Power grids are available all over the world for their necessity in all our everyday activities. Each power grid has its own electrical network frequency pattern which is considered as a signature or a finger print of this power grid. Any audio/video recorded in a power grid, whether directly connected to the power mains or not, is affected by the ENF signature of this grid. Post processing could take place to extract the ENF pattern from recordings which can be used in localization of different media signals recorded among grids. This can be done through computing the cross correlation between extracted ENF and the original ENFs of the grids. Finding the time of recording becomes possible. There are much more applications for ENF extraction and recognition.
In this project we have worked on data set of 9 (A to I) different grids provided by the contest. ENFs of these power grids are extracted for power and audio recordings then characteristic features of these ENFs are extracted and accurately selected. These features are used to feed two machine learning models: SVM and neural network. These machine learning models are used in correlating the practice data set, provided by the contest, and the grids from A to I. We also implemented a simple hardware circuit to record about 24 hours from our grid mains. Here we present our work methodology and propose our developed code for ENF extraction and recognition.