Processors White Papers
Numerical Performance of Fixed-Point DSP Data Paths, or, What are Guard Bits and Why Should I Care?
Overview This paper explores the ways in which quantization is introduced into fixed-point implementations of DSP algorithms, using FIR filtering as an example. On fixed-point DSP processors, quantization usually occurs when a signal is scaled to an appropriate range. The design of the processor's data path often dictates points in the computation where such scaling is necessary or places restrictions on where scaling is possible. The relationships between architectural features of fixed-point DSP processors and quantization in DSP algorithms are discussed. A comparison is made between use of guard bits, scaling of intermediate results, and scaling of filter inputs or coefficients to eliminate the possibility of overflow.
| Publisher | Berkeley Design Technology, Inc. (BDTI) | File Format | HTML |
|---|---|---|---|
| Date Published | August 2003 | Downloads | 3 |
| Format | White Papers | ||
| Topics | |||



