Pre-emphasis

From the broadcast.audio wiki · Chapter 04 of 9

Part of a series on
Broadcast audio processing

Chapters

Reference

Contents

1The noise is not flat
2The fix, and it is an old one
375 microseconds and 50 microseconds
4Why this is a processing problem, not just a filter
5What that looks like in practice
6What de-emphasis gives back, and what it doesn't
7One product note, kept narrow
8See also

If you want to understand why FM sounds like FM — not "warm", not "nostalgic", but the specific forward, dense, controlled thing it does — this is the chapter. Almost all of it comes from one filter that nobody hears directly.

The noise is not flat

Start with a fact about frequency modulation that has no equivalent in the digital world.

In an FM system the information is carried by how fast the carrier's frequency is moving. At the receiver, the demodulator recovers audio by measuring that rate of change. Noise picked up along the way perturbs the carrier's phase; because the demodulator differentiates phase to get audio, a given amount of phase noise produces an audio-band noise voltage proportional to the audio frequency.

The consequence: the noise floor of a demodulated FM signal rises at about 6 dB per octave. It is often drawn as a triangle, and that is exactly what it looks like. At 500 Hz there is essentially nothing. At 15 kHz there is a great deal. Hiss — always hiss, always in the top end. That is why a weak FM station sounds hissy rather than noisy: the noise it adds lives almost entirely where cymbals and sibilance live.

Meanwhile, programme material does the opposite. Most music has far less energy at 10 kHz than at 1 kHz. So the worst signal-to-noise ratio in the system occurs exactly where the signal is weakest and the noise is strongest.

The fix, and it is an old one

Boost the treble before the noise gets added; cut it by the same amount after. The signal comes out flat, and the noise — which was only added in the middle — gets cut by the amount of the treble reduction.

This is the same trick as tape noise reduction, RIAA on a record, and every other analogue medium that had a noise problem in a predictable part of the spectrum. On FM it is called pre-emphasis at the transmitter and de-emphasis at the receiver, and it is not optional or negotiable: it is built into the broadcast standard, and every receiver ever sold applies de-emphasis whether the station emphasised or not.

The curve is deliberately simple — a first-order rise, defined by a single time constant, τ. The corner frequency where the boost starts to take hold is:

f = 1 / (2πτ)

Above that corner the response climbs at roughly 6 dB per octave. The receiver applies the exact complement — a first-order roll-off with the same time constant — and the two cancel.

75 microseconds and 50 microseconds

Two values are in use, and which one applies depends on where you are:

75 µs — North and South America, and South Korea. Corner at about 2.1 kHz. Boost at 15 kHz: about +17 dB.
50 µs — Europe, Africa, most of Asia including Japan, Australia and New Zealand. Corner at about 3.2 kHz. Boost at 15 kHz: about +13.7 dB.

Same idea, different aggression. The 75 µs curve buys more noise improvement and costs more headroom; 50 µs is the more conservative trade. Receivers are built for their market and are not switchable, so the two systems are not interchangeable: a 75 µs signal played through a 50 µs receiver arrives with a rising top end, and a 50 µs signal through a 75 µs receiver arrives dull. In practice this only bites people importing receivers or listening to a stream of an off-air recording made in another region.

Here in Israel, and across Europe, 50 µs is the constant. If you grew up in the United States, the FM you remember was shaped a little harder than the FM I grew up with — and, as the rest of this chapter explains, that difference propagated all the way through to how stations in each region sound.

Why this is a processing problem, not just a filter

Here is the part that turns an innocuous noise-reduction scheme into the central fact of broadcast processing.

The transmitter's modulation limit applies to the emphasised signal. The FM system permits a maximum frequency deviation (chapter 8), and that limit is enforced on whatever actually reaches the modulator — which is the audio after the pre-emphasis boost. There is no allowance for the fact that the treble was artificially lifted on the way in.

So consider what a bright record does. Its cymbals, which on your meters looked perfectly reasonable, are now 13 to 17 dB louder than they were. They, not the kick drum, not the vocal, now define the peak level of the whole programme. And to keep them legal you would have to turn the entire signal down by that much.

That is an unacceptable trade. You would be giving up 15 dB of loudness on everything, permanently, because of hi-hats. No station would ever do it, and the earliest broadcast processing gear existed largely to avoid doing it.

The alternative is to control the high frequencies specifically, which means the processor has to make its peak decisions in the emphasised domain — apply the pre-emphasis first, then look at the peaks, then act. There is no way around this. A peak decision made on flat audio is a decision about the wrong signal.

What that looks like in practice

The classic arrangement is:

Apply pre-emphasis. The signal is now tilted, and its peaks are dominated by treble.
High-frequency limiting. A fast, treble-specific gain reduction that keeps the emphasised top end from running away. It is fast — it has to catch individual transients — and it acts on the emphasised content rather than on the whole signal, so the midrange and bass are not pulled down when a cymbal arrives.
Clipping, in that same emphasised domain (see chapter 5 for why a clipper rather than another limiter).
De-emphasis, if the following stage expects flat audio — or no de-emphasis at all, and hand the emphasised signal straight to a stereo generator that is expecting it.

Step 2 is the one you can hear, and it is a large part of the FM signature. Heavily processed FM has a top end that is always present but never gets away from you. Cymbals do not get brighter when they get louder. A vocal with a lot of sibilance gets its sibilance held to a ceiling. It is a very particular kind of controlled brightness, and once you have heard it described you cannot un-hear it on a commercial station.

Step 3 has a subtler and rather elegant consequence. Clipping generates distortion products, mostly harmonics above the frequency being clipped. If you clip in the emphasised domain, those products are created in the treble — and then the de-emphasis (whether in the processor or in the listener's receiver) attenuates the treble, and attenuates the distortion along with it. Emphasised-domain clipping is therefore less audible in the final result than the same amount of clipping done flat, while also being the clipping that the modulation limit actually requires. You get to spend your distortion budget where it is both needed and cheapest.

What de-emphasis gives back, and what it doesn't

De-emphasis restores the tonal balance. It does not restore the dynamics.

The limiting and clipping happened while the signal was tilted. Once treble peaks have been held to a ceiling, de-emphasising the result brings the tone back to flat but leaves the treble compressed. Once clipping has filled in the space between the peaks, no filter removes it.

That residue is the FM sound. Forward, dense, coherent, with a top end that sits at a consistent level regardless of programme, and a midrange that feels closer to you than the same record does at home. It is not a flaw and it was never a side effect — it is the audible trace of a chain doing exactly the job the physics of the medium required.

One product note, kept narrow

TRANSMIT Tower runs this sequence, in the audio domain, when EXTREME is engaged: a 50 µs pre-emphasis curve, high-frequency limiting against that curve, clipping while the signal is still emphasised, then the complementary de-emphasis to bring the tone back. What comes out is ordinary flat stereo audio — the point is the character the sequence leaves behind, not compliance with a transmitter.

Two honest qualifications. The curve is fixed at 50 µs; there is no 75 µs option. And with EXTREME off there is no pre-emphasis in the signal at all — which does not mean there is no clipping. The clipper runs on every preset; without emphasis it simply clips a flat signal, which is a different sound and a different set of trade-offs. Emphasis is what changes where the clipper spends its effort, not whether it works.