Implicit Adaptation Control for Beamforming
In Chap. 3, we have presented the statistically optimum LCMV beamforming and the Wiener solution (3.22). The latter involves the second-order statistics of the input signals, which are unknown in general. They might be estimated from the data under the assumption of stationary ergodic input microphone signals. However, in our application, the acoustic environment may change over time, for example when the speakers move. Furthermore, speech signals are nonstationary, requiring an adaptive approach.
On the one hand, continuous adaptation is desirable to find and to track the time-variant optimum filter coefficients. On the other hand, the filter estimation needs to be carried out only when the interferer is dominant relative to the target. This latter requirement may be satisfied with an explicit adaptation control based on a double-talk detector (DTD): We adapt only when some estimate of the input SIR is below a certain threshold. However, as we have seen in Chap. 3, the design of a robust and reliable DTD-based adaptation control is often difficult.
KeywordsTarget Signal Interference Canceler Online Performance Contraction Factor Normalize Little Mean Square
Unable to display preview. Download preview PDF.