[5] E.P. Baars, "Digitale Registrierung von Radar-Echosignalenmit dem RADICORD-Gerat," Narchrichtentech. Z., vol. 21,pp. 756-758, 1968.

[6] W.D. Wirth, "Clutter- und Signalspektren aus Radicord-Aufnahmen und die Bewegtzielentdeckung durch Impuls-Radar," Nachrichtentech. Z., vol. 21, pp. 759-765, 1968.

[7] W. Buihring and R. Klemm, "Ein adaptives Filter zur Unter-driuckung van Radar-st&rungen mit unbekanntem Spektrum,"Frequenz, vol. 30, no. 9, pp. 238-243, 1976.

[8] R. Klemm, "Suboptimal simulation and suppression ofclutter signals," IEEE Trans. Aerosp. Electron. Syst., vol.12,pp. 210-212, 1976.

[9] , "Ein System zur adaptiven Clutteriinterdruckungbei Rundsichtradargeriten," DGON-Symp. Radartechnik,Munich, Germany, pp. 274-286, 1974.

[101 W. Biihring and W.D. Wirth, "Signalverarbeitung beiabgetasteten Radar-Signalgruppen," DGON-Symp. Radar-technik, Munich, Germany, pp. 288-303, 1974.

Corrections to "Radio Interference in Helicopter-BornePulse Doppler Radars"'

Equation (2) should read:

fd -2R/X. (2)

In the notations following equation (46) make thefollowing correction:

Fm average pulse repetition frequency of the modulat-ing signal = nxn /60;

'M.K. Moaveni, IEEE Trans. Aerospace and Electronic Systems,vol. AES-14, pp. 319-328, March 1978.

A Conditional-Test Sign Detector for Nonparametric Radar

Abstract

The performance of a four-level sign detector in constant false-alarm

rate (CFAR) radar is investigated. The detector uses a single inde-

pendent reference noise observation per sweep and the nonparam-etric property is obtained through the use of a conditional statistical

test. Detection performance is evaluated for the specific case of

white Gaussian background noise with Rician statistics for signalplus noise. Both the asymptotic and finite-sample detection perfor-mance is determined and found to be significantly better comparedwith the sign detector.

Manuscript received August 15, 1977.

This work was supported in part by the Air Force Office ofScientiflc Research under Grant AFOSR 77-3154.

0018-9251/78/0700-0688 $00.75 © 1978 IEEE

I ntroduction

For radar applications it is sometimes desirable to imple-ment systems which can provide a constant false-alarm rate(CFAR) under only rather general assumptions on thestatistics of the underlying noise sources. Several types ofnonparametric radar detectors can be designed to give CFARcharacteristics over classes of noise probability distributionfunctions which cannot be parameterized by a finite num-ber of parameters. Nonparametric detectors are usefulwhen, due either to lack of complete information or per-haps nonstationarity in the statistics of the noise processes,optimum detectors cannot be built based on a parametricformulation of the noise statistics.

Most nonparametric or CFAR detectors are based onthe use of signs (polarities) or ranks of received observations;some examples of nonparametric radar detectors may befound in [1-3]. In particular, the generalized sign testprocessor described by Hansen and Olsen [2] results in asimple detector achieving reasonable detector power. Theirprocessor operates onM independent video samples X'j,X2, ..., .XM for each range bin, the samples being outputsof an envelope detector following matched filtering. Totest for target presence at a certain range, a generalizedsign test statistic TGS is compared with a threshold, TG sbeing defined by

M

TGS = Z.f=1IJ

whereN

i = U(X jk)1 k=lJIJk

(1)

(2)

with u the unit step function. In forming the Zj it isassumed that for each sweep (one observation X] beinggenerated at the jth sweep) we can observe N independentreference noise samples, that is, data representing the out-puts of the envelope detector when no target is present. In(2) the terms X7'k, k = 1, 2, ...,N denote the independentreference observations at the jth sweep. Hansen and Olsenhave shown how the asymptotic performance of theirdetector improves with N, as compared with the weak-signal-optimum square-law detector on the basis of asymp-totic relative efficiency (ARE) for the case of Rayleighnoise densities. The ARE starts at 0.25 forN = 1 and forlargeN approaches 0.75.

It is conceivable that in several applications the acquisi-tion of more than one independent reference noise samplemay not be practical. In such a case the generalized signtest processor of [2] has a rather poor performance. Inthis correspondence we show how performance for thecaseN = 1 can be significantly improved, without sacrifilceof the CFAR property. The essential improvement is in thereplacement of the coarse, two-level quantizer u in (2) bya four-level quantizer, and the use of the technique of con-ditional testing [4] to maintain nonparametric operation.In the next sections we describe and analyze this form ofgeneralized sign test detection and present comparisons

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-14, NO. 4 JULY 1978688