Distortion measurements

[dweeeeb2]

Hi, Im trying to improve my understanding of the distortion graphs (EDIT: speaker distortion). I believe its basically a comparison of signal in v’s measured output. Does this mean a sweep is performed and measured as per FR graph?
Does the deviations from flat on a FR graph show up as distortion?
thanks

 

[DonH56]

Distortion is often shown in the form of a frequency sweep to show how it varies over frequency. That is not the same as a frequency response plot, which shows how the output deviates from an ideal flat horizontal line at the test frequency. Distortion is caused by nonlinearities in the amplifier (or whatever) and adds frequencies not in the original source. Deviations in the frequency response are not usually nonlinear; they change the amplitude over frequency, but do not add other distortion products (no additional frequencies). They are different tests measuring different things.

Another way to think of it is that a frequency response plot sends out a series of frequencies, one at a time, and measures the response at just that frequency to generate a plot. E.g. step frequency 10 Hz, 20 Hz,.... to 20 kHz or beyond (usually more points are taken). Any change from a flat line is recorded but that can be due to a linear thing, like something loading the output down at a particular frequency, that reduces the output but does not add distortion. And if it did, the frequency response plot would not measure it. Each point in the frequency response plot is just the amplitude of the signal at that single fundamental frequency.

In contrast, at each point of a distortion plot a number (typically 10) frequencies are measured to assess the distortion at each frequency. At 10 Hz, there will be the fundamental tone of 10 Hz measured for reference, and then measurements at 20, 30, 40, 50, 60, 70, 80, 90, 100 Hz to capture harmonics of the 10 Hz input to see how much distortion is generated. If the amplifier were perfect, all the harmonics would measure 0 amplitude. You root-sum-square (RSS, sqrt(a^2 + b^2 + ...) for however many terms you wish to include) the energy in all the harmonics and compare that to the fundamental signal level to calculate the distortion. Each point on the distortion plot is a ratio of the signal to the RSS of all the distortion terms (THD typically uses ten points).

HTH - Don

 

[dweeeeb2]

Thankyou Don

Am I right in thinking that the data required for a FR graph is actually captured when performing a distortion measurement? ie the amplitude of the fundamental?
Also, with reference to only measuring the fundamental frequency for FR data, does this mean that what is heard could be quite different? ie: we can hear harmonics but this is not recorded in a FR graph.

thankyou for your time.

 

[DonH56]

Thankyou Don

Am I right in thinking that the data required for a FR graph is actually captured when performing a distortion measurement? ie the amplitude of the fundamental?
Also, with reference to only measuring the fundamental frequency for FR data, does this mean that what is heard could be quite different? ie: we can hear harmonics but this is not recorded in a FR graph.

thankyou for your time.

It could be but IME they are done as separate measurements.

If the deviation in frequency response is large enough you could hear it; imagine if some frequencies were raised or lowered enough to be audible.

Distortion can be heard if large enough but is not measured by a frequency response plot. I suppose if some frequency was boosted or lowered enough at a particular harmonic you might hear it. Imagine if 3 kHz was boosted by 10 dB (unrealistically large); then, the third harmonic distortion of a 1 kHz tone would be twice as loud compared to a flat frequency response.

But you are sort of comparing apples and oranges. A FR curve is meant to show amplitude flatness, not distortion. A distortion plot may hide frequency deviations since distortion is relative to the fundamental. In the example above, if the FR was boosted at 3 kHz and you measured the distortion of a 1 kHz tone, then it would appear worse since the third harmonic is boosted. In that sense the FR affects the amount of distortion.

 

[DVDdoug]

Amir's SINAD graphs are signal-to-noise and distortion together. They are plotted as a frequency spectrum so you can see the frequencies of whatever noise & distortion is added. Typically, he feeds-in a 1kHz signal so a perfect result would be a single vertical line at 1kHz. (There is some imperfection in the measurement/calculation so a perfect device won't show that ideal single straight line... You'll see that spreading-out at the bottom low-levels.)

Harmonic distortion shows-up as spikes at the harmonics (multiples) at 2, 3, 4kHz, etc. Spikes at 60 or 120Hz are power line noise (and the associated harmonics) and there may be other noise spikes. The fuzzy-wiggly stuff along the bottom is more-random noise.

 

[RayDunzl]

Am I right in thinking that the data required for a FR graph is actually captured when performing a distortion measurement? ie the amplitude of the fundamental?

If you are using REW (or I suppose, any similar analysis software), a single measurement sweep provides all the data it needs for all of its measurements.

SPL and Phase
Distortion
Impulse Response.
Group Delay
Step Response
Noise
etc.

The mathematics involved in calculating these things from the swept sine tone are beyond my payscale. Ask someone else about that, if interested.

 

[ernestcarl]

The mathematics involved in calculating these things from the swept sine tone are beyond my payscale. Ask someone else about that, if interested.

Or just go read Farina's paper directly (yea, I'm not nearly smart and invested enough to want to understand the equations step by step so I'd skim that part ... it might as well be another language to me )

https://www.researchgate.net/publication/2456363_Simultaneous_Measurement_of_Impulse_Response_and_Distortion_With_a_Swept-Sine_Technique

 

[amirm]

The Farina (log Chirp) sweeps do produce harmonic distortion. It kind of falls out of the measurement in the way the harmonics appear in "negative time." They do have an accuracy limit however. They also create boundary condition errors so not suitable for determining distortion at the beginning and ending sweep frequencies. Much also depends on parameters used for the sweep. Longer ones for example give higher frequency resolution.

 

[Steven Holt]

You root-sum-square (RSS, sqrt(a^2 + b^2 + ...) for however many terms you wish to include) the energy in all the harmonics and compare that to the fundamental signal level to calculate the distortion

Yes, this will give you TOTAL Harmonic Distortion, which is a sacred yardstick in our hobby. But I wonder, is it the best? RSS sqrt(a^2+b^2+c^2+d^2+e^2)/5 gives you the AVERAGE Harmonic Distortion. (Nobody in our hobby goes past the fifth harmonic). It seems to me, that this would give people a better idea of how harmonic distortion affects the sound -- or am I off base here? Thanks for your post and excellent discussion on this topic.

 

[DonH56]

Yes, this will give you TOTAL Harmonic Distortion, which is a sacred yardstick in our hobby. But I wonder, is it the best? RSS sqrt(a^2+b^2+c^2+d^2+e^2)/5 gives you the AVERAGE Harmonic Distortion. (Nobody in our hobby goes past the fifth harmonic). It seems to me, that this would give people a better idea of how harmonic distortion affects the sound -- or am I off base here? Thanks for your post and excellent discussion on this topic.

All the measurements I have performed and seen over the years used ten terms for THD and is based upon an RSS of harmonics compared to (ratio of) the fundamental level. Your formula would reduce the absolute harmonic distortion level but I do not see how a scalar reduction of the distortion term would give people better understanding of harmonic distortion. THD as a single metric does not necessarily say how it affects the sound. The frequency content, or amplitude of the harmonics over frequency, is also very important. Not just odd vs. even, but also low- vs. high-order harmonics and so forth. To me that would provide more understanding than a single number.

 

[amirm]

Yes, this will give you TOTAL Harmonic Distortion, which is a sacred yardstick in our hobby. But I wonder, is it the best? RSS sqrt(a^2+b^2+c^2+d^2+e^2)/5 gives you the AVERAGE Harmonic Distortion. (Nobody in our hobby goes past the fifth harmonic). It seems to me, that this would give people a better idea of how harmonic distortion affects the sound -- or am I off base here? Thanks for your post and excellent discussion on this topic.

As Don mentions, audibility actually becomes more important with higher order harmonics as they escape the masking effect of the primary tone.

 

[dweeeeb2]

As Don mentions, audibility actually becomes more important with higher order harmonics as they escape the masking effect of the primary tone.

These harmonics are missed in the FR graph correct. So to expand further, if 2 speakers were to magically have the same FR they could still sound different (disregarding dispersion)?

 

[DonH56]

These harmonics are missed in the FR graph correct. So to expand further, if 2 speakers were to magically have the same FR they could still sound different (disregarding dispersion)?

In general a graph of frequency response does not include distortion so yes. Again, a FR plot shows the amplitude at each frequency, while a distortion plot shows the ratio of fundamental amplitude to the distortion products at each frequency. A FR plot shows one frequency at each point; a distortion plot includes a series of frequencies at each point.

Example:
FR plot at 10 Hz = amplitude at 10 Hz
THD plot at 10 Hz = amplitude at 10 Hz / (mathematical function summing terms at 20, 30, 40, 50, 60, 70, 80, 90, 100 Hz)

 

[Steven Holt]

Don and Amir, thank you. I'm going a adjust my thinking on this.

The frequency content, or amplitude of the harmonics over frequency, is also very important. Not just odd vs. even, but also low- vs. high-order harmonics and so forth. To me that would provide more understanding than a single number.

^^THIS!!

 

[dweeeeb2]

Thank you, I feel like I have a much better (all be it at beginner level) understanding by of what each graph represents. cheers

 

[sam_adams]

The below will either enlighten or add to the confusion, but it is definitely of some interest:

Steve F. Temme, "How To Graph Distortion Measurements" 94th AES convention March, 1993.

 

[Yorkshire Mouth]

As Don mentions, audibility actually becomes more important with higher order harmonics as they escape the masking effect of the primary tone.

For a real-world example. I’m interested in the JBL 308Ps.

The distortion graph for 86db looks ‘okay’, the 96db looks poor.

But I’ll be using them at lower volumes, and I see the worst distortion is from 2nd, and to a lesser extent 3rd order harmonics (which drive total quite high), whilst 4th and 5th look fine, apart from very low bass, where it’s apparently less important (I intend to use a sub anyway).

Would this distortion be noticeable at 86db and below?

 

[Digital_Thor]

So... I tried to compare 2 very differently priced 8" drivers.
Can someone guess which one is the SB23NRX and which is the Purifi PTT 8.0 ?? Because I'm in doubt what to look for