SLNet: A Spectrogram Learning Neural Network for Deep Wireless Sensing
summary
SLNet generates super-resolution spectrograms with a learned spectrogram enhancement network (SEN) trained on synthetic 1–5 component spectra, then applies a polarized complex-valued CNN (PCN) that modulates linear phase across frequency. Evaluated on four Wi‑Fi sensing tasks—gesture recognition (1 Tx, 6 Rx), human ID, fall detection, breathing estimation—the pipeline yields top-tier accuracy while being one of the smallest, lowest‑compute models reported; breathing estimation error drops by about 70% after enhancement.
AI-written summary · about
chapters (9)
- 0:00Introduction to RF sensing
- 2:13Challenges in signal data
- 4:00Spectral leakage problems
- 5:54Signal enhancement network
- 7:19Custom neural network design
- 9:35System architecture overview
- 10:43Applications and datasets
- 12:32Performance validation
- 14:40Results and conclusion
related talks
13mEnhanced EEG Data Classification with 3D SNN and On-Line Learning ESN
06mTeaching RF to Sense without RF Training Measurements
06mDeepRange: Acoustic Ranging via Deep Learning
24mAn Open Channel Identifier using GNU Radio
45mThe Sound of Your Secrets: Teaching Your Model to Spy So You Can Learn to Defend
28mHow to use Deep Learning to process audio data and classify sounds.
06mVibroSense: Recognizing home activities by deep learning subtle vibrations on an interior ...
36mU-Net-style neural networks for feature identification in 1D time-series