dc-dc converter supply opamp ultra low noise amplifier

[Yrrah]

Hello,

I am working on a phase noise measuring system (UKW Berichte 4/2015...). It deploys an ultra low noise amplifier (10 Hz - 90 kHz, measuring range, the amp itself goes to about 500 kHz and beyond). I will use a 12 V lead battery to supply the instrument. I need +/- 8V to supply low noise opamps (e.g. LM4562), obviously of high quality (low noise, no spikes, reasonably stable). I plan to use the well known 723 for that, because of low noise. To generate the negative bias a dc-dc converter is advised, I bought a Mean Well SMU02M-12 for the negative bias. Switching freq over 100 kHz so above the highest measuring frequency of 90 kHz. That will be followed by the 723 neg reg. I have a testboard on the bench. Battery, 47 uF at the input, DC-DC converter; double PI filter at the output: 47 uF,  in each + and - lead: 1 mH and 100 uF. Load is 120 ohm resistor. Scope (Tek2235) at battery: every 8usec (about 120 kHz) a big spike of about 800 mV tt.  Scope at resistor is almost the same. No 723 involved yet.  I am unsure whether this is pure radiation, do I need more filtering? As even the the battery terminals are polluted I am a bit unsure how to move on. I plan to place the dc-dc converter with 723 in one diecast box (12V in, -8V out).
Any advice welcome as are pointers to relevant postings or literature.
Thanks in advance,
Yrrah

[ogden]

LM723 is old, PSRR specified only up-to 10KHz. You may consider LT3042 and LT3093 instead.
This article as well: http://www.wenzel.com/documents/finesse.html

[Marco]

That's a bit of a ramble.

If certainty of low noise is important enough to start using batteries and linear regulators for the positive rails I don't see how it makes sense to suddenly start using a switcher for the negative side. It's not that you can't get interference and ripple arbitrarily low ... but you've already decided to take the easy way out for the positive rail, why suddenly stop for the negative rail? Just use two batteries.

[ejeffrey]

A schematic would be much more helpful than listing the component values.  Even just a photo of napkin sketch.  A photo of the actual circuit would also be helpful as the layout is very important for things like this.

Make sure you are filtering the input as well as the output, and put the filters as close as possible to the pins of the converter.  Your filter components might be too big such that either you can't get them close enough or the parasitic values compromise their behavior.  For instance a bunch of small value capacitors in parallel will have less ESL and one of them can be placed right across the pins.

For such a small package radiated emissions should be negligible -- even at the 1000'th harmonic (100 MHz) and I doubt that converter has substantial noise power at 1 GHz.  However, you might get near field magnetic coupling, so magnetic shielding around it might help.

You also have to use good probing techniques.  You could be seeing common mode conversion by your probe or magnetic pickup from the ground loop if you are using the little alligator clip lead.

Alternately, get a second 12V battery for the negative supply and get rid of the switch mode converter altogether.

Or, for a low power application use a capacitive doubler based inverter to generate the negative supply.  This can be considerably easier to filter than a standard switching converter.

[schmitt trigger]

I ignore whether you can answer the following questions by looking at the Meanwell unit, but here they go:

Is it operating in DCM or CCM mode? DCM impose demanding switching characteristics that can cause noise spikes.
Is the main magnetic component a toroid? Or a shielded core like a PQ core? The reason is that the cheaper EE cores have significant stray fields.
If you can slow the main Mosfet's rise and fall times, noise can be reduced, at the expense of lowered efficiency.


Lastly, related to the measurement techniques, which you may or may not be aware, but I am going to ask anyways: Are you aware that with a scope probe, the long ground lead picks up a lot of noise? There is an EXCELLENT app note from Jim Williams, and I believe from Bob Pease too, which describes scope probing techniques for ultra-low noise measurements.

[ogden]

Alternately, get a second 12V battery for the negative supply and get rid of the switch mode converter altogether.

Right. Two batteries shall be primary objective and DC-DC converter shall be considered only when for some strange reasons you can't have two batteries. Why 12V lead? - Use two 9V alkaline batteries instead.

[Yrrah]

Thanks for your contributions.
I use a battery mainly to avoid any mains based psu problems. I will not incorporate a psu inside the instrument for the same reasons. At this point I am testing some sub systems. I know I can use two or more batteries, whatever I want. I made already a ulna (Wenzel based) with very good results. Battery fed. Refer to picture. I can measure the thermal noise of a 50 ohm resistor..... You may or may not like my construction technique..... A beer for anyone that locates the smd component correctly  In the phase noise measuring instrument a similar ulna will be used after a diode ring mixer.
With this lna I measured the noise from a 78xx voltage regulator: horrible. Inside a Faraday cage of course.
In the article I referred to, the author uses a dcdc converter to generate the negative voltage and I thought to give it a chance. It is very well possible that my scope connection is wrong. However I did not expect such high levels of stray signals. Even just the probe shorted with the earth clip around the small pcb gives 60-80 mV pulses. Screening is probably the answer but I am open for any suggestions, please? If nothing works out successfully I might end up with a battery for the negative voltage, and charge this one when the instrument is off (but connected to the pos battery of course). A picture of my setup is attached. The 120 ohm resistor simulates the expected load. At the battery I measure 200 mVtt, at the load: 900 mVtt. Can I get rid of this with better screening, more filtering or both? What is the best approach?
And yes the 723 is old. I used them back in the early seventies as a student in one of my first adjustable power supplies. But it seems it can have very low noise levels. I need to verify that. 
If anyone with experience with these dcdc converters has some useful tips, please let me know.
Well, so far for now, cheers,
Yrrah

[Yrrah]

Thanks Schmidt Trigger to point me to the excellent app note by Jim Williams. While reading I realized I have a Tek P6021 current probe. That shows a nice dc signal, almost without any rubbish. So the disputable use of the voltage probe was the problem. Current and voltage at a 120 ohm resistor should be quite similar at these low frequencies, of course. I am convinced now by using a proper metal case I can use the dc converter without problem.
Thanks again.
Kind regards,
Harke