Primarily for use when the filter coefficients have been specified using non-integer real numbers, this option quantizes the coefficients to the specified coefficient bit width. The coefficient values are rounded to the nearest quantum using a simple round towards zero algorithm. The coefficient word is split into integer and fractional bits. The integer width is determined by analyzing the filter coefficients to find the maximum integer value. The remaining bits are allocated to represent the fractional portion of the coefficient values. When the specified coefficient bit width is less than the required integer bit width, coefficients are appropriately rounded. The default value for the Coefficient Fractional Bits parameter is set to maximize the precision of the coefficients, but you can reduce it. In this circumstance, more bits are allocated to the integer portion of the word, and the coefficient values are sign extended appropriately. When all the specified coefficients are between 1 and -1, only a single integer bit is required (to convey sign information), with the remainder of the coefficient word being used for fractional bits. When the coefficient range reduces further, the fractional bit width can be specified to a value greater than or equal to the coefficient width. See the Best Precision Fractional Length section for further explanation.
The frequency response of the quantized filter coefficients are compared to the ideal response on the Frequency Response Tab. This enables you to explore the trade-off between filter performance and resources by varying the coefficient width parameter.