15.01.2013 Views

U. Glaeser

U. Glaeser

U. Glaeser

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

The various components of (34.3) are also known as branch metrics.The Viterbi detector accomplishes<br />

the minimization in an efficient manner using a trellis-based search rather than an exhaustive search.<br />

The search is effectively performed over a finite window known as the decision delay or path memory<br />

length of the Viterbi detector. Increasing the window length beyond a certain value leads to only insignificant<br />

improvements of the bit detection reliability or bit error rate (BER).<br />

Despite the efficient nature of the Viterbi algorithm the complexity of a Viterbi detector increases<br />

exponentially with the channel memory of the PR target. A target with channel memory of L − 1 requires<br />

for example a 2 L−1 state Viterbi detector trellis. For a fully parallel Viterbi implementation, each Viterbi<br />

state contains an add-compare-select (ACS) computational unit, which is used to sum up the branch<br />

metrics of (34.4) and keep the minimum metric paths for different bit sequences. Also required for the<br />

hardware is a 2 L−1 ⋅P bit memory to keep a history of potential bit sequences considered across the finite<br />

decision delay window.<br />

Timing Recovery<br />

A phase-locked loop (PLL) is used to regenerate a synchronous clock from the data stream. The PRML<br />

detector use decision directed timing recovery typically with a digital loop filter. The digital loop filter<br />

parameters can be easily controlled using programmable registers and changed when read channel<br />

switches from acquisition to tracking mode. Because a significant pipelining is necessary in the loop logic<br />

to operate at high speeds, the digital loop filter architecture exhibits a relatively large amount of latency.<br />

It can affect considerably the acquisition time when the timing loop must acquire significant phase and<br />

frequency offsets. To ensure that only small frequency offsets are present, the synchronizer VCO is phaselocked<br />

to the synthesizer during nonread times. For fast initial adjustment of the sampling phase, a<br />

known preamble is recorded prior to user data. The time adjustment scheme is obtained by applying the<br />

stochastic gradient technique to minimize the mean squared difference between equalized samples and<br />

data signal estimates. To compensate for offset between the rate of the signal received and the frequency<br />

of the local timing source the loop filter design allows for a factor ∆T n to be introduced, so that the<br />

sample at discrete time n is taken T + ∆T n seconds after the sample at discrete time n − 1. In acquisition<br />

mode, in order to quickly adjust the timing phase, large values for loop gains are chosen. In tracking<br />

mode, the loop gains are lowered to reduce loop bandwidth.<br />

Read Channel Servo Detection<br />

In an embedded servo system (introduced in the previous article), the radial position of the read head is<br />

estimated from two sequences recorded on servo wedges: track addresses and servo-bursts. The track<br />

address provides a unique number for every track on a disk, while a servo-burst pattern is repeated on<br />

each track or on a group of tracks. Determining the head position using only the track number is not<br />

sufficient because the head has to be centered exactly on a given track. Therefore, the servo-burst<br />

waveform is used in conjunction with the track address to determine the head position. Using the servoburst<br />

pattern, it is possible to determine the off-track position of a head with respect to a given track<br />

with a high resolution. While positioning the head over a surface, the disk drive can be in either seeking<br />

or tracking operation mode. In a seeking mode, the head moves over multiple tracks, trying to reach the<br />

track with a desired address as quickly as possible, while in a tracking mode, the head tries to maintain<br />

its position over a track. The track addresses are therefore used mostly in the seeking mode, while servoburst<br />

information is usually used in the tracking mode [25,30].<br />

In read channels, periodic servo-burst waveforms are detected and used to estimate radial position.<br />

The radial position error signal is calculated based on the current estimated position and the position<br />

of the track to be followed, and then used in an external head positioning servo system. Generally, two<br />

types of position estimators are in use: maximum likelihood estimators based on a matched filtering and<br />

sub-optimal estimators based on averaging the area, or the peaks, of the incoming periodic servo-burst<br />

waveform. A variety of techniques have been used to demodulate servo bursts including amplitude, phase,<br />

© 2002 by CRC Press LLC

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!