Browse Prior Art Database

Coding a Band Split Signal

IP.com Disclosure Number: IPCOM000087559D
Original Publication Date: 1977-Feb-01
Included in the Prior Art Database: 2005-Mar-03
Document File: 2 page(s) / 52K

Publishing Venue

IBM

Related People

Croisier, A: AUTHOR [+4]

Abstract

In a coder, the allocation of coding resources may be optimized by splitting the original voice signal into a number of adjacent sub-bands, and nonsignificant sub-bands before quantizing the remaining bands.

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 54% of the total text.

Page 1 of 2

Coding a Band Split Signal

In a coder, the allocation of coding resources may be optimized by splitting the original voice signal into a number of adjacent sub-bands, and nonsignificant sub-bands before quantizing the remaining bands.

If performed in a standard way, this method leads to noise effects on the reconstructed signal. However, such a noise can be avoided by using quadrature-mirror filters in a tree decimation/interpolation structure. In other words, the original signal will be first split into two adjacent sub-bands using one low-pass digital filter H(1) and one high-pass digital filter H(2), H(2) being the quadrature mirror of H(1), i.e.,

(Image Omitted)

where H(1)(omega) .e/jphi/1/(omega)/ denotes the complex frequency response of the filter H(1), and omega(s) = 2 pif(s) with f(s) being the sampling frequency of the input signal. On each sub-band, the sampling rate will be halved by decimating every other sample. This results in a total information rate equivalent to the input rate, but distributed over two adjacent channels. Then, each sub- band will be split and processed in a way similar to the one above, and so on. Finally, an energy measurement will designate the channels to be selected for quantization.

At the receiving end, an inverse tree structure will enable the reconstruction of the original signal with no echo-type noise. In order to do so, each branch of the tree will include two mirror filters H'(1) and H'(2) which, for any value of frequency f = 2pi over omega, will be...