syo6e@hudson.acc.virginia.edu (othman sinan younis) (06/19/89)
Please forgive me if this is not the right newsgroup to post the following
request. Here it goes....
------------------------------------------------------------------------
I need an *efficient* and complete TMS320C30 (Texas Instruments'
third-generation digital signal processor) code for minimum-distortion search
between an input vector of dimensionality k and an n-number of "codebook"
entries each of dimensionality k also. (Documentation would be most welcome.)
(Y11,Y12,....,Y1k)
(Y21,Y22,....,Y2k)
(X1,X2,....,Xk) .......
(Yn1,Yn2,....,Ynk)
input vector Codebook
The distortiom measures I am interested in are the mean square error and the
absolute error. So the task is to find j such that
k k
___ ---
\ 2 \
/ (Xi-Yji) or / |Xi-Yji|
--- ---
i=1 i=1
is minimum.
Note that some distortion calculations may be exited after few terms (<k) if
the cumulative error is found to be greater than the current minimum distortion.
This technique may help to reduce the total number of computations but
clearly requires frequent comparisons. My question is: am I right in thinking
that these frequent comparisons are not suited for TMS implementation and that
it is faster to perform each distortion computation in its entirety?
Your help is greatly appreciated.
Sinan Othman