Demodulator window widths

[Anonymous]

I recommend demodulator window widths of 10 us, 50 us, and multiples of 50 us, and not other timings unless the application actually requires it. If other window widths are desired, I recommend producing their equivalents by adding or subtracting demodulated signals.
The purpose of these timings is to minimize in-band electrical interference.
It is impossible to know in advance what kind of electrical interference will be present. However, three major sources that concern us are LORAN-C at 100 kHz, television raster at 15.7 kHz, and military submarine communications systems the most powerful of which which are clustered in the range of 17 to 25 kHz.
Multiples of 10 microseconds creates a notch filter at 100 kHz, getting rid of Loran-C. Multiples of 50 microseconds creates a notch at 20 kHz, which is broad enough to help out substantially with raster and subcom.
The delay between demodulator windows the demodulated signals of which will be added or subtracted (for instance discrimination and/or earth field cancellation) should also be considered, as should total sequence time. Proper attention to these details will result in maximum cancellation of the most probable interfering signals.
--Dave J.

 

[Anonymous]

Dave,
I made my earlier experimental PI's with adjustable demodulator windows. I use sample and hold's rather than integrators. I found that without any question at all, a 1uS sample window or less not only loweres the noise but actualy increases the detectors sensitivity.
I also use differential detectors which have an excellent common mode rejection ratio. Eric Foster put me on to another trick which can be a life saver. This is to make the transmit frequency variable.
The result is that my prototype detectors work fine in the lab. I must admit that TV's can cause a little noise but not enough to worry about. A good tip for working in the lab is the use of a small coil. I made a four inch coil for bench testing. The coil picks up a nickel at 7 -8 inches and due to it's small size, does not pick up a lot of noise.

 

[Anonymous]

Thanks for your alternative viewpoint, Dave E.
If you're really going for the tiny stuff and pushing up close to the flyback, then a sample shorter than 10 us would be expected to buy you sensitivity.
In most cases, esp. in urban areas, the primary sources of interference will not be at 100 kHz and in the 15.7-25 kHz band, and so using 10 and 50 us windows will not offer any improvement from that standpoint. Nor is it possible to say in advance what window width might be optimum.
Using a sampling window of at least several microseconds width provides some integration which will reduce interference from sources above (say) 100 kHz, as compared to (for instance) 1 microsecond samples. This is not so relevant if antialiasing filters are provided.
In the case where short samples are being taken (for instance if the signal is being A/D'd for demodulation in software), it is customary to add and subtract samples at certain delays in order to provide the functions of earth field cancellation and ground cancellation and/or discrimination. Given the case where the primary interfering signals that one might be able to do anything about are at 100 kHz and 20 kHz, samples which will be subtracted at equal gain should be 10 us and 50 us apart, or multiples thereof. Samples which will be summed at equal gain should be an odd multiple of 5 us, and/or an odd multiple of 25 us.
You mention "differential detectors". I'm presuming you mean differential (far field null) loops, which of course do knock out most electrical interference.
Synchronous interference that can be minimized by shifting frequency is a common enough occurrence that I think frequency shifting ought to be offered as a standard feature on high end metal detectors, especially pulse which is generally more vulnerable to interference than VLF. It's not an uncommon feature on VLF units.
--Dave J.

 

[Anonymous]

Hi Dave's
One of the best moves I ever made was to put a pulse frequency control on my detectors. It only needs a small shift to minimise the beat with local repetitive signals. Fixed frequency PI

 

[Anonymous]

Hi Eric,
Yes sir, the spacing between the + sample and the - sample form a comb looking filter with notches occuring at multiples of the sample spacing. Of course another filter which would be easy to forget is formed from the - sample to the next + sample (across the transmit pulse) unless this spacing is exactly the same as the + sample to - sample spacing after the transmit pulse, which might pass the frequencies the other effective filter notches out.
Amplitude Modulated (AM) radio stations seem to be what I pick up the best (560,000 HZ to 1,600,000 HZ), they are strong almost everywhere. The good news is they can be (must be) filtered out to some extent before demodulation.
JC

 

[Anonymous]

Hi everyone,
When one builds a two sample PI receiver, subtracting the second pulse from the first pulse forms a filter. If all the pulses are spaced evenly apart, and the sample width (yet another filter) is small compared to the spacing (actually doesn't have to be), then the first notch occurs at 1/(2*(spacing time)).
In other words if the sample spacing is 100us apart, the first notch occurs at 5,000 Hz.
The second notch is at 10,000 Hz. At 7,500 Hz the gain is two because the two samples are reinforcing. (Drawing a picture of this will help)
This filter also starts at zero for DC and ramps up to the first maximum at 2,500 Hz.
The sample width filtering doesn't follow this formula and has to do with integrating the signal over the period of its width, and is of higher frequency.
Anyway, this maybe of interest to anyone who has target information they are trying to recover at 5 KHz, or doing the fourier analysis thing.
If your sample spacing is aperiodic as it often is then this gets messier.
JC

 

[Anonymous]

[No message]