low noise 5v symmetric power supply

[BrunoGelli]

Hello, fellow geeks! This is my first post here, although I'm not new in this forum.

I'm designing a low noise +5v and -5v power supply to power some op-amps. This op-amps are placed in a SiPM biasing board, which is being used to count photons in my physics lab.

As the application is really sensitive, the noise level has to be really small! Specially in the frequency range the signal is expected (a few MHz).

The schematics are attached, as well as a board layout. The input for this circuit is a center-tapped 15v transformer directly powered from the mains voltage. C5, C6, C7 and C8 are tantalum caps for low ESR (My major insecurity in this design, perhaps ceramic would be better?) C1, C2, C3 and C4 are common electrolytic caps.

I would love to hear how I can improve this design! Feel free to comment.

Thanks a lot! This will be part of my Ph.D. theses and any help will be in the acknowledgement section 

Bruno Gelli

[Kleinstein]

It is  a good idea to start with a full wave rectifier / bridge to reduce the ripple.
The 7805 regulators don't need so much capacitance at the ouput side, so no real need such a large electrolytic cap. The 7905 needs some care with the ESR, a bit like a low drop regulator: not too much low ESR capacitance, but more like 100 nF ceramic + some electrolytic with the right range óf ESR.

How much fitler capüacitance is needed on the input side is depends on the current drawn. With a full wave rectifier the 470 µF are good for some 200 mA, maybe a littel more with a 15 V tranformer which is bit on the high side. Normally 7805 would be used with about 10 V AC.
The resistors in series are good, especially with the high voltage.

OPs are usually not that sensitive to noise. The really critical part should have local filtering / decoupling after the regulator. So the demand on the regulator / power supply is no that high. Photon counting is more about the detector, not much about the supply for the OPs. The supply to the detector can be quite important.

[David Hess]

The low ESR tantalums are not going to do anything useful.  Do not add or use ceramic capacitors except on the inputs if the bulk input capacitor is located remotely.

Well designed circuit should be very tolerant of power supply noise but there are some things you can do.  Replacing the 7805/7905 with the adjustable 317/337 will result in lower noise when the adjustment pins are bypassed to ground.  A further improvement can be made by replacing the 317/337 with lower noise alternatives.

A step beyond the above is to use a low noise reference to control the regulator through an operational amplifier.  This can be extended to making a tracking regulator so that noise on the positive and negative rails is correlated.

[jonpaul]

Bonjour:

Use series beads or small u H L at the opamps board, with 100 nF mon bypass caps.

See Jim Williams and other app notes re low noise regulators.

No need for tantalum or paired electrolytic.

Very often an electrolytic is enough or with a 100 nF mon in parallel.

Find the Op-Amp data sheet for the parts being used, and get the PSRR = power supply rejection ratio curve vs freq.

Enjoy,

Jon

[Ian.M]

You've got plenty of headroom, so a pair of capacitance multipliers between the bulk capacitors C1,C2 and the respective regulator inputs can easily reduce ripple at the inputs by at least an order of magnitude with a corresponding improvement in PSRR.  See: https://www.eevblog.com/forum/blog/eevblog-1116-the-capacitance-multiplier/

Edit: Also a capacitance multiplier has *MUCH* better low HF band noise rejection than most linear regulators, only limited by the impedance vs frequency characteristic of its base capacitor and the cutoff frequency of the transistor.

[gbaddeley]

C5-C8 can be much lower value, and can use ordinary 100nF. You dont need big 470uF elec caps after the regulators. If you want low noise in MHz region, add an LC filter after the regs with ceramic caps.

[Ian.M]

Also, see https://www.eevblog.com/forum/projects/film-cap-across-bridge-rectifier-in-xxi-century/ for alleviation of a possible issue of broadband impulse noise due to snap recovery of the rectifier diodes.

[BrunoGelli]

Thanks a lot! I know the component choices are not optimal, but is the best we can have of the shelf here in Brazil. Otherwise it has to be imported, which is expensive and takes a long time.

The OPs are somewhat resistant to noise, but the board manufacturer made some weird choices with the tracing. The traces for the 5v and the -5v rails are right beside the traces for the shaped signal. The power supply noise in this rails is easily seen in the output, and can even overpower the signal from the photons.

I hope this little power supply can help, otherwise we will be back to photomultiplier tubes town (They are bulky, but they work amazingly!)

[David Hess]

C5-C8 can be much lower value, and can use ordinary 100nF. You dont need big 470uF elec caps after the regulators. If you want low noise in MHz region, add an LC filter after the regs with ceramic caps.

You sure do not want to do that with a 337 which because of its common emitter output has special requirements for its output capacitor like a modern low dropout regulator.  The 317 may not be happy with it either.

[Kleinstein]

AFAIK the 7905 is not that different from the LM337: it also is a bit picky about the output capacitor and usually wants some electrolytic for the ESR / damping.
I don't see an advantag in LM317 over 7805  - the noise level depends on the 7805 manufacturer. If just 5 V the 7805 is simpler.

[BrunoGelli]

I see! Thanks a lot, these tips are really useful! I guess version 2.1 is coming soon

I would like to add another question to this trend, if you gentleman are still interested in this setup.

After the supply voltages are generated, they have to be connected into a darkbox (completely light tight metal box) about 2 meters away.

I'm used to use coaxial cables, either with BNC or LEMO 00 connectors. But in this case I'm a little afraid about the ground loop generated for using 2 coaxial cables from the PS to the darkbox (+5v and ground in the first one and -5v and ground in the second one). I was thinking that a DB9 connector with a shielded cable could be a better option, since this connector is light tight and I would keep some spare feedthrough pins for future use.

Perhaps a twisted pair would work as well and I'm overthinking the problem?

[Kleinstein]

I see no problem with a DB9 connector and twisted pair cables - though the +-5 V would need 3 wired, so likely +5V/GND and GND/-5 V as 2 twisted pairs.
A ground loop in the supply should not be that bad, unless the area is extra large.

I would not be so sure the DB9 connectors are all 100% light tight. It can help to still have a dark cover - the simple grey plastic cases are likely still a little transparent.

[Doctorandus_P]

There has been quite a lot of going on "super regulators" for audio applications.
You can use a general search for that:
https://html.duckduckgo.com/html?q=%22audio%22+%22super+regulator%22

Or look directly on https://www.diyaudio.com/ which has loads of audio specific stuff.

[gdewitte]

https://linearaudio.nl/silentswitcher?

[Manul]

These are really low noise linear regulators: LT3042, LT3045 (positive) and LT3093, LT3094 (negative) and with good PSRR. Used them myself. I assume that you are not trying to make your circuit the cheapest possible, so why not, even if you don't exactly need them. Still, good circuit design of your amplifier, low value resistors, low noise op amps, maybe discrete front end, layout, shielding, ferrite beads, etc. can make a huge difference. It is unlikely that power supply noise will be limitting factor, until you achieve perfection everywhere else, or your amplifier has really bad PSRR.

[David Hess]

I don't see an advantag in LM317 over 7805  - the noise level depends on the 7805 manufacturer. If just 5 V the 7805 is simpler.

The advantage is that bypassing the adjustment pin to ground lowers the high frequency output noise above the breakpoint because it lowers the noise gain from the feedback divider.  Some fixed regulators support this with an extra pin to the internal divider but obviously not the 3-pin TO-220 ones.

This is also why the "high stability" circuit variation using the 317 use a low noise reference between ground and adjust without the feedback divider.  The same can be done with the 7805 at 5 volts only by enclosing it within some other feedback loop, which I have done to good effect but there is little reason not to use a 317 the same way and it is usually more convenient. (1)

The above of course works just as well with the 337 giving it the same noise advantage over the 7905.

(1) I have also done it with the 7805 for a 10 volt output but here, the 5 volts at the common pin and high common pin current are actually an advantage for circuit biasing.