A Controlled Approach to Cable Conditioning: Signal Design for Repeatable Burn-In

1. Introduction

Most burn-in practice reduces to "play music loudly for a few days". Music is a poor conditioning signal: its spectrum is uncontrolled, its crest factor varies from track to track, and its energy is concentrated where the mastering put it, not where the cable and load will actually operate. If conditioning does anything at all, and we take no position on audibility beyond what controlled listening can demonstrate, it should at least be done with a signal whose properties are known and repeatable.

The requirements we set were: full audio-band coverage with energy distributed predictably across octaves; a known and bounded crest factor; no net DC component over a complete program cycle; a level reference that survives translation between playback systems; and a structure that does not demand continuous maximum excursion of the connected loudspeakers.

2. Signal design rationale

2.1 Equal energy per octave

Audio perception and most driver behaviour are organised on a logarithmic frequency axis. White noise places equal energy per hertz, which means the top octave of the audio band carries as much energy as everything below it combined. We shape noise to approximately -3 dB per octave (pink noise), giving equal energy per octave, using a cascaded first-order IIR approximation of the type commonly attributed to Paul Kellet. Each octave of the spectrum then does equal work during conditioning, and the tilt readout on the tool provides a live check that the playback chain is passing the intended spectrum.

2.2 Sweep rate and loop construction

The sweep is logarithmic: instantaneous frequency rises by a fixed number of octaves per second from 5 Hz to 40 kHz. A logarithmic rate spends equal time in each octave, matching the noise program's energy distribution. A nominal 16-second cycle is fast enough that any resonance in the cable, connectors or load is visited thousands of times over a long session, and slow enough that each traversal is essentially quasi-static. The cycle length is quantised so the phase at the loop point is an exact integer multiple of one cycle, making the loop seamless and click-free. Because generation happens at the playback device's sample rate, the upper bound is clamped to 90% of Nyquist; a 44.1 kHz chain sweeps to about 19.8 kHz.

2.3 Duty cycling

Continuous full-spectrum drive is thermally unkind to voice coils and offers no recovery interval. The Cascade program therefore duty-cycles: sweep and noise blocks of six minutes each are separated by three-minute rest blocks, giving a 60% duty cycle across a 30-minute cycle. Rest blocks also make the program self-limiting in the presence of thermal compression: drivers return to ambient between exposures.

2.4 Crest factor control

The sweep has the crest factor of a sine wave, about 3 dB. The shaped noise measures roughly 10 to 12 dB. Both are measured, not assumed: the tool computes the crest factor of each rendered buffer and uses it to enforce the level ceiling. A controlled crest factor matters because peak excursion, not average level, is what stresses mechanical clearances and clips electronics.

2.5 Polarity alternation

Any real playback chain has some DC offset and asymmetric behaviour. Driven for hundreds of hours in one polarity only, a conditioning program could in principle impart a small net magnetisation or mechanical bias. The Cascade program inverts signal polarity on alternating signal blocks, so over a full cycle the net DC component is nominally zero and long-term mechanical bias cancels.

2.6 Level calibration

Program level is specified in dBFS RMS, defaulting to -20 dBFS RMS. dBFS is a digital-domain reference: it fixes the signal relative to full scale at the DAC, which is the only level reference that survives translation between playback systems unchanged. The analogue level after the volume control is deliberately out of scope, because conditioning does not require high acoustic output. A hard cap keeps peaks below -6 dBFS regardless of the requested RMS level, using the measured crest factor of the active signal.

2.7 Red (Brownian) noise

Red noise integrates white noise through a leaky integrator, giving roughly -6 dB per octave: energy concentrated in the bass, where woofer suspensions and large-signal mechanical behaviour live. The leak term bounds the DC wander that a pure integrator would accumulate. We include it because pink noise still puts substantial energy into the top octaves that a woofer cannot use; red noise spends nearly all of the session's energy where driver excursion actually occurs.

2.8 Stepped warble tones

The warble program steps through 1/3-octave bands from 20 Hz upward with a dwell of about five seconds per band, in the tradition of the bass-decade tracks on commercial test discs such as the Stereophile test CDs, extended here to cover the full band. Each tone is frequency-modulated by +/-5% at 2 Hz. The modulation is the point: a static sine parks on whatever room or driver resonance it lands on, while a warble straddles it, so conditioning energy is delivered across each band rather than into a single standing wave.

2.9 Maximum-length sequence

A maximum-length sequence (MLS) is a deterministic pseudo-random binary sequence generated by a linear feedback shift register over a primitive polynomial. Its spectrum is flat like white noise, but it is exactly periodic: the buffer holds one complete sequence period, so the loop is seamless by construction and every repeat is bit-identical. Its crest factor is 0 dB, the lowest of any signal here, which lets it deliver the highest RMS energy of the set within the peak ceiling. Determinism also makes it the most repeatable broadband stimulus in the set: two sessions differ only in level.

2.10 Amplitude-modulated sweep

The AM sweep multiplies the logarithmic sweep by a slow tremolo at about 0.5 Hz and 20 dB of depth, with the tremolo rate quantised to an integer number of cycles per loop to keep the seam silent. Static tones and steady noise exercise a cable and its load at constant power; the tremolo adds a dynamic envelope, cycling the system repeatedly between low and high drive. The rationale is the same one that argues for music over test tones, but with the spectrum and the modulation rate still fully specified.

2.11 Multitone cluster

The multitone program sums 20 discrete tones, log-spaced across the band, each frequency quantised to an integer number of cycles per loop so the cluster repeats seamlessly. Where a sweep visits each frequency briefly and in sequence, a multitone exercises all of them simultaneously and continuously, which also exercises intermodulation behaviour that a sequential sweep never produces. The tone count is a deliberate compromise: dense enough to cover the band, sparse enough that individual tones remain resolvable.

2.12 High-crest bursts

The burst program plays a 0.4-second pink-noise burst every four seconds: a 10% duty cycle of high-level transients with long recovery gaps. It exists for dynamic-impact conditioning, exercising attack behaviour rather than steady-state drive, and it is calibrated differently on purpose: the level setting refers to the burst itself, not the cycle average, so the bursts play at the selected level while the average is about 10 dB lower. This is the most aggressive program in the set, and it carries a corresponding warning in the interface.

3. Session definition

A conditioning session is defined by four parameters: program (Cascade, or any of the individual signals described in section 2), duration (24, 100 or 300 hours, or custom), level in dBFS RMS, and the playback chain itself. Because the first three are exact, a session can be interrupted and resumed, or repeated months later, with no ambiguity about what was delivered. Session state is kept in the browser's local storage for exactly this reason.

4. Limitations

This paper specifies a signal, not an outcome. It does not claim that conditioning produces an audible change in a given cable; that is a question for controlled listening tests such as the ABX protocol described in our companion paper. The digital-domain level reference says nothing about acoustic level, which depends on amplifier gain and loudspeaker sensitivity, so users must set playback volume conservatively. The Nyquist clamp means the extreme top of the specified band is only exercised on chains running above 44.1 kHz. The burst program is calibrated on its active window rather than its cycle average, so its subjective loudness exceeds what the level setting implies for the other programs. Finally, loudspeakers and headphones can be damaged by full-spectrum signals at high level, and nothing in the signal design removes that responsibility from the operator.

5. Conclusion

Conditioning with an uncontrolled signal produces an uncontrolled result. The signal set defined here is spectrally predictable, bounded in crest factor, free of net DC, duty-cycled for thermal safety and calibrated against a digital reference. Whatever conditioning does, it can now be done the same way twice.

References

1. ITU-R BS.1116-3, Methods for the subjective assessment of small impairments in audio systems, International Telecommunication Union, 2015.
2. ITU-R BS.1534-3, Method for the subjective assessment of intermediate quality level of audio systems (MUSHRA), International Telecommunication Union, 2015.
3. IEC 60268-5, Sound system equipment, Part 5: Loudspeakers, International Electrotechnical Commission.
4. Stereophile Test CD 3, Stereophile magazine (reference for the bass-decade stepped-tone tradition extended in section 2.8).
5. Audio Engineering Society convention papers on listening-test methodology and double-blind testing practice (various years).