We explore two new applications of the filter-diagonalization method (FDM) for harmonic inversion of time cross-correlation functions arising in various contexts in molecular dynamics calculations. We show that the Chebyshev cross-correlation functions c(i alpha)(n)=(Phi(alpha)\T-n((H) over cap)Phi(i)) obtained by propagation of a single initial wave packet Phi(i) correlated with a set of final states Phi(alpha), can be harmonically inverted to yield a complete description of the system dynamics in terms of the spectral parameters. In particular, all S-matrix elements can be obtained in such a way. Compared to the conventional way of spectral analysis, when only a column of the S-matrix is extracted from a single wave packet propagation, this approach leads to a significant, numerical saving especially for resonance dominated multichannel scattering. The second application of FDM is based on the;harmonic inversion of semiclassically computed time cross-correlation matrices. The main assumption is that for a not-too-long time semiclassical propagator can be approximated by an effective quantum one, exp[-i (H) over cap(eff)]. The adequate dynamical information can be extracted from an LxL short-time cross-correlation matrix whose informational content is by about a factor of L larger than that of a single time correlation function.