Sample Rate Conversion: A Primer Part I
Digital systems are not continuously variable. They are comprised of a multitude of individual slices or data points that when systematically presented to a mechanism to reconstruct them, we get the original continuously variable signal back. The rate at which the individual slices are produced is called the sample rate. PCM and DSD are both digital music formats and despite the claims of the 1-bit format advocates that DSD is somehow more “analog-like” (which is marketing spin and decidedly untrue), both formats require an accurate clocking system during the capture and the reproduction phases.
I’m going to stick with PCM in my discussion of the conversion between sample rates. I honestly can’t say that I’ve done any real work with DSD because of the lack of professional tools…or in fact, any tools to modify native DSD data. If processing is required when working on DSD recordings, the files are converted to PCM or the work is done using analog means prior to conversion to DSD.
When an engineer is contemplating what technologies and what specific equipment to use during a new production, there are lots of considerations to keep in mind. Some of them are more important then others. Is the project going to be recorded using analog or digital recording equipment? Does the plan include multitracking many layers of sound or perhaps the project is a live recording. However, one of the most important technical decisions that an engineer must make when working in the digital domain is which sample rate to choose.
It matters because despite claims to the contrary, the fidelity of a recording is “locked in” at the time of the original recording and cannot be fundamentally enhanced during any subsequent post production or re-mastering. In the world of PCM, the Nyquist Theorem puts a cap on the highest frequency that can be captured and interpolating new samples through upconverting or adding new samples doesn’t change that fact. Upsampling does have other benefits when working with digital samples but getting extended high frequencies isn’t one of them.
I would be remiss if I didn’t say that post production digital processing of older audio recordings isn’t able to elevate or enhance audio fidelity. While I would insist that the fundamental sonic parameters of the original recording would remain in tact, there are some miraculous tools that engineers have at their disposal that can accomplish fantastic improvements.
During the early part of my career, I owned a comprehensive DSP suite of tools designed and implemented by Andy Moorer of Sonic Solutions. Andy is one of the top DSP experts in the world when it comes to audio processing. The NoNoise, DeKrackle and DeClick add-ons that were available from Sonic Solutions cost over $60K but they performed audio magic on troublesome older analog recordings.
As their names imply, these tools were able to detect noise, clicks and other imperfections in an audio recordings AND remove or repair them. In fact, the company was originally formed as a service bureau that offered audio restoration to clients at many hundreds of dollars per processed minute. They were able to do things that no one else could because of the brilliance of Andy’s algorithms and the robustness and power of the new nuBus (the original Mac expansion bus) DSP cards that came with the Sonic Solution CD PreMastering system (each card had 4 Motorola 56000 DSP chips and 4 Megs of memory…that was a rocket ship in those days).
By isolating a sample of the noise signature, NoNoise could then chew through the entire file attenuating the noise by 2-3 dB. The DeClick process would synthesize a small number of audio samples by looking at the surrounding milliseconds and allow operators to remove a click, pop or other brief transient error. This was pure digital magic and clients were usually willing to shell out big bugs to scrub their older analog tracks clean.
Tomorrow, I’ll continue with a discussion of upconverting and downconverting sample rates between standard rates.
