BER worst pattern of waveform discrimination using 2-7 code neural networks

An amplitude discrimination method to distinguish "1ldquo\´or "0" by height of the amplitude is used for the discrimination method of a reproduced signal in a current hard disk. The SNR often deteriorates by the thermal noise that a head emits in this method. As a result, waveform interference may cause, and we cannot recognize a high-density signal precisely. Because the neural network based waveform discrimination method used in this study judges "1ldquo\´or "0" with a characteristic of a waveform, it is expected that it can recognize a high-density signal. By a conventional experiment, we obtained an error rate characteristic. In this study, we extracted a worst pattern of an error rate characteristic. As a result, it is shown that the signal waveform includes its characteristic and there is "easiness" and "difficulty" to reproduce. In addition, under the condition; a first-order differential waveform, 8-6 hidden layer and normalized linear density K = 4.0, the form of a BER (bit error rate) curve, of all patterns almost accorded with the mean of bad pattern BER. According to this result, it is considered that the error rate evaluation time of a signal becomes shorter.