth on the ToF estimation.
The spectral response of the chirp signal after passing the
bandlimited channel is presented in Fig.8.

0.0 50.0k 100.0k 150.0k 200.0k 250.0k 300.0k
A
m
p
l
i
t
u
d
e
,

A
.
U
.
Frequency, Hz

Fig. 8. The power spectral density of the simulated chirp signal
It should be noted that the signal energy was kept
constant by keeping the same signal amplitude. Such
restriction allows to separate the filter effects on the ToF
variation.
Sampling influence minimization
Sampling the analog signal introduces convolution in
the frequency domain with a period of the sampling
frequency f
s
. The aliasing will occur for any frequency
component (both signal and noise) falling outside the f
s
/2.
This region between zero and the f
s
/2 is named the Nyquist
zone. The antialiasing filter must be used in order to avoid
aliasing [6]. Furthermore, aliasing of the noise will
decrease the SNR. The sampling frequency was chosen
high enough so that the antialiasing filter did not have
influence on experimental results. The sampling frequency
selection was based on the initial investigation aim of
which was to establish the sampling frequency where
effect of aliasing is minimized (Fig. 9).

200k 400k 600k 800k 2M 4M 6M 8M
100.0p
150.0p
200.0p
250.0p
300.0p
350.0p
400.0p
450.0p
500.0p
En
Tn
s
t
d
(
T
O
F
)
,
s
Sampling frequency, Hz

Fig. 9. ToF deviation versus the sampling frequency
The sampling frequency was chosen 2 MHz, so that
aliasing does not have influence on experimental results
and the ToF prediction using Eq. 1 and experimental
results are the same.
Results for constant center frequency
The first case experiments were carried out with the
chirp spectral response centered at the constant frequency
f
0
. The transducer has been simulated using the I-st order
Butterworth filter with a bandpass filter, the cut off
frequencies of which were from 100 kHz to 200 kHz. The
center frequency f
0
was 150 kHz and the requested chirp
signal bandwidth span was from 30 kHz to 300 kHz (Fig.
10). The chirp signal duration was 2 ms.
In Fig.10 f
c1
and f
c2
denote the filter cut off
frequencies. The lines f
1
and f
2
show the lowest and the
highest chirp signal spectrum frequency.
Fig. 11 shows the resulting F
e
changes and the
constructing components: the envelope bandwidth and
the center frequency f
0
.

+
XCorr Max Interp. TOF
Chirp
gen.
f1, f2
fc1, fc2
unRMS
Noise
x 10000 Std (TOF)
+
XCorr Max Interp. TOF
Chirp
gen.
f1, f2
fc1, fc2
unRMS
Noise
x 10000 Std (TOF)
+
XCorr Max Interp. TOF
Chirp
gen.
f1, f2
unRMS
Noise
x 10000 Std (TOF)
ISSN 1392-2114 ULTRAGARSAS (ULTRASOUND), Vol. 63, No. 4, 2008.
33
50.0k 100.0k 150.0k 200.0k 250.0k 300.0k
0.0
50.0k
100.0k
150.0k
200.0k
250.0k
300.0k
Chirp center frequency
Filter cutt off frequencies
f
2
f
1
f
c1
f
c2
R
e
s
u
l
t
i
n
g

m
a
r
g
i
n
a
l

f
r
e
q
u
e
n
c
i
e
s
,

H
z
Requested sweep bandwidth, Hz

Fig. 10. Chirp signal: f
1
and f
2
chosenm so f
0
is constant

0.0 50.0k 100.0k 150.0k 200.0k 250.0k 300.0k
0.0
50.0k
100.0k
150.0k
200.0k
f
0
F
e

R
e
s
u
l
t
i
n
g

f
r
e
q
u
e
n
c
i
e
s
,

H
z
Requested sweep bandwidth, Hz

Fig. 11. Filtered Chirp signal f
0
, beta and F
e

From Fig. 11 it is seen that if f
0
is kept constant, the
envelope bandwidth is lower than the center frequency f
0
,
the resulting increase of the effective bandwidth F
e
is
insignificant. It can be expected that the resulting ToF
variation decrease should also be negligible (Fig. 12).

0.0 50.0k 100.0k 150.0k 200.0k 250.0k 300.0k
200.0p
210.0p
220.0p
230.0p
240.0p
s
t
d
(
T
O
F
)
,
s
Requested sweep bandwidth, Hz
Theory
Experiment

Fig. 12. Unfiltered chirp signal performance
The results confirm the assumption above: if the
increase of F
e
is small, then the total reduction of the
standard deviation of ToF is also low.
Another two experiments were aimed to account for
the ultrasonic transducer bandlimiting influence. This was
accomplished by applying the filter on the measurement
signal. Type 2 experiment used the unfiltered reference,
but measurement signal was filtered. Type 3 experiment
used both the filtered measurement signal and the
reference. The obtained results are presented in Fig.13. The
experiment results indicate that there is no improve of the
ToF variance. This can be explained by the results on Fig.
11 and 12: the increase of the effective bandwidth F
e
is
insignificant even for unfiltered case.
Furthermore, there is and increase of the ToF variance.
This increase is more significant for the reference signal
having a wider bandwidth (unfiltered reference). We
explain this by a wider bandwidth of the matching wilter
which in turn gives a higher level of a penetrating noise.
0.0 50.0k 100.0k 150.0k 200.0k 250.0k 300.0k
200.0p
250.0p
300.0p
350.0p
s
t
d
(
T
O
F
)
,
s
Requested sweep bandwidth, Hz
Theory
UR - unfiltered reference
F - full filtering

Fig. 13. Filtered chirp signal performance
The graph in Fig.14 is used to demonstrate the time
shift induced by not even filtering of the reference and the
measurement signals. The cross correlation function was
normalized by the signals mutual energy geometric
average (MATLAB option coeff).
-1.5 -1.0 -500.0n 0.0 500.0n 1.0 1.5
0.0
0.2
0.4
0.6
0.8
1.0
Signal filtered,
Reference NOT filtered
Signal NOT filtered,
Reference NOT filtered
Signal filtered,
Reference filtered
C
r
o
s
s

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

m
u
t
u
a
l

e
n
e
r
g
y

n
o
r
m
a
l
i
s
e
d
Time, s

Fig. 14. The correlation function peak area for all three experiments
The conclusion can be drawn that the reference signal
should be filtered in the bandlimited correlation processing
in order to get the optimal results.
Results for increasing center frequency
The previous experiments indicated that the envelope
bandwidth should be comparable to the center frequency
ISSN 1392-2114 ULTRAGARSAS (ULTRASOUND), Vol. 63, No. 4, 2008.
34
f
0
in order to have a significant influence on the resulting
increase of the effective bandwidth F
e
, so the resulting in
ToF variation decrease.
A new set of experiments was aimed to investigate the
case when the upper chirp sweep frequency was increasing
so the center frequency was no longer the same and it was
increasing. With the starting sweep frequency f
1
held at
100 kHz, the upper limit f
2
of the sweep was increasing.
The increase of the center frequency suppose to cause the
increase of the effective bandwidth F
e
resulting in the ToF
variation decrease. Fig.15 is used to demonstrate the effect
of such modulation on the envelope bandwidth and the
center frequency f
0
and the effective bandwidth F
e
change.

0.0 50.0k 100.0k 150.0k 200.0k
0.0
50.0k
100.0k
150.0k

f
0
F
e
R
e
s
u
l
t
i
n
g

f
r
e
q
u
e
n
c
i
e
s
,

H
z
Requested sweep bandwidth, Hz

Fig. 15. Filtered Chirp with f
1
=100 kHz, f
2
increasing
Again, despite the increase of the envelope bandwidth
its influence on the effective bandwidth F
e
is negligible
since the center frequency f
0
is higher. But here there is an
increase of the center frequency f
0
and the resulting
effective bandwidth F
e
change is significant. The resulting
the ToF variation can be examined in Fig. 16.

0.0 50.0k 100.0k 150.0k 200.0k
200.0p
250.0p
300.0p
350.0p
s
t
d
(
T
O
F
)
,
s
Requested sweep bandwidth, Hz
Theory, I-ord Butterworth
Experiment,I-ord Butterworth
Theory, II-ord Butterworth
Experiment, II-ord Butterworth

Fig. 16. Filtered Chirp performs better
A clear optimum can be located here. In addition, there
is a nice match between the theory prediction by Eq. 1 and
the numerical experiment results. The optimum is located
at the filter passband (refer to Fig. 10). The higher order
filter (steeper transition band) has optimum in a higher
span. For it there is an optimum which is beyond the filter
bandwidth.
Conclusions
It has been assumed that for spread spectrum signals
there should be some bonus in ToF random errors
reduction thanks to the effective bandwidth increase even
if the chirp spectrum exceeds the filter bandwidth. The
investigation revealed that if chirp sweep frequencies are
centered on a passband center frequency, then the effective
bandwidth gain (so the TOF deviation) is insignificant.
Numerical experiments indicate that if the chirp
spectrum is shifted towards higher frequencies, then there
is some gain in the effective bandwidth, therefore the
random ToF estimation errors are reduced. There is an
optimum in the case of a higher order filter which is
beyond the filter bandwidth.
References
1. Kays R. J., Svilainis L., Maeika L. Application of orthogonal
ultrasonic signals and binaural processing for imaging of the
environment. Ultrasonics. 2000. Vol. 38 P. 171-175.
2. Roh H., Han J., Lee J. et .al. Development of a new localization
method for mobile robots. IEEE Systems and Control conf. proc.,
Texas, USA. 2008. P.383-388.
3. Tonga F., Tsoa S. K., Xub T. Z. A high precision ultrasonic docking
system used for automatic guided vehicle. Sensors and Actuators.
2005. Vol. 118. P.183189.
4. Rao C. Information and the accuracy attainable in the estimation of
statistical parameters. Bulletin of Calcutta Mathematics Society.
1945. Vol. 37. P. 8189.
5. Cramer H. Mathematical methods of statistics. Princeton, NJ:
Princeton Univ. Press. 1946.
6. Svilainis L., Dumbrava V. The time-of-flight estimation accuracy
versus digitization parameters. Ultragarsas. 2008. Vol. 63. No.1.
P.12-17.
7. Minkoff J. Signal processing fundamentals and applications for
communications and sensing systems. Norwood, MA, USA: Artech
House. 2002.
8. Nowicki A., Trots I., Lewin P.A. et. al. Influence of the ultrasound
transducer bandwidth on selection of the complementary Golay bit
code length. Ultrasonics. 2007. Vol.47. P.64-73
9. Cowell D. M. J., Freear S. Quinary excitation method for pulse
compression ultrasound measurements. Ultrasonics. 2008. Vol.48.
No.2. P.98-108.
10. Gan T. H., Hutchins D. A., Billson D. R., Schindel D. W. The use
of broadband acoustic transducers and pulse-compression techniques
for air-coupled ultrasonic imaging. Ultrasonics. 2001. Vol. 39. No.3.
P.181-194
11. Pollakowski M. and Ermert H. Chirp signal matching and signal
power optimization in pulse-echo mode ultrasonic nondestructive
testing. IEEE transaction on ultrasonics, ferroeletrics and frequency
control. 1994. Vol. 41(5). P.655-659.
L. Svilainis, G. Motiejnas
Skleisto spektro signal efektyvumo ribotos juostos kanaluose
vertinimas
Rezium
Buvo tiriamas skleisto spektro signal panaudojimas ribotos juostos
kanaluose, aptarti galimi spektro ipltimo metodai. Skleisto spektro
signalo efektyvumui vertinti naudotas signalo sklidimo laiko vertinimo
tikslumas, apskaiiuotas tiesiogins koreliacijos metodu. Straipsnyje
pateikiami skaitmeninio modeliavimo rezultatai. Modeliuojant
ultragarsinis keitiklis buvo interpretuojamas kaip ribotos pralaidos juostos
kanalas. Nagrinta, kaip sklidimo laiko nustatymo tikslum veikia
skleisto spektro signalai, pereinantys per ribotos pralaidos juostos kanal.
Analizuoti du atvejai: i) kai signalo juosta pleiama abi puses apie
centrin dan taip, kad spektro svorio centras nekist; ii) kai pleiamo
signalo spektro svorio centras slenka auktesni dani pus. Nustatyta,
kad pleiant skleisto spektro signalo juost simetrikai apie centrin dan,
signalo sklidimo laiko nustatymo atsitiktins paklaidos kinta maai. Jei
pleiamo signalo spektro svorio centras slenka auktesni dani pus,
tada atsitiktins sklidimo laiko nustatymo paklaidos gali bti sumaintos.
Tyrimo rezultatai parod, jog tam tikram ribotos juostos kanalui parinkto
optimalaus skleisto spektro signalo juosta gali bti ir platesn u kanalo
juost.

Pateikta spaudai 2008 12 03