Adaptive Retransmission with Balanced Load for Fault-Tolerant Distributed Detection in Wireless Sensor Networks

A fault-tolerant classification system in wireless sensor networks combining distributed detection with error-correcting codes has recently been proposed. Associated with a decision pattern, each sensor makes a local decision based on its detection result and a set of decision thresholds. The detection result must then be transmitted to a fusion center to make a final decision. The probability of misclassification when adopting this approach is high when the transmission channel is highly noisy. This work first describes an adaptive retransmission algorithm to reduce the misclassification probability. The fusion center calculates the reliability of each local decision received while making the final decision. If the final decision is not reliable, then the fusion center asks the sensor that has sent the received local decision with the lowest reliability to retransmit its decision. However, when some sensors have unrecognized faults, the fusion center tends to query the sensor with the same decision pattern as the faulty sensor to retransmit its decision. This tendency causes unbalanced network load. This work further presents a novel adaptive retransmission algorithm with balanced load to combat this problem. Each sensor carries all sets of decision thresholds. A sensor is randomly selected when the decision based on a set of thresholds must be retransmitted. The selected sensor then makes its new decision according to the threshold set and its detection result. The random selection means the load is balanced. Simulation results show that the misclassification probability can be effectively decreased through the retransmission with a balanced load.