AN ADAPTIVE CORRELATOR FOR UNDERWATER MEASUREMENTS 
by DR. ALFRED A, WOLF, Director of Research 
and J.H. DIETZ, Fellow Engineer 
Emertron, Inc. 
Silver Spring, Maryland 
ABSTRACT 
The measurement and display of the correla- 
tion functions of quasi-stationary random proc- 
esses containing energy in the frequency range 
0,01-100,000 eps have become increasing impor- 
tant in oceanography and underwater acoustics, 
Correlation techniques yield useful statistical 
descriptions of sea state, ocean background 
noise, and the acoustical properties of bodies 
of water, In addition, since the correlation 
functions of periodic processes are also peri- 
odic, correlation may be used to separate weak 
Signals from noise, 
INTRODUCTION 
At present correlation functions are usually 
determined by means of a digital computer, 
which employs sampled records of the processes 
to compute approximating sums for the time inte- 
grals that define the correlation functions, 
Disadvantages of this method are (1) the need 
for the computer itself, (2) the fact that the 
required number of samples must be determined 
experimentally, (3) the bandwidth limitation 
imposed by sampling, and (4) the discrete dis- 
play of the correlation functions, 
The defining time integral may also be ap- 
proximated by analogue techniques, The re- 
quired time delays are introduced by means of 
tapped delay lines; analogue mrltipliers are 
347 
used to obtain the products of the delayed and 
undelayed signals; and averaging is accomplished 
in a low-pass filter, Like the digital method, 
this scheme gives discrete values of the correla- 
tion functions, The bandwidths of the signals to 
be correlated are restricted by the pass band of 
the signal multiplier, The principal disadvan- 
tage of this method is the need for low-distor- 
tion delay lines capable of producing delays of 
order of seconds and in some cases minutes, 
In this paper an analogue method requiring no 
delay lines and yielding a continuous approxima- 
tion of the correlation function is considered, 
An outgrowth of the work of Wolf and Dietz“ in 
system identification, the method consists in ex- 
panding the correlation function in a series of 
orthonormal functions the coefficients of which 
are determined by analogue techniques, A simi- 
lar but less general method was independently 
developed by Lampard ,2 Since in practice the 
series must always contain a finite number of 
terms, the problem of an optimum approximation 
to the correlation function is treated, The 
filters that generate the orthonormal functions 
automatically adjust their transmission charac- 
teristics to give an optimum approximation in the 
minimum-integral-square-error sense, 
