PI Max Frequency?

[xycy]

Did you make it with 10khz or 5khz TX frequency? I have a problem with stability of the frequency now on ~500hz working with batteries. What would be on 10khz?.... It needs quartz generator to stop the drift. Is somebody build such scheme?

[mschmahl]

@xycy i had make it with 5kHz on Surf-PI Pro without quartz and no problems with drift. I think it is PPS (output of 555).

[Tinkerer]

Did you make it with 10khz or 5khz TX frequency? I have a problem with stability of the frequency now on ~500hz working with batteries. What would be on 10khz?.... It needs quartz generator to stop the drift. Is somebody build such scheme?

I use a PIC running at 20MHz for the timing. The timing accuracy is 200ns.

Tinkerer

[Midas]

High frequency PI.
Rather we should say high pulse repetition rate (PRR) PI. Or high pulse per second PI (PPS).
Here is one way to look at it:
If I use 10,000 PRR, I get a signal response of: 1782 for a target of 10us TC
267 for a target of 100us TC
162 for a target of 500us TC
For large deeper targets, I want to increase the TX power, so I make a longer TX time and use 5,000 PRR
Now I get a signal response that is near double, at 3440 for a target of 10us
593 for a target of 100us
349 for a target of 500us
Keeping the same fast response time, at 50ms, I can integrate or stack 250 samples or 500 samples.
The more samples we stack, the better is the S/N. But this equation is not linear.
Would somebody be so kind, to calculate and post the mathematical formula that shows the “point of diminishing returns”?
What I mean to say, up to a certain amount of stacked samples, there is a considerable amount of improvement in S/N. After that, increasing the amount of integrated samples, produces an ever diminishing amount of improvement.

Tinkerer

Hi Tinkerer,

Noise reduction from averaging multiple samples is a function of the sqrt of the number of samples. So say you had an initial peak to peak noise of 1000, and then averaged 500 samples you could reduce the noise to 1000/sqrt(500)=44.72 a pretty significant reduction. But double the number of samples to 1000 and you still only get down to 31.62, as you say very much diminishing returns. You can't exactly calculate 'the point' where increased PPS is no longer worth it though, since its dependent on the penalties your paying in terms of TX strength.

Midas

[Sean_Goddard]

Hi Eric, Good to see you back in the land of the living.

All, from what I see here I can use an analogy of the Petrol V. Diesel engine development over the last few years, the latter has come along FAR more than the old gasoline engine has. Also look at Valves and transistors, I believe the first junction diode was invented way before the valve (tube to our US cousins) but the transistor won out. I see the same happening here, VLF is dead, and has gone as far as it can go (even Deus) unless someone realises what REALLY using a DSP can do but the PI will be pushed further and further until the members of THIS FORUM, NOT any company like Limebin etc, will crack all aspects of what a PI can offer.

Collectively, the people on this forum form the most knowledgeable and skilled development team known to man and Eric is right in what he says. It REALLY is a joy to come here and read what everyone is working on and, best of all, forum members do it to help others FREE instead of making vast profits BRAVO ALL !!

[Tinkerer]

Hi Eric, Good to see you back in the land of the living.

All, from what I see here I can use an analogy of the Petrol V. Diesel engine development over the last few years, the latter has come along FAR more than the old gasoline engine has. Also look at Valves and transistors, I believe the first junction diode was invented way before the valve (tube to our US cousins) but the transistor won out. I see the same happening here, VLF is dead, and has gone as far as it can go (even Deus) unless someone realises what REALLY using a DSP can do but the PI will be pushed further and further until the members of THIS FORUM, NOT any company like Limebin etc, will crack all aspects of what a PI can offer.

Collectively, the people on this forum form the most knowledgeable and skilled development team known to man and Eric is right in what he says. It REALLY is a joy to come here and read what everyone is working on and, best of all, forum members do it to help others FREE instead of making vast profits BRAVO ALL !!

I would like to add to that, that the biggest boost to the development of PI technology has been given by Eric Foster, with his "PI Technology Classroom".

Many thanks, Eric.

Tinkerer

[xycy]

Tnx to all.
@mschmahl
Did you use SM with batteries? Im working with it almost every day and I tune the delay many times. I supose that is the frequency drift, I haven't freq. meter with SM. Its too unlikely to be anything else.
Regards.

[mschmahl]

@xycy
I use 8x1,5 V batteries (12V). If i set my delay, than i have no problem´s with drift. Can it be, that you have fast changing soil where you use your SM? After my changes on SM i had messured the frequency. I don´t have a freq.-meter with my SM.

[Midas]

I would like to add to that, that the biggest boost to the development of PI technology has been given by Eric Foster, with his "PI Technology Classroom".

Many thanks, Eric.

Tinkerer

I'm glad I could bump the thread up for you guys with my boring post so we can all remember to thank Eric. Thanks Eric!

[Tinkerer]

Hi Tinkerer,

Noise reduction from averaging multiple samples is a function of the sqrt of the number of samples. So say you had an initial peak to peak noise of 1000, and then averaged 500 samples you could reduce the noise to 1000/sqrt(500)=44.72 a pretty significant reduction. But double the number of samples to 1000 and you still only get down to 31.62, as you say very much diminishing returns. You can't exactly calculate 'the point' where increased PPS is no longer worth it though, since its dependent on the penalties your paying in terms of TX strength.

Midas

Midas, thank you for the answer.
I just found that my reply to your post got somewhere lost in the clouds. So here I try again.

The information you gave is of great help. However, I still need more help.

I am trying to find the sweet spot, where we get the most efficient TX.
What I mean with the "most efficient TX", would be the TX that is:
-the most power efficient
-excites the highest target response
-gives the best S/N

There is always a certain amount of noise present. If our target response is 1/1 with the noise, we can hope to extract the desired signal with considerable RX processing.
If we manage to get a S/N, of 10/1, obviously we have a much better situation. Therefore it is interesting to have a strong TX.
If I look at the amount of Joules in the TX, it does not make much difference if these Joules were produced by a higher voltage TX, or a longer TX pulse time. The target response is about the same.

Low PRR, show a lot of irregularity in the target response. We can call this noise. I do not know exactly why there is this noise, but this is why we integrate/average many samples.

Looking at the above numbers, we could say noise = 1000, signal = 1000, or S/N = 1/1. Integrating 10 samples S/N = about 3/1 This would be a reasonable response (100ms) detector at 1000 PRR.

Running at 10,000 PRR, we could integrate 100 samples and get a S/N = 10/1.

However, the 10,000 PRR come at a cost. More power consumption. Less time between a first and a second sample, which reduces the relative target response amplitude for longer TC targets.
We also have shorter sampling time. A longer sampling time does produce some degree of integration for HF noise which is to our advantage.

If we use a higher voltage TX, we get more Joules into the coil in a shorter time, again at a higher power consumption.

Of course there are limits too. Maximum Flyback voltage and battery weight just to name the most obvious.

But, somewhere there, there is a sweet spot, where the best compromise is happening, for a specific purpose detector.

It would seem that the "best PI detector" should aim at the best sweet spot.

The morale of the long story? It takes time and work to find that sweet spot. If I do it alone it may take me one year. If a few of us work on it, it will take much less time. When we find that sweet spot, we might be able to define a mathematical formula for the sweet spot.

The formula would then make it possible to calculate the sweet spot for different purpose detectors.

Tinkerer

[Midas]

Midas, thank you for the answer.
I just found that my reply to your post got somewhere lost in the clouds. So here I try again.

The information you gave is of great help. However, I still need more help.

I am trying to find the sweet spot, where we get the most efficient TX.
What I mean with the "most efficient TX", would be the TX that is:
-the most power efficient
-excites the highest target response
-gives the best S/N

There is always a certain amount of noise present. If our target response is 1/1 with the noise, we can hope to extract the desired signal with considerable RX processing.
If we manage to get a S/N, of 10/1, obviously we have a much better situation. Therefore it is interesting to have a strong TX.
If I look at the amount of Joules in the TX, it does not make much difference if these Joules were produced by a higher voltage TX, or a longer TX pulse time. The target response is about the same.

Low PRR, show a lot of irregularity in the target response. We can call this noise. I do not know exactly why there is this noise, but this is why we integrate/average many samples.

Looking at the above numbers, we could say noise = 1000, signal = 1000, or S/N = 1/1. Integrating 10 samples S/N = about 3/1 This would be a reasonable response (100ms) detector at 1000 PRR.

Running at 10,000 PRR, we could integrate 100 samples and get a S/N = 10/1.

However, the 10,000 PRR come at a cost. More power consumption. Less time between a first and a second sample, which reduces the relative target response amplitude for longer TC targets.
We also have shorter sampling time. A longer sampling time does produce some degree of integration for HF noise which is to our advantage.

If we use a higher voltage TX, we get more Joules into the coil in a shorter time, again at a higher power consumption.

Of course there are limits too. Maximum Flyback voltage and battery weight just to name the most obvious.

But, somewhere there, there is a sweet spot, where the best compromise is happening, for a specific purpose detector.

It would seem that the "best PI detector" should aim at the best sweet spot.

The morale of the long story? It takes time and work to find that sweet spot. If I do it alone it may take me one year. If a few of us work on it, it will take much less time. When we find that sweet spot, we might be able to define a mathematical formula for the sweet spot.

The formula would then make it possible to calculate the sweet spot for different purpose detectors.

Tinkerer

Well that's a much bigger question, and not one I'm able to help with yet. I'm still working on the hardware.

I am interested though, assuming you had an army of willing and able volunteers. How do you actually propose to split up the task so it can be run in parallel?
It strikes me that it might be quite an iterative sequential type process and hence quite difficult to divide.

Midas

[Tinkerer]

Well that's a much bigger question, and not one I'm able to help with yet. I'm still working on the hardware.

I am interested though, assuming you had an army of willing and able volunteers. How do you actually propose to split up the task so it can be run in parallel?
It strikes me that it might be quite an iterative sequential type process and hence quite difficult to divide.

Midas

Thanks for the feedback.

Before I answer, I need to correct my mistaken numbers, above. Actually, the numbers are wrong, but the proportions are about right, so I just leave it at that.

The process of finding the sweet spot? Mathematics are not friendly with me, so I must do things the wrong way, by trial and error. This takes years.
The right way is to develop a mathematical formula that takes into account the most important factors, initially, and find an approximation to the real life behavior.

Once the calculations, using the formula, shows a definite similarity to the real life results, then it is a matter of refining the formula, maybe adding some more factors etc. to refine the formula for more accurate results.

Well, this is how I imagine it should work. Of course, I don't know nothing about mathematics, so it is just idle talk.

Which are the most important factors?

Ampere turns
Coil surface area
TX Joules
Power consumption
Number of samples averaged

Some of these are complex, like the joules. We can get the Joules into the coil with TX time or TX voltage. And then the Joules are closely linked with the Ampere Turns.

There are a few more that play a major role and then there are a few that play minor roles.

A happy compromise between all factors produces the sweet spot. Easy, if only I could remember where I left my thinking cap.

Tinkerer

[Davor]

There is a S/N representation that uses energy ratio, Eb/No. Hence twice the current or twice the samples would lead to the same result. Perhaps the way to go is finding the maximum charging current after which you have diminishing Rx increase. Such a knee should be observed for various Rx sample delays, and different Tx transistors, and there you have it. I expect some oohs and aahs and occasional splats on a forehead.

[Tinkerer]

There is a S/N representation that uses energy ratio, Eb/No. Hence twice the current or twice the samples would lead to the same result. Perhaps the way to go is finding the maximum charging current after which you have diminishing Rx increase. Such a knee should be observed for various Rx sample delays, and different Tx transistors, and there you have it. I expect some oohs and aahs and occasional splats on a forehead.

Davor, thanks for the feedback.

The response of a target stops to increase significantly, when the time of the TX reaches 2 target TC. However, this needs a TX pulse ramp that is near linear.
A perfectly linear TX ramp would have many advantages, but I have not found a way to do it, without wasting a lot of power.

There we have a good challenge, create a perfectly linear TX ramp that is also very energy efficient.

Tinkerer

[Davor]

For reasonably small x you'll find sin(x)≈x and it is energy efficient, and already works predictably ... a TEM signal.
Also d/dx of a TEM source has a nice voltage zero crossing for X component detection, and maybe playing with TEM asymmetry. It also has aforementioned sin(x)≈x ramp around 0 and pi that result in a nice voltage plateau prior to a sudden drop.
I played with such exciter soon after I joined this forum. The caveat is that it required high voltage to work properly, and also some means of gate isolation. If I scramble it a bit, I'll have asymmetric TEM that will resemble nowadays PI, but also with energy conservation and perfect cut off.