Speaker
Description
For the first part of my talk, I will be focusing on different radio frequency interference (RFI) mitigation techniques currently being developed at Brown University and the University of Washington. At Brown, my home institution, our group's efforts are directed towards developing both traditional and machine learning (ML) algorithms to detect RFI for MWA data. On the traditional side, we have developed a novel RFI flagging method based on the $\chi^2$ metric from redundant calibration, which has been shown to identify a significant amount of RFI in the Australian DTV7 band that popular flaggers such as $\texttt{SSINS}$ and $\texttt{AOFlagger}$ fail to detect. We have also applied the watershed algorithm used by HERA as an additional RFI detector for MWA data. On the ML side, we are developing a U-Net model that processes waterfall plots to generate flag arrays. This project shows significant promise, with the overall goal being to implement a weak supervision framework that intelligently combines flags produced by traditional flagging algorithms in order to provide robust labels to the U-Net. This will enable it to detect RFI independently flagged by various traditional methods, which may not always overlap.
At the University of Washington, researchers are exploring the image-space representation of RFI. Preliminary results indicate that some forms of stationary RFI, often overlooked by traditional flagging methods, could be more easily detectable in image space. These findings suggest the need for new approaches to RFI flagging based on MWA data images.
For the second part of my talk, I will present my ongoing PhD research, which aims to develop a new method for estimating an RFI emitter's altitude using near-field corrections. Near-field corrections are used to change an interferometric array's focal distance from infinity to a location much closer, in the so-called near field of the instrument. The data used to conduct this work is a two-minute 2013 Phase I MWA observation, during which an RFI-emitting object briefly crosses the field of view. This emission can be visualized in image space using an imaging software such as $\texttt{WSClean}$. In addition to producing the image, $\texttt{WSClean}$ also conveniently outputs a source list containing the right ascension (Ra) and declination (Dec) coordinates of all sources it identifies during the cleaning process. By combining the RFI-emitting object's Ra and Dec with a variable focal distance parameter, which represents the distance between the object and the MWA array, we can apply near-field corrections to bring it into focus. This technique aims to fully localize an RFI emitter, with the ultimate goal being to subtract or peel it from the observation altogether.
Applying this novel technique to the observation in question, we estimate the RFI emitter's altitude to be $8.53 \pm 0.06$ km. Combining this measurement to the object's angular displacement as a function of time, also tracked using $\texttt{WSClean}$, we are able to estimate its speed to $579 \pm 4$ km/h. These findings allow us to confidently classify the previously unknown RFI emitter as an airplane.
| Timeslot preferences | Speaking timeslot length preference: 30min. No scheduling requirements! |
|---|
