Linear Prediction

Reference:  Barkhuijsen, H.; de Beer, R.; Bovee, W.M.M.J.; van 
Ormondt, D.; Journal of Magnetic Resonance, 61, 465-481 (1985).

For another description of use: Using linear prediction (from Texas AMU)

Linear prediction should not be used recklessly because you can
create artifacts.  Linear prediction can be used to predict
forward or backward and will not fix grossly undersampled data. 

To add linear prediction parameters to your experiment, addpar('lp')  
(for np dimension) or parlp.  The standard setup will be for 
backward prediction of the first point in the np dimension and for 
forward prediction in the ni dimension to correction for truncation.

To process using the linear prediction, set proc='lp'  (default 
is proc='ft').
To display the linear prediction parameters, dglp.

lpopt  - set to 'b' for backwards prediction/correction of points 
or to 'f' for forward prediction for next points.

lpalg - linear prediction algorithm, recommended set to 'lpfft' 
which does a least-squares calculation (in np dimension)

lpext - linear prediction data extension, number of data points 
to predict in np dimension

lpfilt - linear prediction coefficients, set to 8, 16, 32, 
larger #, smaller calculation

lpnupts - linear prediction number of data points in np dimension, 
number of points to use to do the calculation from, must be >= 2*lpfilt 

strtlp - the starting point to use for calculating the linear prediction

strtext - specifies the starting point for the linear prediction 
extension in np dimension

Adding 1 as an extension to these parameters addresses linear prediction in the ni dimension for example:
addpar('lp',1) - add linear prediction for ni dimension
lpalg1 - linear prediction algorithm in ni dimension
lpext1 - number of data points to predict in ni dimension
proc1='lp' - adds linear prediction to processing in ni dimension
...

The command >setLP1  will set linear prediction in the ni 
dimension for the completed number of FIDs during an acquisition.  This 
also set the lpext1 to 3X the ni.  If your data set is large, this will 
take a very long time to process.  Try decreasing your lpext1.  Sample 
processing might be:
>setLP1
>gaussian
>wft2da

Some sample parameters for a 2D experiment with ni=128:
	LP				LP 1
lpalg		lpfft		lpalg1		lpfft
lpopt		b		lpopt1		f
lpfilt		16		lpfilt1	 	  8
lpnupts	   64    	   lpnupts1	  128
strtlp		4		strtlp1		128
lpext		3		lpext1		384
strtext		3		strtext1	129


 

Back