What does dither do? Why do digital systems need it?

In the analog world, as a signal dies away, it does so smoothly (if nothing is wrong with the system). As the level drops, the signal gets progressively quieter. At some point it reaches the same level as the noise. But importantly, if the signal level continues to drop, you can still hear it, although it lies below the “noise floor”. This constitutes an important aspect of the way that analog signals behave – you can hear coherent audio information even when it has a significantly lower level than random noise.

In the raw digital environment, everything seems different. As the level of a signal drops, fewer and fewer binary digits represent the audio information. Ultimately, you simply run out of bits, and when this happens, the signal just stops, and in a 16-bit system, this happens at an audible level. This behavior forms one of the sever- al factors that gave early digital recordings a bad name, and led some pundits to claim that digital audio sounded fundamentally inferior to analog.

Adding noise to the signal provided one solution to the problem. At low levels, this effectively turns the last few bits on and off at random, smoothing out the sound and ensuring that everything does not simply disappear as the level falls. We call this noise “dither noise” or simply “dither”. The word literally means to tremble or quiver – a reference presumably to the least significant bits turning on and off at random.

The disadvantage of this process: you introduce noise into the system and therefore effectively degrade its performance. More than that: the noise actually sounds quite objectionable. Truly random (white) noise contains all frequencies and sounds particularly obnoxious. As a result, several manufacturers and researchers have attempted to improve the situation by developing methods of hiding or “shaping” the noise created by the dithering process.

This may appear even more important when you record a signal, say, in 24-bit form and want to reduce it to 16-bit for Compact Disc. If we could preserve the detail of a 24-bit recording by making the noise floor more transparent – more like analog – then we would achieve an audible improvement in the quality of the final CD, and we would enjoy audible benefits by recording beyond the 16-bit level, even for a conventional 16-bit Compact Disc.

Most of these “noise shaping” techniques rely on the fact that the ear seems more sensitive to midrange frequencies (around 4 kHz) than to either low or high frequencies. In transferring a 20-bit recording to the 16- bit world of Compact Disc, for example, we remove the last four bits of the 20-bit signal and feed them back into the input signal through a filter that both adds dither and changes the spectral shape. Originally, the filter shape proposed by researchers (primarily at the Audio Research Group at Waterloo University, Ontario), and based on psychoacoustic principles, added more noise in the upper frequencies while lowering the noise floor at around 4 kHz – the frequency at which measurements indicate the ear’s maximum sensitivity.

In fact, they could have achieved even better results by adding the noise back in at low frequencies, where the ear measures as even less sensitive – remember “loudness” controls? – but this requires significantly more processing.

More recently, a number of manufacturers have claimed that their own proprietary filter shapes sound audibly superior to the theoretical designs. Unfortunately, these noise shaping techniques can cause problems. First, although they lower the noise floor at the most audible frequency, they unavoidably increase the overall noise. In addition, according to independent measurements, they add audible artifacts to the sound.

The fact remains that “dither” is an important tool for digital audio. But how can we do it so that the results sound good? Apogee has the answer. Apogee takes a completely different approach with the UV22® process.

UV-22 does not constitute a “new flavor” of dither noise. Instead, UV22 essentially modulates the data from the least significant bits of a signal on to the 16-bit signal according to a special algorithm, which adds an inaudible high-frequency “bias” to the digital bit stream, placing a “clump” of energy at around 22 kHz. This results in an essentially flat noise floor, at the theoretical 16-bit level – 4 to 5 dB below that of conventional “flat dither. In addition, the noise floor does not have the distinctive and annoying “hissiness” of conventional dither. Thus the UV22 noise floor sounds audibly quieter and less objectionable than other techniques. In addition, you cannot hear any audible artifacts. Yet, as with analog, you can hear coherent audio signals several dB below the noise “floor” – thus retaining much of the detail and audio quality inherent in the original signal.