Location Identification Using Power and Audio Data Based on Temporal Variation of Electric Network Frequency and Its Harmonics

Abstract

During the past decade electric network frequency(ENF) is being exploited in forensic research. In this paper, spatio-temporal variation of the electric network frequency which is the supply frequency of a power distribution line (50 or 60 Hz), is efficiently utilized to develop an authentic region-of-recording identification scheme. The ENF data extracted from power grids vary with respect to time-stamp and location-stamp because of load fluctuations in different locations. These ENF variations hold some recognizable patterns of a specific grid. In the proposed work, a precise method of ENF detection from the raw power and audio signals is presented based on Root MUSIC algorithm. Moreover, various spectro-temporal features are extracted from the extracted ENF and its harmonics. Finally a set of features is proposed to utilize in multi-stage supervised classification scheme with a binary support vector machine classifier. The classification performance is tested on power and audio data collected from nine grids corresponding to different countries and a very satisfactory classification performance is obtained.