Least Squares Method in the Statistic Analysis
of Periodically Correlated Random Processes
I. N. Yavorskyj
1
, R. M. Yuzefovych
1
, I. B. Kravets
1
, and Z. Zakrzewski
2
1
Karpenko Physico-Mechanical Institute of NASU, Lviv, Ukraine
2
Institute of Telecommunications of The University of Technology and Life Sciences (UTLS), Bydgoszcz, Poland
Received in final form June 24, 2010
Abstract—The properties of least-squares estimates of mathematical expectations and correlation
function of periodically correlated random processes (mathematical model of stochastic oscillations)
have been investigated. The formulas defining the statistical characteristics of estimates were analyzed.
In addition, examples were presented for illustrating the analysis of modulated signals.
DOI: 10.3103/S0735272711010079
Periodically correlated random processes (PCRP) represent a mathematical model for a wide range of
physical phenomena, since they describe both the repeatability and the stochasticity of temporal variability
[1–5]. Taking account of the periodic correlation of signals applied in telecommunications and telemetry
makes it possible to solve more effectively the problems of signal analysis, transformation, and processing
[3, 4]. The analysis of vibration signals based on a model in the form of PCRP enables us to improve the
efficiency of diagnostics, in particular uncover defects of mechanisms at early stages of their development
[6, 7]. Mathematical expectation of PCRP mt E t () () =x , distribution averaging operator E, and also
correlation function btu E t t u (,) ()( ) = + xx
o o
, x x
o
() () () t t mt = - are periodic functions of time t and therefore
can be presented by the Fourier series:
mt m
k
k t
k
() =
Î
å
e
i w
0
¢
, btu B u
k
k t
k
(,) () =
Î
å
e
i w
0
¢
,
where w p/
0
2 = T , T is the period. The objective of the correlation statistical analysis is to determine
functions mt ( )and btu ( , ) (as functions of two variables: time t and shift u) and also their Fourier coefficients
m
k
and B u
k
( ) (the latter are also called correlation components) on the basis of experimental data. Such
determination can be performed by using either coherent [8] or component [9] methods. The first of them is
based on averaging the readings of process realization over period T:
$ () ( ) mt
N
mt nT
n
N
= +
=
-
å
1
0
1
,
[ ]
$
(,) ( ) $ ( ) btu
N
t nT mt nT
n
N
= + - +
=
-
å
1
0
1
x
[ ]
x( ) $ ( ) t u nT mt u nT ++ - ++ ,
here N is the number of periods that are averaged, while the second method is based on using the
trigonometric interpolation
$ () $ mt
N
m
k
k t
k N
N
=
=-
å
1
0
1
1
e
i w
,
45
ISSN 0735-2727, Radioelectronics and Communications Systems, 2011, Vol. 54, No. 1, pp. 45–59. © Allerton Press, Inc., 2011.
Original Russian Text © I.N. Yavorskyj, R.M. Yuzefovych, I.B. Kravets, Z. Zakrzewski, 2011, published in Izv. Vyssh. Uchebn. Zaved., Radioelektron., 2011, Vol. 54,
No. 1, pp. 54–64.