ESTIMATING OPTIMUM FILTER LENGTH IN LINEAR PREDICTION

In this article a new method to estimate optimum filter Length in line ar prediction is described, Linear prediction was used to enhance reso lution of a spectrum. In particular, the dependence of prediction erro r on filter length has been studied, With calculations of simulated sp ectra it is shown that the prediction error falls rapidly when the fil ter length attains its optimum value. This effect is quite pronounced when the spectrum has a good signal-to-noise ratio and the modified co variance method is used to calculate prediction filter coefficients, T he method is illustrated with applications to real Raman spectra.