PI noise

[green]

Made more layered targets. Charted integrator out at three target delay times with GEB off and GEB on(aluminum foil_5) One and two layer(1.4 and 2.8usec TC's) are a lot better with the 6usec target delay. Eight layer(10usec TC) about the same as a US nickel is a little better at the lower target delays. TC chart,(aluminum foil_4)

I measure a -1.37 slope for the ground I have. Entering the (A) sample times(aluminum foil _5) in the Excel simulator I posted awhile back a -1.33 slope cancels ground, close to what I measure. For ground the average reading between 6 and 16.3 is the same as a sample at 10usec. The GEB sample average between 21.3 and 121.3 is the same as a sample at 55usec, 9.85 times lower than the sample at 10usec. Using the simulator to calculate signal increase for the 1 and 2 layer targets(1.43 and 2.86usec TC)sampling at 6usec(instead of between 6 and 16.3usec) with an ADC would give a signal increase of 6.8 and 3.6 times. I'm reading integrator out with a digital voltmeter LSD=100uv. Integrator gain =1(10usec sample, target sample), =10(100usec sample, GEB sample). Amplifier gain =400. LSD =.25uv(target sample) 25nv(GEB sample). Trying to figure total bit resolution needed(amplifier + ADC). If you are confused after reading this you know how I feel trying to write it. With GEB on the LSD toggles one count short term maybe two or three counts long term(minutes). With an ADC sample at 6usec the GEB sample would be at 17usec for a gain of 4 or at 28.7usec for a gain of 8 effecting where the holes fall.

[Ferric Toes]

I went through a stage of worrying about the holes in the response of a PI with GEB, but tests showed that the notch in the response was in fact very narrow. This was demonstrated when I carefully filed down a piece of lead until it was undetectable in GEB mode. However, it became detectable again when retested about 15 minutes later. Drift in the electronics I wondered? Then I notice that when holding the lead twixt finger and thumb the detectability slowly vanished again. I came to the conclusion that filing the lead caused it to warm up and change the conductivity slightly so that, although undetectable immediately after filing, when left on the bench it had cooled down, conductivity went up and it became detectable. The heat of my fingers when retesting, reduced the conductivity and the lead once again disappeared off the detector's radar.

Detect a certain area on a hot day and you will find different targets to that on a cold day. I should mention that the mineralised ground response changes too.

Eric.

[Teleno]

I went through a stage of worrying about the holes in the response of a PI with GEB

Holes can be avoided by processing 3 consecutive samples as follows:



where





and






Plot for different ratios of target to ground signals (orange: 100% target, green: 90% target, 10% .. and so on).

[Ferric Toes]

Yes, I saw this in the other thread on GB holes and it looks very interesting. In a practical system, I presume all three samples are narrow and of equal width. Am I correct in thinking that movement of the coil in earth's field is cancelled in the s2/s1 and s3/s2 operations? It looks to be so as the amplitudes will all be the same over that time span. Target ratios could be extracted for target ID of coins. Ratios should be fixed for non-ferrous objects when scanning coil over them, and variable for ferrous?

Eric.

[Teleno]

Yes, I saw this in the other thread on GB holes and it looks very interesting. In a practical system, I presume all three samples are narrow and of equal width. Am I correct in thinking that movement of the coil in earth's field is cancelled in the s2/s1 and s3/s2 operations? It looks to be so as the amplitudes will all be the same over that time span.

EF should be subtratced form each sample before the calculation.

Target ratios could be extracted for target ID of coins. Ratios should be fixed for non-ferrous objects when scanning coil over them, and variable for ferrous?

I guess ratios could indeed be used for target classification. In any case that's beyond the purpose of this method which is simply to cancel the ground signal.

Ferrous response in the time domain appears to be in the form of a power function in early time and a decaying exponential in late time.

See eq. 8 in this paper: Subsurface discrimination using electromagnetic induction sensors.

Still not figured out how to handle this.

[green]

Wondering what effect sample rate and % sample time has on integrator output noise level. Thinking of doing an experiment. Including schematic of test, replace V3 with input amplifier. Test with amplifier input shorted and with a 8inch mono coil connected to input(Tx off). Any thoughts on the test or what I should expect? I've searched but haven't found anything, maybe someone knows where to look.

[bbsailor]

Wondering what effect sample rate and % sample time has on integrator output noise level. Thinking of doing an experiment. Including schematic of test, replace V3 with input amplifier. Test with amplifier input shorted and with a 8inch mono coil connected to input(Tx off). Any thoughts on the test or what I should expect? I've searched but haven't found anything, maybe someone knows where to look.

Green,

Integrating many RX signals improves the signal to noise ratio. See this web link: http://www.thinksrs.com/downloads/PD.../AboutLIAs.pdf Also do a web search on "lock-in amplifier tutorial" to see other on line resources. Since the integrated RX signals are synchronized with the TX signals, a PI detector that integrates many signals while the coil is over the target takes advantage of the principal of the lock-in amplifier. The limitations and trade offs are (1) the diameter of the coil, (2) TX frequency, and (3) sweep speed that all determine how many pulses are integrated while the target is being detected. If your TX pulse rate is 3K PPS to 10K PPS then you could be integrating from about 1000 to 3000 pulses depending of how fast you sweep over the target. Target size and coil size then also become critical variables.

Eric Foster used this method to beef up the RX side of detecting rather than just pump more energy into the TX pulse to make a more sensitive PI machine while still operating at a low delay.

I hope this helps?

Joseph J. Rogowski

[green]

Green,

Integrating many RX signals improves the signal to noise ratio. See this web link: http://www.thinksrs.com/downloads/PD.../AboutLIAs.pdf Also do a web search on "lock-in amplifier tutorial" to see other on line resources. Since the integrated RX signals are synchronized with the TX signals, a PI detector that integrates many signals while the coil is over the target takes advantage of the principal of the lock-in amplifier. The limitations and trade offs are (1) the diameter of the coil, (2) TX frequency, and (3) sweep speed that all determine how many pulses are integrated while the target is being detected. If your TX pulse rate is 3K PPS to 10K PPS then you could be integrating from about 1000 to 3000 pulses depending of how fast you sweep over the target. Target size and coil size then also become critical variables.

Eric Foster used this method to beef up the RX side of detecting rather than just pump more energy into the TX pulse to make a more sensitive PI machine while still operating at a low delay.

I hope this helps?

Joseph J. Rogowski

If the pulse rate is increased from 1000 to 10000, how much less is the integrator out peak noise level(1/3, 1/10)? Does taking a 10usec vs a 1usec target sample lower integrator out peak noise level and if so by how much?

[Lock-in amplifiers use a technique known as phase-sensitive detection to single out the component of the signal at a specific reference frequency and phase. Noise signals, at frequencies other than the reference frequency, are rejected and do not affect the measurement] from above link
What is the reference frequency and phase signal we are looking for with a PI detector?

[bbsailor]

If the pulse rate is increased from 1000 to 10000, how much less is the integrator out peak noise level(1/3, 1/10)? Does taking a 10usec vs a 1usec target sample lower integrator out peak noise level and if so by how much?

[Lock-in amplifiers use a technique known as phase-sensitive detection to single out the component of the signal at a specific reference frequency and phase. Noise signals, at frequencies other than the reference frequency, are rejected and do not affect the measurement] from above link
What is the reference frequency and phase signal we are looking for with a PI detector?

The RX signal is delayed until the TX flyback pulse and damping subsides that is at a constant phase with the TX signal which is also the reference frequency in a lock-in amplifier comparison. Noise is a broad band signal source but the integrated RX signal is only detected consistently in the same location on each cycle. It is the number of integrated signals that determines the ultimate improvement in signal to noise ratio. Once you lay out the TX and RX signal on a pulse train drawing and create a good mental model of how and where the PI detector senses a target, you will see the similarity with how a lock-in amplifier works to extract a signal in the presence of noise. Repeated for emphasis: how many samples are integrated indicates the potential signal to noise improvement.

Joseph J. Rogowski