Application-driven adaptive fixed-point refinement for SDRs

Abstract

Wireless interfaces implement and increasing number of different standards. For cost effectiveness, flexible radio implementations are preferred over the multiplication of dedicated solutions. Software Defined Radios (SDR) have been introduced as the ultimate way to achieve such flexibility. However, the reduced energy budget required by battery-powered solutions makes the typical worst-case static dimensioning unaffordable under highly dynamic operating conditions. Instead, energy-scalable algorithms and implementations are entailed to provide flexibility while maintaining the required energy efficiency. Particularly, energy-scalable implementations can exploit data-format properties to offer different tradeoffs between accuracy and energy. In this paper, an application-driven adaptive fixed-point refinement methodology is proposed. The latter derives the minimum word-lengths which respect a user-defined degradation on the application performance. This technique is applied to the fixed-point refinement of a Near-ML MIMO (Multiple Inputs, Multiple Outputs) detector. Variations on the minimum required precision depending on external conditions are made explicit. Finally, on a processor platform these variations can be translated into reduced cycles and energy by leveraging on sub-word parallel implementations.