The increased dynamic range allowed by new loudness measurement techniques is often overlooked.
Thanks to the ITU-R BS.1770 multichannel loudness algorithm, audio is the new Cinderella. As broadcasters and content producers seek to comply with legislation such as the CALM Act in the United States, or implement the latest standards to enable material interchange or avert the imposition of new local legislation, audio is being thrust from comparative obscurity into the spotlight of the video world. What is less understood is that ITU-R BS.1770 also includes an algorithm for calculating the true-peak amplitude of a signal and now defines the peak program level delivery requirement in terms of this new true-peak measurement, dBTP (dB relative to digital full scale, measured as true-peak) rather than traditional PPM-type meters.
The need for true-peak measurement
Before the development of ITU-R BS.1770, peak program level was used to control loudness, albeit far from perfectly. I stress the imperfection because content producers soon realized that peak program level and subjective loudness have a weak correlation, which can easily be manipulated through simple tools like audio level compression. In many cases, this has led to the introduction of variable peak program level limits depending on the type of audio content. Speech may be allowed to peak at PPM 6 on a BBC Type II PPM meter, for example, whereas pop music, which is usually subject to level compression, may be limited to PPM 5.
A limit of PPM 6 retains 10dB of headroom in the digital domain, not because 10dB of headroom is really necessary, but because it provided a degree of restriction on the subjective loudness of any material. Maintaining that much headroom means that a substantial amount of the potential dynamic range is being sacrificed, but with the advent of a subjective loudness algorithm, much of this excessive headroom can come back into use. ITU-R BS.1770 allows for a maximum peak program level of -2dBTP (dB relative to full scale, measured as a true-peak value) and the more recent EBU R-128 recommendations, which are due to be incorporated into ITU-R BS.1770, push this even further to -1dBTP, giving content producers incredible scope for dramatic use of audio dynamics.
Of course, there is a catch. When you are working with digital audio, you are not dealing directly with the underlying continuous time signal that will eventually be played out through the viewer's audio system; rather, you are dealing with just a discrete time (sampled) representation of it. Some forms of audio processing, such as sample rate conversion — which has to recreate the underlying continuous time signal at points between the existing samples — can lead to significant changes in sample levels.
When you have 10dB of headroom, you can get away with being a bit imprecise in your metering without it causing any significant problems. Real PPM meters are pseudo-peak meters that use specific meter ballistics to provide an approximation of the underlying continuous time signal. But many PPM-type peak program level displays are really peak-sample meters that are just displaying the maximum values of the discrete time audio samples rather than the maximum value of the continuous time signal the samples represent.
It's easy to find situations where peak-sample meters can display under-reads of several dB compared to the underlying continuous time signal. (See Figure 1 on page 12.) Depending on the spectral content, this can be significantly larger. While a couple of dB error is far from ideal, it may have been acceptable when you had 10dB of headroom. Start pushing your dynamics into the additional range available under ITU-R BS.1770 or EBU R-128 though, and even a couple of dB under-read is going to get you into a whole lot of trouble.
Identifying true-peak metering
There is a simple test that will check whether a meter is a true-peak meter or just a peak-sample meter, and it demonstrates how off the reading can be when using a peak-sample meter. You need a test tone that is just off an integer division of the sampling frequency. So, for 48,000Hz audio as is generally used in broadcast video, a good choice is 12,000.1Hz sine. The extra 0.1Hz means the audio sine wave will appear to shift, relative to the sampling clock, by a whole cycle every 10 seconds. This makes the result of the shift easy to view on the meter.
Now we know that the amplitude of our test tone should be constant irrespective of the phase shift, and you can double-check this either using a known true-peak meter or by attaching an oscilloscope to a good quality analog output of the audio system and viewing the output sine wave directly. Now play the test tone through your metering system and watch the peak program meters you want to check. If they are true-peak meters, you will see them hold a steady level as the audio plays. The additional intersample values created by the true-peak algorithm maintain the reading even as the phase shifts. (See Figure 2.) If they are peak-sample meters, you will see the reading dip four times every 10 seconds as the phase shifts, which causes the under-read. (See Figure 3.)
Ensuring true-peak compliance
One way of ensuring true-peak compliance is to ensure that all metering is true-peak and then manually mix to protect the true-peak limit. Unfortunately, this may not be possible within an NLE or DAW system and could well require suitable IO and an external metering system. Of course, you also probably need to meet loudness requirements so you would have to manually balance the needs of the loudness and true-peak requirements. As most loudness requirements are based on an average of the whole content, this can be time-consuming as you may have to repeatedly play the content through a loudness meter to check the loudness, particularly once the EBU R-128 recommendations are incorporated into the ITU-R BS.1770 specification because these reduce the loudness tolerance to 0.1LU (the equivalent of 0.1dB).
Author's additional note: The +- 1LU allowance in EBU R128 is only '... for programmes where an exact normalisation to Target Level is not achievable practically (for example, live programmes)' - EBU R128 clause (i).
The ATSC A/85 recommendation does allow +- 1LU for all material, which may be cause for confusion.
Be aware that not all loudness meters offer true-peak metering.
An alternative option is to use some form of automatic true-peak correction, most likely coupled with simultaneous loudness correction, which may be available as an NLE or DAW plug-in, or as a stand-alone application. These typically scale the audio uniformly to achieve the required loudness, maintaining the full dynamic range of the source mix, and then, if they provide true-peak correction, apply level compression in the true-peak domain to ensure compliance with the true-peak limits. Because the true-peak requirements provide plenty of dynamic range relative to the loudness requirements, the compression usually only affects a small proportion of the audio and so is not generally noticeable. Be aware that, in the same way that standard normalization will not work for ITU-R BS.1770 loudness control, standard level compressors in NLE and DAW systems generally do not operate in the true-peak domain and do not provide a true-peak limiting solution.
Simon Pegg is research and development director at Eyeheight.