Yesterday, I explained that analog to digital conversion using PCM requires the use of a Low Pass Filter to remove any frequency higher than the Nyquist frequency, which is exactly half of the sampling rate. Bad things happen in the audible band if frequencies greater than SR/2 are allowed to get into the system. But you might be surprised that this isn’t always a bad thing.
I studied electronic and computer music during my years of music school at CSUN, Cal Arts and UCLA. In fact, my doctoral dissertation was the first electronic music composition ever granted by UCLA. During the 80’s digital synthesis and sampling were becoming mainstream. There was the introduction of the Yamaha DX-7, which used FM as the basis for its amazing repertoire of sounds. Dr. John Chowning, who has become a close friend over the past few years, was the head of Stanford’s computer music research facility and developed the instrument. To this day, the DX-7 and FM synthesis have generated more license revenue for Stanford university than any other commercialized IP the university has ever invented.
Digital sampling was also a new technology of the time. The idea behind digital samplers is to capture all of the notes (or a representative group) of an instrument and assign them to the equivalent notes on a keyboard. The sampler would reproduce that actual sound of the sampled instrument at the corresponding pitch allowing composers to have an “orchestra” in their own private studios. Today, sampling is one of the most widely used sound producing techniques available.
Roger Linn used this technology to create the Linn Drum, which applied the sampling techniques describe above to a drum kit and percussion instruments. This revolutionized music production. Instead of having to hire a real drummer and mic up an entire drum set, the Linn Drum made it possible to include great drum sounds and rhythmic “loops” that humans couldn’t possible play. For better or worse drum machines and samplers dominate commercial music production.
So how does foldover or aliasing play a role in digital sampling? How could the unwanted frequencies associated with this PCM problem possibly benefit sampling? Imagine the sound of cymbals. They are very rich sources of high frequencies. Musicians refer to these densely packed high tones as “inharmonic” partials or overtones. The term “inharmonic” means that these tones don’t follow the natural overtone series, which are the basic acoustic building blocks of musical tones. Cymbals and other percussion instruments generate sound that is full of noise…random collections of frequencies.
Sounds like a real opportunity to use the “aliases” as a source of random high frequency sounds when trying to enrich the sound of a cymbal sample. Imagine flooding the AD converter while recording a sample with frequencies that are greater then SR/2, the Nyquist frequency. This means leaving off the LPF that we know is typically required.
All of these high frequency tones “foldover” the Nyquist and appear as a dense cloud of high frequency partials in the audio band. Presto, one of the shortcomings of PCM digital audio can be used to produce better samples without additional hardware of memory capacity.
However, it can also be a difficult problem when we want to record accurate highs. If a signal contains lots of ultrasonic noise and that noise is sent to other components that can’t handle that noise (amplifiers and speakers for example) or lacks the ability to roll it off, your ears will hear it as spurious noise. This actually happens. I heard about this very circumstance in the case of a new cell phone and a problem in the Qualcomm chip set that failed to address the ultrasonic noise resulting is terrible audio. Luckily, an audiophile at the company listened to the demo tracks that were derived from source DSD master (converted to PCM) and alerted the QC department before the new device made it to the press and public. It was a close call.
Another vote to avoid DSD, in my opinion. It all comes down to the quality of the content.