In this thesis parametric and non-parametric schemes for discrete time signal discrimination are considered. Discrete time signal discrimination is the problem of classifying a random discrete time signal into one of two classes. The term discrimination arises from the more specific problem where the two classes are a target of interest and a decoy target.

In this thesis we consider both parametric and non-parametric schemes for discriminating between the two classes. In Chapter 2, we assume that first and second order probability density functions (pdfs) of the data under each class are known. Using these pdfs optimal memoryless quantizer discriminators are constructed. In Chapter 3, it is assumed that the pdfs are not know. Utilizing kernel density estimators and samples data from each class, estimates of the pdfs are formed for each class. Optimal memoryless quantizer discriminators are then constructed using the estimated pdfs and the expressions from Chapter 2. In Chapter 4, a perceptron neural network is trained with a supervised learning algorithm using sample data from each class. The perceptron neural network is utilized by a discriminator which uses memory. Results for simulated radar data are presented for all schemes. Results show that the neural network discrimination scheme performs significantly better than the memoryless quantization schemes.