Packet-Level Failure Classification by Characterizing Failure Patterns in Wireless Sensor Networks

Packet failure classifiers help improve wireless transmission performance by identifying packet failure causes. Since different configurations should be adopted for packet retransmissions according to different packet failure causes, knowing the real-time failure causes can avoid further retransmissions. We have the observation that real-world packet failure causes may interleave in packet level so that adjacent packet failures may occur with different causes. However, existing classifiers introduce a long classification delay, during which the channel condition probably has changed. We propose a packet- level failure classifier, which identifies packet failure causes for each corrupted packet. The classifier characterizes failure patterns of byte-level RSSI and packet-level LQI for packet failure causes of collision and weak link. We have implemented the classifier in TelosB nodes in an indoor environment. The salient feature of the proposed classifier is that the classification delay is reduced to packet-level, as short as 23 ms. Test-bed experiments show that the classifier has a classification accuracy of 93%. Moreover, a throughput improvement of 110% is achieved by avoiding 70% packet retransmissions.