Filtering out ground signal

[Anonymous]

One way to remove some of the ground signal is by subtracting off an old value that was saved before the target signal arrived. This removes the DC component of the ground signal. The purpose of this example is to show that it is possible to remove even more of the ground signal.
For a target signal I am using the function (sin t)^2. For a ground signal I am using 1+.06sin t/2. The DC part of the ground signal is 1 and the AC part is 6% of the DC. The frequency of the ground signal is about 25% of the frequency of the target signal.
Filter 1 just subtracts the DC part of the ground signal.
Filter 2 is shown in a partial diagram below. There are 3 sample and holds that are clocked on each pulse. So they hold data from the last 3 pulses. The resistor values determine the filter coefficients. The filter function is out(t) = -in(t) +2*in(t-1) - in(t-2).
There are 3 graphs. The left graph shows the ground signal and a target signal that has one half the size of the ground signal. This is just to show the shapes of the two signals. In the other graphs the target signal will be .01 times the ground signal size.
The middle graph shows the .01 target signal and the ground signal after being filtered by filter 1.
The right graph shows the .01 target signal and the ground signal after each of them have been filtered by filter 2. Notice that the ground signal in this graph is much smaller than in the previous graph. This filter alters the shape of the target signal so both the X and R signals would have to go through identical filters before calculating an ID.
Filter 2 is very sensitive to noise, so this may not be a practical filter. It was just the simplest filter I could think of that would significantly reduce the ground signal.
Robert

 

[Anonymous]

So, is the concept to continously save an information stream, discarding only when the next few pulses produce the same info, and only save for comparison when there is an anomaly, such as a target would produce. Then the saved info can be id'ed as the ground reference, and the target info can be further analyzed after the ground signal has been deleted from the target signal.
This is somewhat like the "next neighbor " interpolation formula used in digital cameras. The surrounding pixels are compared to each other, and used to fill in the areas that contain
empty space.As the ccd chip is scanned progressively from one side to the other, the
interpolation takes place.
A detecdtor could digitaly compare the neighboring pulses of a sweep, disguarding all data that is the same, only saving data what is followed by a signal of significant deviation from the somewhat stable stream obtained from ground that contained no target.Analyzing only data streams that contain signals having large enough fluctuations to indicate a possible target.
Is it possible to use a sensor on the r/l sides of the search head to signal the direction of sweep, and use a progressive scan technique to compare all data recieved up until a target signal,then use the data just before the target to use for reference , allowing the target to be analyzed in a more isolated manner. If done digitally, a profile for various metals could be used to offer probability of target, based on decay time and strength.
I know nothing about detectors compared to others on this forum, I just saw a sumilarity to another method of gathering information that seemed like it might have some use at some point.
Feel free to ask me to just watch and learn . This may not even be possible without a MIcro-Chip type company's resourses.
Thanks
Brad,... neophite and just really interested.

 

[Anonymous]

Brad
Your first paragraph is a reasonable description of what happens with filter1.
Signal processing in the time domain generally involves remembering some information from the past and combining the current signal with past information to extract what you want from the signal. The type of filter you want determines what you try to remember and what you do with it.
There are a lot of similarities between time domain signal processing and image processing. One difference is that in the time domain you usually only have access to past signal information, not future information. So in the time domain you can usually only look left at older information, while in image processing you can look to the left and the right of the pixel you are working on. In fact, since images are usually 2 dimensional you can look left, right, up, and down.
In image processing terms, filter 1 is like using a levels or curves adjustment on a hazy picture and moving the black point in to reduce the haze. Filter 2 is like using a high pass filter.
The length of time that metal detectors can remember most of the information is on the order of a second or less. So your progressive scan idea is a level beyond what current detectors can do (if I understood what you were talking about). Metal detectors may have some longer term memory, but they don't remember large quantities of information like a camera does.
Robert

 

[Anonymous]

Hi Robert,
Is not the same result achieved by using a simple high pass filter after the integrator in a PI? This eliminates the dc component and by having a variable R, you can if needed, shorten the time constant to increase the attenuation of the low frequency ground response. I have experimented with a second filter after the first, which gives further improvement and gives a very fast target response provided you keep the search coil moving at adequate speed.
Eric.

 

[Anonymous]

The fact that a PIB detector has seperate "R" and "X" channels allows us to use the exact same analog filtering methods that motion VLF discriminators use. Simply add or subtract a portion of the "X" to the "R" signal to provide a complex "X" channel, filter this and the "R" channel in two seperate filters, and apply them to a synchronous detector to obtain an output.

 

[Anonymous]

Eric
Yes, as long as there is no clipping or any other non-linear operation, You can put the filter after the integrator. However, it would take a third order filter after the integrator to get the same results as a second order filter before the integrator. And unless the filter can exactly undo what the integrator did the ID you get will not always be the same as the one you would get from before the integrator.
I should have made a distinction earlier about the signal you listen to and the signal you get the ID from. The filter that does the best job of removing ground signal may not produce a signal that is suitable for listening to. So for the best ID you probably need separate paths for the audio signal and the ID signal.
The graphs below show typical X and R signals from a coin and a bottlecap (with no ground signal present).
In the case of a coin the X and R signals usually have the same shape, so at any point that you calculate an ID from X and R you get the same results. In this case if you integrate samples taken along these curves you should get the same ID as you get from any point along the curves.
In the case of the bottlecap the X and R signals have different shapes and the ID is different at different times. When the coil is not over the target you get a -X which indicates iron. When the coil is directly over the target you may get a 0 or +X which looks like a conductive target. For this case if you integrate the signals, you are adding together samples from different times, and the integrated ID is going to be different than an ID from the raw signals. The ID may still be usable. In fact, for the bottlecap the integrated ID is probably more desirable than an instantaneous ID taken at the peak of the R signal because it looks more like an iron ID. But the point is that unless the filter exactly undoes the integration, an ID obtained from the integrated signal will be different than an ID from the raw samples.
For the audio signal I would think that a critically damped second order high pass filter would give a better result than two first order filters.
By the way, I assume you use a low pass filter to do the integration. What is a typical time constant for the integrator?
Robert

 

[Anonymous]

If the time constant of the integrator is short relative to the frequency of the ground signal you can forget what I said about needing a third order filter after the integrator.
As long as the frequencies of the ground signal are well below the cutoff frequency of the low pass integrator the ground signal will not be integrated. And a second order high pass filter after the integrator will accomplish the same thing as the one I put before the integrator.
Below is a plot of the frequency response of three different high pass filters. The black curve is a single pole RC filter. The blue curve is two identical RC filters in series, which has a sharper cutoff than the black curve. The red curve is a two pole Butterworth filter which has a sharper cutoff than either of the others.
The catch with the Butterworth filter is that the standard form for an active filter has two resistors that depend on the cutoff frequency. So making a variable SAT filter out of it is a problem. I don't know enough about active filters to see any easy way to adjust it with a single R.
Robert

 

[Anonymous]

Robert,
New technology comes to the rescue. There are now a number of digital potentiometers and dual potentiometers available. These pots take the shape of a chip. Although most of them are designed for use with a microcomputer, many of them can be used without one. One last problem for the home constructor is that most of the chips available are surface mount components.
Another good way to tune a filter is to use transconductance amplifiers. 20 years ago, I used to tune various filters using two CA3080's. Those near a library can check out the famous Active-Filter Cookbook by Don Lancaster. Chapter nine covers the use of the CA3080. There are newer transconductance chips than the CA3080 but for the 4Hz max ground signals we are dealing with the old CA3080 will work just fine.
As a footnote, a CA3080 and an op-amp makes a great voltage controlled gain amplifier with a very wide control of gain. How about a PI with an automatic gain control or AGC such as the circuits that radio receivers use?

 

[Anonymous]

Dave
I had not gotten this far yesterday, but it turns out that both variable resistors are proportional to 1/fc. So a conventional dual pot can be used to control, the cutoff frequency.
Robert

 

[Anonymous]

Also, if the time constant of the integrator is short relative to the frequency of the target signal you can forget what I said about the ID changing.
As long as the frequencies of the target signal are below the cutoff frequency of the low pass integrator the envelope of the ground signal will not get integrated, and the integrated signal should give the same ID as the raw samples.
Robert

 

[Anonymous]

Robert,
Dual linear pots are hard to find. It is harder still to find them with a good tolerance and tracking. You can find them but they will be very high end parts and they will be expensive.