Extracting Electrical Network Frequency From Digital Recordings Using Frequency Demodulation

Abstract

Electric power system instantaneous frequency, also referred to as electrical network frequency (ENF), pollutes digital recordings due to a variety of factors (e.g., transformer hum). This unwanted interference has been used as a forensic tool in conjunction with a database of ground truth ENF measurements to determine the time at which a recording was made, as well as to detect tampering. Recent work has suggested that that ENF could be used to geolocate a digital recording under the assumption that the low-energy location-unique power system dynamics are observable, which is not always the case in the presence of noise and interference. This letter aims to take advantage of the intrinsic nature of power system instantaneous frequency in order to frame the ENF extraction problem as one of FM demodulation. By considering the ENF as a sinusoid of nominal power system frequency that is frequency modulated by a weak stochastic signal, a 0-Hz Intermediate Frequency (IF) signal can be created and analyzed rather than direct measurements. In doing so, significant data reduction can be achieved through extensive downsampling, FM demodulation algorithms can be applied to extract the ENF, and the spectrum of the IF signal provides a straightforward way of estimating the maximum observable frequency content of the ENF buried in a digital recording. These findings are supported through the analysis of audio recordings made in the presence of ENF interference in the U.S. Eastern Interconnection power system.