Browse Prior Art Database

EFFICIENT FILTER STRUCTURE FOR SAMPLE RATE CONVERSION INTERPOLATION AND DECIMATION

IP.com Disclosure Number: IPCOM000008707D
Original Publication Date: 1998-Jun-01
Included in the Prior Art Database: 2002-Jul-04
Document File: 3 page(s) / 120K

Publishing Venue

Motorola

Related People

Patrick McAndew: AUTHOR

Abstract

This appendix describes the implementation of an efficient filter structure for use in 8 Khz-16Khz interpolators and decimators which may be used in sample rate converters for audio conversion in emerging video telephony or audio sub-system applications. Specifically, this system was designed to provide conversion between G.71 l/G.728 and G.722 (narrowband to wideband audio) within H.320 for use in the QomsrM program.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Page 1 of 3

0 M

MOTOROLA Technical Developments

EFFICIENT FILTER STRUCTURE FOR SAMPLE RATE CONVERSION INTERPOLATION AND DECIMATION

by Patrick McAndew

  This appendix describes the implementation of an efficient filter structure for use in 8 Khz-16Khz interpolators and decimators which may be used in sample rate converters for audio conversion in emerging video telephony or audio sub-system applications. Specifically, this system was designed to provide conversion between G.71 l/G.728 and
G.722 (narrowband to wideband audio) within
H.320 for use in the QomsrM program.

   As described, this structure is implemented in firmware on the DSP563xx and DSP560xx digital signal processors. In order to minimize the mip requirements, the interpolator is implemented as a textbook polyphase FIR filter type. The decimator is implemented as a normal FIR filter but runs at the lower output sampling rate.

  The structure described above requires a filter of size N words. Although, in a typical implementa- tion, this filter size may be reduced by a factor of 2 by storing the first half of a symmetric filter, this arrangement when implemented on the above digital signal processors requires additional instruc- tions for resequencing of the filter pointers at the symmetric mid-point.

  The following description is of a more novel structure which also reduces the filter size to N/2, but yields an ordering sequence which has no over- head for pointer manipulation for the generation of the filter's second half.

  Firstly, a prototype filter of length N (N coeffi- cients), constrained to be even symmetric, is designed. For a filter of this type,
then,

Figure I shows such a filter and the inherent coefficient symmetry.

Length N

h,(N-1)

= h.(O)

kPJ-2) = ho(l)

ho([N/2]-1) = WJ/*)

a

Fl

= h,(N-2) = h,(N-1)

Fig. 1 Prototype Filter Structure

  From this prototype filter hp(k), a new filter h(k), of length N/2 is created by extracting the even ordered coefficients as follows:

h(i)=h,,(2*j) where j=OK(i-1)

  Due to symmetry, this new filter structure directly provides both even and odd ordered coeffC cients. In addition, both coefficients sets may be accessed by linearly incremented pointers, ptr, and ptrO. Thus from this one physical filter set, two logical filter sets for even and od...