(Page 1 of 2 pages for this article  1 2 >)

Filed under: 3D •Audio •Cameras •compression •Hardware •Post Production •

HPA Tech Retreat 2010 - Day 4

The fourth and final day of the 2010 HPA Tech Retreat started off with the vexing problem of TV loudness, and ended with an inspirational talk by Belden’s irrepressible Steve Lampen.

[Previous coverage: Day 1, Day 2, Day 3.]

(As usual, these are barely-edited notes, scribble-typed as people spoke. Screenshots are copyrighted by their creators; poor reproductions thereof are my own fault.)

Audio Loudness Panel, Moderated by Patrick Waddell, Harmonic

Jim DeFilippis, Fox - The New ATSC RP on Loudness and Its Impact on Production and Post

David Wood, EBU - P/LOUD and ITU-R Loudness

Steve Silva, Fox - Hands-On Experience with BS.1770 Loudness Measurements

Ken Hunold, Dolby - The Boundary Issue

David Wood, Ensemble Designs - AGC that Preserves Dynamics

Why do we now worry about loudness? Digital made it happen. In analog, on a good day, had maybe 40 dB of working range. Digital has >100 dB, and much less noise. ATSC started work on A/85 RP, the loudness standards, in 2007; published in 2009.

The CALM Act requires that commercials are no louder than the program around them; ITU-R BS.1770 (loudness measurement standard).

DeFilippis, Fox: worked on the A/85 committee. We also went from stereo, sort of, to 5.1 audio. Loudness is a complex sensation, e.g. fast talking “sounds louder” than slow speech at he same level. “Loudness is the quality that represents the overall perception of level”; it’s subjective and difficult to quantify. Dependent on environment, too.

Objective measurements: CBS loudness meter. Leq(A), A-weighted loudness equivalent. LKFS (BS.1770), K-weighted loudness related to full scale, shown by subjective testing to come closest to perception. Frequency weighted average over the whole program, power sum across all channels in the program.

A/85 published on November 2009, referenced in CALM Act. Established common method for measurement, guidelines to mixers on how to meet “target” loudness (e.g., -24dBfs for Fox; in the absence of a network spec, -24dB is a good aiming point). Recommends that human monitoring in a proper environment is critical, five different environments listed, from headphones at 74dB to a 20,000 cubic foot room at 82dB. Important: verify mix on small speakers; in stereo; in mono. Strong 5.1 surround channels can overwhelm the center dialog track when downmixed to mono or stereo.

How to do it: long form: measure your dialog level and record that as your Dialog Level. Short form (e.g., commercials): measure the loudness of all content on all channels. Use qualified BS.1770 meter. Live mix: level the dialog at the target level, mix other elements to a pleasing balance.

Loudness is measured on the anchor element (e.g, the dialog channel), but the anchor element may drop below average level; this is OK if it’s appropriate (quiet passage w/o dialog). But don’t let the short-term levels at program boundaries get too far away from the average, because the cut to commercial will be jarring.

Golden Rule: make sure your BS.1770 measured loudness levels match desired dialnorm value.

Wood, EBU: The past was a QPPM (quasi peak program meter), 10mS reaction time, didn’t match human perception well. In analog, you had space above headroom, so everyone pushed levels to the top, with compression.

But now the EBU (European Broadcasting Union) has tackled the problem:

Studies done in Canada (“listening tests not done by moose”) led to ITU.R BS.1770.

1 LU (loudness unit) = 1 dB. “Gating”, the concept of foreground loudness and the need to cope with it: programs with wider dynamic ranges, with the same average loudness as a flatter program, will have more apparent “foreground loudness” at the peaks. Gating out periods of quiet passages to avoid this problem took a lot of testing. Four levels tested: -8 LU below momentary (400 mSec) measurement window was chosen for now. Loudness meters should have an “EBU mode” where they all do the same thing. Suggest terminology change from LKFS to LUFS, “but Jim DeFilippis might have a heart attack!” Windows: momentary 400 mSec, sliding 3 sec, integrated is whole program.

LRA (loudness range), excludes lowest 10% and highest 5%, measures statistical distribution of loudness levels.

Distribution guidelines need to account for differing STBs and output modes. Content supply guidelines must address migration scenarios if legislation sets a new number (changing what’s acceptable overnight), how to monitor, what needs to be changed. Main recommendation R128 with 3 parameters: loudness (-23 LU gated, +/-1 LU for live programs), maximum true peak (-1 to -2 dBTP), loudness range descriptor (metadata if possible).

Silva, Fox: Using the Dolby LM100 (BS.1770) loudness meter. Lots of complaints about loudness both between and within programs. Need to measure, both subjectively (ears) and objectively (LM100). LM100 started out with Leq(A), firmware switched to LKFS in 2008. Measured loudness internally at Fox, at satellite downlink and at ingest stations. What modes to use: infinite or short term? Chose 5-second windows. Found there was no simple, single measurement type that works every time, so used two LM100s set up differently. Found Leq(A) matched LKFS within a dB or two. Measuring center-channel dialog as the sole anchor element didn’t work on more sonically-rich material. When to choose center vs all? Center doesn’t work for wide DR (dynamic range) shows, but all-channels numbers don’t help anyone without an LM100. But all channels measured in infinite mode gave overall loudness; easiest approach. Transition between programs / spots still the hardest thing to fix.

Hunold, Dolby: Measured 70+ programs from four networks (ABC, CBS, NBC, Fox) Sep-Oct 2009, Feb 2010. Used Dolby DP600 program optimizer, media meter, LM100; BS.1770 algorithm. Measured loudness, transmitted dialnorm (dialog normalization metedata; essentially indicates the level the program has been mixed to), peaks, etc.; plotted overall network loudness, commercial loudness, program/commercial difference, etc. Program loudness varies from -16 to -32 LKFS, mostly clustered around -25 LKFS, slightly better then in 2008 but not hugely. Commercials centered around -24 but weighted louder.

• Network W: programs typically 0 to 5 LKFS hotter then dialnorm. Commercials -1 to +12 louder.
• Network X: programs -6 to +5, commercials -6 to +7, avg +2.
• Network Y: programs -4 to +2, commercials -6 to +4, clustered at 0.
• Network Z: programs -3 to +2, commercials -7 to +9, very wide range.

Clearly, more work is needed.

Loudness processing: one station used loudness processing; made progam/commercial cuts more consistent, but compromised program DR.

Strategies for stations: Measure and tag: measure loudness, set dialnorm metadata; measure and scale: change gain to match metadata / dialnorm; target and evaluate: content that doesn’t meet spec is simply rejected.

Wood, Ensemble Designs: Audio AGC in the “real” world, or, How do you get the numbers right when all you have is a rubber ruler?

Avenue LevelTrack automated AGC: adjust the gain but don’t compromise the dynamics.

Tools exist, but not everyone uses the right tools at the right places; “The future is here. It’s just not evenly distributed.” - William Gibson

Many user controls for max/min levels, integration time, etc.:

System includes real-time graphing of instantaneous and average levels, and the resulting gain setting:

“People find that this system works; the controls let them finesse the AGC to match the kinds of content they’re broadcasting.”

Discussion:

• Are we heading to a North American practice and an EBU practice? No easy answer; there may some alignment between different countries/organizations. But there isn’t a lot of difference between our approaches.

• CALM has been modified to address commercial loudness AND program/commercial loudness consistency.

• What happens when you have a soft element and a loud element (“Star Wars” vs. “Love Story”)? You still wind up measuring levels and normalizing.

• What about local vs. network commercials? Not measured, but suspicion is that the local spots may not have been normalized as much.

• “We’re not just talking a few dB, we’re seeing a 14 dB difference: egregious!”

• Having a common set of guidelines is a critical point; advertisers don’t need to be the loudest, they just don’t want to be the softest!

• It seems that (from a guidelines standpoint) EBU’s loudness range is similar to ATSC’s dynamic range.

• “My commercials at home come from three sources: network, local insertion, cable headend. Levels are all over the place!” NCTA has been following the discussion, but there are thousands of MSOs with engineers who need to be brought on-board.

• What about inter-channel loudness? What seems to happen is that station engineers in a market notice a mismatch and work together to even things out.

Michael Bergeron, Panasonic - Reducing Operational Complexity of Stereoscopic-Production Camera Systems.

We’ve been looking at what people have been doing with 3D cameras for ages—maybe a year and a half! Right now, 3D equals stereo glasses for the most part. Passive glasses and micropolarized screens (e.g., line-alternate polarization screen overlay) gives half-res; active-glasses alternate-image systems give full res.

3D cam system must (1) link and sync two cameras; (2) manage parallax or two images; (3) correct optical and parallax errors for good 3D (e.g., fix all problems in the first two things). Getting that last part done is one of the big roadblocks. For (1), lenses must be linked and matched very carefully. (2) Parallax (convergence) management: should convergence track focus? Not necessarily; convergence must be consistent with perspective: close-ups should be closer than wide shots irrespective of focus:

What about IO (interocular, a.k.a. interaxial)? Too narrow, too little 3D effect; too wide, exaggerated and even painful 3D. IO and convergence angles are also constrained by / interrelated with object size, nearest & farthest objects, camera placement. Close shots need tighter IOs, farther shots allow / suggest wider IO. Fixed IOs just limit how close your closest object can be. Scientific research is good, but it’ll also help to just put cameras out there and see what happens.

Side-by-side rigs: limit minimum IO, tight CUs (close-ups) are a problem, minimum IO depends on imager/lens size, but simple and easy to make.

Beamsplitter rigs: no limit on minimum IO, good for CUs, but beamsplitter acts as a polarizer (and you lose a stop of light), and it’s mechanically more challenging to align and keep in alignment.

Optical/parallax errors: vertical misalignment; magnification differences; different L & R distortions (and different focus planes). Perfect images aren’t needed, just images that match each other. Ways to reduce these errors: optical/mechanical (pricey, not necessarily possible in all cases); postproduction (not useful for live TV; can be expensive); realtime electronic processing (typically expensive).

Current 3D rigs require a lot of engineering expertise, and skilled operators. Can we scale this down? With an integrated system with a fixed IO, this gets a lot easier; integrated lenses can be tracked at the factory; consumer cameras are already fixing optical problems in realtime. “Operation should move from ‘Guitar Hero Easy’ (focus, zoom, iris) to ‘Guitar Hero Medium’ (focus, zoom, iris, convergence).” It’s not really different from what’s been happening to get integrated camera/lens systems in 2D (in terms of managing complexity and solving optical problems electronically).

Small imager allows close IO (and deep focus). Lenses tracked during assembly keeps cost down. CA correction:

Why this (1/4”, prosumer-looking) camera first? (And by the way “it’s not a consumer camera!”) The biggest 3D bottleneck is a lack of skilled operators. Buy getting this camera out as cheaply as possible, we get more people trained up on 3D. Like Google does: just go out and collect user data. “A lot of people will shoot a lot of bad 3D (marketing told me not to say that!).” But we’ll all learn from what people do with this camera.

We don’t know everything about this camera yet: IO distance isn’t yet set, nor is sensitivity, max convergence, zoom range.

Next: Sony and Adobe; demo room wrap-up; final thoughts…

(Page 1 of 2 pages for this article  1 2 >)

Perhaps Sony should have shown what their flagship F35 camera looks like on a zone plate. It’s not 4:4:4 by any reasonable definition because:

1) it doesn’t measure full luma resolution, and is corrupted by aliases

2) the red, green and blue channels don’t ever line up properly. On high frequency detail you see chroma moire rainbows because of this.

3) the moire and resolution performance is noticeably worse than in comparison to the RED 4k M-X chip.

There is theory and there is practise - it’s important to shoot real world charts with real world camera systems to expose what is actually happening. Fact is that RGB Stripe is a fail.

Posted by Graeme Nattress  on  02/22  at  09:17 AM

---

Adam,

Awesome as always. There is SOOOOOO much meat in this article that I had to re-read sections to just start to grasp it. And you posted all of these articles (days) so close together when, in blog parlance, you could have separated each person or topic and made each its own post and had more and better content - for months - than the majority of blog articles out there.

Thank you for feeding the inquisitive minds not with answers but with more questions and ideas surrounding the issues many of us are just beginning to grasp at the production level. For it is with deeper questions that we arrive at the right answers.

And thanks for being far above the:
- here was something interesting - [link]
type of author.
You are worth your weight in gold.

Anthony Burokas, http://IEBA.com
Publisher: http://TechThoughts.org

Posted by IEBA  on  02/24  at  09:14 AM

---

Name:

Email:

Location:

URL:

Smileys

Remember my personal information

Notify me of follow-up comments?

Submit the word you see below: