ASR FAQ: Basics & Hard Stuff [Inactive]

[pozz]

Let’s address stuff that’s floating around in our culture. Plenty of it gets mentioned in passing in threads, and plenty of new ones feature repeat topics. I'd like to keep everything easy to follow, graphless and mathless, but also provide everything necessary to begin engaging the technical side of it.

Not sure how to organize it all or even if this is helpful. I'll add links to each question as it goes.

Any suggestions for topics and so forth? Additions and corrections are also helpful. I have a lot to learn about each one, too.

Nonlinear distortion?
Nonlinearity describes a change in output that isn’t proportional to a change the input. The distortion in this case indicates the difference between that nonproportional change and an ideal change.

More specifically, in audio, nonlinear distortion indicates that the signal you sent to the input isn’t the only thing that’s present when you examine it at the output. There's a distribution of power: you are given not only the signal, but something extra. That something extra has a real power value in addition to the power value of the signal. Whenever the audibility of nonlinear distortion is discussed, it's really about whether or not the collected power of that additional spectral content is enough to cross the threshold into sense perception.

Linear distortion?
By definition, and this will bring a little smile to your face, flawless amplification, attenuation, delay or advancement of a signal is a type of linear distortion. Also by definition, linear distortion is completely reversible.

When looking at output frequency response, any deviation from the spectrum of the input, with the amplitude being either higher or lower, or phase either lagging or leading, indicates distortion introduced by the device.

Even vs. odd order and harmonic vs. inharmonic distortion?
Don’t be fooled by the word harmonic, or the emphasis on even order distortion.

Some vocabulary: an octave in acoustics is a doubling of frequency, while an octave in music is the same note shifted upwards one register—the fundamental of that note, its lowest frequency, also doubles. The note itself is a collection of frequencies in different phase, some of which will be harmonically related to fundamental, some not. Certain frequencies, if they are related to the resonances of an acoustic instrument and show up strongly, will be called formants.

Second harmonic distortion (or H2) of a 1kHz tone will produce a 2kHz tone, and the resemblance with acoustic instruments ends there. Playing higher and lower on the latter will excite different harmonics, formants and phase relationships, so examining notes an octave apart won’t show an exact doubling or halving of spectra unless it’s done artificially, through electronic means.

Chords are combinations of notes whose intervals follow certain ratios (4:5:6 for a major chord, 10:12:15 for a minor chord), not necessary based on the octave (2:1)—try 160:192:231 for a diminished chord. The more instruments are added, the more dense the resulting spectrum, the less clear the mathematical relationships in a given musical passage. Overlaying a 2:1 upshifted copy of that spectrum on top of the original could easily produce strong, but inharmonic frequency components. Mix 100% H2 (equal strength original and harmonic) into music and see how well it fits.

The higher order the harmonic produced by playback gear, the less you can count on masking to work. There's little inherently musical found here. It has to be shaped for individual components of the music, applied in some areas but not others.

Class A, B, AB, D, G, H (and onwards) amps?

An amplifier superimposes mains power on the weaker incoming signal. All that’s necessary is for the first to be added to the second in a controlled manner. The operational class defines how this is generally achieved, also called electrical bias.

Like everything, there are understood compromises:

• Class A, B and AB amplifiers are less efficient, with between 30%–70% of power dissipated as heat, which requires heavy heatsinks and large (usually linear) power supplies.
• Class G and H are variations on Class AB designs, and are generally more efficient, dissipating 30%–50% as heat, while Class D amplifiers are even more efficient, with just 10%–20% of power dissipated as heat, and can be quite small. The main downside is that these amplifiers use high frequency switching to generate the necessary power, producing a lot of above audible band noise that can (but in practice doesn’t) interfere with the signal. Certain low cost Class D amplifiers show load dependency, which is a variation in performance depending on the electrical impedance of loudspeakers they're supposed to power.

A perfect amplifier would generate no heat, add none of its own noise or distortion to the signal and never run out of power. But in audio you get a little of bit of everything you don’t want with everything you do. What’s clear is that the letters are important only to explain heat and weight, and that the numbers (the measurements) explain the rest.

White boxes? Black boxes?
These are terms from information theory. A white box is a system that behaves in a known way. A black box behaves in an unknown way.

They are useful for characterizing the behavior of certain audio products, with the operative distinction being exactness. Audio products with a comprehensive set of good measurements can be called white boxes. But the majority are black boxes, where it really is unclear what they do beyond certain basic functions (that it’s a preamp, that it has balanced connections, but not what levels it can accept and output, or how if its volume control maintains consistent balance as you attenuate or boost the input signal).

Preference?
The ability to form a preference should not be assumed outright. Some circumstances really don't permit it. One part of being a good listener is learning to make distinctions and identifying differences. The other is learning when you can't do so and operate with a lighter relationship to what you hear.

If you perform a listening test casually, without attempting to define controls and procedures, then you should treat the results casually as well.