Double Talk Detection in Echo Cancellation Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In telecommunications networks, the presence of double talk complicates echo cancellation, causing echo cancellers to diverge from the actual echo, and existing detection algorithms may lead to false identifications and unnecessary inhibition or delay in adaptation processes.
Innovation Solution
A method is introduced to detect double talk by generating measures of average energy for near-end and far-end signals, using a decision statistic to determine the presence of double talk, and inhibiting or resuming the updating of filter coefficients based on this detection to prevent divergence during echo cancellation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing double talk detection algorithms are used, then double talk can be detected, but false identifications occur and adaptation processes are unnecessarily inhibited or delayed
Solution Approach 1:
The patent changes the detection parameter from simple signal presence to average energy measurement. By calculating the average energy of near-end and far-end signals over time and comparing their ratio to a threshold, the system achieves more reliable double talk detection that reduces false positives and unnecessary adaptation inhibition.
Solution Approach 2:
The patent implements a feedback mechanism where the detection result (double talk presence) directly controls the adaptation process (filter coefficient updating). When double talk is detected, adaptation is temporarily inhibited; when not detected, adaptation resumes, creating a closed-loop control system that improves both reliability and timing.
2Reliability
If echo canceller adaptation is continuously updated, then echo cancellation performance is maintained, but divergence occurs when double talk is present
Solution Approach 1:
The patent makes the adaptation process dynamic by conditionally enabling or disabling filter coefficient updates based on real-time double talk detection. The system transitions between adaptive and non-adaptive states, preventing divergence during double talk while maintaining echo cancellation performance during normal operation.
Solution Approach 2:
The patent extracts the harmful influence of double talk by detecting its presence and removing the adaptation process during these periods. By taking out the coefficient updating mechanism when double talk is detected, the system prevents divergence while preserving stability.
3Stability of the object's composition
If adaptation is inhibited during double talk, then divergence is prevented, but echo cancellation performance degrades when double talk is not present
Solution Approach 1:
The patent segments the operation into distinct phases: adaptation phase (when no double talk is detected) and protection phase (when double talk is detected). This segmentation allows the system to optimize for echo cancellation accuracy during normal operation while ensuring stability during double talk conditions.
Solution Approach 2:
The patent changes the operational parameter of the echo canceller based on detection results. During normal operation, the system uses adaptive filtering for accurate echo estimation; during double talk, it switches to non-adaptive mode for stability, thereby optimizing performance for each condition.
Data Source
AI summary
In one embodiment, the presence of double talk (DT) is detected in a telecommunications network having a near-end user and a far-end user. The energies of both (1) a signal received from the far-end user by the near-end user and (2) a signal to be communicated from the near-end user to the far-end user are computed. An echo return loss (ERL) estimate is calculated based on the energy calculations, and a preliminary decision is made as to whether DT is present based on the ERL estimate and the energy calculations. If DT is detected, then a counter is set to a hangover value. If DT is not detected, then the counter is reduced. This process is repeated, and, for each iteration, a final decision as to whether DT is present is made based on the counter value.