DETECTION OF RAUAR SIGNALS IN NOISE

Likelihood-ratio receiver. The likelihood ratio is an important statistical tool and may be
defined as the ratio of tlie probability-density function corresponding to signal-plus-noise,
p»n(r). l0 'ne probability-density function of noise alone, pB(f).

It is a measure of how likely it is that the receiver envelope i; is due to signal-plus-noise as

compared with noise alone. It is a random variable and depends upon the receiver input. If the
likelihood ratio L,{v) is sufficiently large, it would be reasonable to conclude that the signal
was indeed present. Thus detection may be accomplished by establishing a threshold at the
output of a receiver which computes the likelihood ratio. The selection of the proper threshold
level will depend upon the statistical detection criterion used and by the probabilities of error
desired and their relative importance.
The likelihood ratio is primarily useful in the analysis of the statistical-detection problem.
It is difficult to conceive, however, of a receiver that computes the likelihood ratio directly as
defined by Eq. (10.21). However, in certain cases the receiver which computes the likelihood
ratio is equivalent to a receiver which computes the cross-correlation function,or to one with a
matched-filter characteristic, that is, one which maximizes the output signal-lo-noise ratio.24
The Neyrnan-Pearson Observer is equivalent to examining the likelihood ratio and deter-
mining if Lr(v) s AC, where K is a real, nonnegative number dependent upon the probability of
false alarm selected.

Inverse probability receiver. A detection criterion that has been popular particularly for the
theoretical analysis of statistical detection and for statistical parameter estimation is that
based on inverse probability. The usual detection criteria employ the concept of direct probabil-
ity, which describes the chance of an event happening on a given hypothesis. For example, the
probability that a particular radar will detect a certain target under specified conditions is a
direct probability. On the other hand, if an event actually happens, the problem of forming the
best estimate of the cause of the event is a problem in inverse probability. For example, assume
the event in question to be the output voltage v from a radar receiver. Upon obtaining this
voltage, it is of interest to determine whether the output was caused by noise or by signal in the
presence of noise. The probabilities of obtaining noise and signal-plus-noise before the event
takes place are the a priori probabilities. They represent the initial state of knowledge concerning
the evenj. The probability that the receiver output v was caused by noise or by signal-plus-noise
is an a posteriori probability and represents the state of knowledge obtained as a result of
observing the output.
The method of inverse probability involves the use of the a priori probabilities associated
with each of the possible hypotheses which could explain the event. The a priori probabilities
are used, along with a knowledge of the event, to compute the a posteriori probabilities. A
separate a posteriori probability is computed for each hypothesis. That hypothesis which
results in the largest a posteriori probability is selected as the most likely to explain the event.
This method has been applied by Woodward and Davies to the reception of signals in
noise.2*"2* It is based upon the application of Bayes' rule for the probability of causes.29 The
joint probability of two events x and y is
p(*. y) = p(*)p(y\x) - p(y)p(x\y) (10.22)
where p(x) and p(y) = probabilities of events x and y, respectively
p{y\x) — conditional probability that event y will occur, given that event x has
occurred
piv I !•) — rnnHifuinnl nrnhnhililv of rvt*n1 v pivpn fhnf r has norlirrpH
^ 6 i i INTRODUCTION TO RADAR
SYSTEMS

parallel channels, each containing a delay line corresponding to a particular value of T,, as
well as a multiplier and low-pass filter. In some applications it may be possible to record the
signal on some storage medium, and at a higher playback speed perform the search sequen-
tially with different values of Tr. That is, the playback speed is increased in proportion to the
number of time-delay intervals Tr that are to be tested?
Since the cross-correlation receiver and the matched-filler receiver are equivalent math-
ematically, the choice as to which one to use in a particular radar application is determined by
which is more practical to implement. The matched-filter receiver, or an approximation, has
been generally preferred in the vast majority of applications.

J
10.4 DETECTION CRITERIA

The detection of weak signals in the presence of noise is equivalent to deciding whether the
receiver output is due to noise alone or to signal-plus-noise. This is the type of decision
probably made (subconsciously) by a human operator on the basis of the information present
at the radar indicator. When the detection process is carried out automatically by electronic
means without the aid of an operator, the detection criterion cannot be left to chance and must
be carefully specified and built into the decision-making device by the radar designer.
In Chap. 2 the radar detection process was described in terms of threshold detection.
Almost all radar detection decisions are based upon comparing the output of a receiver with
some threshold level. If the envelope of the receiver output exceeds a pre-established threshold,
a signal is said to be present. The purpose of the threshold is to divide the output into a region
of no detection and a region of detection; or in other words, the threshold detector allows a
choice between one of two hypotheses. One hypothesis is that the receiver output is due to
noise alone; the other is that the output is due to signal-plus-noise. It was shown in Chap. 2
that the dividing line between these two regions depended upon the probability of a false
alarm, which in turn is related to the average time between false alarms.
There are two types of errors that might be made in the decision process. These are
unavoidable with observations of finite duration in the presence of noise. One kind of error is
to mistake noise for signal when only noise is present. This occurs whenever the noise is large
enough to exceed the threshold level. In statistical detection theory it is sometimes called a
type I error. The radar engineer would call it & false alarm. A tyrJe II error is one in which the
signal is erroneously considered to be noise when signal is actually present. This is a missed
detection to the radar engineer. The setting of the threshold represents a compromise between
these two types of errors. A relatively large threshold will reduce the probability of a false
alarm, but there will be more missed detections. The nature of the radar application will
influence to a large extent the relative importance of these two errors and, therefore, the setting
of the threshold.

Neyman-Pearson observer. The threshold level was selected in Chap. 2 so as not to exceed a
specified false-alarm probability; that is, the probability of detection was maximized for a fixed
probability of false alarm. This is equivalent to fixing the probability of a type I error and
minimizing the type II error. It is similar to the Neyman-Pearson test used in statistics
for determining the validity of a specified statistical hypothesis.24'" Therefore this type of
threshold detector is sometimes called a Neyman-Pearson Observer. In statistical terms it is
claimed to be a uniformly most powerful test and is an optimum one, no matter what the a
priori probabilities of signal and noise.31 The Neyman-Pearson criterion is well suited for
radar application and is usually used in practice, whether knowingly or not.