| Summary: | 碩士 === 國立交通大學 === 電信研究所 === 85 === To implement a high recording density MO (magneto-optical) disk
drive, the priorknowledge about the MO read/write channel is
needed. An accurate channel model will make an optimal system
design possible. Linear MO channel models which are often used
in low density recording systems are not suitable for describing
the channel effect in a high density recording system. Nonlinear
channel characteristic which is caused mainly by the interaction
between the two closely allocated recorded domains must be taken
into account. It is known that this nonlinear effect will
seriously degrade theperformance of a linear receiver no matter
how complicated the associated equalizer is. he first part of
this thesis deals with the MO channel identification problem.
Both linear and nonlinear methods are used. We find that the
linear method using PN sequence as the input is not stable.
Channel responses using three linear algorithms are not
consistent. There are indications that strong nonlinearity does
exit in the MO channel. We then use a Volterra-decomposed
nonlinear model to identify MO channels. Numerical results show
that this nonlinear model can charaterize the MO channel to a
very high degree of accuracy. In fact, as far as mean-squared
identification error (MSIE) is concerned, the nonlinear method
yields MSIE 20 times smaller than that resulted from any linear
model. The second part concentrates on the detector design,
assuming a nonlinear channel characteristic. We derive a
maximum-likelihood sequence detector and evaluate the resulting
performance. Based upon the nonlinear channel model, we derive
two nonlinear Viterbi detectors, one for an NRZ-coded (non-
return zero) channel and the other for a (2,7) RLL-coded
nonlinear MODD channel. Using the SAM (sequenced amplitude
margin) scheme we are able to predict the error rate performance
up to 10^-9 with only about 10^5 ~ 10^6 sample bits. Due to the
practical limitation of hardware resolution for both the
transimitter (pattern generator) and front-end digitizer
(digital oscilloscope), there are only finite number of clock
rate from which we can choose. Experiment results indicate that
the error rate performance of the NRZ-coded nonlinear Viterbi
detector is around 10^-6 while the (2,7) RLL-coded nonlinear
Viterbi detector can be almost 10^-9. The selected recording
density is at least 1.15 times higher than the original
specification; in other words, there are almost 15% capacity
enhancement for the (2,7) RLL-coded nonlinear Viterbi detector
over the original linear detector.
|
|---|
