K-Means Initialisation for the Modelling of PLC Noise in Indoor Environment

This paper presents the modelling of the powerline communication (PLC) noise using the Gaussian mixture model (Gaussian MM), where the K-means algorithm is employed to initialise the expectation maximisation (EM) algorithm. The Gaussian MM has been shown to adequately model the unpredictable PLC noise amplitude distribution, which is the primary cause of reduced PLC performance. However, the EM algorithm used in the estimation of the maximum likelihood parameters is known to be highly sensitive to the initial parameters. As a result, the K-means algorithm is proposed as a means of initializing the parameters. The resulting Gaussian MM is then validated through measurements and is observed to provide a good fit to the measured data. The performance of the K-means initialised model is then compared to the randomly initialised Gaussian MM. It is observed that the K-means initialised Gaussian MM generally performs better, although the difference in performance with the randomly initialised model is minimal.