Locally Optimum and Suboptimum Detector Performance in a Non-Gaussian Interference Environment

Since the normally assumed white Gaussian interference is the most destructive in terms of minimizing channel capacity, substantial improvement can usually be obtained if the real-world interference environment (non-Gaussian) is properly taken into account. In this paper, the performance of the locally optimum Bayes detector (LOBD) is compared to the performance of various ad hoc nonlinear detection schemes. The known results are reviewed, and then it is demonstrated that these theoretical results may be misleading due to the assumptions that are required in order to derive them analytically. For a particular type of broad-band impulsive noise, the critical assumptions of "sufficiently" small signal level and large number of samples (large time-bandwidth product so that the central limit theorem applies) are removed; the first analytically, and the second by computer simulation. The thus-derived performance characteristics are then compared, especially as the signal level increases. One result is that there are situations where the bandpass limiter outperforms the LOBD as the signal level increases; that is, the locally optimum detector may not remain "near optimum" in actual operational situations.