How to measure power supply noisefloor

I don't think you're going to get there with this. It may output something, but I doubt that it's real. There's nothing readily available on the market that will measure ripple down into the microVolts, which is what we're interested in for audiophile power supplies.

Here's a good start: Low Noise Measurement Preamplifier. We used Tangent's excellent design to tweak the power supply on the MiniMAX to really sterling performance:
Millett MiniMAX History

Foxy, why do you want to measure the noise from the PS?

A 7805 is the noisiest power source you can get. You can use a LM317 or some LDO regulator or perhaps my super regulator which produces 921 nV.

Hello FOXY,

If your are (like me) interested to PSU noise measurements, you can see the AN83
Linear Technology App here : http://cds.linear.com/docs/Application%20Note/an83f.pdf
This document is very useful.
It also describes all the electronic materials nedded to do this measurement.
I made it, and it works very fine !
You can find other interesting documents on the Linar Web site.
Good luck.

Frex.

Actually, 7805 and LM317 are about the same ballpark with respect to noise.

For example, compare Fairchild's LM7805 and LM317 datasheets, you'll see the LM7805 having 42μV/Vo (typ) noise, which, for a 5V regulator, translates to 210μV. The LM317 is rated at 0.003-0.01%/Vo (typ-max), which for 5V output, works out to 150-500μV. Thus the 7805 is right in that range, but note that the 7805's noise spec was rated between 10Hz-100KHz whereas the LM317's is only from 10Hz-10KHz. If the 7805 was measured with the narrower bandwidth, then I would expect its noise spec to be better, perhaps besting the LM317.

There is also the LM7805A (in the same datasheet) which is a lower noise version, with 10μV/Vo, so for 5V it would be only 50μV. Pretty good for a cheap three-terminal regulator.

Surprised? I was too, given the often-cited "conventional wisdom" that LM317 is "better" than the 78xx series. Now let's look at ripple rejection. LM7805 is 73dB. LM7805A is 68dB. LM317 is 60dB (or 75dB with the addition of the ripple reduction capacitor). These numbers don't demonstrate a clear superiority of LM317. One could only say that LM317 is worse, and only achieve roughly equivalent performance with added parts.

National Semiconductor's LM340T-5.0 and LM340AT-5.0 (pin-compatible with 7805) is even better with noise of 40μV and ripple rejection of 80dB.

There are some who consider the Linear Technology LT1083/1084/1085 series LDO adjustable regulators to be superior substitutes for the LM317, but one look at the datasheet shows that the noise and ripple specs to be the same as the LM317.

These numbers were no doubt taken in lab conditions with fancy shielding and other techniques to minimize external factors from influencing the results, so in reality we'd most likely see a bit more noise than in the spec, but still, the numbers speak for themselves and is a reason why I chose the 78xx series regulator instead of LM317 in my "cheap and simple" σ25 PSU. 78xx-based circuit has the advantage of lower parts count, too.

The σ11/σ22 at 5V is about 20x quieter than this, but we're talking a whole different league of PSU there. In audio circuits with high PSRR, σ25 could be very cost-effective.

When I was designing the TREAD, I started with the idea of using a 7824, since the idea was that the TREAD was to be simpler and cheaper than the STEPS, which I still offered at the time. I didn't want the two to compete with each other, but when I saw that the noise was 4x higher despite having all the recommended bypassing, I decided to go with the LM317 after all. This ended up killing off most of the demand for the STEPS, but I still feel it was the right call.

It could be that I happened to test an inferior 78xx sample, or perhaps there is some difference in manufacturer's renditions of the design to explain it. I don't think the 317 was exceptionally good, because I've not seen large variations in performance among them.

Consider this, too: both the AD797 and the LT1028 claim 0.9 nV/rt.Hz noise in their datasheets, but try both in an LNMP and you'll see the base noise floor be higher with the LT1028. The datasheet doesn't tell all.

Nice writeup, Amb -- any measurements to indicate how these things differ in terms of transient response?

A thread about battery powering a DAC

DIYHiFi.org • View topic - Battery Powered DACs?

If you know your load draw, the zener + emitter follower is apparently good :

Simple Voltage Regulators Part 1: Noise - [English] (part way down the page)

No, I did not measure transient response on any of the 3-terminal regulators. It's an exercise that I would be interested in doing some time down the line. I suspect that the output decoupling cap will have a very large influence on the outcome.

very interesting thread. I wish somebody have the noise data for the eneloop Nimh rechargeables.

I supposed since most opamps have very high PSRR, the power supply noise floor is not super critical for an opamp-based amplifier, correct?

I could measure it for you, but I can tell you that it's an order of magnitude higher than the numbers mentioned in this thread.... give me the parameters you are interested in and I'll measure it for you.

is it really that bad? noisier than even the LM317?


can you measure the noise level of eneloop vs standard Nimh from 500K~1.5Mhz? or maybe it doesn't really matter, as the intended use will be a noisy PWM switching (tripath amp) anyway.

Thanks

Hello,

After this discussion and for my own interest, i have decided to make some noise measurements for differents voltage regulator.
The three tested voltage regulators are :
LM7815CT fixed 15V regulator (ON semi, old Motorola) in TO220 package.
LM317 positive adustable regulator (STmicro) in TO220 package too.
LT1764EQ positive adjustable "low noise" regulator (Linear Tech) in D2PAK package.

All of those voltage regulators have been welded in a copper board with decoupling capacitors, as you can see in the schematic below:



To make those measurements, i used an home made instrument (yes, DIY!).
It's a low noise high gain amplifier with band-pass filtering (10Hz-100kHz ).
The amplifier gain is 80dB (10000x) in frequency range, and it's own noise is only 500nVrms (-127dbVrms), with it's input shorted by a 50Ohms plug.
(This amplifier is fully decribed in the AN83 from Linear Technology).
Then, i used my external audio A to D converter for compute and display the noise spectrum.
The ADC is calibrated, and the spectrum is directly displayed in dbV (0dbV = 1Vrms, -60dBV= 1mVrms.)
The sampling rate of the ADC is 192kHz(so ~90kHz max allowable signal), it is good for a measurement bandwith of 10Hz-100kHz.

Below, the brief description of the measurement procedure :



The FFT software display the averaged noise spectrum and the RMS value.
At verification purpose, the rms value computes by the sofware is always compare with the measurement done by the 9300F millivoltmeter.
For finding the real noise of the regulator, now i must add 80dB at this value (or divide it by 10000).
Measurements results for each regulator are grouped in the table below.


Note:
Column 2 and 3 are the same value but in dBV and mV.
Same for column 4 and 5.
Column 2/3 is the measured voltage at the output of the 80dB amplifier.
Column 4/5 are the calculated equivalent output noise of the regulator (x10000 ratio).

As we can see, the LT1764 specified for low noise operation is the worse!
Without a good ripple rejection capacitor, the LM317 is not the best, but became best with a good filtering capacitor.
The 7815CT is also good.
All of those regulators have low output noise. It is unlikely that this level of noise become a problem with any audio analog designs.

Nevertheless, for the hoobyist who want to design an ultra-low noise PSU (for supply ultra low phase noise oscillator for example),
there are many active noise reduction circuits. This actives circuits allow to reduce the noise up to 40dB(100x).

Below, a second DIY board where 2 voltage regulators are used. One LM317 for +15V and one LM337 for it's opposite of -15V.
At left, you see each regulator, and at right the active noise filter section ..


With this board, the output noise result is listed in the table below.


The first result with noise reduction circuit "OFF" is roughly the same than the previous result (-7dBv).
If the NR is active, the noise become very very low, 27dB lower !..

The graph shows the noise spectrum between 1Hz to 100kHz for each regulator type and configuration
This speaks for itself.



Last year, i had bought a JSR06 "low noise DC supply board" from Per-Anders Sjöström (sjostromaudio.com - Home).
It is a Pcb wich include positive and negative DC supply rail with very low output noise.
For the fun , i have also make some measurements on it's output noise.

The result is nearly the same than the LM317 with noise reduction.



The resulting spectrum is illustrated in the graph below.
In red, i have add the noise floor spectrum of the measurement system.
It's level is down to 500nVrms reffered to input in the measurements bandwidth.




I hope some of this measurements help to clarify some preconceived idea about noise issues.


Frex.