The frequency-domain analysis concerning the response of nested-cascad
e multiple input/output systems requires computation of the cross-spec
tral density matrices that involve the input, intermediate, and output
vectors. Clearly, as the number of nested systems increases, the orde
r of the cross-spectral density matrix increases, demanding additional
computational effort. This feature lessens the computational attracti
veness of the frequency-domain analysis. A stochastic decomposition te
chnique is developed that improves the efficiency of conventional freq
uency-domain analysis by eliminating the intermediate step of estimati
ng cross-spectral density matrices. Central to this technique is the d
ecomposition of a set of correlated random processes into a number of
component random processes. Statistically, any two processes decompose
d in this manner are either fully coherent or noncoherent. A random su
bprocess obtained from this decomposition is expressed in terms of a d
ecomposed spectrum. A theoretical basis for this approach and computat
ional procedures for carrying out such decompositions in probabilistic
dynamics are presented.