TGSL Experiments

[simonbaker]

one channel beep

Ok, I think the reason I was able to get a beep while driving only one channel of the TGSL is that I had the sensitivity pot set very high and there is some kind of coupling over to the GB channel...?

I think enough coupling backwards through R30 (100K), through the JFet, through R26 and R27 (20K), through the other JFet, through R28 (100K) to the input of the U103a (GB channel LM35 maybe?

Or something like that. I can see a little bounce in the output of the GB LM308. Frankly it looks like it is 180 deg out of phase and shouldn't trigger the comparator simultaneously.

But maybe the DC level is right there where the comparator triggers all the time for the GB channel.

Anyway, if I lower the sensitivity a little, no audio beeping.

-SB

[simonbaker]

comparator threshold

Don, does this seem right to you...

Your sensitivity threshold goes down to -28 mV at max sensitivity.

When I short the RX input, the noise at both comparator inverting inputs is about 6 to 10 mV pp, centered very close to zero bias.

Yet at max sensitivity, the audio does not make a steady beep. That means to me the comparator needs more than 28 mV differential to trigger.

Actually there are tiny output trigger blips, so some triggering is happening in the noise range, maybe the threshold is around 32 mV.

You figured this out once -- can't remember the numbers.

-SB

[greybeard]

many many years ago when i actually had hair on top of my head i was shown a technique by one of my mentors to assist with earthing problems.
you get a piece of large copper cable ( adjust size to the equipment your are working on ). connect one end of the cable to an earth near the power source, then while observing the test point of concern, use the other end of the cable to earth various earth points near the relevant circuitry. ie you are bridging the various earth tracks with a large piece of cable and effectively removing earth loopps and common signal paths from the circuit.
this gives you soem good feedback on where you have earthing issues and possible ideas to fix them. sometimes it may require cutting tracks and running wires back to a common point, in other cases a wire jumper may resolve the problem.

this is a similar issue to the one mentioned earlier with regard to earth wiring to the coils.
hth

[simonbaker]

many many years ago when i actually had hair on top of my head i was shown a technique by one of my mentors to assist with earthing problems.
you get a piece of large copper cable ( adjust size to the equipment your are working on ). connect one end of the cable to an earth near the power source, then while observing the test point of concern, use the other end of the cable to earth various earth points near the relevant circuitry. ie you are bridging the various earth tracks with a large piece of cable and effectively removing earth loopps and common signal paths from the circuit.
this gives you soem good feedback on where you have earthing issues and possible ideas to fix them. sometimes it may require cutting tracks and running wires back to a common point, in other cases a wire jumper may resolve the problem.

this is a similar issue to the one mentioned earlier with regard to earth wiring to the coils.
hth

Exactly.

-SB

[simonbaker]

Hi Don:

First, I don't trust any test I do unless I do it at least three to thirty-three times! So will continue to check it out.

Quick answer -- I didn't isolate the audio ground, just jumpered it in parallel, sort of load relief on the main ground buss -- arterial bypass so to speak. Yes, I created a "ground loop".

My test was somewhat artificial as I injected the signal I was studying. Still, adds to my feeling that a better audio ground may be a good idea.

See attached photo. The green lines are jumpers I clipped on to perform the test. One is just to short the RX coil input to keep it from adding another signal. The other is the ground-to-ground shunt that I theorized would relieve some of the ground buss lifting by the audio pulse. In a revised circuit, you would also cut the ground connection from the speaker to it's current ground buss connection.

You can also see where I injected the signal and measured the output.

It was dang tricky to get clips to stay put, so always some room for error.

Now -- why did I get a beep with only one channel? That troubles me, so I need to check to make sure this isn't all BS.

-SB (BS backwards???)

Forgot to document on the image: injected signal freq is 2 Hz.

-SB

[dfbowers]

Don, does this seem right to you...

Your sensitivity threshold goes down to -28 mV at max sensitivity.

When I short the RX input, the noise at both comparator inverting inputs is about 6 to 10 mV pp, centered very close to zero bias.

Yet at max sensitivity, the audio does not make a steady beep. That means to me the comparator needs more than 28 mV differential to trigger.

Actually there are tiny output trigger blips, so some triggering is happening in the noise range, maybe the threshold is around 32 mV.

You figured this out once -- can't remember the numbers.

-SB

Yes, those were my observations almost exactly. I was speculatiing that the comparators may not actually trigger at 0V. In reading the spec sheets, it's only documented that they compare "around 0V".

Now the reason I did the mod was because I could not get the circuit to chatter at all, and shorting R37 seemed to do nothing! So.. wondering how far negative I had to go to get both channels to trigger, I arrived at something less than -20mV. Then you came up with the idea of playing with the offset for the :LM308s

Here's the really odd thing. I was playing around last night and to my horror, when I turned on my circuit I could barely get 20cm for a euro.
Nothing had changed. I could turn my sens almost all the way up and very little noise at all. On checking my null, I could see noise all over the Rx signal.. The noise just seems to kill performance sometimes

Sooo.. I turned my unit off, stepped away from the workbench, being careful not to disturb anything and decided to return the next day..
Today, back to normal!! NOTHING is different except the time. I could not for the life of me determine what the source of the noise was. Sounds like your scenario!

[simonbaker]

Yes, those were my observations almost exactly. I was speculatiing that the comparators may not actually trigger at 0V. In reading the spec sheets, it's only documented that they compare "around 0V".

Now the reason I did the mod was because I could not get the circuit to chatter at all, and shorting R37 seemed to do nothing! So.. wondering how far negative I had to go to get both channels to trigger, I arrived at something less than -20mV. Then you came up with the idea of playing with the offset for the :LM308s

Here's the really odd thing. I was playing around last night and to my horror, when I turned on my circuit I could barely get 20cm for a euro.
Nothing had changed. I could turn my sens almost all the way up and very little noise at all. On checking my null, I could see noise all over the Rx signal.. The noise just seems to kill performance sometimes

Sooo.. I turned my unit off, stepped away from the workbench, being careful not to disturb anything and decided to return the next day..
Today, back to normal!! NOTHING is different except the time. I could not for the life of me determine what the source of the noise was. Sounds like your scenario!

Those Russian submarine networks will get you every time...

It is very interesting how noise reduces depth. My first instinct was that it should just add audible noise to the signal. But based on the circuit, I can see how it could actually reduce depth significantly.

The reason I think is the peculiar low-pass filtering that is done after the comparators. I believe it works like a pulse-length filter. Only pulses longer than a cutoff duration make it to the speaker. So weak target pulses that are nice and long are chopped into small pieces by noise and none of the pieces get through the pulse-length filter. I know I sound like a broken record, said it many times. But it is such an interesting type of filtering I hope to spend some more time thinking about it and trying some variations.

As for where this noise comes from... I could believe change in the local standing wave nulls and peaks when big power loads switch on an off nearby...

-SB

[simonbaker]

More Mysteries "Solved" - Case of the Shrunken Null

If you're not interested in puzzles (and why should you be?), don't read this. It's more about electronics sleuthing than building metal detectors.

Spoiler alert -- some mysteries unraveled, the main question (superior depth) still open.

We last left off with a couple of puzzles:

1. Why does dfbowers PCB have larger null signal (other than some guys just have larger nulls), expecially since TX frequency is exactly the same, capacitor C6 is the same, coil is the same? His null signal is 15% to 25% higher (depending on how you measure).

2. Why (in my workshop) does dfbowers PCB have lower noise signal at LM308 output -- especially since his null signal is larger, suggesting higher gain!

Particularly curious is that mystery (1) suggests higher gain which would predict higher noise at LM308 output, but the opposite is true (mystery 2)! Does his PCB have some magic "noise muffler" somewhere I don't know about?

-----------
preamble
-----------

I previously tested and found that our PCBs have virtually the same main filter section (LM358 and LM308 ) performance - same gain and bandwidth.

There was a tiny difference difference in the shape and gain of very small output signals which I believe is due to the effect of the audio circuit on the ground buss, which fed the circuit back into the op amps. Using a jumper from the audio ground to the battery ground made our two filter sections even more similar.

So I didn't think the difference in our PCBs was due to the main filter section.


-------------- Case of the Shrunken Null ----------------

I'll report on Mystery (1) first. Why was his RX null voltage around 15% to 20% higher using exactly same TX and RX coils?

Actually, his null was about 25% higher when measured at output of LF353.

Then I remembered -- Ivconic substituted 4.7k resistors for the 5.1k resistors in the original schematic for R13, R14. I have stayed with the original 5.1k -- I'll bet dfbowers uses 4.7k, which should give more gain.
Sure enough, a quick check confirmed 4.7k resistors nestled snugly on dfbowers PCB. But that would only account for about 8.5% more gain (I think using simple gain formula R15/R13). There was still about 16% gain unaccounted for.

From now on I decided to measure the RX null voltage directly at the coil leads on the PCB using differential oscilloscope probes (which turned out to be somewhat of a problem on its own).

This problem ended up driving me crazy. I wish I could say I found a clear-cut explanation. However, this is my best understanding to date.

I tried dozens of tests, including:

1. Hypothesis: different C6 capacitor, dfbowers is more resonant.
Results: I tuned my C6 capacitor to be virtually same as his, as checked by in-circuit capacitance measurement. Did not change the symptoms.

2. Hypothesis: C6 capacitance measurement is wrong -- maybe my capacitor has hidden resistance, causing less Q resonance. Confirmed this possibility with LTSpice, 3K parallel or 300 ohm series resistance could cause symptoms.

Results - Using signal generator, I disconnected RX coil and injected 13.7 kHz signal through various sized resistors into C6 for his PCB and mine. If the capacitors were different, the measured voltage and/or phase should be different. No significant difference was found. The capacitors were essentially the same.

3. Hypothesis: different TX oscillator amplitudes (why didn't I think of this first?).

Results: Measured TX oscillator voltage -- his was slightly higher, less than 4%. Would not explain 16% higher null voltage.

4. Hypothesis: current, not voltage, in TX coil is what matters! A more resonant coil can have higher current. Need measurement of TX coil current.

Results: I rigged up a "sniffer" coil to sample the magnetic field of the TX coil. This should be an indicator of the current in the coil. However, it turned out that dfbowers TX coil current was about 4.5% higher than mine. Would not explain 16% higher null voltage.

5. Hypothesis: some signal is leaking across PCB with phase sufficient to reduce my RX null voltage.

Results: Ohm meter showed no unwanted circuit paths.

6. Hypothesis: some signal is leaking from somewhere to reduce the null signal on my PCB.

I decided to do a whole barrage of signal injection tests with the PCB power turned off, looking for cross-talk from the oscillator or TX coil into the null signal.

I wired up some "bypass" USB connectors so I could have the TX coil connected to the PCB while the RX coil was disconnected, and vice versa. Then measure RX null with all combinations of connections and injecting the signal into the coil side or the PCB side.

I could go on for pages writing about all the tests I did and redid over and over. I found weird things about using differential probes, how the choice of vertical gain could totally change the wave form; other weird things due to grounding; or even data I recorded where I had things hooked up wrong.

From various tests, I did begin to get the feeling that somewhere there was another signal being introduced in my circuit that was opposing the RX coil null signal and therefore reducing it slightly.

Test
----

The most convincing test (with least amount of dangling alligator clips and questionable probe contacts) was as follows:

I injected an approximately 16 V pp signal into the TX coil while still connected to the PCB -- in other words, PCB power was turned off, the oscillator was not actually running, the TX signal was coming from an external signal generator, tuned to about 13.7 kHz, same as oscillator freq.


Then I took an additional 6 foot USB extension cable and added it to the existing coil cable, and compared results of the 12 foot cable with the original 6 foot cable.

Interestingly, with the 12 foot cable, my null was larger than dfbowers. With the 6 foot cable, dfbowers null was larger than mine.

6 foot cable: Vnulldfb / Vnullsb = 3.5 / 2.9 = 1.2

12 foot cable: Vnulldfb / vnullsb = 3.1 / 2.5 = .8

It was pretty hard to stabilize these voltage measurements, so fairly approximate.

-----------------
Conclusion
-----------------

My best conclusion is that the cable contributes some capacitive coupling of the TX to RX signal that modifies the null signal. My PCB has a longer PCB cable to the female connector by maybe 9 inches, causing our two cables to be slightly different.

So the higher null signal in dfbowers PCB is not due to more gain or resonance -- just a different cross-coupling in the cable.

A confirmation that it has something to do with that little extra piece of cable came from another test. I injected the TX signal directly into the TX coil at the USB female connector using a little bypass cable, with the TX coil disconnected from the PCB -- so the TX signal did not traverse the small piece of cable between the PCB and the female connector.

In that case, our RX null signals were virtually the same. Of course, it still could mean the other case is affected by the oscillator components (even though not running), so I can't give a certain explanation.

Satisfying? No. Relevant? I don't know. I'm hoping that the exact level of the null signal due to cable effects does not alter target signals in any significant way. But still could be a factor I suppose.

Nulling Issue
------------

One issue it raises is: should we null our coils using the oscillator through the cable, or by an artificial signal directly to the TX coil with no cable. Maybe we're not getting the "magnetic null" that we want when we use the oscillator and cable attached because we also add in the capacitive cross-coupling which can give a "false" minimum.

I have not so far discussed "The Case Of The Hidden Noise". That will be addressed in another message.

-SB

Photos below:

1. TX Sniffer Coil
2. RX bypass cable

[simonbaker]

Case of the Shrunken Null - small correction

6 foot cable: Vnulldfb / Vnullsb = 3.5 / 2.9 = 1.2

12 foot cable: Vnulldfb / vnullsb = 3.1 / 2.5 = .8

Should be:

6 foot cable: Vnulldfb / Vnullsb = 3.5 / 2.9 = 1.2

12 foot cable: Vnulldfb / vnullsb = 2.5 / 3.1 = .8

Also did test with our built-in oscillators and got confirming results.

-SB

[simonbaker]

The Case of the Coy Noise

(Continued from previous message).

Case of the Coy Noise
------------------------

The second mystery was why dfbowers noise signal at LM308 output was slightly lower than mine. It is very difficult to estimate, because the noise in my workshop changes continuously. But I had the distinct impression his noise was maybe 30% to 50% less in magnitude.

First, I established the noise comes from the RX coil, not the PCB. Shorting the RX coil make the noise mostly disappear.

The only explanations I could think of for a difference in noise, since we were using the same coils, were:

1. More gain in his system.

2. My oscillator was creating some of the noise.

or both.


I haven't thought of an easy experiment to distinguish oscillator noise from EMI noise. One way would be to put the TX,RX coils in a perfect Faraday box that shielded them from all EMI -- I don't have one.

Or possibly make a tiny transformer that acts like the TX, RX coils and shield it.


Anecdotally, I was inclined to feel that it was a gain difference rather than oscillator noise, because the symptoms seemed the same at different times of day when the noise was much larger than at others. If it was oscillator noise, then larger EMI noise would swamp it out at times, and when EMI was less, the difference in our noise measurements would become more dramatic; I did not feel I observed that.

However, I could not for the life of me find any area where my PCB would have higher gain. In fact, I knew it had lower gain in the LF353 preamp. So nothing made sense.

Then for no good reason I started thinking about sync pulses. The sync pulse ideally is a 50% duty cycle, or "dwell", square wave pulse. However, not all TGSLs are ideal. And it turns out that the gain of the Synchronous Detector is directly affected by the sync pulse dwell. As the dwell departs from 50%, the gain should diminish.

Now it also turns out I had experience to show that at low DISC pot settings, most TGSLs exhibit some really wild looking sync pulses. I was so alarmed at these waveforms that I devised the "SB mod" to clean up the sync pulse on my TGSL circuits.

Perversely, as nature likes to play with mortal fools, in the end it seems that really horrible looking sync pulses don't seem to harm the TGSL one bit. But it was something I made note of.

Flash forward to the present and reexamine the RX coil noise difference between dfbowers TGSL and mine. If his TGSL had less gain for some reason, that could explain why his noise seems less than mine. Since the sync pulse can affect gain, I decided to take a look there.

I had been testing with both of our DISC pots at minimum. When I compared our sync pulses -- hola! What a difference. dfbowers sync pulse consisted of two separate spikes, while mine looked like the normal slightly rounded square wave. dfbowers TGSL is afflicted with the typical "spiky sync pulse" syndrome when the DISC pot is near minimum. As the DISC pot is turned up, his sync pulse becomes normal.

See the following video comparison of dfbowers sync pulse vs. SB sync pulse while turning DISC pot min-to-max-to-min by hand.

http://www.youtube.com/watch?v=KDzPHPvO_4c

The reason for the difference in the sync pulses is because my PCB has the "SB mod", which works nicely, makes everything look better, but doesn't do a damn to make a better TGSL!

However, it offers an explanation why dfbowers TGSL has slightly lower gain at minimum DISC pot setting.

To confirm this, I observed the noise at the LM308 output while turning dfbowers' DISC pot from min to near max. While not easy to evaluate, it seemed to me that I could clearly see the noise get larger as I turned the DISC pot toward max, due to the more symmetric sync pulse. In fact, the noise looked very similar to my PCB.

----------
Conclusion
----------

So in the end, maybe my PCB and dfbowers PCB have more similar noise than I originally thought, at least when it comes to large EMI environment.

I still think dfbowers PCB gives a little more depth and more solid tone when I test in the mountains, so this leaves open the possiblity of some oscillator noise or PCB noise limiting my circuit at very low noise levels. But at least I have a plausible explanation for the seemingly paradoxical noise discrepancy I observed on my workbench.

-----
Coda
-----

What about these spikey sync pulses at low disc settings? Do we care? Do they provide a perverse service by causing the gain to slightly increase as we apply more discrimination, thereby compensating for the loss of gain when the sync pulse is not centered over our target of interest?

I can't help still being a proponent of my "SB mod" even though it has not proved to be of any benefit, and has not been proved to be harmless either.

-SB

Photos:

In each photo, top trace is TX oscillator voltage, bottom trace is LF353 pin 7 output.


1. dfbowers LF353 pin 7 output -- oscillations due to overdriven LF353 input by TX oscillator. This corresponds to the driving signal when the DISC pot is at minimum.

2. SB LF353 pin 7 output - less oscillations due to "SB mod" voltage divider from TX oscillator. The small blip appears to be an overreaction by the underdamped LF353 to the small bump in the oscillator signal.

[Qiaozhi]

Hi Simon,

Can you also do a comparison between the two PCBs at pin 1 of U102a (LM393)? You might have to temporarily add a pullup resistor there (perhaps 10k would suffice) to see the waveforms correctly. Remember that the LM393 comparators are open-collector.

[simonbaker]

Hi Simon,

Can you also do a comparison between the two PCBs at pin 1 of U102a (LM393)? You might have to temporarily add a pullup resistor there (perhaps 10k would suffice) to see the waveforms correctly. Remember that the LM393 comparators are open-collector.

Hi Qiaozhi:

U102a pin 1 should be the DISC sync pulse, is that what you mean?

Basically I showed that in the YouTube video I put a link to. It shows the sync pulse as I turn the DISC pot. You can see the sync pulse is broken up when driven by the "hashy" signal from dfbowers' LF353 pin 1 at minimum DISC pot setting.

Here's the link again:

http://www.youtube.com/watch?v=KDzPHPvO_4c

Regards,

-SB

P.S. I grabbed frames of sync pulse from video for convenience.

Photo 1: dfbowers DISC sync pulse (U102a pin1) with pot at minimum.

Photo 2: my (SB) DISC sync pulse (U102a pin1) with pot at minimum.

[Qiaozhi]

Hi Qiaozhi:

U102a pin 1 should be the DISC sync pulse, is that what you mean?

Basically I showed that in the YouTube video I put a link to. It shows the sync pulse as I turn the DISC pot. You can see the sync pulse is broken up when driven by the "hashy" signal from dfbowers' LF353 pin 1 at minimum DISC pot setting.

Here's the link again:

http://www.youtube.com/watch?v=KDzPHPvO_4c

Regards,

-SB

P.S. I grabbed frames of sync pulse from video for convenience.

Photo 1: dfbowers DISC sync pulse (U102a pin1) with pot at minimum.

Photo 2: my (SB) DISC sync pulse (U102a pin1) with pot at minimum.

Thanks Simon. I hadn't watched the video as I assumed it just showed the same thing as the photos. Note to self - "Don't make assumptions".

It looks to me like the dbower's circuit gets away with it at the lower DISC setting, because the split sync pulses are positioned either side of the peak on the RX signal. Your own mod looks like a better solution.

[simonbaker]

Thanks Simon. I hadn't watched the video as I assumed it just showed the same thing as the photos. Note to self - "Don't make assumptions".

It looks to me like the dbower's circuit gets away with it at the lower DISC setting, because the split sync pulses are positioned either side of the peak on the RX signal. Your own mod looks like a better solution.

I constantly fret that my mod causes some other problem, but haven't clearly identified any yet. I took pains to tune it with a capacitor to minimize the little phase shift it introduces. Also, I happen to use it to drive the GB channel as well because of layout reasons, but that channel doesn't need my mod and I think does better without it (little more range because it likes being driven by a walloping signal). So I'm cautious about the mod.

In fact, I removed my "SB mod" while investigating the "Case of the Shrunken Null" to make our PCBs more similar. I did note that when I removed it, it was not dragging down the oscillator as I worried -- in fact, the oscillator seems to run a hair bigger with my mod in place.

All this nitpicking I'm doing is to because I want to first gain control over building a standard TGSL before I get on to the more interesting business of experimenting with weird and nifty mods. If I ever get there, I hope to offer far more interesting posts. Lots of ideas floating around to play with thanks to everybody.

-SB

[simonbaker]

interesting dud

I finished a new PCB (last one I'll hand-wire, hopefully) with some insane component values to make the TGSL bandpass filter section (LM358, LM30 center on about 1 Hz (!) instead of 9 Hz. In other words, super slooooooooow response. Just to see what happens.

I had to fix some mistakes and one conductive bridge between "traces" to get it working. So maybe not completely trustworthy yet. But both channels seem to be working correctly.

I wired the USB cable opposite of dfbowers (black/green, white/red), so I need to use a USB crossover cable to connect to dfbowers coil. This could add a little noise maybe. I do not need the crossover to connect to my new coil (the Alien).

In my noisy workshop, I can't test much. However, you can tell it is a very different beast. The noise at the LM308 outputs is lower in frequency, but still quite large, so I think I'm just picking up more of the lower frequency noise and less of the higher, but no real S/N gain here.

The response is indeed very sluggish. The beep is very delayed and long. I like the long length of the tone.

I took the PCB out to the mountains for a quick test -- no time for video, just my observations. I also took my new coil to test it. All my tests were with my new coil (which needs the crossover cable to connect to dfbowers PCB).

My insane PCB is really insane -- it was difficult to adjust the sensitivity threshold to reduce noise because it seemed sluggish. I also had it in a cardboard box and it seemed sensitive to movement.

Anyway, I was not able to get additional depth with it. At depths around 20 to 25 cm it responded in a very sluggish way to my Euro. It did seem more sensitive to slow movements than a normal PCB. The beep is very delayed and long. But it is hard to tell noise from target due to the long delay. Noise also can make a long beep.

My conclusion is that it sucks, but that maybe if less extreme there is a useful mod in there somewhere. The longer beep is definitely appealing. Perhaps Vladimir already found the sweet spot with his mod, which lowers the bandpass center also. If there's time, I'd like to try some different center frequencies. The trick is keeping the gain approximately the same.

I also tested my coil with my normal PCB and with dfbowers PCB.

With dfbowers PCB, my coil seems to get about 26 cm depth. This is not surprising, because the RX inductance is not as large as it should be and requires some C6 trimming. I had sensitivity knob at max, DISC at min. Also, the extra USB connector might play a part. Try tapping the USB connector some time and look at your scope.

With my PCB, I added 2 nF trim cap to C6, and if I adjusted the sensitivity extremely close to the chatter point, could get a very short beep, more like click, at about 29 to 30 cm, with my coil. But if I backed off a little on the sensitivity, I lost depth quickly, to 25 cm.

The problem is that my PCB gives too short a beep most of the time. Only when very close (15 cm) is the beep really satisfying. dfbowers PCB gives a better beep over more range.

I haven't been able to figure out why exactly. The suspect is noise, which can interfere with the front and back edge of the beep signal. But I don't think I have that much more noise, or at least can't understand why I would. I'm not sure how much noise would be required to give that effect either. Another possibility is the audio circuit somehow. I forgot to try the ground jumper test.

So, plenty of stuff to think about and try to improve.

-SB