Cognitive channel selection for Wireless Sensor communications

Abstract

Wireless sensor networks (WSNs) are dense networks affected by severe interference among many devices. They operate in the unlicensed 2.4 GHz band that is shared by different technologies, such as Bluetooth and WiFi, which add interference at higher transmission power. For this reason, interference is an important factor to avoid for reliable communications. Due to the unpredictable nature of the wireless medium, the 802.15.4e standard has introduced the possibility to schedule a channel in frequency using Time Division Multiple Access (TDMA), but the selection of the optimal channel is still an ongoing research. In this paper, a Multilayered Feedforward Neural Network (MFNN) is proposed as a possible solution to make predictions about which channel can offer low latency and high throughput at any given time slot. Controlled experiments were conducted in an anechoic chamber, considering the two scenarios of no interference and interference incurred by other sensors and WiFi. Results show that MFNN is a valid solution, obtaining performance comparable to the best case scenario.