Prediction of long-range dependence in cyclostationary noise in low-voltage PLC networks

Exponential demand for broadband applications has elicited renewed interest in characterizing noise in power line communication (PLC) networks for optimal error performance analysis. PLC is economical and ubiquitous, however due to multipath phenomena and nonstationary impulsive noise, the PLC channel is very harsh to signal transmission. Recent studies reveal that PLC noise is cyclostationary and modelling its characteristic should exploit the spectral and temporal redundancy. In addition, preliminary research shows that PLC noise exhibits self-similarity and can be predicted by fractal theory. In modelling noise employing fractal theory, the Hausdorff dimension (D) needs to be approximated. D is related to Hurst parameter (H), which is a measure of self-similarity. Hence, it is important to estimate H for accurate noise prediction using fractal theory. H parameter estimators assume stationarity in time series data. Therefore, preprocessing is required for the case of PLC noise before estimating H to make it stationary as periodicity and nonstatioanrity affects the H estimators. In this study, we remove the periodicity hidden in the PLC noise obtained from measurement campaign within several university offices, workshops and laboratories. Then long-range dependence is estimated from this pre-processed noise data by employing three estimation algorithms: R/S (rescaled adjusted range) method, aggregated variance method and absolute values of aggregated series method. Results from the said estimation methods though not the same, show that the Hurst parameters obtained from various noise measurements are greater than 0.5, implying presence of self-similarity with long-term memory. Results from this study will find valuable applications in PLC noise modelling using fractal and chaos theory and in error performance analysis in both narrowband and broadband PLC systems.