|
Why does THD give me zero when partial THD is 0.003792%?? What am I doing wrong?
Fourier components of ?v(out)
N-Period=60
DC component:-0.000986685
Harmonic ? ?Frequency ? ? Fourier ? ? Normalized ? ? Phase ? ? ?Normalized
?Number ? ? ? [Hz] ? ? ? Component ? ? Component ? ?[degree] ? ?Phase [deg]
? ? 1 ? ? ? ?2.000e+04 ? ? 1.005e+01 ? ? 1.000e+00 ? ? ? 94.00?° ? ? ? ?0.00?°
? ? 2 ? ? ? ?4.000e+04 ? ? 3.651e-04 ? ? 3.633e-05 ? ? ? 96.11?° ? ? ? ?2.11?°
? ? 3 ? ? ? ?6.000e+04 ? ? 9.642e-05 ? ? 9.596e-06 ? ? ? -0.69?° ? ? ?-94.69?°
? ? 4 ? ? ? ?8.000e+04 ? ? 3.907e-05 ? ? 3.888e-06 ? ? ?-99.48?° ? ? -193.48?°
? ? 5 ? ? ? ?1.000e+05 ? ? 1.996e-05 ? ? 1.986e-06 ? ? ?169.62?° ? ? ? 75.62?°
? ? 6 ? ? ? ?1.200e+05 ? ? 1.518e-05 ? ? 1.511e-06 ? ? ? 77.47?° ? ? ?-16.53?°
? ? 7 ? ? ? ?1.400e+05 ? ? 1.444e-05 ? ? 1.437e-06 ? ? ?-43.42?° ? ? -137.42?°
? ? 8 ? ? ? ?1.600e+05 ? ? 6.600e-06 ? ? 6.568e-07 ? ? -102.82?° ? ? -196.82?°
? ? 9 ? ? ? ?1.800e+05 ? ? 1.347e-05 ? ? 1.341e-06 ? ? -120.04?° ? ? -214.04?°
Partial Harmonic Distortion: 0.003792%
Total Harmonic Distortion: ? 0.000000%
|
Please upload your .asc file, with associated models and symbols,
to the Files->Temp- folder.
Without it, we need a crystal ball.
Le 03/02/2025 à 03:58, rlim701 via
groups.io a écrit?:
toggle quoted message
Show quoted text
Why does THD give me zero when partial THD is 0.003792%??
What am I doing wrong?
Fourier components of ?v(out)
N-Period=60
DC component:-0.000986685
Harmonic ? ?Frequency ? ? Fourier ? ? Normalized ? ? Phase
? ? ?Normalized
?Number ? ? ? [Hz] ? ? ? Component ? ? Component ? ?[degree] ?
?Phase [deg]
? ? 1 ? ? ? ?2.000e+04 ? ? 1.005e+01 ? ? 1.000e+00 ? ? ?
94.00?° ? ? ? ?0.00?°
? ? 2 ? ? ? ?4.000e+04 ? ? 3.651e-04 ? ? 3.633e-05 ? ? ?
96.11?° ? ? ? ?2.11?°
? ? 3 ? ? ? ?6.000e+04 ? ? 9.642e-05 ? ? 9.596e-06 ? ? ?
-0.69?° ? ? ?-94.69?°
? ? 4 ? ? ? ?8.000e+04 ? ? 3.907e-05 ? ? 3.888e-06 ? ?
?-99.48?° ? ? -193.48?°
? ? 5 ? ? ? ?1.000e+05 ? ? 1.996e-05 ? ? 1.986e-06 ? ?
?169.62?° ? ? ? 75.62?°
? ? 6 ? ? ? ?1.200e+05 ? ? 1.518e-05 ? ? 1.511e-06 ? ? ?
77.47?° ? ? ?-16.53?°
? ? 7 ? ? ? ?1.400e+05 ? ? 1.444e-05 ? ? 1.437e-06 ? ?
?-43.42?° ? ? -137.42?°
? ? 8 ? ? ? ?1.600e+05 ? ? 6.600e-06 ? ? 6.568e-07 ? ?
-102.82?° ? ? -196.82?°
? ? 9 ? ? ? ?1.800e+05 ? ? 1.347e-05 ? ? 1.341e-06 ? ?
-120.04?° ? ? -214.04?°
Partial Harmonic Distortion: 0.003792%
Total Harmonic Distortion: ? 0.000000%
|
Which version of LTspice are you using? If
it's 24.1.anything, it may be another bug to report.
On 2025-02-03 02:58, rlim701 via
groups.io wrote:
Why does THD give me zero when partial THD is 0.003792%??
What am I doing wrong?
Fourier components of ?v(out)
N-Period=60
DC component:-0.000986685
Harmonic ? ?Frequency ? ? Fourier ? ? Normalized ? ? Phase
? ? ?Normalized
?Number ? ? ? [Hz] ? ? ? Component ? ? Component ? ?[degree] ?
?Phase [deg]
? ? 1 ? ? ? ?2.000e+04 ? ? 1.005e+01 ? ? 1.000e+00 ? ? ?
94.00?° ? ? ? ?0.00?°
? ? 2 ? ? ? ?4.000e+04 ? ? 3.651e-04 ? ? 3.633e-05 ? ? ?
96.11?° ? ? ? ?2.11?°
? ? 3 ? ? ? ?6.000e+04 ? ? 9.642e-05 ? ? 9.596e-06 ? ? ?
-0.69?° ? ? ?-94.69?°
? ? 4 ? ? ? ?8.000e+04 ? ? 3.907e-05 ? ? 3.888e-06 ? ?
?-99.48?° ? ? -193.48?°
? ? 5 ? ? ? ?1.000e+05 ? ? 1.996e-05 ? ? 1.986e-06 ? ?
?169.62?° ? ? ? 75.62?°
? ? 6 ? ? ? ?1.200e+05 ? ? 1.518e-05 ? ? 1.511e-06 ? ? ?
77.47?° ? ? ?-16.53?°
? ? 7 ? ? ? ?1.400e+05 ? ? 1.444e-05 ? ? 1.437e-06 ? ?
?-43.42?° ? ? -137.42?°
? ? 8 ? ? ? ?1.600e+05 ? ? 6.600e-06 ? ? 6.568e-07 ? ?
-102.82?° ? ? -196.82?°
? ? 9 ? ? ? ?1.800e+05 ? ? 1.347e-05 ? ? 1.341e-06 ? ?
-120.04?° ? ? -214.04?°
Partial Harmonic Distortion: 0.003792%
Total Harmonic Distortion: ? 0.000000%
--
OOO - Own Opinions Only
Best Wishes
John Woodgate
Keep trying
|
No need for sorcery here. Any experienced LTspice user can tell from these numbers that this is most likely a machine precision issue.
.four is a post-processing step. It works with the numbers stored in the .raw file as single precision by default. That's only 6.5 decimal digits.
Try again with
?
.options numdgt=16
?
If that doesn't work, get the crystal ball ready.
?
Best Regards,
Mathias
?
On Mon, Feb 3, 2025 at 09:04 AM, Jerry Lee Marcel wrote:
Please upload your .asc file, with associated models and symbols, to the Files->Temp- folder.
Without it, we need a crystal ball.
Le 03/02/2025 à 03:58, rlim701 via groups.io a écrit?:
Why does THD give me zero when partial THD is 0.003792%?? What am I doing wrong?
Fourier components of ?v(out)
N-Period=60
DC component:-0.000986685
Harmonic ? ?Frequency ? ? Fourier ? ? Normalized ? ? Phase ? ? ?Normalized
?Number ? ? ? [Hz] ? ? ? Component ? ? Component ? ?[degree] ? ?Phase [deg]
? ? 1 ? ? ? ?2.000e+04 ? ? 1.005e+01 ? ? 1.000e+00 ? ? ? 94.00?° ? ? ? ?0.00?°
? ? 2 ? ? ? ?4.000e+04 ? ? 3.651e-04 ? ? 3.633e-05 ? ? ? 96.11?° ? ? ? ?2.11?°
? ? 3 ? ? ? ?6.000e+04 ? ? 9.642e-05 ? ? 9.596e-06 ? ? ? -0.69?° ? ? ?-94.69?°
? ? 4 ? ? ? ?8.000e+04 ? ? 3.907e-05 ? ? 3.888e-06 ? ? ?-99.48?° ? ? -193.48?°
? ? 5 ? ? ? ?1.000e+05 ? ? 1.996e-05 ? ? 1.986e-06 ? ? ?169.62?° ? ? ? 75.62?°
? ? 6 ? ? ? ?1.200e+05 ? ? 1.518e-05 ? ? 1.511e-06 ? ? ? 77.47?° ? ? ?-16.53?°
? ? 7 ? ? ? ?1.400e+05 ? ? 1.444e-05 ? ? 1.437e-06 ? ? ?-43.42?° ? ? -137.42?°
? ? 8 ? ? ? ?1.600e+05 ? ? 6.600e-06 ? ? 6.568e-07 ? ? -102.82?° ? ? -196.82?°
? ? 9 ? ? ? ?1.800e+05 ? ? 1.347e-05 ? ? 1.341e-06 ? ? -120.04?° ? ? -214.04?°
Partial Harmonic Distortion: 0.003792%
Total Harmonic Distortion: ? 0.000000%
|
But surely that issue would not find a THD
less than the partial THD up to the 9th harmonic?
On 2025-02-03 11:11, Mathias Born via
groups.io wrote:
No need for sorcery here. Any experienced LTspice user can
tell from these numbers that this is most likely a machine
precision issue.
.four is a post-processing step. It works with the numbers
stored in the .raw file as single precision by default. That's
only 6.5 decimal digits.
Try again with
?
.options numdgt=16
?
If that doesn't work, get the crystal ball ready.
?
Best Regards,
Mathias
?
On Mon, Feb 3, 2025 at 09:04 AM, Jerry Lee Marcel wrote:
Please upload your .asc file, with associated models and
symbols, to the Files->Temp- folder.
Without it, we need a crystal ball.
Le 03/02/2025 à 03:58, rlim701
via groups.io a écrit?:
Why does THD give me zero when partial THD is
0.003792%?? What am I doing wrong?
Fourier components of ?v(out)
N-Period=60
DC component:-0.000986685
Harmonic ? ?Frequency ? ? Fourier ? ? Normalized ? ?
Phase ? ? ?Normalized
?Number ? ? ? [Hz] ? ? ? Component ? ? Component ?
?[degree] ? ?Phase [deg]
? ? 1 ? ? ? ?2.000e+04 ? ? 1.005e+01 ? ? 1.000e+00 ? ? ?
94.00?° ? ? ? ?0.00?°
? ? 2 ? ? ? ?4.000e+04 ? ? 3.651e-04 ? ? 3.633e-05 ? ? ?
96.11?° ? ? ? ?2.11?°
? ? 3 ? ? ? ?6.000e+04 ? ? 9.642e-05 ? ? 9.596e-06 ? ? ?
-0.69?° ? ? ?-94.69?°
? ? 4 ? ? ? ?8.000e+04 ? ? 3.907e-05 ? ? 3.888e-06 ? ?
?-99.48?° ? ? -193.48?°
? ? 5 ? ? ? ?1.000e+05 ? ? 1.996e-05 ? ? 1.986e-06 ? ?
?169.62?° ? ? ? 75.62?°
? ? 6 ? ? ? ?1.200e+05 ? ? 1.518e-05 ? ? 1.511e-06 ? ? ?
77.47?° ? ? ?-16.53?°
? ? 7 ? ? ? ?1.400e+05 ? ? 1.444e-05 ? ? 1.437e-06 ? ?
?-43.42?° ? ? -137.42?°
? ? 8 ? ? ? ?1.600e+05 ? ? 6.600e-06 ? ? 6.568e-07 ? ?
-102.82?° ? ? -196.82?°
? ? 9 ? ? ? ?1.800e+05 ? ? 1.347e-05 ? ? 1.341e-06 ? ?
-120.04?° ? ? -214.04?°
Partial Harmonic Distortion: 0.003792%
Total Harmonic Distortion: ? 0.000000%
--
OOO - Own Opinions Only
Best Wishes
John Woodgate
Keep trying
|
Don't underestimate the intricacies of floating point math:
?
?
Partial and harmonic distortion are calculated very differently. If the result is dominated by rounding errors, anything is possible.
?
Best Regards,
Mathias
?
On Mon, Feb 3, 2025 at 12:24 PM, John Woodgate wrote:
But surely that issue would not find a THD less than the partial THD up to the 9th harmonic?
|
What could be simpler? ?
On 2025-02-03 12:12, Mathias Born via
groups.io wrote:
Don't underestimate the intricacies of floating point math:
?
?
Partial and harmonic distortion are calculated very
differently. If the result is dominated by rounding errors,
anything is possible.
?
Best Regards,
Mathias
?
On Mon, Feb 3, 2025 at 12:24 PM, John Woodgate wrote:
But surely that issue
would not find a THD less than the partial THD up to the
9th harmonic?
--
OOO - Own Opinions Only
Best Wishes
John Woodgate
Keep trying
|
On Mon, Feb 3, 2025 at 06:24 AM, John Woodgate wrote:
But surely that issue would not find a THD less than the partial THD up to the 9th harmonic?
Those two results are calculated quite differently.? That's why it can be less.
?
The Partial THD calculation that includes only the first N harmonics does it by first calculating the component at each harmonic, then adding them up, RMS-wise.
?
The Total THD calculation ignores the harmonics that could be calculated at each harmonic.? Instead, it compares the level of the calculated fundamental against the level of the whole signal.? Because of limited accuracy/resolution, if the calculated fundamental's amplitude happens to be equal to or greater than the RMS voltage or current of the raw signal, then it has to conclude that it must have zero harmonic content.? Think of it as meaning "less than I can determine".? When the numbers are this close, when the contribution of the harmonics is so small, then it becomes limited by numerical resolution.
?
?
|
Thanks, Andy. I can understand why that was
done a long time ago, when calculation speeds and memory
capacity were far less than now. But it's obviously dangerous in
principle to use two different algorithms, and that has now
surfaced. I have used .FOUR to evaluate over 40 harmonics in a
'noiseless' simulation, with no speed or memory issues. I can
see a problem if LTspice is asked, in error, to calculate THD
over a bandwidth which includes noise at a level greater than
-10dB relative to the nearest (in frequency) harmonic, but that
condition could be caught and flagged.
On 2025-02-03 17:42, Andy I via
groups.io wrote:
On Mon, Feb 3, 2025 at 06:24 AM, John Woodgate wrote:
But surely that issue
would not find a THD less than the partial THD up to the
9th harmonic?
Those two results are calculated quite differently.? That's why
it can be less.
?
The Partial THD calculation that includes only the first N
harmonics does it by first calculating the component at each
harmonic, then adding them up, RMS-wise.
?
The Total THD calculation ignores the harmonics that could be
calculated at each harmonic.? Instead, it compares the level of
the calculated fundamental against the level of the whole
signal.? Because of limited accuracy/resolution, if the
calculated fundamental's amplitude happens to be equal to or
greater than the RMS voltage or current of the raw signal, then
it has to conclude that it must have zero harmonic content.?
Think of it as meaning "less than I can determine".? When the
numbers are this close, when the contribution of the harmonics
is so small, then it becomes limited by numerical resolution.
?
?
--
OOO - Own Opinions Only
Best Wishes
John Woodgate
Keep trying
|
On Mon, Feb 3, 2025 at 01:06 PM, John Woodgate wrote:
Thanks, Andy. I can understand why that was done a long time ago, when calculation speeds and memory capacity were far less than now. But it's obviously dangerous in principle to use two different algorithms, and that has now surfaced. I have used .FOUR to evaluate over 40 harmonics in a 'noiseless' simulation, with no speed or memory issues. I can see a problem if LTspice is asked, in error, to calculate THD over a bandwidth which includes noise at a level greater than -10dB relative to the nearest (in frequency) harmonic, but that condition could be caught and flagged.
Most .TRAN simulations have no noise whatsoever, so that should rarely be a problem.? But there is math 'noise' to contend with.
?
Personally, I would not have called it 'dangerous' that it calculates the two (T)HD numbers differently; but that may be because I know and understand that the methods are different.? In some respects it is not that different than knowing that a smaller timestep almost always helps accuracy, even when all the frequencies are so low that such a small timestep "should" not be necessary.? We who have used LTspice for a while realize that this is not always the case.
?
Andy
?
|
I meant math noise, but of course simulated
electronic noise would have a similar effect.
On 2025-02-03 19:56, Andy I via
groups.io wrote:
On Mon, Feb 3, 2025 at 01:06 PM, John Woodgate wrote:
Thanks, Andy. I can
understand why that was done a long time ago, when
calculation speeds and memory capacity were far less than
now. But it's obviously dangerous in principle to use two
different algorithms, and that has now surfaced. I have
used .FOUR to evaluate over 40 harmonics in a 'noiseless'
simulation, with no speed or memory issues. I can see a
problem if LTspice is asked, in error, to calculate THD
over a bandwidth which includes noise at a level greater
than -10dB relative to the nearest (in frequency)
harmonic, but that condition could be caught and flagged.
Most .TRAN simulations have no noise whatsoever, so that should
rarely be a problem.? But there is math 'noise' to contend with.
?
Personally, I would not have called it 'dangerous' that it
calculates the two (T)HD numbers differently; but that may be
because I know and understand that the methods are different.?
In some respects it is not that different than knowing that a
smaller timestep almost always helps accuracy, even when all the
frequencies are so low that such a small timestep "should" not
be necessary.? We who have used LTspice for a while realize that
this is not always the case.
?
Andy
?
--
OOO - Own Opinions Only
Best Wishes
John Woodgate
Keep trying
|
Alright, guys.? I uploaded heng-24v.asc and heng-models.txt.? I have a 'throw_cycles' parameter that throws away cycles before doing .four.? If I set that to 100, I see THD=0.002% but when I set that to >1000, reported THD is always 0.000000%.? Setting numdgt to 16 did not make a difference.
Thanks in advance,
Richard
|
Richard,
?
You forgot to upload the symbol file LG_single.asy.? We can't see your whole schematic, nor run the simulation, without it.? If you forgot the symbol, you might have forgotten its model too.
?
It makes perfect sense that throwing away more cycles (waiting for things to stabilize) results in lower distortion.
?
It also makes sense that there is a point at which you just can't make the calculations any better.
?
Which version of LTspice are you using?? There are known bugs with NUMDGT in the latest version, 24.1.1 - which is turning out to be a major disaster, by the way.
?
Andy
?
|
Richard,
?
I don't know if this makes any difference (since I can't try your simulation).? I am only suggesting it as something to try.? Your simulation saves data for 61 cycles, then it asks the .FOUR to calculate harmonics over exactly 61 cycles.? But what if round-off causes the simulated interval to be ever slightly smaller than 61 cycles?? Can you save data for 61 cycles, then set the .FOUR's Ncycles to 60?
?
Andy
?
|
Just uploaded LG_single.asc.? It did not make any difference whether I am using version 24.0.12 or 24.1.1.? Andy, I think I see where you are going.? Still trying to digest your post.....
|
Are you sure you uploaded the right schematic? LG_single.asc doesn't
even have an analysis directive. It just looks like a loop probe,
which designed to be used in .AC analyses, anyway.
--
Regards,
Tony
On 03/02/2025 21:57, rlim701 via
groups.io wrote:
toggle quoted message
Show quoted text
Just uploaded
LG_single.asc.? It did not make any difference whether I am using
version 24.0.12 or 24.1.1.? Andy, I think I see where you are
going.? Still trying to digest your post.....
|
Sorry, I missed that it is supposed to be used with another
schematic.
For future reference, it is best to upload all required files
packaged in a zip, not separately. that makes it easier and more
convenient for sharing.
--
Regards,
Tony
On 03/02/2025 23:07, Tony Casey wrote:
toggle quoted message
Show quoted text
Are
you sure you uploaded the right schematic? LG_single.asc doesn't
even have an analysis directive. It just looks like a loop probe,
which designed to be used in .AC analyses, anyway.
|
Is there a good reason why you are performing the the .FOUR over 61
cycles? You don't get a more accurate answer using more cycles.
Unless you have some time-varying perturbation going on, why not use
.FOUR with its default setting, which is 1 cycle? Try that and see
if it fixes your issue.
--
Regards,
Tony
On 03/02/2025 21:14, rlim701 via
groups.io wrote:
toggle quoted message
Show quoted text
Alright, guys.? I
uploaded heng-24v.asc and heng-models.txt.? I have a
'throw_cycles' parameter that throws away cycles before doing
.four.? If I set that to 100, I see THD=0.002% but when I set that
to >1000, reported THD is always 0.000000%.? Setting numdgt to
16 did not make a difference.
|
On Mon, Feb 3, 2025 at 03:57 PM, <rlim701@...> wrote:
Just uploaded LG_single.asc.
That's good -- but it does not fully help.? Your main schematic (heng-24v.asc) wants to load the symbol named "LG_single".? That symbol is a file named "LG_single.asy", which is still on your computer only.? None of the rest of the world can try your simulation because none of the rest of us has that symbol file.
?
Maybe you don't want us to try your simulation, and that is sort of OK if it's true.? But wouldn't it help to give us all the parts so that we can try it?? Sometimes, someone else may discover something that a single pair of eyes doesn't find.
?
As it appears, that symbol file represents the lower-level hierarchical block in the schematic "LG_single.asc".? So we now have THAT file.? But we are still missing the symbol that represents that lower schematic on your main schematic.? Without it, there's a hole in your schematic and some missing current paths, causing the distortion to exceed 1%.
?
Andy
?
|
Sorry.? Just uploaded LG_single.asy.? If missing, it could be shorted out in the schematic.? Anyway, I think Andy is correct in that I was asking .FOUR to perform exactly 61 cycles and I suspect the end time was perhaps one point short of 61 cycles.? when I added 1us to the end time, I was able to get a THD that is non-zero.
Thanks everyone!
?
Richard
|
John Woodgate
Tony Casey