Optimal factoring of FIR filters

1. A method of determining factors of an FIR filter, suitable for building the FIR filter as a cascade of factors, the method comprising:

factoring, using a processing device, the filter's transfer function into 1 st order and 2 nd order factors based on its natural zero-pairings;

creating, using the processing device, a collection of N angle values (θ k ), where k=1, . . . , N, wherein each angle value corresponds to a 2 nd order factor of the filter's transfer function;

organizing, using the processing device, the pairings of N angle values into pairing candidates by creating a table with N rows and N columns, one row for each angle value (θ k ), and one column for each angle value (θ k );

entering, using the processing device, an estimated quantization cost for each pairing candidate in the corresponding row/column position of the table;

reorganizing, using the processing device, the entries in each row of the table to have an increasing estimated quantization cost to produce a reorganized table, and entering in each row/column position of the reorganized table an identifier of the column that the estimated quantization cost came from in the table;

defining a threshold to indicate a number of columns of the reorganized table of estimated quantization costs to employ in the determination of the factors of the filter;

generating, using the processing device, a matrix based on a total number of pairing candidates found in each row of the reorganized table and each column of the reorganized table as defined by the threshold;

determining, using the processing device, a subset of pairing candidates in the matrix that provides a lowest predicted total quantization cost between all pairing candidates represented within the matrix;

using the subset of pairing candidates to determine, using the processing device, coefficients of the filter factors; and

building the FIR filter having factors with the determined coefficients.

2. The method of claim 1 , wherein a given pairing candidate is represented by θp and θq, and a first coefficient of a given factor of the filter is provided as −2·(cos θp+cos θq) while a second coefficient of the given factor of the filter is provided as 2·(1+2·cos θp·cos θq).

3. The method of claim 1 wherein the quantization cost for a given pairing candidate is associated with a number of bits required to represent the angle values of the given paring candidate within a factor of the filter.

4. The method of claim 3 , wherein the number of bits does not include a sign bit.

5. The method of claim 1 , further comprising:

exchanging a first entry of the reorganized table from a column defined by the threshold with a second entry of the reorganized table from a column outside of the threshold, wherein a quantization cost of the first entry is the same as a quantization cost of the second entry.

6. The method of claim 1 , wherein the determining comprises using Binary Integer Programming to determine a binary vector x that minimizes a linear cost function.

7. The method of claim 6 , wherein the binary vector x is constrained by the inequality Ax≦b, where A represents the matrix and b represents a vector consisting of ‘p’ values, where p is a positive real number.

8. The method of claim 7 , wherein p=1.

9. The method of claim 6 , wherein the linear cost function is represented by f T x, where f T is a cost vector.

10. The method of claim 9 , wherein one or more entries of the cost vector are related to one or more quantization costs among pairing candidates.

11. The method of claim 9 , further comprising updating the entries of the cost vector in a feedback loop until the actual total quantization cost is substantially the same as the predicted total quantization cost.

12. The method of claim 1 , wherein the natural zero-pairings comprise 4 th order factors.

13. The method of claim 1 , further comprising:

after determining the coefficients of the filter factors for a given threshold, defining a different threshold to indicate another set of columns of the reorganized table of estimated quantization costs and repeat the generating, determining and using with the columns of the reorganized table defined by the different threshold.