TL;DR
This paper introduces a novel algorithm for digital hearing aids that simultaneously tackles acoustic feedback (howling) and background noise. By intelligently adapting filters to target specific oscillation frequencies and using an oscillation detector, the system significantly improves both feedback cancellation and noise reduction, leading to better sound quality for users.
What Problem Does This Paper Address?
Hearing aids are designed to amplify sound for individuals with hearing loss. However, a common and annoying problem is acoustic feedback, which manifests as an unwanted oscillation or "howling" sound. This occurs when the amplified sound from the hearing aid's receiver leaks back into the microphone, creating a positive feedback loop. Additionally, background noise can degrade the performance of the hearing aid and reduce the clarity of desired sounds. Existing solutions often address these issues separately or use wide-band approaches that can introduce distortion. This paper aims to provide an integrated solution that is more efficient and less prone to distortion by specifically targeting the frequencies where feedback occurs and by employing advanced noise reduction techniques.
Methods
The core of the proposed solution is an integrated algorithm that combines an Adaptive Feedback Cancellation (AFC) module with a General Sidelobe Canceller (GSC) for noise reduction. Both modules utilize a band-limited adaptation strategy. This means the adaptive filters are constrained to operate only within specific frequency ranges, particularly those known to be associated with feedback oscillations.
Key innovations include:
- Band-limited AFC: A band-pass pre-filter is used to focus the AFC on the problematic oscillation frequencies, making it more efficient and reducing distortion.
- Band-limited GSC: A band-stop pre-filter is applied to the GSC to enhance the cancellation of interfering noise while preserving desired speech signals.
- Oscillation Detector: A novel detector is introduced that analyzes the zero-crossing rates of the autocorrelation of sub-band signals. This detector accurately identifies the center frequency of oscillations, allowing the band-limited filters to adapt dynamically to changing feedback conditions. This is crucial because the feedback path can vary with the user's environment.
- Frequency Domain Normalized Least Mean-Square (FNLMS): This algorithm is employed for adaptive filtering in both the AFC and GSC modules, providing efficient adaptation in the frequency domain.
Datasets
Not discussed in the paper. The simulations use synthetic data generated to mimic hearing aid scenarios.
Key Insights and Ideas
- Targeted Adaptation is Key: Constraining adaptive filters to specific frequency bands (band-limited adaptation) is significantly more effective for feedback cancellation and noise reduction than wide-band approaches.
- Oscillation Detection is Crucial for Robustness: A dynamic oscillation detector that can accurately pinpoint feedback frequencies is essential for adapting the filters in real-time, especially when the acoustic environment or feedback path changes.
- Integration Synergizes Benefits: Combining feedback cancellation and noise reduction within a single framework, using complementary band-limited strategies, leads to superior overall performance.
- GSC as an Interference Canceller: The GSC is effectively used to treat the feedback signal as a form of interference, allowing it to steer a null towards the receiver and further improve sound quality.
Key Results
The paper presents simulation results demonstrating the effectiveness of the proposed BL-FNLMS-OSD algorithm compared to standard FNLMS and BL-FNLMS.
- Feedback Cancellation: The BL-FNLMS-OSD algorithm achieved a significant reduction in steady-state error for the feedback canceller (W), with nearly 15 dB improvement over FNLMS and an additional 10 dB improvement over BL-FNLMS after the feedback path was altered.
- Noise Reduction: The convergence speed of the noise canceller (C) was significantly increased by the band-stop pre-filter and the oscillation detector, particularly when the feedback path changed.
- Subjective Evaluation: Informal subjective evaluations indicated that the BL-FNLMS-OSD algorithm provided improved feedback and noise reduction performance, even in noisy environments and at different gain levels.
Limitations and Future Work
- Limitations: The paper acknowledges that the performance of the band-limited filters is dependent on the fixed bandwidth of the pass/stop bands. While the center frequency is dynamically detected, the bandwidth itself is fixed in the current implementation.
- Future Work: The authors suggest that further improvements could be made by dynamically adjusting the bandwidth of the band-pass and band-stop filters based on the detected oscillation characteristics. This could lead to even more precise adaptation and reduced distortion.
Summary updated