Schaltungsanordnung zur Signalumwandlung und Frequenzbandeinengung mit einem Kreuzungsnetzwerk

摘要

816,025. Vocoder systems. WESTERN ELECTRIC CO. Inc. April 9, 1957 [April 24, 1956], No.11538/57. Class 40 (4). In a system requiring the transfer of several sets of information having a high degree of statistical correlation, arrangements are made for combining the information for transmission along a set of conductors in such a way that the conductors may be arranged such that their energy contents are of decreasing order of importance. The outputs from the least important conductors may then be ignored. The invention is applied to a vocoder system of the type described in Specification 466,327, containing the usual pitch determining circuit 2, Fig. 1, for controlling, over lead Sp, a buzz/ biss source 23, 24, Fig. 3, and band-pass filters 3 feeding rectifiers 4 whose outputs e 1 ... e n are required to control modulators 21 fed by the outputs from corresponding band-pass filters 22 to reconstitute the original signal. The signals e 1 ... e n have a high degree of correlalation and accordingly they are fed to the verticals of a matrix 12 whose output leads 11, thus carry different combinations of these signals in accordance with coefficients determined by the transfer elements at the crosspoints. These transfer elements comprise phase splitters 7 having tapped resistances across their outputs, the position of the tapping thus determining the magnitude and sign of the coefficient. To obtain the numerical values of these coefficients a matrix [e r e 8 ] is first formed whose elements are the mean values of all the cross-products of the e's as determined by previous statistical observations. The eigen values of this matrix are then obtained and arranged in decreasing order of magnitude. The transfer coefficients for any particular row of the matrix 12 are then the elements of a particular eigen vector, each row thus being associated with a particular eigen value. The outputs S 1 . . . s n of the matrix 12, feed a matrix 15 at the receiving end, the matrix 15 being the transpose of the matrix 12, to reproduce signals e11 . . . e1n corresponding to the input signals e 1 ... e n . The Specification demonstrates that if the transformation matrices are as described then the a minimum total mean square error #(e r -e1 r )2 is obtained when the outputs of a prescribed number of leads S 1 ... S n (corresponding to the lowest eigen values) are ignored. The Specification also demonstrates that it is advantageous to reduce the values of the signals e 1 ... e n to be transmitted by constant amounts equal to their respective mean values. This is achieved by the insertion of batteries 30, 31 in the outputs of the phase-splitters 7. Compensating batteries 35 are inserted at the receiver. It may be desired to minimize some other positive definite function of the errors, e.g. the function α=#b rs (e r -e1 r )(e s -e1s). (In this case a matrix P is calculated which when multiplied by its transpose gives the matrix [b rs ]. If the signals e 1 ... e n are applied to this matrix, Fig. 6, to give outputs g 1 ... g n it may be shown that the error function Q is now #(g r -g1r)2 so that the g signals may now be applied to a matrix N of the type previously described to produce signals s1 for onward transmission. At the receiver the signals are applied to the matrix N T (the transpose of N) to produce signals g1 which are then applied to a matrix P-1 (the inverse of P) to produce the required signals e1. The matrices P and N may be combined into a single matrix as also the matrices N T and P-1.