From: "Eric Smith" <eric(a)brouhaha.com>
I wrote:
> Without decoding the data, it is proposed to
sample the data at 50 Mbps
> or more.
Andy wrote:
Why so fast? You only need sample at 2x the
bandwidth.
2x is the theoretical minimum; sampling at 2x isn't always adequate
in practice due to various real-world limitations.
The minimum write data pulse width spec for the WD1000 controller
is 60 ns. If we were sampling and reconstructing a sine wave with
a 120 ns period, 16.7 MHz sampling would be adequate. But that
Hi
Even for a sine wave, the brick wall filter would be difficult
to create. You need a little more room.
wouldn't yield good results for square waves, and
there wouldn't be
any margins. In practice, I suspect that 33.3 MHz sampling would be
barely adequate provided that no other problems arise. Operating at
50 MHz seems advisable to have reasonable margins.
The thing to consider is that we are not sampling a "square" wave.
It is a variable pulse width digital signal. The only kind of
processors I know of that could handle this from a port is some
of the latest DSP chips. Using a PC, even one like our ( AMD )
latest Opteron wouldn't do well since it is so highly pipelined
and unpredictable in execution speed at this short a time
interval.
There are a number of DSP evaluation boards out there that places
like ADI and TI sell that could be made to handle this task.
Of course, a simple pulse width counter on the front end could
simplify the sampling process.