The Birth of DSD – Part I

There are only two real components in digital audio. There is the word length and the sample rate. Typical word lengths are 8, 16, and 24-bits. And then there is 1-bit PCM…also known as DSD or Direct Stream Digital. Each additional bit lowers the noise floor of the system by about 6 dB and dithering helps in making any noise present in the system more uniform across the entire frequency spectrum. But you need at least 3-bits in order to use dither. It’s not possible to randomize a single bit.

When consumer digital audio was introduced in October of 1982 the filters used in the first Sony machines were analog and they were steep. These filters were needed to remove any partials that were produced as the 44.1 kHz digital samples were reconstituted back to analog form. According to Nyquist-Shannon, a perfect reconstruction of the source analog waveform could be done with ideal filters at half the sample rate. But Philips and other makers of CD players cranked the sampling rate higher (2x and more) so that they could use very steep digital filters, which resulted in better fidelity than what Sony had accomplished with the analog filters.

Philips could also use fewer bits. In the case of their first consumer CD player, they opted for 14-bits (which was the original word length proposed by Sony and Philips, two of the major patent holders of CD technology). The use of fewer bits means the potential noise floor is 12 dB higher. To bring the noise back in line, they figured out a way to “shift” the in band noise into the ultrasonic frequency range…and thus out of the way of human hearing. Thus was born the idea for DSD…fewer bits and shifted noise.

As converters and digital technology in general advanced over the next 10 years or so, analog to digital and digital to analog converters improved. The digital chipmakers designed and implemented “delta sigma” methods and settled on 2.8224 MHz 1-bit as the preferred specification (64 times the rate of CDs). Virtually all electrical designers employ these chips and use a process called decimation to peel out the desired word length at the desired sample rate…in the case of compact discs, this means 44.1 kHz and 16-bits.

The reverse of decimation is modulation, wherein the 44.1 kHz/16-bit stream is converted to the 2.8224 MHz 1-bit signal on its way back to analog. Seems simple and it is…but it’s not an analog process as many in the DSD community would have you believe. There are still filters and conversions happening.

Concurrent with these new chip designs, the move to high-resolution audio was initiated. In 1995, the DVD Forum began to look past the venerable CD and consider the next physical format for audio delivery. The benefits of higher sample rates and longer word lengths were being discussed. Meanwhile the eventual DVD format was being developed by teams from Sony/Philips and Toshiba/Time Warner. In 1995, both teams had already developed competing and incompatible formats for consumer video delivery. Sony/Philips had MMCD (multimedia CD) and Toshiba/TW had SD (Super Disc). Both formats used PCM audio with higher sample rates and longer words. DSD was not in play as a consumer format at this time.

The DVD format was designed to replace VHS tapes and audio took a secondary role. We did get high-resolution audio in 1-channel stereo and 5.1 surround in lossy audio (Dolby Digital) but DVD was all about spinning discs and delivering better video quality. DVD-Video was a resounding success and is still with us 18 years later.

Figure 1 – The ideal digital recording signal path as described by DSD. [Click to enlarge]

Also in 1995 Sony had come to the conclusion that they could avoid the whole question of sample rates and word lengths and their impact on fidelity by simply using the 2.8224 MHz 1-bit stream directly. They were looking for a technology to archive their vast catalog of 2-channel analog master tapes and they locked onto delta-sigma as the generic method to record the converted tapes. Crystal Semiconductor had developed an ADC chip that allowed engineers to capture and store the 2.8 MHz 1-bit data stream directly. Thus was born the DSD format.

Tomorrow, I’ll tell you how this archive format became a consumer format in the form of SACD. Stay tuned…