Equalizer and equalization method

An equalizer comprising:

• a linear equalizer unit configured to sequentially subject the signal input to linear equalization;
• an adaptive nonlinear equalizer unit implemented by a neural network and configured to sequentially subject the signal input to nonlinear equalization in parallel with the linear equalization by the linear equalizer unit;
• an adder unit configured to add a signal subjected to nonlinear equalization by the nonlinear equalizer unit and a signal subjected to linear equalization by the linear equalizer unit;
• a temporary decision unit configured to sequentially subject a signal resulting from the addition in the adder unit to temporary decision according to a partial response rule;
• and
• a decision unit configured to determine whether the plurality of coefficients of a neural network in the adaptive nonlinear equalizer unit converge,
• wherein the adaptive nonlinear equalizer unit derives a plurality of coefficients using a signal subjected to temporary decision by the temporary decision unit as a teacher signal and without using a training signal, a known input in the presence of a given output, normally required to cause a neural network to learn previously, and performs nonlinear equalization based on the plurality of coefficients, and
• wherein, unless the decision unit determines that convergence takes place, the adder unit outputs the signal subjected to linear equalization by the linear equalizer unit to the temporary decision unit instead of a signal resulting from the addition, and outputs a signal resulting from the addition to the temporary decision unit after the decision unit determines that convergence takes place.

An equalization method comprising:

• sequentially subjecting an input signal to linear equalization;
• sequentially subjecting the input signal to nonlinear equalization in parallel with the linear equalization by means of an adaptive nonlinear equalizer unit implemented by a neural network;
• adding a signal subjected to nonlinear equalization and a signal subjected to linear equalization;
• sequentially subjecting a signal resulting from the addition to temporary decision according to a partial response rule,
• determining whether the plurality of coefficients of a neural network used in the nonlinear equalization converge,
• wherein the subjecting of the input signal to nonlinear equalization comprises deriving a plurality of coefficients using a signal subjected to temporary decision as a teacher signal and without using a training signal, a known input in the presence of a given output, normally required to cause a neural network to learn previously, plus performing nonlinear equalization based on the plurality of coefficients, and
• unless the determining determines that convergence takes place, the adding outputs the signal subjected to linear equalization in the linear equalizing to the temporary decision instead of a signal resulting from the addition, and outputs a signal resulting from the addition to the temporary decision after the determining determines that convergence takes place.

A program embedded in a non-transitory computer-readable recording medium, comprising:

• sequentially subjecting an input signal to linear equalization;
• sequentially subjecting the input signal to nonlinear equalization in parallel with the linear equalization by means of an adaptive nonlinear equalizer unit implemented by a neural network;
• adding a signal subjected to nonlinear equalization and a signal subjected to linear equalization;
• sequentially subjecting a signal resulting from the addition to temporary decision according to a partial response rule,
• determining whether the plurality of coefficients of a neural network used in the nonlinear equalization converge,
• wherein the subjecting of the input signal to nonlinear equalization comprises deriving a plurality of coefficients using a signal subjected to temporary decision as a teacher signal and without using a training signal, a known input in the presence of a given output, normally required to cause a neural network to learn previously, plus performing nonlinear equalization based on the plurality of coefficients, and
• unless the determining determines that convergence takes place, the adding outputs the signal subjected to linear equalization in the linear equalizing to the temporary decision instead of a signal resulting from the addition, and outputs a signal resulting from the addition to the temporary decision after the determining determines that convergence takes place.