Missing Link between pulse and continuous waveform detectors ?

[moodz]

A perfect brick wall bandpass filter would be needed, wouldn't it?

Regards,

-SB

yep ... the spectrum of the sinc pulse published earlier is the shape of the passband response of the filter you would need to regenerate the sinc pulse in the time domain. It has a rectangular shape ... fairly hard to do using discrete components ... but numerically using DSP ... a few lines of code.
moodz ...

[moodz]

1. Regarding Aziz's last comment -- what kind of forcing function would be required to make a typical metal detector coil "exhibit" a sync function? By the way, are we talking about the current in the coil, or what?

we want the current in the coil to have a sinc characteristic. Hence the flux .. hence the target excitation ... hence the detected differential .... though the target response does very interesting things to the received sinc waveform in the t-d and consequently the f-d.

2. A delta pulse is really a limiting case of a sync pulse where the rectangular pulse in the frequency domain is infinitely wide and the sync pulse is infinitely narrow. It seems the advantage of the sync signal over a delta pulse is to limit the bandwidth in the frequency domain and increase the amplitude by compressing the power into that range -- this should make working with the data less noisy. The question is - exactly what frequency band would give us the most useful detection and discrimination.

ah well thats the 64k dollar question .... the ideal band is all the frequencies in the f-d that all detector designs to date have used

It is in our interest to use the narrowest band possible to get the best signal to noise, yet still provide discrimination. Wouldn't it be funny if that turned out to be a sine wave!

yep ... of course a single sine wave can be used ... with sinc f-d analysis you can use what I would call 'comparative descrimination' and 'target signature analysis' .... by illuminating the target with a range of possible frequencies you can find the peak response.

To continue the S/N argument, the sync wave forces us to pick up noise proportional to the bandwidth. Because of that, I'll propose: it may be better to use multiple sine waves for discrimination than a sync wave because of superior signal/noise and almost as good discrimination. I know that's old hat, but it seems to make sense from s/n standpoint.

... I am fairly sure that the use of multiple frequencies has been patented by ML ... the sinc method gets around that because the f-d spectrum is continuous ... eg say I did a million point DFT ... there would be a million sine waves ... of course I would have to up the sample bits. As for s/n the sinc pulse is located in time .... unlike a sine wave which is continuous.to increase the s/n on a sine wave I can narrow the filter arbitrarily however as I narrow it the filter will ring and phase distortion will occur ... for the sinc pulse we use synchronous demod which allows the recovery of frequency / phase / amplitude info much like a PLL does for a sine wave.

Regards,

-SB

moodz

[simonbaker]

moodz

Ok. I'm still interested in the problem of making a TX coil current have a sync wave current, because a TX coil has it's own response, so I don't think you can just feed it a sync wave voltage.

Because most TX coils probably are underdamped resonant circuits, they look a little like a bandpass filter around the resonant frequency -- so ironically, hitting them with a square wave/pulse might make a sync-like response.

A follow-up question is: if you apply this idea to a practical detector, when you analyze the RX signal, would you be mostly looking at the overall shape of the power spectrum, or would the details of the phase component be also analyzed? I'm wondering if the phase stuff would be just too noisy and distortable to be reliable.

If you are right about Minelab patenting multi-frequency detectors, I just have to say it is pretty disgusting what can be patented. Patent law says an invention has to not be "obvious", and I remember decades ago when I first thought about making a better MD my very first reflex thought was "multiple frequencies" -- to me, that should be obvious and not patentable by itself. Oh well...

My gut feel is that using a continuous spectrum may not be as efficient as multiple frequencies because very close frequencies have too much redundant information for the extra noise you pick up. But that would be the subject of experiments, and any new idea is exciting to explore.

-SB

[moodz]

Ok. I'm still interested in the problem of making a TX coil current have a sync wave current, because a TX coil has it's own response, so I don't think you can just feed it a sync wave voltage.

Because most TX coils probably are underdamped resonant circuits, they look a little like a bandpass filter around the resonant frequency -- so ironically, hitting them with a square wave/pulse might make a sync-like response.

A follow-up question is: if you apply this idea to a practical detector, when you analyze the RX signal, would you be mostly looking at the overall shape of the power spectrum, or would the details of the phase component be also analyzed? I'm wondering if the phase stuff would be just too noisy and distortable to be reliable.

If you are right about Minelab patenting multi-frequency detectors, I just have to say it is pretty disgusting what can be patented. Patent law says an invention has to not be "obvious", and I remember decades ago when I first thought about making a better MD my very first reflex thought was "multiple frequencies" -- to me, that should be obvious and not patentable by itself. Oh well...

My gut feel is that using a continuous spectrum may not be as efficient as multiple frequencies because very close frequencies have too much redundant information for the extra noise you pick up. But that would be the subject of experiments, and any new idea is exciting to explore.

-SB

Simon ... have a look at this one ... quite a good read actually ... not as obscure as some of their patents.

20070296415.pdf

.... there is another document published my minelab that has nice graphs of transmit waveforms and target reponses .... funny but the target response looks to be a sinc pulse ... maybe they overlooked the obvious.

To answer your question .... the difference between a bunch of frequencies and sinc is that the bunch does not have a pulse characteristic. If you do pulse them ... ie modulate them ...straight away you are introducing phase / frequency distortion. The sinc is an optimally modulated bunch of frequencies .... hence the ideal power bandwidth in the f-d .... now you mentioned the closely spaced frequencies ... the f-d analysis granularity is dependant on the number of bins in your DFT / FFT so if I want only 8 analyis points ... I only have 8 bins ...

Regarding the coil .... hmm .... you can hit a coil with a 'unit impulse' and produce a sinc like response ... however a numerically generated pulse can have equalisation applied to it by way of feedback from the receiver to ensure a sinc characteristic in the flux density .. however this is probably getting too elaborate .... To correlate with PI technology ... if PI can detect subtle shifts in the t-d waveform in uSec resolution ... from a pulse lasting 10 or so uSec .... can the f-d match that ... I would say easily with room to spare ... just on EMI and hum rejection alone the f-d is vastly superior compared to trying to remove it in the t-d.

This is an unproven idea ... I have done physical 'bench tests' ... it seems to offer alot of promise ... only time will tell.

Regards,

Paul.

[simonbaker]

Simon ... have a look at this one ... quite a good read actually ... not as obscure as some of their patents.

[ATTACH]8853[/ATTACH]

.... there is another document published my minelab that has nice graphs of transmit waveforms and target reponses .... funny but the target response looks to be a sinc pulse ... maybe they overlooked the obvious.

To answer your question .... the difference between a bunch of frequencies and sinc is that the bunch does not have a pulse characteristic. If you do pulse them ... ie modulate them ...straight away you are introducing phase / frequency distortion. The sinc is an optimally modulated bunch of frequencies .... hence the ideal power bandwidth in the f-d .... now you mentioned the closely spaced frequencies ... the f-d analysis granularity is dependant on the number of bins in your DFT / FFT so if I want only 8 analyis points ... I only have 8 bins ...

Regarding the coil .... hmm .... you can hit a coil with a 'unit impulse' and produce a sinc like response ... however a numerically generated pulse can have equalisation applied to it by way of feedback from the receiver to ensure a sinc characteristic in the flux density .. however this is probably getting too elaborate .... To correlate with PI technology ... if PI can detect subtle shifts in the t-d waveform in uSec resolution ... from a pulse lasting 10 or so uSec .... can the f-d match that ... I would say easily with room to spare ... just on EMI and hum rejection alone the f-d is vastly superior compared to trying to remove it in the t-d.

This is an unproven idea ... I have done physical 'bench tests' ... it seems to offer alot of promise ... only time will tell.

Regards,

Paul.

I basically like the idea of multiple frequencies and think the idea seems promising. It would be great if each target showed a nice reliable profile in the frequency domain.

My point (I think) is that if you use fewer bins, each bin is wider, and picks up more noise as well. If instead of a spread spectrum you used a bunch of individual line spectra, you could use extremely narrow (well, limited by sample time window) filters (comb filter?) to pluck out the info much more clearly (it would seem).

As for exact sync wave -- how important is it? Or would anything that produced a "pink noise" spectrum work. Because really, we will experimentally look for signatures, does purity of TX signal matter?

I'll be interested in what you find out.

Regards,

-SB

[moodz]

Hi Simon ...

Well I have just ordered an 8 channel 24 bit DSP board to develop the 'real' prototype on .... I will post results.

regards

moodz.

[simonbaker]

Hi Simon ...

Well I have just ordered an 8 channel 24 bit DSP board to develop the 'real' prototype on .... I will post results.

regards

moodz.

I'm envious. There's all kinds of crazy things you can do with that!

-SB

[Aziz]

Hi Paul,

Hi Simon ...

Well I have just ordered an 8 channel 24 bit DSP board to develop the 'real' prototype on .... I will post results.

regards

moodz.

I had a look into some starter kit boards. It is quite expensive and time consuming process. No way! I still go for the laptop and sound card solution.

Which board did you order?

Regards,
Aziz

[moodz]

... the board is called a symphony soundbite .... based on a 56K family DSP by freescale semiconductor ( formerly motorola ).

SYMSBDEVKITFS.pdf

The 56K is a tried and tested DSP platform and the SDK / IDE is free ... eclipse based.

They cost $US150 ... quite reasonable .... cheaper than a similiar 24 bit sound card that is not user programmable and more channels and gp IO to boot.

If it doesn't work out ... it will make a good real time music processor for guitar etc.

what do you think ?

moodz

[Aziz]

Hi Paul,

the soundbite board seems to be interesting. It should help to test some ideas. The 180 MIPS should also be big enough for some real-time FFT and processing. I like the 192 kHz sampling rate of the sound chip. You have more input channels. So you could drive it in differential mode to increase the SNR.

Aziz

[moodz]

Hi Paul,

the soundbite board seems to be interesting. It should help to test some ideas. The 180 MIPS should also be big enough for some real-time FFT and processing. I like the 192 kHz sampling rate of the sound chip. You have more input channels. So you could drive it in differential mode to increase the SNR.

Aziz

Well I got the board .... quite nice .... however I would not recommend it unless you really like assembly language programming. Which I can do ... but takes longer to debug.
moodz.

[moodz]

well .. the board has arrived ( actually 2 of them ) !

Eight channels ( thats eight in AND eight out folks ) of 24 bit 200ksps goodness at 180 MIPS.

FFT in 40 bytes of memory space ... hows that for efficiency !

I was not completely right about coding everything in assembler ... The C compiler took me some time to get going but it runs nice now .... see screen shot below of single stepping the source code running live on the board whilst having access to register contents and mem variables and can even edit and recompile on fly.

This board could be the engine of software driven MD's.

moodz.

[simonbaker]

well .. the board has arrived ( actually 2 of them ) !

Eight channels ( thats eight in AND eight out folks ) of 24 bit 200ksps goodness at 180 MIPS.

FFT in 40 bytes of memory space ... hows that for efficiency !

I was not completely right about coding everything in assembler ... The C compiler took me some time to get going but it runs nice now .... see screen shot below of single stepping the source code running live on the board whilst having access to register contents and mem variables and can even edit and recompile on fly.

This board could be the engine of software driven MD's.

moodz.

[ATTACH]8923[/ATTACH]

[ATTACH]8924[/ATTACH]

Cutting edge! Interested in what you do with it.

Cheers,

-SB

[moodz]

next generation descriminating pulse detector DPS5000

... slaving away over a hot computer ... work continues on the next gen dps5000 metal detector.

An IB coil system has a TX pulse signal that is ouput by a DAC driven by the DSP software to a 65 watt pulse amplifier. The RX coil feeds a balanced receiver and is amplified and fed to the ADC which pumps data to the DSP. Digital ALC and lockin is applied to the data. The bandwidth here is only 5 kHz. There are 4096 amplitude / phase analysis points .

Took me a while to get that code workin ..haven't touched it since I last posted and forgot what some of it did.

The top half of the screen is the T domain of the rx pulse. The bottom half is the F domain of the rx pulse.

Here is a a screen shot



and with the DigALC applied. The lockin works to amazing small pulse amplitudes ... like -60 db The pulse is centred by the lockin. Note there is some increase in noise due to the digital 'amplification'.



moodz

[moodz]

DSP5000 cranking it out at 48 Khz

The joys of DSP ... no soldering irons required.

Enter a new F domain bandwidth and we have 48 Khz pulses with the same ALC and Lockin performance .... now we are pickin up the really small nuggets.



and small signal targets.



moodz