The digital myths of the audio world
When it comes to myths, digital audio gets a great share. If you want to piss off an analog audio purist, you no longer have to carefully ease into lengthy arguments; just show them this video and the negative/dismissive responses will be guaranteed. Oh well, the lengthy arguments might follow anyway, but you can just smile and walk away smugly.
D/A and A/D | Digital Show and Tell (Monty Montgomery @ xiph.org)
Christopher “Monty” Montgomery, founder of The Xiph.Org Foundation, creator of the Vorbis audio codec or the Ogg software container format (among other things) explains these things well and clearly. His fully analog test equipment consists of a 1978 Hewlett Packard HP 3325A Synthesizer/Function Generator as the main signal generator, a mid-90s Tektronix 2246 100MHz oscilloscope, and a Hewlett Packard 3585A 20 Hz – 40 MHz Spectrum Analyzer. The digital signal comes from a Lenovo ThinkPad x61 laptop, and the digital-to-analog converter is an old, consumer grade Emagic EMI 2|6 USB Audio Interface.
Those pesky legends
With the help of the above equipment, Monty shows us the following things:
– there are no so called “stairsteps” present in a digital sine wave (so using the “lollipop” graph is actually more appropriate)
– digital bit-depth doesn’t effect the smoothness of an audio signal; it determines the amount of noise present
– the noise present in a digital signal (dithered quantization noise) sounds and acts like tape hiss (gaussian dither IS the equivalent of tape hiss), so we can express analog tape hiss with the equivalent gaussian-dithered bit-depth. For example compact cassettes are typically 5-6 bits (9 bits in perfect conditions), vinyl records in perfect condition are 12 bits, and the very best open reel studio tapes are about 13 bits, barely. With noise reduction.
– we get to know what dither is: a specially added noise with known attributes, so we can make sure that the quantization error is independent from the input signal. Without it, we get harmonic distortion. Qantizing to zero = 100% distortion. Dither enables us to hear the audio signal even at levels that are only fractions of bits.
– noise shaping allows us to choose a kind of noise for dithering that’s the least obtrusive for our human hearing.
– 16 bit digital audio already has a very low noise floor, even without dithering.
– we can see what bandlimiting does to a square wave (Gibbs effect), and that running the already rippled signal through the filter the 2nd time will not take away anymore frequencies
– parts of the digital signal that fall between samples are still represented perfectly