Class D is just Dandy!

If there are actually audible benefits to GaN based amplifiers I look forward to hearing them.

There should be. GaNFETs are generally faster than MOSFETs (although MOSFETs aren't done by any means) and so the use of GaNFETs means you can run the amp with less dead time. Whether a designer takes advantage of this fact is another matter but generally speaking GaNFET amps should sound better as they have less distortion on that account.

Like Tube amps, or Pass (who makes it clear this is his goal) the distortion profile may have more to do with euphonic capability than merely lower distortion.

He is certainly right in that regard IMO. But we can hear that difference provided the distortion profile is also correct. The problem is that in most cases the difference between 0.05% and 0.005% isn't the issue- the fact that the distortion that is there is mostly higher ordered stuff that is highly audible even in those trace amounts. The correct distortion profile can mask those higher orders making the amplifier easier to listen to. But a class D amp tends to have a different distortion profile than a traditional solid state amp. Many of them using PWM encoding tend to have lower ordered harmonics too- which is the profile you're looking for.

And then without their output filter, the input wave form will be unrecognizable from the output, because of the HF garbage, and melt your tweeters silently in a nano second.

The operating principle of a class D amplifier is that it uses a filter to eliminate the switching waveform from the output. In this manner Class D shares the function of the filter used at the output of the Berning amplifiers. In either case, this is part of the operating  principle and can't be separated out; the idea that anyone would run one without the filter is ludicrous.

For those that wish to see what goes in and what comes out between Class-D v Linear amplification which is closer to the original input, or closer to "straight wire with gain" as what was the start of this

OK so here is the input, 1KHz

Test input:1khz perfect square wave
http://www.tronola.com/moorepage/Sine/1kHzsquare.jpg

But here we see a 10KHz waveform, which is difficult for many amplifiers. Imposed on top is the residual, which is about 250KHz, which is low by modern class D standards:

Output: Class-D:
https://www.stereophile.com/images/archivesart/406Halfig01.jpg

Note the scale on the left side of the image- its not there! There's no way to tell how strong this signal is, but the residual is a clue; its about 1/4 the total waveform amplitude. Since we don't know the power of the amp, we can't say how high this should be but if its 0.25 that would fit for an amplifier that makes 100 watts. If that is so, the amplitude here is half that of the waveform in the the next image, which is 1KHz instead of 10KHz!

Output:Linear Amp:
https://www.stereophile.com/images/archivesart/999A565fig2.jpg

George's attempt to put class D amps in a bad light on this account is debunked.

It appears instead there is an attempt to mislead people by a of slight of hand, which doesn't win street cred. The factor of 10 between 1KHz and 10KHz is pretty significant! In order to reproduce a 10KHz squarewave the amp needs bandwidth past 100KHz. That's hard for most amplifiers, tube, solid state or class D.

George, if you want to show how class D falls right flat on its face, show a 10KHz input waveform and the same 10KHz output waveform **at the same amplitude** of two example amplifiers. Make sure the outputs are shown at the same amplitude as well. Doing what you've done here is not just sloppy, its looks deceitful as well.

The biggest issue with input/output impedance is the change in frequency response. This is especially bad with tube pres as they usually have a high output impedance. Driving a low impedance input amp can affect the overall response and deviate from ideal. With purist tube pre's even the volume control setting can affect things because they lack additional buffer stages that would prevent this.

This is mostly misleading or outright false.

Output impedance **might** affect frequency response (in some cases, certainly not if the output is direct-coupled) but also affects distortion.

Tube preamps do not necessarily have high output impedance. That depends a lot on the design of the circuit! If it employs feedback, its output impedance will be fairly low and driving a 10K impedance should be no worries- check with the manufacturer (our preamps don't employ feedback and 600 ohms or less is no worries)! In a purist tube preamp, the volume control won't affect things (other than volume) despite there being no buffer; its all about how well the design is executed!

As an example, we make two tube preamps that have balanced outputs (and were the first in the high end audio world with such) and the load that they can drive makes no difference- the frequency response is flat, owing to a direct-coupled output. They also support driving 600 ohms as they were intended to support the old school balanced standard. Yet because they lack loop negative feedback, their output impedance is relatively high (compared to most solid state preamps), yet their distortion is very low driving any amplifier made with a balanced input. I've used our MP-1 to drive loudspeakers directly; how many solid state preamps can do that? 

Just sayin'.

I'm sure you make fine products, but would you then please explain the variance in all the tube preamps in the Stereophile section if not due to output impedance issues?

My prior post was a correction and not an attack. I was merely pointing out where the problems were and stated why, basic engineering principles included. More are below.

We won't allow Stereophile to review our products (we don't agree with their editorial policy which seems to be tied to their advertising and I know this from direct experience) which is one example of why if you limit yourself to their pages, you won't get the full picture.

This has more to do with the choice of coupling capacitor at the output of the preamp than it does the output impedance!

Please note that this phenomena has to do with solid state just as much as tubes.

Of course, the ultimate indicator is a graph of the output impedance vs. frequency. If you see it rising as it approaches 20Hz, this **might** indicate a loss of bass impact depending on the input impedance of the amp. The general rule of thumb is a 10:1 impedance difference between the two; as long as you hit that margin with the amp you have in mind its likely no worries.

The output impedance curve of our balanced preamps looks the same as their frequency response curves; we cut them off at 1Hz. So regardless of the load its driving, the preamp will have flat response from 1Hz to over 200KHz. Ours are not the only tube preamps with direct-coupled outputs that have ever been made- as a result you can't just assume that if it has tubes that it will have troubles making bass into a solid state amp with a 10K input impedance or the like. Generalities are often misleading that way.

A statement of fact is not an attack. You are taking this personally.

Here’s the work:

What, exactly, are you claiming is a result of the choice of coupling cap? Distortion or output impedance?

Here’s a formula for calculating cutoff frequency:
F=1,000,000/CxRx 2Pi

Normally you see this formula with a 1 instead of a million; I used the latter so that f is in Hz (-3db point), R is in ohms and C is in uF.

The coupling cap at the output of a preamp, in concert with the input impedance of the amplifier used determines the cutoff frequency.

example: a solid state preamp has a 10uf output coupling cap. The input impedance of the amp is 10K.

1.59Hz=1,000,000/10K x 10uf x 2Pi

We can see from this example that if a tube preamp has a 10uf coupling cap that it too will have a cutoff of 1.59Hz into the same amp.

This means there will be no appreciable phase shift at 20Hz so bass impact will be unimpaired, since the cutoff is 1/10th the lowest frequency to be played. A cutoff at 20Hz will mean that phase shift exists up to about 200Hz. The phase shift will cause the system to sound lean.

Many tube preamps **do** have such large coupling caps unless the designer has not done their homework (or has figured out that the larger the coupling cap, the more coloration it imposes, and so has elected to limit the capacitor size so as to get greater transparency). As a manufacturer you can’t forecast to what amps the preamp will be paired.

The size of the coupling cap will not affect the output impedance unless one is able to graph the impedance curve; if rising at lower frequencies the culprit will be the output coupling cap and otherwise not the output impedance of the preamp.

Now how much **distortion** the preamp makes can be affected quite a lot by the load that it drives. That is likely the more powerful argument for being careful about what preamp drives what power amp. Tube preamps often have very low distortion; in most cases its a good idea to have them drive a higher impedance so as to take advantage of that fact. Our preamps again are an exception- they regard 10K as an effortless load.

Please note:

This supports, not undermines, my statement.

yes, this is evidence of how this was simply a statement of fact and not an attack.

The biggest issue with input/output impedance is the change in frequency response...

I was very specific in what I was talking about. I did not say "the biggest issue with tube preamp sound quality." I said the issue with "input/output impedance."

As we can see, if one is to point at a tube preamp and blame it for a change in frequency response, the factor is not the output impedance (which is often only stated at 1KHz), its the coupling cap at the output. That is a bit different from ’output impedance’ and that is why I placed the correction.

I don’t "blame" preamps for changes in frequency response. I say high output impedance causes frequency response changes which vary based on the load. This is an irrefutable fact based on simple serial circuit analysis. Anyone with a basic understanding of AC circuit analysis would conclude the same.

And I showed the math for why that is not so: the first half of your quote here is false, the second half being based on the first is thus also false.

                                             **Do the math**.

Its the coupling cap at the output, not the output impedance that governs the frequency response.

Example: I've seen ARC preamps with 20uf output coupling caps. If you put them on a 10K load, they will be as flat as they are on a 100K load in the audio passband. Yet the very same preamp according to ARC should not be asked to drive anything less than 30K.

Your claim to which I was objecting was that the higher output impedance of tube preamps leads to frequency response errors and the simple fact is this is not so- it depends more on the timing constant that may or may not be present at the output of the preamp in question.  I showed the math. If you wish to refute this, then show the math.

Sorry- I've said as much as I can offer at this point:
1) its not a module but our own circuit
2) it looks like we have something to contribute in terms of technology.

Oh, and it will be balanced of course :)

Ralph, it appears to me that what underlies much of the disagreement between you and Erik is that he is viewing the impedance of an output coupling capacitor, if present, as contributing to and being part of the component’s output impedance. While you are not, possibly because you are considering "output impedance" as corresponding to "specified output impedance," which as we all agree is often based on a mid-range frequency such as 1 kHz. The capacitor’s impedance of course being unlikely to be a major contributor to the 1 kHz output impedance in just about any reasonable design.

Thanks Al! To be clear here, this was Off topic (we're working on our own class D circuit that is not based on any modules so you can draw your own conclusion about what my attitude about class D is); my main concern was was to try to express the idea that it was the output coupling cap in a given design that was determining the frequency response variation seen in some designs. Since there are tube preamps with a high output impedance that also do not have the rising impedance as seen (due to the fact that they have larger coupling caps), its hard to allow a generalization like 'high output impedance leads to frequency response errors' or the like. It doesn't have to was my point and I didn't have to look very far to find examples.

Put another way, its the rising impedance at low frequency, not the **overall** higher output impedance that causes the problem; I should have expressed it that way earlier!

+1 Exactly.


If you want to look at another very respected amplifier design that has a residual waveform imposed on the output signal, look no further than the Berning amplifiers. These amps are excellent; they are liked by everyone who hears them. They employ a switching power supply at their output which is modulated (loaded) by the operation of the power tubes. The switching supply thus has the audio signal imposed on its output. The switching frequency is then filtered out, leaving the audio signal to drive the loudspeaker.


If this sounds familiar to those versed in class D amplifier concepts, it should because its a very similar idea!

So how is it that such a respected amplifier as heard by all comers gets a Murphy while class D does not? They both have a residual. The conclusion can only be that the residual is not harmful in either case. 

For that matter, Bolt can also "compete" against a cheetah.

https://www.youtube.com/watch?v=VZuRTNidtCM

:-)

But I think we all know what the others mean by "compete," whether we agree with them or not.

Best regards,
-- Al

FWIW dept.:

@almarg , you might be interested in a book called 'Born To Run' by Christopher McDougall. In it, we find out that we are the dominant species not just because of our brains but also because of our ability to run further than any other animal. Not as fast for sure- its well-known that a cheetah is only good for short bursts. Its a fascinating read.

I've yet to see a class D amp keep up with a good tube amp, but that is on my Classic Audio Loudspeakers, which are not well suited to solid state in general . As a result, I'm sure there are those that would say they are 'difficult to drive' despite being 16 ohms and 98 db 1w/1M  :)   On a different speaker I would expect the situation to be reversed.  Just saying- its important to include the speaker when talking about what is 'competitive'; blanket statements can get tricky in high end audio.
 

You can use a Tube stage to slow down the signal or other circuit designs in solid stage to do that as well.

Agreed- **and** FWIW, there does not have to be anything slow about tubes. Keep in mind that in order for color TV to have existed, tubes had to be able to operate with bandwidth over a range of multiple MHz.

Tubes are only slow because of the design of the circuit, not because of tubes in general. In fact you can build a class D amp using tubes as the output switching devices, and switch them at some really crazy high frequencies- in excess of 100MHz- and likely with no need for dead time circuitry. That's pretty fast! I'm pretty sure solid state isn't there yet (but that's off the top of my head; haven't checked).

Don't even need a working amplifier to try this, just hook up two tubes and have them switch back and forth as you suggest. Please post the pics of the results. By pics, I mean of the smoke. :)

Class D was first proposed in the 1940s. The problem isn't the switching, its finding tubes that could manage the current :)

But any speaker manufacturer worth his salt will tell you that doubling up on low order high power filters at or around the same area has it own set of problems with ringing ect.

While this is true, this does not translate to the filters used in a class D amp, which is operating well outside of the audio band. The filter on a class D amp is blocking the switching noise of the amp and nothing more; it does not pay to have that filter go any lower than need be on order to be effective.

If the amplifier is switching fast enough, the internal inductance of the speaker itself is often enough to block the switching noise.

If this "effect" has been remedied, or if I was just imagining it all, maybe a summer amp?

Design plays a huge role! Some amps are not very involving and some are.

Some class D amps employ Delta-Sigma processing, some are simple Delta, some are self-oscillating (the latter require loop feedback to operate correctly, the former two can operate with zero feedback with careful design).

Scan frequencies can cover a wide range. Some have dynamically variable scan frequencies. Some amps employ op-amp input circuitry for the audio, others are discreet and its even possible to have passive input circuits.

So with all these variants you just have to try one out and see if it works for you.

Delta Sigma processing = digital. :)

Most Class D amps are not digital. Weird. :)

Delta Sigma is not digital. It is a process used for digital encoding and there is a nuance there; its safer to say that data in a memory system is 'digital' (1s and 0s). It is also a good process for encoding for class D operation and is attractive because it allows for direct playback of a class D amp from a digital source, although unless there is some upsampling the clock speeds of most digital sources are far too slow for good class D operation.

All, not most, class D amps are an analog process. The funny thing is, if you want, you can use digital chips (for the most part, inverters) in the class D process. But it is very much an analog process.

@erik_squires 

If you use a delta-sigma encoding process, you are converting analog to digital, no?

No- not exactly. You are converting from analog to a pulse train. Very convenient for going digital from there, which is why its used in the digital encoding process. But despite that its still an analog process.

I get that this is hard to understand but when you see how much its affected by layout, the need to squelch oscillations and the like, that's when its obvious that despite the way its used, its actually an analog process. Digital stuff does not oscillate. Analog stuff can.

A lot of people think delta-sigma (also known as Pulse Density Modulation or PDM) is digital though (specifically 1-bit). I invite them to build the circuit; in solving the problems of making it work despite it being the circuit in a known-good diagram, that's when you really get that most circuitry that handles digital signals is actually analog. Put another way, the resulting signals are *treated* as if they are digital. I think some would say I am cutting the hair pretty fine, but that really is the truth of it.

BTW, we've moved beyond proof of concept and our scan speeds are up to 0.5MHz now (target is 3MHz, but we won't be able to do that with our through-hole prototypes due to afore-mentioned layout issues that are prone to oscillation at the higher rates). Its sounding pretty good too.

@islandmandan 

I'll be very interested to see what any of your replies might have to say.

What sort of noise are you encountering?

@islandmandan , I'm with Al on this one. The amp should be fine with no input connected, go ahead and disconnect  the input and see if you still have that noise. If so => amplifier, if not => input

@islandmandan , I don't have a good answer for that right now- but when we do this test then we will know more.

^^Good work- Glad you got it sorted!

Oversimplifying, 'Hybrid' means that the digital source is upsampled to the scan frequency of the amp.

@erik_squires

That’s not what’s being done here. DSP is used entirely for electrical phase and amplitude. The actual speaker/room response is not considered. Quite a heavy handed approach to making an amp perform like an ideal voltage source.

That may be so as to avoid using global loop feedback, which is known to exacerbate higher ordered harmonics. Speculation on my part...

@erik_squires - worth a shot, class D modules are cheap...

George, its more complex than that.

The reason Technics is switching so fast is to reduce distortion. The filter has little to do with it. At their speeds, the inductance of the speaker is sufficient. So their filters are mainly concerned with preventing RFI.

Keeping the switching speed high is important for resolution, and decreases distortion. The problem is that the faster you switch, the more time has to be alloted to allow the devices to turn off before its mate can be turned on.

This waiting time is called 'dead time' and increases distortion. so there's a bit of a catch-22: the more you try to decrease distortion by increasing switching speed, the more dead time you have to have and that increases distortion.

Technics' solution is by using Galluim Arsenide devices that no-one else can get, which are a lot faster (and can switch considerably faster than they are actually being switched in the amp). The reason they are doing this is to minimize dead time, and so have created one of the lowest distortion Class D amps made.

We're taking a different approach. We found a way to eliminate dead time altogether. This allows us to switch at a lower speed and still get lower distortion, or switch at a speed Technics is doing it, without having to use devices that are as fast.

10 years ago class D had already been around for a while, but back then they weren't bringing home the bacon against the prior art. But it was obvious even ten years ago that it was a rising star.

In the last few years though its become a technology to be taken seriously. So about a year ago we began working on our own design.

Despite this, I can't call it a mature technology, a since price/performance curves define what is mature, and the technology is still improving at a rapid rate.

The major problem, as George points out, is the switching speed, but in recent years that is a problem that is fading. Part of the issue of switching speed has to do with the introduction of dead time, which makes distortion. But there have been a lot of advances in relatively cheap semiconductors recently, and the result has been that for a given switching speed, there is less dead time required because the newer devices are so much faster.

This means also that higher switching speeds are showing up.

Somewhere in this, a threshold is being crossed. We see this with the many responses on this thread. Class D, while like any other technology that has its better and worse executions, has arrived.

Its my opinion that any amp manufacturer that ignores the implications of class D is doing so at their own peril.

Looks like our first patent in the field will be filed soon...

1 - Direct Digital

@erik_squires FWIW the phrase above is a marketing thing. It means that the amp has a DAC as its input, but the amp itself is actually analog, like all class D amps are. IMO/IME marketing terms like this tend to confuse the marketplace.

@tomcarr ’PWM’ stands for Pulse Width Modulation.
Such an amplifier usually has something like a triangle wave oscillator in it. The incoming audio is compared to the triangle wave and this determines how long the output devices are either on or off. The switching frequency (which is otherwise determined by the frequency of the oscillator) is then stripped from the resulting signal and what’s left is the audio.

I don't find good digital amps bright or fatiguing today.

What's a 'digital amp'??
Surely not a class D- I thought this thread was class D only ;)

We've not put a lot of thought into what embodiments the amps will take, although an integrated seems likely. Probably some sort of power amp too.

I think the amp is likely to be less than $4k.