On floating-point implementation of modified wave digital filters

Modified wave digital filters (WDFs) have been developed to allow the optimal implementation of WDFs by standard commercial digital signal processors (DSPs). The basic theoretical principles behind these modifications are recapitulated. The effects of the modifications, concerning the realizability and the transfer characteristics, are demonstrated for allpass sections and lattice WDFs. Numerical examples are presented. Due to the finite register lengths of DSPs, the losslessness of the modified filter structures cannot be strictly preserved, but by adopting simple rules for the quantization of the filter coefficients, passive realizations can be guaranteed. The accuracy requirements for the coefficients and especially for the signals are discussed. It is shown that for obtaining all the available advantageous properties, the implementation should preferably be done floating-point DSPs