LF project

[Davor]

Final part is the audio part with binaural conditioning. With binaural you can use any of the channels to produce a mono audio for a built-in loudspeaker, but otherwise it requires stereo amplification.

Sounds are produced by means of amplified and conditioned GEB channel. As it is aligned to have null at ground response, and freely adjustable to assume any angle of the foreseable ground responses, even the exotic ones, it will be very flexible. The angle is relative to the Tx angle, but not with the frequency, so only minute adjustment may be needed when Tx frequency is changed.

This separate GEB approach, as separated from All metal channel, is beneficial at difficult grounds as they both perform optimally their respective functions. Joint GEB/All metal is useless at difficult ground as it can perform only one of the functions properly ... if you adjust for perfect ground rejection, your discrimination gets screwed up. And vice versa. I don't think this is an original idea, but it makes sense.

Tone is produced by phase shifted switching with switches duty cycle of 50%. More about this later. A consequence is a synthetic sinewave with suppressed harmonics up to the 5th, at GEB amplitude (and phase). There is no real need for rectifying GEB voltage because this works either way. Switches are gated by discrimination signals, so that all combinations of tones are possible; neither, either, both. Elliptic filters take care of the remaining harmonics so that pure sinewave goes out. Tones are phase shifted in binaural fashion.

Here is the schematic

[Davor]

As indicated 2 posts before, achieving perfect phase shifting in a I-Q signal path is an easy feat by means of a linear stereo potentiometer. A diagram below shows a connection and a result in abs() form, because zeroes in phase transition are relevant. This way a near perfect linear transfer is achieved for 180° from potentiometer end to end. Transfer can also be optimised for constant amplitude, but with somewhat compressed phase transfer in the middle (no big deal)

[Davor]

Here is the audio modulator in a simplified form. I used a model of CD4066 to make it closer to reality, but also simplifications of op amps for E1 and E2. Attached is a tone sample (out.zip) without signal variation or gating, simply to illustrate purity of such a tone. Big deal is a suppression of all harmonics up to the 5th, and all remaining harmonics are below -40dBc.

Missing are the other channel, and binaural network, also audio PAs.

[Davor]

I think I have an IF gain block candidate. Trouble with these is that op amps are of little use in a low budget arena as they lack speed, noise performance, or both. So BJTs are a way to go. There is a little struggle over Mr. Miller's capacitances, and the ways to make them less prominent. A compound transistor (Complementary Feedback Pair) comes to the rescue. It extends the useful bandwidth for a few octaves, while reducing THD, so it is a BJT well spent.
LTspice model of a BC337 is not correct in sense that it models Rbb as much higher than it actually is. Simulations suggest input referenced noise at 1,3nV/sqrt(Hz), while in reality it should be lower. Because of "on resistance" of the switchers in a range of ~100ohm even this is an overkill.
THD is of lesser importance in IQ IF amplifiers, as it is truncated to the overall accuracy and 1% is Masha'Allah, but even in that department this li'l gem is a nice one - it provides 0.33% THD at 100kHz and 0.9Vpp output (gain of 52dB). Not bad at all. At lower levels it goes waaaay down.
Next step is convincing it to cooperate in a "Correlated Double Sampling" chopper scheme.

[WM6]

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.

Hi Davor, in which mean you use "IF" acronym here? I am usually lost in all those acronyms trying to follow some contributors posts. Thanks.

[Davor]

The part you quote is the one I thought of when I considered a classic heterodyne as a solution for gain, however, by whim of epiphany, I came up to a different scheme not seen before.

IF stands for intermediate frequency. In a sense of a chopper amplifier and applied terminology, a proper term would be "AC amplifier". However, sampling LF signal in a range between 50kHz and 150kHz by a QSD results in two separate streams of chopped signals, I and Q, at double the carrier frequency. So if I have, say, 100kHz Tx carrier, the two resulting streams would be chopped at 200kHz and with duty cycle of 50%. This signal is amplified by two separate AC amplifiers, and converted to a baseband with an output switch. The whole process is in effect a "Correlated Double Sampling" chopper scheme.

Since 200kHz is a bit high-ish for an AC amplifier, as it's name suggests audio frequencies, I assumed IF is better explaining it's function. QSD-s are often described as "zero-IF" systems, implying that the IF is in fact a good DC amplifier. A significant amount of ink is already spilt on various papers about 0Hz behaviour of such schemes, with various solutions. Mine is employing the ultimate DC amplifier, a chopper.

QSD by it's nature chops the input signal, so instead of LPF-ing it I apply it to an AC amplifier, followed by a synchronous detector which completes a "Correlated Double Sampling" chopper scheme. If someone else did it I'd say it was beautiful - nothing is wasted. As a main consequence 1/f noise is missing in the output signal.

What follows after the I and Q gain blocks are phase arithmetics for discrimination.

[Davor]

This is the improved audio modulation circuit from above. Now it converts the input DC signal (target response) of any polarity into a sinewave with THD of less than 0.3%. It is also gated by a digital type discrimination input so I can apply whatever discrimination scheme I want. I'll have to decide on binaural scheme and output PAs, but this part is done.

[Davor]

This modulator is a real DSB with virtually no intermodulation. I simply entered a 100Hz sinewave as a signal input, and the result is a perfect dual tone. Output spectrum is as clear as with DC on input

[Davor]

I made some further simplifications to the audio circuit. It became less obvious, but it is still in line with the circuit in my previous post. It uses a ring oscillator as a time base instead of a shift register, and it is hilariously simple.

I'll post it once I make it more presentable. Most probably I'll make a PCB of it so it can be used in other projects as well. This one is a keeper.

[Davor]

So here it goes. Instead of a shift register that would grant me signal purity up to the 5th harmonic, I made a ring oscillator that provides time shifts with less than perfect symmetry and at the end produces some 2.8% THD in simulation, dominated by 2nd harmonic. I think it is good enough, and surely it is simple enough.

Instead of a bunch of AND gates I'm using a single 4053 chip, and gating is achieved by it's inhibit function. If it is high, none of the gates is on, ergo silence. One of three gates in 4053 is performing some digital algebra, which is not too common, but works fine.

So in total I have 3 chips: dual op amp, CD4069, and CD4053, and the result is a gated, and perfectly modulated detection voltage, at <3% THD. A perfect sound source.

What will follow are a binaural phase shifting network, and of course audio amplifier(s).

This schematic can be a standalone single channel, but my intention is using two.

In this simulation I'm using my own hierarchical model of 4053, which is a bit idealised for speed. I have a more strict hierarchical model as well, so just ask. I did not include Vss and Vee in this model, but instead I use Gnd. In almost all builds these two pins are grounded anyway, so it was easier this way. It still relies upon a global parameter Vcc.

[Davor]

Just a few thoughts on choice of audible tones. First off - there are no tones any better than others, so the choice is arbitrary. 1kHz is right in a middle of a telephone band, so it can work just fine. It is close to C6.

There is a question of relative pitch when two or more tones are used. It is obvious that frequency relationship of 1:2 (octave) works fine, but there are perhaps some other choices that may work better. In order to visualise our perception pitch is often mapped in a circular form or in a spiral. If we use C6 as a reference (~1046Hz) it's opposites are F#5 at 740Hz and F#6 at 1480Hz. C to F# is related as 1:1.41
In fact all frequencies that fall between 1:1.33 (G:C) and 1:1.5 (F:C) would be quite distant in pitch, but not too distant in frequency. This may come in handy when designing sound output using resonant transducers such as piezo.

To put things into perspective, attached is a typical pitch class spiral. Octaves are separated vertically, but horizontal separation can work too. It also suggests that a VCO that spans in a narrow range of frequencies is quite effective way of audibly accentuating some feature.

[Davor]

Yet another thought about human interfacing. I had a chance to play with Vallon UXO detector. It is equipped with a haptic interface - vibration. It works beautifully.

Vibration is quite common with mobile telephones already, so it is only a logical step to introduce vibration to metal detectors. Haptic interface could take care of some slow phenomena, such as non-motion, or all metal, or whatever else appropriate, while sound is related to some other feature as discrimination.

My idea is to have a haptic interface for underwater operation.

Replacement vibrator motors are wildly available for iPhone at good prices, however, these are not well described by freely circulating specifications. I'd appreciate if someone can share information on supply voltage and current of iPhone vibrators.

I'm a specification of a vibrator motor that I found specification for can be downloaded at DigiKey: http://media.digikey.com/pdf/Data%20.../KHN4NX1RC.pdf
Perhaps it is not that different from iPhone specified motors ?!

[Davor]

From IGSL topic about system noise sources, and induction balance...

...
You may consider AM noise of a well built (Tx) oscillator somewhere at -140dBc/Hz at 10Hz due to components noise contribution only, so you would have plenty of headroom before you match input noise. But somehow you are barely within the limits of input noise with 200mV carrier (e.g. amplitude as seen from the preamp input). Let's see why...

Noise performance of LC oscillators is related to the supply rail voltage purity, hence it must be superb, or otherwise...

Let's ponder a second on it. I found that a simple 7805 regulator gives some 40uV of noise in audio band. Thus it is divided by sqrt(20000) and scaled for 7808, normalised to 200mV amplitude, reduced for gain, and we get some ~10nV/sqrt(Hz), again very close and barely exceeding input noise performance. Right here it is clear that a switching regulator would be a killjoy at this place. It is also clear that halving air signal residual would reduce AM noise injection by half.

Filtering a power supply rail using a low noise reference, such as red LED, would improve power rail noise by factor of 10 or so. IMHO it would be a game changer in PI vs. IB depth battle.

I made a slight oversight in my Tx noise calculation. The 200mV at Rx was a signal after amplification, so in reality situation is much better. This also means that there is ample of freedom to use poorly balanced coils in case of front-end sampling without a preamp. This is great news for extreme band spans in detectors, as slight imbalance will not make problems. Making a Tx low noise is apparently an easy feat.

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 !

Actually ... yes, and with almost no reservations. Even modest induction balance should do.

Main points with amplitude and phase noise of MD Tx are as follows:

- phase noise is of very little relevance, as everything else is referred to Tx anyway. Phase noise exists only in comparison with some exact frequency, but not in a self-referred system such as MD. Reality check... spread spectrum devices work fine as long as their Rx is correlated with Tx.

- amplitude noise is reduced by limitation/compression mechanisms, and ultimately it is barely exceeding power rail noise. Trouble lurks from Tx oscillators that operate under rail amplitude. We may investigate this a bit further.


All free running LC oscillators tend to assume maximum available amplitude set by supply rails. Oscillation criterion assumes system gain greater than 1. Stability is achieved by reducing gain to 1, and there are systems that apply some kind of AGC to keep oscillation stability under rails voltage, but there are some pitfalls of this approach.

It is safe to assume that allowing oscillation to be self-stabilised by means of rail voltage limitation gives superior amplitude noise. There are several mechanisms that tend to reduce gain with approaching a rail amplitude. Such compression produces harmonics, so there is no such thing as free lunch here either. However, this approach provides much harder amplitude limitation than any AGC.
A bit extreme approach is using a binary excitation that is also limited by power rail, and it's noise. Harmonic content of such exciter is very predictable, and by slight manipulation certain harmonics can be easily obliterated.


It will be interesting to evaluate harmonics content as Rx interference. I'll make some simulations to see what are the odds with different synchronous detectors and various harmonics.

[Davor]

I thought of alternative ways of using the sine audio modulator, and realised that a VCO could do the trick.
So instead of using the annoying VCO buzzer that starts from 0Hz, I can have a nice sine with volume proportional to target response, and pitch in range ~1:1.5 related to target phase, e.g. discrimination.
To do that I need a phase meter, and I realised that the only reasonably simple way to make it is to revert back to the IF approach, but made in logamp fashion. It may seem as a bit of overkill, but in fact is a way of using one off building blocks. It avoids 1/f problem and is also all quadrant approach. It would also facilitate the auto-GB. More on that later on.

[Davor]

Right now some essential pieces are connecting the puzzle. The way of IF/log amp/phase detector is about to get a front end in "dirodyne" style ... with a twist. I realised that regular QSD, although efficient and simple, may benefit from investigating some other phase constellations, and right now it seem that a winning combination is a 3-phase system. A "dirodyne" is a kind of a commutating filter that, unlike more traditional heterodyne, has output directly translated to a desired frequency/phase. For an incoming signal such front end appears as an extreme Q tank, and by a small twist in design - with a notch on a carrier frequency that results in "air signal" elimination. On top of all, the phases remain intact, so I can use the IF as a perfectly predictable channel, always operating at the same frequency. This is beneficial on many accounts:
- there is no 1/f noise, hence I can safely run this device as a slow motion detector without any need for tuning of offsets
- phases are always working the same, regardless of the coils/frequencies at the front-end, and I can change them at will, and without much fuss
- motion compensation is done at the very front-end, and "air signal" is removed there, so I can safely pimp the gain FM limiter style, and use a phase detector directly in later stages. It also relaxes balance requirements, as "air signal" is of much less importance.
- as phase is extracted directly and exactly, a single channel suffices for all discrimination purposes.
- ground balance is extracted separate from everything else, so it can work just about everywhere. It is also easy to make auto GB in this configuration.
- It is beneficial for usage with micros, as the IF can be an exact fraction of clock frequency, in fact any clock frequency, so AD becomes much simpler, and less constrained.

Below is a principle schematic that wraps it up.

So a result is a sine tone proportional to target, at a frequency directly proportional to target ID, instantaneous reaction, and 4 quadrant discrimination of course. Some binaural manipulation may be useful for Fe/Non-Fe separation.