Current mode amp & speakers?
Hi all,
Given that:
- a loudspeaker cone moves under the influence of a magnetic field;
- the magnetic field around a coil of wire depends on the current flowing through it;
- a speaker is not purely resistive, ie. voltage and current are not directly proportional to one another:
Has anyone ever built an amp / speaker system which, rather than accurately controlling the voltage across the speaker terminals, instead drives an accurately controlled current through the coil?
I'd have thought that it would provide a way to cancel out all manner of non-linearities, and directly control the magnitude of the force between magnet and cone. Controlling the voltage across something with dynamically changing electrical properties seems a bit odd, and I wonder whether it's done as much out of convention as anything else. Presumably you'd need drivers optimised to be used this way, as well as a special amplifier - but is there any other reason why it's not done?
Andy.
Hi all,
Given that:
- a loudspeaker cone moves under the influence of a magnetic field;
- the magnetic field around a coil of wire depends on the current flowing through it;
- a speaker is not purely resistive, ie. voltage and current are not directly proportional to one another:
Has anyone ever built an amp / speaker system which, rather than accurately controlling the voltage across the speaker terminals, instead drives an accurately controlled current through the coil?
I'd have thought that it would provide a way to cancel out all manner of non-linearities, and directly control the magnitude of the force between magnet and cone. Controlling the voltage across something with dynamically changing electrical properties seems a bit odd, and I wonder whether it's done as much out of convention as anything else. Presumably you'd need drivers optimised to be used this way, as well as a special amplifier - but is there any other reason why it's not done?
Andy.
It's been discussed here many times. Assuming a reasonably ideal controlled current source ("current amplifier"), you will:
- reduce Le(x) modulation distortion -> good
- reduce Bxl(x) influence from quadratic to linear -> good
- obtain a rising frequency response -> must be dealt with in the crossover
- obtain a huge response bump at fs because you essentially make Q_es infinite
This last point is the most troublesome. It can be dealt with by:
- XO response filtering -> not likely to succeed if fs is in band
- pre filtering (such as Linkwitz tranform) -> you will be at the mercy of mechanical tolerances (f_s, Q_ms)
- added mechanical damping -> may work but it is hard to add lots of viscous damping that is really linear with excursion, cone speed and temperature
- motional feedback -> most promising, but also not without challenges
And did I mention it: you' d need to go fully active or completely resdesign your passive XO.
- reduce Le(x) modulation distortion -> good
- reduce Bxl(x) influence from quadratic to linear -> good
- obtain a rising frequency response -> must be dealt with in the crossover
- obtain a huge response bump at fs because you essentially make Q_es infinite
This last point is the most troublesome. It can be dealt with by:
- XO response filtering -> not likely to succeed if fs is in band
- pre filtering (such as Linkwitz tranform) -> you will be at the mercy of mechanical tolerances (f_s, Q_ms)
- added mechanical damping -> may work but it is hard to add lots of viscous damping that is really linear with excursion, cone speed and temperature
- motional feedback -> most promising, but also not without challenges
And did I mention it: you' d need to go fully active or completely resdesign your passive XO.
However, you also asked about the right amplifier for this application.
The only amplifier specifically geared towards this application that I am aware of is described in an article by Mills and Hawksford(Transconductance Power Amplifier Systems for Current-driven loudspeakers, JAES 37, 10, p. 811 (1989), available free from Hawksford's university home page). They use a complicated current dumping topology with floating power supplies and no global feedback that appears pretty academic to me.
The reason they gave was that a conventional voltage feedback amp configured to be a current amplifier would run out of open loop gain at high frequencies, hence no longer acting as a current source. However, the price they paid (apart from parts count) was a distortion performance in the 80 dB ballpark (similar to non-global feedback voltage amp designs), and if you search in this forum, you will find my calculation that even a lowly LM3886 configured as a current source has enough open loop gain to make a decent current source throughout the audio band.
The only amplifier specifically geared towards this application that I am aware of is described in an article by Mills and Hawksford(Transconductance Power Amplifier Systems for Current-driven loudspeakers, JAES 37, 10, p. 811 (1989), available free from Hawksford's university home page). They use a complicated current dumping topology with floating power supplies and no global feedback that appears pretty academic to me.
The reason they gave was that a conventional voltage feedback amp configured to be a current amplifier would run out of open loop gain at high frequencies, hence no longer acting as a current source. However, the price they paid (apart from parts count) was a distortion performance in the 80 dB ballpark (similar to non-global feedback voltage amp designs), and if you search in this forum, you will find my calculation that even a lowly LM3886 configured as a current source has enough open loop gain to make a decent current source throughout the audio band.
Let's develop our own power current amp!
Actually, I had thought about posting this for a while, and your post gave me the nudge to finally do it.
A discrete, high bandwidth voltage feedback amp has more bandwidth than an LM3886, and would hence make an even better current amp.
However, the basic problem with either the LM3886 or a conventional discrete design is that it has inherently a low open loop output impedance, and that we are using the open loop gain and the feedback circuit to transform this into a high impedance.
What we really need is a common emitter symmetrical output stage that has a high open loop output impedance. The CFP has the very output devices configured ok for our needs, but the drivers apply local feedback to drive the impedance down again.
What would work well is the symmetrical current mirror found in most integrated current feedback op amps. However, a discrete current mirror with gain is not something that one designs on the back of an envelope.
Actually, I had thought about posting this for a while, and your post gave me the nudge to finally do it.
A discrete, high bandwidth voltage feedback amp has more bandwidth than an LM3886, and would hence make an even better current amp.
However, the basic problem with either the LM3886 or a conventional discrete design is that it has inherently a low open loop output impedance, and that we are using the open loop gain and the feedback circuit to transform this into a high impedance.
What we really need is a common emitter symmetrical output stage that has a high open loop output impedance. The CFP has the very output devices configured ok for our needs, but the drivers apply local feedback to drive the impedance down again.
What would work well is the symmetrical current mirror found in most integrated current feedback op amps. However, a discrete current mirror with gain is not something that one designs on the back of an envelope.
might be as simple as this
Assuming 2 mA op amp quiescient current and 0.65 V_BE for driver and output transistors, this should work.
The dumping resistor was chosen this large to limit the base emitter voltage for the darlingtons to about 6 V (assuming 30 mA current limit in the op amp and +/- 13 V output voltage).
The RC circuit accross the loudspeaker keeps the impedance low for frequencies way above the audio band. Without this, we'd have HF oscillations, as the noise gain would grow with frequency due to the loudspeaker's inductance.
It may be necessary to have some local feedback around the op amp depending on how fast the output stage really is.
The circuit is quite common, but usually you see another resistor between the op amp output and the real output which would add some local feedback around the output stage, but this would be contrary to our aim of having a high open loop output impedance.
I am not sure what modern op amps would accept to receive their supply voltages through emitter followers without additional decoupling. Decoupling the op amp supply pins to ground is out of the question, as this would slow down the cascodes. I have a hunch that a local decoupling cap just between the supply pins would be ok - takers?
Assuming 2 mA op amp quiescient current and 0.65 V_BE for driver and output transistors, this should work.
The dumping resistor was chosen this large to limit the base emitter voltage for the darlingtons to about 6 V (assuming 30 mA current limit in the op amp and +/- 13 V output voltage).
The RC circuit accross the loudspeaker keeps the impedance low for frequencies way above the audio band. Without this, we'd have HF oscillations, as the noise gain would grow with frequency due to the loudspeaker's inductance.
It may be necessary to have some local feedback around the op amp depending on how fast the output stage really is.
The circuit is quite common, but usually you see another resistor between the op amp output and the real output which would add some local feedback around the output stage, but this would be contrary to our aim of having a high open loop output impedance.
I am not sure what modern op amps would accept to receive their supply voltages through emitter followers without additional decoupling. Decoupling the op amp supply pins to ground is out of the question, as this would slow down the cascodes. I have a hunch that a local decoupling cap just between the supply pins would be ok - takers?
Hm... there's a lesson to be learned here : never ask a question if you wouldn't understand the answer 
Seriously, thanks for the comprehensive reply - I will make the effort to look up the bits I don't understand, I promise!
I'm intrigued by your schematic, though. Personally I'd have just put a small value resistor in series with the speaker, then used a differential amplifier to monitor the voltage across that, in order to give a signal proportional to the coil current. Then use another negative feedback loop to keep that signal proportional to the input voltage, effectively compensating out the non-linearity of the loudspeaker's I/V characteristic.
Of course, if that's why I should be leaving amplifier design to the experts, then so be it...
Seriously, thanks for the comprehensive reply - I will make the effort to look up the bits I don't understand, I promise!
I'm intrigued by your schematic, though. Personally I'd have just put a small value resistor in series with the speaker, then used a differential amplifier to monitor the voltage across that, in order to give a signal proportional to the coil current. Then use another negative feedback loop to keep that signal proportional to the input voltage, effectively compensating out the non-linearity of the loudspeaker's I/V characteristic.
Of course, if that's why I should be leaving amplifier design to the experts, then so be it...
Well, let's see.
The force driving the voice coil is proportional to the magnetic field B inside the gap x the length of wire (number of turns x circumference of each turn) inside the gap, hence Bxl, aka the force factor. Obviously, Bxl is largest when the coil is centered in the gap, and it varies (decreases) as the coil moves in either direction, hence Bxl(x).
For voltage drive, there is a (Bxl)^2 in the denominator, hence Bxl(x), which is nonlinear, contributes even more to nonlinearity. With current drive, there is only Bxl in the denominator, hence nonlinearities of the force factors do not give you as pronounced a nonlinearity as with voltage drive.
Next thing, the voice coil has an inductance, L_e. In a typical motor, the center pole piece is just a rod that fills about 2/3 of the voice coil in its rest position. When the coil moves inwards, there is more iron inside the coil than in the rest positions; when it moves outwards, there is less iron. The iron is like the core of an iron cored inductor, so the inductance changes as a function of the position of the voice coil. This gives rise to quite some third harmonic distortion in the midrange.
Lastly, the cone has a mass, and the suspension is like a spring, so together, they have a natural or resonance frequency f_s. Just like a pendulum, the like to swing at a characteristic frequency, and very little power is needed from a perodic power source at this frequency (just think of pushing somebody sitting on a swing). Because of the the voice coil assembly, the mechanical resonance gets translated into an electical resonance, so the impedance of the loudspeaker increases at f_s. With voltage drive, voltage is constant over freqency, and current is voltage / impedance, hence at f_s, little current and hence power is drawn, so the frequency response stays flat.
With current drive, current is constant over frequency, so the current is force fed into the loudspeaker even at f_s where it is not needed.
This lengthy explanation was the equivalent of saying Q_es approaches infinity with current drive. The reason for this: the resonance is damped my mechanical losses (such as the power needed to keep the surround flexing) and electrical losses (the motion of the cone induces a voltage into the voice coil, but with voltage drive, the output impedance of the amp is close to zero, so this voltage gets shorted, resulting in a very efficient electrical brake). With current drive, the output impedance approaches infinity, so there is no electical damping.
The force driving the voice coil is proportional to the magnetic field B inside the gap x the length of wire (number of turns x circumference of each turn) inside the gap, hence Bxl, aka the force factor. Obviously, Bxl is largest when the coil is centered in the gap, and it varies (decreases) as the coil moves in either direction, hence Bxl(x).
For voltage drive, there is a (Bxl)^2 in the denominator, hence Bxl(x), which is nonlinear, contributes even more to nonlinearity. With current drive, there is only Bxl in the denominator, hence nonlinearities of the force factors do not give you as pronounced a nonlinearity as with voltage drive.
Next thing, the voice coil has an inductance, L_e. In a typical motor, the center pole piece is just a rod that fills about 2/3 of the voice coil in its rest position. When the coil moves inwards, there is more iron inside the coil than in the rest positions; when it moves outwards, there is less iron. The iron is like the core of an iron cored inductor, so the inductance changes as a function of the position of the voice coil. This gives rise to quite some third harmonic distortion in the midrange.
Lastly, the cone has a mass, and the suspension is like a spring, so together, they have a natural or resonance frequency f_s. Just like a pendulum, the like to swing at a characteristic frequency, and very little power is needed from a perodic power source at this frequency (just think of pushing somebody sitting on a swing). Because of the the voice coil assembly, the mechanical resonance gets translated into an electical resonance, so the impedance of the loudspeaker increases at f_s. With voltage drive, voltage is constant over freqency, and current is voltage / impedance, hence at f_s, little current and hence power is drawn, so the frequency response stays flat.
With current drive, current is constant over frequency, so the current is force fed into the loudspeaker even at f_s where it is not needed.
This lengthy explanation was the equivalent of saying Q_es approaches infinity with current drive. The reason for this: the resonance is damped my mechanical losses (such as the power needed to keep the surround flexing) and electrical losses (the motion of the cone induces a voltage into the voice coil, but with voltage drive, the output impedance of the amp is close to zero, so this voltage gets shorted, resulting in a very efficient electrical brake). With current drive, the output impedance approaches infinity, so there is no electical damping.
Oh yes, back to amplifiers. What you suggested can be done without a diff amp, any op amp will do (most power amps are in fact op amps). You just connect your load (speaker) between output and inverting input, and you connect a shunt resistor (current sensing resistor) between inverting input and ground.
You would still have to parallel a small capacitor or capacitor / resistor series connection to the load to avoid oscillation (the impedance of the loudspeaker grows with frequency, and the amplification is impedance of loudspeaker / resistance of shunt).
The drawback is that you make the feedback loop transform a low output impedance into an infinite one, which only works to a certain degree.
You would still have to parallel a small capacitor or capacitor / resistor series connection to the load to avoid oscillation (the impedance of the loudspeaker grows with frequency, and the amplification is impedance of loudspeaker / resistance of shunt).
The drawback is that you make the feedback loop transform a low output impedance into an infinite one, which only works to a certain degree.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.