This work presents an enhanced sensing framework for MEMS gas sensors based on tunable-amplitude periodic modulation, enabling multi-state excitation and feature enrichment without increasing the number of sensing elements. A multi-level periodic driving scheme is introduced to realize sensor virtualization, and the resulting multi-state responses are processed using a short-term baseline-tracking algorithm and a dislocated sparse-sampling strategy to improve feature discrimination. A lightweight multilayer perceptron (MLP) classifier is subsequently optimized and deployed on a field-programmable gate array (FPGA)-based accelerator to enable gas recognition under constrained hardware resources. Experimental results obtained from ternary mixtures of CH4, CO, and H2 demonstrate a classification accuracy of 98.5%, accompanied by a 60% reduction in model size and a fivefold improvement in computational speed on the FPGA accelerator.
Jiao et al. (Tue,) studied this question.