Machine Learning Based Channel Error Diagnostics in Wireless Sensor Networks

Due to the unreliable nature of wireless links, sensor nodes frequently encounter degraded performance or lack of connection. This paper provides a real-time status monitoring and channel error diagnosis method in a wireless sensor network. It uses a poll and echo procedure to efficiently collect some physical layer and link layer statistics from both transmitter side and receiver side. A machine learning approach (k-nearest neighbor) is used for failure isolation. Our scheme involves learning the normal and anomalous behavior of the network via continued observation, and classifying future events and observations as normal or different classes of errors based on past experiences. With implementation on IEEE 802.15.4 SoC, our experimental results show that the proposed framework can diagnose the major wireless transmission errors with high accuracy.