As some of you know, see my other topic in the Class-D forum, I've embarked on another subwoofer project. This one however will be a no-holds-barred approach and hence I'm using the Adire Tumult 15D4 woofer in the design.
In order to have maximum control over the response curve, THD and excursion I'll be using feedback from the actual woofer. The obvious approach is to mount a piezo or MEMS accelerometer onto the woofer cone and put that in the feedback loop for the amplifier. I've been calculating the requirements for such a device and asked for feedback from Dan Wiggins at Adire and he commented there's another, potentially more interesting, relation that could make for a simpeler yet just as effective approach.
I'm however not 100% sure how to approach this so thought I'd post a few questions here to see how my fellow DIY-ers would approach this or whether anybody has any good suggestions. But let's first look at the maximum g force generated by the woofer, and then go from there.
The position of the cone with respect to time, assuming sine wave stimulus, is:
X(t) = Xpk sin(w t)
Xpk obviously is the peak excursion (not peak-to-peak, so it's 33mm here), w
is 2 pi f, and t is time.
The velocity is the first derivative of this:
V(t) = w Xpk cos(w t)
and acceleration is the second derivative of excursion:
A(t) = w^2 Xpk -sin(w t)
Maximum acceleration occurs when -sin(w t) is either 1 or -1, thus we are
left with the maximum acceleration being:
A(max) = w^2 Xpk
Let's assume the system has its largest Xpk at 20 Hz,
w = 2 pi 20 Hz
w = 125.6/s
and thus w^2 is:
w^2 = 15775/s^2
and since Xpk = 33mm, or 0.033m, then:
Apk = 0.033m * 15575/s^2
Apk = 514 m/s^2
and since g = 9.8 m/s^2, then:
Apk = 52 G
Dan commented that acceleration is also given by current times BL divided by mass (BL * i / m). So monitoring the current into the voice coil, as long as you have a constant BL and m, will give you acceleration. This got me thinking that in order to make this work all I'd need to do is to design a voltage controlled current source of sufficient power to drive the woofer and it will automatically flatten the curve, lower THD and keep excursion within acceptable limits.
My question obviously is how to approach this from the amplifier end, do I use a low-resistance sensing resistor in series with the woofer? Or something else? How exactly would I design a voltage controlled current source around the woofer? Anybody got any suggestions? Or a few pointers maybe?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
In order to have maximum control over the response curve, THD and excursion I'll be using feedback from the actual woofer. The obvious approach is to mount a piezo or MEMS accelerometer onto the woofer cone and put that in the feedback loop for the amplifier. I've been calculating the requirements for such a device and asked for feedback from Dan Wiggins at Adire and he commented there's another, potentially more interesting, relation that could make for a simpeler yet just as effective approach.
I'm however not 100% sure how to approach this so thought I'd post a few questions here to see how my fellow DIY-ers would approach this or whether anybody has any good suggestions. But let's first look at the maximum g force generated by the woofer, and then go from there.
The position of the cone with respect to time, assuming sine wave stimulus, is:
X(t) = Xpk sin(w t)
Xpk obviously is the peak excursion (not peak-to-peak, so it's 33mm here), w
is 2 pi f, and t is time.
The velocity is the first derivative of this:
V(t) = w Xpk cos(w t)
and acceleration is the second derivative of excursion:
A(t) = w^2 Xpk -sin(w t)
Maximum acceleration occurs when -sin(w t) is either 1 or -1, thus we are
left with the maximum acceleration being:
A(max) = w^2 Xpk
Let's assume the system has its largest Xpk at 20 Hz,
w = 2 pi 20 Hz
w = 125.6/s
and thus w^2 is:
w^2 = 15775/s^2
and since Xpk = 33mm, or 0.033m, then:
Apk = 0.033m * 15575/s^2
Apk = 514 m/s^2
and since g = 9.8 m/s^2, then:
Apk = 52 G
Dan commented that acceleration is also given by current times BL divided by mass (BL * i / m). So monitoring the current into the voice coil, as long as you have a constant BL and m, will give you acceleration. This got me thinking that in order to make this work all I'd need to do is to design a voltage controlled current source of sufficient power to drive the woofer and it will automatically flatten the curve, lower THD and keep excursion within acceptable limits.
My question obviously is how to approach this from the amplifier end, do I use a low-resistance sensing resistor in series with the woofer? Or something else? How exactly would I design a voltage controlled current source around the woofer? Anybody got any suggestions? Or a few pointers maybe?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Current amplifiers like you describe are occasionally seen. As you have calculated, they do eliminate some sources of distortion (because force is proportional to current), but not as much as much as proper motional feedback.
One problem is that you lose all electrical damping, which raises the overall system Q by a lot. On the plus side you don't have the stability problems of MFB.
You would implement this the same as a normal current feedback amplifier: by taking feeback from a small current-sensing resistor in series with the speaker. You can also mix current and voltage feedback to achieve a specific finite output impedance, mixing the properties of both, allowing control of Q.
One problem is that you lose all electrical damping, which raises the overall system Q by a lot. On the plus side you don't have the stability problems of MFB.
You would implement this the same as a normal current feedback amplifier: by taking feeback from a small current-sensing resistor in series with the speaker. You can also mix current and voltage feedback to achieve a specific finite output impedance, mixing the properties of both, allowing control of Q.
SSassen said:
I think it is not true.
Or only a half of truth.
acceleration=force/m=Bl*i/m
is an inertia of the cone, but other phenomenons occuring are:
resilience:
force(t)=Bl*i(t)=-(1/Cm) *x(t)
where Cm is a Thiel-Small parameter of stiffnes of a membrane
friction
force(t)=-b*V(t)
where b is constant
move of the cone is a superposition of these principles
best regards
Darkfenriz,
Thanks for the reply, but didn't you just give me a half answer
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Thanks for the reply, but didn't you just give me a half answer
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
well, probably I did.
and the conclusion is:
I will insist that the best way to drive a speaker is voltage source, because as long as Bl is constant the voltage is proportional to coil's velocity. And velocity is propotional to acoustic pressure as long as adiabatic thermodynamic process is linear.
With current source you produce force, which may be 'wasted' on resilience, inertia and friction depending on mechanical parameters.
Remember that pressure is what we actually hear and mechanical velocity is the result we want, not the fact that speaker 'did its best' and wasted much force on it.
Problems like temperature-induced compression aren't that crucial in my point of view so no real pros for current source.
hope I didn't make any brainfart here.
end_of_second_half
regards
and the conclusion is:
I will insist that the best way to drive a speaker is voltage source, because as long as Bl is constant the voltage is proportional to coil's velocity. And velocity is propotional to acoustic pressure as long as adiabatic thermodynamic process is linear.
With current source you produce force, which may be 'wasted' on resilience, inertia and friction depending on mechanical parameters.
Remember that pressure is what we actually hear and mechanical velocity is the result we want, not the fact that speaker 'did its best' and wasted much force on it.
Problems like temperature-induced compression aren't that crucial in my point of view so no real pros for current source.
hope I didn't make any brainfart here.
end_of_second_half
regards
Darkfenriz et al,
I’ve given the current sensing subwoofer feedback idea some more thought and came to the conclusion that current sensing is not the way to do it. Why? Simply because what we try to accomplish is to create an ideal loudspeaker which has a superconducting voicecoil, hence can be controlled exactly by changing the voltage across the voicecoil. In this case speed is directly proportional to the voltage applied.
In our case speed is only directly proportional to the voltage applied when BL is indeed fixed, but in reality it isn’t. When the woofer is pushed hard part of the voicecoil will leave the linear part of the magnetic field and hence BL changes. This will reduce the feedback voltage of the current sensing resistor we’ve opted to use, hence the amplifier will push even harder effectively forcing the cone further out of the magnetic field. Hence distortion increases and we’re quickly approaching Xmax. Obviously this is not the way to go.
So how about the accelerometer approach? The output voltage of an accelerometer is the derative of speed and hence below the frequency where its wavelength is smaller than he cone diameter there’s a linear relation between sound pressure and speed. Because we’re using a 15” woofer that frequency is upwards of 400Hz, and we'll only be operating it to about 120Hz, hence we’re well within the limit. So acceleration will be our feedback signal. So what we’ll then end up with is that the acceleration will be tracking the input voltage of the amplifier. The only thing we should watch out for is that for very low frequencies the cone will have very large excursion, hence a high-pass filter (+/- 10Hz) should be used to limit excessive excursion at sub-sonic frequencies.
Any suggestions? Did I miss anything?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
I’ve given the current sensing subwoofer feedback idea some more thought and came to the conclusion that current sensing is not the way to do it. Why? Simply because what we try to accomplish is to create an ideal loudspeaker which has a superconducting voicecoil, hence can be controlled exactly by changing the voltage across the voicecoil. In this case speed is directly proportional to the voltage applied.
In our case speed is only directly proportional to the voltage applied when BL is indeed fixed, but in reality it isn’t. When the woofer is pushed hard part of the voicecoil will leave the linear part of the magnetic field and hence BL changes. This will reduce the feedback voltage of the current sensing resistor we’ve opted to use, hence the amplifier will push even harder effectively forcing the cone further out of the magnetic field. Hence distortion increases and we’re quickly approaching Xmax. Obviously this is not the way to go.
So how about the accelerometer approach? The output voltage of an accelerometer is the derative of speed and hence below the frequency where its wavelength is smaller than he cone diameter there’s a linear relation between sound pressure and speed. Because we’re using a 15” woofer that frequency is upwards of 400Hz, and we'll only be operating it to about 120Hz, hence we’re well within the limit. So acceleration will be our feedback signal. So what we’ll then end up with is that the acceleration will be tracking the input voltage of the amplifier. The only thing we should watch out for is that for very low frequencies the cone will have very large excursion, hence a high-pass filter (+/- 10Hz) should be used to limit excessive excursion at sub-sonic frequencies.
Any suggestions? Did I miss anything?
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Hi,
Have you ever heard of the 3a Triphonic 1200 subwoofer? It used a microphone to "listen" to the outcoming signal from the four 11'' loudspeakers and then compared it with the incoming signal from the source. Of course, it had its own servo amp of some 150 WRMS which was fed with the resulting signal. An active crossover system separated frequencies above 120 Hz to send them to full range speakers. One of the advantages of this design was the size of the enclosure used. Because of the very nature of the servo system, size was kept to a minimum, something like a 0.8 cubic foot enclosure and distorsion figure was very low.
This system was designed in the early '70 and never really caught up with the market for pricing reasons and maybe because the market was not yet ready to accept it. We have to remember that in those days, bass this low (we're talking 20 hertz here) was more a source of distorsion than of music for the turntables, tonearms, cartridges and amps of this era. But still, this technology should still be considered today as it is very actual and seems to be suitable for your project.
Have you ever heard of the 3a Triphonic 1200 subwoofer? It used a microphone to "listen" to the outcoming signal from the four 11'' loudspeakers and then compared it with the incoming signal from the source. Of course, it had its own servo amp of some 150 WRMS which was fed with the resulting signal. An active crossover system separated frequencies above 120 Hz to send them to full range speakers. One of the advantages of this design was the size of the enclosure used. Because of the very nature of the servo system, size was kept to a minimum, something like a 0.8 cubic foot enclosure and distorsion figure was very low.
This system was designed in the early '70 and never really caught up with the market for pricing reasons and maybe because the market was not yet ready to accept it. We have to remember that in those days, bass this low (we're talking 20 hertz here) was more a source of distorsion than of music for the turntables, tonearms, cartridges and amps of this era. But still, this technology should still be considered today as it is very actual and seems to be suitable for your project.
darkfenriz said:
Actually, no. Acoustic pressure is proportional to acceleration below a frequency proportional to the circumference of the cone. Thats why you need 4x (12db) more excursion when you go down an octave for the same SPL. The inertia of the cone nicely cancels out this above resonance when driven by a current source.
The voltage over the "motor" part of the circuit is miniscule compared to the voltage over the voice coil resistance and inductance above the resonance of the driver so the "motor" is driven by an approximation of a current source.
Increasing the motor strength or lowering the dc resistance would give a lower Qes which makes the response above resonance rise. If you could make Qes = 0 the response of the driver would be rising with 6dB/octave because the velocity is the same at all frequencies.
As I have understood it, the advantage of using current drive is that the effects of inductance and inductance modulation are canceled out. 3rd order distortion created by eddy currents is also canceled out. (Not totally sure it was eddy currents though)
The disadvantage is that the Qts becomes Qms = very high. Huuuge peak at resonance. Has anyone tried lowering it back with a damping basket? (damping material compressed against the back of the basket)
This stuff is something i read in an old post by Kuei Yang Wang who seems to be very knowledgeable in the area. The theories do make an awful lot of sense!
/Joakim
Rythmik Audio already build a nice 350W current feedback servo amplifier with a small current sensing resistor in series with the woofer.
Their webpage contain information about the technology and the competitors.
http://www.rythmikaudio.com
Their webpage contain information about the technology and the competitors.
http://www.rythmikaudio.com
I think Rythmic uses a second, smaller coil mounted with the regular voice coil sense motion and provide a feedback signal. A microphone or an acceleramoter might be more accurate since the the relative movement between magnet and voice coil are not perfectly related to cone movement. But perhaps that good enough is the cone is sufficiently stiff. Also its a passive approach which may be simpler to implement.
That leads to another conjecture foy a DIYer: what about using a dual voice coil driver, using one of the coils for feedback sense? Rythmic claims that's suboptimal, but not how far sub optimal.
That leads to another conjecture foy a DIYer: what about using a dual voice coil driver, using one of the coils for feedback sense? Rythmic claims that's suboptimal, but not how far sub optimal.
Why Feedback?
Not so long ago i read an article by Bruno Putzeys (?) on Class D amplifiers and where they were going.
It seems to me that an "all-digital" Class D amplifier could be used to predict, and correct-for, distortions that are built-in to loudspeakers.
So why use feedback? Why not use "feed-forward" digital correction to get the subwoofer to do what you want it to? Perhaps temperature feedback (input) might be necessary.
Not so long ago i read an article by Bruno Putzeys (?) on Class D amplifiers and where they were going.
It seems to me that an "all-digital" Class D amplifier could be used to predict, and correct-for, distortions that are built-in to loudspeakers.
So why use feedback? Why not use "feed-forward" digital correction to get the subwoofer to do what you want it to? Perhaps temperature feedback (input) might be necessary.
Dan Ferguson in Audio X Press had two articles (and complementary informations in the mail section of the magazine) about using a feedback servo based on available dual-voice coil units.
Using a special unit with a dedicated sensing voice coil, smaller than the main one, is used by SubRythmik. As this voice coil is less high than the main one, it should rest in a more linear B region of the gap and may be pretty linear.
When using a conventionnal dual voice-coil, the signal of the sensing coil must be attenuated, it suffers from crosstalk between from the two voice-coils (there is less coil coupling with the Rythmik, I think) and efficiency is halved compared to both coils being active.
I think that the sensing voice-coil feedback is the easiest way to make MFB loudspeakers.
Using accelerometers makes the electronics circuit more complicated and is more tricky.
Feedforward can only be as good as the real loudspeaker bahaves similarly to the model.
I think the Linkwitz transform can be considered as analog feedforward. I have made some simulations with it, introducing little errors between the intended frequency resonance to compensate and the real resonance of the loudspeaker. It gaves sudden dips and jigs in the phase shift.
~~~~~~ Forr
Dan Ferguson's articles were published in AudioXpress issues of november 2003 and september 2004.
§§§
Using a special unit with a dedicated sensing voice coil, smaller than the main one, is used by SubRythmik. As this voice coil is less high than the main one, it should rest in a more linear B region of the gap and may be pretty linear.
When using a conventionnal dual voice-coil, the signal of the sensing coil must be attenuated, it suffers from crosstalk between from the two voice-coils (there is less coil coupling with the Rythmik, I think) and efficiency is halved compared to both coils being active.
I think that the sensing voice-coil feedback is the easiest way to make MFB loudspeakers.
Using accelerometers makes the electronics circuit more complicated and is more tricky.
Feedforward can only be as good as the real loudspeaker bahaves similarly to the model.
I think the Linkwitz transform can be considered as analog feedforward. I have made some simulations with it, introducing little errors between the intended frequency resonance to compensate and the real resonance of the loudspeaker. It gaves sudden dips and jigs in the phase shift.
~~~~~~ Forr
Dan Ferguson's articles were published in AudioXpress issues of november 2003 and september 2004.
§§§
feedforward, feedback
I remember something that the late (?) John Linsley-Hood said in his early-'70's (?) article on his new amplifier design. It was something to this effect: "There are those who wrongly believe that, with enough negative feedback, you can straighten the hind-leg of a donkey."
It seems to me that there is something in wrong in effectively saying: "well we don't really know, nor can predict, the precise behaviour of a speaker cone, but we'll just wrap it in enough negative feedback to solve any problem that might occur."
It's just ugly, it is not an "elegant" solution, I fear.
I remember something that the late (?) John Linsley-Hood said in his early-'70's (?) article on his new amplifier design. It was something to this effect: "There are those who wrongly believe that, with enough negative feedback, you can straighten the hind-leg of a donkey."
It seems to me that there is something in wrong in effectively saying: "well we don't really know, nor can predict, the precise behaviour of a speaker cone, but we'll just wrap it in enough negative feedback to solve any problem that might occur."
It's just ugly, it is not an "elegant" solution, I fear.
If you just solve (or reduce) some problems you are miles ahead. No point in rejecting a technique because it isn't perfect -- you can end up getting nothing done. IMO as long as the reseults are "better" you are making progress.
As for ugly vs. elegant: pushing air around with a cone glued to a glorified solenoid stikes me as one the more in-elegant technologies there is.
HI Roysyboy
Despite what Linsley-Hood said , he employs quite a lot of negative feedback in his amplifer designs, one of them was even inspired, I think, by a technique called "infinite feedback".
Do not forget that humans, animals and plants are all feedback systems, it is not a tehcnique to mask defaults, it's a technique to correct, in real time and within limits, any misbehaviour, as far as it can.
~~~ Forr
§§§
Despite what Linsley-Hood said , he employs quite a lot of negative feedback in his amplifer designs, one of them was even inspired, I think, by a technique called "infinite feedback".
Do not forget that humans, animals and plants are all feedback systems, it is not a tehcnique to mask defaults, it's a technique to correct, in real time and within limits, any misbehaviour, as far as it can.
~~~ Forr
§§§
MIKE E
"Why wont active EQ work to remove the undamped loudspeaker resonance?"
With current drive, an eq would not correct external sound pressures which are damped when using voltage drive. Systems using current drive and dealing with this problem have been proposed. They use motion feedback, the sensor being a secondary voice coil, just like Rythmik, (Hawksford) or an acceleromter, just like Philips MFB, whose signal is differentiated (Greiner). Hawskford has a site where his system is described.
~~~~~~~Forr
§§§
"Why wont active EQ work to remove the undamped loudspeaker resonance?"
With current drive, an eq would not correct external sound pressures which are damped when using voltage drive. Systems using current drive and dealing with this problem have been proposed. They use motion feedback, the sensor being a secondary voice coil, just like Rythmik, (Hawksford) or an acceleromter, just like Philips MFB, whose signal is differentiated (Greiner). Hawskford has a site where his system is described.
~~~~~~~Forr
§§§
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.