Check out the Xilinx SSR FFT block in the system generator blockset. It will make timing on the ZCU111 in 8x4096 at 512 MHz and is very efficient. They also have and HLS SSF FFT that isn’t quite as performant (but has public sources. It is in the Vitis IP library on github.
We are using the former in our overlapped PFB. -Jeb —— Dr. J.I. Bailey, III (Jeb) / Project Scientist Mazin Lab <http://web.physics.ucsb.edu/~bmazin/> / Department of Physics, UCSB jebbailey.com / +1 (734) 389-5143 / skype:spacecolonyone > On Nov 18, 2020, at 10:55 PM, James Smith <jsm...@ska.ac.za> wrote: > > Hello Morag, > > AFAIK, Xilinx's FFT IP block only accepts a single block at a time. I think > that was a large part of the reason why there's a CASPER FFT in the first > place - because we needed multiple samples at once. > > We have used the Xilinx core in narrowband designs in the past - KAT-7's > narrowband modes were done that way, after a first "coarse" channelisation > stage using a Casper PFB, and an async pfb_fir followed by the Xilinx FFT > block. If you want to pick them apart, they are in here: > https://github.com/ska-sa/kat7_fpga/tree/master/fengine_oh > <https://github.com/ska-sa/kat7_fpga/tree/master/fengine_oh> (warning: newer > versions of Matlab / mlib_devel may struggle with these old files. AVN's > narrowband R2 design is somewhat more modern if you have a R2 setup handy, > they are based on similar principles: > https://github.com/ska-sa/AVNRoachGateware/tree/devel/source/NarrowBandSpectrometer > > <https://github.com/ska-sa/AVNRoachGateware/tree/devel/source/NarrowBandSpectrometer>) > > Regards, > James > > > On Wed, Nov 18, 2020 at 5:53 PM Morag Brown <mbr...@ska.ac.za > <mailto:mbr...@ska.ac.za>> wrote: > Hi all, > > Has anyone ever put together a wideband FFT (i.e an FFT that accepts multiple > demuxed samples per FPGA clock cycle) using the Xilinx sysgen FFT block? > Looking at the docs, it seems the IP core can't be configured to accept > multiple parallel inputs (but I could be wrong), so one would need to > construct the 2D decomposition of the 1D FFT using Xilinx FFT blocks as > primitives. > > I'm doing some FFT comparisons for my MSc, and would like to include the > Xilinx core. I thought it'd be good to check if anyone's already implemented > this before embarking on the adventure myself... > > Morag Brown > SARAO > > -- > You received this message because you are subscribed to the Google Groups > "casper@lists.berkeley.edu <mailto:casper@lists.berkeley.edu>" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to casper+unsubscr...@lists.berkeley.edu > <mailto:casper+unsubscr...@lists.berkeley.edu>. > To view this discussion on the web visit > https://groups.google.com/a/lists.berkeley.edu/d/msgid/casper/CAGH-0Teev7EW%2BN6EJ2EPAtE%3DL%2BgBASV7RE8ea0uSkz9ehtwsqA%40mail.gmail.com > > <https://groups.google.com/a/lists.berkeley.edu/d/msgid/casper/CAGH-0Teev7EW%2BN6EJ2EPAtE%3DL%2BgBASV7RE8ea0uSkz9ehtwsqA%40mail.gmail.com?utm_medium=email&utm_source=footer>. > > -- > You received this message because you are subscribed to the Google Groups > "casper@lists.berkeley.edu" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to casper+unsubscr...@lists.berkeley.edu > <mailto:casper+unsubscr...@lists.berkeley.edu>. > To view this discussion on the web visit > https://groups.google.com/a/lists.berkeley.edu/d/msgid/casper/CAG67D34M_7Fnz18OjZ8wHpW3tFuOUyH_RjaKm3jbZVD7dhsiug%40mail.gmail.com > > <https://groups.google.com/a/lists.berkeley.edu/d/msgid/casper/CAG67D34M_7Fnz18OjZ8wHpW3tFuOUyH_RjaKm3jbZVD7dhsiug%40mail.gmail.com?utm_medium=email&utm_source=footer>. -- You received this message because you are subscribed to the Google Groups "casper@lists.berkeley.edu" group. To unsubscribe from this group and stop receiving emails from it, send an email to casper+unsubscr...@lists.berkeley.edu. To view this discussion on the web visit https://groups.google.com/a/lists.berkeley.edu/d/msgid/casper/2EF47892-3603-49B1-A461-091304CE91A5%40ucsb.edu.
