I come with data...
At 1KHz, the difference between 1m and 1cm is 26dB.
Measured according to:
https://kimmosaunisto.net/Software/VituixCAD/VituixCAD_Measurement_REW.pdf
Don't shoot the microphone (messenger)
This have been very educational for me. Thank you @Hörnli and @IamJF
At 1KHz, the difference between 1m and 1cm is 26dB.
Measured according to:
https://kimmosaunisto.net/Software/VituixCAD/VituixCAD_Measurement_REW.pdf
Don't shoot the microphone (messenger)
This have been very educational for me. Thank you @Hörnli and @IamJF
When I do nearfield measurements, I adjust the voltage so that the mic is driven to 100 dB or less. I actually shoot for right at 100 dB. I am not sure what the advantage would be to use a full 2.83 V, which can correspond to very high SPL?
During the merger process, the NF response gets adjusted up/down to match the gated response, so ??? what are we doing here?
During the merger process, the NF response gets adjusted up/down to match the gated response, so ??? what are we doing here?
Totally with you on this @hifijim When looking at frequency response alone, that's all that's needed.
But when looking at distortion, we lose accuracy if trying to measure indoors. At 1m, or 1/2 meter, or even 31.6cm. This applies whether we use the sine sweep method or the stepped sine method...
Below is an excerpt of distortion measurements taken at 40cm, 25cm and nearfield conditions:
"With increasing distance, the room influence becomes more noticeable in both frequency and distortion traces.
[Below] is a direct comparison of % distortion traces for nearfield and 40cm measurements.
Increasing distance not only makes the traces noisier, but it also contributes to overall apparent distortion"
"The examples demonstrate the factors to be considered when making reproducible distortion measurements. They include signal and sound pressure levels, the quality of the microphone and its distance from the speaker, and the contribution of room reflections. The measurement setup should account for these factors as much as possible in order to optimise results."
Reference:
3.2.1. Factors influencing distortion measurements, pages 10-14 of
https://artalabs.hr/AppNotes/STEPS_Tutorial_Version_2_4_English.pdf
To accurately capture distortion, we need to strike a balance between getting close enough to the device, in order to get an accurate distortion measurement, but not too close that the microphone itself starts to distort (because it is observing SPL that is too high)
To be clear, to measure a speaker's distortion when playing x dB @1m, it appears to this author that need the microphone to be able to observe at least (x + 26) dB. Also, it appears that to accurately capture 0.01% distortion, an electret condenser microphone needs to observe some 40dB less than it's maximum SPL rating (if that rating is 3% distortion)
x + 26dB < max SPL (3% distortion) rating of microphone - 40
But when looking at distortion, we lose accuracy if trying to measure indoors. At 1m, or 1/2 meter, or even 31.6cm. This applies whether we use the sine sweep method or the stepped sine method...
Below is an excerpt of distortion measurements taken at 40cm, 25cm and nearfield conditions:
"With increasing distance, the room influence becomes more noticeable in both frequency and distortion traces.
[Below] is a direct comparison of % distortion traces for nearfield and 40cm measurements.
Increasing distance not only makes the traces noisier, but it also contributes to overall apparent distortion"
"The examples demonstrate the factors to be considered when making reproducible distortion measurements. They include signal and sound pressure levels, the quality of the microphone and its distance from the speaker, and the contribution of room reflections. The measurement setup should account for these factors as much as possible in order to optimise results."
Reference:
3.2.1. Factors influencing distortion measurements, pages 10-14 of
https://artalabs.hr/AppNotes/STEPS_Tutorial_Version_2_4_English.pdf
To accurately capture distortion, we need to strike a balance between getting close enough to the device, in order to get an accurate distortion measurement, but not too close that the microphone itself starts to distort (because it is observing SPL that is too high)
To be clear, to measure a speaker's distortion when playing x dB @1m, it appears to this author that need the microphone to be able to observe at least (x + 26) dB. Also, it appears that to accurately capture 0.01% distortion, an electret condenser microphone needs to observe some 40dB less than it's maximum SPL rating (if that rating is 3% distortion)
x + 26dB < max SPL (3% distortion) rating of microphone - 40
Last edited:
Buy microphones, don't burn fuel. By burning fuel you harm the environment and climate, and hence you make life worse for me. Please don't do that.
Indeed... I only buy microphones that were manufactured and transported without the burning of any fuel.
(If anyone did not pick up on the sarcasm there, let me explicit... I am being sarcastic)
Last edited:
@tktran303 - I found that if I smoothed the fundamental with 1/2 octave smoothing, and then calculated the % distortion, much of the room effects were mitigated. But I see your point...
@tktran303
Microphones have way higher noise at low frequencies as at higher frequencies. When measuring 70dB at low frequencies - you don't measure THD, you see microphone noise. The nearfield measurement will give better results.
+26dB is totally in the range - it depends on speaker size. Still no problem for a useful mic. You get 100-110dB SPL in your example, that's easy with a mic that is capable of 140dBSpl. And THD of the speaker will rise quickly at higher levels.
Microphones have way higher noise at low frequencies as at higher frequencies. When measuring 70dB at low frequencies - you don't measure THD, you see microphone noise. The nearfield measurement will give better results.
+26dB is totally in the range - it depends on speaker size. Still no problem for a useful mic. You get 100-110dB SPL in your example, that's easy with a mic that is capable of 140dBSpl. And THD of the speaker will rise quickly at higher levels.
This is what I’m looking for.
what’s the minimum viable solution for an indoor measurement that can more accurately characterise distortion of H2-H5, and other types (not just THD alone from H2/H3)
Something like your 3 way with a single SB29RNXL, @IamJF ‘s speaker with twin SB34RNXL that has a dynamic range of up to ~105-115 dB at the listening position, or @vineethkumar01 upcoming horn-loaded compression driver with twin 15” woofers.
Is it SPL?
Directivity?
Distortion?
What are the contributors to the differences?
(Apart from taste, budget, room etc)
I am not sure it is realistic. 115 dB at 3m -> 115+27.5 = 142.5 dB at the 15" woofer plane.
The surface area of the 34mm tweeter is about 1e-3m^2. Compare it to 4*pi*R^2=113m^2. If I am not mistaken, to have X dB at the listening position, you'll need X+50.5 close to the tweeter center. What is the speed of tweeter dome going to be? Supersonic? What the wave steepening distortions you would have there?
The surface area of the 34mm tweeter is about 1e-3m^2. Compare it to 4*pi*R^2=113m^2. If I am not mistaken, to have X dB at the listening position, you'll need X+50.5 close to the tweeter center. What is the speed of tweeter dome going to be? Supersonic? What the wave steepening distortions you would have there?
You haven't mentioned the alternative of multiple sweeps and then process to statistically reduce noise.
I don't know if any of the commonly available software does this but it should be possible.
The MicW M215 should make this unnecessary in practice, but I am curious to learn what is available, and your comments.
Best wishes
David
David, you may be opening a can of worms. Yes, you can average N spectrograms to get 3 dB of SNR improvement per doubling N. You can also increase the chirp length in N times, the result will be the same. But - you may have to increase the delay between chirps and use more advanced windows, like flat-top. In other words, you have to understand spectral analysis, both parametric and non-parametric methods. Suggesting it to a non-mathematician ...
Since that goes with a 10log, it needs a lot of averages to become effective.
With 16 averages we only gain 12dB of noise. 10*log(16)=12
I find anything over 8 averages not really practical anymore. Which is 9dB
Unless you enjoy spending a lot of time on measurements.
That's Ok, I am a mathematician, well, a mathematical statistician...
But most of the details should be taken care of by the software, just that I haven't looked at the various products to see what options they offer, if any.
What software do you use?
A ~10 dB improvement seems worth some inconvenience
A Farina style swept sine is efficient, so maybe a 10 times slowed sweep would be reasonable, while for less efficient methods 10x may be too slow.
Hence my interest in people's experiences.
Best wishes
David
Hi Dave,
I use REW.
According to the manual
“When measuring a system with high distortion levels use a long sweep setting (e.g. 1M or higher), at shorter sweep lengths the harmonics may affect each other giving misleading results. A spot check can be made at frequencies of interest using the RTA and the signal generator. If discrepancies are observed consider making a stepped sine measurement instead. The noise floor of log sweep distortion measurements rises with frequency. For the lowest noise floor with sweep measurements use multiple sweeps, but note that requires the input and output to be on the same device for reliable results.
Although much, much slower than a log sweep the stepped sine measurement can measure low distortion levels much more accurately than a sweep, particularly at high frequencies and for higher harmonics. Stepped sine distortion measurements show distortion components up to the ninth harmonic, THD and the noise floor, in the same way as the sweep-derived results, but also include THD+N (total harmonic distortion plus noise and non-harmonic distortion) and N (noise and non-harmonic distortion) alone. Note that the noise floor plot shows the spectral content of the noise measured with no signal playing. The 'Noise' in the N and THD+N shows the summed level of all non-harmonic distortions and noise across the frequency span for each test frequency. It consequently sits much higher than the noise floor plot. For stepped level measurements the X axis can be dB SPL, dBFS, dBu, dBV, dBW, V or W showing either the generator or input level.”
Reference:
https://www.roomeqwizard.com/help/help_en-GB/html/graph_distortion.html
6+ dB is definitely worthwhile IMHO.
In practice I find using 4 sweeps of 4M samples not much different between my preferred 2 sweeps of 2M samples between 5Hz and 40KHz.
So for frequency response only I use a 1M sample sweep which takes about 12 seconds.
For distortion I use 2M x 2, particularly on a noisier night.
I've privileged to have my own permanently set up lab, indoors. I
t’s the size of a typical Australian suburban bedroom, which is about 10 m^2, climate controlled etc. with a noise floor of about <30dB(A).
Nevertheless, I need to find a sweet spot between my patience, and my (family's) sanity.
No one likes Brrrrrrrrrrrrreeeeeeep!! going on repeatedly, for up to minutes at a time…
I use REW.
According to the manual
“When measuring a system with high distortion levels use a long sweep setting (e.g. 1M or higher), at shorter sweep lengths the harmonics may affect each other giving misleading results. A spot check can be made at frequencies of interest using the RTA and the signal generator. If discrepancies are observed consider making a stepped sine measurement instead. The noise floor of log sweep distortion measurements rises with frequency. For the lowest noise floor with sweep measurements use multiple sweeps, but note that requires the input and output to be on the same device for reliable results.
Although much, much slower than a log sweep the stepped sine measurement can measure low distortion levels much more accurately than a sweep, particularly at high frequencies and for higher harmonics. Stepped sine distortion measurements show distortion components up to the ninth harmonic, THD and the noise floor, in the same way as the sweep-derived results, but also include THD+N (total harmonic distortion plus noise and non-harmonic distortion) and N (noise and non-harmonic distortion) alone. Note that the noise floor plot shows the spectral content of the noise measured with no signal playing. The 'Noise' in the N and THD+N shows the summed level of all non-harmonic distortions and noise across the frequency span for each test frequency. It consequently sits much higher than the noise floor plot. For stepped level measurements the X axis can be dB SPL, dBFS, dBu, dBV, dBW, V or W showing either the generator or input level.”
Reference:
https://www.roomeqwizard.com/help/help_en-GB/html/graph_distortion.html
6+ dB is definitely worthwhile IMHO.
In practice I find using 4 sweeps of 4M samples not much different between my preferred 2 sweeps of 2M samples between 5Hz and 40KHz.
So for frequency response only I use a 1M sample sweep which takes about 12 seconds.
For distortion I use 2M x 2, particularly on a noisier night.
I've privileged to have my own permanently set up lab, indoors. I
t’s the size of a typical Australian suburban bedroom, which is about 10 m^2, climate controlled etc. with a noise floor of about <30dB(A).
Nevertheless, I need to find a sweet spot between my patience, and my (family's) sanity.
No one likes Brrrrrrrrrrrrreeeeeeep!! going on repeatedly, for up to minutes at a time…
Last edited:
This 10log is fixed, so is the length of measuring for a swepped sine wave as function of the sample rate and sequence length (which is connected to the FFT length).
So it doesn't really matter what software someone is using, the results will be more or less the same.
I think this is more like a visual or mathematical artifact?
But this text makes it sound it's a physical artifact?
Noise and (sometimes) distortion values always differ depending on how long the measurements are as well as what kind FFT window has been used.
Even more so with a live spectrum analyzer, each one has their tradeoffs.
If so, what do you think of :
xspk = original exponential sine sweep + 0.5 sec of silence
xmic = recording of the output
%%
rir1=impzest(xspk,xmic,'WarmupRuns',0);
xres=xmic-filter(rir1,1,xmic);
tres=xres(1:end-0.5*fs);
x1=xspk(1:end-0.5*fs);
x2=x1.*x1;
x2=x2-mean(x2);
x2=x2*std(x1)/std(x2);
x2=cat(1,x2,zeros(fs/2,1));
rir2=impzest(x2,tres,'WarmupRuns',0);
%%
The rir2 would have lots of noise before and after the true IR(second harm) which however can be windowed out. The FFT(windowed rir2) = matched-filter fit estimate for 2nd harmonic
The same can be repeated for Nth harmonic
Thanks, that's exactly the sort of practical information I was after.
Have you tried the stepped sine, or know how much time it takes?
Is it the size of typical Australian suburban bedroom because it's a repurposed typical Australian suburban bedroom?
By a remarkable coincidence I have a spare typical Australian suburban bedroom.
My point was that not all software uses exponential swept sine waves.
So while a 1M sweep takes ~12 seconds, and it's practical to do it 10 times, the stepped sine wave used in REW is probably too slow to do 10 times.
Older software often uses MLS and takes more time to achieve similar S/N as Farina sweep.
So maybe not practical to do 10 times, interested in people's experience.
I don't know, I'd have to see all the details, can you post the complete code, with more information on how the system is set up?
Best wishes
David
Last edited: