Neural Network Compensates Digital Filter Distortions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Noise reduction in moving vehicles is challenging due to variable acoustic noise from vehicle speed, road conditions, and weather, which degrades the quality of speech and music by masking soft sounds and reducing intelligibility and fidelity.

Innovation Solution

A method and system combining a digital filter with a neural network to suppress noise from input signals, where the neural network is trained to compensate for distortions introduced by the digital filter, allowing for effective noise reduction and improved signal quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If a digital filter is used to suppress noise spectrum, then noise reduction is improved, but distortions are introduced in the signal of interest

Engineering Contradiction:
ImprovenoiseVSAvoidsignal quality
Core Design Contradiction:
Object-affected harmful factorsVSManufacturing precision

Solution Approach 1:

A neural network is introduced as an intermediary component between the digital filter and the final output. The neural network receives the filtered signal and processes it to compensate for the distortions introduced by the digital filter, thereby maintaining signal quality while retaining noise suppression benefits

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system dynamically adjusts processing parameters by using a neural network that can adapt to different signal conditions. The neural network learns to optimize the balance between noise suppression and signal fidelity by adjusting its internal parameters based on the input signal characteristics

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If noise suppression is applied to improve signal clarity, then intelligibility is improved, but processing time increases

Engineering Contradiction:
Improvesignal clarityVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The digital filter performs preliminary noise suppression in the frequency domain before the neural network processes the signal. This preliminary action reduces the computational burden on subsequent processing stages and enables more efficient real-time processing

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3698360B1Noise reduction using machine learning
Publication Date: 2024.01.24 BOSE CORP
  • EP3698360B1 patent drawingFigure 1
  • EP3698360B1 patent drawingFigure 2A~2B
  • EP3698360B1 patent drawingFigure 3

AI summary

The technology described in this document can be embodied in a method for processing an input signal that represents a signal of interest captured in the presence of noise to generate a de-noised estimate of the signal of interest. The method includes receiving an input signal representing a signal of interest captured in the presence of noise. The method also includes processing at least a portion of the input signal using a digital filter to generate a filtered signal, the digital filter configured to suppress at least a portion of spectrum of the noise. The method further includes processing the filtered signal using a first neural network to generate a de-noised estimate of the signal of interest, wherein the first neural network is trained to compensate for distortions introduced by the digital filter in the signal of interest.