--- Quote Start ---
I am using FFT 13.0 megacore function for the IFFT end FFT. The modelsim results and the matlab m file results obtained by the megawizard created are the same.
--- Quote End ---
Because these two models are bit-accurate, i.e., the MATLAB model matches the calculations used by the HDL.
--- Quote Start ---
However when I use the "ifft" function in matlab it differs.
--- Quote End ---
That is to be expected, since the MATLAB fft/ifft routines use floating-point math and may have different scale factors.
Here's a document showing how the Altera FFT matches the MATLAB fft within the quantization noise floor of the fixed-point data being transformed;
https://www.ovro.caltech.edu/~dwh/correlator/pdf/fft_tutorial.pdf This document has some slides with hardware FFTs (pages 50-54)
https://www.ovro.caltech.edu/~dwh/correlator/pdf/hawkins_jpl_2014.pdf and this has more (page 85-99)
https://www.ovro.caltech.edu/~dwh/correlator/pdf/ee_live_14_hawkins.pdf The Altera FFT is a streaming FFT. You can download the He and Torkelson paper referenced in the User Guide and it explains the operation.
--- Quote Start ---
What is the "SVSfftmodel" that magacore matlab m file uses.
--- Quote End ---
The bit-accurate model that implements the Altera FFT.
--- Quote Start ---
How does that model works?
--- Quote End ---
That is part of Altera's "secret", but if you read the tutorial linked above, you'll see pretty much how it probably works.
Cheers,
Dave
Honored Contributor
