Computationally efficient QMF filter banks

tion which provides performances comparable to that We show now that a near—perfect replica R(n) of of conventional QMF filter banks, but with a compu— x(n), with s—transform I(s) may be reconstructed tational complexity which is significantly lower from the N channel signals. This is done by insar— than with QNF banks or earlier pseudo-QMF banks. ting N-l zero-valued aamples between auccassiva samples of the channel signals, by filtering the PSEUDO QNF FILTER BANK resulting sequences with the filters h(n)cos[27f(2k+l) (2n-I-N)/8N], and by summing the reWe consider a signal x(t) which is band—limited sulting signals. Hence to a frequency f5/2 and sampled at rate f, with w = 2sf = 27T/T . We want to split the sampled in— 5 a 5 put aignal x(n) into N sub—band signals (Nn) sub—