Schemes for detection of double talk and abrupt changes in the echo path in fading communication channels

In this paper, we present novel algorithms for the detection and discrimination of double talk and abrupt changes in the echo path in a wireless communication system. We model the communication channel with a finite impulse response (FIR) channel with parameters that are randomly time-varying. Fading and scattering effects are described by using a multichannel autoregressive (AR) model and abrupt echo path changes and double talk by including finite state Markov processes in the dynamics. The proposed optimal iterative off-line scheme computes the joint, in a maximum a posteriori (MAP) sense, state and parameter estimates of the system. The estimates are computed via an expectation maximization (EM) based algorithm, which is implemented by interconnecting a hidden Markov model (HMM) smoother and a Kalman smoother to derive state estimates, while Yule-Walker and Baum-Welch type equations yield the abruptly changing FIR parameter estimates and the Markov chain parameter estimates, respectively. Replacing the HMM and Kalman smoothers by the corresponding filtered versions, an on-line scheme is proposed. The on-line detection scheme is verified experimentally by way of computer simulations