home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dsp
- Path: sparky!uunet!hela.iti.org!usc!rpi!sriram
- From: sriram@sun.ipl.rpi.edu (R. Sriram)
- Subject: Reflection Coeffts
- Message-ID: <clp1l1h@rpi.edu>
- Nntp-Posting-Host: ipl.rpi.edu
- Date: Wed, 11 Nov 1992 03:19:15 GMT
- Lines: 36
-
- Hello,
- I've a problem on autoregressive process.
-
- A stationary process may be modeled as an autoregressive process (e.g
- speech). The reconstruction process involves driving the inverse filter
- formed by the polynomial 1 - sum[a_i z^(-i)], summation over the order
- of the polynomial. Now usually quantization of the filter coeffts is
- not done bec' of stability of the IIR filter. So one quantizes the
- reflection coeffts.
- My question is
- 1) How can one represent the mean square error of the quantization of the
- reflection coeffts in terms of the quantization of the filter coeffts ?
- 2) If (1) is possible, how does it translate into a mean square error of
- the reconstructed process ?
-
- My problem is actually a 2-D version where the polynomial in 2-D is not
- even separable. But atleast if I can get a handle of the 1-D case, maybe
- I can make some assumptions.
-
- My final goal is to get a rate-distortion curve for the quantization of
- the reflection coeffts vs the mean square error of the reconstructed process.
-
- Are there any references on this ? I feel this problem has been tackled in
- speech coding based on LPC technique. But don't have any references on that
- either.
-
- Any help is greatly appreciated since I've to complete my thesis by this
- year!!!
-
- Thanks to all in advance.
-
- I can post a summary of responses, if there are enough replies.
- Pl. reply to sriram@ipl.rpi.edu or to the net. Either way is fine.
-
- Sriram
-
-