Collision Recognition in Multihop IEEE 802.15.4-Compliant Wireless Sensor Networks

Collisions caused by the hidden terminal effects may result in severe packet corruption and performance degradation in multihop IEEE 802.15.4-compliant wireless sensor networks (WSNs). In order to avoid such collisions through scheduling protocols, it is important to first recognize these collisions by distinguishing them from some other noncollision cases (e.g., path loss, multipath fading, shadow fading, and IEEE 802.11 interference), which may also lead to similar consequences. In this paper, we focus on the collision recognition problem in multihop IEEE 802.15.4-compliant WSNs. First, through a series of measurements of the error properties in various collision and noncollision scenarios, we investigate the statistical behaviors of error patterns including the bit error rate and error position distribution, which reveal obvious differences between collision and noncollision cases in terms of bit- and symbol-level error position distribution. Based on these observations, we further propose a machine learning-based collision recognition mechanism by inserting some redundant blocks in a data frame. The inserted blocks are known to both the sender and receiver, thereby it enables the receiver to recognize the error patterns only according to the redundant sequences. Moreover, a mutual information-guided byte selection technique is also provided to effectively improve the recognition accuracy. Finally, the proposed mechanism is verified under three different transmission environments. The experimental results show that the proposed mechanism achieves good recognition accuracy over 90% with 94% coding efficiency.