I am Usman and I am looking to apply to GNU Radio for this year's Google Summer of Code.
I am a final year bachelor student at National University of Science and Technology, Pakistan. My final year project is on using passive radars for indoor human motion sensing, during which I have gained extensive familiarity with USRP B200 and GNU Radio. Currently, I am working on a machine learning project that uses deep learning (CNN) to detect signal modulation for 11 modulation schemes. This research is inspired from this work with the exception that we are developing an extensive dataset for various indoor and outdoor environments and we plan to open source this data in around a week.
As a part of GSOC 2017, I want to mainly extend gr-radar and develop various signal processing modules of impact to users. I will be writing a detailed proposal in a few days but for now I have the following ideas:
- Computing cross-ambiguity function and providing 3D visualisation
- Adaptive filter implementation to remove the direct transmitted signal in case of passive radar systems which have reference and surveillance antennas
- Batched cross correlation implementation in real time with overlaps between batches to ensure continuity
- Angle of arrival estimation (although I recently found out about https://github.com/kit-cel/gr-specest which does something similar)
I am also very interested in extending the machine learning capabilities of GNU Radio and create OOT modules that allow users to import models from scikit-learn (such as decision trees and KNNs) and tensorflow (such as CNN and LSTM).
Ideally, I want to implement these signal processing and machine learning techniques together in a single module as I believe they go hand-in-hand. I am also open to any other ideas. Please let me know if there is any feedback as I write the first proposal draft!
Best,
-- Usman