Throughput analysis of interference alignment for a general centralized limited feedback model

In this paper, we consider a general centralized feedback model for interference alignment (IA) algorithms of which precoders and decoders are designed with quantized channel state information (CSI) at a central unit and then fed back to corresponding transmitters and receivers via finite rate links. The significant difference of this model from existing ones is that all the realizations of channels, precoders, and decoders are fed back by finite rates. The signal-to-interference-plus-noise ratio characteristic of this model is analyzed, which helps derive an upper bound of average throughput loss relative to perfect IA. The upper bound is shown to be tight by simulation, especially in high feedback regime. Based on this result, the scaling laws of feedback bits for channels, precoding and decoding vectors to obtain full degree of freedom are derived, which are backward compatible to the existing results for quantized CSI only or quantized precoder only. Simulation results show that the system is interference-limited at high signal-to-noise ratio if the three scaling laws are not satisfied simultaneously, and the throughput performance is limited dominantly by the worst feedback links among the channels, precoders, and decoders.