Chapter 3: Audio Headroom Design
Understanding Reference Level and Performance Standards
It is a calibration standard that establishes peak limits at 105 dB for main channels and 115 dB for low frequency channels, preserving dialog clarity while allowing low frequency effects to retain their impact. This is audio headroom design.
Somewhere along the way, a calibration target became a virtue. Reference Level stopped being a tool and started becoming a sales sacrament. People weren’t asking whether it served their needs anymore. They were becoming seduced by the promise of better by virtue of the Director’s Intent.
In concept and original intent, I think it offers something potentially valuable, if all parameters required to actually execute it are met, from content production and delivery implementation, as well as system capabilities and setup, and a user wants to defer to the director (and the targeted understanding of the average audience needs) for the ‘correct’ playback level.
‘Reference Level’ playback capabilities were part of that, and assumed if you bought ‘certified’ equipment and did the very basic calibration of setting channel trim levels, and ran these systems in ‘normal’ home listening environments. After all, it was ‘Certified’, which kind of implies a promise.
A Promise of what?
For those that don’t already know, ‘Reference Level’ is nothing more than a system gain and output level standardization method both in content production, and system setup calibration - Audio Headroom Design.
At home, as with the original commercial THX program, ‘Reference Level’ potentially allows content producers, when implemented from beginning to end, to provide a means for end users to experience ‘ideal’ suggested playback levels for that content, where listeners can easily understand dialog and environmental/action sounds at the ‘average’ playback level, consistently, comfortably. It also allows enough ‘headroom’ above long-term steady playback content like voices or ambient sound for periodic transient louder content to safely and briefly exercise the top end of a pretty capable system’s dynamic range. Explosions, gun shots, falling objects, Godzilla coughing, etc.
These peaks operate within levels that, for a short duration, are still safe for listeners and most equipment. Even if that equipment fails to meet those output levels uncompromised, at least the equipment probably won’t break. Those peak levels that are safe as short-term occasional transients add some extra excitement and realism versus lower dynamic ceilings that would require either lower overall playback levels (with quieter dialog) or more electronic dynamic compression applied to the audio tracks, which removes the ‘live’ character of those events.
‘Reference Level’ establishes a ceiling to provide ‘headroom’ for the playback of content channels, such that 105 dB is the highest level available for a full scale maximum output sine wave for each of the ‘main’ channels, and the Low Frequency Effects channel, offset by 10 dB in the encoder and decoder, may be as high as 115 dB in peak delivered output, to supplement the ‘Boom Boom’ capabilities of playback without compromising the noise floor of the main channels at the time. Barry Ober called it the ‘Dinosaur Stomp’ channel. Barry Ober also had some interesting commentary regarding the origin of the LFE channel for commercial formats, and why using it at all in a home release when you had 24 bit formats and bass management didn’t actually make any sense, but that’s a topic for another day. I miss that guy.
Combined with dialog normalization, metadata that ‘should’ be included with the audio data, the processor will know how to ‘slide’ the output level of the content under the hood (effectively adjusting the master volume without changing the master volume display) so that regardless of where the dialog level sits relative to the highest peak value (which in digital formats is typically normalized to peak track output levels to maximize signal to noise ratio), dialog can be ‘anchored’ at comfortable and intelligible speaking levels, akin to if you were witnessing a conversation, somewhere in the range of 75-85 dB, and then the dynamic peaks land where they do relative to the target dialog levels.
Brian Florian, my dear Canadian Friend, did a nice article a long, long time ago explaining how dialog normalization works, and why.
Reference Link: Home Theater HiFi “Dialogue Normalization: Friend or Foe”
So, what this means is, if a system is properly calibrated so that ‘Reference Level’ actually IS at reference, and you’re listening to content that properly implements dialog normalization so that ‘Reference Level’ actually has any meaning, when you set the volume control to ‘Reference Level’, or ‘0 dB’ on the reference relative scale, IF the people producing the content adhered to ‘Reference Standards’, you may have playback peaks up to the peak voltage on an amplifier output that would be required to elicit 105 dB with a single sine wave, plus or minus variations due to loudspeakers and room response.
In the original reference thread, Tom Vodhanel brought up a valid piece of minutia, in that 105 dB ceiling relates to a sine wave signal, and that content isn’t that. It is in fact minutia, but I think it deserves a side quest.
The peak of the delivery format is fixed. There is no higher, and peaks are what get clipped when the power amplifier ‘clips’ as the output ‘tries’ to exceed the available rail voltage. But a sine wave RMS value is 0.707… of the peak value, and it is theoretically possible to sneak a bit more RMS content with mutli-frequency signals.
So, this discrepancy can be relevant if we wanted to just get extremely argumentative. We could point out that with a ‘perfect’ square wave (as perfect as the format sampling bandwidth limits will allow) at max signal voltage defined by 0 dBFSD (dB Full Scale Digital) digital format limits, the theoretical maximum SPL ceiling output would actually be 108 dB. A square wave packs odd order harmonics until theoretical Infinite Hz onto that sine wave in a way that keeps the peak the same as that sine wave alone, but the wave is as constant in peak output as the bandwidth will allow, so that where peak waveform amplitude and RMS amplitude are the same thing. As such the point is that it is theoretically possible to pack a content signal with more RMS value, and as such slightly exceed 105 dB, without increasing the peak voltage.
Having interrogated that possibility, practically speaking, I would release it from serious consideration. One could make that argument, sure but that square wave example would only happen in listening content through malicious malfeasance or extreme incompetence, and that would imply the WORST technically possible case scenario. 3 dB of playback headroom would guarantee plenty of margin, and margin nice, but exponentially expensive. 1 dB is probably more than enough, ever, with content. Find me an exception, and I’ll buy you a sandwich for your contribution. To be clear, I don’t think Tom was making the argument that you needed to add headroom because of that, I think he just mentioned it as a point of interest, and I thought it interesting enough to include.
But what you CAN count on is that the peak voltage levels of the electronic waveform will not exceed that required to reproduce 105 dB with a sine wave at the ‘nominal’ output level. By ‘nominal’ I mean that the loudspeaker output itself, even with constant input levels, is all over the place as you change frequency and direction, but somewhere you pick the ‘average’ as it is relevant to your use, and you ‘name’ that value. Within the parameters of what qualification, what that means is that for each main channel, if we are implementing ‘Reference Level’, then 105 dB, that’s the nominal ceiling, period. Whether you want to bump up your requirements in this hypothetical another 1-3 dB is entirely up to you.
But wait, there’s more, because we have complications. RP22 is recommending MOAR headroom, possibly with good reason, if you understand what those recommendations actually mean. Some of the details are unclear, or even undecided.
I say this because with the recent discussion of RP22 in that thread, Schurter was good enough to summarize a lot of talk about the value of ‘headroom’ beyond the actual listening requirements at ‘reference’ level.
After mentioning this reference, I bothered to read through RP22, and I feel compelled to say that it has a lot of good, generalized recommendations for thoughtful consideration, covering a broad swathe of relevant topics in audio reproduction. There is a whole lot to applaud, and I am figuratively applauding into my keyboard literally standing up, at my ‘standing desk’ that is actually a commandeered tall café table.
I also think that the way it is written, as is typical of publications written for audiences with less technically nuanced understandings of topics, is largely a very broad summary of, and an incomplete work at that, it leaves lots of room for misinterpretation. It is also not immune to industry influence that has “how do we capitalize on this?” built into cooperation requirements.
Originally published by Totally Hypeless
Comments