Improving the Listening Experience for SSB-Modulated HF Transmissions Using Neural Networks / von M.Sc. Jens Heitkämper ; Erster Gutachter: Prof. Dr.-Ing. Reinhold Häb-Umbach, Zweiter Gutachter: Prof. Dr.-Ing. Gerhard Schmidt. Paderborn, 2023
Content
- Abstract
- Acknowledgments
- Introduction
- High Frequency Radio Transmission
- Related Work
- High Frequency Radio Database
- Transmission
- Clean signal design
- Data preparation
- Marker detection
- Post-processing
- Database
- IQ Recordings
- Concurrent speakers
- Simulating high frequency radio signals
- Summary
- Speech Activity Detection
- Architecture
- Segment recurrent neural network
- Cost function
- Baseline
- Evaluation
- SAD on optimally demodulated speech
- SAD on recordings with concurrent speakers
- SAD on frequency shifted speech
- SAD on IQ signals
- Summary
- Carrier Frequency Offset Estimation and Correction
- RAKE estimator
- NN-based estimators
- CFO correction
- Evaluation
- Database
- Performance measures
- NN parameter tuning
- CFO estimation
- CFO estimation on IQ signals
- CFO correction
- Summary
- Speech Enhancement
- System overview
- Encoder and decoder
- Mask estimator
- Cost function
- Evaluation
- Database
- Evaluation Measure
- Top- and baseline
- Source separation
- Noise reduction
- Noise reduction on an unseen language
- Summary
- Multi-Stage Processing
- Evaluation of the enhancement stage
- Performance measure for imperfect activity information
- Evaluation of the multi-stage system
- Impact of an unseen language
- Summary
- Conclusion and Outlook
- Appendix
- Analysis of the delay correction on the recording
- Comparison of real to simulated training data ratios
- Evaluation on Russian data with a CFO
- CFO estimation on overlapping speakers
- Enhancement with oracle masks
- Small encoder and decoder windows for NR on real HF signals
- Evaluation of noise reduction on signals with a CFO
- Acronyms
- Bibliography