Disk drive employing iterative learning control for tuning seek servo loop

1. A disk drive comprising:

disk;

an actuator arm;

a head coupled to a distal end of the actuator arm;

a voice coil motor (VCM) for rotating the actuator arm about a pivot; and

control circuitry operable to seek the head from a first radial location over the disk to a second radial location over the disk by:

initializing a reference signal;

initializing a feed-forward control signal;

initiating the seek;

generating and saving a position error signal (PES) representing a position error of the head relative to the reference signal;

filtering the PES with a compensation filter to generate a feed-back control signal;

combining the feed-back control signal with the feed-forward control signal to generate a VCM control signal; and

after the seek, processing the saved PES using an iterative learning control (ILC) algorithm to adjust at least one of the reference signal and the feed-forward control signal.

2. The disk drive as recited in claim 1 , wherein the ILC algorithm adjusts the feed-forward control signal according to:

Uffwd — {n+ 1}( z )=Qu( z )[Uffwd — {n }( z )+Lu( z )PES — {n+ 1}( z )]

where:

Uffwd_{n+1}(z) is the feed-forward control signal vector at {n+1} iteration; and

PES_{n+1}(z) is the position error signal vector at {n+1} iteration.

3. The disk drive as recited in claim 2 , wherein:

Lu( z )=(σ re 2 /( n·σ re 2 +σ ne 2 ))·TRC0( z ) −1

where:

TRC0(z) is a nominal model of a torque rejection curve TRC(z) for the VCM;

n is an index of a current iteration;

σ re is an estimated variance of a repeatable part of the PES_{i} over i iterations; and

σ ne is an estimated variance of a non-repeatable part of the PES_{i} over i iterations.

4. The disk drive as recited in claim 2 , wherein Qu(z) is one.

5. The disk drive as recited in claim 3 , wherein Qu(z) is a filter such that

Qu(z)[1−Lu(z)·TRC(z)]

comprises an infinity norm smaller than one so that the iterative learning loop is stable.

6. The disk drive as recited in claim 1 , wherein the ILC algorithm adjusts the reference signal according to:

R — {n+ 1}( z )=Qr( z )[ R — {n }( z )+Lr( z )PES — {n+ 1}( z )]

where:

R_{n+1}(z) is the reference signal vector at {n+1} iteration; and

PES_{n+1}(z) is the position error signal vector at {n+1} iteration.

7. The disk drive as recited in claim 6 , wherein:

Lr( z )=(σ re 2 /( n·σ re 2 +σ ne 2 ))·ERC0( z ) −1

where:

ERC0(z) is a nominal model of an error rejection curve ERC(z) for the VCM;

n is an index of a current iteration;

σ re is an estimated variance of a repeatable part of the PES_{i} over i iterations; and

σ ne is an estimated variance of a non-repeatable part of the PES_{i} over i iterations.

8. The disk drive as recited in claim 6 , wherein Qr(z) is one.

9. The disk drive as recited in claim 7 , wherein Qr(z) is a filter such that

Qr(z)[1−Lr(z)·ERC(z)]

comprises an infinity norm smaller than one so that the iterative learning loop is stable.

10. A method of operating a disk drive, the disk drive comprising a disk, an actuator arm, a head coupled to a distal end of the actuator arm, and a voice coil motor (VCM) for rotating the actuator arm about a pivot, the method comprising:

initializing a reference signal;

initializing a feed-forward control signal;

initiating a seek;

generating and saving a position error signal (PES) representing a position error of the head relative to the reference signal;

filtering the PES with a compensation filter to generate a feed-back control signal;

combining the feed-back control signal with the feed-forward control signal to generate a VCM control signal; and

after the seek, processing the saved PES using an iterative learning control (ILC) algorithm to adjust at least one of the reference signal and the feed-forward control signal.

11. The method as recited in claim 10 , wherein the ILC algorithm adjusts the feed-forward control signal according to:

Uffwd — {n+ 1}( z )=Qu( z )[Uffwd — n ( z )+Lu( z )PES — {n+ 1}( z )]

where:

Uffwd_{n+1}(z) is the feed-forward control signal vector at {n+1} iteration; and

PES_{n+1}(z) is the position error signal vector at {n+1} iteration.

12. The method as recited in claim 11 , wherein:

Lu( z )=(σ re 2 /( n·σ re 2 +σ ne 2 ))·TRC0( z ) −1

where:

TRC0(z) is a nominal model of a torque rejection curve TRC(z) for the VCM;

n is an index of a current iteration;

σ re is an estimated variance of a repeatable part of the PES_{i} over i iterations; and

σ ne is an estimated variance of a non-repeatable part of the PES_{i} over i iterations.

13. The method as recited in claim 11 , wherein Qu(z) is one.

14. The method as recited in claim 12 , wherein Qu(z) is a filter such that

Qu(z)[1−Lu(z)·TRC(z)]

comprises an infinity norm smaller than one so that the iterative learning loop is stable.

15. The method as recited in claim 10 , wherein the ILC algorithm adjusts the reference signal according to:

R — {n+ 1}( z )=Qr( z )[ R — {n }( z )+Lr( z )PES — {n+ 1}( z )]

where:

R_{n+1}(z) is the reference signal vector at {n+1} iteration; and

PES_{n+1}(z) is the position error signal vector at {n+1} iteration.

16. The method as recited in claim 15 , wherein:

Lr( z )=(σ re 2 /( n·σ re 2 +σ ne 2 ))·ERC0( z ) −1

where:

ERC0(z) is a nominal model of an error rejection curve ERC(z) for the VCM;

n is an index of a current iteration;

σ re is an estimated variance of a repeatable part of the PES_{i} over i iterations; and

σ ne is an estimated variance of a non-repeatable part of the PES_{i} over i iterations.

17. The method as recited in claim 15 , wherein Qr(z) is one.

18. The method as recited in claim 16 , wherein Qr(z) is a filter such that

Qr(z)[1−Lr(z)·ERC(z)]

comprises an infinity norm smaller than one so that the iterative learning loop is stable.