The quantized detection algorithm

An algorithm is presented for designing optimum quantizers for signals at two remote sensors that are to be fused at a central site in order to make a detection decision. Fusion rules are selected as candidates according to their ability to approximate the likelihood ratio test, a continuous curve in the observation space, with stepwise continuous approximations. The number of steps is determined by N , the number of levels of quantization. Results are presented showing the uniform convergence of the algorithm´s performance to that of the likelihood ratio test with increasing N for known signals in Gaussian noise. It is shown that an N of four, or two-bit quantization, performs nearly as well as the likelihood ratio test and is far superior to an N of two, or one-bit quantization, which corresponds to local detection decisions