On Optimum Data Quantization for Signal Detection.

An introduction to quantization and to several important detection problems is given in the initial sections. A detailed review follows of most of the work done on quantization for detection. The equivalence of the criterion of minimum mean-squared error between quantized data and data transformed by the locally optimum nonlinearity and the one of maximum efficacy is shown for the general case of local decisions based on independent samples. In addition, a sufficient condition for optimum detection is derived for the above case. Finally, numerical results are obtained for the locally-optimum quantizer for the case of detecting stochastic signals in generalized Gaussian noise. Author