Low frequency

From the broadcast.audio wiki · Chapter 07 of 9

Part of a series on
Broadcast audio processing

Chapters

Reference

Contents

1Mono bass
2Level-dependent low shelving
3Why low peaks are so expensive
4The bass clipper
5Harmonic generation, as an alternative
6Two things at the edges
7See also

The bottom of the spectrum causes more trouble in a broadcast chain than everything above it combined, for three reasons that stack.

It carries most of the energy — a mix's peak level is usually set by its low end, so the bass decides how much headroom everything else gets. It is where playback systems differ most: a car with a door resonance, a phone speaker that does nothing below 400 Hz, a pair of headphones with a 6 dB shelf built in, all playing the same signal. And low-frequency waveforms are slow, which as we'll see makes them uniquely awkward for any device that controls peaks.

Get the low end right and the rest of the chain has an easier job. Get it wrong and nothing downstream can save you.

Mono bass

Almost every broadcast chain centres the low frequencies, and the reasons are good ones.

Stereo information down there is barely useful. The ear localises low frequencies poorly — wavelengths are long compared with the distance between your ears, and the cues that place a sound get weak and ambiguous. Almost every mix puts kick and bass in the centre anyway.

It is unstable. Low-frequency differences between channels are usually not deliberate imaging but room modes, mic spacing, or a phase error in a transfer. Summed to mono that content partially cancels, and a track loses a chunk of its bottom end on a mono receiver for no reason anyone intended.

It is expensive. Side-channel energy costs peak headroom (chapter 6), and low frequencies have the most energy of anything. On FM it costs modulation directly. Wherever you are transmitting, out-of-phase bass is the least valuable and most expensive content you own. This is an old lesson: vinyl cutting engineers centred the bass long before FM processing existed, because out-of-phase low frequencies made the cutter head move vertically and jump out of the groove.

It makes peak control predictable. If both channels carry identical low frequencies, a clipper sees the same waveform on both and acts equally. If they differ, peak control acts asymmetrically and the image shifts every time the bass hits — a subtle wobble that is hard to identify and impossible to un-hear.

The implementation is simple: high-pass the side channel at a corner typically between 100 and 200 Hz, and leave mid full range. Below the corner both channels get the same thing; above it the stereo picture is untouched. It is a crossover applied to the difference signal only, not a destructive mono-ing of the bottom of the mix.

In TRANSMIT Tower the corner is preset-dependent: 120 Hz on most presets, higher on the speech preset (where there is nothing useful in the low stereo field at all), and considerably lower on the most conservative preset, which is built to leave a recording alone.

Level-dependent low shelving

Here is a perceptual fact with a practical consequence: the ear's sensitivity to low frequencies falls faster than its sensitivity to midrange as things get quieter. This is what the equal-loudness contours describe, and it is why hi-fi amplifiers used to have a "loudness" button.

For a station this means a quiet passage does not merely sound quieter, it sounds thinner — the bass drops out of it perceptually before anything else does, and in a car, where the low-frequency noise floor is already masking your bottom end, the effect is doubled.

A fixed low boost is the wrong answer: it would make quiet passages sound right and turn loud, bass-heavy material into mud, while consuming headroom precisely when there is none to spare.

The right answer is a level-dependent low boost: a few dB of lift applied when the programme is quiet, tapering smoothly to nothing as the programme gets loud. When there is headroom and a perceptual deficit, you fill it; when there is neither, you get out of the way. The detector for this needs to watch programme level over a reasonably long window — hundreds of milliseconds at least — and the boost itself needs to move slowly, or you have accidentally built a low-frequency compressor with an audible pumping character.

TRANSMIT Tower's implementation lifts by a few dB (roughly 2 to 5, depending on preset, around a corner between 60 and 140 Hz), at full strength when the programme is well down — in the region of −35 dBFS — and fading to nothing by roughly −15 dBFS. It is switched off entirely on the speech and most conservative presets, where it would be a liability rather than a help. It is a genuinely useful stage and the one I would miss most if it went away.

Why low peaks are so expensive

Now the awkward part.

A 50 Hz cycle takes 20 milliseconds. Its peak is not a transient — it is a long, slow excursion that occupies a substantial slice of time. This is a problem for anything that controls peaks.

A limiter fast enough to catch it will follow the waveform — gain down on each peak, up in each trough. That is amplitude modulation at the bass frequency: harmonic distortion on the note itself and intermodulation with everything else. You hear a gritty bass and a mix being chewed.

A slower release doesn't follow the waveform, but now the gain stays down for tens or hundreds of milliseconds after each bass note and pulls everything with it. That is the classic sound of a bass line breathing the vocal in and out. More bands help (chapter 3) but do not solve it.

So the low end is simultaneously the largest contributor to your peak level and the hardest thing for a peak controller to handle gracefully. Deal with it deliberately rather than leaving it to the main limiter.

The bass clipper

The specialised answer is to clip the low band on its own.

Split off everything below a low corner — 80 Hz or thereabouts. Clip that band by itself. Then low-pass the result before putting it back, so that most of the harmonics the clipping just created are removed. Substitute the treated band for the original.

This works for a reason specific to low frequencies. A bass note is close to a sine wave, so clipping it produces mostly harmonics well above the fundamental — 150, 250, 350 Hz for a 50 Hz note — far enough up that a filter can remove most of them, which you cannot say about clipping midrange. You get most of the peak reduction for a fraction of the audible cost, with no gain movement anywhere, so nothing above the bass is touched. The harmonic residue you deliberately keep is a bonus: it is what makes bass audible on a small speaker.

The trade: with the harmonics filtered away the band no longer has a flat top, so some of the peak comes back. Band-limited clipping always overshoots. It is a headroom saving, not a hard ceiling, and something downstream still catches the small remainder.

Not every setting wants it. TRANSMIT Tower has a bass clipper of this kind, working around a low corner and replacing the band after shaping, and it is switched off on three of the nine presets — the ones whose job is to leave low-frequency dynamics alone. On the rest, the drive varies from mild to substantial.

Harmonic generation, as an alternative

There is a completely different way to make bass audible, worth knowing about even when you are not using it.

If a playback system cannot reproduce 45 Hz — and a phone speaker cannot, nor a laptop, nor plenty of car speakers — then boosting 45 Hz achieves nothing except wasted headroom, driver excursion and distortion.

But the ear reconstructs pitch from harmonic structure. Given tones at 90, 135 and 180 Hz a listener perceives a pitch of 45 Hz with no 45 Hz present — the "missing fundamental" effect that makes a bass line audible on a transistor radio. So instead of boosting the fundamental you can synthesise its harmonics and let the listener's hearing supply the note.

The costs are real: it is an additive tonal change that sounds wrong on a system that does reproduce the fundamental, it stacks with whatever distortion the chain is already making, and overdone it is a buzz following the bass line. TRANSMIT Tower does not do this — its low-frequency work is the centring, the level-dependent lift and the band clipper above — but if you are processing specifically for phone speakers it is the technique to read about next.

Two things at the edges

High-pass first. Subsonic content — turntable rumble, air conditioning, stage vibration, footfalls, wind — is inaudible on every consumer system and still eats headroom, drives detectors and modulates the whole chain. Removing it at the input is the cheapest improvement in broadcast processing. TRANSMIT Tower high-passes the input between 20 and 35 Hz depending on preset, and much higher on the speech preset. It high-passes again at the end, at a fixed 30 Hz on every preset, because processing puts low-frequency energy back: filters ring, clippers make intermodulation products, band recombination is not perfectly clean.

A footnote on asymmetry. Speech, brass and some other sources have asymmetric waveforms: the peak in one polarity is several dB larger than in the other. Peak control has to work to the larger one, so you lose headroom on a waveform that is not actually louder. Running the signal through a chain of all-pass filters — a phase rotator, which shifts phase without changing magnitude — evens the waveform out, buying a genuine dB or two of headroom for free. It is inaudible on music and quite effective on voice, which is why it is a fixture of broadcast chains and almost unknown outside them.