EEVblog Electronics Community Forum
Products => Test Equipment => Topic started by: stephanm on June 24, 2022, 11:11:32 am
-
Disclaimer: I neither own a HP3314A nor have I ever seen one live. It's a function generator from the 1980's, capable of creating sine, square and triangle output up to 20MHz.
I spent quite some time studying and understanding the triangle generator circuitry of the HP3314A schematics. Over time, I understood more and more of the issues related to generating fast and clean triangle signals, and what HP did to overcome these issues. It's really an amazing piece of engineering artwork that they did more than 40 years ago!
However, there is one thing that keeps puzzling me. For generating the triangle signal, HP basically implemented an integrator where a timing capacitor is alternatively charged by a current +I (for the rising ramp) and -I (for the falling ramp). Frequency is controlled by varying I. The actual circuitry is much more complex, the integrator and much of the other circuitry is working with differential signals, but for the question I have this complexity is not too important I think.
For different frequency ranges, the service manual says that the different timing capacitors (C218, C204...C207, C210) which integrate the current are paralleled by the relais K201...K205. So for the 2-20MHz range, only timing capacitor C218 is used, but for e.g. the 200kHz-2MHz range, relais K205 is closed so that C218 + C210 are paralleled. C218 together with some other capacitors and parasitics is supposed to give 27.7pF, and as C210 = 250pF we get C218 + C210 = 277.7pF, which is exactly a 1:10 ratio, resulting in a 1:10 frequency ratio between the two frequency ranges of the instrument. For the lower frequency ranges, it goes on like this, the next capacitor paralleled into the integrator timing capacitance is C207 = 2500pF, so that we get C218 + C210 + C207 = 2777.7pF, again a 1:10 ratio compared to the previous frequency range.
Now, with those parallel capacitors, the whole arrangement becomes susceptible to ringing. The smaller capacitors will produce a high frequency resonance with the parasitic inductances of the larger capacitances. Given the photos found in https://www.eevblog.com/forum/testgear/hp-3314a-function-generator-teardown-explanation/ (https://www.eevblog.com/forum/testgear/hp-3314a-function-generator-teardown-explanation/) the physical arrangement of the capacitors and relais is somewhat large, so I expect relevant parasitic inductances to be found in this part of the circuit. Since the integrator periodically switches the current through the capacitances from +I to -I and back to +I with a quite impressive slew rate, it is very easy to see this ringing happen it a LTspice simulation with only a few 10s of Nanohenries of parasitics.
If this wouldn't be a triangle generator, a set of countermeasures would be available for damping these resonances, but I don't see any of this implemented in the schematics (which makes sense to me, because all countermeasures I can think of would distort the triangle signal even more severely compared to the ringing.)
With the parasitics I believe to be reasonable, I would estimate ringing to be in the area just above 100MHz. This should be within the bandwidth of the rest of the circuit (you want high bandwidth for a 20MHz triangle signal, triangle signals have lots of non-negligible harmonics.) Ringing amplitude I get in simulations is small but I somehow cannot imagine that HP engineers just accepted this to happen and found on the instrument's output.
What I am curious about is: Does anyone of you have a measurement that shows the triangle signal tips? Do we see high frequency oscillations there on the output? How large is it? Or, if there is no such a thing: How did HP engineers solve this issue, what part of the circuitry is the solution? This keeps bugging me for a really long time already...
-
Interesting question, don't have this generator so hopefully someone can show these waveforms.
Regarding the simulations and actual circuits, suspect the PCB layout and parasitic capacitance, in addition to the parasitic inductance will have a significant effect at the triangle waveform peak where the current direction thru the timing caps is reversed. What speed do your simulations show for the current switching reversal?
Best,
-
What speed do your simulations show for the current switching reversal?
It's around 2ns, but this might be too optimistic. I only modelled parasitics in the timing capacitor area of the differential integrator so far.
As I was actually thinking of (re-)building this circuit with components available today, my simulation just uses many MMBTH10's in places where HP engineers might have spent hours and days to figure out what's the best transistor for the spot. I have also captured the HP3314A differential comparator in my simulation based on NPN+PNP transistor arrays available today. Top speed that I was looking for is 10MHz plus 10% guard band (half of what the HP3314A can do) in the hope that this could ease up some things at least bit. So my timing capacitors are larger, I am running slightly more current through them (the max integrator current in the HP3314A is 2.2mA, I am using 2.5mA), and I increased the triangle p-p voltage a bit (2Vpp differentially for the HP, 2.5Vpp for my attempt.) Oh, and I reduced supply from +/-15V to +/-9V, so my DC level voltage layout is sometimes quite different from where HP has the quiescence voltages throughout the circuit.
Still, I did a few checks with adding parasitics in places where I thought it would matter, and I found many things to be surprisingly immune to my attempts to uncover more areas of concern. But that could just mean that I haven't looked deep enough into this. The only notable exception is the differential integrator.
-
What I am curious about is: Does anyone of you have a measurement that shows the triangle signal tips? Do we see high frequency oscillations there on the output? How large is it? Or, if there is no such a thing: How did HP engineers solve this issue, what part of the circuitry is the solution? This keeps bugging me for a really long time already...
Hi,
below are two images of triangle waveform from my 3314A connected to a 2465B analog scope, output frequency is 10 MHz, in one image delayed trace to zoom details at 'I' change.
[attachimg=2][attachimg=1]
Best, Stephano
-
Hi Stefano,
thanks a lot! The triangle peaks are really remarkably sharp and clean, it's amazing what the HP3314A does. It looks a little as if there would be bumps on the rising and falling slopes, but maybe I'm seeing it wrong.
At 10 MHz timing capacitors are not paralleled. Do you, by accident, have a scope screenshot for e.g. 1MHz or 100kHz? At 1MHz, two timing caps are paralleled, and 3 are paralleled for 100kHz, which gives us the construction that is susceptible to ringing...
Kind regards
Stephan.
-
As an additional data point, my HP 3314A also exhibits no ringing on a (10 MHz) triangle wave.
I have certainly seen many instances of cheap function generators where indeed there is some artifact at the peaks of the triangle waveform where the current in the integrator switches. I don't recall much ringing, although I probably didn't zoom in that much. It basically looked like a small voltage spike. And, of course, with triangle waveforms, one should be concerned with linearity - a place where many vendors fall short.
A minor annoyance that I've found over the years is when you look at a vintage piece, for example from the 1970's, on a modern oscilloscope. The high bandwidth of the scope will show you all kinds of high frequency junk that you wouldn't have seen 50 years ago on a 50-year old scope. And turn down the voltage on one of these function generators and your modern scope may have triggering problems since it is seeing the "noise." It is almost like the designers in that era produced equipment that met specs in the prescribed frequency range and didn't much care if the piece was generating junk tens or hundreds of MHz higher up since few people had the instrumentation to see it. Hopefully I'm wrong about that!
-
I have certainly seen many instances of cheap function generators where indeed there is some artifact at the peaks of the triangle waveform where the current in the integrator switches. I don't recall much ringing, although I probably didn't zoom in that much. It basically looked like a small voltage spike. And, of course, with triangle waveforms, one should be concerned with linearity - a place where many vendors fall short.
Messing up the triangle peaks is indeed easy in a function generator circuit that is based on an integrator. I have seen it both in my simulations and on the breadboard circuits that I have built. You want the peaks to be sharp, this means your current switches must be fast. But then, high speed switching signals can couple into into other parts of the circuit, or the integrator capacitor construction could start ringing. All this introduces artifacts on the triangle peaks. Also, you want to carefully control the amplitude of the peaks to make the instrument produce a signal whose voltage is stable over frequency. And, on top of all this: A common way to generate the sine wave output was to take the triangle signal and run it through a sine shaper circuit. There's a nice patent on such a circuit filed by an HP engineer from Böblingen in 1980, my assumption is that this is what HP actually used in the HP3314A's hybrid sine shaper and amplitude control IC. But... with this circuit, distortion and artifacts of the triangle peaks will lead to distorted peaks of the generated sine wave. So for a function generator like the HP3314A, there was lots of reason for HP to put quite some energy and focus in the triangle wave generator.
Still, it remains unclear to me how they managed to get away without ringing in the integrator when the capacitors are paralleled. In my simulations and breadboard work, I constantly see that it's easier to get away without ringing in the highest frequency range, where only one integrator capacitor is used, not multiple in parallel. That said, waveform photos (or a spectrum analyzer measurement) from the triangle wave below 2MHz would be great.
What I also found is that you can avoid ringing by slowing down the current transition in the integrator. The larger the integration capacitor becomes, the more you need to slow down the transition. The comparator HP implemented in the HP3314A has this property, but, so to speak, only for one half of the transition. The comparator has low gain, it will react as soon as the triangle voltage comes close to the peak, and the slower the triangle voltage approaches it's peak, the earlier (and slower) the comparator will start moving the voltages that control the differential pair current switch (Q208A/B in the circuit diagram I posted.) However, the comparator also requires to provide a bistable behaviour; this is implemented by positive feedback in the comparator. Once the positive feedback kicks in, the comparator will generate steeper slopes for the current switch control voltages. And now I'm back at the place where things start ringing in my simulations ;D
-
Do you, by accident, have a scope screenshot for e.g. 1MHz or 100kHz? At 1MHz, two timing caps are paralleled, and 3 are paralleled for 100kHz, which gives us the construction that is susceptible to ringing...
Kind regards
Stephan.
Hello Stephan,
new images for 100KHz and 1MHz output, no ringing visible, the scope is at max BW (400MHz).
Best, Stephano
-
Thanks Stephano for providing these images. Very helpful and very interesting! Once again, waveforms look remarkably clean, no indication of artifacts at the peaks.
Congratulations to owning this instrument, it must be a pleasure to work with it :)
-
Congratulations to owning this instrument, it must be a pleasure to work with it :)
Not so much. It is big & loud and prone to failures due to age. There are better tools out there nowadays; I have sold my 3314A a long time ago (after replacing all the reed relays).