In low SNR applications like passive surveillance sonar, an optimum detector which maximizes the output SNR is of prime importance. This paper presents a novel practical implementation scheme for an optimum (in Neyman Pearson sense) broadband detector for passive sonar, which includes a post beamformer filter named Eckart filter whose goal is to maximize the SNR at the detector output. The Eckart filter, realized using Maximum Likelihood (ML) estimation helps to mitigate the effect of spatio - temporal variations in the environmental noise. The ideal detection performance has been analysed using the Receiver Operating Characteristics (ROC) curves considering both conventional delay sum (frequency domain) and adaptive MVDR (Minimum Variance Distortionless Response) beamformer. An online scheme is proposed for practical realization of Eckart filter and the algorithm is validated for different spectral shapes. The performance of the optimum detector with Eckart filter is evaluated for different SNRs and target scenarios and the performance improvement compared to the detector without Eckart filter has been verified.