Cyclostationary Noise Modeling Based on Frequency-Shift Filtering in NB-PLC

Abstract

In this paper, we propose a cyclostationary noise model for narrowband power line communication (NB-PLC) based on frequency-shift (FRESH) filtering. The FRESH filters are designed to shape an input white noise spectrum to a cyclic spectrum extracted from experimental noise measurements. The noise modeling problem is considered as a system identification problem that minimizes the time-averaged mean square error (TA-MSE) between a reference measured noise and the model generated noise. We propose the normalized mean square error (NMSE) in the cyclic auto-correlation between the measured and the generated noise as a performance measure. The noise waveform generated by the proposed model shows a good agreement with that of measured noise in terms of the cyclic auto-correlation NMSE. In addition, the numerical results show a performance enhancement, in terms of the cyclic auto-correlation NMSE, over the time-region-based NB-PLC noise model presented in the literature, at the same model complexity. Moreover, the proposed model provides a scalable performance-complexity tradeoff.