PI noise

[green]

I cannot help thinking that you're chasing a red herring here.
If you look at the first scope image (the noisy one), and then look at the third one (where the first stage of the preamp is removed), it looks like there may be a 20x difference in the noise level.
Can you do another scope measurement for the first (noisy) waveform, but with the scope set to 100mV/div?

You are looking at PI_noise_4. I added pictures to show schematic, when I did scope shots PI_noise_4 reply#9 I made an error on the second picture which I corrected in reply#10. Look at first two pictures in PI_noise_6, the first picture bypasses the sampling integrator with a gain of one. The second includes the sampling integrator with a gain of one, noise about four times higher. Why does sampling cause higher noise? Some of the pictures in PI_noise_4 reply#9 were included to answer reply#8

[green]

...hmm you seem to be just connecting a chain of amplifiers and expecting the noise get lower ... well with any amplifier the noise just gets higher LOL. :-) An acceptable practice in the PI signal chain is to use a differential integrator ( see below ) .. maybe you could start with what works .. then make it better. Just a suggestion.

[ATTACH]36572[/ATTACH]

I started with a 2C integrator, #1 of the thread. The 2C integrator doesn't cancel EF as good as the 1C integrator for me so I have been using the 1C integrator. Had the same problem with the 2C integrator, noise higher than it is with the integrator bypassed. I'm not expecting the amplifiers to lower the noise, want to know why the integrator with a gain of one increases the noise.

[moodz]

I started with a 2C integrator, #1 of the thread. The 2C integrator doesn't cancel EF as good as the 1C integrator for me so I have been using the 1C integrator. Had the same problem with the 2C integrator, noise higher than it is with the integrator bypassed. I'm not expecting the amplifiers to lower the noise, want to know why the integrator with a gain of one increases the noise.

You are powered off battery and have a bypass capacitor across the power pins of each opamp with very short leads ( preferably +VE is fed with 10 ohm or so resistor ?

[Ferric Toes]

I started with a 2C integrator, #1 of the thread. The 2C integrator doesn't cancel EF as good as the 1C integrator for me so I have been using the 1C integrator. Had the same problem with the 2C integrator, noise higher than it is with the integrator bypassed. I'm not expecting the amplifiers to lower the noise, want to know why the integrator with a gain of one increases the noise.

I have found that a 2C integrator does not give as good an EF cancellation as a 1C integrator even if the capacitors are accurately matched and 1% resistors used on both sides. I also put a balancing 200R trimmer ahead of the gates to match the ON resistance. The best EF cancellation is to have a cross coupled differential output stage with a gain of 1x after the preamp which then feeds the two gates and then into a 1C integrator. The 4066 need to be upgraded to a device with better matching between gates otherwise the small differences in ON resistance will limit the cancellation. If you want to keep it, use a balancing trimmer again.
To make balancing easier, I inject a sine wave of about 4Hz into the preamp input with a summing resistor (220K in my case) and scope the integrator output. Adjust trimmer for zero movement of the base line. I use an analog scope 10mV/cm and dc coupled.

Use a NE5534A for the preamp and its dual, the NE5532, for the differential stage. The 'A' version of the 5534 has lower lf noise. Even better for the preamp is the LME49990.

Eric.

[moodz]

yup .. interesting about the 1C integrator .. I would have thought you could compensated for 'unbalance' by varying the sample duty cycles +/- x nanoseconds. The LME49990 is nice but apparently going end of life / production.

[Ferric Toes]

yup .. interesting about the 1C integrator .. I would have thought you could compensated for 'unbalance' by varying the sample duty cycles +/- x nanoseconds. The LME49990 is nice but apparently going end of life / production.

You can, but I don't think that is possible with my current pulse generator. I need to vary the sample pulse widths equally for other reasons. Is there a 'better' replacement for the LME49990 in view of it's going out of production?

Eric.

[Mechanic]

HI Eric,


I have tried several lme9990 and found that they all take quite a bit longer to settle(think it was up to 5 or 7us) than the AD797. Though a mate of mine was able to use the lme49990 and get it to settle earlier than I could with the same input impedance and gain, go figure! Though he was using a different coil...
When I did noise tests, I could not see any notable difference between the 797 and the lme. Same goes for using an lm394/5534 preamp. This settles as fast as the 797 and there seems to be no notable output noise difference. Though one thing I am still to try, from one of my initial tests, it looks as though the 797 is not as sensitive as the lm394/5534 preamp, however some other things were set up differently. Currently changing back those differences and will test on the weekend. I will know more after the weekend..

Cheers Mick

[6666]

This is interesting, is a 1C integrator also known as a single ended integrator ?
and how does it cancel EF ?

[green]

You are powered off battery and have a bypass capacitor across the power pins of each opamp with very short leads ( preferably +VE is fed with 10 ohm or so resistor ?

The circuits are battery powered. I have at least a .1u and a 1u on each preamp power pin to common. Hadn't tried the series resistors so I added them this morning, no difference.

[green]

This is interesting, is a 1C integrator also known as a single ended integrator ?
and how does it cancel EF ?

Don't know what all it's called. I removed the inverter and the inverted input from the schematic since I wasn't taking an EF sample for the noise test. A schematic I posted awhile back including the inverter.

[moodz]

yup .. interesting about the 1C integrator .. I would have thought you could compensated for 'unbalance' by varying the sample duty cycles +/- x nanoseconds. The LME49990 is nice but apparently going end of life / production.

OPA1611 OPA1612

[Ferric Toes]

This is interesting, is a 1C integrator also known as a single ended integrator ?
and how does it cancel EF ?

The cross coupled stage after the preamp gives two outputs, one normal and the other inverted. That means that after gating, you can sum them into a single ended integrator and common mode signals, such as EF, cancel. You can use just a simple inverter stage but I believe the cross coupled system gives better accuracy.
Using the NE5532, I made all resistors 1K, except for the 75R which I believe was 180R (not critical). If you want some gain make Rf variable or different value. Exact symmetry is maintained as gain is varied. You could make the whole preamp like this but I preferred an independent front end. My present arrangement is gain 10x for front end and 47x for differential. You don't need the 49.9R resistors of course and would go straight to your integrator input gates with resistors summed on the inverting input of the integrator, which has a single R and C in the feedback.


Eric.

[Monolith]

NOISE SEEN DYNAMICALLY

I like to look at the noise dynamically. Here is the output of the integrator, for a Nickel target pendulum swinging at 13cm above the coil.
The coil is 245mm diameter, but I find it better to define the distance as just over 1r, or half the diameter.

4 different sample windows are taken, the red and the yellow traces are window 2 and 4. The blue trace is the mathematical function of the scope, channel 1 minus channel 2.

The noise:
noise level on the board GND, about 10mV PP at frequencies of 2.5MHz, 2MHz, 4.1MHz. I don't know what I can do about that.
The most worrysome noises were at 50mV, PP, 0.22 Hz and 20mV, PP, 6.3Hz, so I increased the frequency of the high pass filter to 3.2Hz. Now I will look again in detail at the output traces to see if there is any improvement.

The preamp is the LME4999. I am now looking for another opamp with good 1f noise figure.

[green]

Thanks for the replies. Lowering the noise before sampling makes sense. My question is should sampling increase low frequency noise(less than 15Hz). At work we filtered the signal before sampling to prevent aliasing. I haven't found any thing on what to expect if I sample resistor noise with out pre filtering. Maybe there is another cause I'm missing, sampling frequency jitter or some thing else? If I knew sampling resistor noise without pre filtering caused a four times increase in low frequency noise I could quite thinking about it.

[6666]

The cross coupled stage after the preamp gives two outputs, one normal and the other inverted. That means that after gating, you can sum them into a single ended integrator and common mode signals, such as EF, cancel. You can use just a simple inverter stage but I believe the cross coupled system gives better accuracy.
Using the NE5532, I made all resistors 1K, except for the 75R which I believe was 180R (not critical). If you want some gain make Rf variable or different value. Exact symmetry is maintained as gain is varied. You could make the whole preamp like this but I preferred an independent front end. My present arrangement is gain 10x for front end and 47x for differential. You don't need the 49.9R resistors of course and would go straight to your integrator input gates with resistors summed on the inverting input of the integrator, which has a single R and C in the feedback.
[ATTACH]36575[/ATTACH]

Eric.

Thank you Eric ,its good to learn something.