PI noise

[moodz]

The impact of EMI including very low frequencies down to DC should not be underestimated in a PI. The front end of a PI is a natural modulator and signal frequencies in the the frontend are mixed or undergo CONVOLUTION. The pictures below illustrate this ... the first pic is the spectrum of a signal from a PI operating at 1kHz with substantially no noise. The next two pics show the effect of a 60 Hertz signal on the PI signal. You should be able to see that the 60 Hertz is modulated onto the spectral components of the PI signal. The thing about CONVOLUTED signals is that they cannot be removed with a filter. Even though I have used a 60 Hertz signal to illustrate in practice ALL signals ( ie noise / EMI etc etc) from DC to up will be convoluted / modulated onto the PI signal and demodulated with your "wanted" signal.

[green]

Thanks for the replies.

I have tested the detector in the yard. The ground signal is a lot higher than amplifier noise. The detector isn't usable without GEB. Noise goes up when I add GEB so I'm trying to minimize the noise.

I wonder why some detectors add switches to eliminate the 1k input resistor if coil noise is always higher than 1k resistor noise.

I'm learning so what I'm thinking might not be correct. Target signal is less than 10Hz, ground and EF maybe closer to 1Hz so maybe other than line frequency(50 or 60Hz)most other frequencies are above the sampling rate, if not they would be filtered anyway. The maximum frequency after sampling is 1/2 the sampling frequency. If we filtered all frequencies above 10Hz the only higher frequencies that would matter are 10Hz and below and within 10Hz of multiples of the sampling frequency. Sample rate 1kHz(1010,2010,3010,3990 etc.) 5kHz(4990,10010,19990,30010 etc.) maybe one of the reasons higher sampling frequency might have less noise, fewer frequencies to alias on.

I measured p-p noise level at post amplifier out with the coil input shorted with 1kHz and 5kHz sample rate. p-p noise increased about 2 times at 5kHz. Target, ground and EF signal should increase 5 times giving an increase in S/N of 2.5. Why doesn't the noise signal increase 5 times? My guess in above reply. What is the reason?

[moodz]

...not sure why it should increase by 5 times ... dont forget your sampling is synchronous .. the noise is not.

[Qiaozhi]

I haven't studied the circuit closely, but the increased noise might be due to the differentiator, which amplifies the input in direct proportion to the frequency. In other words, the noise level increases with frequency. Of course, that might be nothing to do with it, but it could be worth looking at.

[Ferric Toes]

I measured p-p noise level at post amplifier out with the coil input shorted with 1kHz and 5kHz sample rate. p-p noise increased about 2 times at 5kHz. Target, ground and EF signal should increase 5 times giving an increase in S/N of 2.5. Why doesn't the noise signal increase 5 times? My guess in above reply. What is the reason?

Several things going on here so let's go back a bit. You appear to be saying that you are increasing the whole cycle of TX pulses and RX sampling from 1000 operations per sec. to 5000. As you say, this should give 5 times the signal at the integrator output. It appears that you are keeping the integrator TC the same. If you are happy with the response speed at 1kHz (in radar we used kpps which is more accurate for pulses), then you can increase the TC by a factor of 5 and you will see some improvement in S/N.

Eric.

[green]

I haven't studied the circuit closely, but the increased noise might be due to the differentiator, which amplifies the input in direct proportion to the frequency. In other words, the noise level increases with frequency. Of course, that might be nothing to do with it, but it could be worth looking at.

spice simulation. R7 and C4,(integrator feedback)simulate integrator gain=1.

[green]

Several things going on here so let's go back a bit. You appear to be saying that you are increasing the whole cycle of TX pulses and RX sampling from 1000 operations per sec. to 5000. As you say, this should give 5 times the signal at the integrator output. It appears that you are keeping the integrator TC the same. If you are happy with the response speed at 1kHz (in radar we used kpps which is more accurate for pulses), then you can increase the TC by a factor of 5 and you will see some improvement in S/N.

Eric.

Trying to understand how the integrator feedback TC and sample rate correlate. I was thinking sample rate effects gain and feedback TC effects low pass filter response.

[green]

...not sure why it should increase by 5 times ... dont forget your sampling is synchronous .. the noise is not.

You are suggesting the noise doesn't increase as much as the signal because noise is non synchronous? The noise is non synchronous at 1k and 5k samples/second with the gain 5 times higher at 5k but you still might be right. Don't know how to prove why the noise doesn't increase as much as the gain.

[Ferric Toes]

My prefered analogue integrator

This is an arrangement that works well. It has the benefit that the capacitors 'hold' until the next sample i.e. there is no leak down through a parallel resistor.

In the differential version the loop gain controls the time constant, so you can have an external pot to vary the 'response speed' or 'noise averaging'.



Eric.

[green]

This is an arrangement that works well. It has the benefit that the capacitors 'hold' until the next sample i.e. there is no leak down through a parallel resistor.

In the differential version the loop gain controls the time constant, so you can have an external pot to vary the 'response speed' or 'noise averaging'.

[ATTACH]36677[/ATTACH]

Eric.

http://www.geotech1.com/forums/showt...118#post219118

Teleno is having problems with EF with his Minipulse-plus. It has an integrator like the one you suggested. I've had problems cancelling EF with a normal 2C integrator. Does the 2C with hold have the same problems? Looks like you can't do GEB by adjusting ground sample time like can be done with a 1C integrator, maybe I'm wrong.

[Qiaozhi]

Teleno is having problems with EF with his Minipulse-plus. It has an integrator like the one you suggested.

I've been in communication with Teleno concerning this problem, but it doesn't appear to be anything to do with the EF.

[Ferric Toes]

I've been in communication with Teleno concerning this problem, but it doesn't appear to be anything to do with the EF.

Being absent from the Forum for a while, I have not followed the Minipulse Plus thread and what changes/improvements have been made. I have only just looked at the schematic and done some quick comparisons with the Minipulse B schematic of January 1989. I will say, however, that there was never a problem with EF with the integrate and hold circuit that I started to use around that time. The same arrangement was also used in the Superscan and the waterproof Aquastar when they were in production.

Where a problem can arise is if the coupling capacitor (C12 in the MPP circuit) is too low a value. In the original MP it was 22uF, while in the MPP it is 0.47uF. The Garrett XL500, for which I designed the circuit, had EF problems for the same reason; they substituted a low value capacitor for the larger one I originally used. Luckily I checked an early production one and wrote a report for Charles Garrett.

If you scope across R19, with the 0.47uF, you will likely see that the pulsed waveform no longer has a flat d.c baseline but is a ramp. You need a flat, or substantially flat baseline to get good EF cancellation, so it is either a big capacitor or direct coupling. In the MP I used a tantalum capacitor and offset the preamp output to -0.5V to give the capacitor a bit of dc bias.

Today, with better I.C.s and tighter component tolerances, I generally couple the preamp to the integrator input without a blocking capacitor. With good balance in the differential integrator any small offset changes in the preamp cancel out anyway.

Eric.

[Qiaozhi]

Where a problem can arise is if the coupling capacitor (C12 in the MPP circuit) is too low a value. In the original MP it was 22uF, while in the MPP it is 0.47uF. The Garrett XL500, for which I designed the circuit, had EF problems for the same reason; they substituted a low value capacitor for the larger one I originally used. Luckily I checked an early production one and wrote a report for Charles Garrett.

If you scope across R19, with the 0.47uF, you will likely see that the pulsed waveform no longer has a flat d.c baseline but is a ramp. You need a flat, or substantially flat baseline to get good EF cancellation, so it is either a big capacitor or direct coupling. In the MP I used a tantalum capacitor and offset the preamp output to -0.5V to give the capacitor a bit of dc bias.

I cannot recall why C12 was decreased from 22uF to 470nF, but there is no issue with cancelling the EF.
Perhaps someone else can experiment with this, as I'm up to my eyeballs in other work, and the MPP is buried under a heap of stuff.
It's about time for a massive clear-up.

[Ferric Toes]

I just checked on the Garrett XL500 Seahunter schematic and the relevant capacitor they substituted was 0.1uF which was seriously too small.

Eric.

[Ferric Toes]

It's about time for a massive clear-up.

That's what my wife is telling me.