The problem boils down to matching the overall transfer function from input to output which is the product of the transfer functions of each circuit block. If there is feedback wrapped around one block which results in effective negative output impedance from the amplifier inside that block, it does not directly define the input-to-output behavior. That is defined by the transfer function of the complete circuit block the feedback loop is wrapped around. That transfer function can be emulated with an additional upstream circuit block (ie EQ) if desired.
Two other benefits/uses of MFB that haven’t been mentioned much are discussed at length in the Rythmik patent. One is that you can use a combination of VC and current feedback to adjust the effective woofer moving mass, suspension compliance, and damping. The other is that the target alignment is fairly insensitive to changes in box size and T/S parameters. With EQ alone, if anything changes you would need to re-adjust.
Sadly most if not all of my non-audiophile friends consider any subwoofer larger than about 1 ft^3 a ridiculous monstrosity. Conversion is a slow process
Last edited:
That's an abstract argument.
Does your all-knowing transfer function model errors "from input to output" and everywhere in-between?
Umm, didn't think so.
B.
Last edited:
I thought that this matching was the function of the feedback-loop in a MF system... provided enough gain available.
Edwin
Now tucked away, in this near final paragraph of a series of posts critical of MF thinking, bolserst clarifies what MF is all about. It is the feedback circuit that determines the behaviour of the speaker rather than some starched fabric and a rubber roll.
Glad you said "effective". Obviously nobody is changing the T/S parameters of the driver mechanism (or has for 100 years). But with MF, the speaker acts "fast" that is, acts like a big magnet driving a very hight cone with a flexible suspension*. What do all those hornresp simulations to three decimal places mean now?
Ben
*Could you build a physical sub with those MF parameters?
Last edited:
This is a transform with all adjutable parameters (initial values, fo, Qo; target values fp, Qp; as orignally choosen by Linkwitz for his transform) :
http://jipihorn.files.wordpress.com/2013/06/linkwitz-variable-state.pdf
Author explaining it in french :
https://www.youtube.com/watch?v=1B2jf2s1B2M
I built a prototype to evaluate the noise, it is not a problem.
An idea I had - it seems I was not the first to have it - is to have the high-pass transfer function of the driver in closed box (fo Qo) in the negative feedback circuit of an op-amp.
The frequency response is then flat. This means that gain is increasing at a 12 dB/o rate at low frequencies.
Usually there is a high-pass filter (fp, Qp) elsewhere in the circuit to avoid too large cone displacements.
However, low frequency signals being usually not of very high level, we could discard the high-pass filter (fp, Qp) and replace it with a voltage limiter to avoid the excessive cone displacements.
The system would then be aperiodic across the audio band.
Not yet tested by me.
The T/S parameters can't be considered alone, they always belong to a closed circuit. And it is the behaviour of the whole circuit which determines the transfer function. Erik Stahl made a brilliant demonstration of this.
A very good idea. A simpler version of that thinking is modelling a VC Wheatstone Bridge MF so as to counter-act the parameters of the driver (albeit only the basic electrical parameters).
While the balanced bridges approach has a certain abstract theoretical appeal, I can't imagine it makes much difference to the 2.5 octaves covered by most subs. I experimented with bridge balancing (when a bridge method was easy to do with tubes and transformers) and didn't think it helped. Except theoretically.
Today, you can just use EQ to fix it, as forr indicates.
Ben
[Originally Posted by weltersys
Ben,
1. The IPAL is an adaptive system, it follows the designer's intended frequency response regardless of surroundings or drive level.
2. The math model of the box is not needed to program the DSP, but accurate measurements at various drive levels and voice coil heating are.
3. The IPAL microprocessors certainly are a good bit faster than the much older (and less expensive) Behringer 2496. That said, the latency inherent in any DSP can be virtually eliminated by delaying the pick up transducer by the same amount as the DSP latency.]
Around 20 years ago I purchased an Alesis Masterlink hard drive recorder/CD burner.
Among it's many (new to me at that time) features was a "look forward" limiter, a limiter that could detect peak or RMS levels "before they happen" and thus provide any degree of limiting desired with no "pumping and breathing" associated with fast attack/release high compression ratio analog limiters. Since the digital domain has a fixed hard headroom limit, limiting signal to fit it is much more critical than in the analog domain, where overload often simply results in euphonic distortion- commonly used digital "plug ins" emulate tape saturation, tube (valve) distortion, and tape "wow and flutter".
In the real world, a "look forward" limiter is not (strictly speaking) possible in the analog domain, but is easily accomplished in the digital domain by slightly delaying the output signal compared to a copy of the same that the comparator circuit (in the case of a limiter, the "threshold") "sees". The same principle can be applied to a noise gate (a more adaptive version of a "squelch" control) to allow the gate to open without clipping off the leading transient containing amplitude above the gate threshold, or in the case of Motional Feedback, the pick up sensor's signal compared to the feedback control circuits.
My above point #3 violates no feedback theories, but a "zero latency" feedback loop requires a short (up to a few millisecond) delay applied to compensate for the latency inherent in the A/D (analog to digital) converters. A high speed, close head spacing analog tape loop could also be used for a "look forward" circuit, but would now cost many times more than the digital delay available in cheap DSP.
Having paid tens of thousands of dollars back in the day on analog and early digital delay equipment that would fill 21" of rack space that can now be accomplished in a single rack space for a few hundred dollars truly makes me realize how much things have changed in the world of electronics over the last 30 (or so) years.
Although I still cling to many of my old analog devices, I recognize the truly amazing (and impossible in the analog domain) potential that digital architecture (I only recently learned to use BSS's "London Architect") lends to DSP.
At any rate, an accurate microphone coupled to the right digital architecture, software, class D amplification, and a good loudspeaker can make MF do pretty much anything you want it to do, and do it with latency only equivalent to less than two feet time of flight at the speed of sound.
That little delay applied to the complete audio spectrum makes no audible difference in live sound (other than for in-ear monitoring) and no difference at all for playback systems.
Cheers,
Art
Ben,
1. The IPAL is an adaptive system, it follows the designer's intended frequency response regardless of surroundings or drive level.
2. The math model of the box is not needed to program the DSP, but accurate measurements at various drive levels and voice coil heating are.
3. The IPAL microprocessors certainly are a good bit faster than the much older (and less expensive) Behringer 2496. That said, the latency inherent in any DSP can be virtually eliminated by delaying the pick up transducer by the same amount as the DSP latency.]
Ben,
Around 20 years ago I purchased an Alesis Masterlink hard drive recorder/CD burner.
Among it's many (new to me at that time) features was a "look forward" limiter, a limiter that could detect peak or RMS levels "before they happen" and thus provide any degree of limiting desired with no "pumping and breathing" associated with fast attack/release high compression ratio analog limiters. Since the digital domain has a fixed hard headroom limit, limiting signal to fit it is much more critical than in the analog domain, where overload often simply results in euphonic distortion- commonly used digital "plug ins" emulate tape saturation, tube (valve) distortion, and tape "wow and flutter".
In the real world, a "look forward" limiter is not (strictly speaking) possible in the analog domain, but is easily accomplished in the digital domain by slightly delaying the output signal compared to a copy of the same that the comparator circuit (in the case of a limiter, the "threshold") "sees". The same principle can be applied to a noise gate (a more adaptive version of a "squelch" control) to allow the gate to open without clipping off the leading transient containing amplitude above the gate threshold, or in the case of Motional Feedback, the pick up sensor's signal compared to the feedback control circuits.
My above point #3 violates no feedback theories, but a "zero latency" feedback loop requires a short (up to a few millisecond) delay applied to compensate for the latency inherent in the A/D (analog to digital) converters. A high speed, close head spacing analog tape loop could also be used for a "look forward" circuit, but would now cost many times more than the digital delay available in cheap DSP.
Having paid tens of thousands of dollars back in the day on analog and early digital delay equipment that would fill 21" of rack space that can now be accomplished in a single rack space for a few hundred dollars truly makes me realize how much things have changed in the world of electronics over the last 30 (or so) years.
Although I still cling to many of my old analog devices, I recognize the truly amazing (and impossible in the analog domain) potential that digital architecture (I only recently learned to use BSS's "London Architect") lends to DSP.
At any rate, an accurate microphone coupled to the right digital architecture, software, class D amplification, and a good loudspeaker can make MF do pretty much anything you want it to do, and do it with latency only equivalent to less than two feet time of flight at the speed of sound.
That little delay applied to the complete audio spectrum makes no audible difference in live sound (other than for in-ear monitoring) and no difference at all for playback systems.
Cheers,
Art
Last edited:
Yeah, the suspension is part of the system but let's not forget that big motor that's bolted onto the back of the driver. The motor force is usually more dominant than the suspension (or at least equally important, as the two are parts in a system) in controlling the driver behavior, at least until you push it past the suspension limits.
Also MF is not "all about" changing effective driver t/s. This is a secondary benefit that is rarely even mentioned, and not really even all that useful in the grand scheme, as I will reveal in a minute ...
Well, I'm glad you put that word in quotes. The next step is to stop using that word completely, because "fast" is not an issue here and MF does not make drivers faster.
Good job, it was a stretch to turn this discussion into an attack on simulators but you managed to do it.
Simulators allow you to understand the system. If you don't understand the system you can't apply MF. You clearly don't understand the system if you were trying to apply a simple MF circuit to a Karlson and a Klipschhorn. While it COULD be done to good effect, it can't be done by YOU because you can't quantify the behavior of these speakers because you refuse to use a simulator or measure in a useful way. You also seem to be under the impression that only ported boxes and tapped horns have resonances - but the truth is that ALL speakers have them, and your old Klipschhorn and Karlson had plenty.
Simulators help you to create systems that make sense, and also allow you to create systems that are a bit more complex than a simple sealed box (like a Karlson or a Klipschhorn, or modern better equivalents).
And let's not forget that in this modern world we can access the internet, choose from thousands of different woofers from 2 inches to 40 inches diameter, order them delivered to your door without ever getting out of bed or putting your pants on. This variety of different products have such an incredibly wide and diverse set of characteristics that you can choose whatever product you need for any imaginable project.
You can choose woofers that work well in boxes barely big enough to fit the driver or you can choose woofers that need tremendously huge enclosures to work well. (A simulator is a handy tool to help you figure out which is which, by the way, so you don't need to guess and use rules of thumb like "resonances are evil" and "fs should be below the passband" and "MF can fix anything and make a driver faster". Those rules of thumb are incredibly naive and narrow your choices down to almost nothing.)
With such a diverse set of products available, products that have virtually every t/s parameter combination you could ever want, there IS an existing product (or several existing products) that have the t/s parameters that will give a good result in ANY given situation.
So what is the point of manipulating t/s with MF? It's simply not necessary.
Even if you are dead set on using a driver with poor t/s for the intended use, if you have a simulator you can at least get the most out of a bad situation, which is still better than naively instigating a sub optimal MF circuit because you don't really know how MF works (as proven with your Karlson and Klipschhorn adventures).
Last edited:
Art, last time I checked time-travel was still impossible, not even for a pico-second. I just finished laughing and wetting my pants in the process..
This is utter nonsense!
You changed the subject from distortion reduction to bass extension and transient response. My comments should be viewed in that context. If your woofer distortion is not so great as to cause you concern, then any benefits to bass extension or transient response effected by MFB can be performed equally well by EQ circuit blocks. This is not critical of MFB in general, just clarifying that response flattening and resulting improvement to transient response are possible with design techniques other than MFB.
Of course not, I don’t think anybody suggested otherwise. See comments above.
Absolutely. If you use feedback to manipulate the effective T/S parameters, you can use those effective parameters as inputs to any box modeling program to determine optimum box design. That is exactly what the ACE-bass and IPAL systems (and to some extent Rythmik) are all about.
Last edited:
Your idea is similar in concept to the ELF(Extended Low Frequency) technique used in Bag End subwoofers. They use two integrators to effect the 12dB/oct boost down to 5hz or so. Then, optical feedback is used to adjust the LF knee of the integrators based on how loud the low bass signal is.
Patent US 4,481,662 attached.
Last edited:
Esgigt,
Read the rest of post #123, the simple digital "magic" explains how what you call "time travel" is not "utter nonsense", but quite common in digital equipment in use for over two decades of time.
Perhaps you should make baby steps into the realm of what is possible in the digital world of this century, rather than scoffing those advances.
Or not, you could be like the Native Americans who perceived the ships bearing a "new civilization" only as clouds on the horizon.
Art
Yes, exactly the point. Same mag/phase transfer function == same output. Except for the "fine print" and the overload behavior, as noted by me in post #83: http://www.diyaudio.com/forums/subw...back-woofer-available-sort-2.html#post4818907
I might add a benefit of MF is no performance change from parameter drifts. Like for example, with enough feedback factor the cone movement remain the same no matter what medium we operate in, air, helium, vacuum. Or a broken spider, you won't notice it until the coil starts to rub...
Ben doesn't like simulators, doesn't understand what they do or what they are for, and mocks people that use them. The extent of his subwoofer design routine is "put it in a big leaky sealed box", but he doesn't understand what the resulting response will be even then, as shown by his "all drivers will have a whomp up at fs" in this type of box. When he was told this was only the case with very high q drivers he asked for a professional consensus on the matter. No matter what type of evidence is given, Ben's argument will always be a spiraling attack on moving coil drivers (everything that isn't ESL), simulators (and the people that use them), the evils of resonance and the superiority of MF. This is why I get a bit blunt in these discussions.
For those wanting to try a MF control loop embodiment in the digital domain (and with class-D amps), Axign make a chip perfect for this kind of application.
Axign | Digital Feedback Loops
--------:--------
In the end it all comes down to the sensor design (together with driver design, of course) and a clever way to handle overload without destroying the driver and introducing a controlled distortion characteristic when the driver's limits are approached.
A way to handle this is building the required nonlinearity into the sensor, works fine with a capacitive RF demodulating position sensor (steal the idea from the Sennheiser RF condensor mics) as the linearity of those can be precisly shaped by their geometry.
Axign | Digital Feedback Loops
--------:--------
In the end it all comes down to the sensor design (together with driver design, of course) and a clever way to handle overload without destroying the driver and introducing a controlled distortion characteristic when the driver's limits are approached.
A way to handle this is building the required nonlinearity into the sensor, works fine with a capacitive RF demodulating position sensor (steal the idea from the Sennheiser RF condensor mics) as the linearity of those can be precisly shaped by their geometry.
All good thoughts (again) from KSTR. But there is an important bone of contention showing up clearly now in this thread including posts of bolserst and others.
What is MF for? What are the biggest benefits?
Perhaps these can be divided into three classes:
1. distortion due to non-linearities of the driver and box behaviour
2. as KSTR offers, amelioration from extreme states of the speaker
3. or (for me most important), controlling the boom excursions and pushing the bass action of the driver lower (these might be described as gross cone motion in an otherwise pretty good sub).
I respectfully disagree with KSTR. When a feedback system is pushed outside its normal performance limits, there is no way to tell how it will behave (that means most often self-destructions). Therefore, hoping MF will operate in extreme excursions or other extremes is not easily accomplished.
Ben
As a few people have mentioned, if you use modern high quality drivers, this is not an issue until you try to exceed the mechanical and thermal linear region of the driver. Deal with that however you like - MF or more drivers. The enclosure shouldn't have any problems that MF could be used to address if it's well designed.
Unless the speaker is not designed well or broken or driven well past it's linear safe zone there should never be any extreme states of anything.
Most people deal with this quite effectively (and inexpensively) with good enclosure design (that matches the speaker's response to the room gain curve) and with a bit of eq.
Eq does the same thing unless you are pushing the driver well past it's safe zone. Regardless of what you think, the driver and the frequency response do not go off the rails out of control at resonant frequencies (and the resonant frequency of a sealed box is pretty high anyway, not at the bottom of the passband).
If you want to use MF as a glorified eq, that's up to you, but it would help to simulate the system first to quantify it's behavior so you can properly apply the MF.
Of course the system must always remain in a closed-loop state.
Adjusting the system so that one really gets close to fully exploit the driver limits is the hard part of building a MF system.
At the time I tried MF I didn't use drivers with sufficient x_damage spec (no bottoming and/or suspension collapsing even when the coil is fully outside of the cap, so I wrecked a few).
I've quit my MF adventures since, as there are better options today :
Better woofers, and if you want to push them to "99%" use devices like the Klippel "Controlled Sound" processing.
https://www.klippel.de/our-products/controlled-sound.html
After years of research and development the first units are about to ship, finally.
Klippel Company is world leader in stuff like this (any person dealing with driver development will know them) so you can be 100% sure it is working and it might beat simple MF technology since the monitioring and protection features are quite elaborated.
Adjusting the system so that one really gets close to fully exploit the driver limits is the hard part of building a MF system.
At the time I tried MF I didn't use drivers with sufficient x_damage spec (no bottoming and/or suspension collapsing even when the coil is fully outside of the cap, so I wrecked a few).
I've quit my MF adventures since, as there are better options today :
Better woofers, and if you want to push them to "99%" use devices like the Klippel "Controlled Sound" processing.
https://www.klippel.de/our-products/controlled-sound.html
After years of research and development the first units are about to ship, finally.
Klippel Company is world leader in stuff like this (any person dealing with driver development will know them) so you can be 100% sure it is working and it might beat simple MF technology since the monitioring and protection features are quite elaborated.
Studying the history of audio engineering, it is clear that MF is the final frontier, not faith in starched fabric spiders. Just plain ridiculous to say, "... who needs MF.... aren't we already perfect?".
I like KSTR's advocacy of the Klippel system. Who doesn't admire Klippel? The Klippel MF system, as I understand the link KSTR provided, illustrates two important issues:
1. no problem using just a plain simple series resistor (a VC method) as the sole sensor of feedback (as I did for decades), and
2. almost regardless of speaker characteristics, it is the feedback circuit which determines the performance of the system. Klippel introduce a fancy correction DSP into the feedback loop to correct any darn thing they feel like correcting. That sure might be what IPAL are secretly doing. In other words, as in all feedback systems, it is the model within the feedback loop that defines the system performance (at least to the degree of feedback factor).
A look at the Klippel flow chart will show why there's no box for a future-looking time-travel in this feedback network to compensate for DSP delays. Yes, what used to be called "delayed" sound is possible but seems impossible for MF. Tempting notion. Anyways, hard enough to get HD-FM to synch with analog broadcast signals.
Ben
Last edited: