LF project

[mikebg]

Harold Weber

Here is a more sensitive detector invented by Harold Weber:

[Davor]

Let's put a cork on a sound discussion. There are a few ways to make a binaural sound space, but for a reasonably simple LF project only a few frugal options are viable. Let's see the choices:
a) Cu and Fe channels go together, mono;
b) left(+Cu, +Fe) right(+Cu, -Fe), counterphase;
c) split a) into two allpass networks with ~90° apart, binaural;
d) split b) into two allpass networks with ~90° apart, counterphase binaural.

I did not put simple stereo here because it is awful.

Real binaural with exact delays would make little sense because your coil is in front of you, not some odd places in space. Perhaps these could be somewhat improved, but IMHO every option is better than mono. Maybe the simple (+,+) (+,-) solution will do.

Attached are the sounds for cases a) to d) and a screen shot of a LTspice simulation that created the last one (the Laplace things are the allpass networks). Feel free to comment your impressions.

[Davor]

There is another facet of the nowadays VLFs that may need revisiting in order to make a better LF rig, and those are the gain blocks. They suck at the 1/f low frequency noise region.

So what happens in the gain blocks is ... gain. And filtering. And some limiting action as well. They perform at frequencies well below 100Hz. In case there is a high pass filtering action they are called "motion" compensated, but what they really compensate is the offset due to several sources. As a side-benefit, HPF also reduces 1/f noise. So all is well ... until you wish to go slow motion.

There is a family of DC amplifiers based on chopper action, where offset and 1/f noise are tackled quite successfully, but in a rather pricey fashion. You need as many such amplifiers as you have channels, so it is worth examining other options.

I think the best option around is a heterodyne. If the phase is maintained properly it can provide all the gain and limiting/compression/AGC action prior to splitting such IF into different channels, and you can be sure that all such channels will have perfectly equal gain and group delay. Quadrature sampling detectors/mixers have negligible losses, and they do not spoil 1/f performance, hence the heterodyne gain stage is a superior approach to the traditional gain blocks.

I'd say heterodyne is a superior solution even to ADCs because of superior dynamic range that can be maintained through compression/AGC in the heterodyne gain block.

Of all heterodyne configurations there is one that combines commutating mixers and Weaver method filtering in a most beautiful fashion, dubbed a Dirodyne. It's output frequency is invariant from the carrier frequency, but the phase relationships are preserved. It consists of two commutating mixers with their channels sorted in a back-to-back fashion with low pass filtering elements in between, thus effectively performing the Weaver method filtering. Just add HPF here and you have a perfect motion compensation. The output signal is translated to some IF and there you have a single AC signal with static (non-motion) component removed, and no offsets of any kind. And it is perfectly phase related to the input Rx signal. Just perfect.

Because of low losses at commutating mixers, extreme filtering and static component being removed after the mixing, perhaps the input preamplifier will not be needed at all - not to spoil things.

[Altra]

Davor,

You may have seen this before. Check out the circuit in fig 2 in the attached pdf, looks interesting.

Years ago Compass made an IB/TR detector that ran at about 100khz. I think it was model 77b. Some of
of the old timers raved about how good it ignored rust and nails. What frequencies are you considering?

[Davor]

I am thinking of a rig that would be entirely controlled by a Tx tank, and which would oscillate in a balanced fashion in a range between 60 to 120kHz by switching different tank capacitors. Therefore I need some different approach to sorting Rx discrimination phases. I already invented a tank-locked Tx that is both perfectly balanced and simultaneously produces perfect quadrature, so the Rx is next.

The document you attached is about using choppers to make better DC amplifiers. What I have in mind is very much in line with a chopper design in terms of avoiding 1/f noise, but I chose to use an invariant IF to sort my discrimination phases too. Point is that I don't need perfect amplification, in fact I chose a single IF channel that can also be severely compressed, even clipped, and phase relationships will remain intact. After such amplification I can take whatever number of phase shifted discrimination channels, front, back or sideways, and they'll all be perfect.

Such heterodyne performs as a perfect tracking filter with enormous Q and still with perfectly aligned phase. I'd say that's something new.

[Davor]

I gave this no-preamp approach a further thought and concluded it really is a way to go. Fun part is that by virtue of "air signal" elimination in Dirodyne mixer and less than 100Hz bandwidth, away from 1/f noise prone baseband, this approach could perfectly run even in conditions of high "air signal" residual, thus a perfect candidate for a true wide frequency span device. Removing perfect induction balance as a precondition will make coil production easy and fun. Furthermore, by insisting on a balanced mode input the burden of common mode and shielding is also off the table.

The construction constraints thus move (again) to the Rx front-end which is now a bilateral switch. It will have to offer low on resistance not to spoil input noise. Low noise preamp now becomes a front end of an IF, but relieved of the "air signal" and operating on extreme narrowband signal, it's THD is of no importance and it can provide enormous gain, leading to circuit simplification and less current draw. Traditional gain blocks are gone, so at least 10mA are saved for an extra hour working time.

[Davor]

Today it hit me! It was one of those epiphany moments, so it must be very good.

I realised that Tayloe mixer is kinda unfinished - it is missing a chopper amplifier. Fun part is that half of the circuitry required for a chopper is already there. so once finished Tayloe becomes an integral part of a fine I&Q chopper amplifier. What is even more interesting, an instrumentation chopper amplifier. This configuration will eliminate need for a separate IF, and also a pre-amplifier. Such front end is capable of handling signals at over 2Vpp with virtually no distortion and very little increase in effective source impedance. Chopper that follows is completely free from 1/f noise, and operates in ridiculously oversampled mode.

I'll put a LTspice simulation of this soon, perhaps even tonight if I manage to steal some time - it is half finished.

This will be radically different from anything I've seen so far.

[Thomas]

Is the PLL Quadrature sim part of it? I've been running that one a lot and I like what I see.

[Davor]

Yes, why waste a perfect quadrature!
There is an upgrade of that quadrature generator on the way with integrating loop filter - it fixes a few small pending issues in a beautiful way.

I think you'll like a quadrature sampling chopper even better. What you'll like the best is that it cancels air signal during mixing, so chopper has to amplify only much weaker signals that remain. I realised it is not so much of a Tayloe design in it, but much more D.H. van Graas (PA0DEN) balanced approach as a start, but the rest of it is all mine

I so love when it is balanced.

[Davor]

So here it is, a quadrature chopper mixer. It features -3dB bandwidth of (+ and -) less than 100Hz, it is motion compensated at below 1Hz, and at 10Hz features below 2nV/sqrt(Hz) noise, hence no pre-amplifier will be needed. With good switchers conversion loss is below 2dB. Simultaneously all other parameters go sky high, including IP3 etc. The picture below shows a pure target simulation, and a mix with "air signal" will follow.

The very chopper amplifier is not presented because I wanted the simulation to end while I'm still in reproductive age.

I wonder who else will recognise all the advantages of such front end

[deemon]

So , Davor , as I understand it - you have an "auto balance" of the coil , yeah ? Residual disbalance signal on the RX coil , being converted to DC , is blocked by C6 and C7 and can't overload your amplifier , isn't it ? But motion signals can go further to the chopper .... it's interesting . IMHO very good solution !

[Davor]

Yes, something like that. Big deal is that such front end remains low noise in the IF, and free from both 1/f noise and "air signal". I'd balance my coil anyway, but perfection is of less importance now.

Here are two more screen shots. First one shows the timing. A and C are responsible for I channel, and AC1 for it's chopper action. Once coil sampling is finished, a 33uF capacitor is charged to the sample voltage, which is a difference between voltages Coil+ and Coil-, a perfect differential (balanced) mode sample. If there was no chopper this voltage would float till the next sample which is a difference between Coil- and Coil+ (reversed phase). But it doesn't float because the chopping switches close and reference lower side of this capacitor to the ground, and the other side is in series with the 47uF capacitor through the other closed switch to a chopper amplifier, here depicted only by it's input impedance.

There are two things that happen here, first is the voltage that is unbalanced (!) so the chopper amplifier is a single-ended device. Second is that the 47uF provides "air signal" cancellation by virtue of high pass filtering. The second picture shows what happens to the "air signal" some 0.5s after switch on (only I channel) with the same frequency and 1V (2Vpp) amplitude.

[Davor]

I realised that no preamp approach has some additional merits. The I and Q components are mutually orthogonal, and their noise contributions are also orthogonal. Quadrature noise sources are in fact Gaussian, zero mean equal variance uncorrelated variables.
A switched source that is 50% time off presents itself to the following stage with double impedance. Lightly loaded output of a QSD has a conversion gain of about -0.9dB, and with low Rdson switches, that's about it. Noise from the input is split into two uncorrelated paths at double the impedance (!) which is a favourable fact when calculating IF stages for noise. As bonuses you also get incredible input referenced Q, astronomic powerline, EMI, RFI, etc. suppression, and most important of all - 1/f noise immunity. This fact alone is worth all the effort.

you have an "auto balance" of the coil , yeah ?

As for the elimination of air signal ... there is a quirk that stands out. This approach eliminates the static component, the carrier. Upon the Tx carrier there are two mutually orthogonal noise vectors: AM and PM, and neither are suppressed a tiny bit. Of these PM is rotating the whole system (as if nothing happens), hence even a sweep Tx should produce meaningful response, so PM is of less importance here. AM noise is a bit different. It gets reduced with better balance, and it worst affects low real conductors (small gold, foil, salts). Perhaps even worse than that is the effect on ground balance at beaches where GB is less aligned with ferrites.

Solutions are twofold:
- better balance
- low AM noise Tx.

Balanced Tx and differential Rx should enable better coil balancing (deeper minimum).
However PM is of lesser importance per se, there is a nasty effect that turns PM to AM and it is a basis of off-resonance detectors. This conversion happens in Tx oscillators that operate with less than 50:50 duty cycle and those that are switched at angle. With varying phase (noise) switch-on walks up and down the slope and produces AM noise in a process.

To illustrate Tx AM noise effect I made a picture. Red vectors are Tx, green is GB, and blue is a target. I presented Tx as a combination of a carrier (static, presented as bulleted) and dynamic component (AM noise). While the carrier is completely cancelled by quadrature chopper mixer, this AM noise is superimposed to all other vectors as a component parallel to the Tx, and not suppressed at all. In severe cases this noise screws any possibility of achieving ground balance at beaches, and also makes discrimination of small gold more difficult. In short:
- noisy Tx -> no small gold,
- poorly balanced coil -> no small gold.

[Davor]

It makes sense to wrap up parts of a project in a simple to follow flowchart-like schematic. I divided it into 3 parts:
- front end: a tank locked loop Tx, and a quadrature chopper Rx mixer
- 4 quadrant discrimination and GEB extraction (auto/manual)
- audio part with binaural conditioning

So far I have not envisioned any kind of VDI. Dual tone discrimination served me well by now and I did not feel any need for it. I like to keep things simple.

Here is a picture of a front end:

[Davor]

Next up is a 4 quadrant discriminator, GEB extractor and GEB compander. With I and Q DC channels perfectly aligned with Tx, phase manipulation within 180° is easy by using stereo potentiometers. 3 phases are extracted for discrimination:
- All metal, aligned precisely with Tx and ferrites, same as traditional GEB but precisely at -90°
- Cu, same as coloured metal in traditional detectors, freely adjusted to go well within Fe range
- Fe, opposite in function than Cu, but also freely adjustable so that overlaps and notches are possible with different settings of Fe and Cu

These 3 channels are turned into digital form by means of comparators, and multiplied (!) by means of XOR gates -> that's the easiest 4 quadrant discrimination around. There is no "antichatter" to be seen, and sensitivity is all the way through to 0V.


A ground balance is also extracted here and it is freely adjustable in 0° to -90° range. This GEB is not related to discrimination All metal channel, and can assume perfect balancing of any ground around, including salty beaches (yes!). As there is no anti chatter, and one may wish to enjoy silence, I envisioned a compander with adjustable "knees" so that some level of chatter suppression can be achieved, while simultaneously this detector will not go bezerk with strong targets.

One additional function is choice between slow and fast motion compensation. With ~1Hz HPF cut-off for slow motion, there is no real need for non motion mode.

Here is the schematic: