Echo Suppression Algorithm Selection Based on Signal Magnitude
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing echo suppression devices face a heavy arithmetic load, which hampers effective echo removal during voice communication systems.
Innovation Solution
The solution involves selecting between nonlinear and linear learning algorithms based on the magnitude of the reference signal, with linear processing used for smaller signals to reduce computational load and nonlinear processing for larger signals to effectively remove echo, and optionally converting signals into the frequency domain for more precise echo removal across frequency bands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a learning algorithm using nonlinear processing is used to remove echo, then echo removal effectiveness is improved, but processing load on arithmetic device increases
Solution Approach 1:
The patent dynamically switches between linear and nonlinear learning algorithms based on the magnitude of the reference signal. When the reference signal magnitude is large (indicating potential nonlinear distortion), the nonlinear learning algorithm is selected for effective echo removal. When the reference signal magnitude is small, the linear learning algorithm is selected to reduce processing load. This dynamic adaptation resolves the contradiction by matching algorithm complexity to actual signal conditions.
Solution Approach 2:
The patent changes the processing parameter (algorithm type) based on the magnitude of the reference signal. By monitoring the reference signal magnitude and switching between linear and nonlinear algorithms, the system adapts its processing characteristics to balance echo removal effectiveness with processing load, resolving the technical contradiction between reliability and device complexity.
2Device complexity
If a learning algorithm using linear processing is used to reduce processing load, then processing load on arithmetic device is reduced, but echo removal effectiveness deteriorates in cases of nonlinear distortion
Solution Approach 1:
The system dynamically adjusts the learning algorithm selection based on reference signal magnitude. When nonlinear distortion is detected (large reference signal magnitude), the system switches to nonlinear learning algorithm to maintain echo removal effectiveness. When reference signal magnitude is small, linear algorithm suffices and reduces processing load. This dynamic behavior resolves the contradiction by adapting to actual signal conditions.
Solution Approach 2:
The processing parameter (algorithm selection) is changed based on the magnitude of the reference signal. This parameter change allows the system to use computationally efficient linear processing when appropriate, while switching to more effective nonlinear processing when signal conditions demand it, thereby resolving the contradiction between processing load and echo removal effectiveness.
3Reliability
If echo suppression processing is performed continuously, then echo removal reliability is maintained, but processing load increases during low echo conditions
Solution Approach 1:
The patent applies partial processing by selecting the appropriate level of algorithm complexity based on actual echo conditions. Instead of always applying full nonlinear processing, the system uses linear processing (partial action) when reference signal magnitude is small, reducing processing load while maintaining sufficient echo removal reliability. This resolves the contradiction by avoiding excessive processing during low echo conditions.
Solution Approach 2:
The system changes the processing parameter (algorithm type) based on reference signal magnitude monitoring. When the magnitude is small, linear algorithm is selected to reduce processing load. When magnitude is large, nonlinear algorithm is selected to maintain reliability. This parameter adaptation resolves the contradiction between continuous processing reliability and processing load during low echo conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An echo is allowed to be effectively removed while a processing load on an arithmetic device is reduced. When a magnitude of a reference signal transmitted to a speaker is equal to or greater than a first threshold, a learning algorithm using nonlinear processing is selected. When the magnitude of the reference signal is smaller than the first threshold, a learning algorithm using linear processing is selected. The echo is learnt using the selected learning algorithm to remove an echo included in an input signal picked up by a microphone based on a learning result.