There is a KEY element for a measurement - it needs to be repeatable and compareable! If it can't do that you are drawing pictures and don't do measurements.
The 2nd graph fullfills that criteria - it's repeatable and very well compareable. You can monitor even small changes during a development process.
Your spectogram is a nice graph but needs an additional step to make a measurement out of it. e.g. relation to the source signal. Pick out a time slice and have a closer look. We need to be able to see differences of at least 0,3dB - you now have a 80dB scale with colours ...
Many of the usual measurements are trimmed to exact these attributes - Frequency response with MLS/Sweep is repeatable, robust and precise. THD the same and shows a lot of the characters of a driver.
How they translate into what we are hearing ... is a different story. Klippels full scan gives a pretty good view of a speaker and uses these "ancient" measurements.
Short - distortion residual IS interesting, but we need a MEASUREMENT, not some pictures.
Hi, benefits of current drive have very little to do with the amplifier, but only the circuit impedance that is now in series with the driver. Ideal current drive amplifier would have infinite output impedance. So, any speaker with passive parts have some degree of "current drive", increased circuit impedance in series with driver so that the driver doesn't completely dominate current in the circuit, and current through voice coil.
In this sense industry has done exactly the right thing, as you say, voltage amplifier with passive crossovers, as it also makes the power ampflifier and speaker independent from each other without effect on frequency response. Years forward and DSP can make any filters and in my opinion is mandatory for good sounding system. But it is counterproductive to ditch the passive parts away altogether, especially on DIY land where few passive parts to optimize circuit impedance and reduce distortion makes no meaningful difference in total system cost, while DSP can still make any frequency response. I don't see this in industry, for some reason, perhaps it's not that well understood stuff, perhaps even forgotten. Perhaps it is, I'm not part of the industry.
edit. here is midrange driver in application settings, with only DSP, and the other one is with series inductor and DSP adjusted to match closely. Drop in distortion is quite remarkable, although it's not particularly audible in the first place.
If I'm reading Mikets measuremnet corrently, this distortion that got reduced by increasing series impedance, is nicely visible plotted like so.
It's nice way to see how important excursion is to control distortion, even though the distortion seems nice and low on the midrange on the traditional distortion plot, the distortion products stem from the lows and output at the midrange. Basically the whole bandwidth is distorted if there is lows to make great excursion, no matter how low swept distortion plot shows for the midrange.
Last edited:
Without wishing to appear brash, I contend this statement to be nonsense. The "best we have" has simply been overlooked for decades - I suggest because of a lack of proper understanding of the subject matter. Having said that, it certainly does not make for fun reading, so nobody should be accused of ignorance or berated for wishing to listen to their music instead.
In terms of linear colourations, the Wigner Distribution (WD) has long been advocated for measurement and analysis. It's totality of linear information comprises both time and frequency domain representations, and all those waterfall plots in between. In fact, all those other displays can be derived by smearing the information contained in the WD in some manner.
Historically, the usefulness of the WD was perhaps hampered by the requirement to sample at twice the frequency determined by Shannon's theory - and prior to that, even the need to sample! But in this day and age, it should no longer be a problem.
As I have commented in several other threads on this forum, relating well the WD to audible thresholds does then require a non-linear filtering action (a non-arbitrary smearing if you like to envisage it that way). Such non-linear filtering is able to account for our variable time-frequency resolving capabilities, so distinguishing the audibility of transient artefacts from more prolonged resonances, for example.
Thus we could have a waterfall display that conveys audible linear colourations well.
However, in terms of non-linear distortions, matters get more complicated since we are no longer concerned with a single frequency at a time. Instead here we need show how one frequency component relates to another (if indeed it is related at all). Also as I have alluded to in other threads here, much of the relevant information can be found in the bispectrum, that is moving from second-order spectral analysis to a third-order analysis.
Aside from getting to grips with the mathematical fundamentals (as you may have gathered, I recommend listening to your music instead!), a large problem here is how to display the information we retrieve: Usefully fitting a bispectral waterfall display into three dimensions (or two on paper) is a formidable undertaking!
Nevertheless, we have here the potential to correlate and display non-linear information with its cause. And where an artefact varies with a degree of correlation to its cause, it will likely be inferred as belonging to that cause; Where a correlation is more abstract, such as with the current dependent non-linearities in a moving coil driver that have been alluded to in this thread in the guise of "current drive", the distortion can be more audible as it is perceived as a separate source (a noise if you like, or "hazy glare effect" as has often been reported in this particular case).
Further still, if we overcome all those problems, we then need to account for having two ears separated in space - and likely having more than one loudspeaker contributing too. Here we have tools such as cross-spectra and cross-bispectra available, but given the ever-increasing quantity of information, the notion of combining it all into some single figure of "sound quality" or single display will likely remain folly.
And yet, even if we could do what appears to be impossible, our measurement and analysis system would still embed a fundamental flaw - that being its inability to account for the non-linearity of our hearing, and specifically our predominantly irreversible ability to learn. In effect, we would need to somehow derive new thresholds every time we conducted a test for every person we tested.
And so instead we apply "traditional" methods of loudspeaker measurement and analysis, often several individual different measurements to separate out the information we can reliably interpret - and also that which we can reliably convey to others. There is nothing "antiquated" here.
Nevertheless, adopting the long-established analysis techniques highlighted above could significantly aid measurements and displays that better correlate to abstract, singular assessments of "sound quality". The question to ask is whether it will ever really be worth the effort when you can sit there, listen to some music, and readily know yourself the answers that you seek to show on paper?
Hello tubelectron,
Hi tmuikku,
I am more than well acquainted with the concepts you are mentioning. This thread is not the place to bring up current drive as a topic. But I will say the amplifier design matters a great deal. It is also true, not open to debate as it has been proved by labs and others time and time again, the lower the impedance between a properly designed driver and the energy source, the better and well controlled the performance is. I've been involved in tests and listening panels. I have nothing to prove. So don't even consider asking me to do the research you should be doing. I have looked at both sides (if you will).
-Chris
This only occurs when the test isn't correct or complete. If you do your testing correctly with the right instruments, what tests good does sound good, including the testing better - sounding better argument. Once performance falls to lower levels you will see people break up into groups of what distortion types they can handle, and what they can't.
Hi tmuikku,
I am more than well acquainted with the concepts you are mentioning. This thread is not the place to bring up current drive as a topic. But I will say the amplifier design matters a great deal. It is also true, not open to debate as it has been proved by labs and others time and time again, the lower the impedance between a properly designed driver and the energy source, the better and well controlled the performance is. I've been involved in tests and listening panels. I have nothing to prove. So don't even consider asking me to do the research you should be doing. I have looked at both sides (if you will).
-Chris
They both register the same signal from the driver, but:
1. The measurement mic has 20dB higher noise than the condenser
2. The measurement mic does not insert any significant coloration while the condenser does. However, you can EQ the condenser in s/w
3. The measurement mic has significant non-linear distortions at high SPL, typically 10% at 110 dB SPL (or so, varies a lot). The condenser has much smaller non-linear distortions, typically 1% at 135 dB SPL.
4. The measurement mic is onmi while the condenser can be switched to cardio (4.8 dB DI) or figure-8 (6dB DI). The EQ-ed on-axis response would be the same as for measurement omni, but room reflections will be attenuated - which is great for a residential room adapted for a hobby lab.
Then you feed the recording from the (previously EQed) condenser to the software (Matlab or whatever it will be ported to, GPL), which runs System Identification / Adaptive Filtering algorithms and parses the recording into LTI (Linear Time Invariant) part and non-LTI residual, which is the distortions.
The sine sweep & MLS have been borrowed from radar technology of the 1960s with ~20 years lag. These methods have not been designed for acoustics and loudspeakers. These methods were very easy to implement on slow h/w of those days as they were based on the convolution of special self-orthogonal signals. It was assumed that the distortions on these constant amplitude signals and the real music are essentially the same. This assumption was wrong, and it became much worse with time because the driver vendors used to limit their objective testing to these signals... and it shows.
For the example above, the 300-3000Hz prefiltered excitation (input to the driver) shall have had non-linear distortions on the level of -50dB re linear LTI part, if THD curves and numbers could be trusted. In reality, it is only 24 dB lower than the linear part at 16 sec. Thus, there is 20+ dB of discrepancy between what we measure by traditional archaic methods and real music, and that is the issue of major concern. The size of the discrepancy varies a lot. Some drivers are better than others, and it is not clear why. I tried to contact a few driver vendors, but the answer was NIH.
At the time, in the early 1980s, MLS / sine sweep aka chirp was state of the art, and PC hardware was not capable of doing anything more complicated anyway. The math has evolved since 1960s, the h/w became 10000s times faster and cheaper. I'd argue that there is no need to limit loudspeaker measurements to sine sweep anymore.
Listening experience plays a big role.
When I was young (15 to 30 years) for me a low distortion loudspeaker only sounded not enough loud.
Today many public address systems sound in my ears terrible - they go for loudness perception. Rarely for quality of sound.
Often PA boxes make boring one note boom bass and the people do not complain.
It took me a long time working with diy loudspeakers dsp and measurement assisted to learn to hear and evaluate.
When I was young (15 to 30 years) for me a low distortion loudspeaker only sounded not enough loud.
Today many public address systems sound in my ears terrible - they go for loudness perception. Rarely for quality of sound.
Often PA boxes make boring one note boom bass and the people do not complain.
It took me a long time working with diy loudspeakers dsp and measurement assisted to learn to hear and evaluate.
Hi Anatech,
I'll bet the manufacturers are NOT using more complicated/sophisticated techniques because:
1. They have not been published in AES. Dr. Wolfgang Klippel does not have a theory that explains the distortion measurements in mid and high frequencies. Nobody does. Essentially, REW & Klippel are "the latest and greatest". However, vendors do have better hardware and sometimes, better rooms.
2. There is no evidence in the products that any vendor is using anything more advanced than publicly available like REW
3. There is evidence (first-hand admissions in public) that many loudspeaker vendor companies do NOT perform any measurements and rely solely on golden ears. Many driver vendors' engineers do use measurements but offload the final decision to managers' ears.
How much would you like to bet?
I'll bet the manufacturers are NOT using more complicated/sophisticated techniques because:
1. They have not been published in AES. Dr. Wolfgang Klippel does not have a theory that explains the distortion measurements in mid and high frequencies. Nobody does. Essentially, REW & Klippel are "the latest and greatest". However, vendors do have better hardware and sometimes, better rooms.
2. There is no evidence in the products that any vendor is using anything more advanced than publicly available like REW
3. There is evidence (first-hand admissions in public) that many loudspeaker vendor companies do NOT perform any measurements and rely solely on golden ears. Many driver vendors' engineers do use measurements but offload the final decision to managers' ears.
How much would you like to bet?
Off topic it might appear to be, but this response demonstrates a fundamental misunderstanding of the subject matter.
Current drive is implemented with some form of MFB where coil motion can be controlled by orders of magnitude more than with a coil resistance subject to manufacturing tolerances apparent in more normal voltage driven scenarios. Voltage drive from a low impedance amplifier aims by definition for an output resistance that is negligible to the coil resistance, so it has little effect on the final damping; Current drive with MFB circumvents alignment errors due to coil resistance variations altogether.
But what this response really misses is that current drive (if implemented properly) can substantially eliminate current dependent non-linearities due to magnetic hysteresis. This has absolutely nothing to with Qts, and is normally most audible in the midband. But this example does serve well in illustrating how and why certain distortions are more audible than others (as I indicated in my previous response). And that is relevant to this thread.
Hi mikets42,
I'll take that bet easily.
I knew folks at some major companies who do pursue advanced measurement techniques, and they have invested quite a lot in improving performance, and also line quality control. Many used custom procedures, so you wouldn't see anything published (being a commercial trades secret).
Sure. Many firms don't do much testing. But that isn't all firms.
I'll take that bet easily.
I knew folks at some major companies who do pursue advanced measurement techniques, and they have invested quite a lot in improving performance, and also line quality control. Many used custom procedures, so you wouldn't see anything published (being a commercial trades secret).
Sure. Many firms don't do much testing. But that isn't all firms.
Hi soundbloke,
Motional feedback either takes valuable space on the voice coil former (and adds mass), or an external coil (or whatever) is used. This means the correction must be frequency limited or it does oscillate.
I'm talking current drive. You're bringing in a system. Want to talk systems, okay but bring in all the information. You don't need to call it current drive either, you just adjust the voltage drive. It's the same thing, you have a feedback system sensing cone position and correcting the motion with the drive.
This is not a misconception. Let's be clear on what we are talking about.
Motional feedback either takes valuable space on the voice coil former (and adds mass), or an external coil (or whatever) is used. This means the correction must be frequency limited or it does oscillate.
I'm talking current drive. You're bringing in a system. Want to talk systems, okay but bring in all the information. You don't need to call it current drive either, you just adjust the voltage drive. It's the same thing, you have a feedback system sensing cone position and correcting the motion with the drive.
This is not a misconception. Let's be clear on what we are talking about.
That is not true. Try John Vanderkooy's 1998 seminal JAES paper "A Model of Loudspeaker Driver Impedance Incorporating Eddy Currents in the Pole Structure". Klippel's apparatus and many others can identify these distortion sources perfectly well.
Beyond that the reflection in the impedance of diaphragm break-up is also well-known, just it is normally insignificant in comparison.
To what else you are referring, I have no idea. Hopefully you can enlighten me?
Harmonic distortion is most overrated measurement. It is easy to measure but ... That will not tell much about your speaker sound. There are also many other aspects that color your sound. (Like IMD, stored energy, compression etc.) There is no perfect sound. It depends your taste what you think is good sound.
No that is not correct either. MFB can be implemented via a bridge with no additional coil or space taken up. Birt demonstrated a self-balancing bridge, for example.
There is also no "must" re stability criterion and achieving significant performance gains has been made evident on several occasions such as Mills and Hawksford, for example, with no such oscillations in evidence.
But what is relevant here is amplifier output impedance. A high impedance (current drive) eliminates current dependent non-linearities; voltage drive cannot. Rather with voltage drive, driver motion is dependent on coil impedance, complete with its non-linear modulation by eddy currents. I reiterate once more that this is a different subject to using MFB for Qts control.
Boden, absolutely could - but this question needs to be addressed to the REW team. I know how to make embedded bare metal products that run for 20+ years but I have no idea how to make user-friendly products.
Soundbloke - I am 100% for the current drive and MFB ... but the sensor(s) must be incorporated into the driver by design. It's not easy to use the deep-feedback current drive but possible - with lots of measurements in high freqs because you have to ensure phase margin & feedback loop stability at fT.
Anatech - great! Could you ask them directly if they test on real music?
Jzl - I also measured headphones. Studio-grade headphones like Audiotechnika M50x have non-linear distortions on the level of -80 dB re main. And they sound great. I would love to have my loudspeakers sound like that.
Hi jzl,
Now, what THD measurement are you talking about? A meter pointer, or an audio spectrum? One shows a great deal more than the other and I would guess a drive with lower THD will sound better than one with a bit more distortion.
You can't throw rocks at one measurement technique without describing exactly what you are criticizing. Then again, something is way better than nothing.
Finally, I have never seen one person or group of people measure audio performance without also listening to it as well. That goes for electronics and acoustic audio development. Speaker technology has progressed more than almost anything else in the audio chain. This wasn't done by people listening to the product alone.
Now, what THD measurement are you talking about? A meter pointer, or an audio spectrum? One shows a great deal more than the other and I would guess a drive with lower THD will sound better than one with a bit more distortion.
You can't throw rocks at one measurement technique without describing exactly what you are criticizing. Then again, something is way better than nothing.
Finally, I have never seen one person or group of people measure audio performance without also listening to it as well. That goes for electronics and acoustic audio development. Speaker technology has progressed more than almost anything else in the audio chain. This wasn't done by people listening to the product alone.
IMD is just a different way of measuring and expressing a distortion mechanism. Stored energy is often a linear effect that is responsible for linear frequency response variations. Where stored energy is non-linear, such as in the aforementioned hysteresis distortions, it will be evident in the harmonic distortion measures. Compression similarly. It is how the measures are analysed and reported that is crucial.
No that is not correct. As a proclaimed fan of current drive, I assume you are aware of Birt's self-balancing bridge? If not, you will find it enlightening I am sure.
There are also several methods to avoid loop stability too, probably the easiest is transitioning from MFB to current derived feedback, so maintinaing the advantages of high amplifier impedance at high frequencies, where the MFB is no longer as relevant due to the much reduced displacements in this range.