-
I'd like to verify the performance of various low-noise regulators and filter circuits/schemes (e.g. common-mode chokes, ferrites, filters etc.) for producing low noise rails in instrumentation circuits.
Is the LT AN-83 amplifier still a good design or are there better op-amps or ideas around for this?
http://cds.linear.com/docs/en/application-note/an83f.pdf
Does anyone have a tried and tested 2-layer layout and board design?
Has anyone use this in combination with an audio analyzer, e.g. the QA401 or similar?
https://www.quantasylum.com/content/Products/QA401.aspx
would be cool to have some setup where the instrument noise-floor is 'sufficiently' low, e.g. you connect a 50R terminator to the input and the spec-analyzer trace actually rises up to 900pV/sqrt(Hz)
-
The LT1028 based amplifier in AN83 is a little odd: it uses the LT1028 in inverting amplifier configuration, this is not the lowest noise version. So it was not a good idea back then - the 100 Ohms resistor already gives more noise than the OP. The LT1028 is still good for a low impedance source (e.g. < 300 Ohms, < 100 Ohms at < 10 Hz), which usually means DC coupled if you need low frequencies below 10 Hz or so. So if you need AC coupling and low frequency the current noise is just to high to be practical.
So usually one would start with AC coupling with a rather low corner, than a non inverting amplifier and than the final low pass filter if needed. Much of the filtering can be done in SW today.
Good choices for the low frequency range are:
1) DC coupled amplification (e.g. 2-5 times) with the LT1028 or similar (+ power stage)
- followed by AC coupling: very low noise, but quite some power dissipation in low R divider
2) AC coupling with large electrolytic (e.g. 1000 µF range) and BJT based OP like LT1037
The OP choice depends on impedance an lower frequency limit.
3) AC coupling with foil caps and OPA140 (or similar), possibly 2 or 4 in parallel
4) AC coupling with foil caps and discrete JFET based amplifier (e.g. SK369, BF862 or similar)
5) AC coupling with foil caps and low noise AZ OP (e.g. ADA4522 or similar)
There was a thread a few weeks ago about an AC coupled low noise amplifier with several BF862 in parallel to get lower noise than the LT1028 over a large frequency range. -
I once designed a 2-layer board of this LNA in SMT:
http://www.mikrocontroller.net/topic/250656#2666994
and build it several times:
http://www.mikrocontroller.net/topic/250656#2676221
I have a modified unit (Gain=80dB instead of 60dB) running with a STM32 microcontroller with 12bit ADC, so I have the data directly digital without the need of an oscilloscope and use octave to analyze the data, some kind of FFT analyzer. -
Hello,
after getting one of those AN83-boards from branadic, I finally had some time to build a suitable supply and do some measurements after adjustment.
First one: noise floor with a 50 Ohms terminator at the input.
Result: 4.2uVpp or 483 nV RMS.
So near the ideal value of a 150 Ohms (50 Ohms terminator + 100 Ohm input resistor) resulting in 500nV RMS thermal noise.
Resulting noise density: 1.53 nV/sqrt(Hz)
With best regards
Andreas
-
Since DC accuracy is not critical with that application, a better solution could be to build an AC-coupled amplifier using discrete transistors. Sub 1nV/sqHz is possible. Like :
http://www.janascard.cz/PDF/Design%20of%20ultra%20low%20noise%20amplifiers.pdf
Or if you win the lottery : (Noise ~2nV/sqHz @ flatband)
https://www.picotest.com/products_J2180A.html -
Next step:
measure a freshly charged 9V NiMh block battery (actually 9.3V).
There are some short (0.25m) cables to connect the 9V block within the (metal) cookies box to the amplifier.
And the input capacitor (leakage currents) are also expected to increase the noise.
But what I had not expected are the 50 Hz mains frequency + harmonics.
They could be only switched off by fully switching off the complete lab.
(except the battery supplied laptop and USB-supplied oscilloscope).
So the 50 Hz seems to be some magnetic coupling from a analog transformer.
(The wall of the cookies box should be enforced to some feet thickness).
But since switching off the lab supply does not influence signifficantly the RMS value
I decided to keep my lab supplied through the further measurements.
Result: 9V Block NiMh
Lab Supply on: 5.398 uVpp 639.2nV Eff -142.5 dBV@50 Hz
Lab Supply off 5.242 uVpp 630.5nV Eff -173.2 dbV@50 Hz
noise floor with DC input voltage 2 nV/sqrt(Hz)
So obviously sufficient for many measurement tasks.
with best regards
Andreas
-
Since DC accuracy is not critical with that application, a better solution could be to build an AC-coupled amplifier using discrete transistors. Sub 1nV/sqHz is possible. Like :
Its always good to have some ideas as back up. (the FET amplifier looks very promising).
http://www.janascard.cz/PDF/Design%20of%20ultra%20low%20noise%20amplifiers.pdfOr if you win the lottery : (Noise ~2nV/sqHz @ flatband)
Mhm, one would need 3 of those amplifiers to get the 60 dB of AN83.
https://www.picotest.com/products_J2180A.html
And I do really not need the 100 Mhz. (20uV noise for the full bandwidth).
With best regards
Andreas
-
What I ever wanted to measure.
The broadband noise of a LTZ1000.
According to the diagram in the datasheet around 45nV/sqrt(Hz) above 10 Hz for the typical current of 4 mA.
Since I do not want to connect a 100 Ohms impedance to the unbuffered LTZ1000 I measure the buffered output.
The buffer has around 11nV/sqrt(Hz) so it increases the noise of the LTZ from 45 to around 46 nV/sqrt(Hz) (geometrical addition).
Further I have many filter capacitors 100nF directly at the output of the LTZ and after a 22R resistor at the output of the buffer.
The results: measured on the buffered output of LTZ#3
at 100 kHz bandwidth 8.364 uV Eff which is somewhat low compared to the expected value of 14.2 uVeff from the 46nV/sqrt(Hz).
But the FFT shows already a falling spectrum above around 10 kHz. (most probably due to the filter caps on the LTZ board).
Reducing the bandwith to 10 kHz with a digital filter of the scope shows spot on 4.482uV Eff.
Further reduction of bandwith to 1 kHz gives 1.588uV while expecting 1.423 uV Eff
with best regards
Andreas
-
I'd like to verify the performance of various low-noise regulators and filter circuits/schemes (e.g. common-mode chokes, ferrites, filters etc.) for producing low noise rails in instrumentation circuits.
Probably you'll find AN-159 useful.
And the datasheet of LT 3042.
AN-159 - Measuring 2nV/?Hz Noise and 120dB Supply Rejection on Linear Regulators
http://www.linear.com/docs/47682
Datasheet LT 3042
http://www.linear.com/docs/46159 -
Very nice results, thanks for sharing.
This shows, that building something with surface mounted parts very carefully can result in acceptable results.QuoteAnd the input capacitor (leakage currents) are also expected to increase the noise.
Not necessarily leakage, but several mechanisms are given in Low Frequency Noise of tantalum capacaitors.
-branadic- -
I don't think he's selling them anymore, but I found the Geller J-Can circuit very interesting. It's designed for measuring resistor noise. I built one and it really does approach theory over a wide range.
http://www.gellerlabs.com/jcan%20parts%20and%20kits.htm
-
The J-can circuit is only a limited frequency range and does not go very low (around kHz lower limit). So it usually does not get the 1/f noise part. The choice of using the LT1028 in the first stage is also strange - due to the relatively high source impedance it is not the best choice here. Filtering is also just a chain of 1 st order filters - so a rather smooth transition.
-
Measured transfer function of my AN83 in 80dB version (x100 in each stage) today. It does fit the simulation and also my liddle mod at the filter shows the expected results (reduce 5k62 resistor at LTC1562 to 5k2 and eliminate the peaking that way).
Measurement were done using good old and incredible loud HP4194A Impedance/Gain-Phase-Analyzer.
Edit: Added the data plot captured via GPIB.
Since -3dB point at lower frequency is more 11Hz instead of 10Hz I found that increasing both 330µF caps at the 5Hz single order highpass filters and increasing the 4.7µF cap at the 10Hz 2nd order butterworth highpass to 5.6µF gives a much better -3dB point match at 10Hz in simulation model then the original values. So maybe this is another good modification to increase performance of this noise amplifier.
-branadic-
-
looks really good.
in the mean time I measured my buffered LTZs.
The result is that those references with high 0.1-10 Hz (1/f) noise also have a higher 10-100kHz (wideband) noise.
So no surprise here.
The advantage is that the measurement of wideband noise is done much quicker than 1/f noise.
On the other side the popcorn noise is only visible on the 1/f noise measurements.
with best regards
Andreas
-
Andreas,
Maybe I missed this from previous posts, so forgive me. Your data looks pretty detailed, but usually we're used to seeing the other critical & helpful section included somewhere data reports. I know that's not what you're doing here in a brief summary, and I'm not complaining at all. It might help interested readers later on though - so the following is just a suggestion for "best practices" for anybody reporting low ppm data:
When you publish low-ppm data, it would be very helpful if you can add an equipment list with cal dates, and standard calculation of your lab's overall measurement uncertainty (for absolute value, not relative) @ what confidence level? It's also handy to add a measurement uncertainty for each test and technique separately in a separate column, so one can see what you measured for an absolute value @ what error band size. For instance: DMM in one test, uncertainty 8ppm, and null meter + KVD + 732 on second test, uncertainty 4ppm, etc. Or whatever technique you use to estimate a voltage measure.
It's always interesting to know the error band of each measure, because it's all just an estimate at low ppm - every measure always includes some level of uncertainty and something less than 100% confidence. When you put that information in with your data measures, it makes it more meaningful when comparing experiments.
What do you use for Volt Reference in your lab? - do you have something other than LTZ-based Vref to measure LTZ's against? I see you are making comparisons to LTZ#2, but is that the only reference you have?
-
Hello,
althoug it is quite offtopic here in the AN83 thread:
Of couse it needs more than one reference to get reliable results.
Im doing daily! comparison of 7 references with 4 ADCs. Partly also differential measurements. So I can detect unusual drifts.
Drift data is based on 5-7years calibrations of 4 of those references with calibrated gear (several DMMs and 2 Calibrators).
Standard deviation of ADCs (over 42 days = 1000hours) against LTZ#1 + LTZ#2 (proven to be most stable up to now)
ADC13: < 0.3 ppm
ADC15: < 0.5 ppm
ADC16: < 0.2 ppm (heated to 27 deg C).
And that over my lab temperature range 18-28 deg C
Thats why I can use the 6.5 digit DMMs only in differential mode to improve stability/resolution.
with best regards
Andreas
-
OK, thanks - that explains a lot. I was curious how you get from a relatively inaccurate 6.5 digit DMM and come up with a measure in the 100's parts per trillion range or lower (for instance on your TC you have reported x.x ppb). For those of us in industry, -every- decimal place reported implies you have in instrument capable of at least 5 to 10 times better accuracy on hand - which of course doesn't happen when measuring LTZ. Statistical math averaging should never result in a gain in basic accuracy (or orders of magnitude smaller error band), at least in a real world setting when you have to provide accurate real data. I know you're doing this just for fun and exploration - and again I'm not complaining, just an observation of your reported data.
I think when you go back and include the maximum error band of each endpoint of your TC slope calculation, the estimate of TC will have an error tolerance more in the ppm or higher error band range - it's reported error band should be on the order of uncertainty of the measuring instrument used or possibly bigger. Normally we wouldn't see any measure with a reported accuracy resolution better than 10 or 50 ppb (for voltage) unless it is a very expensive setup with direct access to JJA. That's what startled me when I looked at your data, that's all.
When you say "several dmms and calibrators" what DMM's are you using? What calibrators? Are they in their calibration period?? The reason I ask is that in general those really aren't good enough to measure low ppm LTZ drift changes, unless you've got a solid drift history and predictions on that other equipment. Generally older, more stable calibrated equipment gives you better results for low ppm. Usually the only thing good enough for LTZ is a bank of always calibrated 732a/b + KVD and 3uV or 1uV full range null meter or better...which will get you a relatively small error band for absolute voltage, about the same as you'd get from maybe 4ea or 5ea. 3458a's...something in that very low single digit ppm area. 3458a by itself is not an absolute value voltage transfer device, and never promised to be. That is the promise of several freshly calibrated 732a/b - if you're after very solid and formal lab measurements with best low ppm uncertainty + higher confidence in the 90% or 95% range.
It's always good to keep measuring everything as many ways as possible, and compare to as many LTZ and non-LTZ Vrefs as possible to get the best quality low ppm absolute measures. That will decrease uncertainty, keep the error band as small as possible and keep measuring confidence as high as practical. You'll never get to 100% confidence, but you can get to 95% in a practical lab.
Again, I'm not complaining about what you're doing - just keep in mind that LTZ's will tend to drift one direction, and right now you're just comparing one or two LTZs to another LTZ. They could drift together 50 ppm or 500ppm and you'd -never- see it unless you got back to one of your calibrated devices for a reality check. It is surprising how often a very small sample of LTZ's will drift down right along with each other. -
now its getting really offtopic.
But I´m doing the best I can without spending too much money.
Calibrators are Fluke 5700 where I have access to (once per year) and of course I get their calibration reports to reduce uncertainity.
DMMs are ranging from calibrated K2000, with long calibration history and reports where I know that the drift is below 2 ppm/year, to K2002 and 3458A from volt nut colleagues. (partly with official calibration reports).
Of course you cannot rely on a single measurement in this case. But if you have a near equal trend over 5 years of minimum 2 different calibrated devices.
What you claim is of course right for time-economical working and absolute values.
But: how could our engineers ever build a DMM with higher accuracy than a analog meter (1%) without having a 5 to 10 times better DMM?
With best regards
Andreas -
Hello,
a further noise measurement
this time a AD586 buried zener
noise according to datasheet 100nV/sqrt(Hz)
I have a 47uF aluminium electrolytic on my test setup on the output of the AD586.
This may be the reason for the low pass at around 6 kHz.
Results
100 kHz: 10.42 uV rms
10 kHz: 8.843 uV rms (slightly below datasheet due to output cap)
1 kHz: 3.048 uV rms (corresponding to the 100nV/sqrt(Hz))
with best regards
Andreas
-
Added a plot of the transfer function in post above and gave a hint on how to improve lower frequency -3dB point.
-branadic- -
Hello branadic,
I think it would be better for this type of high-pass to use 2 same capacitor values for C2 + C3 (e.g. 5.1 uF)
The other possibilty would be to increase R9 + R10 (they should have a factor 2:1).
with best regards
Andreas
-
So increasing the resistors to 2.7k and 5.4k is another possibility beside increasing the 4.7µF capacitors to 5.1µF by paralleling 390nF to them.
After both modifications are found there is no room left for further improvement of this amplifier.
-branadic-
-
Hello,
ok that seems to be the better solution.
5.1 uF capacitors are rare as hens teeth.
With best regards
Andreas
-
The inverting amplifier at the input is not the lowest possible input version. Usually the non-inverting amplifier is slightly lower in noise. In addition it can avoid most of the noise of the 100 Ohms at the input.
So the modified version is better / optimized in the transfer function, but lower noise is possible. -
Hello,
in my case the AN83 amplifier is already better to that what I need.
The lowest noise that I want to measure is about 45nV/sqrt(Hz)
So the target spec for me is better than 15nV/sqrt(Hz)
Here we have below 2nV/sqrt(Hz).
So why complain?
With best regards
Andreas