Acoustic echo cancellation using sub-band encoded training sequences

An acoustic echo canceler (AEC) enhances speech quality by removing echoes. An AEC operates by estimating the acoustic path, e.g., transfer function, between a speaker and a microphone. A training sequence of symbols, known a priori, is transmitted to probe the acoustic environment between the speaker and microphone and used to determine the transfer function. The transfer function is used to cancel echoes from the signal produced by the microphone. The performance of an AEC is affected by various factors such as a dynamic acoustic path, multiple acoustic paths, presence of background noise, etc. In certain conditions, the time required for robust estimation of the transfer function can be such that the transfer function changes during the estimation. Repeated transmittal of the training sequence can improve transfer function estimates; however, the training sequence itself occupies bandwidth that can otherwise carry speech payload. Techniques of this disclosure achieve robust AEC performance by transmitting pseudo-random training sequences at spectral regions unoccupied by speech, in a manner imperceptible to the human ear. The transfer function can be estimated and updated nearly continuously and imperceptibly