Orban

A 2-Channel Digital Television Audio Processing Solution

By Robert Orban; thanks to Tim Carroll of Dolby for review, editing, and comments.

With the assistance of Dolby Laboratories, Orban has developed a unique and effective approach to audio processing in a DTV plant. This new system will ease the integration of local and network two channel programming, and fully exploit the extra features offered by the Dolby Digital (AC3-) system for 5.1 channel programming.

As you have previously learned, there are three important pieces of metadata in the AC3 bitstream.

The obvious question that arises is how these signals are to be generated in a real-world operational facility. And, indeed, in which situations they should be generated.

We should remember that the marketing landscape is littered with "features" that seemed to be a good idea at the time, but which proved to be of little or no interest to consumers. CDs have always offered the ability to deliver auxiliary data. According to the original CD hype, you would see the lyrics of the songs scroll by as you played them. And you would see still pictures of the band members by connecting your CD player to your television set. Where are these features now? The answer, of course, is that the public did not find them compelling enough to justify the additional production expense to add them to the CD data stream, or to justify the increase in manufacturing cost necessary to add the video outputs or the LCD screens to the CD players.

Another example is the SAP channel in BTSC stereo television. Very few viewers understand it, yet a number of them manage to turn it on by accident. They then can't understand why the sound becomes low-fidelity mono, and why everyone was suddenly speaking Spanish! Consequently, many consumer manufacturers buried the SAP control very deep in the menu structure of receivers or VCRs to prevent this confusion from occurring in the future.

With regard to the AC3 metadata, Dolby Laboratories wisely specified that dynamic range compression would be the receiver default, because they realized that most consumers would never want full dynamic range audio. Dolby did leave the option for this default to be modified by those who do desire full dynamic range.

Experience has shown that a vast majority of viewers want two things. First, dialog should be comfortably intelligible, and second, commercials should not be irritatingly loud by comparison to program material.

Home theater owners may want the opportunity to watch feature films while hearing a wide dynamic range signal. However, even these viewers usually consume television in a much more passive way when viewing garden-variety programs. If television is to be an acceptable part of the domestic environment, the sound cannot overwhelm household members not interested in viewing (not to mention neighbors, particularly in multi-family dwellings). For a variety of reasons, the dynamic range of sound essential to the intelligibility of the program should not exceed 15dB in a domestic listening environment. Underscoring and ambient sound effects will, of course, be lower than this.

The issue of loud commercials is particularly important-the FCC has been concerned with loud commercials ever since the mid 1960's, and has twice actively investigated the problem since then as a result of viewer complaints. It is against FCC rules to broadcast irritatingly loud commercials.

In current NTSC practice, all audio is applied to a transmission audio processor that automatically controls the average modulation and the peak-to-average ratio. This ensures that the audio will be comfortably listenable. The audio processor also has another crucial function-it smoothes out transitions between one piece of program material and the next. Several currently available transmission audio processors incorporate the CBS Loudness Controller algorithm. This uses a complex algorithm that estimates the amount of perceived loudness in a given piece of program material. If the loudness exceeds a preset threshold, the controller automatically reduces it to that threshold. The main purpose of this circuit is to control the loudness of commercials that have been processed to produce irritating loudness without such control.

Orban believes that the most realistic approach to handling Dolby Digital (AC-3) dialog normalization is a hybrid approach. Most program material, including commercials, live news, sports, most documentaries, game shows, talk shows, soap operas, and pop music videos and concerts, can be passed through an audio processor with a loudness controller very much like the ones currently used for NTSC. These processors control their maximum loudness level very well, so a single dialog normalization value will apply to all program material whenever the processor is online. The advantage of this strategy is that the processor will guarantee that all of this material is comfortably listenable, and that commercials are not excessively loud. With the possible exception of sports, this program material does not rely on extreme dynamic range to make its point, so I do not believe that compression damages the artistic integrity of this programming. No one needs more dynamic range on the Jerry Springer Show, or on the local news.

Prime time dramatic shows, newer feature films, and classical music concerts all use dynamic range for dramatic impact, and therefore are candidates for full-blown exploitation of the Dolby Digital (AC-3) metadata. Each show, film, and concert must have a dialog normalization value pre-assigned to it, ideally derived by referring to a calibrated loudness meter. The uncompressed audio is then applied to the AC3 encoder where Line-Mode and RF-Mode Dynamic Range Control signals are generated to ensure that the receiver can apply compression if the viewer prefers a narrower dynamic range.

Commercials should be processed through the station's audio processor in the usual way. If the fixed dialog normalization value is correctly chosen for all material passed through the audio processor, commercials will automatically be limited in loudness to the average loudness of the dialog and will therefore be unobtrusive regardless of whether the listener is hearing compressed or uncompressed audio.

It is probably impractical to pass through, without review, dialog normalization values created by program and commercial providers, because some commercial providers will inevitably try to game the system to make their commercials excessively loud. Instead, if dialog normalization is to be actively used in transmission, the broadcaster must strip its existing value from the program, and then must preview each piece of program material and replace the value with one that will ensure consistency from one piece of program material to the next.

If the networks have done their job well, they will chose dialog normalization values that ensure consistency from source to source, and also when the viewer changes channels. It is improbable that this can be done by automation. The best we will be able to do is to manually identify dialog or other baseline sounds, measure their loudness with a true loudness meter, and manually adjust the dialog normalization parameter so that these baseline sounds emerge with a standardized loudness. CBS's research into this area showed that no simple meter could do this accurately, including frequency-weighted meters with averaging characteristics. The errors in such measurements were so large that they were not useful in controlling the levels of commercials well enough to eliminate viewer complaints.

The CBS loudness meter divides the signal into seven octave bands and weights the gains of the bands according to the 70-phon equal-loudness curve of the ear. It then averages the output of each band with a 15-millisecond time constant. The averaged outputs of the bands are then added and the sum is applied to a 200-millisecond time constant. This is applied to the meter, which is assumed to have instantaneous response so that it clearly shows the effect of the two previous time constants.

In tests, this meter agreed with average listeners within 2dB. However, it's important to note that listeners disagreed amongst themselves by as much as 4dB when asked to assess the subjective loudness of a given piece of program material. So any loudness meter can only work for an average listener, and may show considerably greater errors when compared to any given listener.


Dynamic Range Control

Dynamic Range Control in the Dolby Digital (AC-3) system is performed over the entire audio bandwidth. The level detector determining the amount of DRC compression can be frequency-contoured to mimic the equal-loudness curves of the ear, and it does have the ability to "look ahead" at upcoming program level changes. This is sufficient for many applications, but may be improved with the addition of a multiband device to handle certain programming that may not get sufficient treatment from a single-band device like that in DRC. Experience has shown that dividing the processing into two bands above and below 200Hz, and then compressing each band independently, is sufficient to prevent audible spectral gain intermodulation when using the relatively mild compression typically applied to television audio.

Multiband compression is also useful in performing an "automatic equalization" function to change the frequency balance of the audio on a program-adaptive basis. In a multiband compressor, frequency bands containing excessive energy are automatically compressed more than other bands. This results in a re-equalization of the program material towards some target spectral balance. In a two-band compressor, it results in control of excessive bass, which can otherwise cause muddy balances. A five-band compressor, such as the one available in Orban's current digital processor for analog audio, can perform more detailed automatic re-equalization that can be very useful for program material such as live news. Approximately 60% of our digital Optimod-TV users are employing two-band compression, with the remaining 40% using five-band compression.

Availability of multiband compression is another argument for passing most program material through a conventional compressor with loudness control even in DTV service. Multiband compression smoothes out not only loudness variations but also variations in equalization, which can be particularly valuable with program material that has to air in a timely manner, where there is no time budgeted for careful audio post-production. Material that airs with full Dynamic Range Control implemented should be refined so that it sounds polished and consistent without further processing. A considerable amount of televised material does not meet this criterion.

Optimod-TV in the 5.1 Channel Environment

In the two-channel world, we stereo-couple the compressors by determining the gain reduction of both channels by the louder of the two channels. This works well for two-channel, but for 5.1 we will have to base the gain reduction on a power summation of the five channels, as weighted by the equal-loudness curves of the ear. In addition, we will have to have five loudness meters for each of the five full-range channel, and sum their outputs to get an estimate of the overall loudness of the sound field. We can use this information, by feedback, to control the maximum subjective loudness of the sound field to a user-specified loudness threshold. The combination of power-summed compression and loudness control will result in consistent perceived loudness regardless of the energy distribution in the 5.1 channels.

By setting the dialog normalization parameter to provide adequate headroom, we should never have to perform peak limiting on the compressed signal. However, if we wish to perform peak limiting for some reason, we should do it with a look-ahead peak limiter that provides low modulation distortion. Such limiting does not significantly smear the spectrum of the unprocessed signal, and therefore is less audible. [Last part of sentence not really relevant to AC-3 encoder limiting since it is calculated on the PCM signal and applied to the decoded signal if necessary. Or is this referring to peak limiting BEFORE AC-3? Since it is implied that dialnorm is taken into account, it would seem best to let the AC-3 encoder handle this.]

We are very fortunate that AC-3 does not use pre-emphasis, because the 75us pre-emphasis that is used in analog television audio has long caused considerable problems. It reduces high frequency headroom by up to 17dB, and therefore requires elaborate high-frequency limiting strategies to ensure consistent loudness. Because the processing for AC-3 will not require high frequency limiting or, in all probability, peak limiting of any sort, it has the potential to be cleaner and more artifact-free than current analog processing.


Conclusions


We have a great deal of experience with conventional transmission compression and limiting, and we know that the current practice satisfies most viewers. We know this is true because well-processed audio generates very few viewer complaints. Therefore, we have to carefully consider what program material will truly benefit from the ability to be heard with unprocessed dynamic range. Any program material that will not so benefit should be processed conventionally so that we can ensure that viewers can hear the audio comfortably without being blasted by loud effects or commercials, or being forced to strain to understand dialog.

Any program that is transmitted with full dynamic range should be pre-auditioned to determine if an appropriate setting for Dialog Normalization has been used. We can only hope that the production houses will mix prime-time dramatic programs appropriately. Concerning feature films, full dynamic range is probably inappropriate for home viewing in many cases, but Dolby Digital does leave the option available to each viewer.

The drawing below depicts the proposed interim solution. Network distributed audio is decoded by a DP-572 Dolby E decoder, then fed to a switch to allow local program material to be inserted. The Dolby E decoder provides up to eight channels of audio, the metadata that the DP-569 Dolby Digital (AC-3) encoder requires, as well as the GPI/O commands to control the processor in/out switch. When the network is sending two-channel program material, the Orban 6200 DTV audio processor controls the loudness to a fixed, preset Dialog Normalization value in the DP-569 Dolby Digital (AC-3) encoder. In this mode, the Dolby Digital (AC-3) encoder is in two-channel mode, and the Center/Low Frequency Effects and Left Surround/Right Surround channels are ignored. When the network switches to 5.1 channel mode, the Orban 6200 DTV audio processor is bypassed, the DP-569 Dolby Digital (AC-3) encoder switches to 5.1 channel mode, and Dialog Normalization is obtained via the metadata connection from the DP-572 Dolby E decoder.

The Orban 6200 is padded to have exactly one frame of delay and accepts house sync through its AES3 sync input. The Dolby Digital encoder is configured so that it has one less frame of delay when the 6200 is active than when it is bypassed.

Orban is currently testing this scenario with the assistance of Dolby Laboratories.

orban nab

Drawing courtesy of Dolby Laboratories Inc.

Back to Tech Topics