Motional feedback: dual voice coil signal

Status
Not open for further replies.
A lot of that analysis would lead a builder to favour the bridge approach. In theory, the bridge balances those various factors - such as coil inductance and maybe even the factor so beloved on this forum, heat compression - and only the pure back-EMF remains. In practice, I got great results just doing resistive balancing, just guessing how far I could squeeze down to a pure signal and how much of that signal I could feed back.

MFB to reduce distortion in a LF driver (e.g. subwoofer) has always piqued my interest. Since you mentioned it, I have done some thinking about the bridge approach.

I have always wondered why only a simple RC or RL network was used in the bridge leg that is opposite the driver. Why not create a network that is more "driver-like", e.g. the electrical equivalent of a driver in a box as seen by the amplifier. For instance, SL derives just such a network for a subwoofer in the section titled "2 - Driver and box model" on this page:
Subwoofer design
This network (or the simplified version below it) can be connected in series with a resistor (just like the driver and sense resistor) to better "model" what the desired driver behavior should be. One would want to impedance scale all the components so that the power consumption is negligible. It seems one could configure this "reference" network to have whatever 2nd order rolloff behavior (Assuming closed box loading) and the bridge signal would give the electrical "correction" that is needed to make it so.

Am I missing something in the above line of thinking?


.
 
This model is linearized i.e. if the cone does not actually move linearly you don't directly sense it.

I'm 100% clear on what you are saying... "this model" I assume is the one from SL's web site, right? I can see how it is a "linearized" small-signal based model. But isn't that what we would like our driver to be, to behave like? Wouldn't a linearized model be a perfect reference for a distorion-less driver?

Can you expand/clarify this a bit?
 
I'm 100% clear on what you are saying... "this model" I assume is the one from SL's web site, right? I can see how it is a "linearized" small-signal based model. But isn't that what we would like our driver to be, to behave like? Wouldn't a linearized model be a perfect reference for a distorion-less driver?

Can you expand/clarify this a bit?

The problem is the model doesn't care if it needs 2" of excursion or heats the voice coil to 200C to get a desired SPL. I would assume all else being equal a speaker under small signal conditions has much less distortion than at high SPL, the model does not know that. Motional feedback involves some form of sensing the actual motion/position of the cone. SL's stuff is very nice it shows you how position and velocity are related and you can see how one needs to design around the fact that you can actually measure the cone position or infer it by history of velocity and/or acceleration.
 
The problem is the model doesn't care if it needs 2" of excursion or heats the voice coil to 200C to get a desired SPL. I would assume all else being equal a speaker under small signal conditions has much less distortion than at high SPL, the model does not know that. Motional feedback involves some form of sensing the actual motion/position of the cone. SL's stuff is very nice it shows you how position and velocity are related and you can see how one needs to design around the fact that you can actually measure the cone position or infer it by history of velocity and/or acceleration.

I don't see any problem here. If, as you say:
the model doesn't care if it needs 2" of excursion or heats the voice coil to 200C to get a desired SPL.
that is not surprising. I would certainly place some responsibility on the designer/implementer to make sure that the system is not running away when excursion (and distortion, compression, etc.) are high.

Let's forget I used the term "motional feedback", since this seems to imply something to you that might not really be the system I am describing. What I am envisioning is a bridge - one leg has the loudspeaker. The corresponding leg of the other branch has the passive network that is a "model" of a loudspeaker in a box comprised of passive elements. Of course the "model" network has no distortion - that is exactly what we want, no? The difference between the two, as returned by a differential signal taken across the bridge, will be distortion (assuming the model is not also describing a different 2nd order HP function than the driver-in-box forms). By "distortion" I mean any behavior/movement that is non-linear. By changing the model network to represent some other 2nd order highpass function, an additional "correction" signal should be generated across the bridge that would force the loudspeaker (as much as its motor permits) to assume the model network transfer function. At least this is how I envision it...

You feed back the signal taken across the bridge to the input. This should push the system towards the (distortionless) model.
 
Granted, even a cheap accelerometer of a few decades ago is better than the 10% distortion and miserable transient behaviour of even high quality shaking-cardboard drivers.

Feedback is not a metaphysical concept; it is a matter of having phase right across the entire relevant passband.
Ben,

High quality high Xmax sub drivers (even in relatively small boxes) have harmonic distortion as low as 1% at levels you are like likely to listen at.

Using FIR filters with your miniDSP, it is possible to correct the phase response of a loudspeaker (or subwoofer) to be flat over it's entire frequency range, without needing motional feedback.

http://www.diyaudio.com/forums/mult...hase-linearization-eq-fir-filtering-tool.html

Although you could achieve even better performance using a motional feedback system (the Powersoft/B&C IPAL system again comes to mind) you can achieve around 80% of the results you are after with about 20% (or less) of the cost of such a system.

Art
 
Last edited:
Here is an approach of the problem by Daniel Ferguson (articles published in AudioXpress between 2003 and 2005) :
He used two dual voice drivers, linked à la Isobarik, that meaning cone facing cone with a very small gap between them.
Only one voice coil of one driver was voltage driven.
He compared the signal induced in the second voice coil of the voltage driven loudspeaker with the signal of one of the voice coils of the second driver.

Interesting 8" products with a specialised sensing coil :
Rythmik Audio 8" servo driver• DS800 Direct Servo driver
 
Some authors (Werner, 1957, and others) thinks as CharlieLaub thinks, it's best to balance the bridge as realistically as possible. But not all that important. Bigger fish to fry such as controlling the passband to keep within a stable region.

Since from an engineering point of view, a Rice-Kellogg sub driver is a big compromise (for want of an expletive) which is asked to play outside its proper range. So using feedback to control it is always a stretch.

As proof, how many have been built? What about the thread two years ago where everybody jumped up and down and we are still waiting for the miniDSP MF to be finished.

I hope one of the Philips enthusiasts will post some data for us. My old R&D in 1968 found storage-scope snaps of blips to speak the loudest.

Welter sys, the issue isn't getting output sound in phase but grabbing a feedback signal that is close enough in phase to the pre-amp signal to be useful as feedback.

Ben
 
I have a pair of these Sony HA-7900 MFB speakers.
Their bass sound is quite decent, especially the sense of 'control' and damping in the bass.
The schematic might give some ideas.
Sony - HA-7900 pic.JPG
Sony - SA-H7900 Schematic.png

Dan.
 
You feed back the signal taken across the bridge to the input. This should push the system towards the (distortionless) model.

I think there are two different issues here, one is getting flat response and the compressive distortion of the driver at high SPL. Since you can't get at the nodes inside the speaker you only have the terminal voltage and series current. If you're talking about a bridge I would think that you would not want to sample the current with anything but a very small resistor in series with the speaker. Do you have a reference showing that the radiated third harmonic distortion shows up as current at the terminals?

If you simply want nominally flat phase compensated response the DSP model possibly makes more sense. A physical model full of L's and C's would have to match each speaker and be a pain to build and if they are constant linear components why not just compute their response? Folks need to outline some solutions and propose exactly how they are doing what they are supposed to do.
 
Last edited:
I think there are two different issues here, one is getting flat response and the compressive distortion of the driver at high SPL. Since you can't get at the nodes inside the speaker you only have the terminal voltage and series current. If you're talking about a bridge I would think that you would not want to sample the current with anything but a very small resistor in series with the speaker. Do you have a reference showing that the radiated third harmonic distortion shows up as current at the terminals?

If you simply want nominally flat phase compensated response the DSP model possibly makes more sense. A physical model full of L's and C's would have to match each speaker and be a pain to build and if they are constant linear components why not just compute their response? Folks need to outline some solutions and propose exactly how they are doing what they are supposed to do.

While I see how you seem to break down the behavior into two issues, frequency response and high-excursion performance, I don't really see this as the heart of the matter. What is at issue is using feedback of some kind, around the driver, to improve the performance. The other things that you mention are simply benefits of that approach.

You can measure (directly) the input voltage, and can (with a series component, e.g. resistor) measure the current passing through the voice coil (including back emf, etc.). I don't see why these can't sense the difference between what the amplifier's signal is telling the driver what to do and what the driver is actually doing under the influence of that signal given a model of the driver's electrical equivalent (including acoustic side).

A balanced bridge is used to do this and of course you use a small value resistor in series to sense current. The driver voice coil and this sense resistor form one leg of the bridge and (in my promulgated configuration) an electrical model of the driver (sure passive or active would be OK) and a sense resistor form the other leg of the bridge. The idea is to feed the difference between the "model" leg and the "loudspeaker" leg back to the input to correct any error. This can help force the loudspeaker to behave more like the "model" (the desired behavior) depending on the amount of feedback used.

If this approach is so inept, then why in Werner's work back in the late 1950's could distortion be reduced from over 20% to almost 2% using feedback from a bridge network? For example see starting on page 193 (page 13 of the pdf):
http://www.mif.pg.gda.pl/homepages/...onics_1958/1958_11_AWV_Radiotronics_23_11.pdf

For an example of the bridge I am describing see Figure 2 on page 7 of this document:
http://aerosmart.umd.edu/TechPubs/pratt_1.pdf
 
Last edited:
When I went for my job interview at Bell Labs, I talked about Werner's article. Must have worked.

Actually, just using a small series resistor (like a fraction of an Ohm) is all the "bridge" you need.

A key obstacle, as I see it, is there's no good way I know of to quantify transient behaviour. Regardless of abstract arguments about FR, transient goodness isn't revealed except by pictures. I hope somebody will prove me wrong.

Ben
 
Again, getting a gallon of sound out of a little speaker. Great idea and applicable to DIYAudio folks too. But for me, it is going low and punch.

B.
I know the former lead-designer of the Philips MFB has plans to design a SUB using MFB. I though do not know the status of that project.

On the down-side, mainly the original Philips speakers are used, but as I mentioned earlier there are also transducers available, designed by enthousiast and out-performing the original transduces. They actually are dirt-cheap... no need for expensive and rather heavy commercially available accelerometers, which do not perfom that well in these kind of applications, due to their low resonant frequencies.
 
If this approach is so inept, then why in Werner's work back in the late 1950's could distortion be reduced from over 20% to almost 2% using feedback from a bridge network?

It looks like much of the distortion reduction is caused by the reduction in excursion at resonance (changing the speaker from heavily underdamped to "properly" damped)

Don't negative output impedance amps have stability concerns? If you are going to generate a feedback signal, why not generate one from something that actually senses cone motion? Of course, MFB has stability concerns as well 😉
 
Don't negative output impedance amps have stability concerns?

The main concern seems to be the inductive compoment of the voice coil entering in resonance with the capactive component of the motional impedance.
They are usually labelled Le and Ces in simulation.
Le has not a constant value, it varies with both the frequency and the position of the voice coil in the gap, so one needs to be conservative when setting compensation for stability in systems using negative impedance amps.
 
The main concern seems to be the inductive compoment of the voice coil entering in resonance with the capactive component of the motional impedance.
They are usually labelled Le and Ces in simulation.
Le has not a constant value, it varies with both the frequency and the position of the voice coil in the gap, so one needs to be conservative when setting compensation for stability in systems using negative impedance amps.

Aren't these small enough to be ignored? Or no real obstacle to testing?

Anybody ready to breadboard or is this all talk-talk?

Previously, I complained I didn't know how to measure transient behaviour. Likewise, there're problems with how to create test conditions. I mean we need stimuli that uniformly cause errors that then need fixing. Long ago, I used spikes (which aren't really well defined since we all know they aren't infinitely loud, brief, or wideband). Also used 4-cycle tone bursts (which are precisely defined and can be fairly precisely delivered to a speaker) in order to eyeball onset an offset transient behaviour.

Ben
 
Status
Not open for further replies.