Power supply noise : measurement ?

[jujun]

Hello,

Is there a way to properly measure the noise of a power supply ?
Because I want to be able to measure it before building filter, to be able to quantify the improvement.

Until now I have added random big value inductor and capacitor, and then I was thinking it was better, but I cannot quantify it.

Thank you
J

[awallin]

Is there a way to properly measure the noise of a power supply ?

something like a Sound Blaster X-Fi, if the frequency-range it has is OK for you, possibly with a more or less fancy LNA between the PSU and the soundcard?
If you have lot's of money then a high-end FFT-analyzer SR760 or similar (does anyone make a modern FFT-analyzer, seems the SRS and HP's are quite outdated models..)

[jujun]

Is there a way to properly measure the noise of a power supply ?

something like a Sound Blaster X-Fi, if the frequency-range it has is OK for you, possibly with a more or less fancy LNA between the PSU and the soundcard?
If you have lot's of money then a high-end FFT-analyzer SR760 or similar (does anyone make a modern FFT-analyzer, seems the SRS and HP's are quite outdated models..)

Maybe a simple Rtl-SDR can do the job. (in direct sampling mode)

[awallin]

Maybe a simple Rtl-SDR can do the job. (in direct sampling mode)

depends on what you want to see. if you want uV or nV then a good LNA and a low-noise floor for the ADC is needed.
SNR is about 6N+1.8 dB for an N-bit ADC, so much better to start with 24-bit slow soundcard than a cheap 8-bit SDR which is too fast for analyzing powersupplies or voltage regulators anyway.

[David Hess]

The most comprehensive way is to make a noise measurement directly in nV/SqrtHz versus frequency using an FFT but most oscilloscopes do not do this correctly and manual corrections need to be applied.  A low noise preamplifier is likely needed also.  This EDN article discusses how to correct the FFT results from the DSO.

A spectrum analyser could be used to do the above but most do not cover or perform well at low frequencies.

I usually just make spot noise measurements using a Tektronix 7A22 vertical amplifier which has adjustable low and high pass cutoff frequencies between DC and 1MHz.  With some math, these can be converted into a rough nV/SqrtHz versus frequency graph.

Spot noise measurements work perfectly fine for measuring improvement and may even be preferred however making a wideband average or RMS AC noise measurement after filtering is not trivial and some DSOs cannot do this correctly on a noise waveform despite how simple it should be for them.  With this in mind, if I used a DSO for this, I would arrange some way to verify that the DSO's measures RMS noise correctly.  This is one of those measurements which counter intuitively is easier on an analog oscilloscope using the tangential method unless the DSO is known to make it correctly.

[Wolfgang]

Here is how I do it:

- Use (if you can, via an adapter) a coaxial path throughout your measurements
- use a coaxial load resistor (or at least a resistor in a shielded box)
- then a BIG decoupling capacitor (in a shielded enclosure, too).
- and a sensitive, low noise LF amplifier up to the frequency limit you need.
- The output of this amp is put into a scope or spectrum analyzer.

Some low noise PSUs are here:

https://electronicprojectsforfun.wordpress.com/making-noise/how-low-can-you-go-low-noise-measurements-at-low-frequencies/

and some tries of LF preamps are here:

https://electronicprojectsforfun.wordpress.com/rf-module-gallery/the-amplifier-module-gallery/a-battery-operated-low-noise-low-frequency-fet-amplifier/

[hwalker]

To measure the noise before the building filter, I would use a line impedance stabilization network (LISN).  The set-up would be, in turn: 1. AC input, 2. building filter, 3. LISN, 4. power supply 5. Passive load rated for maximum and minimum of power supply (The noise of the power supply may actually be higher at lower loads in some cases).  This is generally how conducted emissions testing is performed on stand alone power supplies.  Connect a receiver to the LISN which is capable of measuring between about 10x the power-line frequency and 30 MHz.  Be careful to only connect the LISN measurement port to your receiver after the power supply has settled, as turn-on transients can blow out an unprotected receiver front end.
-Herb

[jujun]

To measure the noise before the building filter, I would use a line impedance stabilization network (LISN).  The set-up would be, in turn: 1. AC input, 2. building filter, 3. LISN, 4. power supply 5. Passive load rated for maximum and minimum of power supply (The noise of the power supply may actually be higher at lower loads in some cases).  This is generally how conducted emissions testing is performed on stand alone power supplies.  Connect a receiver to the LISN which is capable of measuring between about 10x the power-line frequency and 30 MHz.  Be careful to only connect the LISN measurement port to your receiver after the power supply has settled, as turn-on transients can blow out an unprotected receiver front end.
-Herb

Thank you, great advice

Does this apply also to power supply that use a battery instead of the AC main voltage ?

All the information on LISN I found, was for AC main voltage.

[hwalker]

Automotive and Military compliance testing is typically dc powered (12v and 28V, respectively) and they both use LISN's in their test setup.  The battery output is attached to the input of the LISN in place of the filtered AC input.  The LISN you choose should be rated to handle the maximum current that the power supply will draw.  The LISN inserts inductance between the dc source and the power supply and may cause some switched-mode designs to oscillate.


- Herb 

[coppercone2]

The practical way is to use a thermal RMS meter if you want up to say 100MHz or a decent multimeter if you are content to something like 1MHz. It is a power rail after all. It might be enough for your needs.

I like wavetek/khronkite brand bandpass filters to do the job, you will get 1st order to 1MHz with a cheap ebay one. Its enough to triangulate a LC filter problem.

When I looked at the performance of filters, they offered suprizingly good performance even at high frequencies. I was panicked that I needed to select proper magnetics and stuff, but it turned out my RLC low pass filter still offered like 10dB attenuation at 50MHz, when it was tuned to the KHz region. So long you are reasonable with the dampening and layout you probably won't see any kind of crazy peaks outside of your band of interest, just less attenuation, but attenuation is always good, so whatever.

I think its better to design a heavier filter anyway, because you never know what kind of interference you can get upstream of the power supply. I am not sure there is that much point in fine tuning it unless its for sale and you need to keep costs down, so in my opinion if its for home use a AC meter and maybe a filter is all you need. Or I guess you might actually be interested in how a filter behaves and want to study it, like I do sometimes.


For a power rail you can also get good results using the digital filters on an oscilloscope. Even the cheap ones on my rigol 1052e are enough for a power rail, its really obvious on an oscilloscope if your filter is doing anything, just short it out. They are quite versatile an you can get an idea of whats going on just by sweeping your digital filters, and its free with most modern scopes and you don't need extra cabling etc. I found the 1052e FFT horrid in comparison to its digital filters for getting an idea of whats going on, the noise floor is too bad.

I don't know how far I would have gotten with my little home projects if I was pulling out the LISN for every LC filter I designed lol, but it is the proper engineering method. It sounds however that you are already much smarter then the average fish if you are putting extra LC filters on things, I don't see most people bother at all. All this advanced test equipment for frequency response testing is kind of expensive and it will slow you down ALOT to do it properly.

Dave Jones has a few EEVBLOG videos where he shows you how to measure power supply noise without advanced equipment and goes into the nuances.



The stuff listed in this thread is rather advanced. I could see it like throwing calculus books at high school students. Based on how you phrased the question I start to think getting this advanced can hamper you efforts, especially if you have done no measurements at all yet. Make some basic ones then decide if you are interested in going all AC analysis crazy

And if you want to get into the industry it helps being able to just use the most primitive tools to do something because people will look at you like you are smoking crack if you start asking for advanced equipment at work for things that people tend to 'lick the voltage' with.

Take one of your power supplies, connect it to the load of interest, hook up a multimeter in AC volts (like a lab meter or better fluke), then short the filter out with a wire and see what happens. You need to do this first. And don't use random values. Even using a filter calculator like http://sim.okawa-denshi.jp/en/Fkeisan.htm will give you much much better results then random values. And stick to inductors instead of ferrite beads, ferrite beads will drive you crazy.

Then test your filter like this