Comment by milesvp | Hacker News Reader

Comment by milesvp | Hacker News Reader

Parent Full thread

I'm not convinced audio engineers are having that argument anymore (which is more the audience of IEEE). I get the impression from an engineer I know who's specialized in digital signal processing, and who's done a significant amount of analog work, that anyone who's still having some of these arguments doesn't understand some of the basic maths of the problem space. That basically, you can always digitally encode an analog signal at a higher resolution and fidelity than you can reliably read said signal at any given technology level/pricepoint. Which means that using an ADC -> DAC pipeline will almost always give you better audio output with better signal to noise ratios than strict analog -> analog systems (though possibly at the cost of latency).

Where it gets tricky, is the arguments for how things like tube amps sound. There is nothing that analog components add to a sound that can't be simulated with digital transforms, but the trick is figuring out how various analog components effect the signal. For instance, my friend found that certain tube amps would overdrive components and force them out of spec, and this caused a non linearity in the response of the signal coming from the component that gave the amp it's distinctive distortion effect. Once he figured that out he was able to reproduce the effect digitally. Incidently, he's found that some guitar amps can never be made today, because they were built around a specific batch of a given component, and newer batches didn't have the same out of bounds responses. But this seems to be why so many people are still so big into analog amps, because it takes considerable effort (and apparently engineering skill) to reproduce some of the classic sounds.

Mind you, I'm biased by one of the best engineers I've ever met, so I could be wrong about the general consensus with audio engineers. I'm also mostly using a sample size of 1, though I have corroborated some of his assertions with some of the blogging I've read on the topic. One of the best (I wish I'd bookmarked), was a good overview of encoding by a prolific codec writer, who pointed out that even though a digitized sine wave looks lossy at a low sample rate, it's trivial to rebuild that sine wave perfectly, with just the digitized version (so long as the sample rate was above some multiple of the Nyquist frequency?).

Likely that was xiph on why excessively high sample rates don't do much for the listening experience: https://people.xiph.org/~xiphmont/demo/neil-young.html

Amplifier designers most certainly are. Class D has significant energy efficiency advantages, and significant sonic challenges. It’s taking an analog waveform, whether decoded by a DAC or starting as pure analog, and chopping it into a series of pulses (pulse width modulation). The amp is switching transistors on and off very quickly, rather than using them as a variable linear gain stage. It’s not the same as PCM encoded digital audio, and shouldn’t be confused with it.

The energy efficiency advantages in class D are incredible. When my old Technics receiver stopped working because it started overheating, i replaced it with a class D amp on a half-credit card sized circuit board from dx.com and taped it to the back of the TV. 20W per channel, which is plenty for my casual TV watching. The Technics was the size of several stacked pizza boxes, with fans and heat sinks etc.

> Amplifier designers most certainly are.

Amplifier designers have to sell to a market that is rife with misinformation and a very poor signal-to-noise ratio between technical quality, profit margins and overall market success. I wouldn't expect many of them—especially the ones who have any name recognition—to be concerning themselves with objective technical performance.

Whereas the guy who designed the amplifier circuits for a product where the customer isn't being placeboed by the spec sheet—like powered studio monitors, or consumer devices like the Apple HomePod—is far more likely to be focusing on objective measurement rather than faulty market perceptions.

The thing is that transistors’ behvior is more linear and controlled when fully on or off rather than when used as continuously variable. Class-D can and does sound _amazing_ if built with good theoretical understanding. Check out Hypex amps designed by Dutch engineer Bruno Putzeys on strong control theoretical foundations.

That’s precisely and exactly the opposite of what ‘linear’ means when talking about the operating point of transistors. I think you should rephrase your argument.

I'm going with an understanding similar to what is presented e.g. here: https://www.mwrf.com/components/learn-meaning-amplifier-line...

"The term linearity derives from an amplifier's linear relationship of input power to output power which, in an ideal amplifier, would be precisely related by the gain of the amplifier."

I see it as matching my argument.

Could I ask you to give a definition which clashes? Thanks!

Yes, the definition of linearity is that f(A) + f(B) = f(A+B), which you have contradicted in your GP post. If a transistor is fully on, then doubling the input voltage to it will not double the output, because it's already at maximum. That is very precisely not a linear device. That's why linear amplifiers, like Class A, operate the transistor in a partially switched state, where increasing the input signal (about some bias point, and with feedback to turn a real transistor into a more ideally-behaving one) gives a corresponding increase in output signal.

Class D (switching) amplifiers are _not_ linear, they trade linearity for efficiency. The article you linked to actually says this:

"Over the years, a number of different biasing schemes have been developed for amplifiers, from extremely linear Class A operation, in which the active devices essentially remain powered on through the full sinewave cycle of an input signal, to more power-efficient schemes, such as Class D, in which multiple transistors are switched on and off to conduct different portions of an input waveform."

I think what you are doing is confusing the linearity of the transistor with the linearity of the whole audio amplifier treated as a black box, which is why I suggested you rephrase your argument rather than say you are categorically wrong, however what you said was:

"The thing is that transistors’ behavior is more linear and controlled when fully on or off rather than when used as continuously variable."

... which I'm afraid is really just wrong.

Thanks but no thanks.

When a transistor is off, it’s off. Then it takes a certain input voltage to get it going. Then there’s a varying curve to full output.

In contrast, fully switching on or off “gets rid of the curve” and tranforms the problem into one of timing, and one of a modulation scheme for a series of on or off states. In that respect I contend that it is an useful thought to consider transistors’ behavior (in an amp) as more linear and controlled if used in only the on or the off state. The output quality of Class-D amps in relation to price, size, complexity, and power, and cost suggests to me that the gains are not only in power efficiency.

Class D amps do leave the operation of transistors in their linear domain on the table, and do gain efficiency by that. Overall linearity as a control mechanism is something I stand by saying is increased. You may present an alternate take of course.

It's an interesting contradiction to build a more linear amplifier by intentionally using the transistors themselves in a non-linear fashion.

It is :) But it’s also only a contradiction on the surface of the words. The amplifier is quite linear even though the transistors aren’t used in what’s called their “linear region”, but I’d say they’re still used in a linear fashion.

I think you are confused. To get a linear amplifier, you operate your output transistors in a state of Saturation. This gives a linear transfer function to the drain based on the Gate to source voltage and you use a method of choice to hold the source constant. This is what a class A amplifier does. An AB amplifier has 2 transistors to extended the range of inputs on the gate for which the output is linear at the cost some distortion when switching between the A and B (hi and low side) transistors.

A class D amplifier uses a smaller amplifier internally (which can be a single transistor or more complicated) that is driven open loop so that it is either hard off or hard on but very fast and efficient. The hard on state is known as full compression because the output power will not change with input power. This is purely non-linear behavior and can be very easily verified by measuring the high IMD components it generates. Class D amplifiers employ significant filtering to suppress these non-linear terms and allow you to recover the signal. Mathematically, they violate the linearity condition of F(ax) = aF(x)

Also you should reread the article you learned from. It directly contradicts you.

It’s not about me; The “you”-ing is incredibly offputting.

We are simply and obviously applying the word “linear” a little differently. I am aware that it is a specific term for certain behavior / a certain operating mode of transistors. I’m referring to it in another wider context on purpose. As another example “linear” in “linear algebra” isn’t about saturation or gates. I’ve gone through this before in this thread and believe my perspective is now pretty complete and clear here and won’t engage further.

Actually the opposite is true. Transistors are approximately linear in half-open state, as in Class A, and non-linear at the extremes (when either fully open or fully closed).

Reply on news.ycombinator.com