Linear prediction for bandpass signals based on nonuniform past samples

This paper concerns linear prediction of the value of a bandpass signal containing one or more passbands from a finite set of its past samples. The method of choosing prediction coefficients involves the eigenvector corresponding to the smallest eigenvalue of a matrix dependent on a function which is the Fourier transform of the set of intervals making up the passband. The method is developed for a set of arbitrary past samples and applied here to a set of “interlaced” samples that are nonuniform but periodic. The method applies to finite energy signals as well as to bandpass signals of polynomial growth, which connects to the theory of generalized functions. Computational examples are given of prediction coefficient values and of signal predictions