Multichannel Detection Using Higher-Order Statistics

A model-based approach is proposed for multichannel system modeling and detection in the context of radar system applications, using higher-order statistics (HOS) for the identification of time series model parameters. The model innovations sequence is used for the detection decision. In general, HOS-based algorithms offer several advantages over conventional time series methods based on second-order statistics. Higher-order cumulants contain information about the system phase, and thus can be used to model both minimum-phase and non-minimum-phase systems. This allows a larger model class for the modeling of radar return signals. HOS-based algorithms specifically address the cases where the desired signal portion does not exhibit Gaussian statistics, which is the cases in many radar system applications. Additionally, the higher-order cumulants are insensitive to additive Gaussian-distributed noise, such as receiver noise and interference sources. Thus, cumulant-based algorithms offer potential for improved performance in many radar target detection problems.