DC Receiver for PI

[Randy Seden]

I was wondering if anyones tried a Direct Conversion style front end on a PI? The impulses would be detected at a much lower amplitude (I have noticed this on my SSB/AM Amatuer Radio Transceiver over the years).

Kinda like the differance between trying to copy CW in the AM mode vs SSB mode.

Randy Seden

[Robert Hoolko]

Re: DC Receiver for PI

Randy

I am looking at the possibility of doing a direct conversion PI. But I figure it would take a 16 bit A/D at several thousand conversions per second just to equal current machines. Beating them would be difficult.

I am not looking at DC for more sensitivity. I am interested because additional features can be added without adding a lot more hardware. Once you have the basics working, other options can be added just by adding more software (and more memory and CPU cycles).

Robert

[Carl]

Re: DC Receiver for PI

>But I figure it would take a 16 bit A/D at several thousand conversions per second just to equal current machines.

Maybe not. Let's say you sample the signal at the clamp diodes. That's about +/- 0.8V. But we're only interested in one side, so let's say we want to sample a 1v range. A 16-bit converter would only offer about 15uV of resolution, which might not even be enough. Most PI detectors have a preamp gain of 1000, followed by an integrator with a gain of about 500, so we probably want sub-10uV resolution.

But we really don't need to resolve the whole 1V signal decay, just the last few 10's of mV. So we might get away with a 16-bit converter, or maybe even less, if we can set it's full-scale range down low, and we can further clamp the input so the extra voltage range doesn't trash the ADC.

I know, setting the ADC reference down low hurts SNR, mostly through converter DNL. But there might be ways around this as well, like applying out-of-band dither energy that gets filtered later.

Also, you don't need to run the converter continuously, just a short burst in the few uS you want to look at the decay. These bursts can be buffered, and then processed at a lower speed.

- Carl

[Robert Hoolko]

Re: DC Receiver for PI

Carl

I did not mean there would be no amplification before the A/D. In fact I would probably even put an integrator in front of it to integrate 10 usec for each sample. But I meant that there would be no subtractions and no integration across pulses before the A/D. All integration across pulses and subtraction of background and filtering would be done in the CPU.

I am assuming that I would only have to output about 50 times a sec to update the audio. I would take a lot more samples than that and try to pick up at least 3 more bits by integrating the samples.

I was not thinking of taking a lot of samples on each pulse but of having a lot of pulses and taking only a couple of samples from each one.

Robert

[Robert Hoolko]

Re: DC Receiver for PI

I guess my original comments don't make much sense unless you realize that I am not just interested in building a PI but in experimenting with target ID on a PI. I need a lot of bits so I can see a small difference between two large numbers.

Robert

[Carl]

Re: DC Receiver for PI

Why integrate before the ADC? Seems like if you capture a number of samples in a region of interest of the pulse decay, you could then process the samples any way you want, hold on to whatever results from that, then process across the pulses any way you want. Gives more flexibility than analog processing.

Still seems like a preamp is not a must-have in this application. The dynamic of the whole pulse is large, but the dynamic range of the region of interest is not.

- Carl

[Robert Hoolko]

Re: DC Receiver for PI

Carl

I don't see much advantage to putting a microprocessor in a simple PI unit that just integrates the samples and uses that to modulate an audio signal. That is a relatively simple thing to do with analog circuits. Although the microprocessor could provide all the timing the real advantage of the microprocessor comes when you want to do calculations. Detecting the presence of a signal might only take one bit, but doing calculations on the signal generally requires more bits.

I would like to work on PI target ID. With a mono coil that probably means looking at the shape of the decay curve. To detect a target you only have to be able to see a reliable difference in the signal level between when the coil is over the target and when it is not over the target. To get information about the shape of the decay curve I would like about 4 bits more than it takes to detect the target.

Since I have not built or worked with any PI circuits yet I am still guessing about numbers. I figure that the signal from a small target at the center of a mono coil is about 5000 times stronger than when the target is 2 coil diameters away. I would like to handle at least that dynamic range in a single sweep without having to change gain. Twelve bits is almost enough for that, but I would like 4 more for the ID. I don't know what signal levels from the coil are. But VLF detectors work with microvolt signals and PI detectors use more gain than VLF, so I assume we are dealing with sub microvolt signals. I don't see any point in trying to sample the signal without any amplifier at all. That seems much too restrictive.

The plot above is my guess at how signal to noise ratio depends on the sample window width for random noise that is not bandwidth limited and a signal that is a simple exponential decay. This is for integrating the signal over the sample window. The horizontal axis is sample width compared to the time constant of the target. The numbers on the vertical axis don't mean anything, but the higher the curve the better the signal to noise ratio.

I am worried about the left end of this graph where the SNR goes to 0. Of course the bandwidth of the noise is limited by whatever is before the A-D and by the sample and hold. So the SNR does not really go to 0 if you do not use an integrator. It levels off at a value that depends on the bandwidth of everything before the A-D. But I do not know what that bandwidth is. And I do not know how much that bandwidth can be reduced before the target signal starts to get messed up. So to be safe I am assuming a short integrator before the A-D. It is possible that the sample and hold could be made slow enough so that a separate integrator is not necessary. In my VLF project I do not use an integrator, but the bandwidth of my amplifier is much narrower than a PI amp.

You are talking about taking multiple samples on each pulse and combining those samples to improve SNR. I am talking about taking multiple samples on each pulse but keeping them separate to examine the decay shape. In both cases the samples from different pulses would be integrated. I do not expect to have a fast enough A-D and CPU to take many samples on each pulse and group them so samples within each group are combined for SNR but separate groups are kept for shape analysis. It would be nice to have a Mega Hertz sample rate and enough CPU to keep up with it.

I would use a median filter on the samples from separate pulses like the one I use in my VLF project, but I am still worried that each sample would have too much noise because of the wide bandwidth of PI front ends.

Robert

[Robert Hoolko]

Integrator vs Sample and Hold

In the first graph above the red curve represents the target signal decaying. The blue curve represents the integrator sample window. The integrator multiplies the red curve by the blue curve, getting the green curve and calculates the yellow area under the green curve. The size of the yellow area is the output from the integrator.

We often think of an A-D sample as being the amplitude of a signal at a single point in time. But it is not really possible to make an instantaneous measurement. It takes some amount of time to capture the signal. The second graph above represents an A-D sample. The red curve is the signal. The blue curve is the response of the sample and hold circuit. The red curve is multiplied by the blue curve and the yellow area is the signal that is measured by the A-D. I have drawn the blue curve wider than it would usually be for measuring a signal that changes as fast as the target signal does. But it can be made this wide by slowing down the sample and hold. It could be made even wider, but as it gets wider the tail will extend out farther to the left to older signals. Eventually it will run into the ground signal which we would like to avoid.

So far I have been drawing a very clean signal. But we want to be able to read signals right down into the noise. So the signal should be shown looking more like the red curve in the last graph. This is the kind of signal that gets multiplied by the blue curves. You can see that if the blue curve is too narrow there is going to be a lot of variation from one sample to the next (the samples will be noisy). This is what I am worried about when sampling the signal.

Robert

[Robert Hoolko]

Re: DC Receiver for PI

Carl

I am going to take back what I said about not being able to group A-D samples from a single pulse to improve SNR. I was thinking that it would take conversion times of about a microsecond for that to work. But I have been looking at it some more, and I realized that it could be practical even at 5 usec. If the samples within a group are just added together there might not be much load on the CPU. Two samples taken 5 usec apart would help where there is 100 kHz radio interference. That number would not be good where there is 200 kHz interference, but other intervals might work.

When set for a short sample delay (looking for thin rings) two samples is probably the most that could be grouped. But for a long sample delay (looking for copper/silver coins) even more samples could be grouped together. I will give this more thought.

Robert

[Carl]

Re: DC Receiver for PI

I was thinking more of broadband noise. You could integrate a few samples in a short interval -- say, 4 samples in a 1us window, at the normal PI sample delay. However, that only cuts noise in half, not so great.

You can also sample at other delays, to get a look at the decay profile. And integrate each grouping over several pulses. With a judicious selection of pulse rate and sample rate, you can cancel some interference, like you said.

I only favor direct sampling of the raw pulse (I agree a low-noise amp will ease the ADC requirements), before analog integration, simply to allow you to do more varied processing with the signal. Less added noise, too. In radio design, we try to sample as close to the antenna as possible.

- Carl