For historical reason, badly designed amps for low cost consumer or PA class D amps has bad reputation near some sectarian audiophiles. But lot of us knows how good they can measure and sound. up to the highest level of quality.
And it is the future, waiting for direct digital conversion/amplification.
So, let-us use this thread to explore all together, on a theoretical as well as practical point of view all the ideas we can imagine in order to improve further this class of amps.
And it is the future, waiting for direct digital conversion/amplification.
So, let-us use this thread to explore all together, on a theoretical as well as practical point of view all the ideas we can imagine in order to improve further this class of amps.
My first contribution to this thread can be resumed on two ideas.
First, some report a "clinical" sound to the pulse-width modulation used in Class D, same remark they had about the first historical DA converters.
This is due i believe, to the limitation of the analyze capability of the ultra low levels.
So, we can imagine to add in serial with the switching power transistor of a class D amp, a class A transistor, with an analog modulation at low level.
The second idea is to fight against the switching frequency problems.
There is a technological limitation to the switching frequency due to the speed of the available power fets. Those speed will increase with technological progress, but we are stuck with what we can buy.
I do not enter in the details of efficiency and distortion compromise we are obliged to do, in order to optimize the switching frequency. You know that, higher is this frequency, easier is its filtration before we send an analog signal to our beloved loudspeakers.
My idea is very simple. Using two D amps, in bridge configuration (phase opposition) with a common clock, and a quarter of wave delay for one of the amps of the bridge.
it will act like a X2 oversampling method, and, in a way, double the switching frequency.
Waiting for your comments and ideas, with apologizes for my poor English..
First, some report a "clinical" sound to the pulse-width modulation used in Class D, same remark they had about the first historical DA converters.
This is due i believe, to the limitation of the analyze capability of the ultra low levels.
So, we can imagine to add in serial with the switching power transistor of a class D amp, a class A transistor, with an analog modulation at low level.
The second idea is to fight against the switching frequency problems.
There is a technological limitation to the switching frequency due to the speed of the available power fets. Those speed will increase with technological progress, but we are stuck with what we can buy.
I do not enter in the details of efficiency and distortion compromise we are obliged to do, in order to optimize the switching frequency. You know that, higher is this frequency, easier is its filtration before we send an analog signal to our beloved loudspeakers.
My idea is very simple. Using two D amps, in bridge configuration (phase opposition) with a common clock, and a quarter of wave delay for one of the amps of the bridge.
it will act like a X2 oversampling method, and, in a way, double the switching frequency.
Waiting for your comments and ideas, with apologizes for my poor English..
This already exists, check out the Crown BCA amplifier, it runs a 250kHz carrier, but the three level modulation "balanced current amplification" scheme boosts the frequency to 1MHz, according to Crown in their BCA manual.
The "filterless" class d is another example, like the BCA, it also uses three level modulation which doubles the frequency at the output.
The "filterless" class d is another example, like the BCA, it also uses three level modulation which doubles the frequency at the output.
I'm sure, and never pretend any idea i can have as original.
I wonder who can have a totally brand new idea in electronic nowadays, see what i mean ?
My second idea seams not genuine, as well: Home
All that do not impeach-us to evaluate, comment, test.
So what is the reason these two technologies are reserved for high power PA and/or low power such at portable devices ?
PRICE! It would simply cost too much, noone would wanna buy these products. Also it uses twice the amount of components per channel as the commonly used halfbridge designs found in comsumer audio today.
I believe hypex has the cheapest bridgemode class d on the market and quite possible the best sounding as well, but do ppl buy them ? NO! they opt for something cheap made in china, nowdays sound quality is not as big of a concern as price, ppl rather buy a cheaper product than spend big money to get good sound.
PRICE! It would simply cost too much, noone would wanna buy these products. Also it uses twice the amount of components per channel as the commonly used halfbridge designs found in comsumer audio today.
I believe hypex has the cheapest bridgemode class d on the market and quite possible the best sounding as well, but do ppl buy them ? NO! they opt for something cheap made in china, nowdays sound quality is not as big of a concern as price, ppl rather buy a cheaper product than spend big money to get good sound.
Not realty our concern, here.
Means we are supposed to focus our interest in quality for money, and class D amps are very promised in this matter.
I believe there is some interest if the members of this forum can share their experiences, expertises and ideas in exploring the class D amps (the younger continent) in order to improve-it to the top. Hoping that it will lead to real community DIY projects
Is that was a bad idea ?
Component count is another factor, instead of using fullbridge and three level modulation that emulates a doubled switching frequency, a haldfridge is used whith a raised switching frequency, just look at fumac (huygens audio) which pushes his class d amplifiers up to 2MHz switching frequency.
Nico, may-be time to loose your virginity ?
I'm pretty sure that L.C. will join this thread very soon after his listening session in the studio where his SSA operate.
Last edited:
Christophe,
Talking about virginity, it has disappeared between the folds of my stomach and I sometimes spend the whole morning finding it to urinate. Sometimes I give up finding it and then people just think of me as a sweaty old man. So before I can lose it I must find it - is this very important,
Jokes aside, I would very much want to take a stab at a well designed class D amp, I do understand the concept and at times wondered if one cannot make the controller chip using a little 18F6240 Pic, I does have some good features such as 40MHz clock and many in built peripherals.
I spent a number of years working on an AM radio transmitter which used digitally driven class D modulators and adaptive precorrection. We came out with a pretty awesome product in the end, but learned a lot of lessons in the process and had to compensate for several things we never thought of in the beginning to make it perform well.
Here's what we were up against:
- Inductors aren't linear, they suffer from inductance droop and thermal effects.
- Power supplies aren't zero impedance.
- FETs have finite switching time, not to mention a Rds(on) that varies with temperature.
- Stuff creeps. Believe me, "burn in" exists. Cap values and ESR change, inductors change value, etc. Maybe not by much, but in a system that has to achieve a very high performance target this effect can screw things up.
In my opinion, analog self-oscillating amplifiers with post-filter feedback (UcD, etc) are pretty much the best amplifier topology out there right now, because they can compensate for all of these effects and do so with a greater amount of bandwidth for a given switching frequency compared to a fixed frequency amp. I'm not hating on fixed-frequency amplifiers, they're pretty close in performance.
To make a good performing amplifier using digitally generated PWM, you have to compensate for all of the aforementioned effects, and do so with some sort of feedback process to catch thermal effects and creep. Also, making things digital introduces a few extra limitations that don't exist in the analog domain - time quantization, A/D bandwidth/delay limitations, limited PWM resolution, etc. None of this is theoretically insurmountable and you can probably make a very good sounding amp - but I think you'll end up with a very complicated amp at not much benefit.
Anyway.
The "two phase" BTL amplifier you describe has been done, and exists in the form of the class BD aka "filterless" audio amplifier. You can accomplish the same end result by using two half bridges, driving them with the same "two phase" signal you would drive your BTL amplifier with, and paralleling the two half bridges - each stage has its own inductor, but all amplifiers share a common capacitor.
You can extend this to any number of phases - as you add more phases, the amount of ripple current seen at the output capacitor is much less, and it's pushed higher in frequency. You're effectively increasing your switching frequency, allowing higher bandwidth, and higher bandwidth allows for more feedback and better audio quality. You need to design things so you don't get large phase-to-phase currents happening, but that's manageable.
This is similar in concept to the evolution of VRM regulators you find on computer motherboards - current demands are higher, voltages are lower, allowable ripple is lower, load transients can happen much quicker - and the industry handled it by going with multiple phases. The radio transmitter I worked on operated this way, running up to 9 separate phases to the class D modulators.
I think evolution of MOSFETs with higher switching frequencies, lower switching loss, etc. will drive more change in audio performance than new amplifier topologies. I also believe we're pretty close to the limit with class D where we can say that the amplifier is no longer the limiting component of the playback chain - or at least we're very close to that point.
Here's what we were up against:
- Inductors aren't linear, they suffer from inductance droop and thermal effects.
- Power supplies aren't zero impedance.
- FETs have finite switching time, not to mention a Rds(on) that varies with temperature.
- Stuff creeps. Believe me, "burn in" exists. Cap values and ESR change, inductors change value, etc. Maybe not by much, but in a system that has to achieve a very high performance target this effect can screw things up.
In my opinion, analog self-oscillating amplifiers with post-filter feedback (UcD, etc) are pretty much the best amplifier topology out there right now, because they can compensate for all of these effects and do so with a greater amount of bandwidth for a given switching frequency compared to a fixed frequency amp. I'm not hating on fixed-frequency amplifiers, they're pretty close in performance.
To make a good performing amplifier using digitally generated PWM, you have to compensate for all of the aforementioned effects, and do so with some sort of feedback process to catch thermal effects and creep. Also, making things digital introduces a few extra limitations that don't exist in the analog domain - time quantization, A/D bandwidth/delay limitations, limited PWM resolution, etc. None of this is theoretically insurmountable and you can probably make a very good sounding amp - but I think you'll end up with a very complicated amp at not much benefit.
Anyway.
The "two phase" BTL amplifier you describe has been done, and exists in the form of the class BD aka "filterless" audio amplifier. You can accomplish the same end result by using two half bridges, driving them with the same "two phase" signal you would drive your BTL amplifier with, and paralleling the two half bridges - each stage has its own inductor, but all amplifiers share a common capacitor.
You can extend this to any number of phases - as you add more phases, the amount of ripple current seen at the output capacitor is much less, and it's pushed higher in frequency. You're effectively increasing your switching frequency, allowing higher bandwidth, and higher bandwidth allows for more feedback and better audio quality. You need to design things so you don't get large phase-to-phase currents happening, but that's manageable.
This is similar in concept to the evolution of VRM regulators you find on computer motherboards - current demands are higher, voltages are lower, allowable ripple is lower, load transients can happen much quicker - and the industry handled it by going with multiple phases. The radio transmitter I worked on operated this way, running up to 9 separate phases to the class D modulators.
I think evolution of MOSFETs with higher switching frequencies, lower switching loss, etc. will drive more change in audio performance than new amplifier topologies. I also believe we're pretty close to the limit with class D where we can say that the amplifier is no longer the limiting component of the playback chain - or at least we're very close to that point.
Brilliant resume. I agree too with your words about VRM regulators, but i would like to make some personal (not an universal value) remarks about.
First, we are not in the same 5 Ghz terrific frequencies, those i cannot even imagine. I've read that those "9 VRM phases" were more a marketing argument than a real requirement and that motherboards with less phases (4) give better overclock results.
So i believe, yes, that we can achieve a pretty good power supply for class D at a reasonable level of price and complexity. Computer based regulated supply ? You are right in talking about PSU first.
In an other way, some said that class D amps have better sound with analog power supplies... ?
Your experience is very precious, thanks to share-it. What, in your point of view would be the ideal switching frequency for audio ?
What technology for the filtering coil ? Air ? Ferrite ? (Why did inductors change value ?)
What is the maximum residual level of the switching frequency can we allow ?
About the speed of active devices, i agree with you, and i think that some new technologies (nano carbon etc..) will change the landscape in an incredible way quite soon. As far I'm in concern, i believe that speed is the secret, as well in the digital domain that in the analog one.
I'm pretty certain those motherboards with 9 VRM phases are designed for people who are doing extreme overclocking. CPUs can pull 200+ watts in those situations, which ends up being well over 100 amps of current, adding more VRM phases greatly helps.
I was comparing computer VRMs with class D amplifiers themselves. A computer VRM is just a synchronous buck regulator, and a half bridge class D power stage is also a synchronous buck regulator of sorts, but with the low side MOSFET switching to a negative rail instead of ground. Any of the same tricks used in making a better buck power supply (interleaving multiple phases, etc) can be done to class D amplifiers also.
The inductors we were using were ferrite cores with an air gap, they changed value due to temperature rise - and resulting material property changes and/or thermal expansion. Inductance also 'crept' over time, for reasons unknown. Could be material changes, mechanical movement (temperature/vibration can make stuff walk), who knows. Rather than trying to understand everything that was going on we just compensated for it.
Ideal switching frequency for audio: As high as possible.
Maximum residual level: depends on so many things - application, size of the amp (if you're going for an absolute # instead of ratiometric to the amp size), EMI regulations...
I was comparing computer VRMs with class D amplifiers themselves. A computer VRM is just a synchronous buck regulator, and a half bridge class D power stage is also a synchronous buck regulator of sorts, but with the low side MOSFET switching to a negative rail instead of ground. Any of the same tricks used in making a better buck power supply (interleaving multiple phases, etc) can be done to class D amplifiers also.
The inductors we were using were ferrite cores with an air gap, they changed value due to temperature rise - and resulting material property changes and/or thermal expansion. Inductance also 'crept' over time, for reasons unknown. Could be material changes, mechanical movement (temperature/vibration can make stuff walk), who knows. Rather than trying to understand everything that was going on we just compensated for it.
Ideal switching frequency for audio: As high as possible.
Maximum residual level: depends on so many things - application, size of the amp (if you're going for an absolute # instead of ratiometric to the amp size), EMI regulations...
Agree on all except about VRM. I had read several overclocking comparisons with different mother boards, and different VRM phases numbers. The winners where not correlated with this phases number.
I agree with you on all other points.
It is obvious that the quality of a Class D amp depend on two main things (between others ;-): the quality of the triangular signal and the constance of the rail voltage.
About "burn in", just i wanted to point that, un many audiophile's minds, burn'in is a kind of improvement in quality of the parts. See what i mean, like old wine ;-)
As you says, it is just changes in components value, can improve or reduce the desired performance as well.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.