I was just curious, so no apologies needed.😊
I just found my old simulation and it seems like there was an error. The noise from the LM431 reference was not modeled correctly. I changed the reference diode to a model that has a closer match to the noise of a real LM431. That gave me a simulated noise close to what I measured.
Changing the reference filter from 10k/10uF to 10k/47uF reduced the simulated noise by a factor of 4 at 20 Hz.
I will modify my amplifier and make real measurements to see if it achieves the same reduction.
I just found my old simulation and it seems like there was an error. The noise from the LM431 reference was not modeled correctly. I changed the reference diode to a model that has a closer match to the noise of a real LM431. That gave me a simulated noise close to what I measured.
Changing the reference filter from 10k/10uF to 10k/47uF reduced the simulated noise by a factor of 4 at 20 Hz.
I will modify my amplifier and make real measurements to see if it achieves the same reduction.
Post #41:
I re-read AN-124 from LT once more, and notice:
A750 'd settle in no less than 24-hours.
I have some tests with RNL1C152MDS1 1500 uF x16V and get < 5nA in about 2 days.
Better design approach 'd be differential input stage, same way as LT, despite 3dB increase in noise level from BJT. At least one of two huge capacitors eliminated.
Setting 12 caps in parallel doesn't lower leakage current, and there is no difference if its input cap or DC bias filtering. Noise getting injected into first stage. Good things is voltage just 1.25V.
I re-read AN-124 from LT once more, and notice:
A750 'd settle in no less than 24-hours.
I have some tests with RNL1C152MDS1 1500 uF x16V and get < 5nA in about 2 days.
Better design approach 'd be differential input stage, same way as LT, despite 3dB increase in noise level from BJT. At least one of two huge capacitors eliminated.
This is really interesting, thank you for bringing it up. I assumed that the al-poly caps would introduce excess noise when coupling to high impedance nodes as the leakage current is converter to voltage noise across a resistor, but that its impact would be of lesser importance when bypassing low impedances, such as the emitter circuit.
When you have a hammer, everything looks like a nail. I threw together a test circuit where a 9V battery biases a capacitor in series with a 100 ohm resistor. The idea is that any signal appearing across the 100 ohm resistor is either the result of the capacitor's leakage current, or the battery's own voltage noise. I took readings around 5 minutes, 20 hours, and 48 hours with the A750 and at 5 minutes with a Kemet 470u 25V ESE capacitor
As you say, the A750 does settle within 24 hours (which is somewhat problematic; the battery on the preamp won't live that long). But the ESE's noise is much lower even on startup. This is part of why I think it's worth trying conventional electrolytics in the emitter bypass position as well, but I do expect that they will have a smaller impact than the ones in the input coupling position. I'd like to test film caps as well.
I do agree that it's worth investigating the differential design. 3dB is a small price to pay when you're starting with 300pV/sqrt(Hz) of noise. SSM2212 or JFE2140 seem like interesting candidates. ADI did it with a bunch of THAT300s but seem to get a high 1/f corner, and needed 16 transistors to get there.
https://www.analog.com/en/resources/app-notes/an-159.html
I look forward to reading your findings. I had focused on the ZTX851's base current working into the 100k bias resistor at the output of the servo. I very nearly used the filtered LM431 and op amp configuration as you did, but opted for the LT3042 as I had a couple on hand.
I just measured the noise of my low noise amplifier.
The amplifier that I tested performed better than the one measured back in 2018. I will try to re-do the measurements on the other unit. I don't have it available at the moment though.
As stated, the performance was better than my previous measurements. On the other hand, changing the capacitor in the reference filter did not really make a difference. I have attached the latest measurement.
This is the output of the amplifier, so input noise is 60dB lower.
The broadband noise is just below 300pV/rtHz.
The noise at 20 Hz is around 700pV/rtHz.
The 3dB frequency response is from around 0.5 Hz to 2 MHz and it is essentially flat from 3 Hz to 1 MHz. This was measured with a source resistance of 1 ohm.
The amplifier that I tested performed better than the one measured back in 2018. I will try to re-do the measurements on the other unit. I don't have it available at the moment though.
As stated, the performance was better than my previous measurements. On the other hand, changing the capacitor in the reference filter did not really make a difference. I have attached the latest measurement.
This is the output of the amplifier, so input noise is 60dB lower.
The broadband noise is just below 300pV/rtHz.
The noise at 20 Hz is around 700pV/rtHz.
The 3dB frequency response is from around 0.5 Hz to 2 MHz and it is essentially flat from 3 Hz to 1 MHz. This was measured with a source resistance of 1 ohm.
Just made another thread, check it out:
I solved input protection long standing issue. And escape big caps, though I have 10 pcs 35 Farads im stock.
I solved input protection long standing issue. And escape big caps, though I have 10 pcs 35 Farads im stock.
I ran a few more tests to try to optimize this design and wanted to share the results should anyone attempt to go down a similar path.
First off, the emitter bypass caps. I was using the Kemet A470 al-poly caps and found that they had significant leakage current, which resulted in low frequency noise when installed in high impedance nodes. The emitter is not a low impedance node and so I expected similar performance. I replaced 12 A750s with 5 Rubycon ZLQ 4700u 6.3V caps to get approximately the same total capacitance and observe any low frequency effects.
Looks to be the same. The ZLQ caps have slightly higher broadband noise; I would assume this means higher ESR.
Next, I tried replacing the OPA132 with an OPA1655. This op amp has higher open loop gain. I also installed the remaining ZLQ caps.
The remaining ZLQ caps did bring down the low frequency noise a bit. I wouldn't go so far as to say it's a worthwhile improvement. The noise below 10Hz is completely unaffected, so there must be another noise source which is dominant here.
Past 500Hz, there's a significant improvement to the noise floor, bringing things in line with the predicted 270pV/sqrt(Hz). I suspect that's the extra opamp gain.
I did try measuring the impact of the 12 caps before doing the op amp switch, but was getting erratic low frequency noise. I ended up populating a new board and the issue was resolved. My guess is that all the flux created leakage paths despite my best efforts to clean it.
One last measurement. Dick's website shows lower 1/f noise for the ZTX1051A than for the ZTX851, at the expense of higher broadband noise. So, in it goes.
As far as I can measure, the 1/f noise is identical, and broadband noise is slightly degraded. The 851 really is the best bet here.
First off, the emitter bypass caps. I was using the Kemet A470 al-poly caps and found that they had significant leakage current, which resulted in low frequency noise when installed in high impedance nodes. The emitter is not a low impedance node and so I expected similar performance. I replaced 12 A750s with 5 Rubycon ZLQ 4700u 6.3V caps to get approximately the same total capacitance and observe any low frequency effects.
Looks to be the same. The ZLQ caps have slightly higher broadband noise; I would assume this means higher ESR.
Next, I tried replacing the OPA132 with an OPA1655. This op amp has higher open loop gain. I also installed the remaining ZLQ caps.
The remaining ZLQ caps did bring down the low frequency noise a bit. I wouldn't go so far as to say it's a worthwhile improvement. The noise below 10Hz is completely unaffected, so there must be another noise source which is dominant here.
Past 500Hz, there's a significant improvement to the noise floor, bringing things in line with the predicted 270pV/sqrt(Hz). I suspect that's the extra opamp gain.
I did try measuring the impact of the 12 caps before doing the op amp switch, but was getting erratic low frequency noise. I ended up populating a new board and the issue was resolved. My guess is that all the flux created leakage paths despite my best efforts to clean it.
One last measurement. Dick's website shows lower 1/f noise for the ZTX1051A than for the ZTX851, at the expense of higher broadband noise. So, in it goes.
As far as I can measure, the 1/f noise is identical, and broadband noise is slightly degraded. The 851 really is the best bet here.
Running some more test with my circuits, that not much differs, I get unstable results. Sometimes "regular" and expected ~0.8 nV drops down below 0.2 nV. For no apparent reason, and this makes me think that circuits is not correct.
What we have is the composite amplifier, where gain distribution is not defined for each section:
BJT & OPA, but set as common value for both.
OPA has very high AOL gain at low freq. end, 80-120 dB, so BJT is enforced to run not as an amplifier
but as a "repeater", providing Gain = ~1.000xxx.
There is likely to be "a fight", when BJT is trapped to have gain << 1, much less than 1.
A repeater (BJT) configured as Common Base, having 1 Ohm source signal impedance likely to have very low noise,
but it's not what should be measured.
What is needed , noise level of Common Emitter Gain = ~10-20.
I install 1k /100k resistors to have strict 40 dB for OPA and whatever BJT could provide.
It turns out as 57 dB overall, so I inserted another 470 Ohm into NFB path to restore 60 dB.
After that I see 0.8 nV all the time.
What we have is the composite amplifier, where gain distribution is not defined for each section:
BJT & OPA, but set as common value for both.
OPA has very high AOL gain at low freq. end, 80-120 dB, so BJT is enforced to run not as an amplifier
but as a "repeater", providing Gain = ~1.000xxx.
There is likely to be "a fight", when BJT is trapped to have gain << 1, much less than 1.
A repeater (BJT) configured as Common Base, having 1 Ohm source signal impedance likely to have very low noise,
but it's not what should be measured.
What is needed , noise level of Common Emitter Gain = ~10-20.
I install 1k /100k resistors to have strict 40 dB for OPA and whatever BJT could provide.
It turns out as 57 dB overall, so I inserted another 470 Ohm into NFB path to restore 60 dB.
After that I see 0.8 nV all the time.