In Brain Machine Interface (BMI), movement direction can be decoded using intra-cortical recordings such as Local Field Potentials (LFP). Due to the natural instability and non-stationarity of these recordings, it is difficult to develop decoders that remain consistent over time and are not affected by learning. This paper uses qualitative information based on the temporal and spatial distribution of inter-channel ranking. The image block processing technique is exploited, and the distribution of top ranked channels is calculated. We use this spatio-temporal distribution information to decode the movement direction via a maximum likelihood estimator. Our results indicate that the decoding power is consistent over a period of two weeks. On an average, we obtain an average classification accuracy of 51.9% versus 33.2% from traditional state-of-the-art technique over a two week period.