A comparison of decoding techniques for Interference Alignment

Interference Alignment (IA) is a general principle for signaling over an interference Gaussian channel, such that at each receiver the superposition of the interfering signals spans a small-dimensional subspace, linearly independent of the desired signal subspace. While signal subspace IA is known to achieve the degrees of freedom of certain classes of interference channels, in this paper we focus on its performance at finite SNR. We consider scenarios where IA is achieved without symbol extension (i.e., without precoding across time and frequency, but only across multiple antennas) and investigate the performance of the standard linear zero-forcing receiver, of linear integer forcing receivers, and of (non-linear) joint decoding implemented by lattice decoding with MMSE generalized decision feedback preprocessing. We show that the latter provides a significant improvement in terms of uncoded symbol error rate, averaged over the fading channel statistics.