Dual-Component Autoregressive Waveform Modelling of Periodic Impulsive Transients in SWER Networks

This study presents an autoregressive (AR) modelling framework for characterising the waveform of recurring impulsive transients measured in single-wire earth return (SWER) distribution networks. The framework uses separate second-order high-frequency (HF) and low-frequency (LF) polynomials to reproduce the two dominant spectral components of the measured impulse at approximately 45 kHz and 18 kHz, respectively. The coefficients are derived from field measurements and are therefore specific to the network and impulse type used for identification. Model order is selected using an explicit pole-based criterion that balances the target resonance, exclusion of out-of-band modes and model complexity. All reported AR models are stable, with a maximum pole radius of 0.9945. Under the stated evaluation protocol, the dual-component response achieves a mean squared spectral error of 134 dB2 over DC to 2.5 MHz, equivalent to a root mean square (RMS) spectral error of 11.6 dB. The corresponding errors are 387 dB2 for an amplitude-modulated white noise model and 835 dB2 and 880 dB2 for two frequency-shaped rectangular pulse models. The framework reproduces the amplitude, width and dominant spectral components of the measured impulse, and its application to a second SWER network is demonstrated. The present model describes the waveform of an individual recurring impulse; modelling its arrival process, validating on held-out impulses and assessing communication-level performance are identified as future work.