What is a valid speaker IMD test?

[6speed]

I have noticed IMD testing of speakers mentioned a few times, but no results...or even suggestions. As a starting point, can we just use an equally spaced multi-tone signal and run an RTA?

 

[Blumlein 88]

You could do it that way. Or you could do two tones, and pick several frequencies.

My suggestion would be the multi-tone like Amir uses in DAC and amp testing.

And I'd suggest an IMD sweep. Start with sweeps running concurrently at say 2hz to 19,000 hz and 1002 hz to 20,000 hz. Easy enough signal to make. Then record it and you can see what is happening at each frequency this way.

I'm expecting when the two tones are on different drivers woofer and tweeter, you'll not get IMD, and when they are on the woofer or both on the tweeter you'll get some IMD.

 

[6speed]

I didn't even realize REW could import wav files. Well that opens up some possibilities.

 

[Blumlein 88]

Yes REW will generate and record the results of two tones and you can pick the frequencies. You could do several of those as spot checks. It will import sweeps and even wav files yes.

Using the Generator Function and RTA has some nice advantages like displaying for you the IMD distortion levels as they occur.

 

[6speed]

I generated the tones in REW, saved them to a wave file, then reopened in REW. The result doesn't give the same separation at low frequencies as Amir's amplifier test tones. I tried with 16 and 24bit PCM with essentially the same results.

 

[6speed]

I tried running REW RTA on the 31 tone test in yellow in my previous post and here are the results. The mic is just 2 feet from the speaker being driven and the non-level peaks show the room modes. I was surprised to see lower sampling rates with less noise and better low frequency resolution. I tried 192kHz and it was incrementally worse than 96kHz. The noise floor is really high though and I was unable to find a way to lower it. What does rise above the noise floor does not always seem well correlated between the two sampling rates, making it all rather suspect.

I did not have time to run an IM sweep, but will try to soon.

 

[Blumlein 88]

I tried running REW RTA on the 31 tone test in yellow in my previous post and here are the results. The mic is just 2 feet from the speaker being driven and the non-level peaks show the room modes. I was surprised to see lower sampling rates with less noise and better low frequency resolution. I tried 192kHz and it was incrementally worse than 96kHz. The noise floor is really high though and I was unable to find a way to lower it. What does rise above the noise floor does not always seem well correlated between the two sampling rates, making it all rather suspect.

View attachment 61782

I did not have time to run an IM sweep, but will try to soon.

What is happening if you use the same size FFT for higher rates, is each FFT bin is twice or four times as wide. So it has 3db or 6db more noise on the same signal. And this wider FFT bin size means lesser resolution at low frequencies. So your results are what is expected. Does this make sense or should I go into more detail?

 

[A800]

Larger FFT size usually means higher resolution.

 

[Blumlein 88]

Larger FFT size usually means higher resolution.

I don't think that is what was done. I think the same FFT size was used for first 48 khz, and then 96 khz. In the latter case, each FFT bin is twice as wide, has 3 db more noise, and less frequency resolution due to the wider FFT bin size. If you double the FFT size with doubling the sample rate/bandwidth, resolution and noise would have stayed the same.

Please note FFT bin size and FFT size are inversely proportional.

 

[6speed]

So every time I double the sampling rate, I need to double the FFT size to keep the bin width the same?

 

[Blumlein 88]

So every time I double the sampling rate, I need to double the FFT size to keep the bin width the same?

Yes that will do it.

REW in its latest version added 2meg and 4 meg FFT sizes. It said the primary reason was for use with higher sample rates.

 

[6speed]

Thanks for the guidance so far @Blumlein 88, but it looks like REW's sweep generator (the tab meant for measuring) uses logarithmic spacing because sweeps from 2Hz - 19kHz and 1,002Hz - 20kHz are not the same length of time. I also don't think they are quite the same thing as a normal sweep, which means I can't use a third party tool to merge the wave files.

 

[aarons915]

I've thought about this before and it seems to me that the best IMD test would be a tone at whatever frequency causes the highest excursion in your speaker, 100Hz is a good bet since that is right before the port starts helping out, then run a sweep showing the worst case and at what frequency. Anyone know if we are able to import this 100Hz tone and sweep into REW will it measure the IMD distortion?

 

[GelbeMusik]

I have noticed IMD testing of speakers mentioned a few times, but no results...or even suggestions. As a starting point, can we just use an equally spaced multi-tone signal and run an RTA?

IMD seems pathetically complicated, in that it introduces an additional dimension. Two tones, but which? Multi-tone isn't that revealing, because the amplitude of each is fairly low, too low to provoke trouble. I've seen highly respected drivers which show nearly no sign of any non-linearity with multi-tone. But, driven with some elevated bass content, choose the second frequency close to a dip in frequency response, and the IM spectrum broadens about several octaves with unbearable sonic coloration (read: siren). O/k, that's the real effect of the suspensions secondary resonance ... THD was perfect, though.

I personally check tones from the lower bandwith end with close to max excursion against a range of exemplary single higher tones. Additionally I test for potential problems with situations as decribed. That would give a picture of how much the excursion would affect the upper range. Speakers which pass the test, I can trust to support the targeted quality. Or, I recap my decision about the bandwith I want to cover with the particular driver. Kind of a Search And Rescue thing, so far. No automation in sight, for me.

 

[Blumlein 88]

I've pointed out before, you aren't going to get IMD between tones that are reproduced by separate drivers in a multi-way speaker. So that is one big complication.

 

[Blumlein 88]

Thanks for the guidance so far @Blumlein 88, but it looks like REW's sweep generator (the tab meant for measuring) uses logarithmic spacing because sweeps from 2Hz - 19kHz and 1,002Hz - 20kHz are not the same length of time. I also don't think they are quite the same thing as a normal sweep, which means I can't use a third party tool to merge the wave files.

@6speed
I didn't think about those being log sweeps.
https://send.firefox.com/download/ba8a4c07d5d763f1/#ckgs5MJcrcKSZnubYk_Qxg

Here is a link to download a linear sweep I made. The rate of the sweep is such that you can select 2048 samples for an FFT in Audacity, and the IMD products are almost exactly what you'll measure with a fixed pair of tones. You also should use a 2048 FFT size for this. I can also show you how to look at results usefully using a spectrogram if you wish which sometimes is more instructive than the FFT view.

While I'm doing this here is a link to the AP multi-tone test like Amir uses in his reviews. Includes a chart showing frequency and phase of each tone.

https://send.firefox.com/download/fb34ed7b1c9b5de1/#zKEaHQsoFj-hwQX9yqeqxA

These links are good for 7 days or until a certain number of downloads of them have occurred.

 

[6speed]

@Blumlein 88 Thanks. Your AP multitone looks a lot cleaner than the one I generated using REW.

Unfortunately, REW will not use your sweep for measurements. When I import it, REW says I need to use "a measurements sweep saved from this version of REW." With a single tone sweep, REW knows to record the fundamental and 9 harmonics, but does not have an IMD measurement facility, so I do not know how it would record an IMD sweep where we want to be able to see n-number of sum and difference products. Furthermore, I don't know how one would want to display an IMD measurement other than via fixed tones.

 

[LTig]

@Blumlein 88 Thanks. Your AP multitone looks a lot cleaner than the one I generated using REW.

View attachment 61985

Unfortunately, REW will not use your sweep for measurements. When I import it, REW says I need to use "a measurements sweep saved from this version of REW." With a single tone sweep, REW knows to record the fundamental and 9 harmonics, but does not have an IMD measurement facility, so I do not know how it would record an IMD sweep where we want to be able to see n-number of sum and difference products. Furthermore, I don't know how one would want to display an IMD measurement other than via fixed tones.

You can play the multitone wav using a bit perfect player, and REW RTA for record. You won't get the distortion panel though.

 

[6speed]

You can play the multitone wav using a bit perfect player, and REW RTA for record.

I plan on doing that for the multi-tone, but how do you record a 2 tone sweep? You could use the RTA again...and then what the average forever setting (or peak trace) will ensure that the swept tones and IMD products remain?

 

[Blumlein 88]

I plan on doing that for the multi-tone, but how do you record a 2 tone sweep? You could use the RTA again...and then what the average forever setting (or peak trace) will ensure that the swept tones and IMD products remain?

I'm sorry, I should have been more specific. Just play and record the linear sweeps I posted. Then analyze them with Audacity or another program which will do FFT analysis. You won't get the convenience of REW spitting out the numbers for you. You'll have to look at them yourself and see what the levels are, but you can determine how much IMD distortion is occurring.

An FFT requires a certain number of samples to give a reading. In a sweep you don't want too large an FFT size or it will miss finer details in time. For instance a 2048 or 2k FFT will need a minimum of 2048 samples to give you results. I used a sweep rate so that if you use a 2048 FFT and select 2048 samples at any point during the sweep, you'll get results essentially identical to having used a pair of fixed tones. 2048 might seem like a small size, and for many purposes it is, but in this case it works fine to see how IMD is changing with frequency over a sweep.

I'll post an example in a minute.