0dBFS+ Level In Audio Production

Thomas Lund of T.C. Electronic presents the results of T.C. Electronic's investigation of higher than 0dBFS signals on Red Book CDs. Digital PCM, such as the 44.1 kHz, 16-bit system used for CDs, can represent signals higher than 0dBFS, or full-scale digital. T.C. Electronic calls such signals "0dBFS+". A simple way to think about this is to imagine sampling a sine wave not on its peak, but around its peak. If these samples are increased to full-scale digital (for example, through a digital audio editing tool), then the peak, which occurs between the samples, must be reconstructed as higher than full-scale digital. It's a simple task to be +3dBFS with sinusoidal signals, and slightly more exotic signals, such as maximum-length sequence (MLS) signals, which are often used to measure impulse responses, can be as much as +6dBFS. +6dBFS is twice the amplitude of the highest digital sample.

Many CD players and DACs, and a significant number of audio editing tools, will distort significantly when presented with 0dBFS+ signals much lower than +6dBFS. The output analog stages of DACs will often have their gain calibrated for 0dBFS, and will be overloaded and clipped when the reconstruction filter presents it with a 0dBFS+ signal. Digital algorithms can also behave badly, including latching up, when presented with 0dBFS+ signals to process. Audibility of the distortion can be subtle, such as small amounts of distortion causing more fatigue over long listening sessions, to the dramatic, such as clicks and pops. As an example, T.C. Electronic found that the NAD 520 CD player, with a THD+N of 0.04% with normal signals, will have about 3% THD+N distortion when presented with a signal only +0.69dBFS (8% higher than 0dBFS), and will degrade to about 11% THD+N when presented with a +3dBFS (41% higher than 0dBFS).

What can we do about this? Recording and mastering engineers should view their data with interpolating peak meters, rather than meters that just look at sample values. An interpolating meter would reconstruct the waveform from the samples, and pick out the peak level from the samples and the reconstruction, rather than just the samples. Thomas was emphatic that mastering engineers no longer count samples to determine clipping: often, mastering engineers have an arbitrary number of consecutive samples at 0dBFS they'll consider to be clipping. For example, more than 5 samples in a row at 0dBFS could be considered a clip. However, this rule ignores the waveform amplitude between samples, and those 5 consecutive 0dBFS samples, may not represent clipping at all. Samples near or at 0dBFS are an important issue today because of the loudness wars happening in many pop music releases: the music is heavily compressed so they sound as loud as possible, and so that they'll stand out in radio play. Besides destroying any form of subtlety and dynamics in music, such compression also introduces many 0dBFS+ peaks because the samples are now so close to 0dBFS. To reduce the likelihood of generating 0dBFS+ signals, many mastering engineers now master with a maximum sample value of -3dBFS.

Consumers and listeners should demand more awareness of this issue from the pro community. They should also ensure that playback devices such as DACs should have at least 3dB of headroom in their analog stages. However, such a design will degrade SNR measurements by 3dB, and may not be favored by marketing departments, even though such a degradation would be meaningless, and in fact, probably lead to a better sounding DAC. Processing algorithms should not make assumptions about 0dBFS samples, and should behave graciously when presented with 0dBFS+ signals.

Thomas also suggested the use of an oversampling level limiter to control 0 dBFS+ peaks. Such a limiter needs to oversample in order to see the intersample peaks it seeks to limit. []