A Channel-based Authentication Using Machine Learning for Body Sensor Networks

In Body Sensor Networks (BSNs), connected nodes on the body exhibit different channel characteristics than any other remote node, such as an attacker who wishes to disrupt the network. Based on this observation, several channel-based Physical Layer Authentication (PLA) protocols have used Received Signal Strength Indicator (RSSI) for authentication. Supported by experiments performed with ESP32 WiFi modules, we introduce a robust feature set that consists of statistical characteristics of RSSI and Frame Loss Rate (FLR) to design a lightweight authentication scheme. To create a realistic dataset in a BSN (using the IEEE 802.15.6 standard), Castalia was used with a network configuration of 6 sensor nodes, a controller node, and an attacker node. To learn the channel behavior of sensor nodes, we adopt a machine learning approach. The model has been trained with a dataset that includes the RSSI and FLR features of 6 sensors and 1 attacker and is capable of classifying received packets by sender, which allows us to detect any attacker trying to send data to the network. The accuracy of the proposed scheme is 95%. Moreover, true attack detection and false attack detection rate is 88%, and 4.1% respectively.

I. Introduction

Remote monitoring of in-hospital patients or chronically ill patients at home is becoming an essential component of any modern healthcare system. It not only improves the healthcare delivery but also reduces cost of care. The need for this feature is further pronounced with the onslaught of deadly viruses, such as COVID-19, where personnel traffic in and out of hospital units needs to be minimized. Furthermore, sensor-based vital sign monitoring can also help increase accuracy of the data captured.