Incorporating signal history into transfer logic for two-path echo cancelers

Two-path echo cancelers offer an effective mechanism to deal with the double-talk problem. A fixed foreground (FG) filter cancels echo while an adaptive background (BG) filter learns the echo path response. The coefficients of the BG filter are transfered to the FG filter when a set of fitness conditions are satisfied. Whereas previous works used memoryless metrics in the coefficient transfer logic, this paper demonstrates that historical information gives a more complete picture about the convergence and non-convergence of the BG filter. History- based transfer logic yields robust performance during double- talk and barge-in events.