Technically it's possible to sample at different moments in time as well as start the FFT analysis at (slightly) different moments.
That way you effectively improve the resolution already as well as improving noise.
Can you PLEASE stop posting absolute distortion plots?
They are not only extremely deceiving, but counting divisions is getting tyring as well.
Some color would be also nice?
It's so extremely difficult to see the different harmonics with just black and white.
Btw, another very important but overlooked issue with measuring distortion, is an equal frequency response between drivers.
This is even more true for the low-end response, otherwise it most definitely isn't apples vs apples anymore.
Any plot requires a technical understanding to interpret properly, and even the most informative plot can be misunderstood by an ignorant reader. Normalizing to the fundamental tells a slightly different story than an absolute measure, but both styles convey important information. "Extremely deceiving" is a bit of a stretch... yes?
@tktran303 - thank you for your continued efforts in exploring this topic.
This is an interesting conclusion. I had suspected this to be true, but I had no evidence.
No it's not
That is not an opinion either.
There are fundamental things that just don't and will never show up when absolute numbers are being used.
This has been explained and shown a billion times, so I am not gonna explain it again anymore.
Besides the fact that if people are talking so deep into this subject, this should be more than obvious.
Otherwise talking about 1/24 smoothing has absolutely no meaning either. That is again not an opinion.
And as always, no offense again, I have no way of understanding how to phrase these kind of things in your culture without upsetting people.
Last edited:
yes
i assume you mean the good old spectrum analysers? they used the heterodyne principle with a fixed narrows band filter and a sweeping local oscillator plus a mixer. They can also measure transfer functions (network analyser) by feeding in the oscillator to the DUT. I believe it did a continuous sweep so essentially a chirp stimulus and the analyser is a tracking band pass filter. This is obviously smoothed and has no aliasing issues.
It is quite similar to the old star “Time Delay Spectrometry” that Heiser is known for. As I understand it, it was a very convoluted way to do gated measurements. However, one must understand that this was before processing hardware allowed MLS or FFT in real time.
In fact I very deeply disagree with @tktran303 conclusion.
Just compare data from these sources and you will see that data correlates well.
So the evidence actually shows the opposite.
The exception is klang und ton, no idea what they are doing, but their data isn't useable.
So by just looking at the numbers we can just never make that conclusion.
agree with the colours. But the relative HD plot loses information about the fundamental so cannot stand alone. Best to show both IMO.
yes, that is a very important point. The responses should preferably be EQ’ed to a common eg band pass response and pass band SPL for proper comparison. Then the info about the fundamental is covered and only having the relative plot is justified.
This is a bit off topic but I have a sincere question, one that seriously interests me.
I use still use TDS for most acoustical measurements because I am familiar with it and for the noise immunity.
I realize that different measurement platforms have differing strengths and weaknesses and this is no attempt to push the virtues of TDS.
Is there a measurement platform that compares to TDS for noise immunity? With TDS I can run a slow sweep or chirp through a loudspeaker while listening to music through that loud speaker and get the same data as without music playing.
Is this possible with other measurement platforms?
Thank you.
Barry.
hi Barry,
There is a big difference between just measuring the FR of a speaker or to resolve very weak harmonics perhaps 80dB or more down. To get about 1dB accuracy on the FR we just need about 20dB SNR but to resolve harmonics at -80dB with 1dB accuracy we need 100dB SNR.
TDS does not measure harmonics as I understand it (please correct me).
Any chirp or stepped sine method that sweeps slowly has high noise immunity and TDS is in this camp.
There is a big difference between just measuring the FR of a speaker or to resolve very weak harmonics perhaps 80dB or more down. To get about 1dB accuracy on the FR we just need about 20dB SNR but to resolve harmonics at -80dB with 1dB accuracy we need 100dB SNR.
TDS does not measure harmonics as I understand it (please correct me).
Any chirp or stepped sine method that sweeps slowly has high noise immunity and TDS is in this camp.
Hello.
It can. You just set the tracking filter to any fraction or multiple of the fundamental frequency sweep and it looks for energy in those bands.
Here is an example from a JBL test. I use it in the same way on a PWT
Again I didn’t mean to derail this discussion.
It kind of came full circle as I was reading this complete thread. When I first built my PWT and started measuring compression drivers I was shocked to see that the current state of the art drivers I was measuring would not pass the old Bell Labs distortion criteria of the 2nd and 3rd harmonics being at least 30 dB below the fundamental.
It didn’t take long to figure out that at PWT test levels my mics were dirtier than the drivers. That was easily solved.
I am also interested in very low level noise floor testing so I am mining this thread for information to continue my learning.
Thank you again to all who contribute here.
Barry.
It can. You just set the tracking filter to any fraction or multiple of the fundamental frequency sweep and it looks for energy in those bands.
Here is an example from a JBL test. I use it in the same way on a PWT
Again I didn’t mean to derail this discussion.
It kind of came full circle as I was reading this complete thread. When I first built my PWT and started measuring compression drivers I was shocked to see that the current state of the art drivers I was measuring would not pass the old Bell Labs distortion criteria of the 2nd and 3rd harmonics being at least 30 dB below the fundamental.
It didn’t take long to figure out that at PWT test levels my mics were dirtier than the drivers. That was easily solved.
I am also interested in very low level noise floor testing so I am mining this thread for information to continue my learning.
Thank you again to all who contribute here.
Barry.
That sounds about right, but they were called transfer function analyzers, and the one that I recall using applied a stepped-sine approach, rather than a continuous sweep. There was some mention of a needed settling time between each stepped-sine measurement, if I recall correctly.
Forgive me, but I do not understand what you are saying above. Can you point me to some relevant reading material covering the concepts in question?
The apparent noise e.g. at 2khz is -72dB ! (-relative to fundamental)
the D4 there is 0.025% (or lower, the noise knows...)
[ REW (Farina log sine) Sweep, settings 256K samples x8 repetitions: time taken: 45 seconds ]
STEPS, there are no signs of noise to judge
but the D4 at 2kHz reads 0.015%
[ STEPS: 1/48 octave- time taken for test: ~12 mins ]
I trust REW here, as it shows the signs of noise clearly.
I do not care 🙂
Btw. there are different sources of noise in measurement setups, that need to be adressed with different measures.
What counts is the "low enough" apparent noisefloor, that is not hidden from sight.
Best regards,
Bernd
PS referred to the measurements in #66
Last edited:
On the topic of graphs with colors vs B&W
In psychological studies- the use of the colour RED leads to bias. Perhaps this is physiological artefact i.e.
Blood->danger->death! Or an effect of the human visual system i.e. more L (red) cones, than M (green) and S (blue) cones, biologically affecting our increased ability to actually see red.
Hence the reason for monochromatic graphs in many peer reviewed research papers.
Furthermore, 3-4 lines has been shown as the ideal number to convey a point.
Wolfgang Klippel (and others) have found that THD isn’t a reliable predictor of sonic quality, yet the Klippel NFS show THD in red.
I find it hard it really hard to look past or ignore the red when I’m looking at the graph.
Let's not forget the importance of avoiding red+green- the most common form of colour blindness.
In psychological studies- the use of the colour RED leads to bias. Perhaps this is physiological artefact i.e.
Blood->danger->death! Or an effect of the human visual system i.e. more L (red) cones, than M (green) and S (blue) cones, biologically affecting our increased ability to actually see red.
Hence the reason for monochromatic graphs in many peer reviewed research papers.
Furthermore, 3-4 lines has been shown as the ideal number to convey a point.
Wolfgang Klippel (and others) have found that THD isn’t a reliable predictor of sonic quality, yet the Klippel NFS show THD in red.
I find it hard it really hard to look past or ignore the red when I’m looking at the graph.
Let's not forget the importance of avoiding red+green- the most common form of colour blindness.
I have a mild form of color blindness... yet I find that many plots are inscrutable because the automated (default) line colors seem so similar to me, such as pastel colors with small changes in hue.
When I make plots, I use bold colors with high contrast. I also use variations in dash-dot so that even someone with complete color blindness could discern the plot.
Agreed.
On the other topics re: colour graphs vs black and white, or relative distortion vs absolute, I see your point.
Your direct and frank thoughts are always welcome here in my threads @b_force
Last edited:
Hello Boden.
A Larson Davis 2200C 200 Volt power supply for a 910B mic pre and a 2530 capsule.
Good to 176dB SPL and nearly flat from something like 3Hz to 70KHz.
Barry.