[time-nuts] How to measure regulator noise?

[Bruce Griffiths]

Poul-Henning Kamp wrote:
> In message <079101c8510f$af8c35a0$7900a8c0 at athlon1200>, "Dave Brown" writes:
>
>   
>> John
>> You won't need much of a cap for DC blocking as inputs are hi-Z....too 
>> big and the chargeup may well trigger the overrange condition.
>>     
>
> You need to put a shunt resistor after the capacitor if the input is hi-Z
> but an even better way is to offset the gnd clip with a couple of fresh
> batteries so that you can avoid the capacitor.
>
> Batteries have very low, but not zero noise, so what you do is:
>
> Put six fresh 1.5 volt batteries in series, so that three of them
> is the "wrong way" and the resultant voltage is zero and then you
> measure their noise.
>
> Then you put three of them "the wrong way" on your 5v supply, so that
> the output voltage is only .5V and then you measure the noise.
>
> That worked fine for me using a HP6885B
>
>   
This technique only offsets the voltage to within 3/4V of zero when
measuring an arbitrary voltage regulator output.
With a discrete regulator it is usually easy enough for test purposes to
adjust the regulator output so that it is within a few tens of
millivolts of the battery stack voltage.

A low noise preamp with perhaps 40-60dB gain is still required before
the spectrum analyser.
Even residual dc voltages of a few tens of millivolts may saturate such
an amplifier.
A simple dc servo can be used to remove the residual offset (<1V)
without using electrolytic capacitors with their attendant leakage and
noise.
A low frequency cutoff well below 1Hz is possible without adding
significant noise.

Alternatively for measurements in the 10Hz to 100kHz range a preamp like
that used in Linear Technology's AN83:
http://www.linear.com/pc/downloadDocument.do?id=4172
may be useful.
It uses electrolytic coupling capacitors together with low value resistors.

For ultralow (milliHertz) frequency noise measurement AC coupling is
best avoided altogether, if possible.

Bruce