More Bringing up a CPM

From: Richard Erlacher <edick_at_idcomm.com>
Date: Wed Jun 2 20:16:14 1999

please see embedded comments below:

Dick

-----Original Message-----
From: Dwight Elvey <elvey_at_hal.com>
To: Discussion re-collecting of classic computers
<classiccmp_at_u.washington.edu>
Date: Wednesday, June 02, 1999 5:18 PM
Subject: Re[2]: More Bringing up a CPM


>"Richard Erlacher" <edick_at_idcomm.com> wrote:
>--snip--
>> The logic analyzer won't hide significant information about the logic and
>> timing. If you sample at twice the frequency of the fastest harmonic you
>> want to observe, you won't miss a thing.
>
> It is not just frequency issues. Levels can be a problem as
>well. I've seen circuits with mixed HC, LS, 4000 and other.
>You set the input for one and the other is wrong. Also,
>your 2X has to be twice the frequency of the response
>of the IC circuit and not just twice the clock speed.
>Also, the 2X sampling rule is only true for repetitive
>signals. This is a basic sampling law thing.
>100 Mhz isn't really fast enough to truly see what is
>going on in even an old 8080 processor board. Glitch
>catching helps but can still hide correct operation.
> Where I currently work, we are completely out of the
>ball park were current day LA's are of much use but
>that is another problem. We still have them and can,
>under limited conditions, use them.
>
I can assure you that you won't run into much CMOS circuitry in the old
IMSAI and Altair stuff. You're right, of course, if your instrument can't
be programmed to handle different thresholds then you could be persuaded you
see something you really don't, or perhaps be swindled out of seeing what's
really there. I remember looking at some new CMOS circuitry about 15 years
ago when the high-speed CMOS stuff was still new. It appeared, for all the
world, as though the outputs preceded the inputs of a series of gates by
quite a difference. I had set my 2467 display to the lower threshold of a
cmos schmidt trigger and displayed a couple of other signals with respect to
that, including the clock which triggered a sequence. It appeared that the
'Q' of a flipflop preceded the clock by quite a margin, perhaps 3-4 ns. We
quickly concluded that a sufficiently long chain would easily allow us to
read tomorrow's stock quotations today, thereby making us all rich. - - -
it didn't work that way!

The way I get around the random sampling error in observing a crystal driven
microprocessor circuit is that I use a little PLL to multiply the clock on
the board by some constant, say 4, 10, 16 or some such, and sample at that
rate as opposed to what the LA wants to use. That makes similarly long
pulses look similarly long as opposed to suffering from the temporal
distortion introduced by the asynchronism between the LA sampling clock and
the circuit on the board under observation.

Of course, like a 'scope or signal generator, you have to know how to drive
them, just as you must know how to interpret the results. The LA often
points up where to hook the 'scope. If it cuts my search through 60-70
signals down to half a dozen, it's made itself worthwhile.

I find the LA really handy for MCU-driven cirtuitry, since I can use the LA
to take a picture and then spiff it up and print it in the O&M manual.
What's more, with easily simulated FPLD's, it's really handy to compare the
simulation with what the LA sees. It's often quite heartening to see what
you thought you ought to see after working on one of those babies for a
couple of weeks.

In any case, I've troubleshot dozens of DRAM circuits with either the little
front-end sampling mux or a real LA, and have found them to be invaluable in
finding component failures, incorrenct jumpering, wrong delay lines, etc.
What's more, if I didn't make the mistake, I can prove it. If I did, I can
prove I fixed it. That's worth quite a bit of setup.

As far as frequency is concerned, the only problem I've encountered which
doesn't quickly show up on a LA display is metastability. As fast as things
run these days, and with the widespread use of fully synchronous circuitry,
almost everything goes into a device through a two-bit shift register stage
to mitigate metastability. I find that if I sample at twice or quadruple
the clock which drives the pipeline registers, I can catch everything I
need.

Those old circuits like the IMSAI and ALTAIR CPU circuits relied quite a bit
on circuits' propagation delays to build events. Races were common, and
that's why folks used to tack 20-400 pf caps all over those boards in hopes
of making them work better.

Like I wrote before, setup and interpretation is slow, painstaking, and
laborious. It's often wrong the first time. Unless you have half-a-dozen
similar boards to work on, it's hardly worth the trouble.
Received on Wed Jun 02 1999 - 20:16:14 BST