Information Bottleneck Optimization of a Finite Bit Resolution Fast Fourier Transform

Abstract

The trend towards larger bandwidths in mobile communications requires coarse quantization to obtain energy efficient analog-to-digital converters. The question is whether quantization and subsequent processing can be optimized with respect to relevant mutual information. Current standards rely on orthogonal frequency division multiplexing (OFD M) as air interface and require the implementation of an fast Fourier transform (FFT). Inspired by the efficient implementation of low density parity check (LDPC) decoders via lookup-tables optimized by the information bottleneck (IB) method, we suggest to approximate the computation of intermediate quantities inside the FFT with IB-optimized lookup-tables being designed offline. Consequently, the FFT only relies on the mapping quantization indices for which the flow of relevant information is optimized.