Abstract:
The author describes a method for matching a vector source and channel with a frequency-weighted mean-square criterion of reproduction accuracy. This method, which includes optimal linear coding (predistortion and correction) and optimal transposition of the frequency bands of the components of the vector messages, is the best in the class of possible linear transformations; it yields a frequency-weighted mean-square deviation no greater than that provided by any other linear methods of coding and transposition. A procedure is presented for finding optimal characteristics for coders and decoders with correlated vector-message components and noise in the presence of linear distortion in the channel.