Induction Balance Stuff - Single/Multi Frequency Response, GB, Disc, Measurements, Ideas, Fun, etc.

[Aziz]

Hi all,

with e_n = 0.3 nV/rt Hz ULNA (ultra low noise amplifier), we should be able to detect and resolve sub nano Volt RX signals (actually 0.35 nV with Gain=90, EMA-Alpha = 0.05).

[Aziz]

BTW,

the IB nulling by the mixer and the line output noise for IB nulling signal is really neglectable. The e_n gets slightly worse (a few pV / rt Hz only) but it won't affect our RX signals.
The dominating noise source is still the preamp. And may be the EMI.

We can limit the residual IB mismatch spec to lower this noise contribution further.

[Aziz]

Hi all,

I will be heading to Berlin tomorrow. 5 days. 650 km with a slow train. A 12 hour trip.
All devs will always delay.

(All the international top spies have a meeting there. We have a top secret meeting near the "Russisches Haus in Berlin", which has been previously closed due to drones in Leipzig Airport recently). (Psssst! Top secret! )

See you.. or later after being arrested..

[Aziz]

Hi all,

I am really not happy with my discrete transistor amplifier specified for 3.7 V battery voltage. To much distortion. Problems arising if the battery voltage varies between 3 V (almost empty) and 4.5 V. I give it up.

I will look at the old PNP + op-amp designs with 9 V battery voltage.
Transistor collector current Ic around 8 - 12 mA. We can chose either single PNP ZTX951 or two Vbe matched PNP for lower noise.
Aziz

[Aziz]

Can we reach that damn fkn level of en = 0.3 nV/√Hz?
Will we have the lowest-noise front-end in the history of metal detection?


The short answer is yes.

I am still optimizing the pre-amp circuit.

[Aziz]

This is it.

Schematics & Noise performance of the pre-amp:


Input impedance:

Full LTspice files included.

[Aziz]

Hi all,

by paralleling two op-amps, the op-amp noise of NE5532 is beeing improved by sqrt(2) factor.
5 nV/rt Hz -> 3.54 nV/rt Hz
At gain 100, it's contribution to the input is 3.54 nV / 100 -> 35.4 pV/rt Hz. So it is completely irrelevant. The dominating noise source is still the pnp input transistor.

And we have effectively doubled the load resistance of 100 Ohm (Rf2) into 200 Ohm compared to a single op-amp. Or doubled the current drive capability of the single op-amp.

What can we do, if the distortion is too much or want to reduce the gain without lowering the load resistance further (Rf2)?
We could slightly increase Rf1 from 1 Ohm to 2.2 Ohm or higher. But this will cause much more noise and we will never get to the 0.3 nV/rt Hz level.

The other option without losing noise performance (keeping Rf1 = 1 Ohm):
We have to upgrade the NE5532 to an other dual op-amp.
- With more drive current capability or
- Rail-to-Rail output.

Rail-to-Rail output option:
As we don't have the big 2.0 - 2.5 V head room to the power supply rails (0 V, 9 V) with the rail-to-rail output option, we could increase the total gain of the amplifier. So we can rise the load resistor Rf2 (for instance doubling it). At the same time, we could voltage divide the output of the amplifier. This voltage divider can be realised in the feedback loop (splitting Rf2 resistor into two Rf2a, Rf2b resistors). There we have it.

This modification is shown here:



I have doubled the total gain (total Rf2 = 100+100 Ohm) and divided the output voltage by two. If I don't change the op-amp, I am limitted to half dynamic range only.
So the load resistance Rf2 is being rised to 200 Ohms and the op-amp won't hopefully clip or distort the signal at the output. You see, we would require rail-to-rail output op-amp here.

Or we could increase the battery voltage. Hell, don't do it. The power consumption... It will turn into a big heater ..

Aziz

[Aziz]

I have been looking for a better NE5532 spice model (which is simulating its noise and dynamic current draw too). I have found one, but the .noise analysis isn't working properly.
Even the power consumption part isn't working properly.


Anyway, we can use the Excel table for amplifier noise prediction. Or take a similar other improved op-amp spice model.

[Aziz]

Hi all,

we are entering the next level now:
e_n = 0.2 - 0.3 nV/rt Hz

And at the very very KISS-Level.


AI Tip: The Richard Lee's common-base amplifier can be used by impedance matching.
And I was wondering, why my former detector wasn't sensitive enough. I have the answer now.

See AI PDF paper.

This is a break-through moment.

Aziz

[Aziz]

AI Paper: Confidential - Engineering Research Department



pV target signal detection becomes reality soon.
I will revival my Richard Lee's amplifier soon. If I can find the bread board.

And the interesting PDF link to the other implementations and modifications to the Richard Lee's amplifier becomes important (former posting).
Link again: https://www.hifisonix.com/articles/m...it-compendium/

[Aziz]

Holly cow!

The single NPN/PNP ZTX851 / ZTX951 pair with Ic=10 mA each in Richard Lee's Duraglit Special does get directly
e_n = 0.2 nV/rt Hz
With input impedance of approx. 1.4 Ohm.
Mission accomplished.

Vbe matched two NPN, two PNP does even get 0.17 nV/rt Hz.

[Aziz]

Oh man!,

I was really shocked about the simulated "temperature dependency" of the Richard Lee's preamp in LTspice.

With floating power supply rails, where the amplified signal "rides" on these rails, it is difficult to simulate the temperature dependency of the circuit.
After all, I had to switch off any initial operating point definitions (.ic commands) . It worked then.
These are usually required for proper .noise, .ac and .tran time domain analysis.

Now the temperature dependency is really acceptable. The gain varies only appr. 3% with 50 °C temperature difference (0 °C - 50 °C).
Offset voltage? What an offset? We don't need any. We don't want it. We are ignoring it. We have an AC coupled input and output.
Aziz

[Aziz]

Hi all,

I have just checked the impedance matching via series resonance RX coil input. And also the IB nulling (via resistor) with some residual IB-mismatch.
It works very nice. The AI is correct.
The unknown mystery have been solved (my former reduced sensitivity observation).
New limits (en = 0.2 nV/rt Hz) have been broken too.

We can afford to increase the collector currents (Ic). The gain stability and THD improves. Ic can go up to 20 mA or more.
Input impedance decreases with more collector current. I have right now 0.83 Ohm.
The total power consumption will be the collector current.

With RX coil resistance of 1 Ohm: en = 0.185 nV/rt Hz
With RX coil resistance of 2 Ohm: en = 0.227 nV/rt Hz
Either 2x1.5 V standard battery cells or 3x1.2 V NiMH cells or single 3.6/3.7 V Li-Ion cell.
And the circuit is very very KISS.

More later... stay tuned..

[Aziz]

Hi all,

I have increased the collector current to ~20 mA. Single NPN/PNP transistors.
I have determined the dominating noise source for this configuration.
If the source resistance Rs (RX-Coils resistance) is below 0.55 Ohm, the dominating noise sources are the transistors.
If Rs > 0.55 Ohm, the dominating noise source will always be the coil itself (thermal resistor noise of course).
A receive coil with such low resistance Rs will get heavy and unrealistic.
Rs between 2 and 4 Ohms is more realistic. So the RX coils resistance will always dominate the noise.
As a consequence of this: We don't need any paralleling more transistors.

We are limitted by the fkn RX coils resistance. I have reached the max. possible limit now.
(Wait! I can cool down the front-end... with liquid nitrogen.. )
Aziz

[Aziz]

Hi all,

this is the final schematics of the Richard Lee's Ultra-Low-Noise Pre-amp.
Two versions included:
- Pre-amp without RX coil (plain amplifier characteristic without impedance matching)
- Pre-amp with RX coil attached (impedance matched, IB-Nulling, but disabled in the simulation file)

LTspice files also attached.
Cheers