| Name: Kaushal Buch |
| Affiliation: Giant Metrewave Radio Telescope, NCRA-TIFR |
| Conference ID: ASI2026_954 |
| Title: Recent Developments in the Statistical Analysis and Automated Identification of Powerline RFI at uGMRT |
| Abstract Type: Poster |
| Abstract Category: Facilities, Technologies and Data science |
| Author(s) and Co-Author(s) with Affiliation: Kaushal D. Buch(Giant Metrewave Radio Telescope, NCRA-TIFR, Khodad - 410504, India), Saptarshi Nag(Giant Metrewave Radio Telescope, NCRA-TIFR, Khodad - 410504, India), Sudhir Phakatkar(Giant Metrewave Radio Telescope, NCRA-TIFR, Khodad - 410504, India), Divya Oberoi(National Centre for Radio Astrophysics, TIFR, Pune - 411007, India) |
| Abstract: Radio Frequency Interference (RFI) from high-tension power lines is a significant challenge for the upgraded GMRT (uGMRT). A real-time RFI mitigation system has been developed and is routinely in use at the uGMRT. While this system is successful in mitigating the powerline interference in real-time using a statistical signal processing system, we are exploring techniques for further refinements to real-time RFI mitigation. For the next level of techniques, over the last few years, we have been investing efforts towards understanding the powerline RFI signal and its manifestation in the uGMRT receiver. Powerline RFI is a composite signal whose statistical properties vary based on the observing frequency, direction, time of day, season, and propagation effects, making it difficult to generalise its statistical properties. Hence, a large database of high time-resolution signals affected by powerline interference is being gathered to understand its properties. We describe the various components of the signal, their properties, and the variability across different recording runs.
Using this database, we developed a tool to explore and classify powerline RFI as individual RFI events, a ‘bunch’ of RFI events, or a collection of bunches. Primarily, the algorithm uses a combination of the first-order derivative of kurtosis and an adaptive thresholding technique based on the knee-point method. We show that the accuracy of identifying the bunches on the recorded uGMRT data is 0.8588, with balanced precision (0.8735) and recall (0.8572), resulting in an F1 score of 0.8584. The tool is further extended to explore Deep Learning (DL) using the Convolutional Neural Network (CNN) and feature vector extraction on the bunch-detected data. We are working on expanding the existing RFI database in terms of its size and variety, and also implementing a more comprehensive approach to building a wider and richer database for future algorithm development.
|
