Average Decision Threshold of CA CFAR and excision CFAR Detectors in the Presence of Strong Pulse Jamming 1

Similar documents
A Two-Level Detection Algorithm for Optical Fiber Vibration

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

An Application of Fuzzy Hypotheses Testing in Radar Detection

DETERMINATION OF UNCERTAINTY ASSOCIATED WITH QUANTIZATION ERRORS USING THE BAYESIAN APPROACH

Error Probability for M Signals

Multiple Sound Source Location in 3D Space with a Synchronized Neural System

Pop-Click Noise Detection Using Inter-Frame Correlation for Improved Portable Auditory Sensing

Distributed Adaptive CCAWCA CFAR Detector

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

Lecture 16 Statistical Analysis in Biomaterials Research (Part II)

Pulse Coded Modulation

Chapter 13: Multiple Regression

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

Assignment 5. Simulation for Logistics. Monti, N.E. Yunita, T.

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

Credit Card Pricing and Impact of Adverse Selection

MATH 829: Introduction to Data Mining and Analysis The EM algorithm (part 2)

Uncertainty in measurements of power and energy on power networks

Composite Hypotheses testing

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

Chapter 7 Channel Capacity and Coding

Computation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models

Inductance Calculation for Conductors of Arbitrary Shape

RELIABILITY ASSESSMENT

Simulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests

A Lower Bound on SINR Threshold for Call Admission Control in Multiple-Class CDMA Systems with Imperfect Power-Control

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M

Linear Approximation with Regularization and Moving Least Squares

ECONOMICS 351*-A Mid-Term Exam -- Fall Term 2000 Page 1 of 13 pages. QUEEN'S UNIVERSITY AT KINGSTON Department of Economics

Comparison of the Population Variance Estimators. of 2-Parameter Exponential Distribution Based on. Multiple Criteria Decision Making Method

Predictive Analytics : QM901.1x Prof U Dinesh Kumar, IIMB. All Rights Reserved, Indian Institute of Management Bangalore

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS

Negative Binomial Regression

Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

Color Rendering Uncertainty

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for P Charts. Dr. Wayne A. Taylor

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

Orientation Model of Elite Education and Mass Education

Odd/Even Scroll Generation with Inductorless Chua s and Wien Bridge Oscillator Circuits

Research on Modified Root-MUSIC Algorithm of DOA Estimation Based on Covariance Matrix Reconstruction

First Year Examination Department of Statistics, University of Florida

Digital Modems. Lecture 2

On Centralized Composite Detection with Distributed Sensors

A Robust Method for Calculating the Correlation Coefficient

A Particle Filter Algorithm based on Mixing of Prior probability density and UKF as Generate Importance Function

Chapter 7 Channel Capacity and Coding

Fuzzy Boundaries of Sample Selection Model

An Improved multiple fractal algorithm

Code_Aster. Identification of the model of Weibull

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

An Upper Bound on SINR Threshold for Call Admission Control in Multiple-Class CDMA Systems with Imperfect Power-Control

Comments on Detecting Outliers in Gamma Distribution by M. Jabbari Nooghabi et al. (2010)

Maximizing Overlap of Large Primary Sampling Units in Repeated Sampling: A comparison of Ernst s Method with Ohlsson s Method

Double Acceptance Sampling Plan for Time Truncated Life Tests Based on Transmuted Generalized Inverse Weibull Distribution

Chapter 20 Duration Analysis

Numerical Heat and Mass Transfer

Lossy Compression. Compromise accuracy of reconstruction for increased compression.

10.34 Fall 2015 Metropolis Monte Carlo Algorithm

Limited Dependent Variables

On the correction of the h-index for career length

Marginal Effects in Probit Models: Interpretation and Testing. 1. Interpreting Probit Coefficients

Convexity preserving interpolation by splines of arbitrary degree

Multi-user Detection Based on Weight approaching particle filter in Impulsive Noise

Parameter Estimation for Dynamic System using Unscented Kalman filter

Simulation and Random Number Generation

The Minimum Universal Cost Flow in an Infeasible Flow Network

NUMERICAL DIFFERENTIATION

Statistical inference for generalized Pareto distribution based on progressive Type-II censored data with random removals

829. An adaptive method for inertia force identification in cantilever under moving mass

Non-Mixture Cure Model for Interval Censored Data: Simulation Study ABSTRACT

On the Multicriteria Integer Network Flow Problem

LINEAR REGRESSION ANALYSIS. MODULE VIII Lecture Indicator Variables

Scroll Generation with Inductorless Chua s Circuit and Wien Bridge Oscillator

A Bayes Algorithm for the Multitask Pattern Recognition Problem Direct Approach

Hopfield Training Rules 1 N

Constructing Control Process for Wafer Defects Using Data Mining Technique

Goodness of fit and Wilks theorem

Low Complexity Soft-Input Soft-Output Hamming Decoder

COGNITIVE RADIO NETWORKS BASED ON OPPORTUNISTIC BEAMFORMING WITH QUANTIZED FEEDBACK

Conjugacy and the Exponential Family

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

4DVAR, according to the name, is a four-dimensional variational method.

One-sided finite-difference approximations suitable for use with Richardson extrapolation

On an Extension of Stochastic Approximation EM Algorithm for Incomplete Data Problems. Vahid Tadayon 1

CHAPTER 14 GENERAL PERTURBATION THEORY

18.1 Introduction and Recap

Uncertainty and auto-correlation in. Measurement

Probability Theory. The nth coefficient of the Taylor series of f(k), expanded around k = 0, gives the nth moment of x as ( ik) n n!

Chapter 12 Analysis of Covariance

Suppose that there s a measured wndow of data fff k () ; :::; ff k g of a sze w, measured dscretely wth varable dscretzaton step. It s convenent to pl

A Network Intrusion Detection Method Based on Improved K-means Algorithm

x = , so that calculated

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Primer on High-Order Moment Estimators

The Gaussian classifier. Nuno Vasconcelos ECE Department, UCSD

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

A Bound for the Relative Bias of the Design Effect

Transcription:

Average Decson hreshold of CA CFAR and excson CFAR Detectors n the Presence of Strong Pulse Jammng Ivan G. Garvanov and Chrsto A. Kabachev Insttute of Informaton echnologes Bulgaran Academy of Scences Acad. G. Bonchev Str., bl., 3 Sofa, Bulgara Phone: 359/979-9-8 e-mal: cabachev@t.bas.bg, garvanov@t.bas.bg Introducton Cell-Averagng Constant False Alarm Rate (CA CFAR sgnal processng proposed by Fnn and Johnson n [] s often used for radar sgnal detecton. he detecton threshold s determned as a product of the nose level estmate n the reference wndow and a scale factor to acheve the desgn probablty of false alarm. he presence of strong pulse ammng (PJ n both, the test resoluton cell and the reference cells, can cause drastc degradaton n the performance of a CA CFAR processor as shown n []. For eepng of constant false alarm rate n PJ, the CA CFAR processor presented n [8,] s used. For the mnmzaton of CFAR losses n case of pulse ammng, PI or BI s mplemented n CFAR processors as shown n [4,7,9]. he use of excson CFAR detectors, supplemented by a postdetecton ntegrator or a bnary ntegrator as shown n [5,6,9], ncreases the CFAR losses. Mnmum CFAR losses n PJ are obtaned n [4,] wth a CFAR adaptve postdetecton ntegraton (API processor wth adaptve selecton on PJ n reference wndows and approry selecton n test wndows as shown n [4], and adaptve censorng n reference and test wndows as presented n []. In such stuatons, t would be desrable to now the CFAR losses, dependent on the parameters of PJ, for ratng of radar behavor. We use the crteron offered by Rollng and Kassam n [,3], based on the average decson threshold (AD. he AD and the detecton probablty are closely related to each other. he dfference between the two CFAR systems s expressed by the rato between the two ADs measured n db, as shown n [,3]. We assume n ths paper that the nose n the test cell s Ralegh envelope dstrbuted and target returns are fluctuatng accordng to Swerlng II model, as t s n [3,4]. As a dfference from the authors n [4], we assume that the samples of PJ are dstrbuted accordng to the compound exponental law, where weghtng coeffcents are the probabltes of corruptng and non-corruptng of the samples. We have used weghtng coeffcents n the nterval between and. For values of the weghtng coeffcents hgher than.3, the Posson process model s used, but t s rough []. he results n ths case are correct as far as processor behavour s concerned, but the numercal values are not accurate. he bnomal dstrbuton s correct n ths case. In ths paper new analytcal expressons for the average decson threshold (AD of a CA CFAR and excson CFAR detectors n pulse ammng are derved. he results obtaned for the AD of a CA CFAR processor wthout pulse ammng are equal to those presented n [3]. he expermental results show that the nfluence of nterference on the detecton process, when havng CA CFAR and excson CFAR processors n pulse ammng, s smlar to that gven n [6,8,9], obtaned by usng conventonal methods for the calculaton of the CFAR losses. he expermental results show that excson CFAR processors are most sutable for use when the probablty for the appearance of pulse ammng taes values n the nterval ( to.5. In cases when the probablty for the appearance of pulse ammng taes values n the nterval (.5 to, we recommend CA CFAR processors. hs wor s supported by II 44, MPS Ltd. Grant RDR and Bulgaran F SR Grant I 9/99

Performance of CA CFAR and excson CFAR detecton n the presence of pulse ammng. Probablty of detecton and false alarm of CA CFAR detectors Consder a radar detector, n whch the receved sgnal s sampled n range by ( resoluton cells resultng n a vector of ( observatons. he samplng rate s such that the samples are statstcally ndependent. After fltraton, the sgnal s appled to a square-law detector and then processed n the CA CFAR decson element. In condtons of pulse ammng the bacground envronment ncludes random nterferng pulses and the recever nose. herefore the samples surroundng the cell under test (a reference wndow may be drawn from two classes. One class represents the nterference-plus-nose stuaton, whch may appear at the output of the recever wth the probablty. hs probablty can be expressed as = t c F, where F s the average repetton frequency of PJ and t c s the length of pulse transmsson. he other class represents the nose only stuaton, whch may appear at the output of the recever wth the probablty (. he probablty densty functon (pdf of the test resoluton cell s assumed to be dstrbuted accordng to Swerlng II case [8, ]: ( ( ( ( ( x x f x = exp exp ( λ s λ s λ r s λ r s where λ s the average power of the recever nose, r s the average nterference-to-nose rato (IR of pulse ammng, s s the per pulse average sgnal-to-nose rato (SR and s the number of observatons n a reference wndow. In ths case the Posson process model s used, and t s vald only for.3. he probablty densty functon (pdf of the reference wndow outputs can be defned accordng to (, settng s =. he probablty of pulse detecton P D s obtaned n [8] as: PD = ( ( s ( r s ( λ λ where M (. s the moment generatng functon (mgf of the nose level estmate. In a conventonal CA CFAR detector the nose level estmate s formed as a sum of all the outputs of the reference wndow: =. In ths case the mgf of the estmate s defned to be x = x ( ( U = Mx ( U, where M U s obtaned n [8]: M ( U = s the mgf of the random varable x. he mgf of the estmate C ( = ( λ U ( λ ( r U he probablty of target detecton n [8] s computed by usng the followng expresson: PD = C ( = ( r ( r r s r s s s he probablty of false alarm s evaluated by (4, settng s =.. Probablty of detecton and false alarm of excson CFAR detectors In an excson CFAR processor the nose level estmate s formed as an average mean of K nonzero samples at the output of the excsor { y }, that s: = y. Accordng to [5] the = operaton of the excsor s defned as follows: (3 (4

x : x B y = : othewse (5 where B s the excson threshold. he probablty that a sample x survves at the output of the excsor, s gven as: B ( ( B P = exp exp (6 λ λ r he probablty that out of samples of the reference wndow survve at the output of the v = C P P. he mgf of the random varable y at the output of the excsor s gven as: ( ( excsor can be obtaned n [6] by usng the followng expresson: ( exp( R BU exp R BU M y ( U = exp R ( r exp R where R B B = R =. ( λ ; λ r ( ( ( ( ( ( ( ( ( Snce the random varables x ( are ndependent, the mgf of the estmate can be obtaned as follows: M ( U, = M ( U /. In ths artcle we use the moment generatng functon on the excson CFAR processor from [6] where Y ( U = CP ( P ( U, = ( exp( R BU / ( R r / ( ( exp( R BU / ( exp( R ( M U = (, C = ( exp ( ( / (9 he probablty of target detecton for excson CFAR n [6] s computed by the expresson: PD = C P ( P ( ( s,, ( r s ( = λ λ he probablty of false alarm s evaluated by (, settng s =..3 Average decson threshold of CA CFAR and excson CFAR detectors he average decson threshold AD s defned as a normalzed quantty []: ADCFAR = / λ ( ( where the random varable s the result of the estmaton method used n the CFAR system, s the scalng factor for threshold adustment adapted to the estmaton method and requred P FA, and stands for the expectaton. d ( / λ = ( d M / λ = ( For the mathematcal expressng of the AD, we use the moment generaton functons (mgf (3 and (8, presented n [6, 8] and the method descrbed n [, 3]. We obtan new analytcal results for the AD of CA CFAR and excson CFAR processors n strong PJ, as follows:.3. Determnaton of the AD for a CA CFAR processor n PJ Usng (3 and (, we substtute U = / λ and for = we have the AD expresson: (7 (8 3

( AD d ( ( d M CACFAR = = ( = C r = (4 λ λ where s computed by expresson (4, settng s =. For =, or wthout pulse ammng, AD = / = P fa. CACFAR = as n [3], where (.3. Determnaton of the AD for excson CFAR processor n PJ Usng (8 and (, we substtute U = / λ and for = we have the AD expresson: ( AD = C P ( P C = = ( ( exp( R exp( R B ( exp( R exp( R r ( ( exp R where s computed by expresson (, settng s =. ( exp( R ( exp( R ( exp( R exp( R ( B (5 3 umercal results he new analytcal expressons obtaned n the prevous secton mae possble the estmaton of the qualty of CA CFAR and excson CFAR detectors n the presence of very ntensve pulse ammng. he results are accurate for.3, but not precse for >.3. Fg. CA CFAR processor Fg. XC CFAR processor AD,, are for r = 5 db, AD,, are for r = 5 db, AD,, are for r = 3 db AD,, are for r = 3 db he expermental results are obtaned for the followng parameters: average power of the recever nose λ =; average nterference-to-nose rato (IR r =5 and 3 [db]; probablty for the appearance of pulse ammng wth average length n the range cells from to ; number of reference cells for CA CFAR end excson CFAR processors =6; probablty of false alarm P fa = 6 and excson threshold B =. In a CA CFAR processor, the nose level estmate n the reference wndow ncreases wth the ncreasng of the average nterference-to-nose rato and the probablty for the appearance of pulse ammng wth average length n the range cells (Fg.. In order to eep the false alarm probablty constant, the scale factor must be decreased when the PJ frequency ncreases. he average decson threshold (AD ncreases when the probablty for the appearance of pulse ammng taes values from to.5, and then decreases for value of > 5. (Fg.. 4

In the excson CFAR processor, the ammng pulses are censored and the nose level estmate n the reference wndow s ept constant (Fg.. In order to eep the false alarm probablty constant, the scale factor must be ncreased wth the ncreasng of PJ frequency. he average decson threshold (AD s constant when the probablty for the appearance of pulse ammng taes values between and.5, and then ncreases for value > 5. (Fg.. he average decson threshold (AD for an optmal detector s constant when the probablty for the appearance of pulse ammng taes values between and, (Fg. and. 4. Conclusons. ew analytcal expressons for the average decson threshold (AD of CA CFAR and excson CFAR detectors n pulse ammng are derved n ths paper. When CA CFAR and excson CFAR detectors operate wth a fxed scale factor, the detecton probablty s decreased n strong PJ, but the false alarm probablty s not mantaned constant. In our case, the scale factor s adusted to PJ so that the false alarm probablty s mantaned constant. he results obtaned for the AD of a CA CFAR processor wthout pulse ammng are equal to those presented n [3]. he usng of averaged characterstcs for the analyss of dfferent CFAR processors s a very convenent mathematcal apparatus for precse and easy determnaton of detector energy losses. he expermental results show that excson CFAR processors are most sutable for use when the probablty for the appearance of pulse ammng taes values n the nterval ( to.5. In cases when the probablty for the appearance of pulse ammng taes values n the nterval (.5 to, we recommend CA CFAR processors. he results obtaned n ths paper can practcally be used for the desgn of modern radar systems. References [] Fnn, H. M., P. S. Johnson, Adaptve detecton mode wth threshold control as a functon of spatally sampled clutter estmaton, RCA Revew, vol. 9, o 3, 968, pp. 44-464. [] Rohlng H. Radar CFAR hresholdng n Clutter and Multple arget Stuatons, I rans., vol. AS-9, o 4, 983, pp. 68-6. [3] Gandh, P. P., S. A. Kassam, - Analyss of CFAR processors n nonhomogeneous bacground, I rans., vol. AS-4, o 4, 988, pp. 443-454. [4] Hmonas S., CFAR Integraton Processors n Randomly Arrvng Impulse Interference, I rans., vol. AS-3, o 3, 994, pp. 89-86. [5] Goldman H., Analyss and applcaton of the excson CFAR detector, I Proceedngs, vol.35, o 6, 988, pp. 563-575. [6] Behar., C. Kabachev, xcson CFAR Bnary Integraton Processors, Compt. Rend. Acad. Bulg. Sc., vol. 49, o /, 996, pp. 45-48. [7] Kabachev C.,. Behar, - CFAR Radar Image Detecton n Pulse Jammng, I Fourth Int. Symp. ISSSA'96, Manz, Germany, 996, pp. 8-85. [8] Behar., CA CFAR radar sgnal detecton n pulse ammng, Compt. Rend. Acad. Bulg. Sc., vol. 49, o 7/8, 996, pp. 57-6. [9] Kabachev C.,. Behar, echnques for CFAR Radar Image Detecton n Pulse Jammng, I Fourth Int. Symp. UM'96, Praga,Cheh Republc, 996, pp. 347-35. [] Behar., C. Kabachev, L. Duovsa, Adaptve CFAR Processor for Radar arget Detecton n Pulse Jammng, Journal of LSI Sgnal Processong, vol. 6, o /,, pp. 383-396. [] Kabachev C., L. Duovsa, I. Garvanov, Comparatve Analyss of Losses of CA CFAR Processors n Pulse Jammng, CI, o,, pp. -35. [] Amov, P., vstratov, F., Zaharov, S.: Rado Sgnal Detecton, Moscow, Rado and Communcaton, 989, pp. 95-3, (n Russan. 5