Speakers (and room acoustics) are commonly mentioned as the weak link (and rightly so).
REW lets you run a measurement (logarithmic sweep) and choose to see distortion at any point on the measurement display.
REW also lets you generate a steady tone, and and on the RTA display choose to see distortion numbers.
The sweep produces much higher distortion numbers than a steady (single) tone.
I haven't found a discussion or explanation for this, and it makes me curious about the discrepancy.
Which display should I consider more 'correct'?
Why?
If, at a specific frequency in the sweep, the distortion numbers generated by the sweep are far higher than those generated by a steady tone at that frequency, which is correct?
Both are actually useless to the layman, because they don't directly correlate to musical programs. If you had to chose between the two you should select the sweep, as it's not steady state, and will more easily reveal potential resonances, and provide higher THD scores, which are likely to be closer to the performance scores of musical programs, which isn't steady state.
Both have a function at the design and QC level, but once released into the wild, not so much.
I personally don't believe that it's possible for a layman to fully, objectively, evaluate a speaker system. Obvious anomalies are just that and of course can be discerned by many, but outside of the obvious it requires education, experience, the proper test equipment and metric and an anechoic chamber.
Lastly, all scores shift based on amplitude, so if we pause to think about this more; our audio bandwidth has no less than 19,980 center frequencies and based on an 8-ohm nominal load depiction and 100-Watts, gives us no fewer than 282 differing Voltage potentials ( much more actually) which produces no fewer than 5, 634, 360 test permutation to manually work through; when merely using old school steady state or sine wave sweeps to measure sonic goodness. An obvious impossibility, in terms of working your way through them all as to gain a complete picture, to make any objective conclusions upon.
Speaker system design is all about trade-off, looking at one, two or just a few scores, doesn't provide the complete picture required, to discern what trade-offs are afoot and how they workout to form the sonic signature of the system as a whole. We need the complete picture!
The same can be said for amplifiers, preamps, DAC's; however the sonic differences between speakers are overt, making them obvious to virtually everyone. But the underlying science and calculations are virtually identical. It is as it has always been, a question of linearity!
Cook has started to touch on this in his 101 Amplifier Primer Thread:
http://www.avsforum.com/forum/91-au...6169-101-primer-amplifier-specifications.html
The difference is primarily in the signal to noise ratio, the noise floor is higher for the sweep which raises the distortion figures - how much depends on how high the noise floor is, using a longer sweep and multiple sweeps (if the soundcard handles them properly) lowers the noise floor and should bring the two readings closer. Where distortion is high the readings should be much closer as the noise should be less significant.
The device itself may also react differently to the different signal types, with the history of what has gone before affecting the response as the sweep progresses. This can particularly be the case for electromechanical devices like drive units, though tones can be worse due to drive unit heating effects.
If I run a 1M sweep at - 6 dB FS on a loopback of an old USB soundcard on my desk the second and third harmonics stand out. On the sweep at 4 kHz 2nd is 0.005%, 3rd is 0.015% and THD is 0.016%. On the RTA with 128k FFT 2nd harmonic is 0.0047%, 3rd is 0.0153% and THD is 0.0157%, so pretty near identical within the numeric resolution of the sweep readout (3 decimals). However, over the space of a few minutes or so the RTA 3rd harmonic level slowly drops, settling to 0.0095% - warming up seems to help this card. If I do another sweep measurement after a few minutes of playing the 4 kHz tone the 3rd harmonic on the sweep is also lower, reading 0.012%.
Your sound card doesn't seem to be lab grade and therefore not appropriate for use in determining summary observations, which are intended for share, as objective realities.
The main reasons for the increases in distortion can be found in observing the following, as frequency and amplitude change (or even just one of the two change):
1. Changes in Bl
2. Changes in Kms & Cms
3. Changes in Qts
The signal to noise ratio of the testing equipment should have nothing to do with it - if it does, the results are nullified.
1. Bl - the motor force of the drive system is not constant and as it changes so does the linearity of the transducer (convertor). There is a point where it becomes exponentially non-linear.
2. Kms & Cms - the cone is typically not perfectly biased, nor are the mechanical resistive agents linear (Spider & Surround). As a result, the drivers bias and non-linear mechanical resistive qualities will produce non linear distortions, of a higher order, at some frequencies and amplitudes compared to that of others.
3. Qts - is a summary specification based on Qms & Qes, these root scores are constantly changing. Frequency & amplitude are key factors, as they affect the mechanical and electrical resistances, which in turn can promote electromechanical ringing, which spikes THD.
When using a sweep, a driver is in motion: conditions of over shoot, undershoot or both exists, which affects the frequency to follow; these alone spike THD scores, but are in actually the product of the aforementioned specification working in tandem, as to produce the harmonic qualities, under said conditions.
There is obviously much more to the story, but these are the most relevant in speaking to the OP's context.
That is why it was selected, it would be difficult to illustrate a point about distortion by measuring something which didn't distort. The point was to show that under conditions of good signal to noise both sweep measurements and tone measurements can produce identical results when measuring with the same signal levels.
You seem to be missing the point of the original query, which was why should a sweep measurement produce different distortion values than a measurement of a single tone, not why do drive units distort.
All distortion measurements (indeed all measurements of any sort) include a contribution from noise. The type of measurement has a very substantial effect on how significant that contribution is.
And when producing a tone the driver is stationary, is it?
In theory sweep and single tone or stepped sine measurements should produce the same results for distortion. In practice they are usually very similar, but there are sometimes differences. The most significant contributor to the difference is usually the effect of voice coil temperature changes during the measurement, which are much smaller with a sweep than a tone or stepped sine measurement as the overall measurement time is much shorter. The nature of the surround can also have an impact, materials which exhibit significant creep impart a memory effect that alters the cone behaviour depending on the history of what went before. As far as drive unit production QA is concerned sweeps are the usual choice thanks to their speed, reduced heating effect and full coverage of the measurement span.
Hey, I honestly believe that I am speaking directly to the OP's question. I could be mistaken, I guess???
I also honestly believe that you think that you have responded point-by-point, and on point, with what I have posted, but unfortunately I also think that you’re mistaken, sorry.
I appreciate your effort in here, but I must report them to be a skew.
The relationships that I depicted speak directly to the why the THD scores would differ between the stimuli’s, in loudspeakers.
The relationships are mathematical, not anecdotal, and can be demonstrated in theory and measurement, with repeatability. Thiele & Small mapped and proved these theories, many years ago. Joseph D'Apolito, Vance Dickenson, Wolfgang Klippel, and dozens of other that followed have repeatedly validated their model, added to it, and have taken it into the large signal domain.
THD is an additive waveform product, which has many base causation and rates of propagation (I describe the primary root causation in a loud speaker), the sum of which form the harmonic structure of any electronic audio device, irrespective of the intrinsic noise floor. Noise floor is comprised of intrinsic and stray voltages. The two sum as to limit the usable power score of an amplifier - yes, but the OP is asking specifically about harmonics scores, as they relate to to two different signal types, in a loud speaker - I have never seen a signal to noise rating for a loud speaker - ever...
If the S/N rating of the test equipment is even in question, it shouldn't be used: -120,-130dB SN is the requite minimum range for lab grade test equipment.
I could be misunderstanding you, but I don't think that I am.
You are.
Again you are, and from the content of your post my attempt to explain it to you is likely to prove futile, but I'll have a go anyway.
I'm more than familiar with Thiele Small - the model I have implemented in REW to extract parameters is a result of discussion with Claus Futtrup (now CTO at SEAS) and Kund Thorburg on their work to extend the basic model to account for the frequency dependence of compliance and damping, details are in the
REW help.
That is beside the point, however, as the underlying mechanisms of drive unit behaviour and the causes of distortion are not particularly relevant to the original question, which was seeking an explanation for the discrepancy between the distortion figures at any particular frequency obtained by two different methods of measurement, log swept sine and FFT analysis of a pure tone via an RTA. REW was used for the measurements, its default behaviour is to use the same signal level for both, so the level dependence of distortion isn't a factor as the test signal levels are the same. That means we have to look more deeply for the explanation, as in principle the results should be identical: the underlying assumption is that the system being measured isn't changing, so neither should its properties (including its distortion at the particular signal level used) regardless of how they are measured. That provides two avenues to explore, whether the underlying assumption of unchanging behaviour (time invariance) is valid and whether the measurement methods may react differently to some property of the overall system, including the measurement chain.
On the first point, whether the speaker is changing its properties, two factors come into play. The first is thermal, as the speaker is measured the current passing through the drive unit voice coils heats them up. That alters their electrical properties a little and also their mechanical properties, which is much more of an issue with the tone measurement as the duration of the test and the energy put into the speaker are much higher. The second factor is the suspension of the drive units and the extent to which it suffers memory effects. That varies a lot depending on the materials used. Where memory effects are present they affect the results at low frequencies more than mid and high, but Ray doesn't mention the frequencies at which he observed the differences.
The second point concerns the measurement methods and how the results they provide depend on the measurement setup. The overall chain consists of the excitation signal (the signal which REW generates) which then passes through volume controls, a soundcard of some sort and an amplifier before reaching the speaker; the speaker itself; the room within which the measurement is being carried out; the mic capturing the signal; a preamp of some sort (either built into the mic or separately), a soundcard input path with further volume controls and finally the input data capture and processing in the measurement software. Noise affects every part of this chain, from the quantisation noise of the numerical precision of the calculations generating the test signal and processing the captured data to the thermal, shot and 1/f noise in the electronics and the acoustic background noise in the room. The dominant noise sources are the acoustic background noise in the room and the thermal and 1/f noise in the mic preamp.
With properly adjusted volume controls/gains the overall signal to noise of an impulse response measured using a log sweep will be around 60 to 80 dB. Its appearance will be something like this:
The linear part of the system's response is the region from just before 0 to around 1.0 s, where it disappears into the noise. The contributions of distortion harmonics appear at negative time, separated into their harmonic components (the second harmonic, then further back in time the third and so on). Noise is distributed throughout the response. The plots of fundamental and harmonics are produced by taking FFTs of the relevant sections of this response and then determining their ratios to display the measure of distortion. Each plot shows the spectral content of the signal and its accompanying noise, if the noise floor is high and the signal level (such as a harmonic component) is low, noise will dominate the plot of that harmonic.
For the RTA, the test signal is a tone that plays continuously. A segment of it is captured, windowed and analysed with an FFT to show the spectral content, which will have a big peak at the tone frequency, smaller peaks at the harmonics and an overall noise level. Here is an example:
![]()
In both cases the same noise sources are present, but their contribution to the end result is quite different due to the different distributions of the energy in the excitation signal. In the log sweep case we are measuring a whole band of interest, often spanning the full sample rate of the sound card. The energy in the test signal is distributed across that frequency range, only a proportion of it is present at any particular frequency of interest. In the RTA case all the test signal energy is at the test frequency, there is none elsewhere. That means that the ratio of signal at the test frequency to overall noise is much higher for the RTA than it is for the log sweep, and there is correspondingly more energy in the harmonics as well - the energy is a combination of the levels of the signals (which are actually the same for both log sweep and RTA measurements) and the time they are present, which is much longer for the RTA measurement. As a result the RTA is much less affected by the noise contributions and can resolve distortion levels that are much lower than those the log sweep can resolve.
Overall then, there are two answers to Ray's original question. The results may differ due to changes that are occurring in the speaker while it is being measured, primarily due to temperature rise in the voice coils, and if the overall measurement signal to noise ratio is poor the sweep distortion levels will be raised by the contributions of the noise floor.
Hi John
I think that you might feel that I am directly challenging you in your area of expertise. That is not what has occurred, or what is about to occur.
The OP in his original posting, and the first few to follow, presented questions, based on his observations, of his DUT. He didn’t qualify the conditions, what-so-ever. So we were left with predominantly assumption to work with. Each of us gravitating towards different, but equally assumptive paths.
For me, because he didn’t list any of the requisite conditions noted below, I decided to speak to THD in the speaker’s (electromechanical-transducer) themselves, irrespective of all other factor (purely electrical or acoustical). I acknowledged that harmonic distortions are additive (meaning each circuit/device/ trace, etc. (inclusive of air) adds them), within the signal chain, producing a higher final result. I then went on to describe in intrinsic roots of THD in piston type transducers (I assumed he was using a traditional loudspeaker system).
Missing qualifiers/conditions:
1. The make, and model of the DAC frontend – for us to look up
2. The make and model of the mic – for us to look up
3. The state of calibration
4. The micing technique that he employed
a. Near Field – with distances & filters listed
b. Fare Filed – with distances & filters listed
c. Etc.
5. How many speakers were being tested at once?
6. What the relative position of the microphone placement was..
7. What type of speaker was being tested
8. What make and model of speaker being tested – for us to look up
9. What frequency(s) were used for the steady state samples
10. What bandwidth was used for the sweeps
11. What make and model of amplifier for us to look up
12. What output voltage was used – what is consistent
13. What was the reference SPL’s was during the tests – was it consistent
14. What were the actual differences in THD scores?
15. 15. And so-on, and so-on…
In an all things equal test metric (if you will), if everything remained the same, other than the speaker being exchanged one for another, the THD scores would change, irrespective of all the other factors remaining the same. The electronic and electromechanical relationships which exist between the source, preamp, and amplifier, and speaker system, produce THD products, which are summed to produce a unified summary score. After this, one can add acoustic coloration, which result, and vary from one position to another, with a given acoustic space. In fact, if one even turns their head, changes in harmonic structure are commonly evident, making mic and weighting technique, requisite disclosures amongst many.
The speaker’s intrinsic distortion producing qualities are prime, over that of additive distortion gained within the acoustic domain (or even in the purely electronic domains), for no other reason then speaker are the most non-linear link, in the entire chain/circuit, and is the agent of wave propagation in the acoustic domain. The more noise (harmonic content) that a speaker, amp, preamp signal chain inject into the room, the higher the noise floor, and as a result, the higher the THD in the acoustic domain, regardless of stimulus. Hence, when faced with only assumptive paths to choose from, I chose the electromechanical one.
I still honestly believe that I answered the questions presented by the OP, at least to the magnitude and relevance possible, given the many possible contexts, which resulted in the wake of the OP’s insufficient citing of conditions.
His questions are excellent, but the clues he left behind for us to attempt answers with, was, well let’s just say, it put us both to the test.
For me, I have no stake in this, and will not comment further on the subject, unless you would like to continue the discussion further, with me.
I believe my posting to be clear and accurate within the context of which they have been presented.
