[sdiy] A question about regulator noise figures
brianw at audiobanshee.com
Tue Mar 16 02:02:30 CET 2021
Many measurements of performance cannot be converted into a universal number for comparison.
Total Harmonic Distortion is usually 1 kHz at some Wattage, and then reported as a percentage in terms of harmonics versus fundamental. Quite a few amplifier makers leave out one of the parameters, and then you have no idea. It's even worse when they're just reporting the Wattage rating of the amp, at least if they don't tell how much distortion they're allowing for. A few brands allow over 1% distortion, while another manufacturer will stop at 0.03% distortion. I don't think there's any way to compare those numbers unless they provide a graph. Even then, extrapolating a value from a low resolution line on the page is quite rough.
Noise is also something that needs a few parameters. Audio noise is often filtered using A-weighting, B-weighting, or something else. Few know that a different rating is required for different dB SPL levels, although that's not relevant if you're not actually measuring sound pressure levels.
I'm not quite sure how to compare V/root(Hz) but it seems to me that noise is RMS is lacking some parameters if you don't know how it was filtered. I mean, they might report total noise, but they'll probably make the numbers look better if they filter out inaudible noise frequencies. Of course, that's only for audio signals. Regulators don't output audio, so perhaps the total noise spectrum is more important than a weighted value.
Finally, RMS can be deceiving, because it doesn't tell whether the noise has a particular frequency content. That's more likely to be an issue with switching regulators, where a few models use spread spectrum techniques to avoid having all the noise at one frequency. A linear regulator should have a very natural spectrum, not concentrated around a single frequency, and I think that's all uV/root(Hz) is saying: that the noise has a particular spectrum shape (and level).
p.s. One solution is to buy the Evaluation Board from the manufacturer for the regulator chip that you're interested in. Then you can perform your own measurements with an oscilloscope or Audio Precision device and compare all options on equal footing. Of course, that's no easy task, especially if the manufacturer doesn't even make an evaluation board.
On Mar 15, 2021, at 17:42, Tom Wiltshire wrote:
> I also notice now that I’d missed that the figure given is actually 8uV / sqrt(Hz), which they’ve expressed as uV/(Hz)1/2
> I assumed the superscript referred to the notes under the table (which is Microchip’s usual style, to be fair), but that’s not what’s going on here - it’s an exponent.
> On 16 Mar 2021, at 00:29, Tom Wiltshire wrote:
>> Aah! OK, thanks Brian. Yes, flicker noise I know about, but I hadn’t made the connection.
>> I still don’t understand the units though. How could I convert a flicker noise measurement like 8uV/Hz to a RMS measurement across a given bandwidth like the other data sheets offer? Is that even possible? I’m starting to suspect some manufacturers don’t *want* me to be able to compare specs with their competitors! ;)
>> On 16 Mar 2021, at 00:05, Brian Willoughby wrote:
>>> Look up the tern, 1/f noise, a.k.a. Flicker noise.
>>> That sort of noise seems to be reported in V/Hz. In this case, it's µV/Hz.
>>> On Mar 15, 2021, at 16:57, Tom Wiltshire wrote:
>>>> Hi All,
>>>> I’ve been reading data sheets for linear regulators, and I’ve got a question.
>>>> Some regulators specific the output noise voltage from 10Hz to 100KHz as an RMS level, usually uV. However, the Microchip MCP1702 specifies an output noise figure as 8 uV/Hz. What does this mean? What’s the relationship between uV/Hz and RMS uV?
>>>> Here’s the datasheet:
>>>> Thanks for any guidance!
More information about the Synth-diy