I need to test a sensor, and I want to measure its resolution. My main concern regards keeping noise and disturbs (e.g., 50 Hz) as low as possible.

The circuit itself (amplifiers, filters, ...) is fully analog, and is designed to be low-noise. No problem here. The circuit is designed so that all its active components are supplied between ±5 V. The total current consumption is expected to be something around 200 mA.

Now the problem: which is the best way to provide the supply voltage to these active components (op-amps, VGAs, INAs, ...), knowing that I have a bench-top DC Power Supply and some batteries?


±5 V from DC Power Supply. Just 10 uF capacitors close to the connectors, 100 nF close to the devices (or proper value according to datasheets), and that's all. Is this enough to have a good supply voltage?


±10 V from DC Power Supply -> Resistive dividers to get ±5 V -> Passive low-pass filters (e.g., 1 MOhm, 1 uF --> around 150 mHz pole) -> Analog unity-gain buffers supplied between ±10 V from Power Supply. Is it better to use active filters? Or LC passive filters?


±10 V Battery -> Resistive dividers -> Analog unity-gain buffers supplied between ±10 V from DC Power Supply. In this way, the 200 mA will be supplied by the Power Supply, and the battery will last long (I can use high-value resistors for the divider so that its current consumpion is negligible). Is the PSRR of typical op-amps enough to have a good output signal?


±10 V Battery -> regulator. In this way, the 200 mA will be supplied by the regulator, i.e. by the battery, that will be discharged fast! Which regulator should I use? DC-DC switching? Linear? LDOs? LDOs with ±6 V batteries so that the power consumption / heating is minimized?


Say yours!

  • \$\begingroup\$ First you need to make up your mind about how much noise you are willing to accept. \$\endgroup\$
    – PlasmaHH
    Feb 6, 2017 at 12:54
  • \$\begingroup\$ Batteries can be very.low noise. Closed loop regulation inherently means delays and variation. So open loop fixes that. But adds its own issues. So I might use low noise resistor tweaked to provide correct voltage if load is constant. You do not want added shot noise, so no pn junctions, if possible. Really, you need to do a careful analysis. There is no Single panacea answer here. \$\endgroup\$
    – jonk
    Feb 6, 2017 at 13:06
  • 1
    \$\begingroup\$ What precisely does "low noise" mean to you? Even the presence of significant (say 100Hz) power supply noise does not necessarily invalidate a test provided you do a proper analysis (eg. power spectral density of the output). If this is just a test you can use batteries (eg. a couple of lead-acid SLA batteries) which would avoid noise possibly coupled through mains connections. There are a lot of considerations besides the raw supply noise measured at the PSU terminals. \$\endgroup\$ Feb 6, 2017 at 13:17

3 Answers 3


Using 3 cascaded C-FB-C filter circuits, the C-FB-C placed in ALL the VDDs and RTN, and with +15/-15/+6/-6/+5/RTN filtered at the switcher-supplies+DSP / centralClockGen / signalchain we used 6 * 3 = 18 of the C-FB-C circuits, I achieved 15 bit noise-floors at 6Million conversions per second. The floor was 150 microVolts. The sensor was a focal-plane IR sensor, itself isolated from ground with 100micron glass beads in the epoxy, thus 50 pF capacitance from the sensor to the chassis/airplane. In the sensor head, I used 10 ohm resistors into each opamp, plus 0.1uF and 10uF, to ensure all the fast signal-related current-surges were provided locally; thus none of the OpAmps were able to interact, via VDD, with any of the other OpAmps. In other words, VDDs went thru a low-pass-filter to enter any of the signal chain opamps. [concept: have 'local batteries']


simulate this circuit – Schematic created using CircuitLab

For ultimate noise/crosstalk_between_ICs, people use a VDD_Tree, to locally implement private_filtering of VDD for each IC. With the power supply rejection at DC often being 80 or 100 dB (thus 60Hz is rejected by 10,000 or 100,000) but becoming much less at high frequencies, you need to implement low-pass-filters in the VDD, with corner frequencies of 100Hz or so. Thus 100 ohms and 10uF, providing 1milliSec tau and 159Hz corner, is good. Or 10 ohms and 100uF. The typical VDD tree is shown in the final diagram of this answer.

I've read about Shunt Regulators for 1 nanoVolt noise levels, for audio vinyl RIAA playback PreAmps. Their PreAmp has no NO power supply rejection, and a low noise ShuntReg has been key to extremely pleasing music. That discussion is at "Simplistic NJFET RIAA" thread, on 'diyAudio' website. Again, that is for 1 nanoVolt levels, in my opinion; the audio experimentors just savor the music, they don't try to compute the VDD ripple floor.

You can easily model the Power Supply rejection of the OpAmps in your circuit, using individual models or Global models, using the tool Signal Chain Explorer. You describe the PSRR as [DC_attenuation, Frequency_corner where attenuation begins to lessen]; you also have RL-C-R-CCC filter networks customizable for each OpAmp; ESR and ESL are params for each capacitor. This tool is free, from robustcircuitdesign.com If you download and use it, feel free to tell us how convenient you find the menus.

Here is the VDD menu for one opamp; click "power" button after selecting stage. enter image description here

The various capacitors in parallel are used to show the series + parallel resonance dips and peaks of poorly-designed VDD filtering. Here is "SHOW FILTER RESPONSE" of the default-values LRC-RCCC network (without PSRR). enter image description here Observe the resonance of 10nF & 1uF (with their ESLs) at 26MHz, where opamps exhibit 0DB PSRR. Dampening is needed.

Some years back I needed a gain of 1Million at 100,000Hz for a magnetic-beacon prototype. Input was 2uVpp, output was 2 voltspp. I wanted to switch the gain, in binary steps, to keep the ADC signal between -6db and -12dB FullSwing, to tolerate thermal noise.

Major risk was oscillation, with the VDD-tree providing the feedback path; given OpAmp PowerSupplyRejection is poor (if not zerodB) at 100,000Hz, I knew VDD distribution was necessary design task. As teenager, I built numerous audio high-gain amplifiers that almost always "motorboated" puuut-puuut-puuut as the stages were driven (by themselves) from min-out to max-out. Here is the first-pass successful VDD-tree.

Each of the GNDs shown is separate from the other GNDs,

probably by several centimeters,

to minimize coupling. Notice the LPF, 10 Ohm & 1,000uF, is 16Hz.



simulate this circuit

  • \$\begingroup\$ And given the intrinsic noise of a magnetic cartridge, a 1 nV floor is just silly. But there's a lot of silliness in really high-end audiophilia. \$\endgroup\$ Feb 6, 2017 at 18:02
  • \$\begingroup\$ The RIAA rolloff imposes a 50Hz noise-bandwidth. As you likely know, 60_ohm Rnoise is 1nV density; scaled by sqrt(50) produces 7 nanoVolts RMS KT noise. Apparently the auditory cortex can detect tones down by 105dB. The 'diyAudio' folk use MM at 2-5mv, and MC down to 100,000nanoVolt (100uV). Thus suppressing fundamental and harmonics of 60Hz (for those Europeans, 50Hz) to -100dB does requires 1nanoVolt. Broadband noise degrades the highs, from what they say. Ripple/harmonics affect the bass clarity. The ShuntRegs have even the servo loop and reference-circuit running at HIGH CURRENT. \$\endgroup\$ Feb 7, 2017 at 5:23

For really low noise work I like battery power with everything mounted inside a cake tin (Or diecast box), 6 * D size 1.5V batteries will get you +-4.5V or so, add a couple of large caps, job done.

If the +-5V thing is really important, then go to 8 D cells to get you +-6V and use a couple of low noise LDO regs.

How quiet the power rails need to be is a function of circuit PSRR (Which is also worth testing), but note this tends to get MUCH worse at high frequency.

Under no circumstances use a switcher if measuring for ultimate noise matters. It is possible to do this and have it not pose a problem, but you will spend longer developing the filters and screening then testing the device you are interested in, the things are not that hard to work with but it is pain you don't need.

Do bond the output cable screen to the box and connect the internal doings to the box at one point.


Capacitive multipliers. (The transistor version) are my favorite noise "hammers". You can get a few nV/rtHz, without much work. (It might be harder for you to measure the noise.) This only really matters when the power supply noise can couple directly into your signal. Opamaps provide decent PSRR up to somewhere near their GBW, so for opamps, a low pass will often be good enough. (as has been suggested by others.)


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.