It doesn't, it can't, it's already at the thermal noise limit of the input devices, no different than off of battery supplies. Really Ed your being too silly today.
If I short the output the noise goes away.
Funny I though the power supply was in series with the output. Should we try a complete circuit equation and path?
BTY I am running my preamps from battery supplies.
Mostly but when you include the trace lengths and other real world issues they begin to show up at the levels being investigated.
I'm not going to dig it up again but it's on Mr Hill's web site. The "noiseless" transformer feedback technique was verified using ordinary good lab practice at .06nV/rt-Hz. Honestly are you being serious or is this a joke? Those spectra indicate you have created something pathological somewhere.
As usual... It is after forming that the noise drops going from 500 uF of bypass to 25F.
So what are you trying to show, that supercaps have some king of process when first energized like in some batteries where whiskers form and are broken by current or what? I'll repeat, you have a very low noise gain of 100 amplifier looking at a 1K resistor you expect 400nV/rt-Hz at the output. There is no amount of PS filtering that can alter that unless your circuit is so poorly designed that the PS causes >400nV/rt-Hz at the output.
Last edited:
They simply make the PSU non-ideal.
I am a pragmatist - you have to deal with things as you find them.
Ed
What Scott says is that whatever you’ll do to the PSU, measured noise can never go below the thermal noise of a 50 Ohm resistor if the amplifier under test has an equivalent input noise of 50 ohm.
A very relevant article is on EDN-Europe latest volume (Feb 2016)
Measuring 2 nV/√Hz noise and 120 dB supply rejection in linear regulators; the Quest for Quiet
EDN - Industry News, Learning center, electronic design center
George
What Scott says is that whatever you’ll do to the PSU, measured noise can never go below the thermal noise of a 50 Ohm resistor if the amplifier under test has an equivalent input noise of 50 ohm.
A very relevant article is on EDN-Europe latest volume (Feb 2016)
Measuring 2 nV/√Hz noise and 120 dB supply rejection in linear regulators; the Quest for Quiet
EDN - Industry News, Learning center, electronic design center
George
George,
What is dropping is the 1/F noise. It seems that the specified corner frequency was influenced by the bypass capacitors not some other mechanism. The equivalent noise resistance is under 3 ohms. So things get really fussy.
Now the other issue is that not all of the noise is from the input. If you look at the input device as a variable resistor with a noise source at the input there is also some noise contributed by the resistor itself. Now if you feed that resistor from a current source of infinite resistance you will get maximum gain. If you feed it from a matching resistance you will get half the gain and less noise.
No there is also a flat band reduction of ~2dB for which, unless you show us exactly what you are doing, there is no apparent mechanism other than a faulty layout/PS.
"Input device as variable resistor, etc." makes no sense at all, try and illustrate what you mean. 0.1nV noise has been measurable for decades long before super-capacitors.
OK Ed
The 50 Ohm was from your input
Your preamplifier is made out of TSH300
http://www.st.com/web/en/resource/technical/document/datasheet/CD00066247.pdf
each with an Input noise of 0.65 nV/√Hz.
Looking at the pic of your populated PCB there are 25 ICs. (post #78440)
So we are assuming a compounded input noise of 0.132nV/√Hz, equivalent to the thermal noise of an 1.08 Ohm (I used this online calculator for all this Thermal Noise | Equations Calculations Formulas | Calculator )
Looking at the diagram of that post (#78440):
I think that the red line (10 Ohm// to input) verifies that this calculated equivalent input resistance number is very close to what AP measures (notice your remark there: 0dB=1mV)
The knee (crossing of flicker noise line with thermal noise line) of the red curve is at around 500Hz to 1kHz
Now looking at the diagram of post #78628
All the curves in this diagram show one knee at ~150Hz and one at ~1kHz . I don’t know what this shows
George
The 50 Ohm was from your input
Your preamplifier is made out of TSH300
http://www.st.com/web/en/resource/technical/document/datasheet/CD00066247.pdf
each with an Input noise of 0.65 nV/√Hz.
Looking at the pic of your populated PCB there are 25 ICs. (post #78440)
Guessing that one IC is reserved for either DC servo or an output buffer, there are 24 ICs paralleled. This reduces the input noise by √24.
So we are assuming a compounded input noise of 0.132nV/√Hz, equivalent to the thermal noise of an 1.08 Ohm (I used this online calculator for all this Thermal Noise | Equations Calculations Formulas | Calculator )
Looking at the diagram of that post (#78440):
I think that the red line (10 Ohm// to input) verifies that this calculated equivalent input resistance number is very close to what AP measures (notice your remark there: 0dB=1mV)
The knee (crossing of flicker noise line with thermal noise line) of the red curve is at around 500Hz to 1kHz
Now looking at the diagram of post #78628
All the curves in this diagram show one knee at ~150Hz and one at ~1kHz . I don’t know what this shows
George
Last edited:
George,
The one with the 1000 hertz is before I added the 25F capacitors. What bothered me from the first measurements is I did not show the data sheet 1,000 hertz breakpoint as shown in the data sheet. So I got curious to see if that was a function of not enough bypass capacitor value. So I did not fully assemble the second one before testing. Seems it is the additional bypass capacitors changed the knee.
This indicates to me that the input stage was not the only noise source.
In some differential input circuits the input stage gain may be only 10. So 1nV/rtHz there would pass on 10nV to the next stage often the Vas stage. If that stage use a 3nV device and has a gain of 100 then the output noise would be 1.044 uV
Now the other issue not yet measured is the actual power supply noise. The capacitors are rated 30mohm ESR. The batteries 6mohm. But the batteries have a protection circuit and the wire leads to connect them add a bit more. So I do need to measure the before and after power supply noise.
BTY all 25 are in parallel. Still haven't added a servo. Also there is a bit of resistance in the input wiring and connector.
But the results are interesting.
The one with the 1000 hertz is before I added the 25F capacitors. What bothered me from the first measurements is I did not show the data sheet 1,000 hertz breakpoint as shown in the data sheet. So I got curious to see if that was a function of not enough bypass capacitor value. So I did not fully assemble the second one before testing. Seems it is the additional bypass capacitors changed the knee.
This indicates to me that the input stage was not the only noise source.
In some differential input circuits the input stage gain may be only 10. So 1nV/rtHz there would pass on 10nV to the next stage often the Vas stage. If that stage use a 3nV device and has a gain of 100 then the output noise would be 1.044 uV
Now the other issue not yet measured is the actual power supply noise. The capacitors are rated 30mohm ESR. The batteries 6mohm. But the batteries have a protection circuit and the wire leads to connect them add a bit more. So I do need to measure the before and after power supply noise.
BTY all 25 are in parallel. Still haven't added a servo. Also there is a bit of resistance in the input wiring and connector.
But the results are interesting.
Ed, I hadn’t noticed it when I posted the link for the TSH300 datasheet (*) but I noticed it now (I’ve had some rakija snaps meanwhile):
Your noise plots (these before adding the supercaps) compared to the datasheet noise plots, are much quieter at 100Hz than at >1kHz
Approx 20 to 24dB lower at 100Hz and 6 to 10dB lower at >1kHz actually.
And after sipping the snaps, I see only one knee (at ~1kHz) on the plots of your #78628.
What can I say!? Should I start a long-term therapy?
George
(*) and section 5 of the datasheet is very good
As op-amps go Aol and PSRR/CMRR fall far short of what one would expect of a precision amplifier in this chip. The Aol is so low that at a closed loop gain of 1000 there would be serious errors from ideal. As pointed out in the Art of Electronics thread there are discrete NPN's at rbb of 1.7 Ohm so not many in parallel would make an easy to realize a composite amplifier at much less size and power and could be designed for better performance.
- Status
- Not open for further replies.