Loading Profile...
THD
Hi Chris,
I’ve been playing around with the new harmonic distortion graphs. Great feature.
Questions :
- When an IR is normalized, the distortion graphs do not follow the signal (H1). In order to check the difference in level between H1 and the others, I presume it is the non-normalized presentation that’s good?
- I need to get a single number qualifier for distortion (or, at least, per distortion order). I realize THD is traditionaly measured out of a single frequency, but it does seem possible to get the ratio of the total level of added harmonics over the fundamental for the whole spectrum. We could use either averaged level or sum... Is this already implemented in any way in Fuzz? Anybody here knows if this is an officially developped technique for qualifiying THD?
- How about adding the level of H2 and H3 to the export data so we could play with this by ourselves on a spreadsheet?
- Any plan to add more harmonic orders?
I’ve been playing around with the new harmonic distortion graphs. Great feature.
Questions :
- When an IR is normalized, the distortion graphs do not follow the signal (H1). In order to check the difference in level between H1 and the others, I presume it is the non-normalized presentation that’s good?
- I need to get a single number qualifier for distortion (or, at least, per distortion order). I realize THD is traditionaly measured out of a single frequency, but it does seem possible to get the ratio of the total level of added harmonics over the fundamental for the whole spectrum. We could use either averaged level or sum... Is this already implemented in any way in Fuzz? Anybody here knows if this is an officially developped technique for qualifiying THD?
- How about adding the level of H2 and H3 to the export data so we could play with this by ourselves on a spreadsheet?
- Any plan to add more harmonic orders?
1 person has this question
I have this question, too!
Tell me when someone answers.
The more people who ask this question, the more it gets noticed.
The more people who ask this question, the more it gets noticed.
-
Inappropriate?For the distortion, I preferred select myself the IR distortion, because I find the automatic fonction unclear, but it's impossible when the transmission time is short (the microphone close to the source).
I can't understand the frequency shift of H2 and H3. -
Inappropriate?Thend, can you explain the manual method you use that involves “selecting the IR distortion” ? -
-
Inappropriate?Thanks Thend. The image provided there is from a very “linear” system. It sems to me that It’s much more difficult (read “impossible”) to measure the distortion this way on complex IR’s.
Do you simply calculate the ratio from the power of the “distortion spark” over the main impulse?
One more thing, Thrend: You write that, using this way of measuring distortion, you don’t add the reverb’s distortion. But if you measure the acoustics of a room, you do want that distortion (I know I do)...
Can you point me to any literature about this?
Also, Chris, aren’t the distortion curves shifted to the left while they should be to the right? For instance, H2 should be fH1 X 2, but it appears to be fH1 / 2 in Fuzz’ graphs... Or am I not reading this correctly? -
Inappropriate?>Thanks Thend. The image provided there is from a very “linear” system. It >sems to me that It’s much more difficult (read “impossible”) to measure the >distortion this way on complex IR’s.
It's a very close measure, i think is the better way to measure the loudspeaker distortions.
>Do you simply calculate the ratio from the power of the “distortion spark” >over the main impulse?
I don't understand (my english...)
>One more thing, Thrend: You write that, using this way of measuring >distortion, you don’t add the reverb’s distortion. But if you measure the >acoustics of a room, you do want that distortion (I know I do)...
Yes and no, i think the reverb of the main impulse is more sufficient and representative than reverb distortion.
>Can you point me to any literature about this?
I saw that on a forum...
>Also, Chris, aren’t the distortion curves shifted to the left while they should >be to the right? For instance, H2 should be fH1 X 2, but it appears to be fH1 >/ 2 in Fuzz’ graphs...
I agree. -
Inappropriate?The frequency is shifted for a very good reason. If you check out the paper by S. Müller (http://www.anselmgoertz.de/Page10383/...), you'll see a very good explanation of this.
On Müller's paper, see page 47 for the following:
"To relate the frequency contents of one HIR spectrum to the fundamental, a spectral shift operation according to the order of the specific harmonic must be performed. For example, the spectral components of the fifth-order HIR are shifted horizontally to one- fifth of their original frequency."
Then, Fig. 25 shows this in more detail.
If you think about it a little more high-level, consider this. You've measured a signal over a specific frequency range, and thus you have information covering from DC -> 20kHz. How could you possibly know 20kHz's 2nd harmonic (of 40kHz)? You don't have measured data for it.
So, the maximum 2nd harmonic component can be known at 10kHz, maximum 4th harmonic at 5kHz, ...
Is this a little more clear? -
Inappropriate?Here's how I see things:
- No horizontal shift determining the frequency band where the distortion is audible.
- With horizontal shift determining the frequency band where there are a non-linearitys, but with the horizontal shift off FuzzMeasure I do not have this impression.
Or it's the automatic windowing that is the problem -
Inappropriate?Well, in my understanding, with a SR of 32 kHz, for instance, you cannot measure harmonic distorsion above 8 kHz, indeed... But you wouldn’t hear it either since any harmonic distortion components would be filtered out by the AD.
From Müller’s paper, I understand 2 possible reasons for this frequency shift: The first is that it’s a (proposed) convention, in which case personally I don’t agree with it as it’s counter intuitive since you cannot easily relate the harmonic component to its fundamental. The second comes from “After this operation, the shifted spectrum can be divided by the fundamental spectrum, yielding the frequency-dependent distortion fraction”(p47). I don’t understand why this would be, but it would be the way to calculate the HD percentage ...
BTW, it’d be great if Fuzz would perform that calculation as well... -
-
Inappropriate?So it's becoming increasingly clear that you both would prefer that I present the distortion graphs in the full frequency spectrum, as presented in Farina's paper, instead of the way that it's presented in Müller's paper.
Like Jean-Pierre mentioned, the harmonic distortion % is easier to calculate when you do the frequency domain shifting, but because I'm showing the distortion components in relation to their original values, my chosen mode of presentation is making distortion harder to read.
Also, the 'desaturate/darken' method of showing increasing distortion orders is not the best for clarity. Maybe I should make a routine, sort of like the 'create octave band decomposition', which separates out the individual distortion products in to separate measurement records? -
Inappropriate?Yes, Farina over Müller in this case. But again: Can you add HD % so we would see the graph in an easier readout form, plus the number telling the dist in % = best off all worlds...
About the colours, I prefer the single graph per IR as I often work on files with *many* IR’s and “decomposing” each makes the IR list difficult to deal with. Is it possible to 'desaturate/darken' more? Or you could use dedicated colours (that would never be assigned to any IR’s FR). -
Inappropriate?The more I work on this, the more I think that neither approach is good. :)
I think that harmonic distortion graphs are just going to have to display % and only the distortion values. See this graph for an example:
http://www.supermegaultragroovy.com/i...
The lighter green is the 2nd harmonic, and the darker green is the 3rd harmonic.
Thend makes a good point above about the automatic windowing, though. By default I'm using a rectangular window, which isn't ideal. I should instead be using the Tukey-style window as proposed by Müller's paper.
So, there's definitely still some work to be done here, though I think that the % display is already a huge improvement. -
Inappropriate?Personally I don’t agree with this “no-horizontal-shift” display. I find it counter intuitive since the trace is not where you actually hear it, i.e. H#2 for 1 kHz is a sound that is clearly heard at 2 kHz, not at 1 kHz...
And for %, I would prefer a single THD number.
This said, your example is very interesting as well. Maybe we could have both ways available (“Harmonic distortion [Farina]”, and “Harmonic distortion [realigned]”)
By extension, would it be possible to also measure non-harmonic distortion (for PCM converters, for instance)? -
Inappropriate?I think you mean to say you don't agree with a shifted display. The current THD graph that ships with FuzzMeasure 3.0 is shifted so that at 1kHz, you see the distortion components for 1kHz. That means the 2nd harmonic distortion value at 1kHz is the measured value at 2kHz, and the 3rd harmonic distortion value at 1kHz is the measured value at 3kHz.
If I do not shift the display, and leave it untouched as Farina does, it is up to you to mentally determine that the 2nd harmonic distortion value for 1kHz is the value you read at 2kHz. The more I think about it, the more I realize this is actually more confusing because you cannot easily visualize a range of frequencies that are free from distortion.
At any rate, the actual value people can relate to is either (a) harmonic distortion %, or (b) total harmonic distortion %.
Calculating (a) requires that you report the values for individual harmonics, as I currently do. So, you'd show a graph similar to the one I report now with 2nd and 3rd harmonics to properly display (a).
Calculating (b) requires that you figure out harmonic distortion up to some upper limit of harmonic (say, the 8th harmonic) and then combine them to figure out the total harmonic distortion.
As far as I can tell, there is no such thing as 'one number' for THD. When you see a single THD percentage reported in specifications, I'm pretty sure they must be reported with a reference frequency. So, 0.001% THD for a piece of equipment is probably calculated at the 1kHz point, and only up to some arbitrary upper-limit harmonic.
According to IEC 60268-5 (Sound system equipment Part 5: Loudspeakers), THD must be reported along with the frequencies at which they were measured. So, if you want to give just one number, you'd probably just take a graph like the one I displayed above, and report the value at 1kHz.
However, I wouldn't recommend this. If you need accurate specifications to report only a single value for THD, you're best served by the single sine wave method. FuzzMeasure's mechanism for calculating harmonic distortion will give you a very informative overall graph of the distortion behavior of a system—it tells you far more than a single number ever could. -
Inappropriate?> According to IEC 60268-5 (Sound system equipment Part 5: Loudspeakers), THD must
> be reported along with the frequencies at which they were measured. So, if you want
> to give just one number, you'd probably just take a graph like the one I displayed
> above, and report the value at 1kHz.
Correct. But I’m thinking about a way to make a single number out of, say, the mean value of all frequencies... this would be new (I think), and, though still imprecise, better than the actual 1 kHz value that doesn’t say anything about the low end, for instance.
In any case, if you could make the distortion data exportable, it would make it really useful I think. -
Inappropriate?3.1b2 appears to have dropped the dB harmonic distortion display.
I would call this a step backwards. Normal industry standards for plotting
loudspeaker driver distortion use a dB scale, also a "shifted" frequency
scale (as was the case with 3.1b1), and quite often a "raised by 20 or 30dB" amplitude shift for the harmonic components. An -option- to plot in % (vs dB)
might be use for some, but seems an odd choice as a default. -
Inappropriate?Dave,
Could you point me at some industry sources that list the decibel display as preferred display type?
The 3.1 beta is very mouldable right now and I want to be sure I get things like this sorted out sooner rather than later. I am a little bit in the dark when it comes to preferences for these more "exotic" graph types in fuzzmeasure. So, any pointers here would be appreciated. -
Inappropriate?Sure :)
http://www.jblpro.com/pub/technote/tn...
There are certainly other examples (including from JBL) that show
specific distortion artifacts in %, such as when attempting to
illuminate/highlight improvements that have been made in reducing 2nd
harmonic distortion in compression drivers, but those are specialized
case (and apparently are being used as marketing tools as opposed to
scientific investigations). Other vendors (when they are brave enough to
show these plots) tend to follow the dB scale usage, and the vertically
lifted harmonic plots.
BTW, the first run I ran of 3.1b2 using the harmonic distortion function
and comparing with 3.1b1 (converting from % to dB), simply using the
MacBook Pro internal speaker/mic, seemed to be not comparing well. I
didn't spend a lot of time to check this, but will try to this evening,
or tomorrow (% plot seemed to be a bit optimistic above a few hundred
Hz, iirc). I'll re-mail if I come up with a more concrete example.
Thanks and regards,
-dave -
Inappropriate?Never mind about my comparison (% vs dB) scale concern...
Below ~100Hz there was not enough signal from the MacBook
speaker to form a useful harmonic distortion value... i.e., was seeing
solely the "N" portion of "THD +N" :P
That said, I'd still hope to see the distortion products shown as dB relative
to the fundamental, even if as a selectable option. Thanks, -dave
EMPLOYEE
