The Future of Audio Amplification

 Nobody cares about class T?
People consider class T as in the same category with class D?

Yes. 'Class T' is a trademark.

I find Brunos definitions kind of useless.
Most Class D have no Analog to Digital Conversion (ADC) stage, but do have comparators. Hence the feedback required.

Actually the use of a comparator has nothing to do with whether feedback is needed or not. If its a cheap comparator with offset problems feedback is probably a good idea, but IME the resulting amp will have the problems that many class D detractors (often, myself included) complain about.

It would, since feedback is how the switching is controlled, without feedback, there can be no switching.

There are three main types of encoding. One is 'self oscillating' which requires feedback and the feedback sets the switching frequency. So for that type the above statement is true but for Pulse Width Modulation and Pulse Density Modulation the statement is false; the latter two can be operated without feedback.

The Technics model is different though. It uses digital signal processing to compensate for the speaker load, and then is a pure digital amp with no feedback ( I believe )

Its a class D (switching analog amp) with digital control. Bruno Putzeys, one of the top class D designers, has a nice article on why class D is analog and digital isn't:
https://www.hypex.nl/img/upload/doc/an_wp/WP_All_amps_are_analogue.pdf

Bit late to the party Merrill already did it using the GaN transistors, I believe the SE-R1 does also.

Yes, there are a number of zero feedback class D amps at this point.

Ralph, you are more capable than anyone on this thread to explain why class D is the future from a technical standpoint.  Lay it on us.  Also, have you (as George is implying) had a religious conversion to the world of class D to the point of implementing it in your products?

My apologies for a long answer to a short question.

No religious conversion. Its just engineering. We've been watching the development of class D for 20 years; the early amps were a joke, like the one under test that is represented in the little graphics George likes to post.


About 3-5 years ago they started to really come around. I've been using one for about ten years for my keyboard rig in my band, mostly because its light and powerful, and I have to move that stuff myself when we do a show. Its a Crown, and doesn't sound that bad. But for high end it falls short, although it is an older amp.

About two years ago I realized that Atma-Sphere had something to bring to the table so we investigated and it turned out that was the case. As a result we have a patent pending.


So my experience is by listening to our class D side by side against our regular production OTL amplifiers. They are remarkably similar. I have to assume that if Merrill has their ducks in a row, their amp should be very nice also.

As Kosst has pointed out, with any amplifier its all about distortion- what the amp has and what it doesn't have. Traditional solid state has always had a bit of low level higher ordered harmonic distortion, which is the source of the brightness and hardness for which solid state is known (and the basis of the tubes/transistors debate). Class D does not have the mechanism to create that sort of distortion. The higher-ordered harmonic structure of traditional solid state is partially the result of non-linear capacitive aspects imbued in the junctions of most transistors *and* also the feedback needed to linearize many transistor circuits.


Class D relies on switching and so eliminates that capacitive problem- and they can be linear enough that its nearly as easy as using tubes to build one that runs zero feedback.


So think about that last paragraph- the two main reasons transistors sound bright and harsh (which, make no mistake, is a coloration) can be eliminated with this technology. Of course, in solving that problem other problems are created; distortion in class D amps is increased by low switching speeds, deadtime circuits and the precision of the encoding scheme.


That is why we've seen the steady march to higher switching speeds- to reduce distortion and increase bandwidth. The problem has been shoot-through current (this is where both output devices are partially on at the same time, allowing current from the power supply to shoot through both devices at the same time, heating them up quite quickly). We can get the encoding schemes to work pretty well, but some modules employ opamps or other amplification at their inputs which can color the sound. But they are not mandatory and some circuits have the audio proceed directly into the encoding circuits without any processing whatsoever.

Oddly, the thing that most class D amps get attacked for, switching noise, isn't a thing because you can't have a noisy amp channel sharing any kind of circuitry with another noisy amp channel. They will find a way to talk to each other and it gets ugly (oscillations, hetrodynes, intermodulations, etc.). So any commercial design has to have that problem licked and they do.  This is actually a requirement to meet international radiation standards.


So there are a **bunch** of variables! But if the following criteria are met, the amp will have low distortion and in particular will be lacking the higher ordered harmonics and IMD, which means they can sound very 'tube-like', 'organic', 'musical' and so on:

1) High switching speed with low residual
2) no deadtime introduced (deadtime increases distortion)
3) accurate encoding
4) no feedback

Regarding feedback, unlike regular amps in which there are things like phase margin and the like that can cause phase shift issues with feedback (and possibly result in oscillation), class D amps employ switching, which introduces *propagation delay*. This means that the output of the amp is occurring time-wise always slightly behind that of the input. Its a tiny bit, and so gets treated by many designers as a phase shift issue, but in essence the feedback **over the entire bandwidth of the amplifier** is going to always be slightly late. This means it will make distortion; such amps IME will have amusical properties. IOW I don't think applying feedback in a class D amp to be a good idea.

So that makes the scan frequency in this example only 200KHz(!),

Wrong again when have you seen an amp that can do 200khz, have a slewing audio frequency square wave like this, the sides are collapsing in like the leaning tower of pizza

I think you misinterpreted this. The amp at the link you provided was an older amp that was only switching at 200KHz. Nobody has been building amps with switching speeds that low in a long time, unless its for subwoofers only. Also, we don't know why the risetime is as slow as it appears since the data on the amp itself is lacking. That can be caused by measurement errors, but it can also be caused by analog signal processing before the encoding scheme.

So its erroneous to apply the results of this nearly 20-year old circuit to newer amps.

Wrong!!! look at it and think again. It’s Stereophiles, there’s your hint, they’ve NEVER EVER given a Class-D 20khz square wave, as it would look unrecognizable as a square wave.

https://www.stereophile.com/images/1212AM1fig03.jpg

You are correct and I was in error- that is a 10KHz squarewave.

So that makes the scan frequecy in this example only 200KHz(!), which means the class D in question had to have been built in the **1990s**. By the early 2000s, everyone was doing well over 300KHz. That means that the bump on the leading edge is the phase shift caused by a filter that is likely at only 40KHz.
So using this as an example of **all** class D amps is still a Hasty Generalization and like any logical fallacy, is false.

20kHz is hard for any amp to do, much less well???

http://www.firstwatt.com/pdf/prod_f5_man.pdf

Look at page 17 where the FW F5 absolutely nails a 200KHz square wave at 1 watt. A lot of amps are designed to roll off at or past 20KHz more for purposes of self preservation than technical limitations. You’re always going to have that guy who wants to try Litz wire ICs and create a high power oscillator. 

I agree- but the Pass amps IMO are some of the best solid state amps made :)

Our OTLs are quite good at 1 watt 20KHz squarewaves too. And they don't care about the Litz wire; I'd be shocked if the Pass amps do!

Regarding distortion- I agree wholeheartedly that small amounts of particular kinds of distortion are easily heard. This is why I've avoided feedback in our amps; while feedback suppresses distortion in doing so it adds some of its own, and its all IMD and higher ordered harmonics which are easily heard.
But it is also true that you can build a class D with zero feedback and there are a number of them around. At that point, the limit of the distortion becomes the precision of the encoding scheme. For example if you use Pulse Width Modulation using a triangle wave to set the switching frequency, the distortion is all about how precise the triangle wave is and the speed and offsets of the comparitor circuit used. It happens that with that sort of circuit, as the amp approaches full output, it can have a form of soft clipping as the encoding scheme starts to fall apart. So the resulting amp can have very low distortion and a lack of higher ordered harmonics over most of its range.

And let me say, what ever you look at, these are both pathetic looking square wave shots, even for an old cheap 1970’s solid state amp to be producing,

This is the link about which this comment is made:

https://www.stereophile.com/images/1212AM1fig02.jpg

Obviously the above quote is false, being based on a Hasty Generalization:

Unless you didn't do the math, that's actually a very respectable waveform; 0.05milliseconds is 20KHz! Its pretty hard for **any** amplifier to make a good **squarewave** at 20KHz- to make it really square you need bandwidth to at least 10x the fundamental frequency.
We can see a bit of peaking in the waveform- probably the result of the phase shift of the filter, which is probably set fairly low to deal with the rather low 400KHz switching frequency; my guess is about 60KHz.

And the amplitude is pretty low, so the residual is easily seen. The residual is a bit high, about 1/2 volt. The link does not specify if the amp is under load, but if it is, then this is an older amp not representative of amps made in the last 5 years; if not under load that's quite decent despite the low switching frequency.

@georgehifiInstead of made up stories, just stick to the facts.
If the filter is not used on the AP, it goes into slewing protection.

We don't use a filter since we are using oscilloscopes; any waveforms we see have the residual sine wave imposed. Depending on the waveform the residual might just look like hash on the waveform until you open up the horizontal gain and see that its just a sine wave.

Haven’t found a digital which can compete.Most digital are to clean for my ear.

Sterile with no decaying harmonic structure, creating too much dead time, are the words I believe you should be using.

Dead time is part of the design of the amp, not something that arises from its ’harmonic structure’.

Judging by the measurements I keep seeing, great class D amps still haven’t arrived. Most of them have issues with ringing. All of them have ultrasonic noise that sits on the outputs like standing DC. Upwards of .8 volts of noise!?!?!? That’s hifi, is it? A 1kHz sinewave looks horrible through those things. For all intents and purposes, it’s made entirely out of high frequency noise.

@kosst_amojan
Actually a 1KHz since wave looks good on most of them, without any noise. The ultrasonic noise to which you refer is the residual and is itself a low distortion sine wave at the switching frequency.

Show me where it is stated that the Merrill uses higher than 500K for its switching frequency. In fact, Merrill stated (on this site) that higher frequencies for switching were harder to do. I hope Mr. Merrill steps in here and clears this up. What he seems to claim is that the zero dead time using GaNs plus his zero feedback circuit are what makes the great sound.

The advantage of such high performance output devices is that they allow one to build a circuit that has no dead time circuitry in it. This means the circuit is a **lot** simpler! Of course, there is always some shoot-through current, so by going with a lower switching frequency a lot of consequences of that are avoided- one can use a slightly larger heatsink and the shoot-through heating is controlled. But the advantage of reduced distortion via no dead time is profound. So this should be a very nice sounding amplifier.

Go to Stereophile and look at the last 5 or 6 class D amps they’ve measured. Unless JA applies a low pass filter on the AP the output signals look horrendous.

@kosst_amojan
I did and found the review at the link below.
I think you may have misinterpreted something in JA’s measurement comments. The filter to which you refer above has to be there because all class D amps have a residual sine wave component, which has to be filtered in order to prevent the analyzer from having difficulty; in JA’s words:

As class-D amplifiers produce RF noise that could drive my analyzer’s input circuitry into slew-rate limiting, between the test load and the analyzer I inserted Audio Precision’s auxiliary AUX-0025 passive low-pass filter, which eliminates noise above 200kHz.

Note the cutoff frequency of the filter- 200KHz, and why it needs to be used. Nowhere in this article does JA say or imply anything that you suggest above. He seems to think the waveforms look pretty good.

https://www.stereophile.com/content/mytek-brooklyn-amp-power-amplifier-measurements

You need to understand better, it’s not the switching frequency "so much" but the "output filtering" needed to get rid of it that ’s problem, which gives phase shift problems right down to 3-5khz (the mids and highs of the audio band).

@georgehifi
This statement is incorrect.

Many such filters are 12db, but if only 6 db, that is where you use the rule of thumb that the phase effects extend to 1/10th the cutoff frequency. In the case you state above that would be 45KHz, not anywhere in the audio passband! If 12db, the effects are more constrained; IOW 6 db is a worst-case scenario.

@kosst_amojan
The Berning TF-10 is a tube preamp with about 10 tubes in it. It draws about 60 watts; that's about the same as my laptop charger.

Panasonic came out with a new formula for electrolytic capacitors some years back that allowed them to be a lot smaller with good performance. Chinese copies of that part proved to be unreliable but due to their price found their way into a lot of consumer gear- laptops, chargers, that sort of thing. So I've become hesitant to blame electrolytics for failures generally speaking, until I know more about the actual part used. If its one of those knock-offs, its going to fail no matter what.

Laptop chargers, battery chargers, TV's, you name it.

-Which don't draw any more power than the Berning TF-10, FWIW. I had a USB power supply fail on me when I took it on the Grand Divide Mountain Bike Route, but that was likely due to vibration more than anything else- the Tour Divide (as its known) has a way of doing that.

Electrolytic caps aren't famous for having exceptionally low ESR, but they're the only technology that's compact enough to provide the current capacity required. The sub-optimal ESR leads to internal heating of the capacitor. The heat dries them out, the ESR rises, that causes them to make more heat, they begin leaking electrically, and they eventually just short out and start frying silicon.

David Bernnig used a switching supply in his TF-10 preamp which was made in the early 1980s IIRC. I have a customer in town with one of those and it has the original filter caps in place- very much alive after all these years. He seems to have sorted it out- those filter caps run cold to the touch. The first statement in the above quote is false- electrolytics are the only technology that's compact enough to provide the **storage** capacity required, not current. 


SMPSs don't need a lot of capacity because the switching frequency is so much higher. This means that a much smaller capacitor can be employed than in a 60Hz supply. I know some SMPSs did have some problems in the early part of the 2000s, mostly on account of some Chinese caps that were failing not because of switching frequency so much as the parts were just plain defective.

Class D, as soon as your warranty runs out have fun getting somebody to fix it if & when it goes down, with all the surface mount components in new digital amps most techs won't touch them including me, (not wanting to use a magnifying glass or microscope) it becomes garbage or parts. I have one I use for a sub for my TV, but that's it. Very light weight power amps which is great, other than that no thanks, Ill stick to to things I can repair myself.

With these, you can:

https://www.amazon.com/Magnifier-SOONHUA-Head-Mounted-Magnifying-Replaceable/dp/B0742CJJM9
Surface mount can easily be done if you have the right tools. Most class D amp use dedicated chips; for example a single chip that includes encoding scheme, high and low side level shifting, driver circuits and dead time circuits. The outputs don't see the same stresses as conventional power transistors because the heating cycles are less extreme and it is heating cycles that ultimately damage power transistors (although conventional devices can handle well over a million heating cycles so its probably moot).
But most class D amps in high end use some sort of module so if one is needing repair, you just replace the module which is much easier than servicing the unit at the board level.

But no matter what, transistors switching off and on will always make huge amounts of high order distortion.

@kosst_amojan
It doesn't work that way IME. What makes distortion in a class D is how well the encoding scheme works. For example if PWM, any offsets generated by the comparator (or otherwise) can cause distortion. This means you have to be careful in such a circuit to block any DC that might come from your triangle wave generator- and in that circuit, the quality of the triangle wave is pretty important too.
Its not uncommon for the encoding scheme to have troubles at lower levels. For this reason I prefer PWM as its problems are more likely to be at higher power levels rather than at low levels.

Each encoding scheme has its upside and also difficulties! But the switching of the outputs themselves do not contribute to distortion, which is a major advantage of Class D, since the operation of class A, class AB and B this is not the case.

@fsonicsmith
If you get a chance, look into the Tour Divide Race, from Banff to Antelope Wells.
@vt4c
+1

Ah.

I thought we were talking about class D, not SETs. SETs FWIW don't (or shouldn't) get played much at full power; at normal listening levels the distortion can be near or at measurable.

I think it's pretty safe to say that an amp that makes 10% distortion at 2 watts isn't even hi-fi by the definition of the term.

@kosst_amojan
That is true- do you know of such an amplifier?

Having said that, Class D is the future for better or worse. There is quite a bit of cost savings involved for amp makers. They can be pretty good with a tube pre amp as a bandaid.

The main reason class D is the future is simple- the semiconductor industry really doesn't want to make linear devices so much anymore- not like they did in the 70s and 80s! Very many non-audio applications are better handled by switching or digital circuits that were handled by linear circuits just a few years ago- and now audio can be done that way too.

Strictly speaking, encoding is not used in Class D. There is no conversion of one type of data to another. Instead Class-D relies on feedback, and a great deal of it. It is a comparative technique.

That’s the nature of Class D and why it remains overall an analog process.
The Technics uses no feedback. It has at least 3 processing steps:

1 - Initial encoding from Analog to Digital
2 - Signal processing(alters the input signal to match the speaker behavior)
3 - PWM
Away with your nonsense, but I expect you will write four pages of barely related word salad to reply.

Thanks pretty funny!  :)

Feedback (self oscillating) is one of three methods of encoding. The others are PWM and sigma-delta (Pulse Density Modulation). PWM allows the designer to build a zero feedback circuit.

Some form of encoding/modulation is of course required to make a class D circuit. Strictly speaking :)
Since you think this is all hand waving, see:
https://www.analog.com/en/analog-dialogue/articles/class-d-audio-amplifiers.html
(a short introduction to class D amplifiers by a manufacturer of class D amplifier ICs)

There are also some amps like the latest Technics SE-R1, which are true digital amplifiers in that they convert everything from Analog to Digital, allow for phase and amplitude correction in the digital domain, and then produce an output based on a PWM output stage.

Just for the record Eric, such an amplifier is a hybrid. Pulse Width Modulation is an analog technique. In any Class D amp, some form of encoding is needed, and PWM is probably the simplest means.Even if the amplifier is computer controlled to adjust its encoding scheme to accommodate various codecs (in effect a DAC that can drive a speaker), ultimately its still class D and thus also analog.

^^ This is partly true as written.

George, the residual **is** the switching frequency. So this bit

so by the time the residual reaches 20khz what’s left with the 1.5mhz is down nearly 3 lower in amplitude and has less phase shift in the audio band than that of normal 600khz switching frequency.

doesn't make sense as written. I think what you are trying to say is that with a switching frequency 3x higher, the residual will be reduced (given that the filter in question is the same whether the amp is switching at 600KHz or 1.5MHz). Its helpful to express this in db; assuming a 6db slope the residual will be down an additional 9 or so db. Which is nice!

The 20KHz part has nothing to do with it. Whether you measure at 20KHz as opposed to 20Hz (if you change the timebase on the oscilloscope if you are observing either frequency as a test signal) the residual will be the same amplitude.

The Technics is the "blue" uses the GaN transistors, very fast quickly settling and reduced ringing, much closer to resembling the ideal grey square wave.
https://ibb.co/87Kh2mV

And now bench tests of Stereophile, use a special very low power high order -100db line filter between amps output and test gear input.
(Audio Precision’s auxiliary AUX-0025 passive low-pass filter),
Which eliminates noise between amp’s output and test gear input so reader can’t see it anymore.

1khz square wave with AUX-0025 filter in place:
https://www.stereophile.com/images/1212AM1fig03.jpg

1khz square wave without AUX-0025 filter in place:
https://www.stereophile.com/images/1212AM1fig02.jpg

Hm. Seems to make a difference about what you might know about what you are looking at.

In the first link above, no frequency is specified! Its clearly marketing and nothing else.
In the second set of links, what we are seeing appears to be the residual imposed on the square wave. Nothing unexpected. Since the filter is low power, so is the square wave depicted. These links IOW really don’t show anything other than that at the output of a class D amp, there is a residual sine wave as I mentioned before.

Really then no one one yet has been able to rid of this, and they are mightier minds than you. Only Technics has so far been able to reduce it greatly, with double the switching frequency speed and then the normal output low order filtering of it.

:) That’s funny! The reason the Technics has lower residual is because for a given filter turnover frequency (probably about 80KHz), the higher the switching frequency the lower the residual. No mystery- just faster switching times. IOW, what you referred to in the Stereophile links as the ’noise’ is a red herring- its not at all what you **seem** to think you’re talking about.

This morning on the morning show they said electric cars are not going to take off for another 15 years. Where are we at with class D right now? Everyone keeps saying that the future is class D but where is that in the future?

They’ve been saying that since the 1910s- but about 1912 Ford and Edison had an electric that could cruise at 60mph and had a 100 mile range, using Edison’s nickel iron batteries, which had a service life of about 25 years. Imagine what things would look like if Edison’s labratory hadn’t been burnt to the ground 2 weeks after Ford and he made that announcement! See ’Internal Combustion’ by Edwin Black.
Class D is very attractive because of energy, just like electric cars. The difference is that we can have Class D although IMO it still has a way to go.
@georgehifi

this then allows the low order switching noise output filter to do it’s job completely without leaving any effects and any left over switching noise artifacts within the audio band.

I think you don’t understand how the filter works, based on this statement. The Technics filter does not filter out all the residual, which is what the leftover switching artifact is called. The residual is always a very clean sine wave; the real question is what amplitude does it have? The filter really has nothing to do with ’left over switching noise artifacts’!! What can happen in a class D is the switching noise can radiate into other parts of the circuit where it can be rectified and amplified. Again, the filter has nothing to do with this- the noise problem is dealt with through good layout and compact (surface mount) design.
@minorl

In car audio it was really the only way to go to get the power output needed. you can’t do it from 12 VDC.

Just a correction: My Denon car stereo amp that I had in my Bronco (RIP) made 150 watts per channel and was an AB amplifier. It employed an inverter to boost the DC voltage. Inverters were used back in the old days when car radios had tubes in them, although they were a different technology, using something called a ’vibrator’ which was essentially a high speed relay that reversed the DC to the power transformer about 50 times per second.
@kosst_amojan

It’s also nice not having a massive pile of filters trying to turn garbage into a signal again.

This statement is problematic. If you listen to digital, something similar is happening there (and of course its an objection that analog guys often raise). And the Berning amplifiers employ a filter at their output too- and those amps are class A or AB despite a switching component that has to be filtered- and it gets very good reviews!


The filter on a class D is usually quite simple; there is not a ’massive pile’. If you raise the switching frequency sufficiently, the inductance of the speaker itself can be sufficient to attenuate the residual. The filter is there mostly to prevent the speaker cable from acting as an antenna for the switching frequency. Because the filter is usually set somewhere well below the switching frequency, the residual is a simple low distortion sine wave which won’t cause interference to higher frequency (radio) services.