Andrew Siemion wrote: >William Mallard wrote: >>Tom Kuiper wrote: >>>Does anyone have a feeling for how many 32,768 channel 1 GHz wide >>>spectrometers (or 65536 channel 2 GHz spectrometers) could be >>>implemented on a ROACH? >> >>I'm already working on a 3GHz FX correlator, so i forked it, >>truncated it at the end of the F engine, dropped the PFB FIR, >>resized the FFT to 32k channels, and hit 'go'. We can only sample >>1.6GHz of bandwidth at the moment, but we're working on getting it >>up to 3GHz eventually. >> >>For what it's worth, here's the resource utilization for a 32k >>channel (64k point wideband real) FFT at 1.6GHz: ... >>That includes BRAMs for reading out both the raw ADC data and >>spectral powers, and is targeted for an sx95t. Hope that helps.
Hi Andrew, As i said, this is an F Engine, not a spectrometer. The capture blocks happened to already be there for my own debugging purposes, and i haven't gotten to the vector accumulator for my correlator. >Does this design just snapshot the fft values or can it accumulate? It just snapshots the fft values. It does not accumulate. Actually, since i was using the BRAMs to debug the correlator, it snapshots the powers of the fft values. But i could easily drop the power blocks. >Did you resize the snap/vector accumulator for 32k channels? Yes, i resized the capture BRAMs for 32k channels. I'm using custom capture blocks based on the snap blocks from the library, not snaps. >Why drop the FIR? Do you think it won't fit? That's what i thought, but it looks like it might after all. Billy
