Hi,
I was wondering why is it that servo units aren't used more often within Subwoofers? The design seems like a good idea to me but I was curious why they're not used more readily?
What are the downsides of using one? Would using a servo mechanism on another woofer used to control the pressure within the cabinet in order to aid the main woofer avoid these downsides if there are any?
Cheers!
Mike
I was wondering why is it that servo units aren't used more often within Subwoofers? The design seems like a good idea to me but I was curious why they're not used more readily?
What are the downsides of using one? Would using a servo mechanism on another woofer used to control the pressure within the cabinet in order to aid the main woofer avoid these downsides if there are any?
Cheers!
Mike
Drivers are much better these days...
It's mainly about diminishing returns. Our hearing isn't very sensitive to low frequency distortion. The cost and added complexity isn't worth the few extra percentage points of accuracy for most people. Large subs with single digit distortion in the 20's are readily available at typical consumer price points.
Doesn't hurt if you want the best. Just make sure the rest of the system is up to snuff.
It's mainly about diminishing returns. Our hearing isn't very sensitive to low frequency distortion. The cost and added complexity isn't worth the few extra percentage points of accuracy for most people. Large subs with single digit distortion in the 20's are readily available at typical consumer price points.
Doesn't hurt if you want the best. Just make sure the rest of the system is up to snuff.
The reason is one has to understand the basics of feedback and loop stability phase issues. It requires understanding of both woofer mechanics and basic stability issues in electronics and their complex interaction. Only one out of 5.000 audio amateurs seem to do. After some 100+ hrs of reading on the topic I still struggle with some of the issues involved.
These threads may interest you:
Commercial motional feedback woofer available sort of.
MFB for ACI SV12 Drivers using Piratelogic Electronics
Eelco
These threads may interest you:
Commercial motional feedback woofer available sort of.
MFB for ACI SV12 Drivers using Piratelogic Electronics
Eelco
It may have to do with one of the general "sounds better" principles, regarding feedback. In the context of amplifiers, it goes something like "It's better to start with a good sounding topology, add a little feedback to make it perfect - than to start with an inferior circuit and add a ton of feedback to make it acceptable"
The ton of feedback is something that can be heard. Unsure what superlative it shows up as.
In a subwoofer, it probably works best to do a good "open loop" driver / cabinet design that works well - with or without MFB. Than to cram a driver in a 0.5 cu ft enclosure and use 1kW of muscle to force the cone to follow that input no matter what. Oversimplified, but perhaps a relevant explanation.
The ton of feedback is something that can be heard. Unsure what superlative it shows up as.
In a subwoofer, it probably works best to do a good "open loop" driver / cabinet design that works well - with or without MFB. Than to cram a driver in a 0.5 cu ft enclosure and use 1kW of muscle to force the cone to follow that input no matter what. Oversimplified, but perhaps a relevant explanation.
Commercial stuff:
It is expensive to manufacture and the feedback loop must be guaranteed to remain stable under manufacturing tolerances and component ageing. If not, some products will self-destruct right out of the box or after several years and ruin the brand name.
DIY stuff:
It requires knowledge about control theory.
It is expensive to manufacture and the feedback loop must be guaranteed to remain stable under manufacturing tolerances and component ageing. If not, some products will self-destruct right out of the box or after several years and ruin the brand name.
DIY stuff:
It requires knowledge about control theory.
Last edited:
First of all it is rather expensive to be effective. Accelerometers are rather pricy. Sure you could use current drive amplifiers and use the back-EMF, but the back-EMF is proportional to velocity. At low frequencies this signal is much lower.
You could also use a sense coil in the driver, but this requires custom drivers or you lose half of the power rating when you use one of the drive coils. There's also lot's of sensor disturbances with a sense coil because the signal is coupled with the drive coil due to mutual inductance.
Doing motional feedback is VERY effective if done right. The back-EMF method or the sensing coil method only lead to single percentages of improvement. Whereas the accelerometer can lead up to 20dB!
I don't really think the control theory is the problem. You also need control theory for good amplifier design and general analog electronics design. If you have the model you put it in matlab, open up the sisotool, optimize with robust control and you're ready to go! The tuning isn't the problem
I think the problem is more in the modelling of the loudspeaker drive unit and doing this accurately. Besides that, the model will most likely change over time and the manufacturing tolerances are like 20%.
If you are interested in making one yourself I highly recommend reading these papers:
Audio Design Principles
The sub he made for Grimm (of which the papers are based) also sounds awesome. Hands down the best sub I know.
You could also use a sense coil in the driver, but this requires custom drivers or you lose half of the power rating when you use one of the drive coils. There's also lot's of sensor disturbances with a sense coil because the signal is coupled with the drive coil due to mutual inductance.
Doing motional feedback is VERY effective if done right. The back-EMF method or the sensing coil method only lead to single percentages of improvement. Whereas the accelerometer can lead up to 20dB!
I don't really think the control theory is the problem. You also need control theory for good amplifier design and general analog electronics design. If you have the model you put it in matlab, open up the sisotool, optimize with robust control and you're ready to go! The tuning isn't the problem
I think the problem is more in the modelling of the loudspeaker drive unit and doing this accurately. Besides that, the model will most likely change over time and the manufacturing tolerances are like 20%.
If you are interested in making one yourself I highly recommend reading these papers:
Audio Design Principles
The sub he made for Grimm (of which the papers are based) also sounds awesome. Hands down the best sub I know.
Depends on the woofers you are using though. Doubling the amount of woofers is going to reduce distortion by 6dB.
If you can reduce the distortion by 12dB from 20-100Hz, most of the time that's going to be cheaper than buying 3 additional woofers.
Yeah sure you have to spend like 50 euro's on the sensor and maybe 75 euro on the rest of the electronics. But where do you find 3 additional woofers for 125 euro?
Besides that, the MFB sub is going to be much more compact. Force canceling woofers is a big advantage though and can be tricky to implement with MFB. Article 7 of the articles I linked in an earlier comment explains how one could tackle that problem.
There are no kits afaik. Rythmik sells their drivers seperately, but they use the sense coil method. Your best bet it is piratelogic if you want to buy drivers with sensors mounted.
Either solution still requires you do design a controller and modify your amplifier for current drive. The papers I posted earlier by RMS acoustics show one method for current drive. More methods for current drive are mentioned in this book: Current-Drive - The Natural Way of Loudspeaker Operation
I know that Robbert (author of the RMSacoustics papers) and ds23man are working on a mod for the hypex fusion amplifiers for MFB subs. But the updates on the dutch MFB forums (MFBlabs.nl) have sadly been rather quite lately :c.
The main challenge is getting an accurate model of the loudspeaker and creating a good controller. We're still talking about single input single output (SISO) systems. Thus tuning can be done relatively easy with the sisotool of matlab. However the modelling part remains. You can make a model based of T/S parameters, but that's only so accurate.
Last edited:
The woofer does not have to be modelled. All you need is the transfer function from amplifier input to accelerometer output, which is easily measured. It similar to measuring the measuring the frequency response of a loudspeaker with a microphone, but instead of a microphone, an accelerometer is connected.
Last edited:
Much good thought in this thread.
There is a foundational idiocy for all our woofers: they have their system resonance inside their operating band. So MFB makes a very substantial benefit (as everybody who has tried it will swear) to resonance and will tighten-up the bass a lot, but still a struggle to make it work.
For a mid-range driver with the driver resonance below the band (which is clean enough for most people who aren't ESL or ribbon fans), MFB would be helpful icing-on-the-cake, but just icing.
It is illustrative that nobody has been successful with MFB for bass and into the mid-range. Too big a wild-world of phases to work.
There are two separate worlds. First, it is tricky to manufacture in quantity and to work in all climates and around dogs and cats. Second, for DIY, I think this forum illustrates the gulf between carpenters and electricians.
Ace Bass and other current feedback MFB approaches can be manufactured for just a few dollars over non-MFB systems. Philips vintage systems, just pennies more on that. Fancy accelerometers a bit more but not a lot.
B.
There is a foundational idiocy for all our woofers: they have their system resonance inside their operating band. So MFB makes a very substantial benefit (as everybody who has tried it will swear) to resonance and will tighten-up the bass a lot, but still a struggle to make it work.
For a mid-range driver with the driver resonance below the band (which is clean enough for most people who aren't ESL or ribbon fans), MFB would be helpful icing-on-the-cake, but just icing.
It is illustrative that nobody has been successful with MFB for bass and into the mid-range. Too big a wild-world of phases to work.
There are two separate worlds. First, it is tricky to manufacture in quantity and to work in all climates and around dogs and cats. Second, for DIY, I think this forum illustrates the gulf between carpenters and electricians.
Ace Bass and other current feedback MFB approaches can be manufactured for just a few dollars over non-MFB systems. Philips vintage systems, just pennies more on that. Fancy accelerometers a bit more but not a lot.
B.
Last edited:
Well yes, you could do it like that. Though I have no experience in that approach.
Le(x), BL(x) Kms(x) are often not linear and would change based on excursion. This results in a change of root-locus and thus impacts the stability. One could use robust controll techniques to compromise, but ideally you would use optimal control techniques.
Having an accurate model gives you a better insight on the limitations of the product. It also allows you to go more towards optimal control. However I must admit, the rms acoustics papers use t/s parameters with semi inductance as model and this was proven VERY succesfull with peaking loopgain of 30dB!
Getting in midrange territory is very difficult, because you have the bode equality. Besides that it can be tricky to find sensors that have wide bandwidth, low noise and the dynamic range of 100dB.
You also need a very stiff cone(alluminum is nice, because the added sensor weight is relatively small as well). This eliminates very large woofers of 15" and up imo. Though they weren't suited for midrange duty in the first place.
Very small drivers (6.5" and smaller) tend to have much lower moving mass and therefore suffer a lot more from the added mass.
You also need to mount the sensor is a very stiff manner, otherwise you will just measure the mass spring resonance of the sensor instead of what the woofer is doing.
Rob (of RMS acoustics) was working on a sub with 2 woofers (dayton audio RSS210 HO 8). His results are pretty astonishing for those woofers. The results can be found on dutch forum MFBlabs. It's obviously in dutch, but perhaps google translate can help out. Here's a link to the thread:
Mini duosub met Hypex Fusion 252 MFB add-on en 2* Dayton RSS210 HO 8 - Pagina 5 - MfbLabs
Thanks for asking; nice to insert a personal note.
Starting in 1968, did research at Bell Labs (Murray Hill) - had great fun in their anechoic chamber, world's biggest. Used first tube and then solid-state versions then 40 yrs with Karlson, sealed boxes, and Klipschorn bass (which is appropriate for MFB but is already a great woofer).
For last few years, no MFB but want to get organized with experiments with an accelerometer I bought.
I think the question for today is how to measure improvement in transient ("fast") performance, esp if there's a quantitative score possible so as to document performance.
B.
Last edited:
To be honest I really wasn't aware they were expensive anymore? A quick google search showed me you can get them pretty cheap now, although I'm not exactly sure on the kind of accelerometer you'd want, my guess would be something which is as light as possible. But almost every phone in the world has one, if they're that expensive break apart an old broken phone and take it out
I actually wanted go about modelling a loudspeaker, in my mind it was a pretty simple system, its a single degree of freedom system which you can model as a simple mass spring damper system. Only 2 forces really act on it, the force from the voice coil, resistance from the air in front of it and behind, both of which I suppose you could include with in the model of the damper?
The idea I had behind modelling it was I imagined if you place a driver behind the main driver, and had a model of the speaker, if you know the resistive forces plus the forces the speaker itself is able to generate, could you then create a system where the driver behind adjust the pressure to ensure the main speaker can move where it needs to be with exactly enough force to over come the resistive forces. So I suppose the drive behind would be just slightly out of phase so that its creating a pressure differential to create the desired acceleration in the main driver if that makes sense?
Either way I was going to write out the dynamic model and send it to my old dynamics lecturer who also was my control loops lecturer so if you have any questions you want me to ask him I could try. But fair warning I was HORRIBLE at control loops so they'll go over my head.
To be honest I really wasn't aware they were expensive anymore? A quick google search showed me you can get them pretty cheap now, although I'm not exactly sure on the kind of accelerometer you'd want, my guess would be something which is as light as possible. But almost every phone in the world has one, if they're that expensive break apart an old broken phone and take it out
You only need a single degree of freedom. The ones in your phone are vastly different and generally have way more DOF.
I'm actually not TOO concerned about the weight of the sensor. It adds moving mass and thus lowers Fres and sensitivity a bit. When using a stiff alluminum cone (stiff alluminum cones are nice because they pretty much guarantee there is no odd resonance in the range the driver is used) the moving mass is already pretty high.
The papers I mentioned earlier uses a SEAS Extreme series subwoofer. This has a moving mass of 173 gram. The mounting + glue and sensor he used are about 25gram which reduces the efficiency by 1.25dB. Sure this is less than optimal, but it's a sacrifice I am willing to take.
The problem lies in the fact that you have to make a compromies between dynamic range, noise, linearity and of course price.
110dB of 100Hz with a 10" woofer is already a cone acceleration of 1700m/s^2!
CD quality music is 96dB so we want atleast that dynamic range. This can be tricky because of 1/f noise. In these low frequenties the noise tends to be quite a bit higher.
The linearity of the sensor should be significantly better than the driver. Otherwise you will correct for things that the sensor says there are, but aren't there in reality!
Most premium MFB subs I know us the ACH-01 sensor. It's a great sensor, but it's also 50 euro's.
Klippel has quite some papers about modelling loudspeakers. There's also an example of a loudspeaker model in the RMS acoustic paper about sensorless velocity feedback and Rob also explains the operation and distortion sources in paper 1 and 2.
I won't go into depth on this topic as it's not my field of expertise. Just read their research, those guys are MUCH and MUCH smarter than I am.
Interesting idea. I assume you want to use the same enclosure and use the air inside as a spring?
Like this:
An externally hosted image should be here but it was not working when we last tested it.
You could use the second driver as sensor then. This works, but you have to buy a second driver which is often not cheap.
If this is what you ment, this problem is discussed in the paper about sensorless velocity feedback.
The tricky part is that the system is super non-linear. You can't just make a state-space model or transferfunction model because those techniques are based off linear systems. Sure you can do a taylor series approximation, but that only gets you up to the first order. As soon as you go to the second order or higher you get a quadratic function again. If you're going to ask questions I'd ask how you would model a very non-linear driver optimally.
What about the Qtc or damping of the system?
Q
Good concept to identify, measure, and quantify what "fast" means. But how do you measure Etc in-situ acoustically? BTW, there's a lot of loose talk about the how wonderful it is to have "critical damping" in Hornresp etc. But my intuition is quite different. A re-production device should have extreme damping. And, mirable dictu, when you push a MFB cone, it feels dead to the touch and like pushing a wall. In other words, there is no "speaker sound" to hear.
And apropos modelling, in a feedback system, the performance is the profile of the feedback loop, not the device under correction. So in that sense, it doesn't much matter what the speaker system is doing.
Using a Wheatstone Bridge (my personal favourite), you can balance the bridge using a Klippe-likel model of your speaker. Note, except for OBs, you are talking about the driver mounted as a speaker, not the just the driver.
Balancing the bridge may improve the stability of the feedback system (I don't know enough to say for sure). But no benefit to knowing the speaker model except in planning amount and bandwidth of the MFB (which, of course, is important).
B.
Good concept to identify, measure, and quantify what "fast" means. But how do you measure Etc in-situ acoustically? BTW, there's a lot of loose talk about the how wonderful it is to have "critical damping" in Hornresp etc. But my intuition is quite different. A re-production device should have extreme damping. And, mirable dictu, when you push a MFB cone, it feels dead to the touch and like pushing a wall. In other words, there is no "speaker sound" to hear.
And apropos modelling, in a feedback system, the performance is the profile of the feedback loop, not the device under correction. So in that sense, it doesn't much matter what the speaker system is doing.
Using a Wheatstone Bridge (my personal favourite), you can balance the bridge using a Klippe-likel model of your speaker. Note, except for OBs, you are talking about the driver mounted as a speaker, not the just the driver.
Balancing the bridge may improve the stability of the feedback system (I don't know enough to say for sure). But no benefit to knowing the speaker model except in planning amount and bandwidth of the MFB (which, of course, is important).
B.
Can't you do a close in measurement? Rob also uses this in his papers.
You can also do laser measurements for low frequencies. And 3rd you could do a measurement using the accelerometer you've mounted.
I don't have much experience on the 2nd and 3rd method. I've been willing to build a MFB system for a very long time, but I promised myself to not start a new project until at least 2 out of my 3 projects are finished.
What's the point of a Q <0.5. I can understand wanting to go slightly below 0.707, because of tolerances, but I don't understand the added value of going even lower. In fact, a lower Q would result in a earlier roll off and thus reduce negatively affect the FR.
A hit on the woofer can be modelled as a disturbance. Disturbance rejection is just a matter of achieving high loop gain. This is easier said than done though.
I wouldn't say that. Feedback suppresses disturbances of all forms. If the open loop has little distortion it will just get lowered even further.
By wheatstone configuration you mean the configuration I linked in my previous comment? It that case it indeed doesn't work for OB.