Preamp

[Old cart]

I'm just trying to learn. Including a scope picture from reply#52 and a chart from reply #62. I think they define target shape and amplitude for sweeping a 300uH 8 inch diameter coil with 1 amp peak current over a US nickel. About 1 uv change at the coil with the target at 15 inches from the coil. I know the peak noise referenced to the coil has to be less than 1 uv(how much less?) You mentioned filtering reduced noise and reduced required bits. I still think the target defines resolution and noise level. It doesn't matter how many bits I have if the peak noise level referenced to the coil is 5 uv. If the noise level is zero I still need enough bits to define a 1 uv change with the defined target recording shape. Bits hardware + averaging?

I will answer this backward. If there is zero noise then yes you would need a resolution on 1 uV or better to be able to see the small signal. Interestingly you are better off having some noise, particularly if you do signal averaging. This is because the noise sums with the signal and raises it above the minimum detectable limit. For example if you had a 1 uv with 1 uV p-p noise the combined signal would be varying from 0 to 2uV which would ensure an a to d with 2uv resolution would detect the signal, even if only occasionally. If you averaged the signal 2 times and the noise is not synchronized with the signal the noise would "melt away" and you would left with a signal of 1 uV and noise of about 0.6 uv. So by using averaging you can see a signal that is equal to and even below the noise. But it is the noise that allows you to digitize the signal in the first place. Classic averaging can only be used on signals that repeat. It would not be very useful on a low frequency signal like a coil passing over a target. Why? Becuase to average 8 times would require 8 identical passes over the target.
However there is a form of averaging called boxcar integration. Google it. It does not require a repetitive signal but does require closely spaced samples in time. It could be used to good effect in a metal detector to pull the signal from the noise. It not only increases the effective resolution, it has a secondary benefit of acting like a low pass filter.
How do you measure and define noise? You can measure it anywhere you want, it is arbitrary. As long as you define where and under what conditions you are good to go. I mentioned the signal and noise levels at the input to the a-d because that is all the converter sees. It neither knows, or cars, what is going on at the coil. Further if there are signal processing stages like gain, filtering, they will modify how much noise and signal is present. Gain stages general do not affect the signal to noise ratio, as long as they are quiet. Filtering can easily improve s/n as long as the signal and noise frequencies are different.

[6666]

I'm just trying to learn. Including a scope picture from reply#52 and a chart from reply #62. I think they define target shape and amplitude for sweeping a 300uH 8 inch diameter coil with 1 amp peak current over a US nickel. About 1 uv change at the coil with the target at 15 inches from the coil. I know the peak noise referenced to the coil has to be less than 1 uv(how much less?) You mentioned filtering reduced noise and reduced required bits. I still think the target defines resolution and noise level. It doesn't matter how many bits I have if the peak noise level referenced to the coil is 5 uv. If the noise level is zero I still need enough bits to define a 1 uv change with the defined target recording shape. Bits hardware + averaging?

Thanks for the graphs.

[Chet]

Hi Green


The charts are a great help. And your chart of the coil sweeping over a nickel explains the 10 Hz response. Your two receiver coil arrangement appears to do a good job of eliminating common mode EMI.


Does the frequency of the target eddy current slope inside of a 10 usec target sample gate need to be considered in later processing?


Thank you,
Chet

[green]

It was just an example. While what you say is true this example includes how you would select an a to d after measuring what the actual levels would be. A band limited measurement from a spec sheet can be helpful. If you measure at the point the a to d is placed you will be taking into account all the factors including gain, noise level at the conversion point, emi, crosstalk, everything. Some of those are not easily modeled, at least by me. However if you have not built the circuit then you wil have to simulate what the signals wil be and this is, I think, quite complicated. Let me know if this still does not make sense.

I'm just trying to learn. Including a scope picture from reply#52 and a chart from reply #62. I think they define target shape and amplitude for sweeping a 300uH 8 inch diameter coil with 1 amp peak current over a US nickel. About 1 uv change at the coil with the target at 15 inches from the coil. I know the peak noise referenced to the coil has to be less than 1 uv(how much less?) You mentioned filtering reduced noise and reduced required bits. I still think the target defines resolution and noise level. It doesn't matter how many bits I have if the peak noise level referenced to the coil is 5 uv. If the noise level is zero I still need enough bits to define a 1 uv change with the defined target recording shape. Bits hardware + averaging?

[Teleno]

I'm still missing why you give a noise level without referencing it to the input or include a gain.

It's very simple, you asked about the resolution of and ADC that samples the output, therefore it's the noise at output what you must look at. The gain is inmaterial reagarding the resolution of the ADC.

[baum7154]

When measuring noise it is a good idea to amplify the test amp with an external low noise amp. Oscilloscopes are not designed to make measurements on small signals at frequencies this low. Scope external preamps are available, but are rather expensive. For example the Tektronix ADA400 but it only works with modern TEK scopes.
Such a preamp should have very low noise and adjustable gain and bandwidth. Adjustable offset is also nice to have. This group could of course design such a preamp or we could just use the design of the ADA400 as a basis. The schematic is available.
A simpler, cheaper design is available here

http://tangentsoft.net/elec/lnmp/

This design is well documented and even includes gerbers so no real design work would be necessary. It does have one issue though and that is that the input impedance is quite low at 100ohms. To fix this would require a high impedance low noise buffer to be added before the first op amp.
It has several component options explained and should work well with scopes and true RMS multimeters.

Lastly last night I read a post by Eric Foster who said the coil, if used in such a noise test must be horizontal as most noise is polarized. Placing the coil vertical will allow more external noise to enter. If we use a pendulum with a long line, like 2 M, the arc that the target makes wil be very slight and should not affect the results.

----------------------------

I agree with your last statement regarding the coil being Horizontal. All of my air testing the past year or two was done in the basement on my pool table with the coil horizontal, placed on the end of a roll of paper towels. I have found this orientation gives the least interference I can manage and it results in very quiet performance of the detector. It also allows a plastic ruler to be stood near the center of the coil for distance verification.

Dan

[Old cart]

[If you for example want to acquire all of the data out of a preamp that has +/-5 volt supplies and a measured noise level of say 500uV then you would need 10V/500uV or a 20000 to 1 ratio.] I'm still missing why you give a noise level without referencing it to the input or include a gain. A preamp with a gain of 10 or 1000 would make a difference finding the nickel at 15 inches with a 15 bit a to d converter. The noise after filtering should to be referenced to a frequency band that includes the target signal. Analog or digital filtering. Some low noise op amps include a 10 second .1 to 10 Hz scope trace in there spec sheet. I'm thinking the target signal is less than 10 Hz so something similar would make sense, maybe not.

It was just an example. While what you say is true this example includes how you would select an a to d after measuring what the actual levels would be. A band limited measurement from a spec sheet can be helpful. If you measure at the point the a to d is placed you will be taking into account all the factors including gain, noise level at the conversion point, emi, crosstalk, everything. Some of those are not easily modeled, at least by me. However if you have not built the circuit then you wil have to simulate what the signals wil be and this is, I think, quite complicated. Let me know if this still does not make sense.

[green]

Hi Green


This is very interesting. I am trying to get a better understanding of the raw frontend noise and signal levels that would be seen by direct measurement with and ADC.
In your measurements and charting do you have any charts that would show an average noise level figure that we may expect to see? And what target signal levels might be expected to be discernable or considered detected above that noise?


In processing would long time constant targets fall into a low Hz band width and short time constant targets fall into a higher band width range?
I know that you probably encountered considerable noise from in house EMI but it would still help to understand some of the noise and target levels from real measurements.
I am really interested in what noise and detectable target levels would be seen by an induction balanced (IB) receiver coil.


Have a great day,
Chet

Hi Chet

Including a scope shot from another thread and a new one I made awhile back. Test_2, preamp connected to post amp(no integrator) OPA1612 1k resistor connected to +input or shorted, -input less than 200 ohms to common. Both tests the circuit was running with the Tx coil not connected. They show what I'm thinking. I think the diff amp is noisier than the opa1612 with 1k to ground on each input because of higher bias current. Integrator sampling increased noise. Working on a new circuit to try. Thinking of labeling the threshold pot in uv at the coil(should show noise level referenced to coil volts). Was surprised the coil noise wasn't higher, some high frequency noise but didn't increase low frequency amplitude, maybe it would with the integrator connected. The large spikes happen when the fet is turned on, need better PS isolation? IB coil, two 8 inch Rx coils side by side(one inverted for opposite phase) surrounded by a race track Tx coil.

Don't think target time constant effects target band width, time for coil to pass target. Interested in how Teleno's amp is working with a mono coil.

[Chet]

Hi Green


This is very interesting. I am trying to get a better understanding of the raw frontend noise and signal levels that would be seen by direct measurement with and ADC.
In your measurements and charting do you have any charts that would show an average noise level figure that we may expect to see? And what target signal levels might be expected to be discernable or considered detected above that noise?


In processing would long time constant targets fall into a low Hz band width and short time constant targets fall into a higher band width range?
I know that you probably encountered considerable noise from in house EMI but it would still help to understand some of the noise and target levels from real measurements.
I am really interested in what noise and detectable target levels would be seen by an induction balanced (IB) receiver coil.


Have a great day,
Chet

[green]

That is absolutely correct. The resolution you require is really a ratio of the largest signal to the smallest, AT THE PLACE YOU INTEND TO DO THE DIGITIZING. This can be measured using the techniques described in the preamp thread.
If you for example want to acquire all of the data out of a preamp that has +/-5 volt supplies and a measured noise level of say 500uV then you would need 10V/500uV or a 20000 to 1 ratio. Each bit of an a to d converter can describe a 2 to to1 ratio. This means a 15 bit a to d would be required because it has 2^15 (32768 descrete digitizing levels. You should always have at least 1 bit more ( double the resolution) than you need due to the +/- 1 bit uncertainty inherent in all digital processes. All this assumes theoretical ideal processes, no signal averaging or other processing, and that you want to digitize the whole signal which you may not. In the case that you are willing to ignore the larger parts of the signal you just reduce the max signal to that you really want to see. For example if the signal is clipped by diodes at +/-.7v. This would reduce the resolution requirements by a factor of 7 or roughly 3 bits. One caution, if you put more that the rated signal into an a to d it is likely to misbehave or even be destroyed. It is up to the user to determine what those behaviors are limits are...
Of course signal processing after the a to d can dramatically alter the effective resolution. For example, if you average the samples 8 times you effectively add 3 bits of resolution (2^3) AND you reduce the noise by a factor of the square root of 8. This assumes the noise is not synchronized with the sampling (is asynchronous). There are other processes you can use lie boxcar integration and digital filtering, even adaptive filtering, and correlation that can bring the signal out of the noise. These can get quite complicated but a quick Google search will yield a lot of data.

Digital sampling and filtering hold great promise to improve the performance of the detector. MAYBE WE SHOULD ADD ANOTHER SUB TOPIC TO THIS THREAD TO MOVE THIS DISCUSSION TO.

[If you for example want to acquire all of the data out of a preamp that has +/-5 volt supplies and a measured noise level of say 500uV then you would need 10V/500uV or a 20000 to 1 ratio.] I'm still missing why you give a noise level without referencing it to the input or include a gain. A preamp with a gain of 10 or 1000 would make a difference finding the nickel at 15 inches with a 15 bit a to d converter. The noise after filtering should to be referenced to a frequency band that includes the target signal. Analog or digital filtering. Some low noise op amps include a 10 second .1 to 10 Hz scope trace in there spec sheet. I'm thinking the target signal is less than 10 Hz so something similar would make sense, maybe not.

[Old cart]

What does system noise mean without knowing system gain? If system gain is 10000, system noise 5 mv p-p. Noise is .5 uv p-p at input. The target defines resolution and noise level. Detecting a US nickel at 15 inches with a 8 inch diameter coil with 1 amp peak current requires 1 uv resolution referenced to input(coil volts). Higher peak current or a larger diameter coil could reduce required resolution.

That is absolutely correct. The resolution you require is really a ratio of the largest signal to the smallest, AT THE PLACE YOU INTEND TO DO THE DIGITIZING. This can be measured using the techniques described in the preamp thread.
If you for example want to acquire all of the data out of a preamp that has +/-5 volt supplies and a measured noise level of say 500uV then you would need 10V/500uV or a 20000 to 1 ratio. Each bit of an a to d converter can describe a 2 to to1 ratio. This means a 15 bit a to d would be required because it has 2^15 (32768 descrete digitizing levels. You should always have at least 1 bit more ( double the resolution) than you need due to the +/- 1 bit uncertainty inherent in all digital processes. All this assumes theoretical ideal processes, no signal averaging or other processing, and that you want to digitize the whole signal which you may not. In the case that you are willing to ignore the larger parts of the signal you just reduce the max signal to that you really want to see. For example if the signal is clipped by diodes at +/-.7v. This would reduce the resolution requirements by a factor of 7 or roughly 3 bits. One caution, if you put more that the rated signal into an a to d it is likely to misbehave or even be destroyed. It is up to the user to determine what those behaviors are limits are...
Of course signal processing after the a to d can dramatically alter the effective resolution. For example, if you average the samples 8 times you effectively add 3 bits of resolution (2^3) AND you reduce the noise by a factor of the square root of 8. This assumes the noise is not synchronized with the sampling (is asynchronous). There are other processes you can use lie boxcar integration and digital filtering, even adaptive filtering, and correlation that can bring the signal out of the noise. These can get quite complicated but a quick Google search will yield a lot of data.

Digital sampling and filtering hold great promise to improve the performance of the detector. MAYBE WE SHOULD ADD ANOTHER SUB TOPIC TO THIS THREAD TO MOVE THIS DISCUSSION TO.

[green]

In addition a to d resolution it is very important to consider the system noise level. For example if the system has 5 mV P-P noise it does little good to have 5 uV of resolution. Note also that few a-d converters sample fast enough to be able to take multiple samples during a typical 10 uS sample window. This may mean you will have only one sample taken in that period. Worse, if you can not control precisely in time WHERE that sample is take it wil be difficult to make any sense of it at all.
While it is possible to sample the whole receive, and even transmit, waveforms of a PI it seems it would be very expensive to do this with enough time and amplitude resolution to be of great value. Probably better to sample after the integrator with a slow 16 bit converter. This eliminates timing problems and also noise is lower there.

What does system noise mean without knowing system gain? If system gain is 10000, system noise 5 mv p-p. Noise is .5 uv p-p at input. The target defines resolution and noise level. Detecting a US nickel at 15 inches with a 8 inch diameter coil with 1 amp peak current requires 1 uv resolution referenced to input(coil volts). Higher peak current or a larger diameter coil could reduce required resolution.

[Old cart]

http://www.geotech1.com/forums/showt...875#post207875

Trying to learn something about using an adc. Searched to find the sample resolution needed for a PI detector. Five micro volts was mentioned as probably not being enough in the above thread. What would be a good resolution to try for?

In addition a to d resolution it is very important to consider the system noise level. For example if the system has 5 mV P-P noise it does little good to have 5 uV of resolution. Note also that few a-d converters sample fast enough to be able to take multiple samples during a typical 10 uS sample window. This may mean you will have only one sample taken in that period. Worse, if you can not control precisely in time WHERE that sample is take it wil be difficult to make any sense of it at all.
While it is possible to sample the whole receive, and even transmit, waveforms of a PI it seems it would be very expensive to do this with enough time and amplitude resolution to be of great value. Probably better to sample after the integrator with a slow 16 bit converter. This eliminates timing problems and also noise is lower there.

[Teleno]

http://www.geotech1.com/forums/showt...875#post207875

Trying to learn something about using an adc. Searched to find the sample resolution needed for a PI detector. Five micro volts was mentioned as probably not being enough in the above thread. What would be a good resolution to try for?

The resolution is given by the noise level. If your output noise is 2mV then 11 bits is more than enough (with a 5V reference it resolves 2.4mV). You go for a standard 12-bit ADC and scrap the least significant bit.

[green]

http://www.geotech1.com/forums/showthread.php?22655-Sampling-question-in-PI-design&p=207875#post207875

Trying to learn something about using an adc. Searched to find the sample resolution needed for a PI detector. Five micro volts was mentioned as probably not being enough in the above thread. What would be a good resolution to try for?