State Detection Of Acoustic Echo Cancellation And System Realization

In the use of hands-free phone, far-end signals were broadcasted by the loudspeaker, and were reflected by the wall. Then they were picked up by the microphone in the form of direct waves and reflected waves. At last they were sent back to the far-end and generated acoustic echoes. This kind of echoes must be suppressed in the modern communication systems. Otherwise, they will cause invalided states, such as howl and so on. Usually we use adaptive echo cancellations to realize echo suppression, but in the face of time-varying environment, they are vulnerable to failure, particularly in automobile, aircraft and some other small rooms. So we need some states detections which can control them. Some common states detection algorithms can work well in the environment of constant noise; but in some time-varying environment, such as double talk, time-varying noise and time-varying echo path, they perform badly, because they can not track them well. This dissertation attempts to compare these algorithms to find a suitable one, so that echo cancellations are able to work steadily in time-varying environments.In this dissertation, we have discussed and compared these algorithms in the environment of quickly changed echo path, time-varying noise, and double-talk We also have used actual voice data to simulate these states, and have simulated these algorithms in such states. Finally, we have summarized their advantages and disadvantages for the choice of algorithms in practical applications.In order to analyze these algorithms better in the actual environment, we have designed an echo canceller data acquisition system. This system can use corresponding USB software which was programmed by us to save the data to PC for further analysis; it also can send the processed data back to the far-end to realize echo cancellation. At last, we have debugged this system in detail.