Locally optimum (LO) detector is an appropriate technique for removing strong interference signals because its performance asymptotically approaches the global optimum as the signal amplitude goes to zero. However, deriving a LO detector with memory is often practically impossible due to the need to deal with multivariate probability density functions (pdf). Assuming Markovian structure for the interference and subsequently applying an autoregressive (AR) model leads to a straightforward implementation of the LO detector. Here we consider a system where a student t distributed noise drives an AR process and derive a LO detector for it. Further, we compare the performance of this detector to a traditional Wiener filter in terms of bit error rate (BER) and output signal-to-interference ratio (SIR).