The singular value decomposition (SVD) algorithm for the AR spectral estimates of the SMMW Fourier transform spectroscopy data is presented. The SVD algorithm has the same resolution as the least squares algorithm using forward and backward linear prediction (LS algorithm) has

but the SVD algorithm can avoid the spurious spectral peaks and ill-conditions appeared with the LS algorithm

and suppress the noise component in the data. The SVD algorithm is more stable and reliable than the LS algorithm. The selections of the order and the number of the effective singular values of the SVD algorithm are discussed.