John wrote, "...?without stepping because you deleted K L1 L1 K. When you add that back in, to get the steps, it doesn't run. I don't know why, or why the error log has those weird expressions."

Presumably you meant K L1 L2 {K}.

No problem adding it back in on my end.? I am pretty sure you have something else on your end, which is not in the schematic David uploaded.? Where did k1_A0, k1_A1, and k1_A2 come from?? They must have gotten added somehow to your copy of the schematic.? ?I guess you pasted more than you thought.

Andy

开云体育

How can it give identical runs with .step param K list 0.0001 0.01?

But I see why I can't get version D to run with the K expression added. I forgot the curly braces.? Now it runs and shows double peaking at both A and B with the tighter coupling and not with the looser coupling. I will upload it.

DesplitterD.asc runs OK for me.? I don't know what problem John sees, but I don't have that.? I think his simulation got corrupted.

But the stepped parameter K doesn't DO anything.? It gives you two identical runs.

Andy

开云体育

Yes, of course it runs, but without stepping because you deleted K L1 L1 K. When you add that back in, to get the steps, it doesn't run. I don't know why, or why the error log has those weird expressions.

Ok, I downloaded the D version from Groups, and the ONLY change I made was to name it DesplitterD1 so it wouldn't overwrite my other file.
It runs.
I'm running 17.0.220, and I sync'd the release a couple days ago.
Release is dated Feb 19 2021


Circuit: * C:\Users\Dave\Desktop\DesplitterD1.asc

.step k=0.0001
.step k=0.01

Date: Fri Feb 26 14:20:43 2021
Total elapsed time: 0.425 seconds.

tnom = 27
temp = 27
method = trap
totiter = 0
traniter = 0
tranpoints = 0
accept = 0
rejected = 0
matrix size = 10
fillins = 6
solver = Normal

DesplitterD.asc runs OK for me.? I don't know what problem John sees, but I don't have that.? I think his simulation got corrupted.

But the stepped parameter K doesn't DO anything.? It gives you two identical runs.

Andy

The nice thing about digital computers, is that they always give us the same results to the same calculations with the same inputs.? There is no possibility of having a different answer, unless either the inputs differ, or an intentional random difference was added to the calculation.? One LSB difference doesn't just happen on its own.

There can be an LSB of uncertainty between an analog quantity and its digital representation.? Or between an exact calculation (with infinite resolution) and a calculation with limited resolution.? But when I ask you to add integers 4 + 6, you'd better always get an answer of 10, and not 9 or 11 (= 10 +/- 1).

FYI, when LTspice needs to back up, it backs up to the previously saved good point.? Then it makes the timestep smaller.? With SPICE2, it made the timestep 1/8 what it was, but LTspice seems to use a different scaling.? Regardless of what that scale number is, there is no reason for it to be 1/8 sometimes, 1/7.9999999999999 sometimes, and 1/8.0000000000001 sometimes (starting from the same point), unless LTspice intentionally gives it that randomness.? It won't just happen because of round-off errors.? It'll be the same every time, because that's how math works on a digital computer.? Same inputs through same calculations --> same results, down to the very last bit.? Even rounding off a calculation is the same every time.

Andy

Ok, I downloaded the D version from Groups, and the ONLY change I made was to name it DesplitterD1 so it wouldn't overwrite my other file.
It runs.
I'm running 17.0.220, and I sync'd the release a couple days ago.
Release is dated Feb 19 2021


Circuit: * C:\Users\Dave\Desktop\DesplitterD1.asc

.step k=0.0001
.step k=0.01

Date: Fri Feb 26 14:20:43 2021
Total elapsed time: 0.425 seconds.

tnom = 27
temp = 27
method = trap
totiter = 0
traniter = 0
tranpoints = 0
accept = 0
rejected = 0
matrix size = 10
fillins = 6
solver = Normal

开云体育

Without the step K command it runs but only one run of course, no stepping.

Very odd..
I just ran it here, and I ran it before I sent it..

Circuit: * C:\Users\Dave\Desktop\DesplitterD.asc

.step k=0.0001
.step k=0.01

Date: Fri Feb 26 14:18:44 2021
Total elapsed time: 0.401 seconds.

tnom = 27
temp = 27
method = trap
totiter = 0
traniter = 0
tranpoints = 0
accept = 0
rejected = 0
matrix size = 10
fillins = 6
solver = Normal

I'm going to download the version from groups and try that file.

Very odd..
I just ran it here, and I ran it before I sent it..

Circuit: * C:\Users\Dave\Desktop\DesplitterD.asc

.step k=0.0001
.step k=0.01

Date: Fri Feb 26 14:18:44 2021
Total elapsed time: 0.401 seconds.

tnom = 27
temp = 27
method = trap
totiter = 0
traniter = 0
tranpoints = 0
accept = 0
rejected = 0
matrix size = 10
fillins = 6
solver = Normal

I'm going to download the version from groups and try that file.

Andy: Ok, absolutely equal I agree is vanishingly small probability.? Can we agree that say 1% difference is achievable?

开云体育

Actually it was me that said that. It's the only explanation I can think of for David's real-world result. His latest variant D won't run and the error log has weird entries.

David wrote, "So you say that "the transmitter does not couple exactly identically to the two tanks"? If we adjust the angle so that each tank is the same angle to the transmitter, then the intensity and phase of the signal arriving at each tank should be identical, except for the signal radiated by the adjacent tank.? Agreed?"

I have not kept up very well with this.

But the chance of you getting the same signal in both tanks in the physical world, is approximately zero.? Exactly the same, never gonna happen.

Andy

开云体育

It's not intentional, it's inevitable. Just think what happens, or so we are told. LTspice hits a large difference between sample x and sample x-dt,? where dt is the time-step, so it backs up to sample x-ndt, where n might vary? at the LSB level and tries again with a smaller dt; how much smaller might vary at the LSB level. There is inevitably at least an LSB uncertainty about the digital rendition of any quantity.? If the quantity is itself small, the LSB uncertainty is relatively big.

John wrote, "This randomness may be inherent in the routine that adjusts the time-step around discontinuities. The finite bit-depth can result in at least an LSB variation between different runs."

Sorry, but I am not buying it.? For what reason would LTspice add an intentional 1-LSB or few-LSB variation in the timesteps?

Andy

David wrote, "So you say that "the transmitter does not couple exactly identically to the two tanks"? If we adjust the angle so that each tank is the same angle to the transmitter, then the intensity and phase of the signal arriving at each tank should be identical, except for the signal radiated by the adjacent tank.? Agreed?"

I have not kept up very well with this.

But the chance of you getting the same signal in both tanks in the physical world, is approximately zero.? Exactly the same, never gonna happen.

Andy

John wrote, "This randomness may be inherent in the routine that adjusts the time-step around discontinuities. The finite bit-depth can result in at least an LSB variation between different runs."

Sorry, but I am not buying it.? For what reason would LTspice add an intentional 1-LSB or few-LSB variation in the timesteps?

Andy

开云体育

You deleted the directive K1 L1 L2 K, but adding it gives an error message because somehow you are getting this in the error log:

k1_A0 l1 l2
k1_A1 l1 k
k1_A2 l2 k

I have no idea where those are coming from, and they make no sense to me, but the simulation will not run.

"Your simulation cannot be true to your real world set-up, since it gives a different but entirely expected result."
Precisely why I'm asking questions here.

I implemented the changes you suggested, using the .step param list 0.0001 0.01, and increased K2 to 0.00001 leaving K3 at 0.000001

I tried probing both tanks separately, and a differential and I see no resonance splitting at all.
Uploading latest rev.

This may be a definitional problem, but assume the transmit coil is several meters away, and the two tanks are at the origin, mounted at 90 degrees to each other.
The frequency is low, so at this distance we are inside the radiansphere, and inverse cube law applies.
If the transmitter is at an angle where each tank is pointed 45 degrees off to the side of the transmitter, (transmitter bisecting the angle between the tanks), then that more or less copies the real world situation.

Physically, the ends of the two inductors are, and must be, located very close to each other (single digit mm) I can't change this.

So you say that "the transmitter does not couple exactly identically to the two tanks"? If we adjust the angle so that each tank is the same angle to the transmitter, then the intensity and phase of the signal arriving at each tank should be identical, except for the signal radiated by the adjacent tank.? Agreed?

cedrichirschi.21,

I edited the TPS43061 model file to remove all bad instances of the {{double curly braces}}, and remove the unused Td parameter.? I will upload it soon.? It eliminated all the errors.? But the behavior seems similar.

I am puzzled why the switching outputs from the TPS43061 to the CDS86330Q3D never switch when the output voltage stabilizes.? Is that expected?

The amount of overshoot seems to depend on the size of the load resistor.? More load -> less overshoot.

When the Enable pin goes negative, it is for less than 10 picoseconds.? I think it's in the part of the simulation where LTspice struggles, perhaps fighting with bad model formulas.? The timesteps become very small there, less than a femtosecond.??After the output stabilizes, the simulation proceeds much much much faster.

Andy

"Your simulation cannot be true to your real world set-up, since it gives a different but entirely expected result."
Precisely why I'm asking questions here.

I implemented the changes you suggested, using the .step param list 0.0001 0.01, and increased K2 to 0.00001 leaving K3 at 0.000001

I tried probing both tanks separately, and a differential and I see no resonance splitting at all.
Uploading latest rev.

开云体育

This randomness may be inherent in the routine that adjusts the time-step around discontinuities. The finite bit-depth can result in at least an LSB variation between different runs. It could be much larger, but the probability would be a steeply-decreasing function of variation size.

cedrichirschi.21,

Hmm, I need to modify what I wrote before.? If I re-run your simulation, I get different results after the first one.? One of mine even looks similar to your screenshot (png file).

Most SPICE simulations should simulate exactly the same every time -- unless there is an explicit random source (unlikely).? However, some models or simulations do behave a little randomly, which I don't understand.? It has been suggested in this group (but never confirmed) that LTspice has a tiny amount of randomness built-in, which helps prevent it from being stuck when something like a singular matrix comes along.? (Simulated circuits tend to be ideal and exactly balanced.??The randomness nudges it slightly, letting it get past the problem point.)

Andy

cedrichirschi.21,

Hmm, I need to modify what I wrote before.? If I re-run your simulation, I get different results after the first one.? One of mine even looks similar to your screenshot (png file).

Most SPICE simulations should simulate exactly the same every time -- unless there is an explicit random source (unlikely).? However, some models or simulations do behave a little randomly, which I don't understand.? It has been suggested in this group (but never confirmed) that LTspice has a tiny amount of randomness built-in, which helps prevent it from being stuck when something like a singular matrix comes along.? (Simulated circuits tend to be ideal and exactly balanced.??The randomness nudges it slightly, letting it get past the problem point.)

Andy

FYI, when a SPICE model is poorly made or has math errors, it can force the simulator towards extreme voltages and currents.? This can cause those spikes you saw.

Behavior of components in SPICE models should be continuous in both value and derivative.? Many models are not.? Many models, including those from respectable IC manufacturers, have abrupt discontinuities (due to things like IF() statements in formulas) and they can make the simulation go haywire.

Andy