"Beyond" TGSL ....

[Aziz]

Improved version of the synchronous oscillator

http://www.wipo.int/pctdb/images4/PCT-PAGES/2001/152001/01026213/01026213.pdf

Guess, whose invention is it? Same guy.

But this version has a -48 dB noise rejection!!!
Aziz

[Aziz]

Simon,

put this little code into the spice model:

.options Gmin=1e-12
.options reltol=0.00001

And step through the input voltage to see the phase shift of the output.
Make an AC analysis to see the frequency response (at resonance +25 dB, BW approx. 200 Hz)

Aziz

[simonbaker]

Simon,

put this little code into the spice model:

.options Gmin=1e-12
.options reltol=0.00001

And step through the input voltage to see the phase shift of the output.
Make an AC analysis to see the frequency response (at resonance +25 dB, BW approx. 200 Hz)

Aziz

I still get the phase oscillations in the time domain when plotting (Vout - 380000 * V2).

I know nothing about "phase amplifiers", but my worry would be that changes in the TX frequency or TX phase jitter would make some kind of phase change in the output. But I'm not sure how the phase amplifier fits into an MD front end exactly.

Interesting circuit for sure. Please keep explaining all you can, Aziz and moodz. Good to have it as a reference.

Regards,

-SB

[Aziz]

Hi Simon,

the basic idea behind this is as follows:
Just reducing the bandwidth of the amplifier. Yes, a resonant circuit is also an amplifier, which has it's maximum gain at it's resonant frequency. Outside the resonant frequency, the gain of the amplifier suffers. Totally outside, the amplifier attenuates even. The phase shift at resonance frequency is always 90 degree. The Quality (Q) of the resonant circuit is defined as:

Q = freq/BW, where
freq = Resonance frequency
BW = Bandwidth (3 dB mark)

What happens, if you reduce BW? Q gets bigger. This is the objective in such applications. Making Q as much as possible.

Consider, you have a input signal with it's own additional noise on the spectrum. The noise spectrum has a frequency density called noise spectrum density. It's definition is for bandwidth 1 Hz at specified frequency.

Now, what happens, if you reduce your bandwidth? Let's say to 0.01 Hz! Ok, Q gets quite big. What happens to the noise spectrum? Nothing - still there. Noise spectrum stays same, but the effective output noise (integral over frequency bandwidth range) gets lesser. If you have a very high Q, then you go effectively below the noise spectrum density level.

High Q means also that the phase is very sensitive to frequency changes. Look at the pictures below for a series resonance circuit. If you reduce the coils resistance, the BW gets low, its gain increases, Q gets higher.

High Q coils achieve nearly same characteristics.

The circuit (sychronised oscillator) just implements a very short bandwidth filter (more gain at resonance frequency, lower gain outside the resonance frequency).

Aziz

[simonbaker]

Hi Simon,

the basic idea behind this is as follows:
Just reducing the bandwidth of the amplifier. Yes, a resonant circuit is also an amplifier, which has it's maximum gain at it's resonant frequency. Outside the resonant frequency, the gain of the amplifier suffers. Totally outside, the amplifier attenuates even. The phase shift at resonance frequency is always 90 degree. The Quality (Q) of the resonant circuit is defined as:

Q = freq/BW, where
freq = Resonance frequency
BW = Bandwidth (3 dB mark)

What happens, if you reduce BW? Q gets bigger. This is the objective in such applications. Making Q as much as possible.

Consider, you have a input signal with it's own additional noise on the spectrum. The noise spectrum has a frequency density called noise spectrum density. It's definition is for bandwidth 1 Hz at specified frequency.

Now, what happens, if you reduce your bandwidth? Let's say to 0.01 Hz! Ok, Q gets quite big. What happens to the noise spectrum? Nothing - still there. Noise spectrum stays same, but the effective output noise (integral over frequency bandwidth range) gets lesser. If you have a very high Q, then you go effectively below the noise spectrum density level.

High Q means also that the phase is very sensitive to frequency changes. Look at the pictures below for a series resonance circuit. If you reduce the coils resistance, the BW gets low, its gain increases, Q gets higher.

High Q coils achieve nearly same characteristics.

The circuit (sychronised oscillator) just implements a very short bandwidth filter (more gain at resonance frequency, lower gain outside the resonance frequency).

Aziz

It sounds sort of like a taking a feedback amplifier with positive feedback, and instead of stopping just short of oscillation, you let it just barely oscillate, and then feed it an input signal very close to its oscillation, and it tracks over to that signal, even if small and buried in noise.

But how do you stabilize it for phase of the target signal? We want to discriminate by phase. It just seems like a very tempermental circuit phase-wise. I'm not clear on what you use for the TX signal so that fluctuations in frequency and phase of the TX signal don't make phase noise in the RX signal.

What would be a more complete picture of an MD front-end using the phase amplifier?

-SB

[moodz]

Interesting -- it seems either LTSpice can't correctly simulate this circuit or perhaps there is a non-linear "limit cycle" at work here. When I subtract the input from the output, it looks like either the phase is oscillating, or the two frequencies are slightly different.

I changed the time step and the oscillation period changes. Must be numerical simulation problems.

-SB

Hi Simon .... err why are you doing the subtracting ? .... I may have confused you by calling it an amplifier .... it is an amplifier of sorts but in the Frequency domain. Plot a FFT of the input ... then plot a FFT of the output. You will see that the spectrum is "lifted" or amplified by around 100 db or so.

Actually what you have done though illustrates an important principle. addition and subtraction of Frequency domain signals in the time domain is the same as multiplication and division in the F domain. So you have modulated the signal by attempting to subtract it in this way !!!! vice versa addition and subtraction in the F domain corresponds to multiplication / division in the Time domain.

See below how the spectrum is amplified ( ie the amplitudes of the spectral peaks are amplified but the phase and freq is not disturbed .... well not too much ... all amplifiers add noise ).

Below you can see the frequency of interest is "lifted" from -170 db to about +10 db a gain of nearly 180 db !!!! So if you are going to do adding or subtracting to work out gain .... do it in the Frequency domain.

It makes sense that this is not a voltage amplifier or it would lock onto itself as the output signal is 200db higher than the noise floor and you could never isolate such an amplifier.

Because this is a phase "amplifier" it is insensitive to a large degree to input amplitude changes .... I think this is a bonus. To implement in a detector ... you need two oscillators one for TX and one for RX .. they need to be frequency locked but not phase locked ( as the coupling and target info is going to do the phase locking )
For my modification hashup which is not ideal I just set the two osillators to nearly the same frequency and did some testing before the TX oscillator drifted away from the centre frequency of the Rx oscillator but if feel that a frequency locked loop could do the job.

INPUT

OUTPUT

[simonbaker]

Hi Simon .... err why are you doing the subtracting ? .... I may have confused you by calling it an amplifier .... it is an amplifier of sorts but in the Frequency domain. Plot a FFT of the input ... then plot a FFT of the output. You will see that the spectrum is "lifted" or amplified by around 100 db or so.

Actually what you have done though illustrates an important principle. addition and subtraction of Frequency domain signals in the time domain is the same as multiplication and division in the F domain. So you have modulated the signal by attempting to subtract it in this way !!!! vice versa addition and subtraction in the F domain corresponds to multiplication / division in the Time domain.

See below how the spectrum is amplified ( ie the amplitudes of the spectral peaks are amplified but the phase and freq is not disturbed .... well not too much ... all amplifiers add noise ).

Below you can see the frequency of interest is "lifted" from -170 db to about +10 db a gain of nearly 180 db !!!! So if you are going to do adding or subtracting to work out gain .... do it in the Frequency domain.

It makes sense that this is not a voltage amplifier or it would lock onto itself as the output signal is 200db higher than the noise floor and you could never isolate such an amplifier.

Because this is a phase "amplifier" it is insensitive to a large degree to input amplitude changes .... I think this is a bonus. To implement in a detector ... you need two oscillators one for TX and one for RX .. they need to be frequency locked but not phase locked ( as the coupling and target info is going to do the phase locking )
For my modification hashup which is not ideal I just set the two osillators to nearly the same frequency and did some testing before the TX oscillator drifted away from the centre frequency of the Rx oscillator but if feel that a frequency locked loop could do the job.

INPUT
[ATTACH]13070[/ATTACH]
OUTPUT
[ATTACH]13071[/ATTACH]

Thanks for additional explanation.

OK, I'll have to think deeply about this... or remain happily confused.

The reason I subtracted the (scaled) signals was I wanted to see the phase relationship of the output to the input. Again, I'm not sure how we process the output of the "phase amplifier", but I assume we want to be able to discriminate based on the phase of the input signal relative to the TX signal. I also want to see that any phase jitter or frequency drift/jitter in the TX signal does not come through in the demodulated target signal.

So probably seeing a bigger picture of the overall target detection would help.

Regards,

-SB

[moodz]

Hi moodz:

Would you still have a synchronous detector after the "phase amplifier" driven by the TX oscillator? If so, what is the real purpose of the phase amplifier? Or is now the oscillator part of the RX circuit, and you tap off a signal for the TX signal -- which then becomes indirectly the input signal for the "phase amplifier" -- which can shift the RX oscillator -- which controls the TX signal -- my brain is oscillating....

I think I'm not seeing the big picture.

Also, when it "locks onto the input signal", is the oscillator phase exactly the same as the input signal, or is there some lead or lag? And what output would correspond to the phase?

Regards,

-SB

Simon .. I would imagine that the circuitry after the the phase recovery / amplifier would be the same. I think whereas we have got used to seeing the TX oscillator as the master oscillator you may have to get used to the idea of the RX oscillator as the master oscillator. In order to resolve very fine phase for targets etc we dont want that information being destroyed by a 'lock' signal. So the TX oscillator is the one that should be locked. The TX oscillator if it is stable to say 10 Hz or better would stay in the lock range of the RX oscillator which would also require similiar stability. The normal low level coupling via the Tx and Rx coils would provide the 'phase lock' signal between these two oscillators. The phase difference between the two oscillators is then determined by a phase detector in the conventional way. This difference will depend on how the coils are coupled, what the ground / target is. So whether it is leading or lagging does not really matter for lock purposes. There will be a certain phase diff for coil in air. coil over ground. coil over target etc etc.

Another way of putting this is that it is a phase locked loop system where the control loop ( ie the phase information path ) passes through the detection path. The way the phase information path is modified tells us about targets etc.

[Aziz]

Phase Detection

Hi all,

I think the Lock-in amplifier gives more accuracy to detect the phase relationship. It also has a very good linearity. The ADC sampling frequency need not to be cohorent with the signal and reference frequency even. My measurements confirm this behaviour.

Remember: Lock-in amplifier is a phase sensitive detector!

[moodz]

Hi all,

I think the Lock-in amplifier gives more accuracy to detect the phase relationship. It also has a very good linearity. The ADC sampling frequency need not to be cohorent with the signal and reference frequency even. My measurements confirm this behaviour.

Remember: Lock-in amplifier is a phase sensitive detector!

Hi Aziz .... yes a successful implementation of the lock-in amplifier would be good to see however I am only looking at the sync oscillator as it is in line with Simons KISS principle. And I may be able to use it for a two box solution as it is very sensitive.

Simon .... here is my test bed .... a very old VLF / TR ETI 1500 ... I will see what I can do with it

Also attached is a patent describing a Frequency locked loop which uses a different principle than a phase locked looped. So to implement this idea the TX would be frequency locked to the RX and the RX would be phase locked to the TX ... sounds a bit dodgy however in the frequency domain Phase and Frequency can be independant of each other. ( sort of like voltage and current in time domain )
I will however try and find a non digital solution to match the simplicity of the sync oscillator.

Frequency_locked_loop_US4801894.pdf

[simonbaker]

Hi Simon .... err why are you doing the subtracting ? .... I may have confused you by calling it an amplifier .... it is an amplifier of sorts but in the Frequency domain. Plot a FFT of the input ... then plot a FFT of the output. You will see that the spectrum is "lifted" or amplified by around 100 db or so.

Actually what you have done though illustrates an important principle. addition and subtraction of Frequency domain signals in the time domain is the same as multiplication and division in the F domain. So you have modulated the signal by attempting to subtract it in this way !!!! vice versa addition and subtraction in the F domain corresponds to multiplication / division in the Time domain.

Is it not true if I have two signals a(t) and b(t), then

F(a + b) = F(a) + F(b)

where F is the Fourier Transform?

And

F(K*a(t)) = K * F(a(t))

where K is a constant?

So isn't the Fourier Transform of the difference of two time domain signals equal to the difference of their Fourier Transforms in the frequency domain?

And isn't it actually the dB scale that converts multiplication to addition, etc, not the Fourier Transform, in your example?

Regards,

-SB

[simonbaker]

Simon .. I would imagine that the circuitry after the the phase recovery / amplifier would be the same. I think whereas we have got used to seeing the TX oscillator as the master oscillator you may have to get used to the idea of the RX oscillator as the master oscillator. In order to resolve very fine phase for targets etc we dont want that information being destroyed by a 'lock' signal. So the TX oscillator is the one that should be locked. The TX oscillator if it is stable to say 10 Hz or better would stay in the lock range of the RX oscillator which would also require similiar stability. The normal low level coupling via the Tx and Rx coils would provide the 'phase lock' signal between these two oscillators. The phase difference between the two oscillators is then determined by a phase detector in the conventional way. This difference will depend on how the coils are coupled, what the ground / target is. So whether it is leading or lagging does not really matter for lock purposes. There will be a certain phase diff for coil in air. coil over ground. coil over target etc etc.

Another way of putting this is that it is a phase locked loop system where the control loop ( ie the phase information path ) passes through the detection path. The way the phase information path is modified tells us about targets etc.

This "phase amplifier" is definitely interesting but challenging for me to analyze and think about.

If the TX oscillator is derived from the "RX oscillator", I could envision a couple of problems.

One is that the null signal from the coil is at exactly the same frequency as a target, but a different phase. So why would the phase amplifier not "lock onto" the null signal instead of the target?

But that may be moot because we have a sort of "phase feedback" path here. Essentially, if the TX signal is derived from the phase amplifier output, then the phase amplifier output is being fed back into its input via the null signal (or targets). I don't know if this is stable, but if it is, it must satisfy an equation (I would think) where the phases don't fight each other. This implies that there is some wiggle-room phase-wise, so that the system can adjust the frequency so the phase relationships are a multiple of 360 going around the loop, like a normal oscillator.

Once the system has adjusted itself to be stable, what happens when a target signal is added to the null signal? Do we get the detection we want?

We'll look forward to what you find if you build some practical circuits.

Regards,

-SB

[moodz]

Is it not true if I have two signals a(t) and b(t), then

F(a + b) = F(a) + F(b)

where F is the Fourier Transform?

And

F(K*a(t)) = K * F(a(t))

where K is a constant?

So isn't the Fourier Transform of the difference of two time domain signals equal to the difference of their Fourier Transforms in the frequency domain?



And isn't it actually the dB scale that converts multiplication to addition, etc, not the Fourier Transform, in your example?

Regards,

-SB

Hmmm good point ... this circuit is not exactly what I would call a PLL however it behaves similiarly .... in the classic PLL analysis the phase detector is a 'multiplier' ... so one would expect that the incoming signal is multiplied/convoluted by the reference. This is not a linear addition.
The question here is .. can this circuit be analysed in a similiar manner.
This doco is not a bad reference ...

www.complextoreal.com/chapters/pll.pdf

more reading .... sigh.

[moodz]

FFL control solved ... maybe

Hi Simon ... no need for digital control loops ... the FLL loop can be implemented with a simple op amp integrator and two switches. Can you see how it works ...
<<explanation goes here ....>> however its late and I am off to bed.
The TX oscillator and the RX oscillator are fed into the frequency detector. When F1 = F2 the output voltage = 0 if F1 <> F2 then the voltage will be + / - dependant on whether F1 is less / greater than F2. Use this voltage to adjust one of the oscillators. What is important with this circuit is that phase information cannot pass through only frequency control unlike a PLL ... Make the integration time really long so the oscillator frequency will only change slowly compare to target frequency. This circuit gets a KISS rating of 9/10 I have not drawn the VCO loop yet.

Moodz.

[moodz]

That diagram is a bit small ... here is the sim



synchosc001.asc.zip