Error spectrum shaping in narrow-band recursive filters

The error introduced by zero memory quantizers can usually be modeled by an additive white noise component. By incorporating feedback and additional memory elements to these quantizers, it is possible to shape the error spectrum to advantage. Two simple error spectrum shaping quantizers are presented, one to be used with narrow-band low-pass filters and the other with high-pass filters. Simulation examples are presented where the reduction in roundoff errors is equivalent to three bits of data.