I'm not trying to implement something, I already did it. I'm only extending an existing, successful application.
In that case I am simply questioning your definition of successful. I would suggest that without proper system identification the inaccuracies inherent in your finished application leave it fundamentally flawed and add distortion where no audible artefacts need be apparent.
It is also worth remembering that auditioning transient limiters requires a loudspeaker system capable of reproducing the transients in the first place. Such systems are by no means typical but without that benefit there will be little to tell one form of compression from another.
Believe me, the limiting is very specific to the application. The large-signal response of the driver must be well characterized in order to apply this type of limiting properly. You need to know other things about the system. It's intended to be built into a multi-way active loudspeaker as part of the crossover. It's not a standalone application.
Sure. And your point is? What you wrote is basically insinuating that I am not aware of the limitations of the systems I am working with. What is your basis for that???
I think you are completely misunderstanding what I am doing.
Last edited:
I am not insinuating anything. I am simply supplying a few facts.
To quote your first post...
This is my understanding of your objective.
This is not a good estimate. Non-linearities and time variant linear elements prevent sufficient accuracy with a simple model. It is not uncommon in over-driven woofers, for example, to see a 6dB reduction in sensitivity due to thermal compression (in conventional voltage driven mode at least). This is by no means the only issue.
The simplicity of the self-balancing bridge makes an outstanding case for your application - either as online feedback or as a means for offline model calibration. Some form of measurement is a necessity outside of a theoretical exercise and my experience is that the bridge when implemented properly is superior to the addition of physical motion sensors.
Returning to this statement again... How then are you establishing your limiter is "blameless"? It is a separate but very salient point in how do you know what you are listening to and which is correct - or at least the minimally compromised?
This is compression, not protective limiting and a whole different subject. Aside from Kraftwerk's early work featuring a 3dB crest factor, music is typified by a higher peak to rms ratio. Loudness is closer to function of rms power not peak magnitudes. Such compression can also generate harmonic content that becomes more audibly detrimental than the clean peak clipping you are attempting to provide.
Am I misunderstanding anything?
Thanks for your post with the quotes of my previous posts. I think I see where you are coming from now. So let me explain more to clarify where and how I want to use this feedback-less system.
Addressing this last comment, let's re-examine my goal - it is to limit the cone excursion. What you are saying is correct, that a linear model (TS parameter based) of the driver output is not very accurate and if I was trying to reduce distortion it would be totally inappropriate (it is just a linear model!). I would need to monitor the instantaneous driver output (via acoustic or electrical feedback) in order to have a sufficiently fine grained model of the driver's output. But this is not what I am trying to do. Getting an estimate of the cone excursion really doesn't require all that complex of a model - well, that is what I am asserting at least. A linear model in the large-signal regime is just fine. Why? Well we don't really care what is happening in the small signal regime because no limiting is needed there. So we can simply focus on the large signal behavior, and this is something that can be obtained from a Klippel analysis of the driver.
Regarding your comments about the behavior of an "over driver woofer": Again, what you say is strictly correct. But it doesn't really apply to the application where I want to use the limiter/compressor that I have written. My guess is that you are envisioning a small closed box woofer with a large LT boost, driven by a high power amplifier. As you say, this would certainly cause thermal compression. My application is an open baffle loudspeaker woofer or large format midrange. To raise the low end up to the passband level a significant amount of boost is applied (depending on how much of the OB loss you want to correct). But because I typically use high sensitivity drivers (both pro and consumer), the power levels involved to reach Xmax are not high, maybe 50W. With modest playback levels and program material with modest low bass energy I can lift the woofer response to have F3 of 30Hz but at higher playback levels or with bass heavy music this has to be dialed back because the excursion cannot support it. At the same time, because of the higher sensitivity of the driver(s), I only need a couple of Watts in the upper passband (the woofer passband extends up to 500Hz in my system). Some OB users simply would use a 50W amp and drive it into clipping. Because only 50W are on tap, the amp cannot "overdrive" the woofer, but the amp will be clipping very often. This is not what I want to do. I want a higher power amp to stay out of clipping and for transients at higher frequencies where Xmax is not an issue. For music with little bass energy I can boost up the low end still enjoy (for example) the 30Hz F3 while staying within Xmax. But with the higher amp power on tap, there is the possibility to over driver the woofer, so my goal is to monitor the music signal, and "dial back" the low bass when the low bass content is too much for the drivers' Xmax to support. I want to do this is as blameless a fashion as possible, so I am using a lookahead limiter. Again, I am not trying to discriminate excursions to the sub mm level - an estimate to within a few mm is fine for this purpose. This is why I feel that the linear model is good enough.
The only thing that my limiter/compressor does is reduce the level in a sub-band. Essentially it is automatically riding the "bass tone control knob" of the system. The attack and release rate can be ramped up and down over as much time as the user desires due to the look-ahead nature of the algorithm.
I think that you are correct when you call this action compression. I see it as "preventing the excursion from exceeding some limit", which I personally think of as a limiter, but perhaps this is just semantics.
Addressing this last comment, let's re-examine my goal - it is to limit the cone excursion. What you are saying is correct, that a linear model (TS parameter based) of the driver output is not very accurate and if I was trying to reduce distortion it would be totally inappropriate (it is just a linear model!). I would need to monitor the instantaneous driver output (via acoustic or electrical feedback) in order to have a sufficiently fine grained model of the driver's output. But this is not what I am trying to do. Getting an estimate of the cone excursion really doesn't require all that complex of a model - well, that is what I am asserting at least. A linear model in the large-signal regime is just fine. Why? Well we don't really care what is happening in the small signal regime because no limiting is needed there. So we can simply focus on the large signal behavior, and this is something that can be obtained from a Klippel analysis of the driver.
Regarding your comments about the behavior of an "over driver woofer": Again, what you say is strictly correct. But it doesn't really apply to the application where I want to use the limiter/compressor that I have written. My guess is that you are envisioning a small closed box woofer with a large LT boost, driven by a high power amplifier. As you say, this would certainly cause thermal compression. My application is an open baffle loudspeaker woofer or large format midrange. To raise the low end up to the passband level a significant amount of boost is applied (depending on how much of the OB loss you want to correct). But because I typically use high sensitivity drivers (both pro and consumer), the power levels involved to reach Xmax are not high, maybe 50W. With modest playback levels and program material with modest low bass energy I can lift the woofer response to have F3 of 30Hz but at higher playback levels or with bass heavy music this has to be dialed back because the excursion cannot support it. At the same time, because of the higher sensitivity of the driver(s), I only need a couple of Watts in the upper passband (the woofer passband extends up to 500Hz in my system). Some OB users simply would use a 50W amp and drive it into clipping. Because only 50W are on tap, the amp cannot "overdrive" the woofer, but the amp will be clipping very often. This is not what I want to do. I want a higher power amp to stay out of clipping and for transients at higher frequencies where Xmax is not an issue. For music with little bass energy I can boost up the low end still enjoy (for example) the 30Hz F3 while staying within Xmax. But with the higher amp power on tap, there is the possibility to over driver the woofer, so my goal is to monitor the music signal, and "dial back" the low bass when the low bass content is too much for the drivers' Xmax to support. I want to do this is as blameless a fashion as possible, so I am using a lookahead limiter. Again, I am not trying to discriminate excursions to the sub mm level - an estimate to within a few mm is fine for this purpose. This is why I feel that the linear model is good enough.
The only thing that my limiter/compressor does is reduce the level in a sub-band. Essentially it is automatically riding the "bass tone control knob" of the system. The attack and release rate can be ramped up and down over as much time as the user desires due to the look-ahead nature of the algorithm.
I think that you are correct when you call this action compression. I see it as "preventing the excursion from exceeding some limit", which I personally think of as a limiter, but perhaps this is just semantics.
Last edited:
This assertion is not correct. Run your woofer at 50Wrms and measure for yourself how the coil resistance rises as its temperature increases over time. (As a word of warning do not run the test for too long as most drive units will not survive the test for even a couple of minutes despite their pretencious power ratings). But it would not be unsurprising to measure an increase in the order of a factor of 2.
To my knowledge Klippel's voltage driven mirror filtering does not model time variant thermal effects? Instead it appears to model just time invariant non-linearities such as suspension compliance and inductance. Your argument re small/large signal modelling mystifies me at this late hour since you appear to adopt a small signal model to compensate for a large signal measure?
Whilst thermal compression obviously compresses, that is, reduces the output, it appears to be to your advantage. But under such 'high' power working your 'sufficient' model is then throwing away up to HALF of its displacement by prematurely limiting. In such a case it also implies your limiter will be acting as a compressor given its likely extended period of operation. I would also not brush this off as semantics as the audibility of the processes (transient limiting and power compression) are quite different.
Most striking in your reply, however, is what appears to be an under-sized amplifier. Quite obviously 50W is insufficient if clipping occurs often as you state. No amount of tinkering obviates the need for proper system design and I would suggest your time would be better spent optimising this component rather than adding another compromise.
I think you need also consider whether your "50W" represents rms power or "peak power" and how often your limiter is working. If it is rms power then even a 20dB crest factor implies a 500W peak power requirement and a lot of 'limiting'. If it is "peak power" then you must have ridiculously sensitive drivers in very large baffles or listen at very low peak levels.
I also point out something I alluded to previously in this thread, namely that in my experience peak clipping is not particularly objectionable if done cleanly - it simply makes music sound subjectively dull and unlively (read not particularly life-like). I still have reservations therefore about the "blamelessness" of your limiter/system - if indeed blamelessness is a word (?).
FYI also, I am not envisaging maximally driving a closed box system. I too have an open baffle set-up having pursued this long before the current rage for such designs. I use current drive, however, which obviates the issues of thermal compression completely. I have a self-balancing bridge that serves for off-line model calibration and online system protection only. The system affords peak excursion limiting, rms power limiting (compression!) and also peak voltage limiting.
The system by its nature requires setting of the LF cut-off frequency. I rely on the thermal warning indicator (which is a precise indicator of voice coil temperature obtained via the bridge) to say if this requires attention. My impression is now that this is possibly what you are trying to modulate? If so then it is a frequency response limiter and not an excursion limiter (since outputs below the impinging frequency are also curtailed in such a design).
A further advantage of the bridge set-up is being able to drive high peak powers for transients without destroying the drivers. The LF amps have a peak power capacity of over 1kW but are rms limited to 100W (notably with high quality drivers known to sustain such continuous levels for over 24 hours). But the excursion limiting does not compensate for under-sized amplifiers. It never does.
Last edited:
You are so hung up on thermal compression it's pathologic.
If you want to go on about your bridge approach, why don't you start a thread about that instead of injecting it into this one?
It's fine that you don't seem to think that my approach has any real merit, but at the same time I don't get the feeling that you are really listening.
If you want to go on about your bridge approach, why don't you start a thread about that instead of injecting it into this one?
It's fine that you don't seem to think that my approach has any real merit, but at the same time I don't get the feeling that you are really listening.
Firstly, whilst this is not a scientific forum, it is on the whole technically based and personal quips aimed at my mental state are thoroughly out of place - and misguided. I am "injecting" my knowledge here because that is what forums are for and because I have a substantial amount of experience in this area. That you object to an objective response to your work is your choice but not one that is helpful in moving knowledge forward.
I am also not hung up on thermal compression: Its mention here is to point out the mechanism by which you are discarding as much as half the driver excursion you have available - and in an application that is excursion limited - at least where the power amplifiers have been appropriately designed. The bizarre result is that you may very well find your limiter being triggered when there was ample excursion available and hence no need to have the limiter in the first place.
FYI I have considered publishing details of the bridge arrangement I use - I may well do so in the future when I have the chance to document it fully. I have discussed it to some degree in another thread already where it emerged as the optimal way of driving a moving-coil loudspeaker. The details of Birt's original designs are available to anyone capable of using a search engine and thoroughly recommended to anyone with an open mind.
What I meant was that your focus on thermal compression seemed rather excessive. Again, it seems to me that you feel that applying some limiting (what I am doing) will create highly compressed program material and that I will drive the system as hard as possible. This is totally wrong and it would sound like crap that way. My approach, if anything, would REDUCE power to the driver, both peak and average. This is not the same as using high amounts of compression. I don't think that you are seeing this aspect of it.
Regarding your approach - current drive can minimize thermal compression but I always thought that this could be a dangerous positive feedback that continues to raise power input as VC temperature rises. One can not use that with abandon. Just like with my system, you have to apply it judiciously. I could also use an amp with feedback around the driver to increase output impedance (to reap some of the benefits of current drive) because the controller and amp+driver are independent for my system. This kind of feedback system is guaranteed stable, so this might be something I could add on later.
One problem I see with 100% current drive is the large near-resonance frequency response peak that will result. How do you compensate for that?
In any case, I was not able to find any info on the web about Birt's auto-balanced bridge and how that applied to loudspeakers, but I am curious about it. I asked if you could provide some links, but you did not. Perhaps there are some other search terms I can try (I am asking you to suggest them here). If you think your application of it is of merit, then please share it however you see fit.
C' mon Soundbloke, Be reasonable and fair please.
Charlie Laub has provided -both through his website and through diyaudio- much info plus a number of fairly advanced software tools. All this info is shared freely and non -commercially.
Now it is your turn to come up with something specific, solid and traceable. The Mysterious but Highly Talented mr. David Birt's Self Adjusting Bridge is nowhere to be found on the Internet. JAES?
Please provide us at least with a basic block diagram or something comparable or give specific links instead of just hints.
Waiting, anticipating...
Eelco de Bode
Charlie Laub has provided -both through his website and through diyaudio- much info plus a number of fairly advanced software tools. All this info is shared freely and non -commercially.
Now it is your turn to come up with something specific, solid and traceable. The Mysterious but Highly Talented mr. David Birt's Self Adjusting Bridge is nowhere to be found on the Internet. JAES?
Please provide us at least with a basic block diagram or something comparable or give specific links instead of just hints.
Waiting, anticipating...
Eelco de Bode
If your estimate of excursion is double that of what is occurring in reality then your limiter will trigger when it is not required. Given the low crest factor of much modern recorded music, your limiter could be operated sufficiently frequently that it will behave as a compressor. Given the equal loudness contours that you have also mentioned, any harmonic content generated by the limiter may also be more audible than you have envisaged.
No - it eliminates it completely by removing coil resistance from the transfer function.
There is no feedback to become positive unless motional feedback is applied - in which case a self-balancing bridge drives the motion sensing error to zero.
Either using the bridge-derived motional feedback as above or by using the bridge to calibrate an equaliser.
Whilst the former is better in principle, it is compromised by typically semi-inductive components appearing in the motional feedback signal. Previously I have cut slices through pole pieces to eliminate this issue but as their audible effects (very obvious in the mid band) are removed by current drive anyway, the latter, easier approach delivers many of the gains without the need to modify driver motors.
It is also possible to configure the electronics to transition from the motional feedback to current drive with increasing frequency but this again requires the customised drivers for best results. But in both cases the bridge is still available online to measure what the voice coil is doing.
UK patent applications GB2203609 and GB2234880 - although they may also be available with different numbers in other patent territories. I also recall at least one highly readable AES journal paper by the same author.
Last edited:
The patent application numbers you gave (thanx anyway) do not yield anything in Google.
A specific link maybe?
Someone else's I.P. for a non-traceable patent application?Frankly speaking I do not see that point: a patent application has to be awarded firstly before it becomes a patent.
Furthermore, It is certainly not forbidden to describe the workings of someone else I.P. One is not allowed to use/apply patent protected work/circuitry commercially, but any diy er is entirely free to use any patent protected circuit at will. Patents are supposed to seen and checked by all, that why e.g. patensonline does exist.
Eelco
A specific link maybe?
Someone else's I.P. for a non-traceable patent application?Frankly speaking I do not see that point: a patent application has to be awarded firstly before it becomes a patent.
Furthermore, It is certainly not forbidden to describe the workings of someone else I.P. One is not allowed to use/apply patent protected work/circuitry commercially, but any diy er is entirely free to use any patent protected circuit at will. Patents are supposed to seen and checked by all, that why e.g. patensonline does exist.
Eelco
I suppose this is the publication:
A Motion Transducer for Low-Frequency Loudspeakers
A simple motion transducer is described which provides an analogue of voice coil position, and a novel design feature yields good linearity over a useful span. Its output may be used directly for excursion limiting and digital correction of motor nonlinearites. Integration of the position output enables precise mean-position control in high compliance drivers. Differentiation of the position output provides velocity feedback.
Author: Birt, David
AES Convention:91 (October 1991) Paper Number:3196
Publication Date:October 1, 1991 Import into BibTeX
Subject:Loudspeakers
Permalink: AES E-Library A Motion Transducer for Low-Frequency Loudspeakers
A Motion Transducer for Low-Frequency Loudspeakers
A simple motion transducer is described which provides an analogue of voice coil position, and a novel design feature yields good linearity over a useful span. Its output may be used directly for excursion limiting and digital correction of motor nonlinearites. Integration of the position output enables precise mean-position control in high compliance drivers. Differentiation of the position output provides velocity feedback.
Author: Birt, David
AES Convention:91 (October 1991) Paper Number:3196
Publication Date:October 1, 1991 Import into BibTeX
Subject:Loudspeakers
Permalink: AES E-Library A Motion Transducer for Low-Frequency Loudspeakers
AES E-Library
Nonlinearities in Moving-Coil Loudspeakers with Overhung Voice Coils
A technique is described which enables profiles of transduction coefficient B1 versus cost position to be plotted for an assembled loudspeaker, with voice coil current as a parameter. The performance of a system comprising a prototype loudspeaker with extended center pole and compensating winding, combined with digital forward error correction is described. This system simulates a linear motor with very low distortion.
Author: Birt, David R.
JAES Volume 39 Issue 4 pp. 219-231; April 1991
Publication Date:April 1, 1991 Import into BibTeX
Permalink: AES E-Library Nonlinearities in Moving-Coil Loudspeakers with Overhung Voice Coils
Nonlinearities in Moving-Coil Loudspeakers with Overhung Voice Coils
A technique is described which enables profiles of transduction coefficient B1 versus cost position to be plotted for an assembled loudspeaker, with voice coil current as a parameter. The performance of a system comprising a prototype loudspeaker with extended center pole and compensating winding, combined with digital forward error correction is described. This system simulates a linear motor with very low distortion.
Author: Birt, David R.
JAES Volume 39 Issue 4 pp. 219-231; April 1991
Publication Date:April 1, 1991 Import into BibTeX
Permalink: AES E-Library Nonlinearities in Moving-Coil Loudspeakers with Overhung Voice Coils
The patents can be found here:
https://worldwide.espacenet.com/pub...ES&FT=D&date=19881019&CC=GB&NR=2203609A&KC=A#
https://worldwide.espacenet.com/pub...ES&FT=D&date=19910213&CC=GB&NR=2234880A&KC=A#
The AES publication and paper above is not related to the patents here - but it is still a good read nonetheless...
https://worldwide.espacenet.com/pub...ES&FT=D&date=19881019&CC=GB&NR=2203609A&KC=A#
https://worldwide.espacenet.com/pub...ES&FT=D&date=19910213&CC=GB&NR=2234880A&KC=A#
The AES publication and paper above is not related to the patents here - but it is still a good read nonetheless...
I would also add that for anyone considering the self-balancing bridge, under-hung geometries and high suspension linearity make the implementation very much easier. Quite possibly these factors and the issues with third order magnetic non-linearities are why the system is not in widespread use. Quite possibly those considerations prompted the work presented in the AES paper?
Again, you keep belaboring this point about thermal compression. Sure, under certain circumstances of highly compressed program source and hard driving of the speaker you can get thermal compression. But I could just as well argue that this is not a PA system, so severe thermal compression is really not of serious concern. Is thermal compression not be happening in every loudspeaker driven by a voltage source (your usual low output impedance solid state amp)? If it is such a problem, why do we not hear about it more? Finally, if you compensate one driver (e.g. the woofer) for its thermal compression using current drive, what about the other drivers in the systme (e.g. the midrange)? Won't the mid also need similar compensation? If not, it will have decreased output from its own thermal compression when the woofer under current drive will not.
I do see your point that the estimate will be in error if significant thermal compression occurs, but with my approach the user is certainly free to use voltage or current drive as they see fit - the two are independent.
Also, there is the approach where the voice coil DC resistance is monitored in real time. There was an interesting Sterophile article about this in 2006 that you can read HERE. It talks about current drive and voice coil heating, and investigated just how much VC heating occurred. It's an interesting read. The Re data that this kind of monitoring would generate can be used to update the TS parameter model of the driver in my DSP excursion limiter code. This could be used to eliminate the thermal compression effects by increasing the gain without having to resort to current drive and all of the complexities and problems that come with it.
This is only true with pure current drive, which causes large frequency response changes around resonance because only the mechanical component of the resonance remains, and the associated Q is typically quite high. Because the resonance behavior arises from the mechanical system, and that is cone-position dependent, it seems to me that you will get a high Q peak that is modulated by cone movement. It doesn't seem to be very feasible to EQ this out.
I'm not aware of a non-feedback current drive amplifier for a loudspeaker (maybe there is one?). The "balanced bridge" is providing feedback, and even if that is negative feedback it's feedback all the same.
I was able to read the Birt paper in JAES in which he describes the auto-balancing bridge concept. Some of the principles he uses are not all that different than the ones used in that Stereophile article I linked to, above. Birt's bridge is just a way to probe the driver's impedance, and the integrator is acting like a filter, passing only the lowest frequencies. The Stereophile article also uses a current source, capacitive coupling, but uses a 6th order LP filter to remove the audio signal and leave the "DC" component (below a very low corner frequency). I think that both approaches are interesting, and have their pluses and minuses. Also, it is of note that Birt mentions that using a small DC offset in the power amplifier instead of coupling a current source into the AC signal is a convenient way to provide DC sense current.
No matter how one chooses to sense Re, the resulting value can be used to change the amplifier transfer function (I'm intentionally using that term to be general) or adjust values of other circuitry. For example, Stahl's ACE-BASS circuit needs to cancel Re in order to work well. A few years ago I considered constructing a circuit to monitor the driver and adjust a component in Stahl's circuit so that the negative amplifier output impedance would closely track the instantaneous Re value. This kind of thing has been on my mind for awhile. In the end I wasn't really sure that the added complexity was necessary, primarily because of the measurements presented in the Stereophile article.
With the advent of small PIC/ARM platforms that have analog inputs, it would be interesting to see how they can be used. For example, if I can obtain the value of Re from some external signal, I can adjust the DSP processing to maintain constant SPL and to correct for changes to the frequency and Q of the driver resonance. Instead of using expensive multistage analog circuits, high-order low-pass filtering could be done in software to produce a relatively accurate measure of the DC voltage across a sense resistor in series with the driver.
No matter how one chooses to sense Re, the resulting value can be used to change the amplifier transfer function (I'm intentionally using that term to be general) or adjust values of other circuitry. For example, Stahl's ACE-BASS circuit needs to cancel Re in order to work well. A few years ago I considered constructing a circuit to monitor the driver and adjust a component in Stahl's circuit so that the negative amplifier output impedance would closely track the instantaneous Re value. This kind of thing has been on my mind for awhile. In the end I wasn't really sure that the added complexity was necessary, primarily because of the measurements presented in the Stereophile article.
With the advent of small PIC/ARM platforms that have analog inputs, it would be interesting to see how they can be used. For example, if I can obtain the value of Re from some external signal, I can adjust the DSP processing to maintain constant SPL and to correct for changes to the frequency and Q of the driver resonance. Instead of using expensive multistage analog circuits, high-order low-pass filtering could be done in software to produce a relatively accurate measure of the DC voltage across a sense resistor in series with the driver.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.