Abnormal-State-Clustering for In-Vehicle Cable Communication Using Equalizer Parameters and Machine Learning Approach

In this paper, an abnormal-state-clustering method for the in-vehicle communication is proposed. The proposed method uses a full set of an equalizer's tap coefficient parameters that can be obtained from serializer/deserializer devices. The difference in the tap coefficient parameters between the normal state and the abnormal state is calculated and used for a characteristic parameter of abnormal-state-clustering. We also utilize a machine learning approach to obtain high accuracy for abnormal-state-clustering from multidimensional parameters of the difference in the tap coefficient. We verified the effectiveness of our approach in a simulation. Classification accuracy high as 96.4% was achieved by using a decision tree with dimension reduction in the principal component analysis.