Statistical impulse detection of in-vehicle power line noise using hidden Markov model

Control signal networks in vehicles using dedicated communication cables such as twisted wires and fiber optics cables have been built in order to communicate between electronic control units (ECUs). However, there are problems such as the increase in weight of the cable harness and the difficulty to insure the reliability of complicated network architectures. Communication systems using in-vehicle power lines have been investigated. This paper proposes an impulse detection scheme and its four criteria to decide the impulse occurrence in a communication cycle by using the two-state hidden Markovian-Gaussian noise model for noise on in-vehicle power lines. The new scheme involves the maximum a posteriori (MAP) estimation using the Baum-Welch (BW) algorithm and the moment method to estimate the model parameters. The proposed scheme is applied to noise generated from the hidden Markov model and noise observed on in-vehicle power lines and investigate their detection accuracy. It is shown that the detection accuracy is substantially improved by using an impulse occurrence criterion based on the number of free parameters.