Low Noise Direct Coupled Preamp

 Steve Wilson wrote:
=========================

** Horse manure.  

A 1uF film cap ( with SFA leakage) feeding a 10Mohm load gives corner frequency of 15 mHz.  

If you have a point, you did not post it.  


..... Phil  

Phil Allison wrote:


Nah, trying to do this with AC coupling is useless in the sub-audio range.

The tempco of polypropylene is about +200 ppm/K.  With even 1V DC on it,  
that cap will drift -0.2 mV/K.  To get down to the ~1 nV/sqrt(Hz) level,  
the temperature fluctuations have to be below 5 uK/sqrt(Hz). Pretty  
tough at low frequency.

And then there's the drift, which is even worse.

Cheers

Phil Hobbs

 Phil Hobbles Along and Bullshits wrote:

========================

 **  Crapology.  

The OP only spoke of cap leakage  -  fuckhead.  
 --------------------------------------------------------------------------



** Likely never matter in real applications involving DC PSUs,  zeners and the like.  

 Wot a colossal wanker you are.  


........... Phil  

On Thursday, February 25, 2021 at 6:36:40 PM UTC-5, Phil Hobbs wrote:

Who wants 1 nV below a hertz or so?  ,I once had this idea to measure 1/f noise  
from DC, so I made a bridge (All 1 k ohm 0.1% MF resistors to start* IIRC)  And all I remember  
was DC drifting one way or the other.. (Thermal effects maybe) I didn't spend much time on it.  

George H.  
* replace one with carbon composite later.  

George Herold wrote:


At frequencies for which the thermal mass approximation still works, the  
amplitude of thermally-forced noise and drift goes as 1/f.  That extends  
the sensitivity up to surprisingly high frequencies, especially in a  
powered test circuit with people moving around.

Making good measurements at sub-hertz frequencies is surprisingly hard.

Cheers

Phil Hobbs

On Thursday, February 25, 2021 at 8:30:05 PM UTC-5, Phil Hobbs wrote:

I've used Phil's technique, a 1 uF film cap and big R.  
to measure power supply noise.  It worked fine...    
(I still remember being surprised the big R didn't add to the noise,
mostly dominated by the opamp in this case.)  

Grin, for the above case if I tried to measure the power supply noise below  
~100 Hz, at maximum gain, it would start to oscillate as soon as I looked at it  
sideways.  

George H.    

George Herold wrote:
=================


** If the noise coming from your  DC PSU is dominated by the input noise voltage  
 of a modern BiFET op-amp ( 16nV/rtHz )  ....  you need bigger things to worry about.  



.....   Phil

On Thursday, February 25, 2021 at 10:27:35 PM UTC-5, snipped-for-privacy@gmail.com wrote:

Oh in this case the supply was to bias other noise measurements.  
Powering leds or light bulbs for shot noise measurements.  
or just resistors...  
Mostly with 8nV/rtHz opa134's, with some averaging you can measure ~1 nV.
(you need a switch to ground the front end and measure the noise w/o the supply.)
George H.    

Phil Allison wrote:


If all you're attaching is a mic and some speakers, that's true.  In an  
instrument that 16 nV/sqrt(Hz) could easily degrade the noise floor by  
25 dB.  (Discrete circuitry is de rigueur for the lowest noise  
applications, and generally doesn't have much PS rejection.)

Cheers

Phil Hobbs

George Herold wrote:


Sure, I have too.  In his frequency range and noise levels there's no  
reason not to.  But a dedicated noise tester really needs to be able to  
go below 0.1 Hz IME.  Lots of voltage references specify noise in that  
range, and below a few hertz the tempco of the capacitor can easily be a  
serious limitation.

The effect is so large--like 1 mV/K on a 5V supply-- that you don't have  
to slew the temperature of the inside of the cap.  Just the outer foil  
is enough to worry about, and that changes surprisingly fast with air  
currents and so on.

Cheers

Phil Hobbs

On 26/2/21 11:23 am, George Herold wrote:



Seismologists, perhaps?

CH

Clifford Heath wrote:


Getting within 30 dB of there will be a challenge with that circuit.

Cheers

Phil Hobbs

The typical 1/f corner frequency of the LT1028 is 3.5 Hz. This limits the
usable low frequency response of the preamp to 1Hz or so. Multiple
sections can be added in parallel to reduce noise, but this quickly runs
into practical limits of 1/sqrt(N).  

The value of the preamp is measuring sources that have greater flicker
noise than the preamp. Examples are low noise power supplies, zener diode
noise, IR LED noise, etc.  

Steve Wilson wrote:


It gradually degrades the noise floor, but it takes more than two
decades for that 1/f contribution to get up to the flatband of Phil A's
noiseless TL084.

3.5 Hz / (16 nv /1 nV)**2 = 0.014 Hz.



If you have a squint at the battery paper I linked to, you'll find a  
discussion of the two-channel correlation technique.  With CMOS  
amplifiers, you can get the noise floor down way below the amplifier  
noise, though of course you have to average for awhile if you want it to  
work well at very low frequency.



Sure.  But there are a lot more bogies down in the very low baseband  
than just flicker noise.

Cheers

Phil Hobbs

[...]




I don't understand your calculation.  

The typical LT1028 hits 1nV/root(Hz) at 10 Hz, and 0.8nV/root(Hz) at 100  
Hz.  

The preamp starts to roll off around 1 Hz, so 0.014 Hz is well below the  
cutoff frequency.  



Fred Walls has published an amazing number of papers on low noise
techniques.  

The system illustrated in Fig. 1 requires duplicate identical noise
sources, and duplicate amplifiers with positive and negative inputs to
take the difference between the two noise sources.  

The low noise preamp described here has only a single input. It cannot
take the difference between two inputs.  



The preamp starts to roll off around 1 Hz. It does not respond to very low
frequencies.  
  

Steve Wilson wrote:


The 1/f noise goes up as, well, 1/f.  That's in terms of power, so to  
get N times more noise _voltage_, you go N**2 times lower in frequency  
to get N**2 times more PSD.


That's just on account of your circuit topology and parts values, not  
the limitations of the op amp.



Again, that just requires moving a few wires or changing some component  
values.


Cheers

Phil Hobbs

[...]


  

  


I don't want to get more noise power. This calculation has no meaning in  
this application.  
  
[...]



I don't see how moving a few wires would change the low frequency response.

I do not want to measure very low frequencies.  

The 1/f noise would dominate the measurement and make it useless. I want  
to start rolling off around 1 Hz, or even a bit higher. Fotunately, the  
design of this preamp makes it very easy to change the upper and lower  
cutoff frequencies.

A capacitive coupled preamplifier would suffer extreme transients if the  
input capacitor value is changed. The preamplifier could be destroyed if  
the capacitor was suddenly connected to a large voltage source.

Direct coupling as shown here changes the transient into a simple slew rate  
issue.

I could make a NIST correlator to measure very low frequencies, but that is  
a different ball game and out of the scope of this preamplifier,


  
  

Steve Wilson wrote:


<>

You wrote,
"The typical 1/f corner frequency of the LT1028 is 3.5 Hz. This limits
the usable low frequency response of the preamp to 1Hz or so."

But that isn't true, as I demonstrated.  Your circuit is your circuit,  
and does what you want it to, but it isn't the only one in the world.  
That's a separate point.

There's been a lot of fuzzy reasoning in this thread, basically along  
the lines of "my filter cuts off everything past the 3 dB point."  When  
the interfering signal is very strong, as the tempco forcing in biased  
capacitors is, you have to be way down the skirts of the filter before  
it becomes negligible.


Cheers

Phil Hobbs

Yes, it is a single pole filter. The bandwidth is measured at the -3dB  
points, as is normal for filters. If you know it is a single pole and the  
3dB point, you can easily find the response at any point on the curve. The  
LTspice PLT file gives you that curve.  

A single pole filter has a pretty crappy rolloff, but that is not a problem  
in this application. The measurement time is too fast for the capacitor to  
change temperature and affect the reading.

I do not need much low frequency response. All I need is to find the corner  
frequency and the noise floor.

The corner frequency is where the noise starts to rise at about a 45 degree  
angle. The 1/f noise is pretty much the same for most sources, so once you  
have the corner frequency, there is not much more to learn.

Measurement speed is also an issue. I may want to sort through dozens or  
hundreds of LEDS to find the best brand and the quietest ones. If the  
filter frequency response is too low, I will have to wait longer for the  
measurement to settle.  

The noise floor is much more significant. This separates the noisy sources  
from the quiet ones. The quiet ones is what I am after, so that is the main  
emphasis. The noise floor occurs at a much higher frequency than the corner  
frequency, so low frequency response is not a major concern.

I expect the noise sources to be much greater than the noise floor of the  
instrument. If I need a lower noise floor, I can always use multiple units  
in parallel, up to the practical limit set by 1/sqrt(N).

Temperature coefficient of the capacitors is not a concern. The instrument  
must be in a shielded box to eliminate RFI. This also reduces the random  
air currents present on the outside. If tempco becomes an issue, I can dump  
the circuit in a jar of mineral oil to minimize any fluctuations.

Having a higher rolloff frequency also minimizes any effect of capacitor  
tempco. The cutoff frequency is simply too high for the capacitor to change  
temperature much and affect the reading. If it still becomes an issue, I  
can raise the cutoff frequency to allow less time for the capacitor to  
change value.  

All I need is to be able to find the corner frequency, which may be in the  
tens or hundreds of Hz. Not much is going to happen to the capacitor  
temperature in that short of time.

I have presented an approach that eliminates the capacitor coupling at the  
input, that eliminates any danger of destroying the base junction of a  
bipolar or jfet due to current surge, that settles quickly, is easy to  
change the rolloff frequencies, has reasonably low noise due to the LT1028,  
and is simple and inexpensive.

And did I mention - it gets rid of the coupling capacitor at the input.

It was also designed in the Theta state.