Jim Koehler's report

[Sam Scafferi]

Forget all the rhetoric and philosophy for the moment, and answer a very simple question.

What possible influence could the output of the 555 have on distant targets, and why did you lie in your advertising stating that the device was transmitting three separate frequencies --when it clearly is not?

Suppose a judge in a courtroom ask you that question, what would you tell him/her?

[Jan]

Re: Lack of Credit

Dell,
I have thought long before I decided to write this post.
I am against personal attacks on forums, but feel that I have to speak my mind now.
------------------------------------------------

I far as I am concerned you get all the credit that is due to you.

The article on the VR-800 and the answers you gave on this forum has proven to me that you are very untrustworthy and run a business that will sell expensive useless junk.

You do not dare to answer Sam's questions.
You do not even admit that the VR-800 is useless
Instead you put a link to something you call a Ground Penetrating Radar.

This shows me that you will shamelessly continue your scams.

Your way of dealing will hurt the LRL comunity no end, because it gives the Nonbelievers all the ammo they need to prove that LRL will not work.

So, the credit you get WORLDWIDE is that of a man preying on trustworthy people by selling them useless junk for a lot of money.

Yes Dell, one gets the credit one earns.

Jan

[Carl-NC]

Jim Koehler's report

I've just uploaded a fresh version of Jim Koehler's report on proton mags, courtesy of Jim. Go to the Mag-Info section of this web site.

- Carl

[Willy Bayot]

Re: Jim Koehler's report

This is a note for all those on this forum who have got a copy of this document from me while it was off-line.
This is now the version 2, corrected by Jim. Take it and compare with the previous version. You will have some good surprises as it is even better than before.

Thank you Jim and Carl,

Willy



That's the exact link.

[armin]

Re: Jim Koehler's report (fourier analysis)

The document of Jim is definitely the best descriction of PPM. But it contains one common mistake concerning Fourier analysis, I would like to clarify.
It is correct to say, that the frequency resolution is invers proportional to the measuring time.
A measuring time of 0.5 seconds leads to a frequency resolution of 2 Hertz.
But it is wrong to say that the resolution is equal to the ability to determine a frequency.

Resolution only means the width of the frequency response of a pure sine. Determing the frequency means finding the maximum of this frequency curve. It is obvious, that this task can be done much more exact than the width of this curve. If you want to get this true resolution curve, you cannot use FFT, but the old fashioned discrete fourier analysis, where you can use a frequency raster as narrow as you want. You can limit this analysis at the frequency range of interest without problems.
The limiting factor of determing the maximum of a frequency curve is the noise, which deforms the curve. The advantage of a longer measuring time is the more narrow freqency distribution, which contains less noise.

Armin

[Willy Bayot]

Re: Jim Koehler's report (fourier analysis)

I completely agree with these statements.
I am just doing some real experimentations of a frequency bin interpolation algorithm of mine with sine wave signals mixed with an increasing level of backgound, white noise. I should then be able to make the inverse relation between the resolution one could expect from such an algorithm and the level of noise (or better said, the SNR).
But it is also true that one can not expect miracle from a signal with a low SNR. It is also impossible to get a good resolution with the zero-crossing counting/timing method with such a low quality signal.
That's the reason why the analog stages of the PPM are so important in terms of SNR. Garbage In, Garbage Out.

Willy

[Nick]

Re: Jim Koehler's report - calculating frequency

Further to what Jim says about calculating the frequency by measuring the time period of a fixed number of cycles, isn't it possible to reduce the effect of noise even further by the following method (my example uses a measurement period of 512 cycles):-

instead of timing the 1st and 512th zero crossing and then calculating the period of one cycle from (T512-T1)/511

time every zero crossing and add up the first 256 times in one buffer (32 or even 40 bits are necessary to hold the large numbers) and the next 256 in another. Then subtract the first buffer from the second :-

(T257+T258+..+T512) - (T1+T2+..+T256) =
(T257-T1)+(T258-T2)+...+(T512-T256)

you now have the sum of 256 measurements of 256 cycles, so dividing by 256*256 (65536) gives the time period for a single cycle but with an eightfold reduction in the expected error due to the small noise errors in each individual timing (averaging 256 measurements gives a 16 fold improvement but each measurement of 256 cycles has ~2 times the error of the first method due to the smaller number of cycles so the result is 16/2).

I know it seems too simple to be true but if the underlying sine wave (the proton precession signal) is regular in its period and the error comes only from the noise at each individual timing then the method seems to work.

[Willy Bayot]

Re: Jim Koehler's report - calculating frequency

That's a good idea which should be tried in practice. However, there is a small problem as most of the micro-controllers do not have this amount of RAM (at least, 2Kbytes in this case).
I think it would require an external SRAM module to be able to apply this algorithm.

Willy

[Nick]

Re: Jim Koehler's report - calculating frequency

It doesn't need extra RAM actually - it can all be done in registers if you add the numbers up as you go along, there's plenty of time between each zero-crossing to add the measured time to the buffer. So far I've done a test program for an AVR 8535 running at 8Mhz and I used 5 registers for each buffer to get 40 bits all together - but my preamplifier stage is still too noisy so I've only "measured" a test signal !

[Jim Koehler]

Re: Jim Koehler's report - calculating frequency

I've been thinking about this proposed method and don't see any problems with it. One the one hand, it seems like something for nothing but, on the other, I can't see why not.

I do have a small quibble. Using the same figures, measuring the time for 512 zero-crossing is essentially taking the average of 512 measurements of a single cycle so the accuracy will be about root 512 times better than that for a single crossing - about 11 times. Using the proposed algorithm, each measurement of 256 cycles is therefore about 8 times better than a single measurment and then averaging 256 independent measures is another factor of 8 giving about a factor of 64 improvement over a single measurment. Therefore, this method is about 6 times better than just measuring over 512 cycles. This is a SIGNIFICANT improvement.

There is no problem in implementing this algorithm in the time available (for most modern micros, anyway).

[Jim Koehler]

Re: Jim Koehler's report - calculating frequency

Oops. I don't know what type of new math I was using but the improvement, using Nick's technique, should be root(256) times root(356) all divided by root (512) which comes to something like 11, not 5! That is, the errors should be reduced by a factor of 11.

[Nick]

Re: Jim Koehler's report - calculating frequency

I think the something for nothing comes from including more measurements in the calculation in such a way that the errors tend to average out, basically the more independent measurements the better. I've made a spreadsheet that does a simulation of 100 sets of 512 cycles and I usually get a standard deviation for the 100 estimates of the cycle time between ~7 and ~9 smaller than those calculated using the standard formula.

My main worry about using the method in practice is whether in fact some of the error is coming from microfluctations in the earths magnetic field, this may not matter much though in a gradiometer as it would affect both sensors. Also my calculation doesn't take into account the increasing size of the errors.

If you or Willy have a working mag you can try it out on I can send you my test AVR code to look at if you wish.

[iftikhar]

Re: Jim Koehler's report - calculating frequency

hi,
i am working on differential magnetometer which is working plz send your code for verification
thanks

[Nick]

Re: Jim Koehler's report - calculating frequency

Sorry about the delay in replying - I'm on holidays now and unfortunately I didn't have time before I left to get my code from the PC I use for programming micros. I'll be back in late January.

[Willy Bayot]

Re: Jim Koehler's report - calculating frequency

To push this method one step further, I think it would also be possible to reject from the accumulation, the periods whose duration is too far away from the current running average of this measurement cycle. These would probably be spurious or missing zero-crossings due to noise. We would obviously need a count of good values to be used at the last phase of the calculation instead of the 256 constant.
What do you think about that?

Willy