Re: [Discuss-gnuradio] ADC tied directly to CPU?

[Eric Blossom]

On Fri, Jan 21, 2005 at 05:39:27PM -0500, Bryan E. Chafy wrote:
> How feasable would it be to directly interface an AD to a pentium (1,2,or 3)
> class CPU?

You'd hang it off some standard bus, e.g., PCI, PCI express.

> For raw AD sampling say a satellite IF signal, the USB and even PCI is 
> marginal
> at best.

How fast do you want to sample?
64-bit 66 MHz PCI is pretty fast  (~500 MB/sec)

> Also, the protcols are not simple (nothing is easy and synchronous,
> always a burst, always a delay).  PCI/USB make FPGA's almost a requirement.

The FPGA isn't a requirement for PCI, however you might want one
anyway.  They are much, much faster than the host processor for
certain classes of problems.

> PCI Express looks nice and fast, but it's far away from cheap or practical at
> this time.
> 
> I was thinking something like Fig 9-11, or 9-12 in the following page:
> http://216.239.63.104/search?q=cache:b630kpcbVcoJ:www.ee.polyu.edu.hk/staff/eenc
> heun/WebSubject2/chapter9.htm
> It appears to look straightforward enough, but would the same concept
> work with a modern cpu/mobo combination?

It was straight-forward because the microprocessor it was hooked to
was running at something on the order of 1 MHz, not 1 GHz, and had a
very simple bus.

> I dont mean to create a new circuit board, but simply homebrew
> solder an AD converter directly to the CPU pins underneath an ordirary
> motherboard.  Doing an "in" at some hardwired address would read the contents.

I think you'll find it's a bit of a challenge.

For any given processor, there will be a hardware manual will explain
the organization of the bus.  One of the reason PCI exists is so that
there would be a standard bus to plug peripherals into.  What goes on
between the CPU / memory controller / north bridge / south bridge is
*highly* dependent on exactly which CPU chip you've got.  Grab one of
the pentium hardware manuals and take a look.

http://developer.intel.com/


> Since the FSB of a P3 class is 100 or 133mhz, a highspeed AD could be used
> (or slower if the clock is divided down).
> If successful, I know starvation is a problem at these rates, but if more
> than one motherboard were wired up in this way, perhaps some
> parallelism/staggering could solve that.?
> 
> Just a thought on highspeed AD on the cheap.

We had the same goal: getting high speed analog data in and out of the
PC for a reasonable price.  The USRP was the result.

Eric