A topic comming up at a different discussion about driver measurements. So I sacrificed my lunch break and did some measurements to show you.
In short - it's always important to know your Signal to Noise range, especially when measureing THD. Do I get results or am I looking at noise is one of the main questions in acoustic measurements when you get deeper into details.
A quick and simple way to check is to unplug the speaker and run a measurement - you will measure the actual noise inflicting your measurement.
There is the noise behaviour of the measurement microphone, surrounding noise but also the analysis methode! And the differences can be huge! Let me show you here.
I measured the fan of my APX515 system in 5cm distance - what a great noise source 🤓
Stepped frequencylevel sweep. An old but standard measurement signal and analysis. 31 dedicated sine frequencies and you measure the level.
So you measure the broadband noise of the fan at every frequency - pretty bad.
When you do a bandpass stepped sweep the analyser sets a narrow bandpass arround the measurement frequency to eliminate noise of different frequencies. Helps.
There is quite some fluctuation of the signal - it's noise ... But we see the fan produces mainly low frequency noise, you get a good S/N range at high frequencies!
Now a more modern signal - continuous sweep. It's the analysis with FFTs similar to what you use with MLSS and it's implemented very good here (not every software is the same accurate!)
This is the reason I use 1/24th smoothing with these measurements - the "hair" doesn't give information for this measurement.
As we have WAY more resolution we see spikes in the response. But we lack resolution at low frequencies - and have a huge spread at 20Hz. Even the stepped level sweep gives more stable results here.
Increasing resolution and sweep time to 2s gives this graph:
Now we clearly see our spikes - typical for fan noise. And we have enough information at low frequencies to get a more stable result. This was not the behaviour of the noise source before - it's a problem of the too short sequence and FFT calculation! And still at 20Hz we get more derivation as we want, from 30Hz upwards everything is perfect.
When you use small membrane (1/4") or cheap measurement microphones you get A LOT of noise at low frequencies and it will influence your results and dynamic range of your measurement significant. Choose the right measurement methode and check it!
I hope this gives more light to the toppic of noise sensitivity with acoustic measurements and different measurement algorithms as it brings new questions 🤓
In short - it's always important to know your Signal to Noise range, especially when measureing THD. Do I get results or am I looking at noise is one of the main questions in acoustic measurements when you get deeper into details.
A quick and simple way to check is to unplug the speaker and run a measurement - you will measure the actual noise inflicting your measurement.
There is the noise behaviour of the measurement microphone, surrounding noise but also the analysis methode! And the differences can be huge! Let me show you here.
I measured the fan of my APX515 system in 5cm distance - what a great noise source 🤓
Stepped frequency
So you measure the broadband noise of the fan at every frequency - pretty bad.
When you do a bandpass stepped sweep the analyser sets a narrow bandpass arround the measurement frequency to eliminate noise of different frequencies. Helps.
There is quite some fluctuation of the signal - it's noise ... But we see the fan produces mainly low frequency noise, you get a good S/N range at high frequencies!
Now a more modern signal - continuous sweep. It's the analysis with FFTs similar to what you use with MLSS and it's implemented very good here (not every software is the same accurate!)
This is the reason I use 1/24th smoothing with these measurements - the "hair" doesn't give information for this measurement.
As we have WAY more resolution we see spikes in the response. But we lack resolution at low frequencies - and have a huge spread at 20Hz. Even the stepped level sweep gives more stable results here.
Increasing resolution and sweep time to 2s gives this graph:
Now we clearly see our spikes - typical for fan noise. And we have enough information at low frequencies to get a more stable result. This was not the behaviour of the noise source before - it's a problem of the too short sequence and FFT calculation! And still at 20Hz we get more derivation as we want, from 30Hz upwards everything is perfect.
When you use small membrane (1/4") or cheap measurement microphones you get A LOT of noise at low frequencies and it will influence your results and dynamic range of your measurement significant. Choose the right measurement methode and check it!
I hope this gives more light to the toppic of noise sensitivity with acoustic measurements and different measurement algorithms as it brings new questions 🤓
Last edited:
Not sure, but it looks like you (or the software...) applied some sort of deconvolution to the noise recording, and then Fourier transformed it to the frequency domain. That's not what I had in mind when I thought about analysing the background noise recorded with the speaker unplugged.
First of all, if you record the background noise with the loudspeaker unplugged, the result simply CANNOT depend on the test signal, because the speaker did not produce any output at all.
The recording of the background noise is done in the time domain. If you want to look at noise in the frequency domain, you can apply a Fourier transform. So far, so good. However, it seems the time-domain noise data were filtered or deconvolved with the hypothetical test signal. Of course the result will depend on the filtering or the hypothetical test signal. However, I don't understand why such filtering or deconvolution might be suitable to analyze the background noise -- as I wrote before: the noise is not related to the test signal in any way!
First of all, if you record the background noise with the loudspeaker unplugged, the result simply CANNOT depend on the test signal, because the speaker did not produce any output at all.
The recording of the background noise is done in the time domain. If you want to look at noise in the frequency domain, you can apply a Fourier transform. So far, so good. However, it seems the time-domain noise data were filtered or deconvolved with the hypothetical test signal. Of course the result will depend on the filtering or the hypothetical test signal. However, I don't understand why such filtering or deconvolution might be suitable to analyze the background noise -- as I wrote before: the noise is not related to the test signal in any way!
This is a function of the capacitance of the membrane construction and the input impedance of the mic amplifier. The cap builds a filter for the input noise and that works better when the capsule capacitance is high.
Large membrane -> large cap -> lower noise from the mic when the electronics is good.
Hi IamJF,
Even back in the 1980's, the HVAC in the shop would create enough noise to mess up acoustic measurements. I had to turn the system off and do these things at night. Oscillator and meters (HP), logging things on paper. I had a PC I used, but ADCs weren't good enough yet. When you do things with your own hand, you understand exactly what you are measuring. You even had to mess around with your physical setup to make sure you weren't getting systemic aberrations from that.
Exactly! Cheap mics with built in preamps will not perform well enough. So larger diaphragm capsules coupled with good electronics (and calibrate that combination) means you have a decent chance of meaningful measurements. Then of course, configuring your software analysis correctly followed by any post processing.
Even back in the 1980's, the HVAC in the shop would create enough noise to mess up acoustic measurements. I had to turn the system off and do these things at night. Oscillator and meters (HP), logging things on paper. I had a PC I used, but ADCs weren't good enough yet. When you do things with your own hand, you understand exactly what you are measuring. You even had to mess around with your physical setup to make sure you weren't getting systemic aberrations from that.
Hi Bernd,
A sweep is when you take a tone starting at usually a lower frequency and vary it to a higher frequency continuously. You can do it by hand, in the old days it was a ramp (linear or log) feeding a voltage controlled oscillator. Today it is done digitally. The results were displayed either on a chart recorder, or in a video display (today) and the file recorded in a database format. A discrete list of amplitude vs frequency.
A sweep is when you take a tone starting at usually a lower frequency and vary it to a higher frequency continuously. You can do it by hand, in the old days it was a ramp (linear or log) feeding a voltage controlled oscillator. Today it is done digitally. The results were displayed either on a chart recorder, or in a video display (today) and the file recorded in a database format. A discrete list of amplitude vs frequency.
I've heard good things about cheap small Panasonic condenser mikes, but they are not good enough for measurements?
Cheers!
Cheers!
If they were, everyone would be using them.
Transducers are generally not flat in response even if they are expensive. Cheaper units (okay, "less expensive") are not produced with reliability or high quality in mind. Would we all like cheap test equipment? Absolutely!!!
Yet here I am with a bench full of HP / Agilent / Keysight equipment. It's the more expensive stuff and if I could get by with Leader, B&K or other cheaper brands -I would in a heartbeat! We all want the easy / cheap way out. Sometimes that simply will not give us the results we hope for (operational word here is "hope").
Transducers are generally not flat in response even if they are expensive. Cheaper units (okay, "less expensive") are not produced with reliability or high quality in mind. Would we all like cheap test equipment? Absolutely!!!
Yet here I am with a bench full of HP / Agilent / Keysight equipment. It's the more expensive stuff and if I could get by with Leader, B&K or other cheaper brands -I would in a heartbeat! We all want the easy / cheap way out. Sometimes that simply will not give us the results we hope for (operational word here is "hope").
Interesting.......but hardly anything more than random waxings of observations. We quite often check our mics (very expensive to cheap) and compare the FR in our iso booths and there's hardly any deviation in recorded FR except for below 30hz or above 17k or so....neither of which we care about. My $89 Sonarworks hangs with the best of em where accuracy is concerned......when accuracy is desired.
In all fairness, we're using a Focusrite ISA one for this kinda thing where the preamp is as neutral as possible and the phantom power is pure and not from a switch based power supply.
In all fairness, we're using a Focusrite ISA one for this kinda thing where the preamp is as neutral as possible and the phantom power is pure and not from a switch based power supply.
https://www.deepl.com/en/translator#en/de/sweep signal
By the way, REW help file gives excellent info of how REW analyzes to show noise. Long sine sweep is necessary, and distortion analysis is based on stepped signals. Doing multiple sweeps is possible, but I don't know if it helps for distortion analysis. I use only REW and Umik-1 v2 usb-mic, single sweeps...
Last edited:
Hi mayhem13,
The preamp simply must be flat with little noise. Same requirements for any signal amplifier. As for accuracy, the definition depends on your limits and confidence level. If your mic / amp combination works for you, then great. It may or may not for others. Since we are testing speakers here I guess it depends on what we are testing for and to what limits. In your case, you don't care about higher frequencies. Lower frequency measurements are difficult. So, it's all good.
But don't discount the experience and requirements of others.
No, sorry. Experience, hard won. I also put my money where my mouth is.
The preamp simply must be flat with little noise. Same requirements for any signal amplifier. As for accuracy, the definition depends on your limits and confidence level. If your mic / amp combination works for you, then great. It may or may not for others. Since we are testing speakers here I guess it depends on what we are testing for and to what limits. In your case, you don't care about higher frequencies. Lower frequency measurements are difficult. So, it's all good.
But don't discount the experience and requirements of others.
By the way German word for distortion is Klirr - sounds just like it!
Hörnli et al. - Sengpielaudio has German and English info about electronics and audio, specially about microphones and measurements!
https://sengpielaudio.com/index.html
https://sengpielaudio.com/Calculations03.htm
Hörnli et al. - Sengpielaudio has German and English info about electronics and audio, specially about microphones and measurements!
https://sengpielaudio.com/index.html
https://sengpielaudio.com/Calculations03.htm
I see.
The original post that triggered the whole discussion was:
To me, "checking the noisefloor" seemed like one would record the noise to check how much there is and what it looks like. I don't see the point of applying filters or of deconvolving it with the (inexistent) test signal.
To me it seems quite (or even more) revealing what the measurement filtering does with the noise and what is being displayed.
That is the first time I read someone writing B&K is cheap. Gras, Microtech Gefell and Aco Pacific probably are too.
it’s all relative.
You can get an automotive stethoscope
A nursing stethoscope
A doctor’s stethoscope
A cardiologist’s stethoscope
A digital stethoscope
Price difference can be a factor of 100
Can you listen to your cabinet vibrations with any/all of these?
Sure.
But a lot is about the build materials, and designed for durability, reliability, repeatability, ability to be have replacement parts etc. But sometimes it about market position too.
You can get an automotive stethoscope
A nursing stethoscope
A doctor’s stethoscope
A cardiologist’s stethoscope
A digital stethoscope
Price difference can be a factor of 100
Can you listen to your cabinet vibrations with any/all of these?
Sure.
But a lot is about the build materials, and designed for durability, reliability, repeatability, ability to be have replacement parts etc. But sometimes it about market position too.
Last edited:
B&K (sometimes B+K) does not mean Bruel and Kjaer in this context, different company.
I hoped it was You haven't mentioned the alternative of multiple sweeps and then process to statistically reduce noise.
that I asked @IamJF
Best wishes
David
Last edited:
MLS is usually Maximum Length Sequence and MLSSA is an MLS based Software Analyser or some similar acronym.
I am not sure which one you mean by MLSS, if either?
In either case it's not a continuous sweep so I don't understand your comment... is it a typo?
Best wishes
David