DIYHiFi.org

Curiouser and curiouser...

Lets's start with the comparator. Looking at the correct data sheet, we can deduce an ouput impedance in one of at least three ways:

a) Voltage drop versus output current. (This is what I suggested Hifi do for his 'HC86). In this case at (say) 9 mA output current we get and output impedance of 31 ohms when pulling down and 200 ohms (!) when pulling up. This just doesn't seem right...

b) Let's modify the first method by noting that even with no load there is a voltage drop across the output stage. So instead of comparing the output voltage to the rail voltage, let's compare it to the unloaded output voltage. This method gives an output impedance of 11 ohms when pulling down and 88 ohms when pulling up. This is closer, but still doesn't seem right. Also note that this method doesn't help on the 'HC86, as at least according to the Fairchild datasheet there is *no* voltage drop when the current draw is <= 20 uA.

c) Let's go a step further and note that the output impedance is the slope of the curve given in the AD8561 datasheet. But these curves are curving a fair amount, especially for the "pulling up" curve. When pulling down at 9 mA output current, it looks like Zo = 10 ohms is about right, and about 20 ohms for pulling up. This is about the same as Jocko is reporting, but still not what George is reporting.

Finally, let's look at George's waveform posting. The transition at the top of the waveform is textbook perfect, while that at the bottom shows more rounding. Based on the points noted above, I would actually expect the reverse to be occurring. George, would it be possible to show us the waveforms for deliberately undersized and oversized series resistors? Say 25 ohms and 100 ohms?

Charles,

AD8561 -

Vhiew, I will never learn to cite myself, better said, my own notes taken on the flight without deeper digging into..
My only excuse is that while noting this behaviour, that is, that with 27 ohm I got a more distorted waveshape, is still true, but.. only in those special circumstances, which occurred in that setup. And I was concentrating on other type of problems, and it was just a side note. [Was trying to have a go with my phase noise setup..:))]

So, I went back and had a look again.
For consistency, here it is the same setup like before, only the cable is 5nsecs now, the oscillator is 10MHz now, and the clock PS is a battery.
So there are minor shifts, but I would say, generally the same pic.
AD8561 -50ohm- rg174 5nsec - scope 50ohm input.

The strange shape of waveform is a general characteristic of the AD8561.
Once I have gone through all the posted pics at diya, and it's always the same when loaded with 50 +50 =100ohm.

Now I've changed to 27 ohm. Basically the same again! though one can note the higher output, and also the stronger assymmetric distortion of the signal, because of the higher load current.
Bur no major reflections!
What the heck!

Now I was forced to think.. Never do that until the situation is not desperate!

By the way the final blow was the 120 ohm"termination" - everything just got smoother, less assym. distortion, no reflections!! [also higher output]

The problem is that in this type of setup at the OTHER end, at the scope,
I'm terminating really good. And here we got to - maybe - even a useful conclusion: until at least ONE end in a point to point cable setup is terminated well, you can get away with quite a big amount of sheet..

By the way, small secondary reflections are visible even in this setup - to have a look at them, we should dive into "TDR" mode. AM I right, Jocko?
I think the TDR way would be to compare this bump value to the original signal amplitude, and get a percentual value. Here it should be 12mV/848mV, that is, 1,41%. Though this is really just a guess, Jocko maybe could explain well.

One more thing, the bump in the previous pic is exactly 10,2 nSec away from the rising edge - twice the signal cable length [5nsec].

So, but if I would like to see better the changes caused by my improper termination at the generator end? Well, for this I should not terminate at the receiver side! and so wreak a total havoc..
And this is what was happening to me previously - I was not using the scope, but my home made DB schottky mixer for the receiver end termination - and it was far from perfect.. So I saw a stronger difference.
Let's have a look!
AD8561 - 27ohm termination - rg174 5ns - 1Mohm at the scope.

And the same thing with 50 ohm.
That is, AD8561 - 50ohm - rg174 5ns - 1Mohm at scope.

Finally, "padding" is a general trick to improve the termination at your device.
This means that small divider / attenuator network what Jocko has shown several times, and Hifi included in his last config. Here I used an industrial high quality "barrel" attenuator chain, because padding in more steps [cascading] gives even better results. I've used a 10dB /1db /1db chain, that is, 12 db attenuating on my clock output with 27 ohm.

AD8561 - 27ohm - 12db padding [barrel att.] - rg174 5ns - 1Mohm scope end

So, the conclusion[s] - One can NOT terminate properly an AD8561 output, as Charles had pointed out correctly - It has an assymmetric drive, which is very well visible in the 50 ohm pic. There the rising edge has a much higher output impedance, and there are strong reflections generated. while at the falling edge the drive is much stronger, the output impedance drops to a few ohms or below, because a quite clean sequence is visible following there, and only a minor reflection. The level of that, by guessing, is ~ 5-10%, so we are several ohms, ~5ohm ??, away from the 50 ohm nominal value..

On the other hand, with the 27 ohm I see an equally unbalanced situation folowing both edges - this gives some hint why I was able to observe a better shaped signal with the 50 ohm term. in my mixer setup.

Though, if You can do padding, You might be saved - if you can permit the signal loss, which is the case with the SPDIF drivers..

Faithfully yours,
George

Wow, George! Thank you for your fine work!

Your second set of measurements with 1 Meg 'scope termination is more reflective of real-world CMOS usage. I don't think anybody uses double-terminations for internal circuit traces with CMOS logic. Maybe for ECL, and of course for an external S/PDIF connection.

But even if you are using double termination, it is much easier to see what is happening with the source impedance with the 1 Meg 'scope termination. The pictures you posted were (almost - see below) textbook examples of what happens with a single-terminated cable with an impedance mis-match.

In all cases, the source impedance was too high -- resulting in a stair-step function to approach the final output level. If the source impedance had been too low, then you would see an effect that looks like overshoot and ringing, but is actually also made of stair steps.

You could also see that there was less stair-stepping on the bottom transitions. This is because Zo is significantly lower on the pull-down side for the AD8561.

The only think I didn't understand was why you had *more* stair-stepping with the 27 ohm resistor than with the 50 ohm resistor. These resistors are in series with the output of the AD8561, aren't they? Is it possible that you reversed the graphs?

Thanks again, George -- excellent work!

Thanks, Charles, for the kind words. Though I could have been better prepaired.. so I don't consider this a great achievement, but maybe a public way of learning from my own faults..:)

No, this time that error I did not commit - just controlled again, the pics are in the right order. The 27ohm do produce that waveform..

Finally, the cmos osc. output, with 33 ohm termination, with the previous parameters:
OSC [HC] - 33ohm -rg174 5nsec - 1Mohm scope end

Also here, I'm not proud, it should be better done - there are two pigtails included here, no ground plane etc. But I would say it's not much off from the right value [that is, ~17 ohm ? OSC Zo]

I think I am the one at fault. The method I have been using (and suggested to Hifi) seems to be incorrect. The nice thing about your measurements is that with the very fast 'scope you used, it clearly shows the stair-stepping that each round-trip reflection creates.



That still is puzzling to me. If you look at the stair steps on the positive-going edge of the 50 ohm measurement, you can see that each stair step cuts the error almost exactly in half -- the first step reaches 1/2, the second 3/4, the third 7/8, and so forth. I would think that this means the impedance mismatch is 2:1, so the total output impedance would be 100 ohms, with 50 ohms due to Zo of the comparator (at least for positive-going edges).


But the measurement with the 27 ohm resistor seems to me to show a *greater* mismatch, as each step reaches *less* than half the remaining value. Any ideas, anybody???



This is just what Jocko said. So Jocko, is this true for all logic families? Even the latest ultra-fast ones with propagation delays of only 1 nS?

Thanks again, George and Jocko.

I have not tested the "ultra-fast" ones, as they have no place in anything we are doing. I can't recall all the values, but I seem to recall that VHC is a bit higher than HC.

If you want a particular gate/family measured, let me know.

BTW......Charlie............you "top secret" thingie is still on my bench. Too much crud stacked up this week.

Jocko

Hi all,

I just wanted to show some additional results:
This little puzzle with the AD8561 didn't leave me to rest. Though Charles has provided us with a much valid analysis based on what we could see, I wished to get some cross -check. Though I don't have the TDR of Jocko, I have an additional reflection /transmission test set available with the analyzer.
With this set it is possible to make calibrated one port measurements, just like the way it is done in the N2PK project: that is, measuring the DUT relative impedance referred to a reference load, after an Open/Short/Load calibration process.
It looks like this:

So, with this jig the return loss curve [as a function of freq.] of that previous CMOS 10MHz oscillator with the 33 ohm back terminating resistor [and with pigtails.. ] looks like this:

When I switch to the calculated impedance [as referred to the 50 ohm ref. load], it gives 47,5 ohm up to that zero visible on the previous curve, then starts to rise [seriously only after ~ 20 MHz] up to 72 ohm. But that rise is due to the pigtails, not the output impedance. So I would say it's 47.5 ohm-33ohm=14,5 ohm for this CMOS oscillator.

For a reference, in this same position that 12dB barrel attenuator applied earlier on gave 48,7 ohm value. The ref load put back later gave 50,1 ohm full freq. range [flat]

Now, with the AD8561 there is that problem of the assymmetrical ouput impedance. I could have had an exact value by, for example, stopping the oscillator and measuring the impedance in both high - low output cases.
But I was lazy.. And was also looking for a more dynamic value. So I had just adjusted the duty cycle once down to ~15%, then up to 75 %. In these two cases I have seen both time a "capacitive" Zo behaviour, Zo continously decreasing with frequency, in the 15 % duty cycle value going from ~350 Ohm down to 33 ohm;;
And in he case of the 75% duty cycle going from ~1,8kOhm down to 33 ohm at 200MHz.
So, Charles is right - we just continously have a too high output impedance with this chip. The actual waveform will always depend on the setup: for example, with that 5nSec cable I got the curves shown here;
with a 1nsec cable I got very nice, though assymmetric waveform, without reflections - and this is the explanation why I had been seeing an acceptable waveform at the very beginnings with 50 ohm back termination.

[Unfortunately, no pics here - the analyzer just shows a flat line, the values should be red from the marker at each freq. position]

Ciao, George