Waveform equalizer apparatus formed of neural network, and method of designing same
AssigneeVictor Company of Japan, Ltd.
InventorSusumu Takahashi
Life of this patent
Estimating the term…
Abstract
A waveform equalizer for reducing distortion of a digital signal produced from a digital data recording and playback system or transmission system is formed of a neural network having fixed weighting coefficients. Respective values for the coefficients are established by generating a corresponding simulated neuron network, by software implementation using a computer, and by executing a neuron network learning operation using input values obtained from a distorted digital signal and teaching values obtained from an original digital signal which resulted in the distorted digital signal.
Claims10
A method of designing a waveform equalizer for waveform equalization of a distorted digital signal produced from a digital signal recording and playback apparatus, said waveform equalizer being formed as a neural network comprising a plurality of neuron units which are interconnected by linking elements providing respective fixed weighting coefficients, wherein respective values for said fixed weighting coefficients are mutually independently established by:
- generating a set of original data, and storing said original data in a first memory means;
- recording said original data as digital data values on a recording medium, and subsequently executing playback of said recorded original data, to obtain a playback digital signal;
- periodically sampling said playback digital signal with a sampling period that is less than or equal to a data period of said digital data values, to obtain successive digital sample values to be used as learning input values, and storing said learning input values in a second memory means;
- generating a simulated neural network having variable weighting coefficients, by using a computer, said simulated neural network being an equivalent circuit of said waveform equalizer neural network;
- supplying successive ones of said learning input values from said second memory means to said computer to be sequentially inputted to said simulated neural network;
- supplying successive data values of said original data, respectively corresponding to said learning input values, from said first memory means to said computer to be used as teaching signal values for comparison with respective output values produced from said simulated neural network;
- repetitively executing a learning algorithm utilizing results obtained from said comparison, to successively alter said variable weighting coefficients of the simulated neural network, until a predetermined degree of convergence is obtained for values of said variable weighting coefficients;
- and
- establishing respective values for said fixed weighting coefficients of the waveform equalizer neural network, based upon final values obtained for corresponding ones of said variable weighting coefficients of the simulated neural network.
A waveform equalizer for processing an input distorted digital signal, comprising:
- delay means coupled to receive said input distorted digital signal, for delaying said distorted digital signal by successive fixed amounts to obtain a plurality of delayed distorted digital signals;
- and
- a plurality of neuron units, and a plurality of linking elements having respectively fixed weighting coefficients, said linking elements interconnecting said neuron units to form a neural network which is coupled to receive said input distorted digital signal and delayed distorted digital signals as input signals thereto, and which produces a waveform-equalized output digital signal in response to said input signals;
- wherein each of said neuron units comprises a pair of series-connected signal inverting elements, and wherein of a plurality of weighted input signals supplied to said each neuron unit, each of said weighted input signals is selectively supplied to a first one or a second one of said inverting elements in accordance with whether an effectively positive or effectively negative value of weighting coefficient is to be applied to said each input signal.
Dependencies are inferred from the claim text.