Multi-tone intermodulation distortion testing

[Kees de Visser]

Scott Wurcer has demonstrated a reverse test of this sort- he puts multiple narrow notches into a white noise signal. Very clever, IMO.

That's interesting. I've DIY'd a similar test myself, not knowing it existed already, to test IMD differences between hi-res and redbook audio. I created an audio file with alternating hi-res/lo-res and made several versions with different notches. The files were played back over speakers and recorded with a high quality microphone. The SNR in my setup (room) was insufficient to observe any differences in the notched bands, but I think the test is basically valid. A very quiet room might have helped

 

[DonH56]

For stability impulse or step response testing is often used. A VNA (vector network analyzer, capable of showing magnitude and phase) can also show things like Nyquist plots, K-factors, and other measures of stability.

Two-tone (or multitone) testing provides more insight into both time and frequency domain performance and how/what nonlinearity terms are present. Classic distortion response in very predictable since there is a mathematical relationship between single-tone and two(multi)-tone distortion products and the order of the nonlinearity. Deviations can indicate power compression or other issues (than power compression, things like hystersis, ISI-type distortion, nonlinear frequency response, etc.) that are harder to see in a single-tone test.

 

[andreasmaaan]

I think it would be nice to see a two-tone (or multi-tone) graph for each of the devices @amirm tests here

 

[KSTR]

Scott Wurcer has demonstrated a reverse test of this sort- he puts multiple narrow notches into a white noise signal. Very clever, IMO.

Something related, a "silence sweep" by Angelo Farina (father of the now widely used logsweep+convolution method to obtain impulse response incl. pre-pulses of the distortion components): http://pcfarina.eng.unipr.it/Public/Papers/247-AES126.pdf

 

[Werner]

The SNR in my setup (room) was insufficient to observe any differences in the notched bands, but I think the test is basically valid. A very quiet room might have helped

A bit OT, but you could try with headphones instead of speakers. The closer proximity of the microphone will help suppressing the room.

 

[Kees de Visser]

A bit OT, but you could try with headphones instead of speakers. The closer proximity of the microphone will help suppressing the room.

I've thought about that too, but it probably requires small (miniature?) microphones with inherent higher self noise levels. It also involves making a dyi coupler, something like a wooden plank with a hole. The question remains, if I can't measure distortion products in a deadly silent (24 bit dither) notched band under normal listening conditions, are we supposed to be able to hear them anyway ?
Good suggestion though, thanks !

 

[solderdude]

When someone has those notched files I can run it through some headphones and check for differences.
How wide are those notches ?

 

[Kees de Visser]

I usually make them quickly with iZotope RX, so it's easy to make any notch(es) you prefer. Any suggestions ?

 

[solderdude]

SF10 will also make them I reckon. Will try myself... thought files like this were already available.
Thinking of notching around problem areas and known good areas of headphones.