Echo Cancellation During Double-Talk Using Receive-Side Attenuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing echo cancellers in telecommunications systems are ineffective during double-talk conditions, especially when the echo path impulse response is beyond the coverage length of the convolution processor or is non-linear and time-variant, leading to imperfect echo cancellation.
Innovation Solution
The introduction of a receive-side attenuator that is activated during double-talk conditions to provide additional attenuation, coupled with a double-talk detector and convolution processor, which adjusts the amount of attenuation based on the residual echo and echo return loss enhancement values, and deactivates when double-talk ends.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the NLP is eliminated during double-talk to avoid degrading near-end speech, then speech quality is preserved, but echo cancellation effectiveness deteriorates
Solution Approach 1:
The echo canceller is divided into separate functional components: a convolution processor (CP) for linear echo cancellation and a non-linear processor (NLP) for residual echo suppression. During double-talk, the NLP can be selectively deactivated while the CP continues to operate, allowing speech quality preservation while maintaining partial echo cancellation capability.
Solution Approach 2:
The system dynamically adjusts the operation mode of the NLP based on double-talk detection. The Double-talk Detector (DTD) controls the NLP activation state, switching it off during double-talk periods and on during single-talk periods, optimizing performance for each condition.
2Stability of the object's composition
If the CP stops updating impulse response estimate during double-talk to avoid convergence issues, then algorithm stability is maintained, but echo cancellation effectiveness deteriorates
Solution Approach 1:
The CP adaptation is dynamically controlled based on talker activity detection. During double-talk, the CP updates are paused to prevent convergence to incorrect impulse response estimates, while during single-talk, full adaptation is enabled to maintain accurate echo modeling.
Solution Approach 2:
The Double-talk Detector provides feedback control to the CP adaptation process, monitoring the signal conditions and controlling whether impulse response updates should proceed, thereby preventing degradation while maintaining effectiveness when conditions are favorable.
3Length of stationary object
If the echo path impulse response is beyond the coverage length of the CP, then long echo paths are handled, but cancellation effectiveness deteriorates when the path is non-linear or time-variant
Solution Approach 1:
The echo cancellation function is segmented into two processors: the CP handles linear, time-invariant echo paths within its coverage length, while the NLP handles non-linear and time-variant components. This segmentation allows each processor to be optimized for its specific function.
Solution Approach 2:
The echo canceller uses a composite processing approach combining linear convolution processing and non-linear processing. The CP provides baseline linear echo cancellation while the NLP adds non-linear suppression capabilities, creating a composite system that handles both linear and non-linear echo paths effectively.
Data Source
AI summary
An echo canceller apparatus comprises a receive side attenuator coupled in a receive side signal path that is configured to couple from a conference call bridge to a caller; a convolution processor coupled to the receive side signal path at a convolution processor pick-off point; a double-talk detector coupled to the receive side signal path and to a sending side signal path that is configured to couple from the caller to the conference call bridge; and logic coupled to the receive side attenuator which when executed is responsive to a double-talk condition detected by the double-talk detector and operable to determine a level of echo canceled by the convolution processor, to determine an additional amount of attenuation to introduce, and to activate the receive side attenuator to introduce the additional attenuation.