VLF MD with digital signal processing : Bee-Buzz 1

[Carl-NC]

Keep in mind what you call SNR isn't really SNR, it's a relative noise level and its value depends on the depth of the FFT. To get SNR, you have to integrate all those noise bins over the BW of interest. Here's an FFT or the AD6645, a chip I helped design. You can see the noise floor is around -120dB while the SNR is only -75dB. From this, you can calculate the size of the FFT to be 64k. Using a smaller FFT will raise the noise floor but the SNR stays the same.

[Aziz]

Hi Carl,

yep. I should have called it noise floor level instead of SNR.
When I reduce the block size to 128 (instead of 1024), the frequency bins integrate more noise as the bandwitdh of the frequency bins gets greater.


It is at least 10 dB more noise at each frequency bin.
Oh my god!, I can process only 1/3 of the frames (495 of 1500). 2/3 will be discarded as the process-thread can't make it as fast as the IO-thread delivers data.
But this is the old process-thread with 1001 features. The new process-thread with the Goertzel decoders will do that with ease.

Aziz

[Aziz]

Hi friends,

I have lost a lot of time in investigating LTSpice noise calculation error. But I have found the reason finally. You have to set the operating point voltages on critical nodes correctly. If you have (large) capacitors in your circuit, the initial voltage on its node is very very important. If not set correct, you get wrong noise calculations in LTSpice.

BTW, the 0.2 nV/rtHz noise voltage density is very very difficult to achive with the NE 5534A op-amp. No, I'm not going to use other expensive op-amps (1nV/rtHz). This is not necessary.
Under very critical conditions, I can get 0.175 nV/rtHz at gain x1000 with the NE 5534A op-amp.
The gain of 1000 is too much I think. Gain of 200 - 500 would be ok I think. Let's investigate how much gain we really need. And then optimize the pre-amp for the required gain.
Me LTSpice
Cheers

[Aziz]

Hi all,

have a look.
I have adapted and arranged Dick's Ultra-Low-Noise Pre-Amp:
see source http://www.dicks-website.eu/low_nois...rt3/part3.html



This nice pre-amp with gain of 1000 and 9 V single supply operating voltage (9 V block battery), achieves with only 1x NPN (ZTX 815) an input voltage noise density (en)
en of 0.225 nV/rtHz at 1 kHz and
en of 0.214 nV/rtHz at 10 kHz

Not bad. With a cheap NE 5534A op-amp. Even with a 500 Ohm load on the output of the op-amp.
Now the critical conditions, which allow the superior noise specs:
Collector current set to approx. 9 mA
Source resistance is max 0.5 Ohm (RX-Coil with 0.5 Ohm series resistance)
Gain setting resistor RG1 is 2x1 Ohm parallel (=0.5 Ohm). Must be fixed at this value.
ESR of the capacitors must be each max. 20 mOhm.
The real gain setting resistor RG2 can be calculated according to the formula above (must be approx. 500 Ohm)
Bandwidth is well above 100 kHz. Enough for me.
Power consumption approx. 12-13 mA.

No LT1028 or AD797 required at all.


I have versions with 2xNPN and 3xNPN in parallel. These are a bit better. But the power consumption is too much for a 9 V block battery. I like the single transistor version. I don't need to go below en 0.22 nV/rtHz.

Cheers

[Olly]

Very nice, but I'm not sure that an RX coil with such a low series resistance (0.5 Ohms) is practical.

[Aziz]

Very nice, but I'm not sure that an RX coil with such a low series resistance (0.5 Ohms) is practical.

Hi Olly,

no problem. You can of course use high resistance coils too. But it will produce more thermal (resistor) noise.
Look at the table below for source resistor RS (in Ohms) and input voltage noise density en (in nV/rtHz)

RS    en
==========
1    0.243
2    0.276
4    0.331
6    0.380
10   0.460

The figures are tanken from spice simulation. And this pre-amp is still the quitest front-end you can get in the metal detecting technology.
Due to the high gain of 1000, you can use low inductance RX coils, which will have lower resistance.

I see, I have forgotten to put 2 anti-parallel 1N4148 to protect the input of the pre-amp. This is required.
Cheers

[Aziz]

Hi all,

the recently shown pre-amp is an AC-coupled pre-amp. Only IB configurations with seperate TX and RX coil must be used.

What about a DC-coupled pre-amp for pulse induction front-end?
This is good news. With two hand selected and matched (measured) ZTX 815 in a differential configuration with a current mirror and bipolar power supply, a very good noise performance can be achieved.
The NPN pair must be thermally coupled together (glued together) to have the same temperature to minimize temperature drifts. An output offset adjustment circuit is required too. You have to buy 50 or more of ZTX 815 from the same batch, to find the best matched and measured pair.

I think an en of 0.3 to 0.4 nV/rtHz is easily possible with the ZTX 815 pair. But this is a design challenge for enthusiasts. Everybody is invited to make this possible.
Cheers

[Detectorist#1]

Aziz, do you believe that side sources of the noise will be lower than these levels?

[Aziz]

Aziz, do you believe that side sources of the noise will be lower than these levels?

No I don't think. EMI noise can be much higher. It is depending on the location you are detecting and sun and weather activitiy too.
When I look at the FFT spectrum a while, I can even see my neighbors activity. Switching lights on and off and activity of other devices and so on.

But, if you have a quiet location, the signals of interest are usually below the noise floor level of the ADC system (in my case < -120 dB). We have to rise the signal just above the noise floor level of the ADC system. That would be enough.

In my case, the noise floor level of my Sound BlasterX G6 reaches -100 dB at 70 kHz. I would like to go upto 70.. 80 kHz for my LF detector. So I would need at least 40 dB gain (x100). But I don't know the exact value yet. Something in the region of 40 dB to 60 dB gain (x100 - x1000). This is the next goal to determine the required gain and inductance of the RX coil.

The noise is not our enemy. White noise is our friend to increase the SNR of the signal by averaging the decoded samples. And it helps detecting deep and weak targets.

The effort of the ultra low noise design is just a sports or challenge. We are making it not because we really need it. We are just making it because we can do it. Even a very low-cost version.

Cheers
Aziz

[Aziz]

Bull**** happened due to ltspice noise error calculation. I have deleted the post.

[Aziz]

Hi all,

LTspice produced again wrong noise calculations. There must be a bug in LTspice.
Anyway.
The question was: What is better?
a) High pre-amp gain, low RX coil inductance (less turns, less series resistor) or
b) low pre-amp gain, high RX coil inductance (higher turns and higher series resistor) regarding noise generation in the pre-amp.

I got misleading results due to LTspice bug and did wrong conclusions in my previous deleted post.
A third example calculation revealed the buggy LTspice noise calculations.

Oh man!, be very very careful in noise calculation in LTspice.
First make the circuit work for the parameters you want to analyse and set the initial conditions for critical nodes. Then make the noise calculations without changing any parameter.
This is the rule!

So the answer to the question above is: b)
Reduce the gain of the pre-amp and increase the number of turns of the RX coil (to compensate the gain reduction). Lower pre-amp gain produces lower thermal noise. It is of benefit to take high inductance RX coils, which do not produce as much noise as using lower resistance and less RX inductance would do with high pre-amp gain.

Cheers,
Aziz

[Aziz]

I have made three working configurations in LTspice and looked at the noise at the output of the pre-amp. The noise density is measured at 10 kHz. The bandwidth of my noise calculation is 200 Hz. So we integrate from 10.0 kHz to 10.2 kHz over the noise density to get the output noise voltage of the pre-amp. Or simply Noise = ouput noise density * sqrt(200 Hz).
Let's look at different configurations.

Configuration 1:
Pre-amp gain G=500
RS = 4 Ohm (RX coil resistance)
Ouput noise density = 160.55 nV/rtHz (at 10 kHz)
Noise = 2.271 µV (rms)

Configuration 2:
Pre-amp gain G=1000
RS = 2 Ohm (RX coil resistance), RX coil has twice less turns as we have twice more pre-amp gain
Ouput noise density = 264 nV/rtHz (at 10 kHz)
Noise = 3.73 µV (rms)

Configuration 3:
Pre-amp gain G=2000
RS = 1 Ohm (RX coil resistance)
Ouput noise density = 460.2 nV/rtHz (at 10 kHz)
Noise = 6.51 µV (rms)

You see the answer now why. But this does not anwer my question yet. How much gain is required at what RX coil inductance? I have to plug a pre-amp and an RX coil to see the answer in the FFT spectrum. Trial and error method at best.

Cheers,
Aziz

[Aziz]

Now the very very interesting question:
Do we really need an ultra low noise pre-amp?

The answer is a big YES!

A simple NE 5534 AC amplifier will produce at least 20 times more noise in the above calculations. The more the gain, the more the difference.

Following noise calculations for a NE 5534 AC-amplifier:


Configuration 1: (see previous post)
Noise = 45.2 µV (rms), 20 times more noise

Configuration 2:
Noise = 89 µV (rms), 24 times more noise

Configuration 3:
Noise = 170 µV (rms), 26 times more noise

You see the difference and the answer now.
Cheers

[Aziz]

Hi all,

the pre-amp gain may not be larger than 100. Even lower (50 - 100). RG1 and RG2 gets critical due to drive capability of the NE5534. RG2 may not go below 470 Ohm. Otherwise, we would have more non-linearity at the pre-amp output. RX-Coil series resistance may go up to 10 Ohm. We can not get the 0.2 nV/rtHz level. But 0.4 - 0.6 nV/rtHz is easily possible with 1x ZTX851 depending on the source resistance. Paralleling more ZTX851 makes no sense.

I've forgotten, that the RX-coil will have some resudial voltage due to imperfect TX/RX IB-balance. So the faint RX signals are riding at these levels and they are well above the noise floor level and should be detected with quite low pre-amp gain. Maybe there isn't any pre-amp required at all. I know this, when my TX-circuit is ready to test (choke L1 not yet done).

Next time, I show you the EMI noise spectrum with a pre-amp gain 100 for a 2.1 mH RX coil.
Cheers,
Aziz

[moodz]

Hi all,

the pre-amp gain may not be larger than 100. Even lower (50 - 100). RG1 and RG2 gets critical due to drive capability of the NE5534. RG2 may not go below 470 Ohm. Otherwise, we would have more non-linearity at the pre-amp output. RX-Coil series resistance may go up to 10 Ohm. We can not get the 0.2 nV/rtHz level. But 0.4 - 0.6 nV/rtHz is easily possible with 1x ZTX851 depending on the source resistance. Paralleling more ZTX851 makes no sense.

I've forgotten, that the RX-coil will have some resudial voltage due to imperfect TX/RX IB-balance. So the faint RX signals are riding at these levels and they are well above the noise floor level and should be detected with quite low pre-amp gain. Maybe there isn't any pre-amp required at all. I know this, when my TX-circuit is ready to test (choke L1 not yet done).

Next time, I show you the EMI noise spectrum with a pre-amp gain 100 for a 2.1 mH RX coil.
Cheers,
Aziz

I already told you no amplifier ( beyond differential buffer ) is required LOL. With an IB balance the problem is to analyze difference signals ... not an noise / amplification problem.
It is very important to understand the problem before providing a solution.