Performance comparison between matched filter and locally optimal detector for composite hypothesis test with inaccurate noise

This study considers the performance comparison between the matched filter (MF) and the locally optimal (LO) detector when the noise information is not accurate. For a given binary hypothesis testing problem, the authors derive the condition under which the LO detector performs worse than the MF with the assumptions of large number samples and low signal-to-noise ratio. The condition is an inequality, which involves the noise probability density function (pdf) used in the LO detector, the real noise pdf, and the noise variance. Simulations show that the condition is theoretically sound under both known and unknown noise pdf's.