> Note that I don't think this affects the BEE2 design you sent out a > little while ago because you are doing a 2^13 point FFT with 8 points > at a time, so internally that is broken up into a 2^10 length biplex > FFT and a 2^3 direct FFT. Similarly, the PFB FIR will internally be 8 > 2^10 length filters.
OK. I didn't see where it had a problem, so that explains it. We are looking into doing *really* big FFT's on the bee for coherent dedispersion, so we may be looking for ways to use DRAMs instead of BRAMs in our designs. Thanks! John > Glenn > > On 5/6/08, G Jones <glenn.calt...@gmail.com> wrote: >> If you try to use a PFB FIR or FFT block with a large number of taps ( >> > 4K maybe), it doesn't work because you end up with long arays of >> coefficients which the drawing script tries to stuff in the field of a >> RAM/ROM block to initialize it. The fields have a finite maximum >> length, so you end up with an array that looks like [1 2 3 4 with no >> closing ]. Apparently it can cause Matlab to crash, but I've never had >> that experience. >> I tried to make a single input complex FFT last night w/ 2^13 channels >> and found that the initialization vector in the map ROM in the reorder >> block in the unscrambler was truncated. >> Glenn >> >> >> On 5/6/08, John Ford <jf...@nrao.edu> wrote: >> > > We are going to put that fix in the ported version of the library >> > > (it's not in the current green blocks). Right now you need to >> change >> > > those values to a variable and keep the backpopulate_mask script >> from >> > > removing that variable to get around the problem. >> > > -Terry >> > > >> > > >> > > On May 5, 2008, at 6:51 PM, G Jones wrote: >> > > >> > >> Hello, >> > >> Was a work around ever found for using the green blocks for long >> > >> FFTs/PFBs? I am running into the problem where the initial value >> > >> vector in a block ram is too long so it gets truncated. >> > >> > >> > Can someone expound on this? What is a "long" pfb/fft? Are there >> any >> > notes on this anywhere? >> > >> > John >> > >> > >> > >> > >> >> > >> Thanks, >> > >> Glenn >> > >> >> > > >> > > >> > >> > >> > >> >
