Echo Cancellation for Hands-Free Systems

Abstract

Echo is defined as the delayed and attenuated version of the original signal produced by some device, such as a loudspeaker. As a consequence a person listens to a delayed replica of its own voice signal. This is an undesired effect that appears whenever the output signal is fed back into the system’s input and it can be quite disturbing on voice conversations. Echo arises in long distance communication scenarios such as hands-free systems Hansler (1994); Jeannes et al. (2001); Liu (1994), voice over internet protocol (VoIP) Witowsky (1999), teleconferencing Kuo & Pan (1994), mobile phone conversation, and satellite communications among others. In order to minimize or even remove the presence of echo in communications, echo suppression and echo cancellation techniques have been proposed in the last three decades Sondhi (2006). An echo suppressor is a voice-operated switch that disconnects the communication path (or introduces a very large attenuation) whenever some decision mechanism indicates that we are in the presence of echo. The emitting circuit is disconnected whenever we have signal on the reception part of the circuit; the reception circuit is disconnected whenever we have signal emission. Their behavior is not adequate for cross conversation (full duplex) scenarios. Echo suppressorswere the first approach to this problem. In the last decade, due to their unsatisfactory results, they have been replaced by digital echo cancelers. An echo canceler device, as opposed to an echo suppressor, does not interrupt the echo path; it operates by removing (subtracting) the detected echo replicas from the information signal. The term usually coined for the cancellation of echoes with acoustic coupling is acoustic echo cancellation (AEC) Gilloire & Hansler (1994). In the past years, adaptive filtering techniques Haykin (2002); Sayed (2003); Widrow& Stearns (1985) have been employed for the purpose of AEC Breining (1999); Widrow et al. (1975). Typically, these techniques rely on the use of finite impulse response (FIR) filters Oppenheim & Schafer (1999); Veen & Haykin (1999) whose coefficients are updated along the time by an efficient rule guided by some statistical criterion. Usually, one employs a gradient descent technique in order to minimize some cost (error) function. The most popular of these techniques is the Widrow-Hoff least mean squares (LMS) algorithm as well as its variants, that minimize the mean square error (MSE) between two signals. Moreover, in many cases such as real-time conversations over mobile phones, AEC algorithms must run in real-time to be useful. We thus have the need for efficient implementations of echo cancellation techniques on digital embedded devices like field programmable gate array (FPGA) and/or digital signal processor (DSP), to fulfill real-time requirements of many applications, these days. 14