Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

435

RADAR EMITTER RECOGNITION


USING INTRAPULSE DATA
Adam KAWALEC', Robert OWCZAREKt

Abstract: Automatic emitter recognition is one of the most difficult tasks in the radar signal analysis. In most
cases,the modem ESMELINT systems cannot recognize the different devices of the same type or class. New
method of the radar identification with a very high precision recognizing is the Specific Emitter Identification
(SEI). The main task is to find non-intentional modulations in the receiving signals. This paper provides
an overview of the new methods of measurement emitter signal features parameters in the time and the frequency
domain. This paper presents some aspects of radar signal features extraction, selection and classification.

1. Introduction
The radar signal is the source of information about recognized object, which is propagated in free space.
Radar emitter identification based on a collection of received radar signals is a subject of wide interest in both
civil and military applications. The received signals usually consist of sequences of pulses emitted from multiple
radar transmitters. If different radars transmit pulses with different radio frequencies (RF) or pulse repetition
intervals (PRI), then it is not difficult to distinguish them one from another. However, in modem radar systems,
more sophisticated signal waveforms are used and only inter-pulse information may not be enough to separate
those received pulses according to their originations. To classify radar emitters in such an environment, we need
to explore the detailed structure inside each pulse, called intra-pulse information. This is because an emitter has
its own electrical signal structure inside each of its transmitted pulse due to both intentional and unintentional
modulations. This part of radar signal analysis is called Specific Emitter Identification (SEI).
It is a composite task that involves pulse measurements, features extraction, normalization, selection,
classification (recognition) and verification.
2. Signal analysis
A radar intercept receiver passively collects incoming pulse samples from a number of unknown emitters.
The information such as pulse repetition interval(s), angle(s) of arrival, carrier frequencies, and Doppler shifts
are not usable. Our objectives are to: (1) determine the number of emitters present; (2) classify the incoming
pulses according to the emitters.

Sipnal Model
We have the signal representation of the received pulses from K distinct emitters. The radar receiver intercepts
altogether N non-overlapping pulses from the emitters. We designate the nth received pulse by x,,(t;q,) where
n = 1,2, ..., N . Here q, is association parameter which assumes an integer value, q. E {1,2,..., K } , such that if
qn = k ,then the nth pulse is from the kth emitter. We can therefore express the nth pulse as

where A, denotes the initial amplitude of the received pulse; yndenotes the added phase of the received pulse

' Military University of Technology, Institute of Radar Technology, 2, Kaliskiego Str., 00-908 Warsaw, phone: (+48 22) 6839080, fax: (+48
22) 6837461, e-mail: akawalect2wat.edu.Dl,
' Military University of Technology, Instimlc of Radar Technology, 2, Kaliskiego Str., 00-908 Warsaw, phone: (+48 22) 6839243, fax: (+48
22) 6837461, e-mail: robeno(ii,wel.wat.edu.ol,
436

after transmission; r, denotes the time delay of received pulse with respect to the reference; W. denotes the
residual carrier frequency of the nth pulse; a,,(t) is the original envelope for the nth pulse such that
a,,@) E (a,(t),a2(l),..., a K ( l ) }; &"(I) IS the original phase for the nth pulse such that
@ ( I )= { 4 ( 1 ) , ~ (...,
r )~ ,K ( ~;) vn(l)
} is the Gaussian noise accompanying the nth pulse.

The received pulse in Eq.( 1) contains several nuisance parameters: A, , y, , rn and on. In our application these
parameters carry no usefil information for the determination a number of emitters and the pulse classification,
and should therefore be removed. The removal of these parameters necessitates the pulses pre-processing. After
that, the data can be expressed as

In practice, pre-processing is done in discrete time. Rewriting Eq.(2), we have


y,(mT,q,) =s,(mT)+%(mT), (3)
where T is the sampling interval of the pulses. The samples of the receiving pulses can be expressed as two
vectors: YAM- from demodulator AM (amplitude) and YFM- from demodulator FM (frequency)
T
YAM = [Y,l .Yii21....Y"p>Yo(p+,) 2 ..., Yn" I > (4)
r
YFM= b / i , Y . ~ ~ 1 ).
z - ~ ~ j p , ~ j ( ~Y/.+ l
I...1 (5)

3. Features extraction, selection and classification


The result of the feature extraction process is the set of intra-pulse radar signal parameters. Some of them shows
figure 2.
bi

Figure 2. Feature exfracrion. a) rise rime,fime of slope, and lime offall


h) rise angle ondfoll angle c) line of regression
All features measured in the microscale of time (in the time and the frequency domain) are combined to the
feature vector x = [ x , . x ,,..., xN]' .

Feature selection is the process of choosing a subset of the original predictive variables X by eliminating
redundant and uninformative ones. By extracting as much information as possible from a given data set while
using the smallest number of features, we can save significant computing time and often build models that
generalize better to unseen points. In our research, we used the two most popular methods of features selection to
check the goodness of the new parameters of radar signal.
Linear discriminant analysis (LDA) is a well-known classical technique to find the projection that
maximizes the ratio of scatter among the data of different classes to scatter within the data of the same class.
Initially, we consider a within-class scatter matrix for the within-class scatter. A within-class scatter matrix S,
is defined as

where c is the number of classes, C, is a set of data within the ith class, and mi is the mean of the ith class. The
within-class scatter matrix represents the degree of scatter within classes as a snmmation of covariance matrices
of each class. Next, we investigate a total scatter matrix in order to consider a between-class scatter matrix for
the between-class scatter. A total scatter matrix S, and a total mean vector m are defined as
437

where n is the number of total data and ni is the number of data within the ith class. After calculation we obtain

T
S, = S, + S, , S, = x n i ( m i-m)(mi -m) (8)
i=l

where S, is defined as a between-class scatter matrix, which represents the degree of scatter between classes as
a covariance matrix of means ofeach class.
The projection from a d-dimensional space to ap-dimensional is accomplished by
y = WTX (9)
where W is a d x p transformation matrix. (10)

For the scalar measure of scatter, we use the determinant of scatter matrices such as S
IM and 1-S ,I . Using this
measure, the criterion function J ( W ) can be defined as

We can obtain the transformation matrix W as one that maximizes the criterion function J ( W ) , The columns of
optimal W is the generalized eigenvector wi that correspond to the largest eigenvalues Ai in

S b W j = ,l;s,,wi (14
The second method is based on the principal component analysis (PCA) called Karhunen-Loeve
transformation ( T U ) . The essence of TKL is to reduce dimension of feature space. This expansion is an optimal
feature representation method to minimize mean-square error. The vector of feature X can be represent as a sum
of N linear independent vectors, without errors
Y
X = x y , t , = T'Y .
jil

We can calculate the covariance matrix oftransformation TKL as

After a few calculations we can obtain

...A#),
P=~[TX"'-T~][TX'"-T~]'=diog(A,,~,
,=I

where A,&, ...,AN are an eigenvalues of T. It means that the elements of matrix KL are uncorrelated and we
can calculate the coefficient of distance
d ( X )= zTz = (X'"-X)TP-'(X(,) -X), (16)
where z is normalized in this way

The distance d between tested feature and pattern is the weighted Mahalanobis distance in the feature space
The main task of the classificatory is making a decision about membership of tested vector to the given class
in the pattern space. The decision function can be expressed as

d,(X)=maxd,(X)+ X E ~ (18)
438

In our research we used one of the most popular measures of distance and similarity as Euclidean distance
d,‘(X,?) = (x-Y)r(x-Y) (19)
or “Cosine” similarity

The set of samples that we had at our disposal had been finite. They were utilizing for both: training and
testing. We used to Holdout Method to testing data. The available data set was divided into two subset, the third
part for training and the two thirds for testing. The correct identification coefficient ( C I Q was used to evaluation
of quality recognition

where L, is a number of good classifications, and L is a number of all classifications.

4. Experimental results
The samples were taken from the nine emitters of the same class. LDA and TKL gave almost the same
results. We used an old features as RF, PRI and PW (regularly using in the ESM devices) to compare them with
our features. Experimental results indicate our new intra-pulse features as the most discriminant and important
(see Fig.3.a). Using the new features it was easy to recognize all emitters because the patterns are safe distant
between each other (see Fig.3.b). We have successfully tried to recognize three emitters using their signature in
FM signal (see Fig.3.c). The CIC for the new features equals 0.98 and for the old 0.47. These marks
~ ~

eliminate the old features from the specific emitter identification.

Figure 3. Graphical results of experiment. a)features selection


b) emitter identification c) emitters recognition using one pulse
5. Conclusion
Specific Emitter Identification using intra-pulse features makes it a useful tool for electronic warfare
devices, to he used when reliable and fast information about different recognition radar emitters of the same type
is needed. Our features were checked by two different methods of data selection. Both LDA and TKL gave the
similar results. Measurement results shows that a new intra-pulse features including about 90% recognizing
information needed to resolve the problem of SEI. The results of speed and numerical stability of algorithms
seems to be enough to put them into practice in the ESM devices. In the future, we will attempt to solve the
problems of emitter identification using only one pulse.
References
[l] Duda R. O., Hart P. E., Stork D. G.: Pattern classification. John Wiley & Sons, Inc, New York 2000.
[2] Fukunaga K.: Introduction to statisticalpattern recognition. seconded. Academic Press, New York 1990.
[3] Langley L. E.: “Specific emitter identification (SEI) and classical parameter fusion technology”, fEEE
WESCOW93 Conference Record, San Francisco, 28-30 September 1993.
[4] Owczarek R.: ,,Problemy specyficznej identyfikacji ir6del emisji radarowych”, VKonferencja Naukowo-
Techniczna nf.: ,,Systemy rozpoznania i wa/ki radioelektronicznej”, Warszawa, 18-19 listopada 2002.
[SI Owczarek R.: ,,Automatyczne rozpoznawanie radarow na podstawie analizy ich sygnatow w mikroskali
czasowej”, IVKrajowa Konferencja nt.: ,.Melody i systemy komputerowe w badaniach naukowych
iprojektowaniu injnierskim”, Krakbw, 26-28 listopada 2003.

You might also like