Intelligent Filtering Tuning in Edge Computing

Abstract

An Artificial Intelligence approach to determine the best narrow input filter for a fundamental frequency extractor is presented. The algorithms are implemented on board a water current velocity sensor node. The target sensor node is based on an ARM Cortex-M0+ without DSP and FPU hardware support. The implementation is studied in detail in domains of real and integer variables. The results demonstrate that the proposed ANN-based solution is at least 5.5 and 15 times better than the published FFT solutions when using real variables and integer variables, respectively, for an input bandwidth reduction factor of 7.