Group of 4 inch driver performance evaluation: an IMD perspective

Sorry I don't understand what you mean.

Linear, mathematical summing of 2 frequencies will not have voice tone amplitude differences between the crest and trough part of the bass tone. They will be equal, hence 100% amplitude stability.

But I admit amplitude stability is a random term came up in my mind, nothing academic. IMD2 side band seems somewhat correlated, but I feel IMD2 is not very reliable, for example 830854 have almost no IMD2 side band on the left, yet "amplitude stability" is still easy to derive.
If you present two tones different frequency, with equal amplitude, you will have amplitude modulation.

Beating as it is known. The beat frequency, I.e. amplitude modulation frequency is a simple calculation. This will occur in purely electrical domain too...

Make a potential divider, centre tap to oscilliscope; inject 50Hz in one tap, 51Hz in the other tap. What do you observe? Amplitude modulation.

I'm not sure how intelligent folk, do not know this.

Ok, more relevant, 4:1 ratio will reduce this, but it will still exist, whether or not it fulfills the mind numbingly tight definition of IMD, it is still AM caused by IM. The D part is pretty moot.

Amplitude modulation is pretty audible.

Next expert please.
 
If you present two tones different frequency, with equal amplitude, you will have amplitude modulation.

Beating as it is known. The beat frequency, I.e. amplitude modulation frequency is a simple calculation. This will occur in purely electrical domain too...

Make a potential divider, centre tap to oscilliscope; inject 50Hz in one tap, 51Hz in the other tap. What do you observe? Amplitude modulation.

I'm not sure how intelligent folk, do not know this.

Ok, more relevant, 4:1 ratio will reduce this, but it will still exist, whether or not it fulfills the mind numbingly tight definition of IMD, it is still AM caused by IM. The D part is pretty moot.

Amplitude modulation is pretty audible.

Next expert please.
You misunderstood the technique. There's no beating here.

50hz + 51hz will have beating but it's a different problem, because they are very close in frequency and is not a frequency of Nx multiply of each other.

I use 80+680hz because they are 8.5x frequency of each other. Same as 30+255hz. They are not integer multiply, because we avoid bass tone's harmonics overlapped with voice tone.

In case you are still curious, here's the electrical input for 80+680hz 1:1. Do you observe any beating pattern?
1660746137249.png


Wikipedia please:
In acoustics, a beat is an interference pattern between two sounds of slightly different frequencies.
 
Last edited:
  • Thank You
Reactions: tktran303
we do the 30Hz+255Hz IMD test to reveal position dependent amplitude modulation effects in both the motor and soft parts, ie mainly Bl(x) and Sd(x). The 30Hz is a bit arbitrary, could be lower as well. The main point is that the deep tone controls the excuriosn and then we have the 255Hz voice tone superimposed which has low excursion (excursion drops with frequency squared when abbove fs). The test is heavily inspired by HifiCompass. We can also do multitone analysis but the results can be quite hard to interpret. For the two tone 30Hz+ voice tone we can check how close we are at the Doppler limit (the intermodulation caused by the motion of the mebrane).

We have a Klippel but developed our own analyser that is more precise and gives more information in order to separate the distortion sources (eg. the 'Cunningham' distortion aka flux or reluctance modulation)

Cheers

Lars (co-founder at Purifi)
I'd love to see multitone distortion provided in the datasheet. I think that's the ultimate test for a finished speaker system. For driver datasheet I believe there're many challenges and needs a lot of educations to the public. For example.. What bandwidth has what kind of effects? What kind of sidebands are doppler and what kind of sideband are not? Shaped envelope like EIA-426B or equal loudness? ETC.

But Purifi has done it before, the super comprehensive datasheets, maybe it's time to raise it again👍 😉
 
thanks, I agree with your points. The education part is essential. We have a long overdue task in doing a paper explaining the data sheet and how various distortion mechanisms are expected to show up etc. We also have versions of the two tone IMD plot where the side bands are decomposed in to their AM and FM components. it is a bit of a balance between providing data and not confuse/overload people. I like the 30Hz plus voice tone since it reveals the main excursion driven amplitude modulation mechanisms (Bl(x), Sd(x), Bl(i), L(x)). Ideally the amplitude of the bass tone should be swept to reveal when the linear range is exceeded. The multi tone plots tend to be difficult to interpret since it mixes so many tones with multiple nonlinear mechanisms. We also consider to play back music and remove the linear responses to distill the distortion and post the audio files.
 
‘’π÷9 = 0.349
this entry point promotes Helmholtz resonance. having the driver entry perfectly at this point facilitates the harmonics with double and triple the wavelength of the fundamental, especially when the TL has 2 folds (3 sections), which causes reinforcement of the fundamental through harmonic subtraction. the reinforcement is perceived as the difference between the double and triple wavelength and thus our ear hears an additional 'ghost' fundamental, or rather a boost of that frequency. the frequencies in the box are not 'fighting' each other but rather coalescing on every node in the TL. it's like the Tacoma Narrows bridge, resonance became so strong the entire structure started waving. this is ideally what one would want, smaller-scale, in a TL’’.
 
  • Like
Reactions: camplo
thanks, I agree with your points. The education part is essential. We have a long overdue task in doing a paper explaining the data sheet and how various distortion mechanisms are expected to show up etc. We also have versions of the two tone IMD plot where the side bands are decomposed in to their AM and FM components. it is a bit of a balance between providing data and not confuse/overload people. I like the 30Hz plus voice tone since it reveals the main excursion driven amplitude modulation mechanisms (Bl(x), Sd(x), Bl(i), L(x)). Ideally the amplitude of the bass tone should be swept to reveal when the linear range is exceeded. The multi tone plots tend to be difficult to interpret since it mixes so many tones with multiple nonlinear mechanisms. We also consider to play back music and remove the linear responses to distill the distortion and post the audio files.
Thanks Lars, that's a lot of cool ideas!
Distortion-only audio is especially interesting.

Yeah multi-tone could be difficult to understand(I don't). Erin did some great work in demonstrating the effect, but still maybe it's a bit too far. And one may came up with a conclusion that "why not build a 10 way speaker with cheap drivers only" which kind of defeat the purpose of great driver engineering.
 
  • Like
Reactions: lrisbo
Until I'm interested to do a 30hz+255hz(4:1) test my self. Test condition is not ideal but made as fair as possible. I test free raw driver with nearfield MIC, same distance for both driver, with about 3.5mm~4mm one way excursion on SB15 and match the bass tone(not voice tone) on 15WU
I think it would be interesting to see what would happen if you tested the following dedicated midranges:
Scan Speak 15M/4531K - 3mm xmax
SB Satori Satori MR13P-4 - 3mm xmax
Scan Speak 12MU/4731T-00 - 3.5mm xmax
Purifi PTT6.5M04-NFA-01 - 3.9 mm xmax

By 2.8mm, the larger SB17 already tightens compliance wise 25%, causing 10% distortion. The SB15CAC30 should be really viewed more like a midrange and not a midwoofer. It's a midwoofer only based on xmech imo, since the suspension seems to make all the right tradeoffs for midrange use. Granted 4mm of excursion is within what the manufacturer describes as linear, but the Klippel will show it's quite nonlinear by that point. I think you give up 1db in air moved in low octaves compared to Scan Speak Illuminator due to linear and xmech limitations but the midrange performance should be cleaner using an appropriate high pass filter.

If the test as is setup will prove all the midrange drivers to be underperforming compared to woofer versions by the same manufacturer, does that mean that midranges are always inferior to woofers? Or maybe that this test necessarily tests for a good "woofer" or "midwoofer?"

 
  • Like
Reactions: Arez
How applicable is this testing if I aim to keep excursion on my drivers to less than 1 mm?

I guess the IMD effect will be minimal. You are basically testing distortion without excursion effects (LE(i) instead of LE(x), for example).

By 2.8mm, the larger SB17 already tightens compliance wise 25%, causing 10% distortion. The SB15CAC30 should be really viewed more like a midrange and not a midwoofer.

I'm not entirely sure of that. Yes, suspension causes bass HD (which is already show at spikes of 60hz and 90hz as HD2 and HD3 in the case of 30hz+255hz), but the idea behind IMD test is to isolate distortion caused by CMS(x) and BL(x) (also probably LE(x)). Because suspension nonlinearities only affect low frequencies.

https://www.klippel.de/fileadmin/_migrated/content_uploads/Loudspeaker_Nonlinearities–Causes_Parameters_Symptoms_01.pdf
klippel.png



As a proof, You can see some driver with poor suspension perform well in the IMD test, most notably W4-1720 with very bad suspension (I suspect ~2mm of linear range) but very long magnet gap.

It's a midwoofer only based on xmech imo, since the suspension seems to make all the right tradeoffs for midrange use. Granted 4mm of excursion is within what the manufacturer describes as linear

No, SB15 is clearly designed as a midwoofer. The xmax is determined by magnet system, and the 35hz FS is clearly in midwoofer territory. There's no point to design a driver with 5mm overhung but only intended to use 1mm. it's waste lots of sensitivity(which is very important for a dedicated midrange)

If the test as is setup will prove all the midrange drivers to be underperforming compared to woofer versions by the same manufacturer, does that mean that midranges are always inferior to woofers? Or maybe that this test necessarily tests for a good "woofer" or "midwoofer?"

? This does not make sense to me. There's no midrange driver here.
 
Last edited:
I'm not entirely sure of that. Yes, suspension causes bass HD (which is already show at spikes of 60hz and 90hz as HD2 and HD3 in the case of 30hz+255hz), but the idea behind IMD test is to isolate distortion caused by CMS(x) and BL(x) (also probably LE(x)). Because suspension nonlinearities only affect low frequencies.

https://www.klippel.de/fileadmin/_migrated/content_uploads/Loudspeaker_Nonlinearities–Causes_Parameters_Symptoms_01.pdf
View attachment 1220016


As a proof, You can see some driver with poor suspension perform well in the IMD test, most notably W4-1720 with very bad suspension (I suspect ~2mm of linear range) but very long magnet gap.



No, SB15 is clearly designed as a midwoofer. The xmax is determined by magnet system, and the 35hz FS is clearly in midwoofer territory. There's no point to design a driver with 5mm overhung but only intended to use 1mm. it's waste lots of sensitivity(which is very important for a dedicated midrange)



? This does not make sense to me. There's no midrange driver here.
Agreed, I picked the worst of the parameters to make a point here. Kms(x) is not the main culprit. Also agree that none of midrange drivers were tested here. The SB15 may be marketed as a midwoofer, and I'm sure it would hold its own against other midwoofers. But given the way it tests I would use it as a midrange only. I think the low FS and longer coil are nice in a bind but not sufficient when Purifi and Scan Speak make proper midwoofers.

If I can reframe. Let's look at BL(x). The 15WU has substantially more linear throw than the SB15, i.e. 8.7mm at 82% threshold vs <5.1mm (SB17 was tested at 5.1mm and has a longer coil than SB15). Since the 30hz tone pumps the driver to 4mm excursion we should see some BL(X) effects already, maybe even 10% loss in motor force for the SB.

Purifi, Scan, and SB make some drivers in both midwoofer and midrange applications. The midranges I listed all shorten the coil in reference to the midwoofer version, hence BL(x) will be substantially neutered. In fact, any of the midranges I listed will likely have worse BL(x) than SB15.

Do you expect this IMD test to show lower distortion for midranges compared to woofers in the same line of products? Even outside the same line of products, if you tested Scan Speak 15M/4531K - 3mm xmax, would you expected it to best the 15WU in this test? If the midranges generally show worse distortion, then how do we interpret the results?
 
@cvjoint

Yes I think midrange will test worse in IMD compared to Mid-Woofer.

But that's the exact purpose of a mid-woofer (or full-ranges): Working on a bass and mid at the same time. Hence IMD is more important for mid-woofer, while HD/sensitivity is probably enough to judge mid-range driver objectively(which I think SB15/SB17 could do well and I have a few pairs of them).

I just find it generally lacking IMD information on the internet, and try to find it out myself.

BTW the SB15 and SB17 range are still used as mid-woofer in many successful DIY build and commercial products. I even saw SB15NAC used as a single driver full-range in a commercial product covering 50-20000hz.
 
@racingpht The tradeoffs we make for higher sensitivity right! Better make sure not to drive pure midranges into high excurion.

We're lucky to have Scan Speak sell to diy. I've found their 10" Revelator as a perfect midwoofer, can do high cross points and low cross points for very simple effective projects. Before that I've always done nth degree way projects to reduce IMD. The Illuminator line is just bliss, taking it to extremes.

I'm totally split on whether to use 18wu or SB17 for a line array. Based on what you showed here, I'd say it's 60hz HP for 18wu or 100hz HP for SB17. Is it worth 3x cost? This kind of testing goes a long way. IMD is not an entry level concept, harder to grasp than HD, but certainly very important. Worth the sweat.

I've used the Satori line full range for surround sound, works reasonably well. However, I'm sure it would have sounded better as a 3way implementation but I got lazy and ran out of room.

A natural progression here is to see how things would look like at less excursion, to find the sweetspot where the SB just matches the Scan in IMD.
 
One variation that comes to mind could be a multi-tone test with 2 very closely spaced frequencies + a 3rd F, so that it's like a 2-tone test with swept amplitude.
Eg:
30 Hz
31 Hz
255 Hz
So effectively you get 30.5 Hz with a slow tremolo (assuming equal amplitude for 30 + 31, so there's maximum dynamic range), and then observe how the IMD changes at varying levels of disturbance.

It could be that the different drivers have different IMD curves relating to Xmax.

I'm not sure how such a test could be executed off the top of my head. I'd probably do a sliding FFT window by hand-selecting different parts of the recording, until I find a more automated way of doing it.
 
@racingpht The tradeoffs we make for higher sensitivity right! Better make sure not to drive pure midranges into high excurion.

A natural progression here is to see how things would look like at less excursion, to find the sweetspot where the SB just matches the Scan in IMD.
There's such info on Erin's website, for example:

https://www.erinsaudiocorner.com/driveunits/sbacoustics_sb17cac35-4/
https://www.erinsaudiocorner.com/im...7CAC35-4/SB Acoustics SB17CAC35-4_IMD_HPF.gif

The SB17 tested very good in IMD tests, almost matching 18WU. If you listen to complex music such as Orchestra, maybe Purifi is better choice consider the price is similar to 18wu.

@abstract
Interesting concept if 1hz tremolo. I'm not sure what signal is best for IMD test.. 1hz could be hard to converge in time domain though.
 
@racingpht I've looked at those tests in detail. To my eyes every test up to 7.8mm excursion the SB is matching or beating the Scan. I'd say that's in contrast to what we see here at half of that excursion. However, if you look at higher displacement, we can see some tests at 17.3mm one-way, which is impossible. That driver is likely limited to 12mm xmech. So I'm not sure how to interpret the results. Either the calibration is off by a lot and actual excursion is half of what is quoted. In this scenario I'd say the tests are corroborating your findings. Or the calibration is fine, but 17.3m is just a nonsense reading as the coil is bottoming out. In this scenario I'd say the SB looks to be superior unless bottomed out.
 
  • Like
Reactions: racingpht
@cvjoint I agree. The excursion number is probably an estimation based on SPL or input voltage or something like that.

At 17.3mm one-way, the driver is likely destroyed. The stiff suspension of SBA should compress the excursion as other drivers do.. But we never know.

An interesting example from Klippel, the excursion in one direction could even decrease, with increasing input:
csm_motor_stability_4eabb07f33.jpg
 
That's what you get sometimes with a half roll surround when it collapses on itself. It reminds of the CSS SDX7. I received them already crumpled up from factory at x [mm] == 0! Had to fight the factory to take them back, they claimed it was normal. Makes you appreciate the precision we get with Scan and SS.

See for example these CSS drivers for sale:
https://www.diyaudio.com/community/attachments/photo-4-jpg.473917/
https://www.diyaudio.com/community/threads/fs-css-sdx7-enabled-css-el70enabled-alpair-10-2.272021/

More commonly occurs in tall boy surrounds on high excursion car audio subwoofers.
 
I'm no longer split on whether to use 18wu or SB17 for a line array. I've been staring at how flat the Klippel curves are for the 18WU up to 5mm, no change in parameters. That and Madisound had a special deal on them for Black Friday. Got 12 aluminum coned 4 ohm drivers for $180/piece!
 
  • Like
Reactions: racingpht and GM