2 Measuring the invisible
A measurement looks like the moment a judgement becomes a fact. Noise, on the account just given, is a relation — a sound met by a person in a place — and defining it is already to decide what will count. One step further along that decision sits the instrument. When an acoustician places a meter at a façade and reads a number in decibels, the number seems to settle the matter: whatever the neighbours feel, this is how loud it is. But the number is not a bare fact about the air. It is the output of a chain of choices about which frequencies to count, how heavily, and over what span of time, and each of those choices was made by people, for reasons, with consequences. The decibel that governs European noise policy has a filter built into it, and the filter decides in advance that some of the most disturbing sound in the modern city will not register. The claim is not that the metric is wrong, but that its blind spot is a choice — and that others benefit from what it cannot see.
2.1 The decibel is a choice
Start with the weighting. A raw sound-level meter, measuring the actual pressure variation across the whole audible range, would report a quantity that bears little relation to how loud a sound seems, because the ear is not an even instrument. We hear a tone at 1,000 hertz far more readily than a tone of the same physical intensity at 50 hertz; the low end of the spectrum has to carry much more energy to seem equally loud. The foundational measurement of this unevenness was made by Harvey Fletcher and Wilden Munson at Bell Labs, who in 1933 had listeners match tones across the spectrum against a reference and mapped the contours of equal loudness — the curves that show how many decibels each frequency needs to sound as loud as the next (Fletcher & Munson, 1933). Their contours were refined over the following decades and are now maintained as an international standard (International Organization for Standardization, 2023), but the basic finding has never been overturned: human hearing is markedly less sensitive to low frequencies, especially at moderate levels.
A-weighting is the engineering response to that finding. The filter, standardised in the specifications for sound-level meters (International Electrotechnical Commission, 2013), discounts each frequency roughly in proportion to the ear’s insensitivity to it. Its shape is, in effect, the inverse of one of those equal-loudness contours — the one at about 40 phons, which corresponds to a quiet sound. Apply it and a measurement in “A-weighted decibels”, written dB(A), tracks perceived loudness far better than the raw figure would. So far this is entirely reasonable. The difficulty is what A-weighting does at the bottom of the spectrum, and the fact that it was calibrated for quiet sound and then applied to everything.
How much it discounts, and where, decides everything that follows. At 1,000 hertz the weighting does nothing, by definition. At 500 hertz it removes about three decibels. But descend into the low frequencies and the discount becomes severe: at 100 hertz A-weighting subtracts roughly nineteen decibels, at 58 hertz about twenty-eight, at 50 hertz over thirty. A sound sitting at 50 hertz is, after weighting, treated as though more than thirty decibels of its energy were not there. For a mid-frequency conversation this correction is a fair approximation of hearing. For a deep, sustained hum it is close to erasure. The correction applies by default to every environmental measurement, regardless of whether the sound in question is a voice or a drone. The choice made for the ordinary case governs the extraordinary one silently.
An unweighted option does exist. Z-weighting, a flat response provided in the same meter standard, reports the actual acoustic energy with no frequency shaping at all; it is simply not what environmental regulation uses. The decision to weight, and to weight this way, is a decision to see the spectrum through the ear’s daytime sensitivity and to let the rest fall away.
2.2 What \(L_\text{den}\) averages away
Weighting handles the frequency dimension. A second set of choices governs time, and they compound the first. European noise policy does not report the sound at a moment; it reports a long-term average called \(L_\text{den}\), the day–evening–night level, defined in the Environmental Noise Directive (European Parliament and Council of the European Union, 2002). The indicator takes the A-weighted level and averages it across a representative year, adding a penalty to the quieter periods — five decibels to the evening, ten to the night — on the reasoning that a given sound intrudes more when the surroundings are quiet. As a summary of overall annoyance across a population, \(L_\text{den}\) is a considered and defensible construction.
But averaging is itself a filter, and it removes a particular kind of information: the difference between a steady sound and an intermittent one. A continuous drone that never lets up and a sound that is silent for most of the hour but punctuated by sharp events can return the same averaged figure, while being experienced as entirely different afflictions. The metric that reports their equality is not lying; it is answering a question — how much acoustic energy, on average, weighted for the ear — that happens not to be the question a sleepless resident is asking. What persists, what is tonal, what recurs at three in the morning: these are the properties that a single long-term average is built to smooth over.
And beneath the averaging sits a hard boundary that is easy to miss. The Directive’s own assessment methods define the calculation over octave bands from 63 hertz to 8 kilohertz (European Parliament and Council of the European Union, 2002). Everything below that floor — the deep low-frequency region, and the whole of infrasound beneath about 20 hertz — falls outside the range the official method computes at all. The metric does not quietly discount the lowest frequencies. By regulation, it does not look there. A source whose energy lives largely below the floor can be, in the terms the law recognises, almost silent.
2.3 A 58 hertz tone that disappears
A single tone makes this concrete. Imagine a steady tone at 58 hertz — the sort of deep, felt hum that a large piece of machinery can radiate — arriving at a bedroom window at a genuinely disturbing physical level. Measured without weighting, it is plainly there. Pass it through A-weighting and nearly twenty-eight decibels are subtracted, because the filter treats 58 hertz as something the ear barely registers. Express the result as a day–evening–night average and the tone, present every second of every night, is folded into a figure dominated by whatever mid-frequency sound shares the record. On paper the location may sit comfortably below the threshold that would trigger any regulatory concern. In the bedroom the hum is the reason no one sleeps.
The simulator below runs the calculation on any tone or spectrum the reader sets, applying the A-weighting curve and reporting how much of the low-frequency energy the single dB(A) figure discards.
Source · A-weighting per IEC 61672-1:2013, Annex E (International Electrotechnical Commission, 2013); equal-loudness basis after Fletcher & Munson (1933) and ISO 226:2023 (International Organization for Standardization, 2023). Weighting values computed from the standard analytical expression. Accessed 2026-07-11.
None of this is a flaw in the arithmetic. A-weighting does what it was designed to do, and \(L_\text{den}\) summarises what it was designed to summarise. The problem is the gap between what the instrument measures and what harms a person — and that gap is not evenly distributed across the spectrum. It opens widest exactly where the low-frequency, tonal, night-time sources live.
From late 2014, residents of the Brittany Heights neighbourhood in Chandler, Arizona, lived beside a data centre that produced a constant hum, one that did not stop at night. Noise-cancelling headphones and earplugs made no difference, and complaints to the authorities went nowhere, though the city eventually amended its zoning code and, years later, rejected a further facility, with noise central to the opposition (Environmental and Energy Study Institute, 2025). What had stalled the complaints was the measurement itself. Cooling-plant noise is concentrated in the low-frequency range, so a standard decibel meter captures it poorly, and without a reliable measurement the ordinance cannot be enforced. A facility can read as compliant while a neighbourhood cannot sleep — the concentrated form of this source is the subject of Chapter 7.
Environmental health officers in the United Kingdom have long met a particular complaint: a resident describes a low-frequency noise that is intense, even deafening, while a visitor standing in the same room hears nothing at all. Because the established noise descriptors are built on the A-weighted level, which discounts exactly this part of the spectrum, the standard criteria proved inappropriate for such cases. The response was an institutional admission of the gap. The Department for Environment, Food and Rural Affairs commissioned a team at the University of Salford to build a dedicated procedure for assessing low-frequency noise complaints, since the ordinary metric could not see the sound that was tormenting people (Moorhouse et al., 2005).
An illustrative case. Picture a resident beside a mixed-use building whose ventilation plant runs a steady tone through the night. A council officer visits at eleven in the morning, takes a fifteen-minute dB(A) reading at the façade, finds it below the ordinance limit, and closes the file: no breach. The snapshot is accurate and beside the point. It cannot register the tone at three in the morning, present every night, discounted by A-weighting and folded into a daytime average. Nor can a quarter-hour visit capture what months of fragmented sleep do to blood pressure and cardiovascular risk, the mechanisms set out in Chapter 3. Measured compliance and accumulating harm are not in contradiction; they answer different questions. A spot measurement asks whether a limit was exceeded on one morning. The resident’s body is integrating a dose over years, and the health toll can be far graver than any single reading suggests.
2.4 Who benefits from the invisible
That the standard metric is a poor guide to low-frequency annoyance is not a heterodox claim but the settled conclusion of the specialist literature. Reviewing the evidence on low-frequency noise, Leventhall states plainly that assessment methods based on the A-weighted equivalent level are inadequate for it, that they underestimate its effects, and that they lead regulatory authorities to incorrect decisions (Leventhall, 2004). The complaint and the measurement come apart, and they come apart in a systematic direction: the sound that people find hardest to live with is the sound the meter is least equipped to record.
When a metric cannot see a harm, the harm does not thereby cease; it becomes, administratively, a non-event. A resident kept awake by a hum that the official measurement cannot find is not merely unlucky. They are on the wrong side of a boundary that was drawn, for good reasons in another context, in a way that happens to exempt the source troubling them. And someone is always on the other side of that boundary: the operator whose emission is lawful because it is unmetred, the development that passes an assessment it would fail if the assessment looked lower. No conspiracy is required. An instrument distributes visibility in the ordinary course of its working, and with it, responsibility.
What follows depends on it, because the sources changing our acoustic world fastest are concentrated precisely in the region the metric cannot see. The continuous low-frequency drone of data-centre and cooling infrastructure is almost purpose-built to exploit this blind spot 7. And the governance that is supposed to contain such sources inherits the blind spot wholesale, since it is built on the very metric that cannot register them 11. The A-weighted decibel is where the book’s spine begins: a measurement presented as neutral, carrying a choice about what counts, made visible only by asking who gains from what it leaves out.