Detection and classification of frequency-hopped spread spectrum signals

This dissertation proposes a means of defeating Frequency-Hopped (FH) spread spectrum modulation using an intercept receiver capable of fast emission detection and classification. By continuously analyzing the spectrum and classifying detected emissions, the receiver is capable of following a FH signal as it changes frequency. No a priori knowledge of the FH signals is needed by the receiver.; The basic structure used for emission detection is the radiometer. The tradeoffs between integration time and detection bandwidth examined. A suitable design for the intercept receiver is selected, and a digital architecture using Fast Fourier Transforms (FFT's) to implement the radiometer function is proposed. The detection performance of the digital receiver is predicted and compared with conventional analog radiometric receivers. Constraints on the time-bandwidth product imposed by the use of FFT's are also examined.; Correct emission classification is the second problem that must be addressed before signal interception is possible. Two classification algorithms developed using Baysian decision theory are discussed. The algorithms use data with arbitrary distributions calculated from samples of detected emissions to classify emissions. The first classification algorithm considered is the well-known maximum likelihood rule. The form of the decision rule, and the classification accuracy that can be expected from it are discussed. The epoch classification algorithm is then developed to eliminate errors which occur using the maximum likelihood classification algorithm. The probability of classification error for the epoch classification algorithm is found to be considerably less than that obtained using maximum likelihood classification.; To illustrate the application of the classification algorithms, emission frequency is used with hop frequency order statistics to classify emissions. The use of emission frequency is shown to eliminate or significantly reduce the probability of classification error when hopping spans from different FH signals are distinct or only partially overlap.