Calculating a voltage or sound pressure level to dBV or dB(SPL) is done with the 20*log(V/Vref) or 20*log(p/pref) formulas. As REW shows, -40dB equals 1% of the reference signal. And 1% HD is audible in certain conditions by most listeners. But not audible by default.
I want to clarify myself. IMD cannot be measured with REW sweep because by definition it requires playing multiple frequencies at once. So what I should say that if you already found that the cone breakup is causing major IMD with listening tests - because to listen is much easier than to run 500 tests permutating 5 different frequencies across the ranges - you can use the REW sweep to tune the crossover.
The real music with 10 instruments each playing 20 harmonics is passing 200 frequencies down to the amplifier. If the speaker is not capable of playing 200 different frequencies you will hear it.
Then taking above as the example, if you already know that the 3 kHz cone breakup resonance is causing IMD, you can use the simple REW sweep to test different crossovers. If I already measured that with main signal at 84 Db and HD3 distortion 42 Db at 1 kHz it is extremely audible, then I should tweak the XO until I put this 1 kHz peak from 42 Db to say 35 Db. I don't need to listen every iteration of the XO, I can just measure it.
Then when it's done I can start 4 hours of listening tests running all genres of music to verify I cannot hear it.
I hope this clarifies.
The real music with 10 instruments each playing 20 harmonics is passing 200 frequencies down to the amplifier. If the speaker is not capable of playing 200 different frequencies you will hear it.
Then taking above as the example, if you already know that the 3 kHz cone breakup resonance is causing IMD, you can use the simple REW sweep to test different crossovers. If I already measured that with main signal at 84 Db and HD3 distortion 42 Db at 1 kHz it is extremely audible, then I should tweak the XO until I put this 1 kHz peak from 42 Db to say 35 Db. I don't need to listen every iteration of the XO, I can just measure it.
Then when it's done I can start 4 hours of listening tests running all genres of music to verify I cannot hear it.
I hope this clarifies.
About music, this weekend I found that Tidal has 20 variants of Verdi's Messa da Requiem. It is a totally badass music with 50 instruments, 100 people choir and huge dynamics - so perfect to test the speakers. This amount of simultaneous sources is probably generating like 2,000 frequencies at once. And what I discovered is that CD resolution is just not enough to record that much data. I was listening Verdi for 5 hours and found how much the HD music variants are just superior to any of the CD recordings.
So don't forget, for a good listening tests you need only HD music, not CD.
So don't forget, for a good listening tests you need only HD music, not CD.
I am sorry, but there is no "audibility threshold" to either IMD or THD. It simply does not exist since neither of these measures is at all indicative of sound quality (i.e. there is no correlation between either one and perception.)
I don't see much that I can offer. That's because THD and IMD are both based on a false premise - that the ear processes nonlinearities the same way that a Fourier transform does, which is far from being true. Unless the "metric" has some foundation based on the way the ear actually works, no amount of tweaking can make it relevant.
Member
Joined 2003
Fwiw IMD is easily measured in REW using the signal generator and RTA. Select "show distortion" in RTA and save the measurement, then look at the distortion tab to view IMD result with fundamental removed.
The catch with IMD is that the frequencies used and separation and number of tones provides very different results, so the analysis can be a fair bit more complex than just run a sweep and see the result.
Better option may be to try the new FSAF measurement in the REW beta. It allows a measurement using any audio at all, pick your favourite stress tracks and view a total distortion result that includes harmonics, intermodulation, energy storage, everything, and you can listen to the residual audio to easily correlate the distortion result with actual listening experience. Only downside is that this measurement requires a good competent mic that provides low self-noise and low distortion and a very quiet recording space. Think more along the lines of creating a recording booth for speaker testing as any background noise and room reverb will be present in the residual audio.
The catch with IMD is that the frequencies used and separation and number of tones provides very different results, so the analysis can be a fair bit more complex than just run a sweep and see the result.
Better option may be to try the new FSAF measurement in the REW beta. It allows a measurement using any audio at all, pick your favourite stress tracks and view a total distortion result that includes harmonics, intermodulation, energy storage, everything, and you can listen to the residual audio to easily correlate the distortion result with actual listening experience. Only downside is that this measurement requires a good competent mic that provides low self-noise and low distortion and a very quiet recording space. Think more along the lines of creating a recording booth for speaker testing as any background noise and room reverb will be present in the residual audio.
FSAF thread here https://www.diyaudio.com/community/...sing-music-michael-tsiroulnikovs-fsaf.418843/
So far as I know there is no research for relevance of FSAF result/level versus audible distortion in music. It is just a way of making IMD separation residual audible, isolated from signal. Obviously this was just a sideline from his daytime job. https://www.researchgate.net/profile/Michael-Tsyrulnikov
As Geddes said, relevance of HD and IMD of loudspeakers is open too. Harmonic distortion audibility in music is verified to some extent but number of subjects (persons) tested are very small, not even near to scientific good practise. Sine wave distortion detection is easy to test, but speech or music...
https://www.researchgate.net/public...ibility_and_listener_preference_in_headphones
https://www.jstage.jst.go.jp/article/ast1980/11/1/11_1_29/_pdf
https://www.aes.org/technical/documentDownloads.cfm?docID=650
https://www.axiomaudio.com/blog/distortion
So far as I know there is no research for relevance of FSAF result/level versus audible distortion in music. It is just a way of making IMD separation residual audible, isolated from signal. Obviously this was just a sideline from his daytime job. https://www.researchgate.net/profile/Michael-Tsyrulnikov
As Geddes said, relevance of HD and IMD of loudspeakers is open too. Harmonic distortion audibility in music is verified to some extent but number of subjects (persons) tested are very small, not even near to scientific good practise. Sine wave distortion detection is easy to test, but speech or music...
https://www.researchgate.net/public...ibility_and_listener_preference_in_headphones
https://www.jstage.jst.go.jp/article/ast1980/11/1/11_1_29/_pdf
https://www.aes.org/technical/documentDownloads.cfm?docID=650
https://www.axiomaudio.com/blog/distortion
Last edited:
ps. Linear distortion ie. deviation from flat (or desired) spl response is quite easy to hear (2dB commonly stated). Cone resonance is mostly linear distortion. In regard of M/T crossover typically cone resonance will be more or less masked by tweeter´s higher spl at same frequency. And this is sometimes the most challenging task when designing 2-way speakers, perhaps more important than directivity match IMO.
Harmonic distortions can be heard in sine sweeps or constant sine at specific problematic fundamental F. Like previously stated human hearing is most sensitive around 1-5kHz, so when distortion procuct matches that region it is easy to hear even in music as eg. different timbre of a trumpet or violin playing a long note. Just like in androxylo´s case, but you must have a reference then because different instruments and musicians have different timbre. But I still believe that in his case it is more about linear distortion, so 3-4kHz resonance peak attenuation and lower/steeper xo should help.
In 2013 I attened a presentation by Wolfgang Klippel, he demonstrated audibility of THD by playing music clips which had added distortion throughout the spectrum. If I remember right I noticed 10%, most of audience lifted hands at 5%... The loudspeaker was a small 2-way monitor.
Another important aspect of (cone and other mechanical) resonances is decay of them, most often visualized as CSD. Differences in this are huge, but assessment and audibility tests are nonexistent.
HifiCompass is a wonderful resource for this info too.
https://hificompass.com/en/speakers/measurements/dayton-audio/dayton-audio-rs180-4
Harmonic distortions can be heard in sine sweeps or constant sine at specific problematic fundamental F. Like previously stated human hearing is most sensitive around 1-5kHz, so when distortion procuct matches that region it is easy to hear even in music as eg. different timbre of a trumpet or violin playing a long note. Just like in androxylo´s case, but you must have a reference then because different instruments and musicians have different timbre. But I still believe that in his case it is more about linear distortion, so 3-4kHz resonance peak attenuation and lower/steeper xo should help.
In 2013 I attened a presentation by Wolfgang Klippel, he demonstrated audibility of THD by playing music clips which had added distortion throughout the spectrum. If I remember right I noticed 10%, most of audience lifted hands at 5%... The loudspeaker was a small 2-way monitor.
Another important aspect of (cone and other mechanical) resonances is decay of them, most often visualized as CSD. Differences in this are huge, but assessment and audibility tests are nonexistent.
HifiCompass is a wonderful resource for this info too.
https://hificompass.com/en/speakers/measurements/dayton-audio/dayton-audio-rs180-4
Last edited:
The REW docs says that to use this new FSAF requires a very good microphone, like much more than $100.
This guy had put his $80 microphone to test and discovered it's incapable of measuring properly:
As I am not a speaker builder, I am building one speaker for myself, I am definitely not throwing money on a mic. So for me only the listening tests remain.
I cannot detect the level of distortion from a short clip as you had to do above, my listening tests require at least 30 minutes of music, and then I would listen various genres for another 3 hours. It requires the brain to be in a calm and concentrated state. I am talking about good XO when distortion is already low, not the obvious case.
I hope that people will pickup this new feature soon and start a wave of new reporting.
This guy had put his $80 microphone to test and discovered it's incapable of measuring properly:
I cannot detect the level of distortion from a short clip as you had to do above, my listening tests require at least 30 minutes of music, and then I would listen various genres for another 3 hours. It requires the brain to be in a calm and concentrated state. I am talking about good XO when distortion is already low, not the obvious case.
I hope that people will pickup this new feature soon and start a wave of new reporting.
Probably because within reason, severe concentrated resonances like in the RS180-4 mentioned aren't very important because the driver will never be listened to (except by a handful of extremists) as-is, so it's the filtered performance that counts.
It's difficult to link properly to John's site when something is buried within a wider page, so quoting, with clear reference to John Krutke as the author (www.zaphaudio.com), without any selective editing, & using the examples he provided in order:
[These are the labelled 01 & 02 images attached]
[These are the labelled 03 & 04 images attached]
As far as linear distortion goes, I basically agree with John on this (not that my opinion is worth much / anything). There may be [harmonic] distortion amplification lower down in the range, most obviously from the rigid cone's main breakup mode / modes & unless that's addressed with a high impedance notch in series with the driver, it's going to remain -whether it's an issue though will likely depend on its level relative to the fundamental & where you're crossing relative to those distortion spikes -if it's below the HD3 amplification in particular you're probably okay. Above -you might be. Possibly. Perhaps. Maybe. YMMV (etc.). 😉
Attachments
As you say filtering remedies, but it's not all filters, only some. Let's simplify cone breakup as a peak in on axis response. The breakup peak can be thought of as an amplifier in acoustic domain which happens no matter what the filtering is. If you measured voice coil current it would likely be ~flat over that frequency band, but the acoustic output shows a peak. This resonance peak, or amplification, would be removed only by physically modifying the driver not to resonate, so the filtes don't actually do nothing to the resonance itself they can only precondition signal going to the driver, to be amplified by the resonance. So, if your DSP PEQ fixes the measured peak in acoustic domain, current through voice coil would show a dip there.
So, what's the difference between filters? If one puts a DSP or active filter before a power amp and it would lowpass, or PEQ, the main signal but all distortion products and noise generated in the amplifier or in the transducer are not attenuated at all so get amplified by the breakup peak. So, if you think how it sounds, you've low passed the music for example at 1kHz and breakup peak is at 3kHz boosting all the distortion/noise landing on it making say a 10db peak kind of stick out, highlight, as there is no music masking it. Music would be on the tweeter, but tweeter is not at same location and the breakup peak likely is very beaming, near on-axis only. Hear it? possibly, at least thought this way if any distortion or noise is audible it's here on the cone resonance peak because it is a resonance and boosts any distortion there is in the current.
However, if you do filtering passively with a series filter, which also EQ:s any distortion current produced in the amp or by the driver motor, the distortion products or noise would not stick out like that but be low passed / treated as well just like the music is. It should be less audible, if it was audible in the first place. So only one type of filter can do this, passive filter in series with the driver, that makes high impedance in series with driver at this brekaup frequency.
For example typically used 2nd order low pass filter would not work very well for this because it has low impedance path through the cap for driver distortion current to flow, but 1st or 3rd order would be more effective because there is an inductor in series with the driver. Or if you used a parallel notch in series with the driver, would work well. Series filter in parallel with driver would not work, although it could give same attenuation in acoustic measurement it provides low impedance in series with the driver allowing driver distortion current to flow and get amplified by the cone resonance.
This stuff is already written million times in various forms, but hopefully it helps anyone think it through, who already haven't.
So, what's the difference between filters? If one puts a DSP or active filter before a power amp and it would lowpass, or PEQ, the main signal but all distortion products and noise generated in the amplifier or in the transducer are not attenuated at all so get amplified by the breakup peak. So, if you think how it sounds, you've low passed the music for example at 1kHz and breakup peak is at 3kHz boosting all the distortion/noise landing on it making say a 10db peak kind of stick out, highlight, as there is no music masking it. Music would be on the tweeter, but tweeter is not at same location and the breakup peak likely is very beaming, near on-axis only. Hear it? possibly, at least thought this way if any distortion or noise is audible it's here on the cone resonance peak because it is a resonance and boosts any distortion there is in the current.
However, if you do filtering passively with a series filter, which also EQ:s any distortion current produced in the amp or by the driver motor, the distortion products or noise would not stick out like that but be low passed / treated as well just like the music is. It should be less audible, if it was audible in the first place. So only one type of filter can do this, passive filter in series with the driver, that makes high impedance in series with driver at this brekaup frequency.
For example typically used 2nd order low pass filter would not work very well for this because it has low impedance path through the cap for driver distortion current to flow, but 1st or 3rd order would be more effective because there is an inductor in series with the driver. Or if you used a parallel notch in series with the driver, would work well. Series filter in parallel with driver would not work, although it could give same attenuation in acoustic measurement it provides low impedance in series with the driver allowing driver distortion current to flow and get amplified by the cone resonance.
This stuff is already written million times in various forms, but hopefully it helps anyone think it through, who already haven't.
Last edited:
Wait, I am confused. If the resonance happens in the middle of the cone intended frequency indeed any filter will still be sending the signal there so the distortion cannot be removed.
But we are talking about cone breakup in this thread, the distortion that is supposed to be suppressed by moving the crossover frequency lower below the area the cone is not capable of reproducing. The same at the low range, 5 inch drivers cannot play below 200 Hz and must be high passed.
The task is just to properly find where to cut, exactly, the absolute amount of the distortion is not even important here, only its location.
But we are talking about cone breakup in this thread, the distortion that is supposed to be suppressed by moving the crossover frequency lower below the area the cone is not capable of reproducing. The same at the low range, 5 inch drivers cannot play below 200 Hz and must be high passed.
The task is just to properly find where to cut, exactly, the absolute amount of the distortion is not even important here, only its location.
Yeah it depends on type of filter 😀 Small thought experiement: take very steep 96db/oct low pass with DSP at 1kHz, to make sure that no signal is output beyond that and there should not be any sound at the breakup frequency around 3kHz, right? So a cone resonance at 3kHz would get no sound to it to boost? Wrong, the amplifier adds some noise, and the driver likely adds harmonics due to electrical parameters vary with excursion, 3rd harmonic among others, especially if there is any low frequency sounds making good amount of excursion which basically makes the whole passband to distort. So sounds on the passband generate harmonics beyond passband. 3rd harmonic generated by say Le(x) at 1kHz sound would land smack on the breakup peak and happily output and get amplified.
DSP low pass before amplifier ( and driver ) does not low pass harmonics generated in the amp or driver because it's before them. What you need is low pass (or notch filter) that is in series with the driver reducing current through voice coil at the breakup frequency. So, a passive low pass filter would low pass the harmonics as well, as long as it increases series impedance for the driver
DSP low pass before amplifier ( and driver ) does not low pass harmonics generated in the amp or driver because it's before them. What you need is low pass (or notch filter) that is in series with the driver reducing current through voice coil at the breakup frequency. So, a passive low pass filter would low pass the harmonics as well, as long as it increases series impedance for the driver
Last edited:
Member
Joined 2003
Why rely on research of others when you can listen for yourself and draw your own conclusions. It is "just a way" to separate all non-intended audio from the original signal in a recording, and freely available for public use. It can make for easily identifying those small nuances that people have been picking apart for years in audio, as well as providing audible correlation to what the charts and graphs show to what you hear. Other benefits include the ability to generate accurate frequency response transfer function from any audio, if you are tired of hearing that annoying sweep sound. So you can test with your favourite song and listen to the residual audio, or pink noise,, or M-Noise, or a sinne sweep, or multitone signal, whatever you like. For myself, evaluation of speaker performance can be achieved with real audio with realistic crest factor has proved invaluable. One major downside of sine sweep is that crest factor is much different than real audio, requiring testing and multiple levels to provide any real depiction of reality (often 86dB and 96dB/1m SPL are used), and a common misunderstanding that a sine sweep at 86dB is representative of speaker performance at 86dB with real audio, which is not true due to that crest factor thing.
I've found FSAF to be an incredibly useful tool in my own research of driver performance as well and design philosophy. It has highlighted the quantity of modulated distortion products in bass that even high end drivers produce, as well as allowed me to better analyse performance benefits of various crossover circuits. Of course, it can be difficult for productive discussions when Mike has been silenced from this forum, so progress has already been stifled here. You'll have to go to the REW support forum to chat with him.
To ber honest that FSAF thread you link to is a bit of a mess, derailed by people who just want to talk theory, and no one seemed willing to run any real measurements and share their results. I hope that more users will take the plunge in the future.
I hope this isn't intended to convey threshold of audibility of distortion in any way. With such insight we must be perfectly well safisfied with little bluetooth boomboxes.
To be honest that is a reality of good recording requiring good equipment. If we are spending thousands on drivers and woodworking tools to design and build loudspeakers as a hobby, why evaluate them with the cheapest measurement mics available? Tiny little condensers are going to run into problems at high SPL, just like a tiny speaker driver will, and the reality of measuring with real world crest factor is that you will encounter peaks in audio that are a fair bit higher than the RMS level, and you'll want to be certain that the distortion encountered originates from the equipment being tested, not from the test equipment itself. There are some fairly low cost options that provide excellent results. For example I am using a Line Audio Omni1 which performs extremely well for its cost, I believe its one of the best value mics on the market today.
This is a very valuable input for a noob like me, thank you. I do plan to slap a 14 gauge 2 mH inductor in line with the driver and nothing else on the signal path. All the filters I come with will be parallel with the driver. According to what you just said this is the way to go, and I have zero experience to come with something like this on my own.
Hi, yes it's not the most important thing on a system, it's more important to get the transfer function right, and especially the whole system concept right, what ever these are. If you can take this stuff into account, in addition, it's great. Few passive parts with active system can lake at least measurements look better, but audibility, not sure, depends on if the system is oversized or undersized, pushed to limits.