Hi JanIt sounds like you want the wideband real pfb. This exists in the 10.1 toolchain as pfb_fir_real. Select 2^2 number of inputs and you'll get four real inputs. If you make it biplex, you'll get another four inputs basically for free which you can use for a second input.
You can't do an xlModelUpdate on the libraries because these blocks are actually drawn by matlab scripts. When the script next runs, it will simply revert the block to it's outdated form. If you can't get the 7.1 toolflow setup EXACTLY as outlined at http://casper.berkeley.edu/wiki/index.php/MSSGE_Toolflow , then you need to download the new 10.1 libraries. These are still in heavy active development and I strongly suggest you get an SVN client and do daily updates. Expect some bugs. Anonymous checkouts are possible: on Linux, svn co http://casper.berkeley.edu/svn/trunk. On windows, try TortoiseSVN and enter http://casper.berkeley.edu/svn/ trunk as the SVN root on the server.
I have attached a model file for a basic spectrometer on the new ROACH hardware using the 10.1 flow which seems to do what you want. Ensure you have the correct service packs installed and the right version of Matlab. Only two are Matlabs are supported. I recommend 2007b because it's what I use :) See http://casper.berkeley.edu/wiki/index.php/MSSGE_Toolflow_10.1_Notes . 64-bit versions don't work.
To compile this design for the IBOB, simply change the target platform to IBOB in XSG core config and the LED lines and it compiles for the IBOB no problem.
roach_spec_no10gbe.mdl
Description: Binary data
Jason On 03 Feb 2009, at 17:30, Jan Wagner wrote:
On Tue, 3 Feb 2009, Jason Manley wrote:ASTRO lib won't work with the 9.1 tools :-( 7.1 will only work with Matlab R14 :-(Ookay :-/If you want to feed the 1:4 demuxed ADC data directly into the PFB_FIR, then you need the wideband version. You'll notice in the Medicina packetised F engine that we're using a DDC up front which extracts 1/4 of the band (hence only one complex input to the PFB_FIR).Yep. I was kind of wondering about that DDC part.Are you trying to do an FFT on a real-sampled signal (in this case pick the pfb_wideband_real and set the number of simultaneous inputs to 2^2 and feed in the four real signals from the ADC),Is the pfb_wideband_real found in the beta/devel mlib_devel_10_1 for 10.1i?Does it work on the iBob?or are you doing complex sampling on the ADC's Q/I inputs with a 90deg out of phase clock sampling each input? There are different versions of the PFB_FIR for each situation.Just trying something really simple for starters.Get the real-sampled 512Msamp/sec "I" input signal that is spread to i0..i3. Low- or bandpass-filter with a 31/32-tap 4:1 decimating FIR in polyphase decomposition. The Xilinx DAFIR v9.0 compiler supports multirate decimating FIR with a single input, but not parallel/ polyphase -- so i used four xilinx fir blocks plus adders to accomplish this now.The closest from the interface point of view is the Casper / downconverter/dec_fir but I realize now that it is 2-tap only.And PFBs/pfb_fir_real would do the job if I'd extract the desired band(s) from the output after the filterbank output cornerturn. Too bad the PFBdoesn't work in our tool versions :-/(I tried casper 8.2 for our Xilinx 9.1i but the required xlUpdateModel('casper_library') causes Matlab to segfault after a few minutes of running)We managed to get the Xilinx 10.1i today for free.If the pfb_wideband_real is found in the 10.1 casper i'll try it out tomorrow. Maybe things wont crash over there in 10.1 ... ;-)- Jan
