The FM signal path

From the broadcast.audio wiki · Chapter 08 of 9

Part of a series on
Broadcast audio processing

Chapters

Reference

Contents

11. Programme audio
22. The audio processing chain
33. Pre-emphasis
44. The 15 kHz brickwall
55. The stereo generator — the multiplex
66. Deviation — the hard limit
77. Composite clipping
88. Exciter, transmitter, antenna
9Where a plugin's responsibility stops
10See also

This chapter is the map. Everything in chapters 2 through 7 happens somewhere on it, and knowing where each thing sits explains a lot of otherwise arbitrary-looking design decisions — including where a plugin's job ends, which is the last section and the honest one.

We go from the studio output to the antenna, in order.

1. Programme audio

Two channels leaving the studio, at whatever level the desk and the playout system produce. Mixed sources, mixed eras, mixed mastering. Everything in chapter 1.

2. The audio processing chain

Levelling, multiband, low-frequency work, stereo work, peak control. Chapters 2, 3, 5, 6 and 7. This is the part with the taste in it, the part stations spend money and argument on, and the part a plugin can genuinely be.

By the end of it the programme is at a consistent loudness, spectrally consistent, and peak-controlled. What it is not yet is legal to transmit.

3. Pre-emphasis

The treble tilt from chapter 4: 75 µs in the Americas and South Korea, 50 µs across Europe, Africa, most of Asia and Australasia.

Where this physically happens varies. Some processors apply it internally, do their peak work in the emphasised domain and then de-emphasise before their output, leaving the actual emphasis to a later box. Others apply it and hand the emphasised signal straight downstream. Either way, the ordering constraint from chapter 4 is absolute: the peak decisions must be made after the emphasis, because that is the signal the modulator sees.

4. The 15 kHz brickwall

Every FM audio path ends at a hard low-pass at 15 kHz, and the reason is not fidelity. It is that the space above 15 kHz is already spoken for.

The stereo pilot sits at 19 kHz. The difference-signal subcarrier occupies 23 to 53 kHz. Programme audio above 15 kHz would run into the pilot's frequency and into the lower edge of the difference signal, which does three bad things: it degrades stereo separation, it can interfere with the receiver's pilot detection (so the receiver's stereo indicator flickers, or it blends in and out), and in some receivers it beats against the pilot to produce an audible whistle. A 16 kHz tone against a 19 kHz pilot gives you 3 kHz, right in the middle of the vocal range.

So: a steep filter at 15 kHz, and it has to be steep, because 15 kHz of passband and 19 kHz of stopband is not much room. Steep filters ring, ringing is overshoot, and overshoot is deviation you did not budget for — which is why broadcast band-limiting filters treat overshoot control as a first-class concern rather than being ordinary low-passes. The 15 kHz ceiling is also, quietly, part of the FM sound. There is simply nothing above it, ever.

5. The stereo generator — the multiplex

Now the two-channel audio becomes one composite signal, usually called the MPX or the composite baseband. It is assembled like this:

0 to 15 kHz: the mono sum, L+R. Sent as plain audio. This is what every mono receiver hears, and it is why FM stereo was backward compatible from the day it launched — a mono receiver simply ignores everything above 15 kHz.
19 kHz: the pilot tone. A small, constant sine wave, typically taking 8 to 10 % of the total modulation. It does two jobs: it tells the receiver that a stereo signal is present, and it provides the phase reference the receiver needs to recover the difference signal.
23 to 53 kHz: the difference signal, L−R, amplitude-modulated onto a 38 kHz subcarrier with the carrier suppressed, phase-locked to exactly twice the pilot. The receiver regenerates 38 kHz by doubling the pilot and uses it to demodulate this band.
57 kHz and above: data and auxiliary services. The radio data system (RDS, or RBDS in North America) sits at 57 kHz — three times the pilot, again for phase reference — carrying the station name, programme type and traffic flags. Some regions permit further subcarriers above that, at 67 kHz and 92 kHz, for subscription audio or telemetry services.

The receiver reverses this: take the 0–15 kHz baseband as M, demodulate 23–53 kHz as S, then L = M + S and R = M − S. Which is exactly the mid/side arithmetic from chapter 6, implemented in radio.

It also explains the stereo noise penalty from that chapter. The difference signal is recovered from a subcarrier far up the baseband, where the demodulated noise is much higher (chapter 4's triangular spectrum), so stereo reception is roughly 20 dB worse in signal-to-noise than mono under the same conditions. Receivers blend toward mono when the signal weakens for precisely this reason.

6. Deviation — the hard limit

The composite signal modulates the carrier's frequency. How far the frequency swings is the deviation, and this is where all the loudness pressure in broadcast ultimately comes from.

In the standard FM broadcast service, ±75 kHz of peak deviation is defined as 100 % modulation. That is the number. Exceed it and your signal splatters into the adjacent channel, which is both an engineering problem for your neighbours and a regulatory one for you.

The budget is shared. The pilot's 8–10 % comes out of it. Any data or auxiliary subcarriers take their share. What remains — typically around 90 % or a little less — is all the programme audio gets. Some regulators allow the total to be raised slightly above 100 % when subcarriers are present, but the principle stands: it is a fixed allowance, divided between everything you are sending.

The occupied bandwidth that results is estimated by Carson's rule — roughly twice the sum of the peak deviation and the highest baseband frequency, so about 2 × (75 + 53) ≈ 256 kHz. That is why FM channels are spaced 200 kHz apart in the Americas and 100 kHz across much of Europe, with careful frequency planning to keep neighbours apart.

The essential point for a processing engineer: deviation is a peak limit, and loudness is an average. Everything in chapters 3, 5 and 7 exists to raise the second without exceeding the first. It is the same zero-sum game as digital full scale, with regulators instead of a limiter to enforce it.

Some regions add a second, different constraint. Parts of Europe apply a limit on the average power of the multiplex, measured over a rolling 60-second window. A station operating under that rule cannot simply clip its way to loudness — pushing the average up hits the limit even when peak deviation is fine — and the processing strategy there ends up genuinely different from an unrestricted market. It is one of the reasons European FM has historically sounded less compressed than American FM, alongside the 50 µs versus 75 µs difference from chapter 4.

7. Composite clipping

Because deviation is a peak limit, one last trick is available, and it is applied to the composite signal after the stereo generator has assembled it.

The composite waveform's peaks are often the result of the main channel and the difference subcarrier lining up momentarily rather than any single loud thing in the audio. Shaving a small percentage off those composite peaks buys apparent modulation that no amount of audio-domain work could, because those peaks do not correspond to anything a listener hears.

It is not free. Clipping the composite generates harmonics of the composite, and the composite is full of things you must not disturb: land energy on 19 kHz and the pilot is corrupted and stereo decoding wobbles; land it at 57 kHz and data reception degrades. It also reduces stereo separation. So it is used in small amounts — a few percent — usually with a protection filter notching the pilot region, and it remains one of the more argued-about practices in the field.

The thing to hold on to: composite clipping can only exist after the stereo encoder. It operates on a signal that contains a pilot and subcarriers. Nothing upstream of the stereo generator can do it, by definition.

8. Exciter, transmitter, antenna

The exciter modulates the carrier with the composite. The power amplifier raises it to the licensed power. A mask filter keeps out-of-band emissions within the regulatory envelope. The feedline carries it up the tower and the antenna radiates it.

Then physics takes over: terrain shadowing, multipath from buildings and hills, the receiver's own front end, and the four-dollar amplifier in the dashboard. None of this is under anyone's control, and all of it is the reason chapter 1 exists.

Where a plugin's responsibility stops

Plainly, because it matters.

A plugin runs in a host and produces ordinary stereo audio. It works in the domain of stages 1 through 4 — programme audio, processing, emphasis and band-limiting. It cannot do anything at stage 5 or beyond.

Specifically: a plugin cannot generate a composite signal. It has no pilot, no 38 kHz subcarrier, no RDS. It cannot perform composite clipping, because there is no composite to clip. It cannot set or measure deviation, because deviation is a property of a modulated radio carrier and there is no carrier. And it cannot replace the stereo generator in a real transmission chain — if you are feeding an actual transmitter, the exciter still needs its own multiplex generation and its own composite-domain protection, and those remain entirely that equipment's job.

TRANSMIT Tower stops at the audio, deliberately. Its EXTREME section runs the emphasis, high-frequency limiting, clipping and de-emphasis sequence from chapter 4 in the audio domain, and hands you flat stereo audio at the other end. That sequence is where the recognisable character comes from, which is the part that is worth having in a plugin. The composite-domain work is where the licence comes from, which is not.

Any tool that claims otherwise about a plugin is describing something it cannot be doing.