I do not see this behaviour with tut3 or any of my other designs using the simple_bram_vacc. I don't believe this is a "CASPER" problem, but rather something specific to your implementation or design.
There were bugs in the FFT in early 2011 (January and April), which were fixed in SKA-SA's git repo in April. It seems there was a github merge in July at least. So I'd guess your September checkout should included the fixes, and have known-good DSP blocks. Based on this, you might want to start checking your design itself (bitwidths, casting, reinterpreting etc are typical errors that cause this sort of behaviour). If you don't have time to debug your design and want an off-the-shelf, known good high-res spectrometer with high dynamic range that "just works", consider RATTY1. This does 16k channels over 900MHz and includes a full software suite that is properly calibrated. https://github.com/ska-sa/ratty1 We use this for RFI monitoring and EMC characterisation. There's a 32k channel version available for ROACH2 as well (RATTY2) and we've got a 1-million channel model in the works for ROACH-2... watch this space! Jason On 28 May 2013, at 22:35, [email protected] wrote: > Various people urged me to post this. > > My doctoral thesis is due a couple of > days from today and I have problems with > CASPER stuff that worry both my advisor > and me. We have two spectrometers: one > with lower resolution but all the > features I wanted and one based on > CASPER Tutorial 3 but transformed into a > higher-resolution spectrometer. > > The lower-resolution, full-featured one > uses a simple_bram_vacc but the output is > a pattern which forms groups of four data > words in a pattern that repeats through > the entire output. The values don't > exactly repeat, but the overall pattern > repeats. For example, sometimes the first > value is large, the second is small and > the last two values in each group of four > are medium (neither large nor small). I > believe this is a timing issue. > > The higher-resolution spectrometer uses a > Bee2 ADC (iADC). NO matter what signal we > put inot this spectrometer, the output > FFT has the same amplitude. This prevents > us from labelling the vertical axis with > units. The thesis committee has already > said graphs without proper units are not > acceptable. Is the problem caused by an > Automatic Gain Control or could this > problem also be a timing issue with the > accumulator? > > Any comments? People have been asking > questions before I can even finish typing > this message. > > >
