AR #30937 - LogiCORE Fast Fourier Transform (FFT) v5.0 - Why does the description of the convergent rounding appear to be opposite of the conventional implementation of convergent rounding (round-to-even)?

AR#	30937
Part	IP-DSP Horizontal
Last Modified	2008-05-14 00:00:00.0
Status	Active
Keywords	convergent rounding, rounding type, round-to-even, round-to-odd, documentation, v3.2, v4.0, v4.1, v5.0

Solution

The description in the data sheet is incorrect. The FFT does implement a conventional convergent rounding process, also known as round-to-even.

The description of the FFT Rounding Modes should read as follows:

Rounding Modes

At the output of the butterfly, the LSBs in the data path need to be trimmed. These bits can be truncated or rounded using convergent rounding, an unbiased rounding scheme. When the fractional part of a number is equal to exactly one half, convergent rounding rounds up if the number is odd, and rounds down if the number is even. Convergent rounding can be used to avoid the DC bias that would be introduced by truncation.


See (Xilinx Answer 29209) for a detailed list of LogiCORE Fast Fourier Transform (FFT) Release Notes and Known Issues.