DEEPER PI DETECTION DEPTH

[Tinkerer]

I would remove the RC filter R7,9 C8 - The RC filters loose signal power in the resistive elements - the LC ones dont to speak of.

The order of this filter is a little more than I sketched, The input impedance I chose 500R and the output z to op amp came out as 560R

S

Thank you for the filter.
What are the specs for the inductors?
A question about the impedance:
I have a variable voltage divider between R9, U7, R13. I use this to adjust both legs of the RX coil to the same signal amplitude.
The diodes should not conduct, they are there as a precaution only for when there is a problem with the induction balance.
R34, is to adjust the offset caused by the shield of some types of cables.

Increasing the impedance, reduces the amplitude of the signal, which allows for more amplification in the preamp, at the cost of more noise.

Eliminating more noise with the filters, before the input of the preamp is going to help.

I feel there is a sweet spot, where the S/N is best.
How do I determine that spot?

Tinkerer

[golfnut]

I would remove the RC filter R7,9 C8 - The RC filters loose signal power in the resistive elements - the LC ones dont to speak of.

The order of this filter is a little more than I sketched, The input impedance I chose 500R and the output z to op amp came out as 560R

S

[Tinkerer]

HI, with the cct you posted - If LP filters were what U required I would fit them here.

Having them close to the amp gives the best chance of catching noise.

If U wanted this topology I could get U some values for a roll-off corner of your choice.

Steve

Thanks for the feedback.

I would like to try 2 different frequencies. About 100kHz and about 150kHz to see what the influence on the small and very small target response is.

The RX coil is center tapped, about 300uH, but I still want to try different inductance.

Thanks for the help

Tinkerer

[golfnut]

HI, with the cct you posted - If LP filters were what U required I would fit them here.

Having them close to the amp gives the best chance of catching noise.

If U wanted this topology I could get U some values for a roll-off corner of your choice.

Steve

[Tinkerer]

The filter would be duplicated, one per input. The common center terminal of the filter containing the filtered signals goes to gnd.

If the filter has symmetry there would be no penalty to the common mode rejection.

steve

Hi Steve,

would you like to help me improve this RX schematic of mine?

It works, but I am sure it could be improved a lot.

I would much appreciate.

Tinkerer

[golfnut]

The filter would be duplicated, one per input. The common center terminal of the filter containing the filtered signals goes to gnd.

If the filter has symmetry there would be no penalty to the common mode rejection.

steve

[Tinkerer]

I guess the 4kTR also has another factor B - bandwidth.


If the Input of the Rx is not restricted to the Usable Rx BW of the system then you let extra noise in. The noise floor goes up as it gets hit with noise from a wider frequency span. S/N suffers = you would see less targets as they are in the noise.

Restricting the Rx input bandwidth with an appropriate filter will limit noise from mains tv cell too.

Separating the Rx from the Tx is a good move...
1) You could run the Rx amp common mode to cancel a load of picked up noise
2) The Rx signals are not being burned away as heat by the De Q resistor on the Tx coil
3) You would not need the inverse pair of diode limiters -as your not Txing up your Rx
Always a bad move

S

The signal of a small target has a TC of less than 5us. To capture such signals, the bandwidth of the coil and preamp should be above 100kHz.
When people used the NE5534 preamp with a gain of 1000, this automatically reduced the bandwidth. However, using better opamps with more bandwidth, a low pass filter before the preamp if the configuration allows, and /or bandwidth limiting the preamp reduces the noise significantly.

I like to use a differential input preamp. Now how does a lowpass filter in front of the preamp influence the common mode rejection?

Tinkerer

[golfnut]

I guess the 4kTR also has another factor B - bandwidth.


If the Input of the Rx is not restricted to the Usable Rx BW of the system then you let extra noise in. The noise floor goes up as it gets hit with noise from a wider frequency span. S/N suffers = you would see less targets as they are in the noise.

Restricting the Rx input bandwidth with an appropriate filter will limit noise from mains tv cell too.

Separating the Rx from the Tx is a good move...
1) You could run the Rx amp common mode to cancel a load of picked up noise
2) The Rx signals are not being burned away as heat by the De Q resistor on the Tx coil
3) You would not need the inverse pair of diode limiters -as your not Txing up your Rx
Always a bad move

S

[simonbaker]

An example?
E-Motor switching on (just imagine, you are detecting in an industrial area). You don't know, when the E-motor is switched on (random/gaussian behavior). But if switched on, it's response characteristics is more or less definite (depends on the type of noise source). You can predict it's behaviour if you can detect such an event.

More examples?
Other inductive electrical devices during switch on/off. Switching lights on/off. ....

There are more types of noise sources of this behaviour. And they give you the chance to cancel them.

Aziz

PS:
A good example now:
Swinging the coil over a strong magnet! The detector doesn't know, when you do it. The magnet could be hidden in the ground. But it generates a wide band response, which can't be handled by simple filters!!!

I probably should have said "white noise" or "pink noise" rather than just call it gaussian, which really refers to the amplitude statistics I think.

I agree, if you have good "a priori" knowledge of a noise signal "profile", then all you have to do is detect and estimate it as it occurs and apply the profile you know from experience. And you should be able to detect/estimate a known signal with a few points outside a certain band, I would think. There will probably be a new type of "false positive" error, where you think you detect this signal but are wrong, and you apply the profile, creating a greater noise signal.

I'm not sure how swinging a coil over a strong magnet is very wide band unless you do it super fast, like a bullet. But I see what you are getting at; and I believe it does require a lot of historical information about this signal that may arise. If the noise is a totally unknown signal, you wouldn't be able to predict the content in one band based on another without more information, I'm pretty sure.

Regards,

-SB

[Aziz]

The noise in the out of band correlates with the region of interest band.

- I don't think that is correct for gaussian noise. Maybe some other kind of noise.

An example?
E-Motor switching on (just imagine, you are detecting in an industrial area). You don't know, when the E-motor is switched on (random/gaussian behavior). But if switched on, it's response characteristics is more or less definite (depends on the type of noise source). You can predict it's behaviour if you can detect such an event.

More examples?
Other inductive electrical devices during switch on/off. Switching lights on/off. ....

There are more types of noise sources of this behaviour. And they give you the chance to cancel them.

Aziz

PS:
A good example now:
Swinging the coil over a strong magnet! The detector doesn't know, when you do it. The magnet could be hidden in the ground. But it generates a wide band response, which can't be handled by simple filters!!!

[Tinkerer]

Hmmm, looks to me like twice the TX current as well. Post #93 shows ~450mA single-ended, but #97 shows +/-450mA.

In any case, I am working on a circuit to measure this phenomenon. I'll post results next week.

- Carl

It is twice the same current, charge and discharge of the same capacitor.

With LTSpice, TEM.asc, I use the inductor L7 to measure the current. I click Control L7 to get the average current, it is about 8mA. For the Sawtooth TX it is about 26mA.

The slope, or rate of change is slower, so one would expect a lesser magnetic moment, but the target response shows that the longer slope produces more eddy currents.

There are several differences in the circuits. TEM allows for a much higher pulse repetition rate and results in a heavier coil to get low power losses.

It also gives very good FE discrimination.

Tinkerer

[Carl-NC]

Below is another simulation.
With the same TX coil current I get twice the RX signal.

Hmmm, looks to me like twice the TX current as well. Post #93 shows ~450mA single-ended, but #97 shows +/-450mA.

In any case, I am working on a circuit to measure this phenomenon. I'll post results next week.

- Carl

[Carl-NC]

Anyone heard of JOHNSON noise? It pertains to resistors (specifically carbon) so, USE METAL FILM IN THE FRONT END!!!!!! Formula is SQROOT (value of resistor) * 4nV/Hz

Johnson noise applies to ALL types of resistors. Bulk carbon resistors have additional noise on top of that, carbon film are a little better, and metal film the best. But even metal film can't escape 4kTR.

- Carl

[Tinkerer]

Hi Tinkerer:

You are doing yeomans work, it is always appreciated.

Can you attach your LTSpice file each time you post an output graph so we can better understand all the factors involved and how you model your system?

Best regards,

-SB

SB,

here is the simulation of the Sawtooth TX and the TEM TX side by side. Both simulations use about the same TX peak current, but the TEM simulation recycles the current and uses only a fraction of the power that the Sawtooth TX uses.

Also attached are the LTSpice files of each.

As you can see on the .asc files, the difference in the schematic is minimal, but the TEM gives twice the signal amplitude and uses a fraction of the power.

The blue trace is the Flyback voltage.
The green trace is the coil current.
The red and pink traces are 2 targets with different TC.

Tinkerer

[simonbaker]

There are simple/advanced rules for avoiding or coping with (EMI) noise:
(besides the PCB rules, avoiding ground loops, decoupling, low-noise designs, etc.)

<turning on the novice mode>

1. Avoiding: Avoid it, i.e., don't measure it if possible (shield your coil and cables).

<turning on the advanced mode>

2. Cancelling: Make the noise common mode to cancel it in a differential stage. Or use an anti-interference coil (figure-8 coil, or use a "reference" coil).

3. Band limitting: Don't measure out of interest bandwidth. You don't need it.

<turning on the expert mode>

4. Synchronizing: If the noise source isn't a gaussian nature (gaussian noise can't be modelled and predicted), then synchronize your measurement to the noise source to avoid the modulation of your measurement (the noise modulates your measurement). The noise source typically comes from the mains power source, radio stations carrier frequency, switched power supplies, CPU (program code dependent), other switching devices, etc.
Well, if the noise source is a gaussian nature, sh1t happens. You can't predict it, when it comes.

<turning on the Einstein/genius mode>
(to cope particularly with the evil gaussian noise)

5. Predicting: It's beyond the scope of most users now (even experts):
Use the out of interest band to cancel/predict the noise in the region of interest band. Some noise sources are wide band gaussian nature (like the wide band random pulses), which intersects with the band of interest region. The noise in the out of band correlates with the region of interest band.
Oh man!, you need a lot of cpu processing & brain power for this.

<turning off the brain demolution mode>

The list isn't complete. Just to give you an idea.


Aziz

- I think it is a good point to distinguish EMI noise problems from circuit noise problems. Circuit noise problems follow the usual rules about focusing on the front-end components, making them as low-noise as possible.

- Certainly a good point about "synchronizing" to any predictable noise signals -- subtract them out, integrate them out, map them to a constant voltage (SD), etc.

The noise in the out of band correlates with the region of interest band.

- I don't think that is correct for gaussian noise. Maybe some other kind of noise.

- Band limiting is always a good idea, just have to watch out your filter doesn't distort your signal, or at least you know how to compensate.

- As for differential mode, that's great to knock down common mode noise such as the cable acting like an antenna. However, can it really help with noise picked up by the coil? There must be noise that looks just like the target signal, so you can't have it both ways -- block the noise, you block the target.

Not knowing anything, my impression of PI detectors is that the real challenge is calibration. It seems to boil down to picking off exact voltage points at exact time delays. Some kind of normalizing circuitry/algorithm would seem to be desirable to enable looking for smaller changes.

Cheers,

-SB