Amplitude noise in LC oscillators

[Teleno]

Not much info on this subject to be found in Google, so I decided to give this forum a try.

My project involves the "loading effect" of a metallic object upon a free-running LC oscillator. This means that the amplitude - and to a lesser extent the frequency - of the oscillator changes as the object gets closer to the inductance.

My oscillator produces a 40Vpp sine wave. Before applying the peak detector I subtract 37V DC so the peak is 3V maximum. This is like "zooming in" into the peaks, which allows to detect small amplitude variations of the order of a millivolt in the 40V range.

You would assume that the amplitude of the oscillator (the envelope after peak detection) would be constant if the power supply is well regulated. Well, this is not the case, instead, the amplitude behaves like a "random walk" (Brownian motion) drifting up and down in what appears to be huge "pink noise" also known as "flicker noise" or "1/f" noise. In this case the drift is in the order of the 100's of millivolts, so large it buries the signal.

Google taught me that flicker noise is the integral of white noise, and since the LC circuit is in fact an integrator, any noise injected by the power supply or the components of the oscillator will result in 1/f noise.

I've tried to remedy this by using a battery as a power supply + big capacitor, since regulators are noisy. The drift is a bit less but still remains.

Any ideas on how to minimize this annoying effect?

Attached: LTSpice simulation. I've added the white noise source BV to show what the effect looks like (plot V(peak)) as it happens in the real circuit without BV.

 

[kubeek]

That is a real graph or simulated? What in the hell are you doing that the deflection is 300 kilometers?
Are you sure that the noise you are injecting into the simulation is what your power supply actually produces? Could you use your power supply as a pre-regulator and use better filtereing and an LDO to get the noise floor way down?
AFAIK, usually oscillators are tied by some non-linear impedance such as the famous H&P's light bulb, so that the amplitude relies less on power supply rails and clipping and is more predictable.

Scratch that, this is what I get with maximum time step of 1ns. I gues your result comes mostly from some truncation error in the solver.

 

[unclejed613]

in your sim, you added the noise to the power supply, instead of putting a regulator there. this does absolutely nothing to track down the source of the noise, because the source of the noise in this case IS the noise source. however, there are potential sources within some regulators that could be the source of your noise.

one of the noise sources in power supplies is noise in the voltage reference, which in many cases is a zener diode. a zener has two modes of operation (determined by the actual device materials). in zener mode, there is a small amount of noise generated in the junction, this mode of operation is dominant in zeners with breakdown voltages less than 6V. zeners above 6V operate in avalanche mode, and this generates a lot of noise in the junction. either type of zener generates noise, but the avalanche mode zener generates a lot more noise. that is the reason that in most regulators that use zeners as a voltage reference, there's a large value electrolytic cap across the zener, to kill the junction noise.

also, if the regulator is a 3-terminal regulator like a 7805, most designers bury the regulator in electrolytic caps (there's usually a 10uF cap at the input, and one at the output very close to the regulator itself, plus anywhere from 200-4700 uf on the unregulated side, and 50-1000uF on the output side, further away from the regulator).

 

[Teleno]

LTSpice without a noise source is not able to reproduce the phenomenon. The real circuit shows this noise even when powered by a battery, which has the lowest noise figure of any power source.

The function of the white noise source BV in LTSpice is to prove that white noise gets converted to 1/f noise by the oscillator, something I suspected but wanted to confirm.

The true cause of the flicker, I suspect, is the sum (integral) over each half-period of the noise caused by the driving transistor that's conducting during that half period.

P.S. The graph shows real measures. A "300 Km." swing translates into roughly 300mV. I didn't bother to change the units, sorry.

This effect is a project killer, dammit!

 

[kubeek]

P.S. The graph shows real measures. A "300 Km." swing translates into roughly 300mV. I didn't bother to change the units, sorry.

300mV of what? I thought you were measuring some distance or something. Can you please label what each axis actually is and how it relates to the circuit?
Have you tried changing the transistors for some lower noise ones? I am no expert in this, but measuring the amplitude seems way more complicated to do accurately than measuring change of frequency. Yes the base frequency would depend on temperature, but I think the amplitude will be even more dependant.

 

[unclejed613]

AFAIK, usually oscillators are tied by some non-linear impedance such as the famous H&P's light bulb, so that the amplitude relies less on power supply rails and clipping and is more predictable.

that would negate what the OP seems to be trying to do, he's using the output amplitude to measure the proximity to an external capacitance.

The real circuit shows this noise even when powered by a battery, which has the lowest noise figure of any power source.

actually a chemical process such as that in a battery is a noise source. the noise is generated by the random nature of the recombination of individual ions and electrons at the battery electrodes. the voltage of this noise is lower than what you would get with a zener, but it is still there nonetheless.

 

[Teleno]

actually a chemical process such as that in a battery is a noise source.

Well, everything is a noise source, but chemical power sources are the quietest. I've done my homework and checked the facts. NiCd are the quietest of all. Not that it helps suppress this noise though.

 

[Teleno]

that would negate what the OP seems to be trying to do, he's using the output amplitude to measure the proximity to an external capacitance.

Not a capacitance, but an inductive coupling. Look at the object to be detected as the secondary of a transformer. The circuit does its job at short range very well, the problem being the high noise level that greatly reduces its usefulness at medium range and beyond.

 

[kubeek]

And that is at laboratory conditions, what stray magnetic fields are going to do to such free running oscillator is probably going to be much worse. Maybe you could average out the amplitude, maybe some other approach like fixed frequency driver and measuring impedance could do better. I think that a lock-in amplifier measuring the impedance should be more robust against noise, but I have never practically tried it.

 

[unclejed613]

Not a capacitance, but an inductive coupling. Look at the object to be detected as the secondary of a transformer. The circuit does its job at short range very well, the problem being the high noise level that greatly reduces its usefulness at medium range and beyond.

you might not be able to use it beyond a certain distance. the electromagnetic field around L1 is governed by the inverse square law, so there's going to be a finite distance from L1 where an object won't have an effect on L1.

 

[Teleno]

you might not be able to use it beyond a certain distance. the electromagnetic field around L1 is governed by the inverse square law, so there's going to be a finite distance from L1 where an object won't have an effect on L1.

The limit is always the signal to noise ratio, if I get rid of this noise the detection distance increases accordingly.

 

[BobW]

I can see a couple of problems with the circuit.
1. By subtracting a fixed 37 volts, the relative noise voltage increases significantly. If the noise voltage is x then x/3 is much greater than x/40.
2. Using a current source and a resistor as a means to subtract the fixed voltage is a potential noise source. Have you considered using a 37V zener? Yes, zeners are noisy too, but it would be interesting to see a comparison.

What you are doing strikes me as a strange way to detect the effect of nearby metal. There are a number of proven methods.

 

[Beau Schwabe]

the electromagnetic field around L1 is governed by the inverse square law, so there's going to be a finite distance from L1 where an object won't have an effect on L1

Correct, but I happen to hold the record at R.M.I.S.C. (Rock Mountain Information Security Conference) for correctly reading a passive RFID tag at a distance of 25 feet. You just have to know where to look in the noise floor. Over sampling techniques are a way to increase the signal to noise ratio by a process known as Ensemble averaging. Basically your SN ratio is equal to the square root of the number of samples you acquire.

All that aside, I suspect much of the "noise" is simply from external RF interference making its way into the coil.

Perhaps I missed it, but I was not able to find a schematic that the OP is using. That might help my assessment of what might be going on.

EDIT: Ahh I see... it's in a SPICE file. Would someone be kind enough to post an actual schematic please?

 

[BobW]

Would someone be kind enough to post an actual schematic please?

 

[Beau Schwabe]

L1 and C2 are essentially tuned to about 577kHz ... There is plenty of noise there. What your circuit may be doing is something of an Active Antenna effect. Even though your driving the LC you wouldn't think it can act as a receiver, but that's exactly what some Active Antenna's do. Any wires leading into or out of the LC will exaggerate the effect.

Believe it or not, in your oscillator circuit I see a path that could act like a decent receiver..... I use a similar one transistor version in some of my receiver designs. What you have effectively is a symmetry of my receiver on each rail. See attachment for similarities, I believe the Base follower receiver is what you essentially have.

 

[Teleno]

L1 and C2 are essentially tuned to about 577kHz ... There is plenty of noise there. What your circuit may be doing is something of an Active Antenna effect. Even though your driving the LC you wouldn't think it can act as a receiver, but that's exactly what some Active Antenna's do. Any wires leading into or out of the LC will exaggerate the effect.

Believe it or not, in your oscillator circuit I see a path that could act like a decent receiver..... I use a similar one transistor version in some of my receiver designs. What you have effectively is a symmetry of my receiver on each rail. See attachment for similarities, I believe the Base follower receiver is what you essentially have.

I appreciate your comments. By the way, the LTSpice schematic is attached to the first post.

I think if the noise were due to the antenna effect it would be periodic rather than a random walk. The sine wave looks pretty clean on the scope, but here's we're talking tens of millivolts on top of a 40V peak, so radio interference might as well be hiding there.

Let me expand on the details:

- The circuit is a serial LC resonator around L1 and C3. Resonant frequency is 81KHz.
- L1 is shielded and the shield is grounded.
- C2 is the parasitic capacitance of the coil. Impossible to remove.
- R1 is the resistance of the coil. Rseries is the ESR of capacitor C3

Apart from shielding, which I already have done, what other measures can be taken to avoid the antenna effect?

 

[Teleno]

I can see a couple of problems with the circuit.
1. By subtracting a fixed 37 volts, the relative noise voltage increases significantly. If the noise voltage is x then x/3 is much greater than x/40.

There is a reason. I'm not interested in the oscillator signal itself, but in measuring tiny changes in the amplitude of the oscillator induced by a nearby metallic object. The intention is to maximize the sensitivity, and that's where noise gets in the way.

2. Using a current source and a resistor as a means to subtract the fixed voltage is a potential noise source. Have you considered using a 37V zener? Yes, zeners are noisy too, but it would be interesting to see a comparison.

Emitter degeneration resistors (R11 and R10) drastically reduce the noise of the current mirror, much lower than a zener's. Unlike a zener, the current mirror is adjustable, keep in mind that the applitude of an oscillator is hard to predict, changes with temperature, etc. An MCU can modify the current dynamically to maintain the working point.

What you are doing strikes me as a strange way to detect the effect of nearby metal. There are a number of proven methods.

This method is very common and widely known, it's called "loading effect". TI even has chips implementing it (see for example LDC1000 **broken link removed**) but they are low power and very short range. What I'm trying to do is just to increase the power and the sensitivity.

 

[Teleno]

Over sampling techniques are a way to increase the signal to noise ratio by a process known as Ensemble averaging. Basically your SN ratio is equal to the square root of the number of samples you acquire.

This is how I'm treating the readings right now. I oversample and decimate to obtain 13 bits from the 10 bit ADC converter of an Atmega328P. I deal with the flicker noise by differentiating to remove DC (basically subtracting two succesive samples) and band-pass filtering. It's quite sensitive and works relatively well but it requires moving the coil or the object. Static operation is only possible at a much reduced sensitivity. If I only could reduce this flicker it would be perfect.

 

[Beau Schwabe]

A low pass filter may be all you need .... place a 22pF cap from the base of Q7 to ground. A 22pF in combination with the existing 41k resistor forms a low pass filter at about 176kHz .... roughly twice your 81kHz so that should filter through just fine, but anything above 176kHz should be attenuated.

Reference Calculator:
https://www.trance-cat.com/electrical-circuit-calculators/en/rc-low-pass-filter-calculator.php

 

[unclejed613]

Correct, but I happen to hold the record at R.M.I.S.C. (Rock Mountain Information Security Conference) for correctly reading a passive RFID tag at a distance of 25 feet. You just have to know where to look in the noise floor. Over sampling techniques are a way to increase the signal to noise ratio by a process known as Ensemble averaging. Basically your SN ratio is equal to the square root of the number of samples you acquire.

All that aside, I suspect much of the "noise" is simply from external RF interference making its way into the coil.

Perhaps I missed it, but I was not able to find a schematic that the OP is using. That might help my assessment of what might be going on.

EDIT: Ahh I see... it's in a SPICE file. Would someone be kind enough to post an actual schematic please?

the OP is trying to detect the presence of metallic objects by having the object act like a shorted turn and changing the voltage and frequency of the oscillator.

one problem with this question, is the OP has not mentioned what the detection distance is that he's trying to achieve, or the dimensions of L1, so it's impossible to make any solid recommendations, even for what possible noise sources to eliminate.