Difference betweeen Class D and "Digital Amplifiers"

[roffe]

Since it matters whether your signal is 1's and 0's and not 0.9973...'s and -0.002...'s so to speak, the signal is not digital, because not the encoded bits themselves matter, but the voltage and the timing. You are also suggesting that it is the filter that turns the signal into analogue. Not so. It removes the HF from a squarewave, leaving the LF. I've yet to see a filter that takes a digital input signal and produces an analogue output. It takes a squarewave and produces a smooth signal but that's quite a different affair.

Your dictionary is entirely right. Since you cannot write down your reference node signal other than in a very precise oscillogram (because the difference between 0.9973... and "1" matters), your signal is analogue.

It's so simple once you get your head round it.

First of all, I would like to thank Bruno for sharing his vast experience with us. I find the digital vs analogue discussion interesting, but one point is bugging me. Bruno stated in his first posting:

Digital means "symbolic". A digital signal symbolises a stream of symbols. A symbol is an element of a limited set. The nice thing about digital is that inaccuracies in the transmission of an analogue representation can be rectified as long as you can still recognise what the symbols are. Only when the inaccuracy is so great that you can no longer tell what the transmitted symbol was, the message gets corrupted.

Now, while it is correct in the electronic world that even a binary symbol (i.e. 0 or 1) have deviations from the reference values, I find it hard to agree that bitstreams even with distortions are analogue. The reason is that both the human eye (when presenting the bitstream om an oscilloscope, for instance) and a computer easily can interpret the bitstream as the symbols they are (0 and 1). By using electronic buffering and accurate re-clocking of the bitstream, the original signal will be free from distortion for any practical purposes, and truly are symbols (one can even use them for digital calculation, if desired).

I agree that a signal is only digital if the receiver "perceives" the signal as digital, but in my opinion an "interpretation" stage with comparators, buffering and re-clocking actually do perform exactly this interpretation. And in that case, the bitstream is digital. Further processing in the amplifier, whether digital or analogue is then done on a clean signal, and even if further processing is analogue, there complete unit must be said to accept a digital signal rather than an analogue signal, due to the interpretational nature of a comparator/buffer/reclocking module. This do in my opinion constitute a "digital amplifier" in the sense that it takes a digital signal as input.

Now, for the quality of class D vs traditional class A/AB amps. Bruno writes:

As for listening pleasure, analogue wins hands down. Some of the initial appeal in digital class D turns out to be a boost at the end of the frequency range caused by the fact that normal speakers have an impedance way above nominal at 20k, under-damping the filter. This is a classical mastering engineer's trick to put some "air" in a lacklustre mix, but it's not something your amp should do for you. Once you recognise that, the sheer neutrality and transparency of a good analogue class D amp coupled to a decent DAC is something you'll never want to trade back in.

I am under the impression that most successful class D amps use an internal frequency so high that does not require an output filter close to the audible frequency spectrum. Natural variations of impedance causing underdamping of the output lowpass filter will then, if I get this right, not be an issue in the audible spectrum anyway. This filter is present in most class D amps, whether the inner operation is digital or analogue if I have got it right.

I do however agree that a class D amp works better with some loads than others. My most recent DIY loudspeakers are built and tested using ICEpower-based monoblocks, and they work quite well, outperforming any class AB amps I have tried in listening tests. But since the loudspeakers have been created using said monoblocks as reference, this is to be expected. As a side note, the ribbon tweeter on these loudspeakers have a flat, resistive impedance of 8 ohms, and the output filter will then work quite well, NOT showing any underdamped characteristics.

Bottom line when it comes to amplifiers an their loads (loudspeakers) I suspect to be as it always have been:
It is a matter of matching the two for the best end result.

 

[Bruno Putzeys]

Now, while it is correct in the electronic world that even a binary symbol (i.e. 0 or 1) have deviations from the reference values, I find it hard to agree that bitstreams even with distortions are analogue.

I'm saying quite something different about bit streams. A bit stream is digital when the subsequent process only looks at the bits encoded in them. It is analogue when the subsequent process is interested in the shapes. The input data stream to a computer is digital, because the computer is only interested in ones and zeroes. The computer couldn't care less about the exact shape of the data, as long as the distortion isn't so gross it misreads the bits. A bit stream going into an analogue lowpass filter is analogue, because distortions in the signal will result in distortion of the output of the filter. So, whether a signal is digital or analogue depends on what is done to it next, not on what it looks like. A signal can look very analogue (e.g. output of an ADSL modem), yet be digital (minor errors do not alter the meaning). Likewise, a signal can look very digital (a PWM signal), yet be analogue (any error, however minor, finds it way to the output). It's clear that you see things the same way, but there was some confusion about the paragraph you are quoting. It pertained to a signal found in a certain design after the last reclocking stage and before the first analogue processing stage (a control loop), and argued therefore that the amplifier couldn't possibly be termed digital since it contained explicit analogue processing between the final reclocker (in fact a DAC) and the power stage.

I agree that a signal is only digital if the receiver "perceives" the signal as digital, but in my opinion an "interpretation" stage with comparators, buffering and re-clocking actually do perform exactly this interpretation. And in that case, the bitstream is digital. Further processing in the amplifier, whether digital or analogue is then done on a clean signal, and even if further processing is analogue, there complete unit must be said to accept a digital signal rather than an analogue signal, due to the interpretational nature of a comparator/buffer/reclocking module. This do in my opinion constitute a "digital amplifier" in the sense that it takes a digital signal as input.

The input signal to a "digital" amplifier is certainly digital. Earlier I presented the signal flow of such an amplifier, showing that the signal "becomes analogue" after the PWM stage and before the MOSFET driver. Yes, the input is digital, but the amplifier certainly isn't.

I am under the impression that most successful class D amps use an internal frequency so high that does not require an output filter close to the audible frequency spectrum. Natural variations of impedance causing underdamping of the output lowpass filter will then, if I get this right, not be an issue in the audible spectrum anyway. This filter is present in most class D amps, whether the inner operation is digital or analogue if I have got it right.

The filter is instrumental in allowing the amplifier to operate as a "power converter". The coil and the inductor provide temporary storage of energy throughout the switching cycle. So-called "filterless" amplifiers rely on the inductance of the speaker coil for energy storage. They are ridiculously inefficient into purely resistive loads (although the extra energy is dissipated in the load, which explains why it's so often overlooked).
The filter response with load variations is not necessarily a problem. At least not in analogue class D's. Analogue class D's can be designed to include the output filter in the control loop. Quite a few are. Mueta and UcD amplifiers produce a frequency response which is completely independent of the load and of the filter components and hence they do not suffer from speaker matching problems at all. IcePower amps are less load-independent, but still much more so than any digital class D amplifier. Once you learn how to make the output filter part of the control strategy, you start seeing it as a solution, not as a problem.

 

[itsmrdavid]

Hi all...

I've read through this, and thought it over a bit, and wanted to
put in my own two cents on this silly topic...

Class D is most certainly NOT a "digital amplifier."

According to me, "digital" referrs to symbols. It is a meaning or
intention which is superimposed on any media.
This could be anything, including morse code over radio;
radio teletype; baudot; computer modems sending tones
over phone wires; voltage on wires between computers, which
we call a "network; light flashes between navy ships, spoken
word on a radio, cassette tape or CD; messages conveyed
by written letters and words on paper; etc...
The "digital" part is the actual message contained in a medium.

"Analog" is simply a signal which can be manipulated in a
number of ways, including amplification. It exists
independent of any meaning which is intended.
It may contain "digital" data which can be reproduced
in some fashion by a recipient.


Illustrative example:

Man "A" sits down at a radio and sends out the number "60" by
morse code.
Man "B" recieves the number "60" by interpreting the morse code
recieved via his radio, which was sent by man A.

Now - what has been amplified?
Was the "60" amplified?
Or was the signal that carried the "60" amplified?

The answer should be obvious.
"60" was the intended message, and was not altered
in any fashion.

How, then, would you go about amplifying that "60"?
If you're talking to an accountant... you may find that
"60" amplified by a factor of two is "120".
But that's amplifying the meaning, not the signal.
If you want to amplify the signal, you'll end up yelling
at your accountant.

"analog" is a fluent ever-changing thing, which exists
solely as itself, and can also carry a digital message.
In the case of music, such an analog communication can
convey things as esoteric as emotions and aesthetics;
while also conveying a message digital in nature, like
words, numbers or concepts.

"Digital" is simply a static message, represented by symbols,
regardless of how they're sent or recieved.
Red is always red no matter which language is spoken.


Now, closer to the topic - consider what happens when music
is recorded: you have an analog signal containing whatever
meaning may have been intended by the artist; this is
quantized as to voltage levels by a digital device. It can then
be re-encoded in a variety of media, including a DAT, a CD,
.wav file, or the almighty MP3 on your hard drive.
That is, an analog media called a song --> digital encoding
--> analog storage media for digital information, like a hard drive.

Now reversing this process, the digital pattern of the music is
interpreted from the media it was recorded in, and reconstituted
in it's original analog form.

The "digital" part is never amplified. It is only encoded and
interpreted. It is not the media, or the signal conveyed by
the media. It is the meaning placed on the signal or media.

So - there is no such thing as a "digital amplifier".
That's a terrible misnomer.

~David

 

[Bruno Putzeys]

Amen

 

[roffe]

Now, closer to the topic - consider what happens when music
is recorded: you have an analog signal containing whatever
meaning may have been intended by the artist; this is
quantized as to voltage levels by a digital device. It can then
be re-encoded in a variety of media, including a DAT, a CD,
.wav file, or the almighty MP3 on your hard drive.
That is, an analog media called a song --> digital encoding
--> analog storage media for digital information, like a hard drive.

Now reversing this process, the digital pattern of the music is
interpreted from the media it was recorded in, and reconstituted
in it's original analog form.

The "digital" part is never amplified. It is only encoded and
interpreted. It is not the media, or the signal conveyed by
the media. It is the meaning placed on the signal or media.

So - there is no such thing as a "digital amplifier".
That's a terrible misnomer.

I'll have to disagree with you here. It is true that digital means something represented by discretely distinguishable symbols, as opposed to analogue representation, where there are no fixed values.

But what you are talking about is amplifying data, regardless of representation. Data is representation without meaning. Now, let us move from data to information. When we know what the data represents, it gives more meaning to talk about manipulating it. It would perhaps seem like amplification would be appliccable to uninterpreted analogue signals, but this is simply not true. Take the case of an AM or FM radio signal, for instance. Or even a television broadcast.

I think it is pretty clear that you cannot talk about amplifying DATA, but you can talk about amplifying INFORMATION (as long as the information is a quantification of something). This is regardless of its data representation, which may be digital or analogue.

In fact digital amplification and processing is done every day, not only in pure computers, but also in Home Theater Processors. A digital volume control operating on digitally represented audio is in fact a digital amplifier.

But I do agree that most class D power amplifiers are not digital amplifiers. That said, it doesn't mean that the term "digital amplification" is meaningless.

 

[Bruno Putzeys]

Whoa! Deep semantics here! It sounds a bit of a stretch to me to describe digital gain (a multiplyer with a constant factor) as "amplification". I think we need to consult the IEEE dictionary here but my hunch is that amplification as a term applies to analogue signals only.

To effect gain in the analogue domain, one uses an amplifier. To effect gain in the digital domain, one uses a multiplier. Amplification is a voltage and current thing, multiplication is dimensionless.

(What is called a multiplier in the analogue world is an amplifier with a voltage controlled gain and where "gain per volt" is required information, because you can't get volts squared out.)

 

[Pano]

Would not an amplifier with 1 channel per bit be a digital amplifier? You know the type, e.g. 16 switches assigned one to each of the 16 bits. From LSB to MSB the voltage switched is increased as we move up the bits.

I ask, because back in the early 90's I tried to build such an amp as a means to an end - to drive a digital loudspeaker. My results were poor, but the concept is good. Even if used to drive an analog speaker, it would still be like a giant A/D convertor. Isn't that digital "enough."

 

[Bruno Putzeys]

Your circuit produces a voltage and delivers a current and switches at certain times. Voltages are analogue. Currents are analogue. Time is analogue. Only numbers are digital.

I'm reminded of how difficult it is to make engineers think in the current domain. Everyone just looks at voltage. Here too. Just because the voltage in a class D amplifier looks like a square wave, people think it's digital. If only they ever thought of the currents in the power stage...

If they were to measure the current in the MOSFETs of any switching amplifier, they'd drop the idea of calling it digital straight away. The current in the MOSFET is not a square wave. Its a chopped up version of the loudspeaker current! Now, is the loudspeaker current digital? No. Therefore, is the amplifier digital? No.

 

[Pano]

So Bruno,
for you only the infomation can be digital? Digital can never do any work, i.e. supply current?

Also: If time is analog, what happens when we mix it with a digital signal? Do we get a hybrid? To store digital information, or even to transmit it, we are not tied to time. But to record or play back audio we are. (The same could be said of of a vinyl disc.)

At what point do we "break the plane" from analog to digital, then back again?

 

[SSassen]

Sigh, it is soo simple really:

Information is that which you'd like to get across, the medium by which you do this doesn't really matter as long as the information gets transferred from A to B.

If the information is in digital form, ie. 0s and 1s, this is just the form by which the information is represented. But for the sake of the argument lets call this digital information.

If that digital information is transferred from A to B over a piece of copper wire with the 0 represented by a 0-volt signal and the 1 by a 5-volt signal then the signal travelling across the wire from A to B is analog.

Why? Well, although we labelled it digital information it still is an analog voltage that switches from 0 to 5-volt. Try this analogy instead: if we had opted to use a optical cable between A and B, would you be calling the flashes of light digital as well? No, they too are analog.

Ergo conclusio: the term digital doesn't describe the actual signal, all signals are analog by nature, it is just the tag we attribute to the information we try to get across. Digital is the discrete mathematical form, analog is the real world carrier of the information.

Best regards,

Sander Sassen
http://www.hardwareanalysis.com

 

[itsmrdavid]

Take the case of an AM or FM radio signal, for instance. Or even a television broadcast.

I think it is pretty clear that you cannot talk about amplifying DATA, but you can talk about amplifying INFORMATION (as long as the information is a quantification of something).

I understand where you're coming from regarding data and
information. But i think you may have missed my point.

Yes - information can be amplified -- Of course.
But then it's not the same information. Digital, by definition
is a quantification of something. If this quantification is
"amplified" as you say, it's not the same quantity.

It's like a reporter taking what you say and embellishing it
for the tabloids...
"Actor kisses actress" becomes "actor gets down and dirty
with actress for two hours in a dimly lit restaurant."
It just isn't the same information. It's been "amplified".

Contrast that with the search for clarity and lack of distortion
in signal amplification. This, optimally, preserves the essence
of the signal. If we amplify it, we want to preserve its integrity.
This is analog (analogue if you please). When amplified, it is
similar, or analogous, but at a larger volume, voltage level,
or whatever.

Whoa! Deep semantics here! It sounds a bit of a stretch to me to describe digital gain (a multiplyer with a constant factor) as "amplification". I think we need to consult the IEEE dictionary here but my hunch is that amplification as a term applies to analogue signals only.

Yeah - sematics... sorry about that.

Here's an interesting point though: an "op-amp" or operational
amplifier is called so because it approximates a mathmatical
"operation".
An example is when an op-amp is conencted as a simple inverting
amplifier. If the input resistor and the feedback resistor are the
same, the gain is zero. If the feedback resistor is double the
input resistor, the gain is 2x. Simple. Fairly mathmatical...
But even so - this is only an approximation to a mathmatical operation. It is "analogous", hence the term "analogue".

Digital comes from latin "digitus" meaning "fingers".
It is now used to refer to numbers and numbering systems.
In terms of information, this would be symbols defining
some message or communication.
Such messages are superimposed on analog media, and
must be interpreted.

Amplifying numbers makes DIFFERENT NUMBERS.
Enough said.

Ergo conclusio: the term digital doesn't describe the actual signal, all signals are analog by nature, it is just the tag we attribute to the information we try to get across. Digital is the discrete mathematical form, analog is the real world carrier of the information.

Well said!


/me scrys digital messages from the gods in his alphabet soup.
mmmmm... (joke)

 

[BWRX]

An example is when an op-amp is conencted as a simple inverting
amplifier. If the input resistor and the feedback resistor are the
same, the gain is zero.

I think you meant to say the gain would be -1. Although zero gain could be read and interpreted as no gain, which could mean the output is the same as the input, which would be unity gain.

I'm not touching the rest of this discussion with a 10 foot pole

 

[Pano]

Sigh, it is soo simple really:

Sure, it's simple. Digital is abstract information in the form of numbers. Usually in binary form. Everything else is analog.

But the term "Digital", like any other word, suffers from definition creep. Engineers may like to define the term very strictly, but it rarely gets used that way. E.G. a digital recording. Certainly the information is stored in an analog manner - pits and lands, magnetic pulses, whatever. But would it be wrong to call it a digital recording because the medium and the transportation of the signal are fundamentally analog? If it is wrong to call it such, then how do we distinguish it from a purely analog recording? Do we call the digital signal "DCA"? Digitally Coded Analog?

What to make, then, of my "Digital Ready" headphones? Where will I ever find a digital amp to power them???

 

[Bruno Putzeys]

Definition creep is a particularly insiduous feature of "common language". In science and engineering, terms are precisely defined specifically to prevent this. If a marketing person writes an ad using a word in an incorrect context, we, as practitioners should not follow suit.

 

[SSassen]

Definition creep is a particularly insiduous feature of "common language". In science and engineering, terms are precisely defined specifically to prevent this. If a marketing person writes an ad using a word in an incorrect context, we, as practitioners should not follow suit.

Hear, hear! I second that

Sander Sassen
http://www.hardwareanalysis.com

 

[Yves Smolders]

I had the discussion in class a long time ago... analog vs digital, and it works here as well:

"Numbers" not = digital.

To represent a digital signal you not only need "numbers" in a row - you also need to define how to interpret them. A predefined timescale for example, or a matrix of some sort to hold them.

For example PCM=16 bit numbers... but it's also defined with 44khz sampling interval. Digital. For example, the 7th sample of the 1st second of audio has value 32768.

A TIFF or JPG also has 2 dimensions and values for each cross of these dimensions. Digital. For example: Row 10 and column 7 has an RGB value of 3FA.

When you are running a comparator against a analog signal, you can't say it's a train of 0's and 1's... it's turning on or off a transistor, but *not* for a fixed, defined time interval - there is none. The transistor goes on and off, but not in a predefined scale in time. This is analog. There is no buffer, nowhere the 0 or 1 is stored or put in a register (well you could see the transistor as a register...) Writing down the on and off values as a list of 0100101 won't help you a bit - you don't know how long the on and off states were held in the transistor, it's predefined nor stored. That's how I see class-d as analog, except when the modulation is derived from a digital signal directly.

It's like taking a photograph: A bunch of cristalline stuff showing you colors, it's analog. there's no intensity value to the cristals. Yet it does give you an image of a point in time.

A comparator also checks if a signal is higher or lower in amplitude than the wave you're comparing with - it doesn't really need to put a "value" on it.

 

[el`Ol]

I find the question noise-shaping or not more important than the question digital or analog.
44100*2^16 is about a hundred times as much as 350000*256, so PWM without noise-shaping and a slow time-base in my opinion is only useful where less high-frequent parts, less loss and less heat is desired and the dynamic range is not the point.
I find it interesting that both all the digital amps and the analog mueta concept based on hysteresis work with about 350000 Hz and noise-shaping.
Seems to be the best compromise for high fidelity.

The difference that remains is the load-independent frequency response of the analog concept because of post-filter-feedback.

 

[Yves Smolders]

haven't heard any of these type class-d's before (at least not that I'm aware of, the things are hidden everywhere...) - only UcD types, so can't comment on that.

Never got the concept of noise shaping btw...

 

[Bruno Putzeys]

Some people make a big thing about the difference between noise shapers and control loops. The maths (at least in their linearised form) are pretty similar though. A noise shaper can be viewed as a feedback loop in which the error to be corrected is a sampling quantiser (the rest of the loop is usually sampled too). The most notable difference between a sampled and a non-sampled control loop is that as the error is reduced in one frequency range (usually the audio range), it is compounded at higher frequencies. Gerzon's noise shaping theorem states this quite nicely. It says that if you make a plot on a decibel y scale with the frequency on a linear x scale of the noise floor without noise shaping (flat) and the noise floor after noise shaping, the area of the shaped curve above the unshaped curve is equal to the area of the shaped curve below the unshaped curve.

With unsampled control loops things are less obvious because there isn't a "hard" frequency limit, it's usually just a second pole that compromises phase margin. You do get an increase of the error around the unity gain frequency.

Class D amplifiers are sampled by nature. However, analogue class D amps like UcD and Mueta do not have sampled quantizers. Because of that, we don't call the control loop of a class D amplifier a noise shaper. These amps do not have quantisation noise.

Noise shapers are used in digitally controlled amps, because the modulation is time-discrete and hence represents a sampled quantiser. These noise shapers operate without heeding the output filter or power stage, so they do not fulfil the function of control loop, although they could double as one if feedback were taken using an A/D converter (discussed earlier in this thread).

 

[Yves Smolders]

Bruno,

Thanks for your comments on noise shaping. It was hard to understand (wiskunde was nooit m'n beste vak...) but your post and the Noise Shaping page on Wikipedia led to some understanding!