Restricted Boltzmann Machine for Interference Pattern Learning in Broadband Receivers.

Interference and noise mitigation is a critical component of many broadband communication systems. However, interference is often nonstationary, heavily dependent on the environment, and statistical a priori knowledge is not generally available. We propose a restricted Boltzmann machine (RBM) for unsupervised learning of time/frequency interference patterns in orthogonal frequency-division multiplexing (OFDM) receivers. Capable of learning patterns without statistical a priori knowledge, an RBM can be combined with a factor graph underlying a turbo or low-density parity-check decoder. We demonstrate the benefits of the proposed approach using the example of turbo encoded OFDM signal frames exposed to different forms of interference.