Hardware Reference
In-Depth Information
on the known RRO trajectory, inverting the RRO signal against the transfer
function allows us to inject a signal into the loop such that the RRO is can-
celled. Such method is based on pre-calculation of the reference signal and is
implemented in feedforward manner. This method does not affect closed loop
stability. The disadvantage of this method is that the values of the compen-
sation signal need to be stored in a look-up-table (LUT) after pre-calculation;
when there is a change in the RRO profile due to motor aging, disk slip etc, the
LUT need to be re-calculated and updated. The RRO compensation system
can be designed to follow the servo-burst-defined track centers. Alternatively,
knowing that the RRO reflects the deviation of the servo-burst-defined track
centers from perfect circles, we can redefine the actual data track centers which
are not coincident with the servo-burst-defined track centers by not following
the RRO signal in PES. For the later scheme, interested readers may refer to
the zero-acceleration-path or ZAP scheme reported in [33].
Feedback control based peak filters of limited peak gains, for both single
and multiple peaks, as well as delay-generated multiple frequency peaks, allow
partial cancelation of RRO signals. Peak filters of infinity peak gain (and their
equivalent automatic feedforward control or AFC) allow complete cancelations
of a specific RRO frequencies. All these schemes affect the control loop gain
and hence affect the system stability.
In addition to the accuracy of the RRO compensation scheme, time taken
by effective compensation is a critical issue as it affects the access performance
of the HDD. When the head moves from one track to another, the time taken
to arrive at the desired data block is anywhere between the seek-settle time
plus zero and seek-settle time plus the time for one full revolution. If, for
example, an AFC scheme takes one full revolution to converge in order to
cancel the necessary percentage of a target RRO component, then all the data
read/write actions need to wait for the full rotational period in addition to
the average latency of half of the rotational period. However he AFC scheme,
which typically takes 10 ms or more to converge, can be used to learn in
real time or offline the RRO pattern which varies slowly. The RRO pattern
learntcanthenbeusedtoupdatetheRRO cancelation LUT which is used for
realtime RRO correction.
3.5.4 Multirate Control
A multi-rate system is a discrete time system in which more than one sampling
rate is used with different sampling rates for different sections [105, 116]. For
example, in many practical systems, the output is sampled at a rate of f O Hz
where as the input is updated at a rate f i = f O Hz [204]. Multi-rate system
has been studied for control systems with different time-scale loops, and differ-
ent sampling rate and control update rate. While designing controller for such
systems, it can be first converted to an equivalent single rate system, either at
the slowest common base rate or at the fastest common rate. Then the conven-
 
Search WWH ::




Custom Search