Flexible FSK Learning Demodulator
Paul Gorday, Nurgün Erdöl, Hanqi Zhuang · 2018
Frequency Shift Key (FSK) modulation formats have been widely used in wireless personal area networks such as Bluetooth LE and IEEE 802.15.4. FSK demodulation often involves nonlinear processing to recover phase or frequency information from the carrier. This complicates analysis and derivation of optimum detectors, but it is well suited to learning-based approaches such as neural networks. In this paper, we propose a neural network approach for the design of FSK Learning demodulators, or simply FSKL demodulators. FSKL demodulators offer several potential benefits, including adaptability to non-Gaussian, non-stationary, or non-linear channels. Their performance relative to conventional FSK demodulators is attractive for low-cost, low-power, wireless Internet-of-Things applications, and the flexible topology is a step toward more general, reconfigurable IP that supports multiple standards and allows performance updates as modeling and techniques improve.