Noise predictive maximum likelihood (NPML) detection methods and apparatus based thereon

• US 6,104,766 A
• Filed: 09/17/1996
• Issued: 08/15/2000
• Est. Priority Date: 09/18/1995
• Status: Expired due to Term

• Current Assignee: Western Digital Technologies Incorporated (Western Digital Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Coker, Jonathan Darrel, Galbraith, Richard Leo, Hirt, Walter, Eleftheriou, Evangelos Stavros
• Primary Examiner(s): Chin, Stephen
• Assistant Examiner(s): Fan, Chieh M.
• Application Number:
  US08/715,174
• Time in Patent Office: 1,428 Days
• Field of Search: 324/76.28, 324/76.29, 342/203-204, 375/231-233, 375/327, 375/341, 375/350, 375/367, 375/262, 375/263, 375/265, 360/39, 360/40, 360/55, 360/65, 714/794
• US Class Current: 375/341
• CPC Class Codes: G11B 20/10009 Improvement or modification of read or write signals
• CPC Class Codes: G11B 20/10055 using partial response filtering when writing the signal to the medium or reading it therefrom
• CPC Class Codes: G11B 20/10175 PR4, PR(1,0,-1), i.e. partial response class 4, polynomial (1+D)(1-D)=(1-D2)
• CPC Class Codes: H04L 25/497 by correlative coding, e.g. partial response coding or echo modulation coding transmitters and receivers for partial response systems transversal equalizers at the transmitter end H04L25/03343

Abstract

The invention is family of noise predictive maximum liklihood (NPML) symbol detectors that are particularly useful in direct access storage devices. Various embodiments representative of the family of detectors are described. In general, the NPML detectors include a sequence detector with imbedded feedback which may be preceded by a filter, e.g., a prediction error filter. The sequence of detectors, which may be Viterbi detectors, have an imbedded filter whose coefficients are determined by the convolution of the partial response (PR) or generalized PR function with the predictor coefficients and do not require multiplications and, thus, allows a simple RAM look-up for intersymbol-interference ISI cancelation. In one class of embodiments, the NPML detector comprises a prediction error filter of length N, cascaded with a sequence detector having 2^K states, and a feedback filter imbedded in the sequence detector, having a length N+P-K, where P is length of the generalized PR shaping polynomial (P=2 yields PR4), and 0≦K≦N+P.

Claims

1. A detector comprising:

a prediction error filter comprising a predictor having a length;

a sequence detector, coupled to the prediction error filter, the sequence detector having a number of states; and

a feedback filter, coupled to the sequence detector, having a length which is a function of the predictor length, the number of states of the sequence detector, and a length of a generalized partial response shaping polynomial.

2. The detector of claim 1, wherein

the predictor has length N;

the sequence detector has 2^K states; and

the feedback filter has a length N+P-1, where P is the length of a generalized partial response shaping polynomial, and 0≦

K≦

N+P.

3. The detector of claim 2, wherein the feedback filter has a length of N+P-K for each state, and the sequence detector comprises a maximum likelihood sequence detector.

4. The detector of claim 1, wherein the feedback filter uses path history decisions.

5. The detector of claim 1, wherein the generalized partial response shaping polynomial is partial response class IV.

6. The detector of claim 1, wherein the feedback filter comprises a finite impulse response filter or a RAM.

7. The detector of claim 1, wherein the feedback filter comprises a RAM and the sequence detector comprises means for adaptively altering contents of the RAM.

8. The detector of claim 1, wherein the prediction error filter whitens a noise component of a signal.

9. The detector of claim 1, wherein the sequence detector further comprises a survivor path memory for storing path history decisions.

10. The detector of claim 9, wherein the feedback filter uses precomputed cancellation terms, and retrieves the cancellation terms by applying the path history decisions as addresses.

11. The detector of claim 1, wherein

the prediction error filter whitens a noise component of a received signal sample having a generalized partial response signal component, and adds intersymbol-interference components;

the sequence detector, further comprises a survivor path means for storing path history decisions; and

the feedback filter cancels a portion of the intersymbol-interference components using precomputed intersymbol-interference cancellation terms by applying the path history decisions as addresses.

12. The detector of claim 1, wherein an output signal of the sequence detector is coupled to an inverse precoder.

13. The detector of claim 1, further comprising means for adaptive setting of a transfer characteristic of the prediction error filter.

14. The detector of claim 1, wherein the sequence detector comprises a memoryless detector.

15. The detector of claim 14, wherein the memoryless detector determines a nominal expected value, and further comprising:

means for estimating a noise contribution in a plurality of signal samples by subtracting a value of a signal sample from the nominal expected value;

means for predicting a noise contribution of a current signal sample using the estimated noise contribution;

means for modifying the predicted noise contribution of the current signal sample; and

means for feeding the output of the means for modifying to a partial response maximum likelihood detector.

16. A data storage device, comprising:

means for reading encoded sequences of data; and

a decoder comprising

a prediction error filter comprising a predictor having a length,

a sequence detector, coupled to the prediction error filter, the sequence detector having a number of states, and

a feedback filter, coupled to the sequence detector, having a length which is a function of the predictor length, the number of states of the sequence detector, and a length of a generalized partial response shaping polynomial.

17. The data storage device of claim 16, wherein

the predictor has length N;

the sequence detector has 2^K states; and

the feedback filter has a length N+P-K, where P is the length of a generalized partial response shaping polynomial, and 0≦

K≦

N+P.

18. The data storage device of claim 17, wherein the feedback filter has a length of N+P-K for each state.

19. The data storage device of claim 16, wherein the generalized partial response shaping polynomial is partial response class IV.

20. The data storage device of claim 16, wherein the sequence detector comprises a maximum likelihood sequence detector.

21. The data storage device of claim 16, wherein the feedback filter comprises a RAM or a finite impulse response filter.

22. The data storage device of claim 16, wherein the sequence detector comprises a survivor path memory for storing path history decisions, and the feedback filter uses precomputed and stored cancellation terms, and retrieves the cancellation terms by applying the path history decisions as addresses.

23. The data storage device of claim 16, wherein the prediction error filter whitens a noise component and adds intersymbol-interference components to a signal.

24. A method for symbol detection, comprising the steps of:

receiving a signal sample having a noise component;

filtering the signal sample using a prediction error filter having a length, the filtered signal having intersymbol-interference components; and

determining a symbol using a sequence detector having a number of states, and also using a feedback filter, coupled to the sequence detector, having a length which is a function of the predictor length, the number of states of the sequence detector, and a length of a generalized partial response shaping polynomial.

25. The method of claim 24 wherein

the prediction error filter has a length N;

the sequence detector has 2^K states; and

the feedback filter has a length N+P-K, where P is a length of a generalized partial response shaping polynomial, and 0≦

K≦

N+P.

26. The method of claim 24, further comprising the step of canceling a portion of the intersymbol-interference components from the filtered signal using predetermined intersymbol-interference cancellation terms.

27. The method of claim 24, wherein the noise component of the received signal sample is filtered by whitening.

28. The method of claim 25, wherein

the signal sample comprises a partial response signal component corrupted by a noise component;

and

the step of determining further comprises the step of performing a branch metric computation based on a 2^K -State Viterbi algorithm to determine a most likely sequence corresponding to the signal sample.

29. The method of claim 24, further comprising the steps of:

storing path history decisions from a sequence detector in a memory;

retrieving pre-computed intersymbol-interference cancellation terms using the path history decisions as addresses to storage locations; and

canceling intersymbol-interference components in the signal sample using the intersymbol-interference cancellation terms.

30. The method of claim 24, further comprising the steps of:

estimating a noise contribution in a plurality of past digital signal samples by subtracting a value of a signal sample from a nominal expected value, the nominal expected value being determined by memoryless detection;

using the noise contribution of the plurality of the past digital signal samples to predict a noise contribution of the signal sample; and

subtracting the predicted noise contribution from the signal sample.