Re: [Discuss-gnuradio] Alternative interface from PC to a USRP.

[Lamar Owen]

On Monday 11 July 2005 12:43, Matt Ettus wrote:
> Sounds like a very interesting application.  However, getting all those
> bits in and out of the computer would only solve half the problem.  You
> need to be able to perform computations on that data, and your PC would
> not be able to keep up.  The best solution is to preprocess in the FPGA,
> which you are doing with the MAPstation.  What is the data rate coming
> out of the MAPstation and into the computer?

Well, the MAPstation packs three Xilinx XC2V6000's (there are a couple of 
other options; see http://www.srccomputers.com/HardwareSpecs.htm#MAP for 
details). The MAPstation integrates a MAP Direct Execution Logic (DEL) 
coprocessing engine with a high-end Pentium 4.  For the algorithm we're 
using, it should be equal to or faster than an 800GHz P4. 

We're probably not going to push more than about 100MB/s to the main CPU for 
final data massaging (CLEAN).

The application is autocorrelation, and all channels must be simultaneously 
autocorrelated.  Don't know if the EP1C12 is powerful enough to do it, even 
with only 4 channels at 64MS/s.  The MAP's two user-available XC2V6000's 
should be able to handle it, even with 16 channels spewing 200MS/s per 
channel.  (Yes, 4.8GB/s).

The big advantage SRC gives is the Carte programming environment that lets you 
write your algorithm in straight C or Fortran and have the compiler target 
the critical paths to the FPGAs for direct execution on the hardware (without 
writing a single line of Verilog).  If you want to write a Verilog block to 
help, you can, but you're not forced into it.

Incidentally, the 'SRC' in SRC Computers is a man's initials.  A relative 
unknown by the name of Seymour R. Cray.
-- 
Lamar Owen
Director of Information Technology
Pisgah Astronomical Research Institute
1 PARI Drive
Rosman, NC  28772
(828)862-5554
www.pari.edu