The advantages of composite clipping

Compared to traditional left/right clipping, composite clipping can give you more than 2 dB's of extra loudness without sacrificing audio quality or dynamics. In this blog posts I'll explain why that's the case.

The pilot and RDS take up space

If you have a signal that's clipped in left/right, and after that is stereo modulated, you have a signal with a fixed peak level. But after that, you still need to add the pilot and RDS signals, which together require about 13.5% of the total signal (assuming 9% pilot + 4.5% RDS injection). The image below shows what happens: You have the full stereo modulated audio signal (the gray block), and you have to add the pilot and RDS (the green line) to that. The green area surrounding the gray block is an area that has been unnecessarily cleared by the left/right clipping - had you known in advance that the pilot and RDS combined were moving in the opposite direction, you could have raised the clipping thresholds at that point.

To clarify: Say your total signal must stay below 1.0. If at some point, the value of the pilot+RDS is -0.1, then your audio could have gone upto 1.1 at that point. Instead, because the pilot+RDS can go upto 0.135, the audio was clipped at 0.8865. That's a difference of more than 0.2, which is 2 dB. Note that this is not possible all of the time, there are (rare) locations where you actually do need to keep the audio below 0.8865. But it's never less than that - and almost always more.

Making use of the stereo modulation

It gets worse though. Consider a waveform that peaks much higher on one channel than the other. See the 2 images on the left top. The thin line is the center of the waveform, the thick blue line is the clipping threshold. If you were to clip it like that, you would get the results that you see in the bottom right.

In the 2nd row, you can see what can happen to the L-R after stereo modulation: If it's short enough, the peak has a 50% chance of going in the opposite direction. On the right you can see what this means after modulation (without clipping): There's almost no overshoot left anymore, so clipping doesn't have much impact yet.

In the 3rd row, you can see the resulting signals. Left shows the demodulated audio after composite clipping. The peaks are way louder than the threshold, and the levels are different, so the stereo effect is maintained well. On the right you can see what it looks like after traditional L/R clipping: A lot more audio has been removed and the result is almost mono.

If you were to clip it like that, you would get this result. A lot of the signal is gone, so you'll get a very clipped sound, and on top of that, the result is nearly mono - the fact that one channel was much louder that the other isn't really visible anymore in the end result.

So what do we do when we create a composite signal? Instead of working with the left and right channel, we're going to work with the L+R (just the mono sound) and L-R (the stereo sound, the difference between the two channels) signals. You can see what those signals look like on the right.

(Typically we actually use (L+R)/2 and (L-R)/2).

So let's first check what happens in stereo modulation. Basically, the L-R signal is multiplied by a 38 kHz sine wave. Which means that there's a good chance that the resulting waveform is less loud (the average sample value is 3 dB's lower), and if it's short enough, there's actually a 50% chance that the resulting waveform points in the opposite direction.

If you add the L+R and modulated L-R signals back together again, the result is always the same or lower as the peaks of the left and right signal, and often by a substantial amount. Clipping this resulting signal will have far less impact on the audio than clipping the separate left and right channels has.

And indeed, this is what the demodulated composite clipped signal looks like in the above example. As you can see, it's very close to the original signal, and the stereo difference has not been reduced as was the case for the left/right clipped signal.

Making use of lower and upper sideband differences

We're still not done! There's another trick that we can use to make the end result even louder, with even less clipping applied. Normally, the area around 38 kHz is perfectly symmetrical: The upper and lower sidebands are exactly the same, just mirrorred. It is possible to transmit only one of the two sidebands at twice the level. Most receivers (there are some exceptions) will work perfectly fine if you do that.

So, the next trick that we can use is to analyse both sidebands separately, and when one of the two causes a bigger spike than the other, quickly put in a bit less of the sideband that causes the spike and a bit more of the other. The image below shows the difference between the two sidebands, if you can dynamically - and very quickly - switch between the two, that will greatly reduce the amount of clipping that's being done and hence increase the output volume and dynamics further. The arrows show which is the optimal sideband to choose at those points in time.

Because some receivers really don't like single sideband transmissions and will start to blend to mono, we have to limit this effect. We typically keep it below at most 10% of change to the resulting sample values. At levels below 20%, no receivers seem to suffer from any negative effects. And you still gain close to 1 dB of extra loudness compared to transmitting a perfectly symmetrical signal.

Making use of lower and upper sideband differences

A very small amount of asymmetry at the top part of the L-R area

Here, the whole L-R area is clearly asymmetrical

Improved reception

If the RF signal that's being transmitted gets wider, this tends to cause reception issues, especially in fringe areas an when there are multipath reception issues, and potentially interference to stations at nearbby frequencies. The ITU has published a recommendation (ITU-R SM.1268) that the RF signal should ideally stay inside of. For mono content without overmodulation, this is guaranteed, but when the stereo pilot, modulated stereo signal and RDS subcarriers are added, that's no longer the case. If you broadcast outside of this mask, you're broadcasting in an area where most receivers won't be listening - they will filter it out, which causes distortion in the audio.

A composite clipper has access to all these subcarriers. That's why, inside Stereo Tool, we have added a software-based FM exciter and a spectrum analyzer. These analyze the output of the composite clipper, and adjust the clipper parameters on the fly to force it to stay inside the ITU mask. That does add a small amount of extra clipping, but only in the brief moments where the signal would otherwise exceed the ITU mask.

This filter can be enabled or disabled. In Stereo Tool it's called "Stokkemask (ITU-R SM.1268)". In the Thimeo STXtreme, Omnia.9, Omnia,9sg and Omnia.SST, it's called "RF bandwidth limiter".

The images below show the resulting RF bandwidth without (left) and with (right) this protection enabled.

Without RF bandwidth protection. Each horizontal line is 10 dB, so the overshoots are close to 15 dB.

With RF bandwidth protection. The signal never exceeds the mask.

The total effect on the audio

We started this blog post with the claim that composite clippingis advantageous to the audio. So how advantageous is it really? The biggest effect that it has is that it allows far more high frequencies to get through. In tests with extreme amounts of clipping on white noise, we measured that the high frequencies were more than 5 dB louder with composite clipping than with left/right clipping. While the lows and mids don't directly benefit from composite clipping, they do benefit from the fact that the highs get "out of the way", so they are a bit louder as well. For music and with "sane" levels of clipping, the effect is less extreme, but there's still a very clear difference in the audio level, the audio quality and the amount of dynamics in the audio.

The images below show - for several songs - the demodulated output of left/right clipping, standard composite clipping, and composite clipping with asymmetry between the upper and lower sidebands, processed in Stereo Tool. It's clearly visible that the peak levels are much higher - and hence the amount of clipping is much lower - when using composite clipping, and the effect gets even bigger when asymmetry is added.

This blog post is loosely based on a paper that Hans van Zutphen presented for The Telos Alliance at the 2017 NAB Broadcast Engineering & IT Conference in Las Vegas.

< Back to overview