This thesis is to investigate the performance of the maximum likelihood sequence estimation (MLSE) of the run-length limited (RLL) duobinary codes of minimum bandwidth. It is well known that since the Viterbi algorithm had been proposed as the decoding method for the convolutional codes, it has been applied in various fields. It can be also applied to the sequence estimation of partial response (PR) codes such as duobinary code.
For the RLL duobinary codes designed by means of the variation of the duobinary code, exact analysis of the decoding performance is difficult because of their nonlinearity. It is possible, however, to regard the encoder of an RLL duobinary code as a finite state sequential machine. Therefore the MLSE can be achieved by using the Viterbi algorithm.
The comparison of simulated results with those of the bit-by-bit detection method showes that the SNR is improved about 1.7 dB at $10^{-4}$ error rate. The buffer size for the decision delay is required as much as 150 symbols for VDB5 code an 120 symbols for VDB4 code. An analogy of the constraint length of the convolutional codes for the RLL duobinary codes exhibits the required buffer size of 10 times of their constraint length.