Optimal design of multibit radar detection networks

Detection networks have received increasing attention recently, but most research has focused on cases where each local sensor makes a binary decision based on its own observation and then transmits this decision to the fusion center. The restriction of the sensor output to one bit certainly implies a substantial information loss. We, therefore, investigate optimal decision rules in a multi-bit detection network. A multi-bit decision is equivalent to multiple level quantization of the likelihood ratio. In this paper, we develop an iterative algorithm to determine the quantization levels in each of the independent local sensors. Our method is efficient and its asymptotic convergence is guaranteed. We demonstrate the feasibility of the proposed approach through the application to a radar detection problem that is CFAR detection of Rayleigh fluctuating targets in Gaussian noise. Simulation results are presented to show how the additional bits from local sensors could result in a better detection performance.