Thesis progress report

On

A comparative study of Adaptive filters for noise reduction in hearing

aids
Submitted in Partial Fulfillment of the Requirements for the award of degree of

Master of Technology in Signal Processing

Submitted by:

BAMIN DIMING

(2024SP06)

Under the guidance of

DR. SUMIT KUMAR JHA

DEPARTMENT OF ELECTRONICS & COMMUNICATIONS ENGINEERING

MOTILAL NEHRU NATIONAL INSTITUTE OF TECHNOLOGY ALLAHABAD

PRAYAGRAJ-211004, INDIA
A COMPARATIVE STUDY OF ADAPTIVE FILTERS FOR
NOISE REDUCTION IN HEARING AIDS.
ABSTRACT
Hearing aids require efficient noise reduction to enhance speech intelligibility in various
acoustic environments. Adaptive filters, such as LMS (Least Mean Squares), NLMS
(Normalized LMS), and RLS (Recursive Least Squares), are widely used for active noise
reduction due to their ability to dynamically adjust filter coefficients based on signal
characteristics. This research proposes a comparative study of these adaptive filtering
algorithms in the context of hearing aid systems. The study will simulate these filters in
MATLAB and evaluate their performance using real-world and synthetic noisy speech
signals. Metrics like convergence rate, mean squared error (MSE), signal-to-noise ratio
(SNR), and computational complexity will be used for analysis. The goal is to identify
the most suitable adaptive algorithm for real-time, low-power implementation in digital
hearing aids.

INTRODUCTION
Hearing impairment is an all-too-common problem for millions of individuals worldwide,
having far-ranging consequences for their ability to communicate, interact socially, and
maintain quality of life. For those suffering from hearing loss, hearing aids represent an
important tool that magnifies sound and assists in the regeneration of the perception of
sound, facilitating easier engagement in everyday activities and enhanced mental health.
Hearing aids greatly benefit the quality of life for those suffering from hearing impairments
in that desired sounds—essentially speech—can be amplified, while undesirable background
noises can be reduced. Unfortunately, real-world environments, being acoustically
complicated and dynamic, cause enormous challenges in the process of noise reduction. Of
the many methods used for this problem, the method using adaptive filtering has proven
powerful and versatile.

All their innovations notwithstanding, hearing aids continue to experience significant

challenges in real environments—top among them being the problem of noise interference.
Under noisy environments like busy roads, eateries, and public transport, background noises
overpower the target speech sound, rendering the user unable to follow conversations.
Beyond speech intelligibility reduction, it causes user fatigue and frustration, ultimately
resulting in device abandonment.

Adaptive filters are widely implemented in contemporary hearing aids for real-time reduction
of noise. Adaptive filters change their parameters in response to varying acoustic conditions
in order to attenuate undesirable sound while retaining the clarity of speech. Adaptive filters
greatly contribute towards improving speech intelligibility and overall user experience in
hearing aids through the increase in the signal-to-noise ratio (SNR).
Adaptive filters possess the special ability to adapt their parameters in real-time according to
the statistical properties of the input signal. Due to the same, they are best suited for
application in hearing aids, where the speech and the characteristics of the noise may vary
very rapidly and in an unpredictable manner. In recent years, the adaptive filtering algorithms
have been extensively studied and implemented for the removal of noises, and the Least
Mean Squares (LMS), Normalized Least Mean Squares (NLMS), and Recursive Least
Squares (RLS) algorithms were the most widely studied among them.

BACKGROUND AND MOTIVATION

Hearing aids are typically small embedded systems equipped with microphones, amplifiers,
processors, and output speakers. While traditional hearing aids amplify all incoming sounds,
this can lead to discomfort in noisy environments. Adaptive filtering enables selective
enhancement of desired speech signals while suppressing background noise.

Among adaptive filters:

 LMS is simple and widely used but has slower convergence.

 NLMS improves LMS by normalizing the input signal, yielding more stable
performance.
 RLS achieves faster convergence and better accuracy at the cost of computational
complexity.

The motivation behind this work is to evaluate and compare these three algorithms
specifically for hearing aid applications using real-world crowd noise data, thus bridging the
gap between theoretical understanding and practical implementation for biomedical use.

SYSTEM ARCHITECTURE
The system designed in Simulink consists of the following key components:

1. Input Signals

 Primary Input: Crowd noise (interfering background noise)

 Desired Signal: Mixture of clean speech + crowd noise
 Goal: Extract clean speech by minimizing the error between output and desired signal

2. Adaptive Filter Block

 One of the LMS, NLMS, or RLS blocks

 Takes input and desired signal
 Outputs the error signal, which approximates the clean speech

3. Analysis Blocks

 Time Scope: To visualize signal behaviour

 Audio Writer: To hear the output and error signal
 To workspace block: To calculate:
o Signal-to-Noise Ratio (SNR)
o Mean Squared Error (MSE)
o Mean Absolute Error (MAE)
o Peak Signal to Noise Ratio (PSNR
IMPLEMENTATION IN MATLAB AND SIMULINK
[Link]

1 Audio was down sampled to 16000Hz and converted to mono.

2 Time-series format: first column is time, second is audio signal.
3 Assigned to Simulink via From Workspace.

2. Filter Configuration

 LMS Filter
o Step size: μ = 0.001
o Filter length: L = 64
 NLMS Filter
o Normalized step size: Ensures stability regardless of input power.
 RLS Filter
o Forgetting factor: λ = [0.99]
o Initial condition: zeros(L,1) where L is the filter length.

3. Performance Evaluation

 Output (error signal) passed through:

o To Time and Audio Writer block.
o To Workspace block to calculate SNR, MAE, MSR and PSNR.
o Compared values for all three filters under identical conditions.

ADVANTAGES AND LIMITATIONS

Advantages

 Provides a practical simulation of hearing aid performance using real crowd noise.
 Allows fair comparison between LMS, NLMS, and RLS.
 Simulink provides real-time audio feedback and visual signal tracking.
 Objective metrics (SNR, MSE, MAE) give measurable insight into performance.

Limitations

 Simulation uses offline pre-recorded signals; not tested with live microphone input.
 RLS, while most accurate, is computationally heavy—not ideal for real-time low
power devices like hearing aids.
 Does not account for hardware limitations like latency, microphone placement, etc.

WORK COMPLETED SO FAR

LMS Algorithm

 Implemented successfully in Simulink.

 Shows low MSE and MAE, meaning minimal speech distortion.
 SNR improvement: ~7.8 dB
  Spectrum shows moderate noise suppression but not aggressive filtering.

NLMS Algorithm

 Improved stability and convergence over LMS.

 Automatically adapts to input signal power.
 SNR improvement: ~20.88 dB (highest among all three)
 Spectrum analysis shows strong noise reduction across speech frequencies.

RLS Algorithm

 Fastest convergence but computationally heavy.

 SNR improvement: ~4.10 dB, lowest among three despite theoretical advantages.
 Spectrum analysis indicates uneven filtering and some instability.

CONCLUSION

Based on the completed simulations and analysis:

 NLMS provides the most balanced performance: high noise suppression, stability,
and low computational cost.
 LMS is simple and energy-efficient but less effective in noisy environments.
 RLS, although theoretically superior, did not perform as well in practical simulations
due to instability and overfitting.

Thus, NLMS emerges as the most suitable adaptive algorithm for noise reduction in hearing
aids.

FUTURE WORK

Hybrid Adaptive Filtering System

 Combine LMS, NLMS, and RLS using dynamic weighting or switching.

 The system can automatically select the best algorithm based on SNR or error.
 Expected benefits:
o LMS → stability
o NLMS → robustness
o RLS → rapid convergence

REFERENCES
[1] Haykin, S. (2002). Adaptive Filter Theory (4th ed.). Prentice Hall.

[2] Widrow, B., & Stearns, S. D. (1985). Adaptive Signal Processing. Prentice Hall.

[3] Benesty, J., Sondhi, M. M., & Huang, Y. (2008). Springer Handbook of Speech
Processing. Springer.

[4] Kuo, S. M., Lee, B. H., & Tian, W. (2006). Real-Time Digital Signal Processing:
Implementations and Applications. Wiley.
[5] Loizou, P. C. (2013). Speech Enhancement: Theory and Practice. CRC Press.

[6] Kolbæk, M., Yu, D., Tan, Z.-H., & Jensen, J. (2017). Speech Enhancement Using
Deep Learning with Temporal Context. IEEE/ACM Transactions on Audio, Speech, and
Language Processing.

[7] Harshitha L, Sindhu Kumari ND, Rekha C “Implementation of Noise Cancellation

using Adaptive Algorithms in GNU Radio”, 2022 IEEE International Conference on Data
Science and Information System (ICDSIS).

[8] Nikhil Prasad, Devesh Prakash and Shubhra Dixit “Implementation of Optimized
Adaptive LMS Noise Cancellation System to Enhance Signal to Noise ratio”, The Second
International Conference on Applied Artificial Intelligence and Computing (ICAAIC
2023).

[9] Parvathy, A. K. and Gayathri Narayanan, “Comparative Study of Adaptive

Algorithms Using Matlab and Verilog”, International Conference on
Wireless
[10] M Sugadev, Malladi Kaushik ,V Vijaya Kumar and KT Ilayaraaja , “Performance
Analysis of Adaptive filter Algorithms on Different Noise Sources”, 2022 International
Conference on Computer Communication and Informatics (ICCCI ).
[11] G K Girisha1 , S L Pinjare, “Performance Analysis of Adaptive Filters for Noise
Cancellation in Audio Signal for Hearing Aid Application”, International Journal of Science
and Research (IJSR).
[12] Adarsh J, P. Vishak and R. Gandhiraj, “Adaptive noise cancellation using NLMS
algorithm in GNU radio”, 4th International Conference on Advanced Computing and
Communication Systems (ICACCS), IEEE
.

SUBMITTED BY: SUMITTED TO:

BAMIN DIMING DR. SUMIT KUMAR JHA

Reg. No. - 2024SP06 (Supervisor)