In linear amps the output transistors are the components really taking the heat when the current and voltage are out of phase. How is this in class-D amps? The switching transistors are either fully on - or off and have pretty much no voltage over it when conducting the load current. So all though the load current increases with say a cap in parallel with the load resistance, the power dissipation in the transistors are not much increased...? How about the output filter? Does it too only see an increase in current and not much more power dissipation? Assuming the ESR of the inductor is relatively benign.
Ignoring the loading issues (frequency response, peaking, etc) on the filter due to complex loading, where if anywhere, is the extra power dissipation taking place in a class-D amp with complex load? Is it just the extra current that we take into account or will there be anywhere with a larger voltage drop as the current is high giving the same dissipation issues as with linear amps?
Ignoring the loading issues (frequency response, peaking, etc) on the filter due to complex loading, where if anywhere, is the extra power dissipation taking place in a class-D amp with complex load? Is it just the extra current that we take into account or will there be anywhere with a larger voltage drop as the current is high giving the same dissipation issues as with linear amps?
Much unlike a linear amplifier, a switching amplifier transfers power in both directions i.e. from source to load and back. With a reactive load, the resulting reverse (algebraically negative) power is simply transferred back to the DC bus, as opposed to being turned into heat.
In case of a full-bridge amplifier, even this "reverse" power is perceived as forward (by the inverter), as the load is connected between two legs. Thus, even though the load perceives currents in both directions (AC), the power transfer in such cases is forward (only).
In case of a full-bridge amplifier, even this "reverse" power is perceived as forward (by the inverter), as the load is connected between two legs. Thus, even though the load perceives currents in both directions (AC), the power transfer in such cases is forward (only).
+1 on what @newvirus2008 wrote above.
In fact there was an interesting paper (I think Don Keele was an author) that came out a couple of years ago that used this very principle of class-D half bridge amplifiers (often called bus pumping) in tandem with a very low Qts high sensitivity pro audio woofer, with the result being an extremely efficient (in terms of real power consumed per acoustic Watt produced) by the loudspeaker. One example of a driver that can be used in this sort of application is the iPAL 21. Initially you would think that the dropping response from the very low Q driver (Qts<0.2) would mean that you have to apply a lot MORE power to lift the low end. But it turns out that when you look at real power consumed it is the opposite. This is because a low Q driver has a very wide impedance peak, and it is the impedance of the load that matters, not the Re value. Also, due to bus pumping, the amplifier gets back a lot of energy that it delivery to the driver at low frequencies, and this makes the overall efficiency very high. It's a very intersting principle, but there are few drivers that low a low enough Qts to take advantage of it. The design also requires amplifiers that can deliver very high voltage to the driver, but the PS does not need to have very high power capablity. Perfect for pro audio, but could also be used in domestic settings for e.g. a subwoofer.
Edit: found the paper. It's a JAES convention paper from 2003, so not all that recent after all. Attached.
In fact there was an interesting paper (I think Don Keele was an author) that came out a couple of years ago that used this very principle of class-D half bridge amplifiers (often called bus pumping) in tandem with a very low Qts high sensitivity pro audio woofer, with the result being an extremely efficient (in terms of real power consumed per acoustic Watt produced) by the loudspeaker. One example of a driver that can be used in this sort of application is the iPAL 21. Initially you would think that the dropping response from the very low Q driver (Qts<0.2) would mean that you have to apply a lot MORE power to lift the low end. But it turns out that when you look at real power consumed it is the opposite. This is because a low Q driver has a very wide impedance peak, and it is the impedance of the load that matters, not the Re value. Also, due to bus pumping, the amplifier gets back a lot of energy that it delivery to the driver at low frequencies, and this makes the overall efficiency very high. It's a very intersting principle, but there are few drivers that low a low enough Qts to take advantage of it. The design also requires amplifiers that can deliver very high voltage to the driver, but the PS does not need to have very high power capablity. Perfect for pro audio, but could also be used in domestic settings for e.g. a subwoofer.
Edit: found the paper. It's a JAES convention paper from 2003, so not all that recent after all. Attached.
Like a stepping motor, speaker inductance pushes current backwards into the amp, aka bus pumping. Switching Transistors have to have body diodes or external clamping diodes to conduct this current back into the power supply, which PWM supplies may not handle well. In linear amps, the "off" transistor comes on, vs class-D where the output inductor pushes the voltage to reverse the voltage on the "on" transistor.
Thanks for replies.
So I can expect mostly the same power dissipation in the amp's output devices (FETs and output filter) even with very capacitive or inductive loads and no need for extra cooling. But the power supply rail might 'see' the effects of reactive loads more than the amp output stage and thus I should look out for issues there.
So I can expect mostly the same power dissipation in the amp's output devices (FETs and output filter) even with very capacitive or inductive loads and no need for extra cooling. But the power supply rail might 'see' the effects of reactive loads more than the amp output stage and thus I should look out for issues there.
If that were the case, the output devices wouldn't require a heat sink.
Mosfets have an on resistance (that increases as they heat up) so there is a voltage drop across the drain-source that increases with current.
Also, they do not switch on and off instantly, rather there is a rise and fall time on the gate drive.
The other concern with high output currents is saturation of the output inductor. You don't want this.
Mostly, but for switching losses only. When the load is highly reactive, almost all power is returned to the supply, with very less real power being transferred to the load. However, the situation would be much better than what it would be with a linear amplifier.
That is not usually a problem, as the loop gain / PSRR has a decent value at low frequencies, which is where the "pumping" is significant. With larger capacitors this effect is further reduced, as it takes more energy to raise the bus voltage by the same amount.
If desired, a bidirectional rectifier (vs. diode-based) could also be used, which would start putting this reverse power right back into the grid, as soon it senses an over-voltage at the DC bus (its output).
On second thoughts, yes, mostly for conduction losses, but there could be some variation in the switching losses due to changes in switching times.
newvirus2008 said:
Please note that the above mainly applies to IGBTs where the reverse current path is through the antiparallel diode. With changing power factor, the conduction losses would shift between the IGBT and the diode.
In MOSFETs, both forward and reverse currents flow through the channel (assuming that the gate is charged) and therefore the conduction losses should remain more or less the same, even with highly reactive loads.
Switching time depends on the plateau part of the gate charge, which is very dependent on current in the channel (the gate voltage plateau is when the channel charge density builds up, being the opposite polarity from the charge on the gate, but roughly the same amout).
In order to set a switching time that prevents cross-conduction at full load, you have to be conservative, and this means most of the time you have switching losses higher than would be optimal (the alternative is to risk cross-conduction (aka "shoot-through") which typically explodes MOSFETs violently.
When the switching time you give is larger than the optimal, current spends the extra time flowing through the body diodes which typically drop more than 1V. And during the actual switching the devices pass through a point of both high current and voltage, usually the bulk of the losses are due to this.
And the I-squared-R relationship means even though modern MOSFETs are very low on-resistance, as current increases simple conduction losses may rise dramatically. There's a fair amount of literature out there, often in datasheets, for estimating total losses for various switching situations. For energy conversion resonant converters are favoured as this allows switching close to the voltage zero-crossing point, greatly reducing switching loss.
IGBTs switch too slowly to be a good match to class D audio, but are heavily used for high voltage motor control being much more robust than MOSFETs, and for motor control switching frequencies are a few kHz to a few tens of kHz only.
In order to set a switching time that prevents cross-conduction at full load, you have to be conservative, and this means most of the time you have switching losses higher than would be optimal (the alternative is to risk cross-conduction (aka "shoot-through") which typically explodes MOSFETs violently.
When the switching time you give is larger than the optimal, current spends the extra time flowing through the body diodes which typically drop more than 1V. And during the actual switching the devices pass through a point of both high current and voltage, usually the bulk of the losses are due to this.
And the I-squared-R relationship means even though modern MOSFETs are very low on-resistance, as current increases simple conduction losses may rise dramatically. There's a fair amount of literature out there, often in datasheets, for estimating total losses for various switching situations. For energy conversion resonant converters are favoured as this allows switching close to the voltage zero-crossing point, greatly reducing switching loss.
IGBTs switch too slowly to be a good match to class D audio, but are heavily used for high voltage motor control being much more robust than MOSFETs, and for motor control switching frequencies are a few kHz to a few tens of kHz only.
Yes, that's the point and the variations in plateau voltage have also been exploilted for current sensing purposes like in Section 4 of the below article.Mark Tillotson said:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3649430/