Forward Linear Prediction

Forward Linear Prediction adds data points with ampltitude to the end of the FID. It is similar to Zero-Fill, but adding points with amplitude can lead to sharper resonances especially if the FID was truncated due to too short of an acquisition time- the nuclei were not fully relaxed at the end of the FID. The linear prediction is accomplished by predicting what the amplitude of the points should have been after the last real data point acquired.

The Linear Prediction menu is located under the Processing tab then Zero-Filling/LP. Zero-Filling is discussed here:
Zero-Fill

Backward Linear Prediction is discussed here:
Backward Linear Prediction


1) To accomplish Forward Linear Prediction, click on LP Filling, then select Forward in the box below the Zero-Fill.

2) In the Linear Prediction box on the lower right, when you click on LP Filling, the software should choose the From, and To and the Method as well as the Basis Points and the Coefficients. The From is where to start predicting points, the To is the amount of points to predict to, the Method is the way the prediction is accomplished, the Basis points are the amount of points being used, and the Coefficients are the amount of coefficients being used for the prediction.

From = usually the prediction is started at the last acquired data point (Original).

To = the amount of points to predict to, normally 2 times to 4 times the amount of points acquired (Original).

Method = Typically Zhu-Bax is best, but there are other options.

Basis Points = Amount of points acquired used to predict the points. Normally, this is all acquired points (Original).

Coefficients = Typically the default value is fine.

3) The Spectrum Size option in the upper right of the window needs to be at least equal to the To points in the lower right. Any number of points in Spectrum Size that is greater the To points is zero-fill; however, if the Spectrum Size were smaller, then the linear predicted points will be ignored.

4) In 1D data, normally linear prediction should not be necessary as enough data points should have been acquired to prevent truncation of the FID. However, if you see sinusoidal side bands off of sharp resonances, the linear prediction option can potentially eliminate them. Thus, to confirm the effectiveness of linear prediction, zoom in on a sharp resonance, such as a solvent peak and see how that resonance changes with application of forward linear prediction.