Look-ahead delta sigma modulator with pruning of output candidate vectors using quantization error minimization pruning techniques

• US 7,148,830 B2
• Filed: 01/26/2005
• Issued: 12/12/2006
• Est. Priority Date: 01/26/2004
• Status: Active Grant

• Current Assignee: Cirrus Logic Incorporated
• Original Assignee: Cirrus Logic Incorporated
• Inventors: Melanson, John L.
• Primary Examiner(s): Barnie, Rexford
• Assistant Examiner(s): Nguyen, Khai M.
• Application Number:
  US11/043,719
• Publication Number:
  US 20050162285A1
• Time in Patent Office: 685 Days
• Field of Search: 341/143
• US Class Current: 341/143
• CPC Class Codes: H03M 3/366 in feed-forward mode, e.g. using look-ahead circuits
• CPC Class Codes: H03M 3/458 Analogue/digital converters using delta-sigma modulation as an intermediate step
• CPC Class Codes: H03M 7/3011 of non-linear distortion, e.g. by temporarily adapting the operation upon detection of instability conditions avoiding instability by structural design H03M7/3035

Abstract

A look-ahead delta sigma modulator utilizes pruning techniques to reduce the number of possible output candidate vectors used to determine quantization output values. The look-ahead delta-sigma modulators can also combine the pruning techniques with a reduction in the amount of processing by simplifying the computations used to generate quantizer output values. In one embodiment, the pruning techniques exploit application of superposition techniques to a loop filter response of the look-ahead delta sigma modulator. The set of possible pattern response vectors can be pruned to form a subset of P candidate pattern response vectors. If an element value of a pattern response vector proximally matches a value of the corresponding reference element of the natural response vector, a quantizer uses the proximally matching pattern response vector to determine a quantization output value.

Claims

1. A method of processing input signal data using a look-ahead delta sigma modulator of depth N using natural and pruned pattern loop filter responses, wherein N is greater than or equal to two, the method comprising:

determining a set of pattern responses of a loop filter of the look-ahead delta sigma modulator for each output candidate vector by setting signal input data to the loop filter of the delta sigma modulator to at least approximately zero and filtering a set of output candidate vectors;

determining a natural response of the loop filter of the look-ahead delta sigma modulator by setting feedback data to the loop filter to at least approximately zero and filtering a set of input data signal samples;

pruning the set of pattern responses to determine a set of one or more pruned pattern responses; and

quantizing each input signal data sample by applying predetermined decision criteria to determine a best match between a member of the set of one or more pruned pattern responses and the natural response and selecting quantization output data from the output candidate vector associated with the pruned pattern response used to determine the best match.

2. The method as in claim 1 wherein the natural response and each pattern response are respective vectors and pruning the set of pattern responses comprises:

selecting each pattern response vector that includes a vector element that proximally matches a corresponding reference element of the natural response vector.

3. The method as in claim 2 wherein selecting each pattern response vector that includes a vector element that proximally matches a corresponding reference element of the natural response vector comprises:

determining if the vector element of the pattern response vector that corresponds to the reference element of the natural response vector is within a prescribed linear distance.

4. The method as in claim 2 wherein selecting each pattern response vector that includes a vector element that proximally matches a corresponding reference element of the natural response vector comprises:

determining a prescribed number of pattern response vectors that include respective elements having shortest linear distances to a corresponding natural response vector reference element value.

5. The method as in claim 1 wherein:

applying the predetermined decision criteria to determine the best match comprises determining a lowest power of the difference between a member of the set of one or more pruned pattern responses and the natural response; and

selecting a quantization output value from the output candidate vector associated with the pruned pattern response used to determine the best match comprises selecting the output candidate vector associated with the lowest power difference.

6. The method as in claim 5 wherein determining a lowest power match between one of the pruned pattern responses and the natural response comprises determining a dot product for each pruned pattern response and the natural response.

7. The method as in claim 1 wherein:

applying the predetermined decision criteria to determine the best match comprises determining a lowest power match between a member of the set of one or more pruned pattern responses and the natural response wherein at least one element of the pruned pattern response is weighted with a weight different than any other element of the pruned pattern response; and

selecting output data from the output candidate vector associated with the pruned pattern response used to determine the best match comprises selecting the output candidate vector associated with the lowest power match.

8. The method as in claim 1 wherein the set of pattern responses excludes pruned pattern responses that provide duplicate best match determinations.

9. The method as in claim 1 wherein determining the set of pattern responses of the loop filter of the look-ahead delta sigma modulator further comprises precomputing the set of pattern responses before quantizing each input signal data sample.

10. The method of claim 9 further comprising:

storing the precomputed set of pattern responses in a memory for use in subsequently quantizing each input signal data sample.

11. The method as in claim 1 further comprising:

computing the natural response only once for quantization of each filtered input signal data sample.

12. The method as in claim 1 wherein the set of input data signal samples used to determine the natural response of the loop filter equals N.

13. The method as in claim 1 wherein the set of output candidate vectors is reduced to less than 2^N without decreasing quantization accuracy.

14. The method as in claim 1 wherein the input signal data sample comprises audio input signal data.

15. The method as in claim 1 further comprising:

recording quantized input signal data on storage media.

16. The method as in claim 1 further comprising:

for each time t, determining a dot product of each pruned pattern response with the natural response using substantially only addition operations.

17. A signal processing system comprising:

a look-ahead delta sigma modulator comprising;

a loop filter;

a pattern response generator coupled to the loop filter to determine a set of pattern responses by the loop filter obtained by setting signal input data to the loop filter to at least approximately zero and filtering a set of output candidate vectors;

a natural response generator to determine a natural response of the loop filter by setting feedback data to the loop filter to at least approximately zero and filtering a set of input data signal samples;

a pruning generator to determine a set of one or more pruned pattern responses; and

a quantizer to receive output data from the loop filter and the set of pruned pattern responses, wherein the quantizer includes a function generator to quantize each input signal data sample by applying predetermined decision criteria to determine a best match between a member of the set of one or more pruned pattern responses and selecting output data from the output candidate vector associated with the pruned pattern response used to determine the best match.

18. The signal processing system as in claim 17 wherein the natural response and each pattern response are respective vectors and the pruning generator is configured to select each pattern response vector that includes a vector element that proximally matches a corresponding reference element of the natural response vector.

19. The signal processing system as in claim 18 wherein the pruning generator is further configured to determine if the vector element of the pattern response vector that corresponds to the reference element of the natural response vector is within a prescribed linear distance.

20. The signal processing system as in claim 18 wherein the pruning generator is further configured to determine a prescribed number of pattern response vectors that include respective elements having shortest linear distances to a corresponding natural response vector reference element value.

21. The signal processing system as in claim 17 further comprising:

a memory coupled to the loop filter to store the determined set of pattern responses of the loop filter for each output candidate vector.

22. The signal processing system as in claim 17 wherein:

applying the predetermined decision criteria to determine the best match comprises determining a lowest power match between a member of the set of one or more pruned pattern responses and the natural response; and

selecting output data from the output candidate vector associated with the pruned pattern response used to determine the best match comprises selecting the output candidate vector associated with the lowest power match.

23. The signal processing system as in claim 22 wherein determining a lowest power match between one of the pruned pattern responses and the natural response comprises determining a dot product for each pruned pattern response and the natural response.

24. The signal processing system as in claim 17 wherein:

applying the predetermined decision criteria to determine the best match comprises determining a lowest power match between a member of the set of one or more pruned pattern responses and the natural response wherein at least one element of the pruned pattern response is weighted with a weight different than any other element of the pruned pattern response; and

selecting output data from the output candidate vector associated with the pruned pattern response used to determine the best match comprises selecting the output candidate vector associated with the lowest power match.

25. The signal processing system as in claim 17 wherein the set of pattern responses does not include pattern responses that provide duplicate best match determinations.

26. The signal processing system as in claim 17 wherein to determine the set of pattern responses of the loop filter of the look-ahead delta sigma modulator the natural response generator includes components to precompute the set of pattern responses before quantizing each input signal data sample.

27. The signal processing system as in claim 17 wherein the natural response generator generates the natural response only once for quantization of each filtered input signal data sample.

28. The signal processing system as in claim 17 wherein the set of input data signal samples used to determine the natural response of the loop filter equals N, wherein N is a look-ahead depth and is greater than or equal to two (2).

29. The signal processing system as in claim 17 wherein the set of output candidate vectors is reduced to less than 2^N without decreasing quantization accuracy, wherein N is a look-ahead depth and is greater than or equal to two (2).

30. The signal processing system as in claim 17 wherein the input signal data sample comprises audio input signal data.

31. The signal processing system as in claim 17 further comprising:

signal processing and recording equipment to process output data from the quantizer and record the processed output data on storage media.

32. The signal processing system as in claim 17 wherein the quantizer further comprises:

an add/compare/select system to combine each pruned pattern response with the natural response using substantially only addition operations, compare the combinations, and select the output data.

33. An apparatus for processing input signal data using a look-ahead delta sigma modulator of depth N using natural and pattern loop filter responses, wherein N is greater than or equal to two (2), the apparatus comprising:

means for determining a set of pattern responses of a loop filter of the look-ahead delta sigma modulator for each output candidate vector by setting signal input data to the loop filter of the delta sigma modulator to at least approximately zero and filtering a set of output candidate vectors;

means for determining a natural response of the loop filter of the look-ahead delta sigma modulator by setting feedback data to the loop filter to at least approximately zero and filtering a set of input data signal samples;

means for pruning the set of pattern responses to determine a set of one or more pruned pattern responses; and

means for quantizing each input signal data sample by applying predetermined decision criteria to determine a best match between a member of the set of one or more pruned pattern responses and the natural response and selecting quantization output data from the output candidate vector associated with the pruned pattern response used to determine the best match.

34. The apparatus as in claim 33 further comprising:

means to filter each output candidate vector and the input data signals.

35. A method of processing an input signal using a look-ahead delta sigma modulator of depth N, wherein N is greater than or equal to two, the method comprising:

pruning a set of candidate vectors by selecting a subset of the candidate vectors, wherein the subset of candidate vectors minimize quantization error of the look-ahead delta sigma modulator at a future time t+n, t is the current time, and n≧

2; and

quantizing the input signal using the subset of candidate vectors.

36. The method of claim 35 further comprising:

determining a response of a loop filter of the look-ahead delta-sigma modulator from a natural response of the loop filter to a set of N input data signal samples and a forced response of the loop filter to the set of output candidate vectors;

wherein pruning a set of candidate vectors by selecting a subset of the candidate vectors, wherein the subset of candidate vectors minimize quantization error of the look-ahead delta sigma modulator at a future time t+n comprises pruning the set of pattern responses to determine a set of one or more pruned pattern responses; and

wherein quantizing the input signal using the subset of candidate vectors comprises quantizing each input signal data sample by applying predetermined decision criteria to determine a best match between a member of the set of one or more pruned pattern responses and the natural response and selecting output data from the output candidate vector associated with the forced response used to determine the best match.

37. The method as in claim 35 wherein the input signal data sample comprises audio input signal data.

38. The method as in claim 35 further comprising:

recording quantized input signal on storage media.