Measuring noise in high voltage supplies

Measuring noise in high voltage supplies
Posted by Atomillo on 09 Jul, 2021 15:38
Hello:

I'm starting to work on the design of various high voltage bias supplies for things like gaseous radiation detectors and I'm running into the issue of noise.

Of course, the best resource I could find on the topic is an App Note by Jim Williams, more specifically "Application Note 118: High Voltage, Low Noise, DC/DC Converters". There it is mentioned that a protection network as well as 40dB preamp has to be used before seeing the signal on the oscilloscope. I specially need this preamp, since my scope bandwith limits itself in the two lowest V/div and I want a 50Mhz bandwitdh noise measurement.

However, the HP-461A used by Jim goes for quite a bit on eBay. Is it easy to make with modern components an inexpensive 50Mhz amp with relatively low noise (also, I would like to use THT components)?
A superficial look at Mouser leaded to the LT1222, specifically designed to be stable at a gain of 10 with a BW of 50Mhz (from a GBW of 500Mhz).

Thus, could I cascade two non inverting amplifier stages like those shown in the Fig 1 of the Datasheet (without the clamping network)? Or are there any "traps for young players" (like myself) I'm missing?

I appreciate any help!

FAIL: I just came across EEVblog #572 Cascading Opamps For Increased Bandwidth and saw that the actual BW of the system would be 32Mhz! Thus, it seems like I need an even faster op amp. Any suggestions?

#1 Reply
Posted by TimFox on 09 Jul, 2021 15:47
The -hp- 461A is very useful, and it is a shame that it is no longer made.
(I own two, plus the higher-frequency units that have 50 ohm inputs.)
You might want to clone it from the original diagram:
http://bee.mif.pg.gda.pl/ciasteczkowypotwor/ESI/461A.pdf

#2 Reply
Posted by Atomillo on 09 Jul, 2021 15:59
I see that it is fully discrete and that each stage was adjusted individually.
Nowadays can't this be done with op amps? Or am I perhaps too naive about the virtues of op amps.

#3 Reply
Posted by TimFox on 09 Jul, 2021 16:13
Modern op amps can be used a relatively high frequencies, but you need to take much more care on parasitic elements (such as shunt capacitance of components and to ground) than with op amp circuits at audio frequencies.
The fundamental parameter for a high-frequency op amp is the GBW. To oversimplify, the gain you can get at a given frequency is k_V = (GBW)/f. To get x10 (20 dB) voltage gain at 100 MHz therefore requires at least 1000 MHz = 1 GHz GBW. If you cascade stages, then the overall bandwidth adds in quadrature:

(BW)² = 1/(1/BW₁² + 1/BW₂² + ...)

#4 Reply
Posted by Atomillo on 09 Jul, 2021 16:28
Thanks for the kind reply.

Regarding parasitics, in the various datasheets I've read they mentioned that the various resistors should be of low value, so that parasitic capacitances don't lower the bandwidth. Are there any important tips regarding parasitics I should know?

Yes, I've just finished watching Dave's video. Assuming I only 2 stages with G=10 for each, that means I need an op amp with a GBP of 600MHz, ideally in dual package (¿correct?)

Is anyone aware of any such op amps?

#5 Reply
Posted by TimFox on 09 Jul, 2021 17:34
A given parasitic capacitance will have less effect on a low-value resistor, by simple circuit theory.
The construction of a bulk-foil resistor has less parasitic capacitance than some other constructions.
See page 8 of http://www.vishaypg.com/docs/49788/10reasns.pdf

#6 Reply
Posted by Atomillo on 09 Jul, 2021 17:49
Tim:

Thanks for the document. I will print it out Monday and I will read it online in the mean time.

I must say, I restricted myself A LOT by only choosing DIP... A quick search reveals easy to solder SO-8 packages with much much better specs.

However since to a first prototype I plan to use a tht board instead of a pcb I wonder how much the parasitics of a SO-8 to DIP adapter would hurt me

#7 Reply
Posted by Kleinstein on 09 Jul, 2021 18:16
For testing a high voltage supply (as opposed to a relatively low voltage reference) should not really need much amplification. The noise level is expected to scale at least like the voltage. So the main point is more some protection that does not effect the BW very much and avoids excessive reduction in the amplitude.

I would consider AC coupling followed by maybe a divided by 2 divider also used for the protection. Amplification (a buffer, e.g. source follower) may be useful for bootstrapping the clamping part. The higher frequency part may need lower impedance to reduce the effect of parasitic capacitance. The low frequency part may need higher resistors to get away without very large capacitors.
So it may be easier to divide the frequency range and use a separate circuits for the higher frequency (e.g. > 1 kHz) and lower frequency (e.g. < 10 kHz) part. The high frequency part may not need much gain.

The really fast OPs are rarely available in DIP - more like SO8 and smaller. The adapter would add capacitance and inductance in the path for the decoupling cap at the supply. One can often add some supply decouling to the adatper boad and this way at least reduce the 2nd problem.
An alternative to a classical adater board could be a small full copper board and than use dead bug / manhaten style for the really fast part.

#8 Reply
Posted by TimFox on 09 Jul, 2021 18:26
For a single unit, I would second Kleinstein's suggestion for "dead bug" construction. Unfortunately, really good modern op-amps are rarely available in DIP anymore.

#9 Reply
Posted by Atomillo on 09 Jul, 2021 18:32
Kleinstein:

I need the amplification because the noise of HV power supplies intended to power radiation detectors must be quite small: down in single milivolts or microvolts.
In a lot of Jim Williams design in the app note I mentioned earlier he gets to a noise level of 100uV for example.

Regarding dead bug construction, the problem are my manual skills. So right now I must choose whether to bite the bullet learn how use KiCAD and pay for a PCB or face my fears and learn dead bug construction. One issue I see is that SMD leads are quite short. Do they still lean themselves to easy soldering?

Thanks all!

#10 Reply
Posted by Kleinstein on 10 Jul, 2021 07:54
Dead bug with SO8 parts is a bit tricky and needs some practice. So maybe try it out with some old parts, like LM358 or even just solder practice with random non working (e.g. removed for repair or from scrap) parts. With only a few used pins and a fine tip soldering iron it should be doable. This is one of the few jobs that actially needs a rather fine tip. For most part the OPs are quite robust and take quite some heat. I would still used leaded solder - so the temperature can be a bit lower.

A good scope input should be at some 10-20 nV/sqrt(Hz). A spearate DIY amplifier would hardly be 10 times better, even with low impedance - so there is no real need for more than a 10 fold gain in the high frequency range. Even just 3 fold gain may be enough. The point is more like getting the protection and avoiding extra noise from a 10:1 probe.

The low frequency part may use more gain, as the scopes noise is higher there, but this can use DIP parts and SO-DIP adapters. The low frequency noise of the high voltage supply would also be relatively high (unless one uses the crayzy stack of batteries) - voltage references have there own noise and even something like 20 x LM399 or Ref102 in series would be relatively high noise compared to the scope input.

#11 Reply
Posted by Atomillo on 11 Jul, 2021 09:17
I ordered some copper boards in Amazon which should arrive by Monday.

I will follow your advise and waste a couple of boards just practicing!

#12 Reply
Posted by magic on 11 Jul, 2021 15:41
What's the minimum frequency you care about? There are some interesting problems with AC coupling and current noise of amplifiers.
What about gain accuracy? An opamp at 10x closed loop gain is -3dB or 30% off at GBW/10.

#13 Reply
Posted by Atomillo on 12 Jul, 2021 18:27
Low frequency noise isn't really any concern.
I expected a lowest frequency in the 10s of KHz, that way I also eliminate the ugly low frequency noise of op amps as well as the enC noise.
And Gain accuracy is also relatively unimportant. As long as it stays constant during a measurement of course!

#14 Reply
Posted by TimFox on 12 Jul, 2021 20:22
The basic problem with AC-coupled inputs and noise is that at low frequencies, even though the high-pass filter of the capacitor and input resistance is at a lower frequency, the op amp's input current noise generator flowing through the high reactance of the capacitor (which is not itself a noise source, absent leakage) can be larger than the input voltage noise generator of the amplifier. For example, the standard input impedance of an RIAA phono stage is 47 k $\$\Omega\$$ , but a 100 nF capacitor at 50 Hz is -j64 k $\$\Omega\$$ .

#15 Reply
Posted by perieanuo on 13 Jul, 2021 07:31
i'm no expert, but i'd bet the load physics need to be veeeeery well known. what can the load 'take' as noise from your ps?
after this first step, you buy the tools to measure the noise. can't skip this step

#16 Reply
Posted by Kleinstein on 13 Jul, 2021 11:36
An example for a sensitive detector is a PMT used for gamma spectroskopy, so in the linear mode with need for a stable gain. Some 10 V change in the supply can double the gain and one may want the gain to be stable to some 0.1 % if the demand is high. So this would be a required stability of some 10 mV (at some 700-1000 V). The upper BW limit would be set by the peak width (before pulse forming), so maybe some 10 MHz BW.