An energy-aware framework for cascaded detection algorithms

Abstract

Low-power, scalable detection systems require aggressive techniques to achieve energy efficiency. Algorithmic methods that can reduce energy consumption by compromising performance are known as being energy-aware. We propose a framework that imposes energy-awareness on cascaded detection algorithms. This is done by setting the detectors' thresholds to make a systematic trade-off between energy consumption and detection performance. The thresholds are determined by solving our proposed energy-constrained version of the Neyman-Pearson detection criterion. Our proposed optimization method systematically determines the energy-optimal thresholds and dynamically adjusts to time-varying system requirements. This framework is applied to a two-stage cascade, and simulations show that our energy-aware cascaded detectors outperform an energy-aware detection algorithm based on incremental refinement. Finally, combining our framework with incremental refinement reveals a promising approach to the design of energy-efficient detection systems.