AMX TX

[Carl-NC]

Yes but this subtraction is made in PI from INTEGRATED samples which is not the case here. You just cut two separate windows. They would need to be integrated first by an analog integrator for this subtraction to work.

If you make a digital integration over such a long duration, I fear that the reaction of the compensation system will not be fast enough to correct the fast ground variations and the variations of coil height over the ground.

The diagram I made in post 107 has 2 sample/holds which can effectively do the same thing if the taus are slow. If the taus are fast and you time the ADC to sample exactly after the 2nd sample then you can adjust on a pulse-by-pulse basis.

[Willy Bayot]

The diagram I made in post 107 has 2 sample/holds which can effectively do the same thing if the taus are slow. If the taus are fast and you time the ADC to sample exactly after the 2nd sample then you can adjust on a pulse-by-pulse basis.

Sorry, I did not realize that each switch was followed by a sample-and-hold circuit

[ivconic]

I don't know where to ask this question...maybe it won't bother here...
Carl I saw your reply about AF108 coil inductance on another thread...
The TX stage of the AF108 has nothing to do with the TX stage of the AMX.
But... are there any analogies and similarities in timings?
Is it also important there that the impulses do not happen at the same time or...?
I guess you understand why I'm asking this...

[Carl-NC]

No, I'm not sure of the 'why'. Yes, they are different methods. The AF108 is bipolar but with traditional passive charging of the coil, and a dead time between each pulse:



The AMX uses a bipolar current square wave with no dead time:



In the AF108 the pulse timing isn't critical since there is so much dead time between the (+) and (-) pulses. In the AMX timing may end up very critical.

[ivconic]

You take things too literally. I speak in principle, and you cling to every word and explain it over and over again.
There is too much talk about that AMX TX. You even put together a schematic.
But you still don't know the exact timing (pause between positive and negative pulse).
I wonder if that TX is worth the effort? What are the benefits it brings compared to other approaches...
Doesn't "dead time" at AF have a similar role as the pause between pulses at AMX?
That was "why", btw.
Based on the whole story so far, about the AMX TX, I concluded that it will be a very critical assembly,
which will be very difficult to "immunize" from problems in operation.
And I can again list a lot of examples, which if it happens; such a TX will start producing white smoke ...
It's just my impression, I don't claim to be right.
Why insist on something that is so complex and untested in practice so far.
At the same time, it is not a small matter to make it and work around it.
Things are very complicated with that TX and I'm afraid it's only going to get more complicated...
I can't get ANY of the components from your schematic in local stores. Nor in neighboring countries.
And when I look at the prices... it's going to be a very expensive assembly in the end!
At the same time... very "wobbly" and insecure in work. Because there are a lot of critical points.
When I say this; my last intention is to doubt your expertise.

[Carl-NC]

I'm never sure how much of the operational theory you understand, so I tend to keep things a bit on the light side. You're right, the whole AMX thing is very experimental, and it may take some effort to get it to work. I tend to think it won't be too bad because Paul has already built it and proven the TX to work, though there may still be some beans he hasn't spilled. A major purpose of this initial effort (for me) is to see what it will take to make it work.

The dead time in the AF108 is to allow the current to settle out (critically damped response) to zero, same as we do in a regular (unipolar) PI. The AMX TX has no current dead time, no settling, and no damping. However, the clock pulses probably at least need to be very slightly non-overlapping with maybe 10s of ns dead time. I don't know, I can't predict it, and I don't trust simulations in that kind of detail. So I'll design it with a micro that gives me all the control I think I'll need.

The whole reason for using this TX circuit is that it eliminates the dead-time settling of a traditional bipolar PI (like the AF10 which means it will run at a much higher frequency. 25kHz is the plan, and you get 2 PI responses per cycle so effectively it's 50kHz. I've never seen a PI run above 10kHz (Eric's GoldQuest) so this is substantial. It also offers complete control over the coil current; this initial design is adjustable up to 2A. I'm pretty sure you can build this circuit with any old FETs, including the IRF740. The drawback will be more resistive losses and slower slewing, but it should work fairly well. In the layout I've used SOT23 for Q1/Q2/Q5/Q6 (they can be low VDS parts) and for Q3/Q4 I've placed footprints for TO220, TO247, TO247-4, TO252, TO263, and TO263-7 so folks can shop around for something that will work. Despite the fact that I'm designing for performance by selecting performance parts, I see no reason why the whole thing can't be built with more common parts.

[green]

Playing with LT spice. Wondering how low A/sec needs to be when current is topped.

[Carl-NC]

Wondering how low A/sec needs to be when current is topped.

That's been one of the big debates with this method. Short answer is, I don't know. I've been calling it the "tilt." There are two sources of tilt: losses in the circuit itself, and losses due to reaction with the ground matrix. Therefore, the tilt has a static and a dynamic component. I've added tilt correction to the TX design, it will be under the control of the micro so response speed is TBD. In the end, it may not even be necessary.

[moodz]

I'm never sure how much of the operational theory you understand, so I tend to keep things a bit on the light side. You're right, the whole AMX thing is very experimental, and it may take some effort to get it to work. I tend to think it won't be too bad because Paul has already built it and proven the TX to work, though there may still be some beans he hasn't spilled. A major purpose of this initial effort (for me) is to see what it will take to make it work.

The dead time in the AF108 is to allow the current to settle out (critically damped response) to zero, same as we do in a regular (unipolar) PI. The AMX TX has no current dead time, no settling, and no damping. However, the clock pulses probably at least need to be very slightly non-overlapping with maybe 10s of ns dead time. I don't know, I can't predict it, and I don't trust simulations in that kind of detail. So I'll design it with a micro that gives me all the control I think I'll need.

The whole reason for using this TX circuit is that it eliminates the dead-time settling of a traditional bipolar PI (like the AF10 which means it will run at a much higher frequency. 25kHz is the plan, and you get 2 PI responses per cycle so effectively it's 50kHz. I've never seen a PI run above 10kHz (Eric's GoldQuest) so this is substantial. It also offers complete control over the coil current; this initial design is adjustable up to 2A. I'm pretty sure you can build this circuit with any old FETs, including the IRF740. The drawback will be more resistive losses and slower slewing, but it should work fairly well. In the layout I've used SOT23 for Q1/Q2/Q5/Q6 (they can be low VDS parts) and for Q3/Q4 I've placed footprints for TO220, TO247, TO247-4, TO252, TO263, and TO263-7 so folks can shop around for something that will work. Despite the fact that I'm designing for performance by selecting performance parts, I see no reason why the whole thing can't be built with more common parts.

There are no beans left to spill in the can. The tx circuit def works with junkbox parts and no deadtime on the gate drive. If there was "shoot through" occurring then I would imagine the energy conservation would not work as the flybacks would be shorted out by this event. Of course shoot through could be a problem if ultrafast hispec non junkbox parts are used.
Moodz

[ivconic]

I'm never sure how much of the operational theory you understand, so I tend to keep things a bit on the light side. You're right, the whole AMX thing is very experimental, and it may take some effort to get it to work. I tend to think it won't be too bad because Paul has already built it and proven the TX to work, though there may still be some beans he hasn't spilled. A major purpose of this initial effort (for me) is to see what it will take to make it work.

The dead time in the AF108 is to allow the current to settle out (critically damped response) to zero, same as we do in a regular (unipolar) PI. The AMX TX has no current dead time, no settling, and no damping. However, the clock pulses probably at least need to be very slightly non-overlapping with maybe 10s of ns dead time. I don't know, I can't predict it, and I don't trust simulations in that kind of detail. So I'll design it with a micro that gives me all the control I think I'll need.

The whole reason for using this TX circuit is that it eliminates the dead-time settling of a traditional bipolar PI (like the AF10 which means it will run at a much higher frequency. 25kHz is the plan, and you get 2 PI responses per cycle so effectively it's 50kHz. I've never seen a PI run above 10kHz (Eric's GoldQuest) so this is substantial. It also offers complete control over the coil current; this initial design is adjustable up to 2A. I'm pretty sure you can build this circuit with any old FETs, including the IRF740. The drawback will be more resistive losses and slower slewing, but it should work fairly well. In the layout I've used SOT23 for Q1/Q2/Q5/Q6 (they can be low VDS parts) and for Q3/Q4 I've placed footprints for TO220, TO247, TO247-4, TO252, TO263, and TO263-7 so folks can shop around for something that will work. Despite the fact that I'm designing for performance by selecting performance parts, I see no reason why the whole thing can't be built with more common parts.

Again good and easy to understand explanations, thanks!



"...maybe 10s of ns dead time. I don't know..."

That's why I though the simple 4069 inverter might do the job just fine, with the 30nS latency. If you are right, than 30nS would be too much.

[ivconic]

That's been one of the big debates with this method. Short answer is, I don't know. I've been calling it the "tilt." There are two sources of tilt: losses in the circuit itself, and losses due to reaction with the ground matrix. Therefore, the tilt has a static and a dynamic component. I've added tilt correction to the TX design, it will be under the control of the micro so response speed is TBD. In the end, it may not even be necessary.

How you will indetify those tilts in a quantity of sampled data?
How those will manifest... as a sampled value between many sampled values?

[ivconic]

There are no beans left to spill in the can. The tx circuit def works with junkbox parts and no deadtime on the gate drive. If there was "shoot through" occurring then I would imagine the energy conservation would not work as the flybacks would be shorted out by this event. Of course shoot through could be a problem if ultrafast hispec non junkbox parts are used.
Moodz

I guess there is no way to measure the time difference between pulses on already working model?
Or I missed to read it somewhere else as usuall?

[moodz]

I guess there is no way to measure the time difference between pulses on already working model?
Or I missed to read it somewhere else as usuall?

I will dig it out tomorrow ☢️

[ivconic]

[ivconic]

I'm never sure how much of the operational theory you understand, so I tend to keep things a bit on the light side. You're right, the whole AMX thing is very experimental, and it may take some effort to get it to work. I tend to think it won't be too bad because Paul has already built it and proven the TX to work, though there may still be some beans he hasn't spilled. A major purpose of this initial effort (for me) is to see what it will take to make it work.

The dead time in the AF108 is to allow the current to settle out (critically damped response) to zero, same as we do in a regular (unipolar) PI. The AMX TX has no current dead time, no settling, and no damping. However, the clock pulses probably at least need to be very slightly non-overlapping with maybe 10s of ns dead time. I don't know, I can't predict it, and I don't trust simulations in that kind of detail. So I'll design it with a micro that gives me all the control I think I'll need.

The whole reason for using this TX circuit is that it eliminates the dead-time settling of a traditional bipolar PI (like the AF10 which means it will run at a much higher frequency. 25kHz is the plan, and you get 2 PI responses per cycle so effectively it's 50kHz. I've never seen a PI run above 10kHz (Eric's GoldQuest) so this is substantial. It also offers complete control over the coil current; this initial design is adjustable up to 2A. I'm pretty sure you can build this circuit with any old FETs, including the IRF740. The drawback will be more resistive losses and slower slewing, but it should work fairly well. In the layout I've used SOT23 for Q1/Q2/Q5/Q6 (they can be low VDS parts) and for Q3/Q4 I've placed footprints for TO220, TO247, TO247-4, TO252, TO263, and TO263-7 so folks can shop around for something that will work. Despite the fact that I'm designing for performance by selecting performance parts, I see no reason why the whole thing can't be built with more common parts.

If I understand well; you are talking about 50 000 samples per cycle, right?
25 000 from "positive" side and 25 000 from "negative" side.
Impressive quantity of raw data!
I now understand why it was insisted on those specs.
Also I see that it will need the rest of processing with serious resources...
Is it wrong to assume that decent percent of those 50k raw samples will be redundant?
Powerfull mcu will process such quantitiy with no effort at all, I don't doubt that.
I was wondering how to initially reduce that amount of data, discard unnecessary ones, and send the rest to an mcu whose performance is good enough to process it.
Don't be surprised by my constant insistence on the "poor man's solution", I always like to see a wider variety of options.