| 摘要 |
A method and system are shown and described for separating, on a communication channel, multiple complex communication signals originating from corresponding multiple communication sources. The method includes estimating separate complex communication signals using an unmixing matrix. The unmixing matrix is recursively updated by transforming the estimated complex communication signals into real numbers for use with a gradient-based training module. The gradient-based training module is configured to use the real numbers to calculate learning rule updates to apply to the unmixing matrix. The gradient-based training module is configured to minimize error in the signal estimation obtained by application of the unmixing matrix. The transformation of the estimated complex communication signals into real numbers may include applying the complex communication signals to a hyperbolic tangent function.
|
