Detection distance for a US nickel and quarter

[green]

Used Excel to integrate a signal similar to a quarter. Quarter decays straight line log-log to about 100us. Generated a decay with -.68 slope and calculated the average for a 10us sample(6 to 16us) and a 110us sample(6 to 110us). Average for 10us sample=2.159. Average for 110us sample=.834. With a 1C integrator the gain for the 110us sample is 11 times the 10us sample, .834*11=9.17 . The longer sample is 9.17/2.159=4.25 the shorter sample, close to the 4 times I got with my real data calculation.

[green]

110us sample gives 1.33 times the signal as a 10us sample for the nickel. 110us sample gives 4 times the signal as a 10us sample for the quarter. A longer sample might help for the quarter not the nickel. Probably can't do ground balance with a 110us target sample.

4 times the signal was based on calculations reply #38. Original bench tester had target sample and ground sample capability. Failed awhile back. Made target response tester, record linear and or log amplitude decay. No integrator. Not sure calculation using target sample and ground sample is same as sampling 10us target sample and 110us sample. Thinking making another bench circuit including integrator. Maybe just sample delay and target sample with EF samples. Adjustable: transmit rate(200 to 2000pps), target delay(5 to 30us), target sample(10 to 500us).

Reply in another forum stated a GPX4500 and GPZ7000 could detect a quarter at 24inches. Wondering what it would take. Dave J stated in another thread a 24inch coil might be a place to start for a target at 24inches. Just a learning exercise to see if I could do it on the bench. Hoping the above bench circuit would be usable. Any thoughts what might need changing or added?

[green]

110us sample gives 1.33 times the signal as a 10us sample for the nickel. 110us sample gives 4 times the signal as a 10us sample for the quarter. A longer sample might help for the quarter not the nickel. Probably can't do ground balance with a 110us target sample.

Depends on integrator if 4 times the signal. 1C integrator gain=R fdbk/R in*sample rate*sample time. Gain is 11 times higher with 110us sample vs 10us sample. 1C integrator out=average sample*gain?
Still trying to figure MPP integrator.

Some spice sims for MPP and 1C integrator, think they make sense. Sample time for 1C integrator changes gain not response. Sample time for MPP integrator effects response, not much gain effect. Changing R5 MPP integrator effects response, not much gain effect.

Thinking 2C integrator is same as 1C integrator?

[green]

It is the Rate at which a new Voltage is applied to the integrator and the time for the integrator is decay.

Carl wrote a good description a while back on how an integrator in a PI detector works. Don't remember the thread but I copied the text. I then did a spreadsheet to do setup the parameters and do the calculations.
Here is Carl's explanation:

Let's say we use a single-ended integrator shown below; R1=1k, R2=100k, C1=0.1uF. 1000 samples/sec. Carl's example

Integrator time constant=.01 seconds. If .01 is correct should we change the time constant if the sample rate changes to 3000 samples/second? Maybe I'm looking at it wrong?

[waltr]

Why does Pulse/sample rate effect optimum integrator time constant?

It is the Rate at which a new Voltage is applied to the integrator and the time for the integrator is decay.

Carl wrote a good description a while back on how an integrator in a PI detector works. Don't remember the thread but I copied the text. I then did a spreadsheet to do setup the parameters and do the calculations.
Here is Carl's explanation:

[green]

Also, the Pulse/sample rate.

Why does Pulse/sample rate effect optimum integrator time constant?

[waltr]

What determines optimum integrator time constants? Maybe sweep speed, coil size and ?

Yes, target response speed, which depends on sweep speed.

Also, the Pulse/sample rate.

[Carl-NC]

For the application I am working on GB is not required

Same with me.

What determines optimum integrator time constants? Maybe sweep speed, coil size and ?

Yes, target response speed, which depends on sweep speed.

[green]

For the application I am working on GB is not required, but better range on higher conductivity coins is needed. With lower conductivity items, the fact that the range is little better than with a short sample pulse, is fine. I am also going to look into the integrator time constants as these may not now be optimum with the long samples.

Eric.

What determines optimum integrator time constants? Maybe sweep speed, coil size and ?

[Ferric Toes]

110us sample gives 1.33 times the signal as a 10us sample for the nickel. 110us sample gives 4 times the signal as a 10us sample for the quarter. A longer sample might help for the quarter not the nickel. Probably can't do ground balance with a 110us target sample.

For the application I am working on GB is not required, but better range on higher conductivity coins is needed. With lower conductivity items, the fact that the range is little better than with a short sample pulse, is fine. I am also going to look into the integrator time constants as these may not now be optimum with the long samples.

Eric.

[green]

Eh, not really.

110us sample gives 1.33 times the signal as a 10us sample for the nickel. 110us sample gives 4 times the signal as a 10us sample for the quarter. A longer sample might help for the quarter not the nickel. Probably can't do ground balance with a 110us target sample.

[Carl-NC]

Hope it makes sense, was getting confused typing it.

Eh, not really.

I will look at this in more detail over the next few days. Is your cache still buried?

No, I dug it up 12 years ago when I moved to the West Coast. Never did bury it again.

[Ferric Toes]

Funny you posted this. For the last couple of weeks I've been optimizing the timing on a new rev of the Fisher pinpointer. It uses bipolar pulsing and with a short (20us) sample width has the obvious problem of detecting nickels well but quarters not so well. I decided to stretch the sample width out to get the quarters better and expected the nickel response to either degrade or get noisier -- it did not. Quarters got better, nickels stayed exactly the same. This uses a sampling integrator which may explain why. Had I used a SHA-type demod then I think the nickel would have been weaker due to averaging. I did not stretch the sample width all the way to the next TX pulse but will try that.

Yes it is worth trying for as long as you can make it. I tried lengthing it on the monopolar board that I used for the earlier plots I posted in this thread. Tx is 350uS and the minimum delay controlled by a pot is 10uS with 20uS sample. Max. delay is 50uS. I put a 22k pot in series with a 10k resistor which is in the timing network for the two sample pulses this would give a range of 40uS - 100uS, so as to see what difference in signal there was with a US Quarter and a similar size UK cupro-nickel Shilling. Unfortunately, the bin in which I keep 10k resistors, had a 100k reel by mistake. This gave me a pair of sample pulses of 200uS width. It seemed to work OK so I did some quick tests and 'lo and behold' it gave considerable improvement in the range of a Quarter and came out a bit better than the shilling, which seemed largely unaffected. Range was considerably improved for a silver dollar too, but looking at the total decay time, it could be better still with greater spacing between samples which currently is 350uS between the end of SA1 to the start of SA2. There is enough space between Tx pulses to accomodate 600uS between samples at the max delay of 50uS and 200uS sample widths.

I am only using an 8 inch coil at the moment, suspended between floor and ceiling as I made a big mistake in having all my new workshop lined, including underfloor, with insulation board which I later found out had a foil surface. Also the electrician fitted LED lights which cause a lot of interference, so I have to switch off the lighting and resort to filament bench lights when doing any testing. These mistakes happened when I was having chemo for lymphoma and was out of the loop for a year.

I will look at this in more detail over the next few days. Is your cache still buried?

Eric.

[green]

Chart I posted awhile back when my bench circuit was working. Integrator out readings, first sample(target sample) and second sample(ground sample + target sample). Ground sample=second sample-first sample.

First two columns for the nickel(first delay, 6us). First sample(10.2us, 62.5mV). Second sample(100us, 42mV). Ground sample -20.5mV. If first sample was 110us with no GB(83mV). 1.33 times signal increase

First two columns for the quarter(first delay, 6us) First sample(10.2us, 20.5mV). Second sample(100us, -41mV). Ground sample -61.5mV. If first sample was 110us with no GB(82mV). 4 times signal increase

Can't do GB if first sample is over 100us?

Hope it makes sense, was getting confused typing it.

[Carl-NC]

The best way to test TFR would be to use a bipolar Tx so that a sample, or samples, could be taken after a short delay from the end of +Tx1 right to the start of -Tx2. This alternate polarity pulsing would also cancel earth's field without any cancellation of the wanted signal. What I would like to know is whether a number of short successive samples which are summed at the integrator would add up to a much larger voltage whose amplitude is proportional to the length of the target decay. Presumably a long single sample would just give an average value?

Funny you posted this. For the last couple of weeks I've been optimizing the timing on a new rev of the Fisher pinpointer. It uses bipolar pulsing and with a short (20us) sample width has the obvious problem of detecting nickels well but quarters not so well. I decided to stretch the sample width out to get the quarters better and expected the nickel response to either degrade or get noisier -- it did not. Quarters got better, nickels stayed exactly the same. This uses a sampling integrator which may explain why. Had I used a SHA-type demod then I think the nickel would have been weaker due to averaging. I did not stretch the sample width all the way to the next TX pulse but will try that.