Author ORCID Identifier

https://orcid.org/0009-0001-4788-0207

Semester

Summer

Date of Graduation

2026

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Natalia A Schmid

Committee Member

Kevin Bandura

Committee Member

Duncan R Lorimer

Committee Member

Matthew C Valenti

Abstract

Radio Frequency Interference (RFI) increasingly contaminates the radio astronomy spectrum, often exceeding astronomical signal amplitudes by 50–70 dB. Reliable detection and mitigation are therefore essential for studies of faint transient phenomena such as pulsars and fast radio bursts (FRBs). Existing practical methods including Spectral Kurtosis (SK), Median Absolute Deviation (MAD), and SumThreshold perform well in many settings but depend on assumptions about the RFI environment and data statistics, limiting their effectiveness for weak, broadband, or nonstationary interference.

We develop a transform based RFI detection method that requires no prior knowledge of the RFI origin or type. The method operates on spectrograms, where the power spectrum is displayed as a two dimensional array of time and frequency. Because each frequency channel is relatively wide (195 kHz), it can be transformed into a secondary spectrogram using a Short Time Fourier Transform (STFT). The magnitude of this secondary spectrogram is treated as an image that can be segmented into RFI / non-RFI regions using a variety of image segmentation methods, producing a binary RFI mask. The mask is multiplied with the STFT transformed data, which are then inverse STFT transformed and reinserted into the original spectrogram. We analyze the performance of several segmentation algorithms including energy based methods, Gaussian Mixture Models (GMM), Iterative Self Organizing Data Analysis Technique (ISODATA), and the Segment Anything Model (SAM) as well as combinations thereof.

Data from the Green Bank Telescope (GBT) containing the well characterized pulsar PSR J1713+0747, with 4096 frequency channels spanning 1.1–1.9 GHz and 5.12 us sampling, are used to test our methods. Performance is assessed using the signal to noise ratio (S/N) of the folded profile of the cleaned PSR J1713+0747 signal. SK, MAD, Spectral Relative Entropy (SRE), Shapiro–Wilk (SW), and SumThreshold serve as baseline methods for comparison. Experimental results show that refining each channel’s frequency content via STFT, followed by segmentation in the STFT domain, yields measurable improvements in RFI suppression over many baseline methods and across differing RFI morphologies.

Download

Share

COinS