Biorthogonal representations have been widely used in signal processin
g. Gabor expansion and wavelet transform are two popular ones. When th
e signal, expressed using orthonormal representation, is to be approxi
mated by partial set of the basis, the criterion to obtain LSE approxi
mation is to discard those basis vectors corresponding to smaller coef
ficients of the representation. However, such a simple strategy for di
scarding/retaining basis vectors does not hold for biorthonormal cases
. The re-calculation for coefficients is needed in order to avoid larg
e distortions. In this paper, we propose and investigate into several
algorithms for finding the suboptimal subspace to represent the signal
s, and present the error analysis pertaining to biorthonormal signal r
epresentation. (C) 1997 Elsevier Science B.V.