Information Bottleneck Receivers for ISI Channels

This paper leverages the information bottleneck method to design receivers for ISI channels working with coarsely quantized messages. The proposed equalizer is based on the forward-backward algorithm. In contrast to conventional approaches, state reliability information is exchanged with a finite alphabet message instead of a real-valued probability vector. Moreover, each node update performs only a single lookup instead of many arithmetic operations. The lookup table construction aims at maximizing the preserved relevant mutual information. We propose a symmetric KL-means IB algorithm applied in an iterative discrete density evolution procedure. Based on the resulting distributions, a new static design technique creates three reusable symmetric lookup tables to perform forward, backward and final node updates. This way, coarse quantization is an integral part of the system design in the first place rather than a separate follow-up process. Two different ISI channel setups show that the proposed equalizers can achieve comparable performance to the high-resolution alternatives. Remarkably, the state metrics require an order of magnitude fewer bits, which potentially improves area and energy efficiency. Also, a lookup table sharing approach is presented to spread the implementation cost across sub-block equalizers operating in parallel.

Options

Information Bottleneck Receivers for ISI Channels