Jurisdictions define acceptable environmental noise levels in different ways. Sound is measured in decibels (dB), a logarithmic ratio of air pressure fluctuations to a reference pressure of 20 micro-Pascals. The decibel, unless otherwise noted, is considered to be un-weighted (dB, or dBZ), meaning no correction has been applied to the amplitude of the sound based on its frequency content. In almost all cases, community sound criteria are given in terms of A-weighted decibels (dBA). The A-weighting adjusts the sound pressure level to account for the differences in human hearing at different frequencies.

Other than the overall A-weighted sound level, the character of a sound also affects community perception. Tonal, intermittent, impulsive, or excessively low-frequency sounds are often judged more negatively than typical broadband noise at the same level. Some regulations therefore use C-weighted decibels (dBC), which include more low-frequency content and are more representative of human response to very loud sounds (>85 dBZ) but less representative than A-weighting for human response to quiet-to-medium sounds. Other approaches may set limits for specific octave bands or define, qualitatively or quantitatively, what kinds of pure tones violate a noise ordinance.

In addition to frequency weighting, sound descriptors are needed to capture variability of sound levels over time. The most common is the A-weighted Equivalent Sound Level (Leq), an average of the sound energy over a given measurement period. Leq provides a single-number representation of environmental sound, as opposed to transient instantaneous sound levels that fluctuate over a wide range. The one-hour Leq, denoted Leq,1-hr, is used as a quantitative metric for many environmental acoustic assessments across the United States. Another metric commonly used to characterize steady state sound levels is the L90, which is defined as the sound level exceeded for 90% of the measurement interval. The L90 is typical of continuous sounds and often similar to the minimum sound level, which can characterize the residual ambient, or background, sound level. It is essential to understand the metrics used by a jurisdiction, not just for the potential sound from the data center, but more importantly for the definition of the ambient, or background sound level. An ambient defined by the Leq will be higher than an ambient defined by the L90 when characterizing the quietest times.

Environmental sound spans a wide range. The example below is from a mid-afternoon measurement in a rural area, near a moderately busy two-lane road. The figure below compares the second-by-second sound levels to the hourly statistical metrics, including Lmax, L10, Leq, L50, L90, and Lmin. The difference between the minimum (34 dBA) and maximum (75 dBA) sound levels observed is over 40 decibels. The average (Leq) in this case is 59 dBA, which is above the median (L50) of 55 dBA and L90 of 43 dBA. The dramatic differences between these various metrics in a single hour of data demonstrate the importance of identifying a quantitative definition of the ambient sound level. Defining “typical” sound is therefore difficult for regulators, and the challenge only increases when considering hourly, daily, or seasonal variability.

Some jurisdictions set fixed sound limits (e.g., 55 dBA), which may be further restricted during evenings, nights, or weekends. Different limits are applied based on the zoning or land use, with lower noise limits for noise-sensitive uses such as homes, schools, daycares, healthcare facilities, and places of worship. Because data centers operate continuously, they must address operations during the most restrictive times of day.

Other criteria are defined based on an allowable increase in sound over existing ambient conditions (e.g., 10 dB increase over ambient). Whether a jurisdiction has a fixed or ambient-based regulation, varying background levels matter: even where fixed limits are generous, low ambient noise can make data center sound more noticeable and may increase the likelihood of community complaints.

Many jurisdictions have weak noise regulations that are easy for developers to meet, but compliance alone rarely ensures project success or strong community relations. Data centers should strive to be good neighbors and understand when it is appropriate to go above and beyond the code minimum.
A cautionary example comes from Virginia, a major data center hub. A jurisdiction had an exemption that stated:

Heating and cooling systems, including but not limited to air conditioners and heat pumps, shall not be subject to the night [noise limits].

This exemption was very favorable to data center developers, as it could be interpreted to include any cooling system typically found at a data center. Developers interpreted this broadly to cover large-scale cooling systems, though the regulation was likely intended for residential units near neighbors. When we see similar exemptions, we strongly recommend against taking this exemption to avoid meeting the applicable noise limits provided in that noise ordinance. This particular Virginia jurisdiction ended up with significant community noise complaints from data center operations, and have since revised their noise ordinance to clarify that commercial and industrial cooling equipment are not exempt.

Appropriate noise goals depend on the character of the neighborhood: is it rural, suburban, or urban? Does the community have any experience with similar industrial noise? Are local regulators and community members aware of the suite of noise mitigation incorporated into the data center design? While activities like emergency generator testing may be exempt, developers can still reduce impacts by adding enclosures and limiting tests to daytime hours. Early communication about how a project will operate and strive to be a good neighbor is critical to building community trust.

In a typical air-cooled system, a computer room air-conditioning (CRAC) unit, located inside the building, circulates refrigerant to an air-cooled chiller outside of the building. These chillers may be located on the roof or at ground level. The chillers include large arrays of fans that pull air across the refrigerant-filled compressors. A large data center may have dozens of air-cooled chillers covering the entire roof to provide sufficient cooling

In a typical water-cooled system, indoor CRACs or large water-cooled chillers reject heat via cooling water sent to an outdoor cooling tower. The cooling towers have a very large cooling capacity, so fewer are needed when compared to air-cooled chiller systems.

In some regions, free and evaporative cooling is sufficient to meet cooling loads. Large air handlers pull air through inlet louvers in the side of the building to be filtered, then cooled and humidified by evaporative coolers. This cool air circulates through the hot aisles of the data center before being pulled out of the building through exhaust fans. Air handlers and exhaust fans are characterized by a broad-band spectrum of mechanical noise, although they can have significant tones at low frequencies if not effectively designed.

While not involved in cooling, the electrical equipment may contribute significantly to the noise from data center operations. Transformers, which connect the data center to the local grid, are known to have a low-frequency hum associated with the mains electricity frequency (60 Hz and its harmonics in the United States).

The tones arise primarily from magnetostriction—the slight dimensional change in the iron core as it magnetizes and demagnetizes twice per AC cycle—as well as electromagnetic forces acting on windings and structural supports. These effects cause the transformer core and enclosure to vibrate at 120 Hz and other harmonic multiples, producing a distinct tonal character that can be particularly bothersome to nearby communities.

A large number of diesel or natural gas generators are required to provide backup power to ensure a data center’s operability, even in the case of an area-wide power outage. Large generators  produce noise much like heavy truck engines, including noise from exhaust and cooling airflow, often creating the rumble typical of diesel equipment.

While full-load emergency operation is typically infrequent, the generators must be tested regularly. This routine maintenance can typically occur during daytime hours, and the units are often tested in series; however, even a single generator can produce significant low-frequency sound.