Roundoff noise properties of lattice filters employing floating-point arithmetic

The roundoff noise properties of general lattice filters are studied for floating-point implementations using the concept of dummy multipliers. The main efforts are devoted to the Gray-Markel recursive structures. It is shown that the roundoff noise levels of these structures are closely related to their well-known low-sensitivity properties, and therefore, the low roundoff noise to signal ratio (RNSR) is obtained. For minimum phase FIR filters, one of the two-multiplier lattice structures is considered for the second-order case. It is found that its floating-point roundoff noise characteristics are better than that of the direct form realization under a certain condition.<
>