Ultra low noise amplifier

[lucifer]

I thought someone else could find that interesting too.

It's an article from Analog Devices on how to achieve extremely low system noise. Four AD8428 opamps in parallel to reduce noise by a factor of 2, gain 2000, 3.5 MHz bandwidth. Looks like a nice configuration for an ultra low noise preamp.

Well.. the price is quite steep even for a single opamp.

Original link of article: http://www.analog.com/library/analog.../nanovolt.html

[homefire]

I guess the Noise is reduced by the Averaging of the noise ?

[lucifer]

I guess the Noise is reduced by the Averaging of the noise ?

Yes, that's right. The noise of each unit is uncorrelated to the noise of the other units while the input signal is the same for all units.

[detectormods]

The noise problem is not in the preamplifier itself but what resistive losses that are placed in front of the low noise preamplifier, any resistance degrades the noise figure, capacitance effects the slew rate. The trick to getting around the noise issues is to build a low noise preamplifier directly inside the coil. Using ceramic board with conductive printing it would not be large enought to be detected, energy can be supplied by a small ""energy thief winding" inside the coil, a DD or concentric design would allow direct connection to the receiver coil without suffering back emf destruction , a bav99 diode pair across the rx coil would save the input from getting hit by back emf. Opamps and low noise transistors can be obtained in die format as to greatly reduce any metallic reactive mass in the coil assembly. You could put the small assembly in a ferrite sleeve so it acts like an electromagnetic coffin.

[Davor]

The ultimate noise source in metal detecting is a coil resistance. 1nv/sqrt(Hz) comes from a 50 ohm resistor. 1/2 of that noise would require a coil at 12.5ohm resistance, and that borders with practical coil values in IB, but there is some room for improvement with PI.

Traditional IB front ends are plagued with quasi-resonant solutions aimed to transforming the coil voltage (and resistance) to noise values of the olden op amps. With a seriously low noise op amp, traditional front end configurations are sub-optimal.

[Skippy]

There's no need to ultra-miniaturize the pre-amp inside a search-coil. Just the usual SMD stuff will make something plenty small enough to fit. It doesn't matter that the search coil sees it, (within reason) as long as it doesn't move around and doesn't have much iron in it. After all, the XP Deus coil has a PCB and a LiPo foil-pouch cell inside it, and I've seen an old C-Scope machine that had loads of electronics inside it.

Re: the noise. There are several sources of noise at the front-end: the RX coil winding resistance; the pre-amp's thermal noise; any gain-setting resistors associated with the pre-amp; plus static noise, which I understand is influenced by the screening. While it would appear obvious to reduce any and all of these, there's a balance needs to be found. There's no point having a very-low resistance coil-winding if the pre-amp is noisy, etc. So manufacturers try and balance it out so that they all contribute equally. Coil-winding resistance can easily be changed (use thicker wire) so the pre-amp is usually the place where the money is spent. If your complete pre-amp is "like a 50 Ohm resistor", then there's little point making the coil have R less than 35 Ohms , it's a waste of copper.
It's slightly more complex than that, as resistors and lengths of wire just generate thermal noise, whereas semiconductors generate 1/f (flicker) noise as well as thermal noise.

Putting an amp in the coil isn't really needed from a noise point of view, the cable only adds 1 Ohm resistance, but it does reduce problems like microphony in the cable, and tolerates poor contact on connectors better, too. Power for the pre-amp is not an issue, there's lots of ways it can be achieved, from simply adding an extra wire in the cable, to things such as sending the signal down the power line (similar to how satellite dish LNB's have power / signal / data / AC tones all sharing the same wire). You can also make clever use of screens, eg. if your TX screen is at +10 Volts, and your RX screen is at 0 Volts, then you've got yourself a 10 V power supply.

[Nupi]

Putting an amp in the coil isn't really needed from a noise point of view, the cable only adds 1 Ohm resistance, but it does reduce problems like microphony in the cable, and tolerates poor contact on connectors better, too. Power for the pre-amp is not an issue, there's lots of ways it can be achieved, from simply adding an extra wire in the cable, to things such as sending the signal down the power line (similar to how satellite dish LNB's have power / signal / data / AC tones all sharing the same wire). You can also make clever use of screens, eg. if your TX screen is at +10 Volts, and your RX screen is at 0 Volts, then you've got yourself a 10 V power supply.

I may be wrong but I thought more to amplify the source signal/voltage for that blended with the receives interference signals from the cabel itself.
It also works well for impedance adjustment on the coil and to the next pre-amplifier, if you are using one .
Also for the same temperature change as the coil itself.
I wonder how do you wil shielding the TX to ground when you put 10 Volt on it.
In my experience you can, depending on the oscillator you use, feed the amplifier with + voltage from the oscillator coil itself without any problems, after filtering.

[Skippy]

Quote:"I wonder how do you shield the TX to ground when you put 10 Volt on it."
Provided there is no contact between the '+10V' and the RX ground, there shouldn't be a problem.
I think it's quite common to have a +ve supply rail as the 'TX cold', with the switching transistor being an npn type to ground. As an example, my Fisher F2 (and hence the F4 also) has about 5 Volts difference between TX cold and RX ground. True, it's a low-end machine with a Bounty-Hunter coil, but it works OK.
In the Fisher F70 / F75 machine, the coil cable comprises:
* Single-core screened cable for TX hot (core) and cold (screen)
* Twin-core twisted-pair with overall screen. Cores are RX ground and RX hot. The screen is connected to circuit ground inside the detector, but not in the coil. In the coil, it's only connected to the coil case shield.
There is no direct connection between TX and RX inside the coil (though there maybe some not-really-essential nulling parts, like a trimmer-cap).
So in this case, it's viable to have the TX shield at a +ve voltage, like +5V, and power an amplifier.

You would probably want to have a negative supply rail, there's lots of ways to do that, of course. A switched capacitor chip; A 555 oscillator and inverting rectifier on it's output; a proper switched-mode inverter voltage regulator. Or for keeping things clean, maybe an optical arrangement: an LED shining on 2 or 3 photo-diodes to give an isolated 1 Volt supply. You don't need much -ve rail, the signal itself will only be 200 mV, I guess.
There are a few integrated opto-devices that do this function, intended for MOSFET/ triac / IGBT gate drive. Panasonic, Toshiba make some, but the current output is flimsy, less than 50 uA.
This sort of thing:
http://eu.mouser.com/ProductDetail/P...ices/APV2111V/
http://eu.mouser.com/ProductDetail/Toshiba/TLP190BUCF/

Gotta say, I'm curious as to how efficient an optical isolated supply can be. Maybe 5% efficiency? Pretty rubbish, but if you're only wanting 2 Volts at 0.5 mA it doesn't matter so much. Might have a play some time. Infra-red LED + BPW34 ? Sounds like a start. Though I will try a modern high efficiency white LED, too, even though it's not spectrally matched to the p.d, and it has double the forwards voltage drop.

Update: a quick test shows white LED + BPW34 is 0.5% efficient. Hmmm, I think the diode is the culprit here. Proper modern photovoltaic cells are needed, the BPW34 is the weak link here.

[Skippy]

I've seen ultra-low noise audio pre-amp circuits (using many paralleled transistors) that are down at the "15 Ohm equivalent" level, and it's certainly practical to make RX winding resistances below 10 ohms. My feeling is that the circuit noise in modern high-end machines is not the problem, it is other issues like the ground signal, which just dwarfs the weakest target signals, and multi-freq detectors still don't get you that much more target sensitivity that circuit noise is a problem.

Opto power supply update: Efficiency of 1.2% achieved with very good white LED and BPW34 photodiode. Sadly I have no decent PV cells, only little ones from bike speed-computers, garden lights and other poor quality consumer stuff.

[detectormods]

With PI Minelab Coil resistance is half an ohm, cable is around half an ohm if you divide the resistance of the inner and outer shield, 1st loss is diodes across the rx input, second loss is N Fet for rx switching of 3 ohms then P of around 1 ohm, dual gate mosfet swithing input to ground is a 5 pf capacitor, the AD797 is referenced to ground with 10 ohms, feedback resistor is 330 or 470 and should be counted in parrelel with the reference on the inverting input. I have experimentedc getting the resistive losses down to 3 ohms and reference to 5 ohms, feedback to 150-270 ohms and use a negative impedance feedback of the opamp to further reduce noise. Does it make a difference? Yes it does, another good improvement is using a servo loop with the correct time constant to alleviate the integrators from being pumped by jitter and with correct filtering remove overshoot on the fast opamps and reduce high frequency noise, these improvements are substantial.

[Ferric Toes]

What improvement in detection range have you measured in field conditions, when incorporating these modifications?

Eric.

[crane]

With PI Minelab Coil resistance is half an ohm, cable is around half an ohm if you divide the resistance of the inner and outer shield, 1st loss is diodes across the rx input, second loss is N Fet for rx switching of 3 ohms then P of around 1 ohm, dual gate mosfet swithing input to ground is a 5 pf capacitor, the AD797 is referenced to ground with 10 ohms, feedback resistor is 330 or 470 and should be counted in parrelel with the reference on the inverting input. I have experimentedc getting the resistive losses down to 3 ohms and reference to 5 ohms, feedback to 150-270 ohms and use a negative impedance feedback of the opamp to further reduce noise. Does it make a difference? Yes it does, another good improvement is using a servo loop with the correct time constant to alleviate the integrators from being pumped by jitter and with correct filtering remove overshoot on the fast opamps and reduce high frequency noise, these improvements are substantial.

Woody,
A Minelab mono coil plus cable resistance is less than 0.5 ohms provided we use an in spec coil, some third party coils go as high as 1 ohm but don't perform well.
As I think Mechanic pointed out, we can wind the gains up to maximum on the later models in a noise free environment and not hear any significant circuit noise, none that could possibly harm performance.

We can then fit a DD with aprox 20 ohms RX coil resistance and not hear any difference whatsoever. Minelab apply higher gains than anyone else so the limiting factor is the ground signal and no amount of fiddling with the RX circuit will ever fix this.

And Woody, by "noise free environment" I mean noise that occurs at the detector's fundamental operating frequency, not just all or any noise you can measure at the front end.

[detectormods]

Actually there is a net gain that can be obtained by altering the time delay on the receiver turn on time, yes we all know that ground signal elimination is paramount to negate while retaining target signals. There is a sweet spot in altering the tx on time, the rx off time and the rx on time and with this goes changing the gain profile of the detector within these variables. One of the biggest problems with most current Pi detectors is the high slew rate non frequency limited rx front ends that allow signals in way past the relevant passband for low frequency signals. Nearly every attempt to remove these signals entering the detector and causing intermodulation results in slew rate limiting and effects the detectors performance, even to the point that it will not ground balance. I found that most intrinsic noise in these PI detectors is the noise generated in the N Mosfet on the input switching circuit but this can be partially cured by changing the fet drive in relation to real ground, ie not two diode junctions above ground. To digress, A lot more gain can be used as well as shorter off times in non reactive mediums such as pipe clay, i actually tried this with the rx off time set to 12 us and input gain close on 70 times. It cannot be done with a standard detector unless it has a redesigned input stage and alternative clamping is used as not to swamp the integrator stage. At present standard detectors have to be "fiddled with " to obtain these extra-ordinary results. PS There is a lot of Gold in pipe clay, especially very hard to detect filament and specimens.

[green]

With PI Minelab Coil resistance is half an ohm, cable is around half an ohm if you divide the resistance of the inner and outer shield, 1st loss is diodes across the rx input, second loss is N Fet for rx switching of 3 ohms then P of around 1 ohm, dual gate mosfet swithing input to ground is a 5 pf capacitor, the AD797 is referenced to ground with 10 ohms, feedback resistor is 330 or 470 and should be counted in parrelel with the reference on the inverting input. I have experimentedc getting the resistive losses down to 3 ohms and reference to 5 ohms, feedback to 150-270 ohms and use a negative impedance feedback of the opamp to further reduce noise. Does it make a difference? Yes it does, another good improvement is using a servo loop with the correct time constant to alleviate the integrators from being pumped by jitter and with correct filtering remove overshoot on the fast opamps and reduce high frequency noise, these improvements are substantial.

[another good improvement is using a servo loop with the correct time constant to alleviate the integrators from being pumped by jitter]

Sampling with the integrator is causing some noise that I don't understand. I'm trying to learn. Could you explain the above statement little more?

[detectormods]

[another good improvement is using a servo loop with the correct time constant to alleviate the integrators from being pumped by jitter]

Sampling with the integrator is causing some noise that I don't understand. I'm trying to learn. Could you explain the above statement little more?

I am not going to go into great detail as it is just giving away knowledge but... even small amounts of electrical noise can cause the integrator capacitors to be pumped up to a voltage that is greater than the voltage derived from a real target and thus obscures the target information. Most detectors of a modern vintage take care of this in the ADC section after the horse has bolted, it is much better to catch it in early stages of the detector and take care of it, you can put a capacitor accross the 1st stage FB resistor to the inverting input of the opamp or base of the transistor in the mirror of a differential type discrete preamplifier. Just be careful not to dampen the slew rate or sensitivity will suffer. You can remove most noise in a wide band system by incorporating a 50 or 60hz notch filter type servo to give lots of negative feedback at the mains frequency of interest, same can be done as a bandpass filter to remove am radio stations. If you want to remove mains the filter will need to be very sharp and i found it better to be done over 2 or 3 gain stages, be careful with phase changes as you can end up with an oscillator.