A
Anonymous
Guest
VLF/MF-FD PRACTICE: NO PROB BOB
Earth field pickup in this class of machines is usually no problem. The signal path is linear (nothing that could modulate quasi-DC signals up into the demodulator passband), and quasi-DC signals are eliminated by of AC coupling and often by full wave demodulation as well.
Demodulation is continuous, so the signals in the channels are synchronous and remain so unless timing mismatch is introduced by (for instance) mismatched capacitors in low pass filters or hardware differentiators. Mismatch can also be introduced by A/D sampling sequences but I presume that most engineers pay enough attention to this detail so that problems do not arise from it.
In principle it is possible to feed the signal from the front end AC amplifiers directly into an A/D converter and do the demodulation in software. However, that is not as easy in practice as it is in principle, so I don't think anyone is doing that yet in commercial high-performance metal detectors.
PULSE INDUCTION (some of this also applies to non-PI time domain machines, and some does not apply to bipolar PI)
In PI, signals are demodulated in a sequence. The most common sequence is target now, earth field balance later. If the demodulated signals are summed/subtracted into the same continuous-time low pass filter, the time relationship between the signals is preserved. This is how it normally is in all-analog PI machines.
PI machines are better suited to demodulation in software than VLF/MF machines. Since the reactive signals are eliminated, high gain amplification can be used, making it possible to raise the noise level high enough to get a good A/D conversion noise figure. The natural tendency of PI to separate things in time lends itself to interrupt-driven A/D conversion, allowing demodulation and low pass filtering to be done in software.
Common sense might say to sample the target signal, then sample the earth field signal, and subtract the latter from the former. NOPE! Because the target signal was saved in memory longer than the earth field signal, when the latter is subtracted from the former, it will create a differentiator. Not much of one, but if the machine has a large loop and has high sensitivity esp. to high-conductivity targets, you're likely to hear earth field if the axis of the coil is subjected to angular rotation.
The same problem occurs when subtracting any other pair of signals, as in discrimination/ID, or reactive ground balancing if a reactive signal has been provided.
Probably the simplest solution to this problem is time-proportioned balancing (interpolation. The following example is concerned only with earth field balancing, but the principle can be extended to discrimination and reactive ground balancing.
Suppose we have an earth field signal which is sampled 700 microseconds after the target signal, and further suppose the sampling sequence is 1 millisecond long. You might think that to be an extreme example, but in an earlier post today I described a system in which such numbers are quite realistic.
We maintain a bucket-brigade for the signal samples. The earth field is the last sample in the sequence. Then a new sequence begins by turning on the transmitter. Sampling can be suspended and numbers can be crunched while the transmitter is on and the receiver is shut down.
In the discussion which follows, remember that we are not looking at signals in real time. We are inspecting and analyzing a historical record.
If we just subtract the most recent earth field sample from the most recent target signal, there'll be a 700 microsecond difference in the earth field being balanced. "Apples will be subtracted from oranges" and the signals will differentiate.
Looking at it from the target sample's point of view, it could have been balanced against the prior earth field sample, which occurred only 300 microseconds prior, rather than 700 microseconds later. That's a better deal, but not good enough. What we do is subtract 0.3 of the most recent earth field sample and 0.7 of the prior earth field sample. By interpolating in this manner, we eliminate unwanted time mismatch in the historical record, restoring synchronization.
What I have illustrated above is a very simple interpolation, which would be followed by low pass filtering and probably at least one stage of differentiation to eliminate offsets. In principle, it would be possible to use long bucket-brigades and do the filtering and time-interpolation simultaneously; however I don't see any advantage in that approach.
--Dave J.
Earth field pickup in this class of machines is usually no problem. The signal path is linear (nothing that could modulate quasi-DC signals up into the demodulator passband), and quasi-DC signals are eliminated by of AC coupling and often by full wave demodulation as well.
Demodulation is continuous, so the signals in the channels are synchronous and remain so unless timing mismatch is introduced by (for instance) mismatched capacitors in low pass filters or hardware differentiators. Mismatch can also be introduced by A/D sampling sequences but I presume that most engineers pay enough attention to this detail so that problems do not arise from it.
In principle it is possible to feed the signal from the front end AC amplifiers directly into an A/D converter and do the demodulation in software. However, that is not as easy in practice as it is in principle, so I don't think anyone is doing that yet in commercial high-performance metal detectors.
PULSE INDUCTION (some of this also applies to non-PI time domain machines, and some does not apply to bipolar PI)
In PI, signals are demodulated in a sequence. The most common sequence is target now, earth field balance later. If the demodulated signals are summed/subtracted into the same continuous-time low pass filter, the time relationship between the signals is preserved. This is how it normally is in all-analog PI machines.
PI machines are better suited to demodulation in software than VLF/MF machines. Since the reactive signals are eliminated, high gain amplification can be used, making it possible to raise the noise level high enough to get a good A/D conversion noise figure. The natural tendency of PI to separate things in time lends itself to interrupt-driven A/D conversion, allowing demodulation and low pass filtering to be done in software.
Common sense might say to sample the target signal, then sample the earth field signal, and subtract the latter from the former. NOPE! Because the target signal was saved in memory longer than the earth field signal, when the latter is subtracted from the former, it will create a differentiator. Not much of one, but if the machine has a large loop and has high sensitivity esp. to high-conductivity targets, you're likely to hear earth field if the axis of the coil is subjected to angular rotation.
The same problem occurs when subtracting any other pair of signals, as in discrimination/ID, or reactive ground balancing if a reactive signal has been provided.
Probably the simplest solution to this problem is time-proportioned balancing (interpolation. The following example is concerned only with earth field balancing, but the principle can be extended to discrimination and reactive ground balancing.
Suppose we have an earth field signal which is sampled 700 microseconds after the target signal, and further suppose the sampling sequence is 1 millisecond long. You might think that to be an extreme example, but in an earlier post today I described a system in which such numbers are quite realistic.
We maintain a bucket-brigade for the signal samples. The earth field is the last sample in the sequence. Then a new sequence begins by turning on the transmitter. Sampling can be suspended and numbers can be crunched while the transmitter is on and the receiver is shut down.
In the discussion which follows, remember that we are not looking at signals in real time. We are inspecting and analyzing a historical record.
If we just subtract the most recent earth field sample from the most recent target signal, there'll be a 700 microsecond difference in the earth field being balanced. "Apples will be subtracted from oranges" and the signals will differentiate.
Looking at it from the target sample's point of view, it could have been balanced against the prior earth field sample, which occurred only 300 microseconds prior, rather than 700 microseconds later. That's a better deal, but not good enough. What we do is subtract 0.3 of the most recent earth field sample and 0.7 of the prior earth field sample. By interpolating in this manner, we eliminate unwanted time mismatch in the historical record, restoring synchronization.
What I have illustrated above is a very simple interpolation, which would be followed by low pass filtering and probably at least one stage of differentiation to eliminate offsets. In principle, it would be possible to use long bucket-brigades and do the filtering and time-interpolation simultaneously; however I don't see any advantage in that approach.
--Dave J.