Professional Documents
Culture Documents
A. DĂNIŞOR
“Ovidius” University of Constanta
Received December 21, 2004
This paper proposes a new method of signal to noise ratio maximization for
better underwater targets detection. Studying the properties of the underwater noise we
couldn’t conclude anything about his stationery, even in the restraint sense. Due this
fact, the time-frequency signal processing, signal affected by the underwater noise is
justified. This paper proposes a new method of signal to noise ratio optimization,
method based on a quadratic time-frequency processing (Wigner-Ville transform) of the
SONAR signals. This optimization is realized using a time-frequency matched filter,
similar with the classic matched filter. Using this quadratic matched filter we obtained a
signal to noise ratio visible better than the signal to noise ratio in the time domain,
using the classic matched filter.
INTRODUCTION
Time frequency methods have long been used for detection in applications
such as sonars. Time-frequency methods are of interest because of nonstationary
nature of these signals. The underwater noise is a nonstationary one and that’s why
it is necessary to use quadratic detectors for better performances.
Quadratic detectors theory has recently been exploited to derive optimal
quadratic time frequency detectors. Unfortunately, for an efficient implementation
of the matched time frequency filter presented in this paper it is necessary to use
performant computers.
*
Paper presented at the 5th International Balkan Workshop on Applied Physics, 5–7 July
2004, Constanţa, Romania.
Rom. Journ. Phys., Vol. 51, Nos. 1–2, P. 101–109, Bucharest, 2006
102 A. Dănişor 2
frequency filter that maximizes the signal to noise ratio, a time frequency matched
filter. The time frequency transfer function of this filter reseambles to the matched
filter impulse response of the linear detectors.
It is known that the quadratic detectors have good frequency selectivity, short
time transient response and noise suppression. If we consider the signal received by
a sonar system:
y (t ) = s (t ) + n(t )
if the target is present and:
y (t ) = n(t )
if the target is absent, where:
• y(t) is the received signal
• s(t) the reflected signal
• n(t) the noise
the quadratic detector’s response is:
∞ ∞
e= ∫ ∫ Q(t ,τ ) R (t ,τ )dtdτ
−∞ −∞
y (1)
τ τ
R y ( t ,τ ) = y t + y ∗ t – (2)
2 2
and Q(t ,τ ) may be considered the impulse response of the nonstationary detector.
The statistical autocorrelation function is, in fact, the statistical mean of the
instantaneous autocorrelation function:
τ τ
ry = E y t + y* t − = E ( Ry (t )) (3)
2 2
E(*) denotes the statistical mean.
Using the Wigner-Ville distribution’s definition:
∞
∫ R (t ,τ )e
− j 2π f τ
WVy (t , f ) = y dτ (4)
−∞
τ τ
Q(t ,τ ) = h t − , t + (6)
2 2
and using the Weyl symbol of the impulse response of the nonstationary detector,
defined as:
∞
τ τ
∫ h t + 2 , t − 2 e
− j 2πντ
LH (t ,ν ) = dτ (7)
−∞
Using the relations (5) and (8) the response (1) becomes:
∞ ∞ ∞ ∞
e= ∫ ∫ ∫ ∫L (t ,ν )WVy (t , f )e j 2π ( f −ν )τ dtdfdν dτ
*
H (9)
−∞ −∞ −∞ −∞
Equation (9) shows the relation between the response o the quadratic detector
and the Weyl symbol associated with the detector. Using Fourier transform of the
Dirac impulse relation (9) becomes more simple:
The essential problem is to find a suitable formula for Weyl symbol of the time-
frequency filter LH(t,f), formula which maximize the response of the quadratic
detector. Like the impulse response formula of the classic matched filter we choose
for the Weyl symbol an expression similar with the symmetric Wigner-Ville
distribution of the reflected signal. The symmetry between the Weyl symbol and
the Wigner-Ville distribution is in time domain as in frequency domain:
L*H (t , f ) = WVs* (−t , − f ) (11)
It is known that the Wigner-Ville distribution of a real signal is real too, previous
equation (11) can be wrote as:
L*H (t , f ) = WVs (−t , − f ) (12)
104 A. Dănişor 4
t =T
f =F
In consequence the maximum power of the signal at the output of the matched time
frequency filter is:
Ps = e2 (17)
5 Signal to noise ratio optimization using time-frequency 105
The medium power of the time frequency matched filter output is:
T F
1
T ∫0 ∫0
Pn = o(t , f ) y (t )= n (t ) dtdf (18)
where o(t,f) is the output of the time frequency matched filter in absence of the
reflected signal (target absent).
We can obtain a very interesting result using the ambiguity function. It is
known that the ambiguity function of the received signal y (t ) is defined as:
∞
Ay (τ ,ν ) = ∫ R (t ,τ )e
− j 2πν t
y dt (19)
−∞
WVy ( f , t ) = ∫ ∫ Ay (τ ,ν )e − j 2π ( f τ −ν t ) dτ dν
∫ ∫ A (ν ,τ ) A (ν ,τ )e
j 2π (ν t −τ f )
o(t , f ) = s y dν dτ (20)
−∞ −∞
But it is known that the relation (13) defines an unrealisable filter and that is why
the output o(t , f ) in the relation (12) is unrealisable too. For a realisable quadratic
filter, which output is expressed in the equation (14), using the ambiguity function
we obtain:
∞ ∞
∫ ∫ A (ν ,τ ) A (ν ,τ )e dν dτ
j 2π (ν ( t −T ) −τ ( f − F ))
o(t , f ) = y s (21)
−∞ −∞
The maximum value of the output filter in the relation (21) is:
∞ ∞
e = o(t , f ) t =T , f = F = ∫ ∫ A (ν ,τ ) A (ν ,τ )dτ dν
−∞ −∞
y s
106 A. Dănişor 6
EXPERIMENTAL RESULTS
Fig. 4 – Maximum response of the time frequency matched filter. Sinus signal.
Fig. 5 – Maximum response of the time frequency matched filter. Chirp signal.
9 Signal to noise ratio optimization using time-frequency 109
REFERENCES
1. Bouvet, M., Traitements des signaux pour les systems sonar- Ed. Masson, Paris 1991.
2. Coates M., Time frequency modeling - University of Cambridge, PhD., 1998.
3. Dănişor A., The study of the underwater noise – “Ovidius” University. Annals of Physics,
Constanţa, 2000.
4. Dănişor A, Optimizarea detecţiei ţintelor submarine – Telecomunicaţii, Bucureşti, 2000.
5. Maurice, A., Time frequency analysis; a real time differential spectral methods for an application
on sonar signals - J.Accostic Society of America, Aug., 1992.
6. Sayeed, A., Optimal kernels for nonstationary spectral estimation - IEEE Transaction on
information theory, Feb. 1995.