Message Alignment for an Information Bottleneck Optimized FFT in OFDM Receivers

Abstract

The trend towards larger bandwidths to accommodate higher data rates is unbroken in mobile communications. High sampling rates resulting from these large bandwidths can lead to critical energy consumption in the analog-to-digital converter (ADC) particularly in mobile devices. As the ADC energy demand depends exponentially on the bit resolution, an increase in energy consumption can be avoided by coarse quantization. The ADC can be designed to maximize the mutual information between its output and the channel input using the information bottleneck (IB) method. This approach can also be used to replace computationally expensive arithmetic operations with mutual information maximizing look-up tables (LUTs). We consider an orthogonal frequency division multiplexing communication link and substitute arithmetic computations inside the fast Fourier transform (FFT) with LUTs. Replacing every FFT butterfly operation with an individual LUT quickly becomes infeasible for large FFT sizes due to exploding memory requirements. Instead, a technique called message alignment is used ensuring similar meanings of cluster indexes at different positions in the IB-FFT butterfly network. In the proposed message aligned IB-FFT, butterfly nodes are replaced by separately optimized LUTs for multiplication with twiddle factors and for addition. Message alignment allows to reuse these LUTs saving valuable memory resources.