Classifying WLAN Packets from the RF Envelope: Towards More Efficient Wireless Network Performance

Abstract

This paper describes Packet Assay, a power efficient sparse neural network (NN) that can discriminate between wireless transmissions, such as WLAN packets, based solely on the RF signal envelope, a feature that can be measured with much less power than fully demodulating and decoding the packets. The NN was trained on a Wireless Local Area Networks (WLAN) dataset developed in-house with over 600K labeled samples and achieved above 88% accuracy while maintaining a memory footprint of only 4.9KB. This approach can reduce the power consumption of wireless modules (WM), can minimize the signal processing in IoT devices, and provides a foundation for future protocol development.