ELF Monitoring Station with Real Time Streaming

Thanks for your suggestions.

In the above excerpt, it appears you mean signals below 40Hz are not substantially affected by filter-induced phase distortion.

If so, is there a particular type of filter that would optimize this effect for processing complex signals like geomagnetic micropulsations.

They somewhat resemble raw EEG traces in composition.

Mark Harris

Correct.

I'd look at a Bessel filter, as these have a very linear phase response all across their passband.

Take your original signal, apply a several-stage Bessel low-pass filter with a knee frequency of 500 Hz or so, digitize at 44100 samples/second using your DC-coupled sound card, then downsample at a

Maybe a non-causal filter, AKA time machine will help. ;-)

All filters add delay. C'est la vie.

Tim has a good point on logging the time. I suggest a GPSDO. While I know how to get a PC clock synced to a GPS, I don't know how one gets accurate time stamping since a computer has delay. I'm thinking the easiest way to do this is to get a sound card and run the GPSDO 1PPS into one of the channels. The sound card would have to be hacked to accept low frequency data, basically hack it to accept DC. One channel is the ELF signal, and the other is the 1PPS.

You can make linear phase FIR filters with a number of computer programs. Read up on

It can be done in Matlab, but I certainly wouldn't buy the program just to do that. The code is open source. Hell, they wrote it in Fortran it is so stinkin' old.

Here are some tips. The FIR filter should be symmetric for linear phase. That way you can check that the program isn't spewing total nonsense. But what you probably want to do (I suspect) is insure all the FIR filter coefficients are positive. That is a requirement for the filter not to ring. [If you think about that, it is obvious.]

Any decent compression method should do that. Even a simple differential PCM could do that.

And similar problems to the network at that -- modern preemptive multitasking OS's are nondeterministic.

Supposedly, Windows runs on a 1ms loop or something like that, so you might expect to see your program's code serviced as often (or less -- potentially never, if it's at a lower priority setting than something chewing a lot of CPU). Best case is minimal background tasks (so there's plenty of CPU cycles to spare).

If everything still must be guaranteed to very tight specs, it must be embedded. PCs can't do that kind of thing (and haven't been made that way for decades). Grab an Arduino, a GPS shield, and a 16 bit ADC (or audio something or other). It'll be custom (to the extent of buying plug-in hardware and writing code), but it will do what you expect of it.

Tim

GNU Octave.

Tim

email to where?

Mark, did you not read the information I posted? For measuring the earth's field you do NOT want a flat V/B meter. At 1 Hz the earth's field's approx noise spectral density is 1pT/rtHz, at 0.01Hz the approx noise spectral density is 1nT/rtHz. You want to match your sensor to the signal so your ADC is used most effectively. Therefore, a simple field coil with buffer/gain OpAmps into your ADC is the most reasonable match.

I'm not sure why you're cutting off near 40Hz, because there's a lot happening up beyond 10kHz. Be that as it may, cutting off at 40Hz has the advantage of more easily removing AC mains fields.

From experience at trying to control parameters, I recommend an air core coil of significant diameter. at least 18 inch. You can get aggressive with a 6 to 12 foot coil.

A magnetic field into a coil generates voltage ev = N B Area 2pif where N is number of turns B is field in Teslas Area is sq meters and 2pif is frequncy in radians per second

See how a simple coil will maintain the signal amplitude into your ADC? Watch out for 1/f noise in your amplifiers.

The resistance of the coil determines the Johnson noise and you want that to be less than the signal you're trying to measure, so select wire gauge appropriately. Received Voltage is proportional to N, but noise is proportional to sqrt(N) so each turn helps and there's hope. Your coil will probably weigh around 10 to 20 pounds. Cross section 1 to 2 sq inches. Copper is EXPENSIVE! you can esily spend $50 to $100 on the sensor. A good source of wire was Fay Wire, bought out by EIS Electrical Insulation Suppliers

Make certain you have stable construction. A coil in the earth's field can easily become a microphone.

Simpler to use, but have 'fixed' noise floors, you might want to look at 'active' sensors, like GMRs from NVE Corporation (800) GMR-7141 (800) 467-7141

But the sound cards have the secret sauce to make a steady stream, albeit buffered to get around the PC jitters. That was why I suggested simultaneous recording of a time base with the signal.

It would be a bit of work, but if you listen to this podcast. the guest mentions a sound recording done with carbon black on paper. It was two channel recording. Voice on one channel, and a time base comprised of a tuning fork on the other channel. This was in the 1800's. There was no way to decode the recorded signal at the time. But I couldn't believe someone in that era had the brains to record a time reference with the signal. This sound recording technique predates Edison.

Interesting stuff, but why air core? Also I presume you would want some differential input, i.e. center tap the loop?

There's public domain Fortran code for Parks-McClellan designs (their own, I think). It's published in an IEEE tome on DSP that I used to have.

Cheers

Phil Hobbs

What good does a metal core do? and with the tiniest bit of coercivity will you even 'see' the metal core in very low signals?

Actually metal core can do some things, if it gets activated, like lower the resistance for the same amount of voltage that gets generated. Think like 'spatial distortion' the metal core is like compressing air down into a small volume. Where if left as air you have a certain amount of flux, but with a metal core all that flux [originally spread throughout the air] can be concentrated down inside a smaller diameter coil at the permeability ratio - and therefore the lower resistance.

On the last page of some data sheet from some company [ferroxcube, fairite, ??] there is a plot of 'effective' inductance vs length to diameter ratio for various permeability core materials. For 'short cores, it doesn't matter if the perm is 100, or 1,000,000 the increase in inductance to the coil is the same.

Why center tap, the whole thing is floating. you just have parasitic capacitance to the outer loops anyway.

Simply electrostatically shield and be done with it.

For the preliminary work I had a set of XY coils scrounged from an Antarctic monitoring station. An AMP01 instrumentation amp was used for gain. The setup detects a spoon waving across the lab.

Yes, those AC fileds.

My coils are 20" long and 3" diameter. Mu metal core and electrostatic shield overall.

Thanks for the tip. I have emailed them requesting information on sensitivity.

Mark Harris

Would this hold true for a complex signal with composite frequencies that are effectively higher than the 40Hz cut-off?

This sounds like the most straightforward approach so far. Assuming that some of those accessing the data will be on the opposite side of the globe, what would you estimate to be the range of encountered delay between sender and receiver?

That would be between sites with world standard transfer speeds.

Mark Harris

I'm really puzzled by what you mean by "composite frequencies"? Can you explain further?

A signal which consists of components only below 40 Hz... well, it has no signal energy above 40 Hz. Period.

Tens to a few hundreds of milliseconds, at a guess. Worse if the signal has to go through a satellite link.

On a sunny day (Thu, 31 Oct 2013 13:19:25 +1100) it happened snipped-for-privacy@comprodex.com wrote in :

How about these?

Are your OpAmps or ADC overloaded by this local mains field ?

If not, have you tried a comb filter to remove the mains and its harmonics (particularly the 3rd) ?

You might need some ability to adjust the notch frequency slightly to lock it with the actual mains frequency at a particular time.

The sharper you try to make the cutoff at 40Hz the worse the phase distortion in the pass band must necessarily become. You will have to decide on an acceptable tradeoff between rejecting unwanted out of band signals and passband phase distortion.

Boxcar averaging for the subsampling, or even better a convolution with a suitable low pass symmetric kernel to output samples at a rate of say

Again there is a conflict here since to get the best phase behaviour you need to delay output of the signal in realtime.

You might also want to add a digital tracking phase locked anti 50/60Hz mains filter into the capture stream since it will be omni present unless you are a very long way from civilisation.

Depending on server load and network congestion something between 1s and infinity. Plenty of free newservers time out when US is awake.

Most times a remote site that isn't very busy establishes a connection within about 10s and serves packets in under 1s but you can get much worse behaviour to US when they wake up.

Why do you want to give people realtime access to a signal that will be pretty much like watching paint dry? Wouldn't it be better to save it locally and make available any interesting snippets from that day?

Doesn't WWLLN already do something like this with interpretted data?

Whistlers and the like require a much wider bandwidth than you are proposing to see interesting frequency structure.

I suspect he means aliased frequencies from the noisy signal. In the real world it is impossible to implement a true brick wall low pass filter at 40Hz and so any residual traces of components at a frequency

Attenuating these well enough and maintaining a flat phase response in the pass band is I think the question that the OP needs to decide upon. One strategy is digitise fast enough to capture all significant non zero frequency components and then DSP filter to the wanted bandwidth.

Most of the time unless the route suffers congestion. It is worst when one of the worlds major financial centres is waking up or closing down and huge transatlantic net traffic is flowing (private and public).

The odd packet taking almost forever is quite common on streaming video from major well resourced video streaming services at busy times of day. That is why they buffer it before playback.