The dB(A) Blind Spot | Why Standard Noise Meters Miss Low-Frequency Noise & Infrasound

The whole problem in one number

Almost every noise ordinance, landlord, and police meter measures sound in dB(A). "A-weighting" is a filter that mimics how the ear hears quiet mid-range sound, and to do that it deliberately subtracts low frequencies before the number is reported. At 20 Hz it subtracts about 50 dB. A dB(C) meter subtracts about 6 dB at the same frequency, and dB(Z) subtracts nothing.

So the same bass, hum, or vibration can produce a "quiet" dB(A) reading and a loud dB(C) reading at the same moment, in the same room. Neither meter is lying. They are answering different questions, and the question most rules ask is the wrong one for low-frequency noise.

Standard IEC 61672-1 weighting corrections. The shaded band (below 20 Hz) is infrasound; roughly 20–200 Hz is what most people call bass.

What each weighting subtracts

Frequency	dB(A) correction	dB(C) correction	Range
10 Hz	-70.4 dB	-14.3 dB	infrasound
16 Hz	-56.7 dB	-8.5 dB	infrasound
20 Hz	-50.5 dB	-6.2 dB	bass
31.5 Hz	-39.4 dB	-3.0 dB	bass
50 Hz	-30.2 dB	-1.3 dB	bass
63 Hz	-26.2 dB	-0.8 dB	bass
100 Hz	-19.1 dB	-0.3 dB	bass
200 Hz	-10.9 dB	+0.0 dB	bass
1000 Hz	+0.0 dB	+0.0 dB

dB(Z) ("zero" weighting) applies no correction at all and is the honest baseline. G-weighting (ISO 7196:1995) is the international standard built specifically for infrasound: it is most sensitive between about 1 and 20 Hz and nearly blind above that, which is the mirror image of dB(A).

Better than any single number is a 1/3-octave spectrum, which reports the level in each frequency band separately. That is how an acoustician finds a 36 Hz tone or a pulsing blade-pass signature that a single dB(A) figure averages away, and it is what to ask for when a measurement is going to be relied on.

What it looks like in practice

Granbury, Texas — the quiet reading next to the complaints

More than 40 residents near a bitcoin mine described migraines, vertigo, tinnitus, and lost sleep. A county-commissioned study measured only 35–53 dBA at homes. Both things can be true at once: the study measured in the weighting that discounts exactly the sound residents were describing. The residents' nuisance suit survived a motion to dismiss; health causation remains their belief, not an established finding. Documented Cases

The shared wall — why a neighbor's subwoofer passes the test

Low frequencies pass through walls and floors far more easily than speech or music mid-range does, and small rooms can amplify particular bass frequencies by 10–20 dB at a resonant spot, such as a bed against the shared wall. An officer with a standard meter in the hallway may read well under the limit while the bedroom is the loudest place in the building. Detection & Evidence

Where the gap has been recognised

Prince William County, Virginia — a dB(C) rule (2025)

After years of data-center hum complaints, county supervisors adopted a C-weighted noise standard for data centers (73 dB(C) day / 68 dB(C) night, effective May 1, 2026), so the low-frequency component is measured directly instead of being filtered out. Residents called the limits too weak. (Bristow Beat)

Ontario — a court ordered one specific tone fixed (2025)

In Frankel v. York Region Condominium Corporation No. 664, 2025 ONSC 719, an 80-year-old owner's sleep was disturbed by a 160 Hz tonal hum from the building's hot-water pumps above his bedroom. A jointly commissioned acoustic report found the overall level below ASHRAE and WHO guidelines, but the 160 Hz tone stood out. The Superior Court ordered the corporation to attenuate the 160 Hz component within three months, with the acoustician certifying the result at the corporation's cost, and awarded $32,500 compensation plus costs. The single number passed; the tone did not. (CanLII · decision PDF)

Chandler, Arizona — measure the baseline first (2022)

After years of complaints about a hum from a data-center campus, the city's Ordinance No. 5033 (adopted December 2022) requires a new data center to commission a third-party baseline sound study at the nearest homes before it is built, to be designed so its emitted sound does not exceed that baseline, and to repeat the study after opening and annually for five years. A different route to the same end: measure what residents actually hear, rather than only a single weighted number. (Ordinance 5033 · City of Chandler)

Wisconsin — a class action puts the gap in the pleading (2026)

In Ostergaard v. Microsoft Corporation (E.D. Wis., No. 2:26-cv-01169), residents near the Mount Pleasant AI data center allege that much of the "low hum" they live with is low-frequency sound "that is not detected and reflected in dBA measurements", and quote Microsoft's own April 2026 statement that neighbours "have noticed a tonal humming sound" from cooling fans running at high speed. The parties told the court in September 2026 they had reached a settlement in principle; terms are not yet public. An allegation, not a finding, but it shows the argument has reached federal pleadings. (complaint · docket · case record)

Why overstating it backfires

The blind spot is real, and so is the temptation to overclaim on the other side. The research on direct health harm from infrasound at typical residential levels is early, small, and mixed; the strongest findings come from the lab at levels well above what a neighbor's equipment produces. Low-frequency noise reliably causes annoyance and sleep disruption. That is already enough to act on, and it does not need embellishing. See the Research Library and Emerging & Contested Claims.

Germany, 2021 — a 36 dB error that lived for over a decade

The Federal Institute for Geosciences and Natural Resources (BGR) admitted in April 2021 that a programming error in its widely cited wind-turbine infrasound study had overstated levels by about 36 dB (roughly 100 dB reported, about 64 dB real). Those inflated numbers had been quoted by campaigners and in court for years. The Economics Minister apologised publicly. Every later argument that leaned on the figure was weakened with it. (Der Spiegel · taz)

Finland, 2020 — when infrasound was tested blind, it did not explain the symptoms

A Finnish government project measured infrasound near wind farms for 308 days, surveyed residents, and ran double-blind listening tests. Participants could not detect wind-turbine infrasound, and its presence had no effect on reported annoyance or on autonomic nervous-system responses. The lab sample was small (27 people) and the exposures short, but it is the most careful test of its kind so far. (Maijala et al., JASA 2021)

None of this means low-frequency noise is harmless. It means the measurement problem is real and the health claims are still being worked out, and that mixing the two up is how a good complaint gets dismissed.

What to ask for

1.dB(C) or dB(Z), not only dB(A). If a council, landlord, or officer measures, ask them to report C- or Z-weighted levels alongside the A-weighted figure. Many meters do both.
2.A 1/3-octave spectrum for anything that will be relied on, and G-weighting (ISO 7196) if infrasound below 20 Hz is suspected.
3.Time. Low-frequency problems are intermittent and worst at night. Continuous logging over 48–72 hours, with a timestamped incident log beside it, is far stronger than a single visit.
4.Measure in the worst room, at the position where you experience it, with windows closed, and note the suspected source's activity.
5.Keep the record in one place. The free, private Evidence Packet builder logs incidents, imports meter exports, and prints them as one dated document.