I have the following circuit which generates digital noise... circuit

The reason for the 2 diodes has to do with stochastic uniformity of the noise output and I don't really want address this aspect. The comparator is like a LM311 with a <500ns propogation delay. Some other supporting bits have been ommited for clarity. The noise coming off the Zeners is around 1Vp-p when measured 10,000 times.

I intend to have a 8 sets of this circuit so I am interested in reducing the component count and not having a huge circuit board. I have bread boarded this, and it seems to work. I have measured the noise signal running at up to 8MHz bandwidth. In reality, I would be expecting to read the noise signal at around 2Mhz.

My question is, should I have (op-amp) buffers on the inputs to the comparator given the presence of the 200K resistors? What would best practice recommend?

  • 1
    \$\begingroup\$ If the comparator can pick up the variations in voltage, and the output is within the voltage range you expect, I see no need to have an op-amp or buffer involved. I have worked on similar random noise generators that used op-amps to amplify the noise to usable levels, nothing more. If the output is the level you want, then you need nothing more. \$\endgroup\$ – Puffafish Dec 7 '16 at 14:16
  • \$\begingroup\$ I don't see why, if you've already established you can read to 8MHz, and only need 2MHz. \$\endgroup\$ – Neil_UK Dec 7 '16 at 14:17
  • \$\begingroup\$ @Neil_UK I'm reading it with an Arduino as I don't have access to an oscilloscope of sufficient bandwidth. So I can only see the signal indirectly, and I'm concerned that it might just be comparator oscillations due to high source impedance. Isn't 200K a bit high for a comparator? \$\endgroup\$ – Paul Uszak Dec 7 '16 at 14:37
  • 3
    \$\begingroup\$ I will be surprised if this circuit will work reliably across component variations, there will be a bias to the output. The tolerance of the zener diodes could easily cause the comparator to be stuck in one state or the other - what if one diode is 24V and the other 24.1V. Normally this sort of circuit uses AC coupling with a single diode. \$\endgroup\$ – Kevin White Dec 7 '16 at 14:58
  • \$\begingroup\$ @KevinWhite You can hand match the diodes to well <0.1V, and that variation gets swamped by the 1V noise anyway. The reason I'm doing it this way is exactly because this isn't the typical approach... \$\endgroup\$ – Paul Uszak Dec 7 '16 at 16:00

Typical noise from a 24V zener at optimum current (much higher than you are running them at) is less than 200mV RMS (maybe 1 or 1.5 volts p-p).

A 24V zener might have 5% tolerance, meaning that your two zeners may be mismatched by +/-2.4V, meaning no signal at all.

You could consider replacing one of the zeners with an RC low-pass filter connected to the remaining zener so that it settles near the average voltage.

If you're depending on this being truly random you will want to make sure the PSD is fairly flat.

  • \$\begingroup\$ But do I need buffers as I asked? \$\endgroup\$ – Paul Uszak Dec 7 '16 at 16:31
  • \$\begingroup\$ Unity gain buffers won't be of much help (or harm). An LM311 has input bias currents of less than 150nA and those currents will just subtract a bit from your zener current. If the buffer had a different bias current it would behave differently, but hardly likely to be measurable. Buffer frequency response might reduce the frequency of the noise, depending on the buffer again, but it's not going to increase the frequency with gain =1. \$\endgroup\$ – Spehro Pefhany Dec 7 '16 at 16:49
  • \$\begingroup\$ Just reviewing this. Do I have sufficient DC bias return paths considering the placement of the 2 diodes and no connection to ground? \$\endgroup\$ – Paul Uszak Apr 1 '17 at 0:15

The reason for the 2 diodes has to do with statocastic uniformity of the noise output and I don't really want address this aspect.

Well, but they impact the discussion severely, so they need to be discussed. The simple answer is that you do need a "buffer" circuit (actually, a level shifter) on one of the inputs. As has been mentioned, for ideal results the two zeners must be perfectly matched. I assume you have done such matching for your test circuit, or else you just got lucky using two zeners which came from the same wafer. Matching 8 pairs is going to take a good deal more work, not to mention a certain level of wastage due to discarding the ones you can't match. Plus, of course, there is no guarantee that the two will track with temperature.

More importantly, you seem to want very good noise statistics, what you seem to call stochastic uniformity. I assume you want uniform distribution (equal numbers of 1's and 0's). You need to be aware that any difference in mean zener levels will degrade this, and waving your hands and invoking 0.1 volt matching will do nothing to change the matter. Of course, you can always compensate by taking your samples in pairs and looking for bit differences, but in that case why worry about uniformity?

I also suggest you revisit the data sheet and look at figures 3 and 4 on page 7. Response times for a 311 are typically under 200 nsec.

  • \$\begingroup\$ Why are you only suggesting 1 buffer rather than 2? Both inputs to the comparator have to deal with the same source impedance. \$\endgroup\$ – Paul Uszak Dec 7 '16 at 22:19
  • \$\begingroup\$ @PaulUszak - You've lost me. Why is source impedance important in this case? \$\endgroup\$ – WhatRoughBeast Dec 7 '16 at 23:32
  • \$\begingroup\$ Does the operation of the comparator not depend (at least somewhat) on the source impedance? If not, then great I'll drop it from the circuit as component count was part of the original question... \$\endgroup\$ – Paul Uszak Dec 8 '16 at 2:26
  • \$\begingroup\$ LM311 offset current is 70 nA max, which through 200k gives a maximum offset of 14 mV. This is a good deal less than your hypothetical matching. But you're still ignoring the effect of mismatched levels on your noise statistics. Or have you stopped caring about it? \$\endgroup\$ – WhatRoughBeast Dec 8 '16 at 4:44
  • \$\begingroup\$ I think that this may be an example of the reverse XY problem. You seem to be determined to answer what you think the OP should have asked, rather that what he did ask :-) Thanks for the nano amperage though. That's really useful. \$\endgroup\$ – Paul Uszak Dec 8 '16 at 11:08

Check out the Miller Input Capacity of your comparator pins, right around 0V differential input. Some comparators have cascaded inputs (AM685) as do some opamps (UA715). The LM111 NPN diff pair has no cascodes, thus there is lots of Cmiller, but that CMiller is driven from PNP emitter followers with tiny pullup currents. The Cmiller may be enough to cause a "hesitation", a halt to slewing, and a disruption of the needed independence of samples.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.