A09.02 Estimating Low-Frequency Noise from a Typical Data Centre

Key Takeaways

• Data centres generate low-frequency noise primarily through cooling towers, chillers, pumps and transformers rather than the servers themselves.
• Engineers estimate environmental noise by combining the sound power of individual equipment before modelling how sound propagates through the surrounding environment.
• Larger data centres require more cooling equipment, but the resulting increase in low-frequency sound power is much smaller than the increase in facility capacity because acoustic energy combines logarithmically.
• Distance, terrain and Malaysian weather conditions—particularly stable nighttime atmospheric stratification—can significantly influence how low-frequency sound propagates.
• Engineering calculations provide valuable estimates of community sound exposure, but actual experience depends on both measurable acoustic conditions and human perception.
• Low-frequency noise is not solely a site-level issue; its potential to affect surrounding communities warrants careful engineering assessment, informed planning and ongoing monitoring.

Introduction

This article forms part of the Malaysia Data Centre Observatory (MDCO) Analyse Series on Low-Frequency Noise (LFN), →A09 Understanding Low-Frequency Noise from Data Centres, from which additional articles examining the engineering, acoustics, health research, governance, regulation and mitigation of low-frequency noise are also accessible.

In →E02.01 How Data Centres Generate Low-Frequency Noise, we examined the engineering mechanisms through which cooling towers, chillers, pumps, transformers and other mechanical systems generate low-frequency sound. However, one question naturally follows:

How much low-frequency noise does a typical data centre actually produce?

The answer is less straightforward than it might first appear because noise is not determined by electrical capacity alone.

Unlike the electrical capacity of a data centre, which can usually be expressed as a single figure such as 100 MW or 300 MW, noise cannot be represented by one number alone. The sound experienced by nearby communities depends not only on the equipment installed, but also on how many units are operating, their arrangement, their acoustic design, the surrounding terrain, weather conditions and the distance between the facility and the receiver.

For this reason, professional environmental noise assessments do not begin by measuring sound at the property boundary. Instead, engineers start with the facility design itself. They estimate the cooling demand, identify the mechanical equipment required, obtain the acoustic characteristics of that equipment from manufacturer data or recognised engineering references, and then predict how the sound propagates through the surrounding environment.

This article follows that same engineering approach.

Using a representative 100 MW hyperscale data centre as a worked example, we estimate the principal outdoor sources of low-frequency noise before examining how those sound levels may change with increasing facility capacity, increasing distance and typical Malaysian weather conditions.

The objective is not to predict the performance of any particular data centre. Every facility differs in its cooling technology, equipment selection, site layout and operating conditions. Rather, the calculations presented here illustrate the engineering principles used to estimate low-frequency noise.

Throughout the article, emphasis is placed on the principal continuous sources of operational low-frequency noise, particularly outdoor cooling equipment. Emergency generators, construction activities and temporary maintenance operations are excluded because they are intermittent rather than continuous and are generally assessed under separate regulatory frameworks.

MDCO Insight: Engineers estimate low-frequency noise by analysing the design of the cooling system before considering how the resulting sound propagates through the surrounding environment.

From IT Load to Cooling Demand

Before estimating the sound produced by a data centre, it is first necessary to understand why the cooling system exists (→E03.01 Understanding Data Centre Cooling Systems).

Almost all electrical energy consumed by servers ultimately appears as heat within the data centre environment.

Processors perform billions of calculations every second by switching electrical transistors on and off. Although these computations create valuable digital services, they also generate heat as an unavoidable consequence of electrical resistance. Memory modules, storage devices, network switches and power supplies contribute additional heat, with virtually all of the electrical energy consumed inside the data hall eventually becoming thermal energy.

From an engineering perspective, a data centre can therefore be viewed as a very large heat-generating system.

The role of the cooling infrastructure is to remove this heat continuously so that equipment remains within its designed operating temperature range. If the cooling system fails to reject heat at the same rate that it is generated, internal temperatures rise rapidly, increasing the risk of equipment failure or automatic shutdown.

Consequently, estimating low-frequency noise begins not with fans or cooling towers, but with the amount of heat that must be removed.

Understanding IT Load and Facility Load

Data centre capacity is typically quoted as IT load.

A 100 MW data centre, for example, normally refers to the maximum electrical power available to servers, storage systems and networking equipment.

However, servers are not the only equipment consuming electricity.

Additional electrical power is required for:

• cooling systems;
• pumps;
• cooling towers;
• fans;
• transformers;
• lighting;
• building management systems;
• security systems;
• auxiliary infrastructure.

The total electrical demand of the entire facility is therefore greater than the IT load alone.

This relationship is commonly expressed using Power Usage Effectiveness (PUE).

PUE is defined as:

Total Facility Power ÷ IT Equipment Power

A theoretical PUE of 1.0 would mean that every watt entering the facility is used directly by IT equipment, with no additional energy consumed by cooling or supporting infrastructure. In practice, this is not achievable because all data centres require mechanical and electrical systems to operate safely and reliably.

Modern hyperscale facilities may achieve annual PUE values between approximately 1.2 and 1.4, although actual performance depends strongly on climate, cooling technology and operating strategy. Facilities operating in warmer and more humid environments may experience different cooling energy requirements compared with facilities in cooler climates.

For the purpose of this illustrative example, we assume a representative PUE of 1.3.

This gives:

Parameter	Assumed Value
IT Load	100 MW
PUE	1.3
Total Facility Load	130 MW

The additional 30 MW represents the energy consumed by cooling systems and other supporting infrastructure.

Almost All Electricity Eventually Becomes Heat

One of the most important engineering principles in data centre design is that almost all electrical energy entering the facility eventually appears as heat somewhere within the facility or its immediate surroundings. However, the location where this heat is released determines whether it contributes directly to the cooling load.

The electricity powering servers is not stored permanently inside the equipment. Instead, it is converted into useful computation and then dissipated as thermal energy through electronic components.

Likewise, electricity consumed by pumps becomes hydraulic work before eventually dissipating as heat through friction. Fans convert electrical energy into moving air, but that energy also ultimately becomes heat within the surrounding environment.

Consequently, from an energy perspective, approximately 130 MW entering the facility will ultimately be released to the surrounding environment as heat.

However, this does not mean that the cooling system must remove the full 130 MW.

The 100 MW IT load is almost entirely converted into heat within the data halls and therefore requires continuous cooling. In addition, a proportion of the supporting infrastructure—including UPS losses, power distribution equipment, lighting and other conditioned plant rooms—also contributes to the cooling demand.

For this illustrative example, we assume that these indoor ancillary systems contribute a further 10 MW of heat, giving a representative cooling load of 110 MW.

The remaining energy is associated with equipment located outside the conditioned spaces or losses dissipated directly to the external environment, such as outdoor electrical equipment, cooling tower fans and portions of the condenser system.

Stage	Representative Value
IT Equipment Load	100 MW
Additional Facility Power	30 MW
Total Facility Electrical Load	130 MW
Heat Requiring Active Cooling	110 MW
Heat Dissipated Directly to External Environment	≈20 MW

In other words, although a 100 MW data centre ultimately releases approximately 130 MW of heat to the surrounding environment, the cooling system in this worked example is assumed to remove 110 MW of heat from the conditioned spaces.

This 110 MW cooling load forms the basis for the illustrative calculations presented in the remainder of this article. Actual facilities will vary depending on their electrical architecture, cooling technology and equipment layout.

MDCO Insight: Total facility power determines the overall heat released to the environment, but the cooling system is sized to remove only the heat generated within the conditioned spaces.

Estimating the Cooling Plant

Once the required heat rejection is known, engineers can estimate the cooling infrastructure needed to support the facility.

Although every data centre is designed differently, modern hyperscale facilities generally follow similar engineering principles. The required cooling capacity is distributed across multiple pieces of equipment rather than relying on a single large machine. This provides operational resilience, improves maintenance flexibility and allows equipment to operate more efficiently under varying IT loads.

For this worked example, we consider a representative 100 MW hyperscale data centre employing a conventional water-cooled chilled-water system, as this remains one of the most common cooling architectures for large facilities requiring high energy efficiency.

Based on published industry guidance and representative manufacturer specifications, such a facility would typically comprise:

• multiple water-cooled chillers;
• multiple cooling towers;
• condenser water pumps;
• chilled water pumps;
• heat exchangers and associated pipework;
• Computer Room Air Handlers (CRAHs) within the data halls;
• electrical transformers and switchgear.

Among these systems, the principal outdoor contributors to continuous low-frequency noise are generally:

• cooling towers;
• outdoor chillers (where applicable);
• condenser water pumps (where applicable) and associated plant;
• transformers.

Indoor equipment such as CRAHs also generates low-frequency sound, but much of this is attenuated by the building envelope before reaching nearby communities.

Accordingly, the remainder of this article focuses primarily on the outdoor mechanical plant, which typically dominates environmental noise assessments.

In the next section, we examine each major equipment type individually, review published acoustic data from representative commercial systems and estimate the range of low-frequency sound that each may contribute to the overall acoustic environment of a 100 MW data centre.

Estimating the Low-Frequency Noise Produced by Each Component

Once the cooling plant has been identified, the next step is to estimate the sound generated by each major item of equipment.

Professional acoustic assessments do not rely solely on measurements at the site boundary. Instead, they begin by understanding the acoustic characteristics of the source equipment.

This distinction is important.

Sound power describes the total acoustic energy emitted by a machine. It is an intrinsic property of the equipment and does not depend on where the measurement is taken.

Sound pressure, by contrast, is what a person or microphone experiences at a particular location. It depends not only on the source itself, but also on distance, weather, terrain, buildings and other environmental factors.

For this reason, engineers first determine the sound power emitted by each major item of equipment before predicting how that sound propagates through the surrounding environment.

Because this article examines low-frequency noise (LFN) rather than general environmental noise, the calculations focus on the low-frequency portion of the sound spectrum instead of the overall broadband sound power.

Rather than analysing every individual frequency, environmental acoustics groups sound into standard octave bands, each representing a range of frequencies.

For the worked example presented in this article, we adopt the 63 Hz octave band, which spans approximately 45 Hz to 90 Hz. This band captures a significant portion of the low-frequency energy generated by large mechanical equipment commonly found in data centres.

The 63 Hz octave band is widely used in environmental acoustic research and guidance, including DIN 45680, the UK DEFRA low-frequency noise procedure, and numerous published studies on environmental low-frequency noise. It provides a practical balance between engineering accuracy and measurement simplicity while allowing meaningful comparison with existing research.

Accordingly, all illustrative calculations presented in this article use the 63 Hz octave band as the representative indicator of low-frequency noise.

Because manufacturers publish acoustic data using different standards, equipment configurations and test conditions, no single sound power value applies universally. Instead, the values presented below are derived from representative manufacturer specifications and published engineering literature.

They should therefore be interpreted as illustrative engineering ranges intended to demonstrate the calculation methodology rather than predict the acoustic performance of any specific data centre.

MDCO Insight: Environmental noise assessments begin with understanding the sound power emitted by individual equipment, before estimating how that acoustic energy propagates through the surrounding environment.

Cooling Towers

Why They Are Important

For large water-cooled data centres, cooling towers are generally the most significant continuous outdoor source of low-frequency noise.

Unlike many other mechanical systems, cooling towers combine two independent noise mechanisms:

• large axial fans moving enormous volumes of air; and
• circulating water falling through the tower.

As discussed in →E02.01 How Data Centres Generate Low-Frequency Noise, the large-diameter fans are the principal source of low-frequency sound, while the falling water contributes mainly to broadband noise extending across a wider frequency range, but is generally less dominant in the low-frequency spectrum.

Because cooling towers operate continuously whenever heat must be rejected, they are often the dominant source considered in environmental acoustic assessments.

Representative Cooling Tower Installation

For the worked example presented in this article, we assume the data centre requires a representative 110 MW cooling load.

Modern hyperscale facilities typically distribute this cooling duty across multiple cooling tower cells to improve operational resilience, maintenance flexibility and energy efficiency.

For this illustrative model, each induced-draft cooling tower cell is assumed to reject approximately 5 MW of heat under full design conditions. This results in approximately 22 cooling tower cells for a 110 MW cooling load.

Each cooling tower is assumed to employ a large variable-speed axial fan approximately 8–10 metres in diameter operating at around 90–120 RPM. Although these rotational speeds appear relatively low, the large fan diameter and blade-pass frequency naturally generate significant low-frequency sound.

Actual hyperscale facilities may employ fewer larger cells or more smaller cells depending on manufacturer selection, redundancy philosophy and site constraints. The purpose of this assumption is not to represent a universal design, but to provide a realistic engineering basis for acoustic calculations.

Representative Low-Frequency Sound Power

Representative acoustic data from cooling tower manufacturers indicate that a modern induced-draft cooling tower typically produces a 63 Hz octave-band sound power level of approximately:

80–92 dB (L_w, 63 Hz octave band)

The range reflects differences in tower capacity, fan diameter, rotational speed, blade geometry and the application of low-noise technologies such as variable-speed drives and acoustic attenuators.

For the representative 110 MW cooling load, we assume the facility employs 22 identical cooling tower cells.

Because sound is measured on a logarithmic scale, the sound power of multiple independent sources cannot be added using ordinary arithmetic. Instead, engineers apply logarithmic addition:

L_total = L_single + 10 log₁₀(N)

where:

• L_single = sound power of one cooling tower;
• N = number of identical cooling towers.

For 22 cooling tower cells:

10 log₁₀(22) ≈ 13.4 dB

Accordingly:

Individual Cooling Tower	22 Cooling Towers Combined
80 dB	93.4 dB
92 dB	105.4 dB

For simplicity, we round these values to:

93–105 dB (L_w, 63 Hz octave band)

This representative range forms the basis for the subsequent facility-level low-frequency sound power calculations.

MDCO Insight: Doubling the number of cooling towers does not double the sound level. Because sound combines logarithmically, each additional tower contributes progressively smaller increases to the total noise.

Technology Matters

Cooling tower design has a significant influence on both the level and frequency characteristics of the emitted sound.

For example:

• larger fans operating at lower rotational speeds generally improve energy efficiency but may shift more acoustic energy towards lower frequencies;
• direct-drive fans eliminate gearbox noise;
• variable-speed operation reduces sound during periods of lower cooling demand;
• optimised fan blade geometry reduces blade-pass tones;
• acoustic attenuators and silencers reduce sound emissions without affecting cooling performance.

Consequently, two cooling tower installations providing the same cooling capacity may exhibit noticeably different low-frequency characteristics.

MDCO Insight: For large water-cooled data centres, cooling towers are typically the dominant source of continuous low-frequency noise, although their acoustic performance depends as much on engineering design as on cooling capacity.

Chillers

Why They Are Important

Chillers remove heat from the chilled-water circuit before transferring it to the condenser water system for final rejection through the cooling towers.

For the representative data centre considered in this article, the chilled-water system is responsible for transferring approximately 110 MW of heat away from the conditioned spaces, with chillers providing the primary heat transfer mechanism.

From an environmental noise perspective, however, chillers generally contribute less low-frequency noise than cooling towers.

This is because most hyperscale facilities employ water-cooled chillers installed within enclosed mechanical plant buildings. The building structure provides significant acoustic attenuation before the sound reaches the surrounding environment.

By contrast, air-cooled chillers are installed outdoors and incorporate large condenser fans, making them substantially more significant sources of environmental noise. Air-cooled systems are more common in smaller facilities and regions where water availability is limited.

Representative Chiller Installation

Modern hyperscale facilities normally distribute the cooling load across multiple chillers to improve reliability and operational flexibility.

For this worked example, we assume:

• representative cooling load: 110 MW;
• individual water-cooled chiller capacity: 10 MW;
• installed chillers: 11 units.

This 11 × 10 MW assumption is an illustrative simplification. In practice, a 110 MW cooling load would likely be served by 12–16 chillers (including N+1 redundancy), with only 11 operating at full load. That preserves engineering realism while keeping the arithmetic simple for readers.

Each chiller employs a large centrifugal or magnetic-bearing compressor operating under variable load throughout the day.

Unlike cooling towers, chillers do not rely on large outdoor fans to reject heat. Their principal low-frequency sound therefore originates from compressor operation, refrigerant pressure pulsations and structural vibration.

Representative Low-Frequency Sound Power

Published manufacturer data indicate that modern large water-cooled centrifugal chillers typically produce a 63 Hz octave-band sound power level of approximately:

75–88 dB (L_w, 63 Hz octave band)

The range depends on several engineering factors, including:

• compressor technology;
• compressor rotational speed;
• cooling capacity;
• vibration isolation;
• acoustic treatment;
• equipment enclosure.

Assuming 11 identical chillers, the combined sound power is obtained using logarithmic addition:

L_total = L_single + 10 log₁₀(11)

Since:

10 log₁₀(11) ≈ 10.4 dB

the combined installation produces:

Individual Chiller	11 Chillers Combined
75 dB	85.4 dB
88 dB	98.4 dB

For simplicity, we round these values to:

85–98 dB (L_w, 63 Hz octave band)

These values represent the sound emitted by the chillers themselves before any attenuation provided by the plant building.

Because most water-cooled chillers are installed indoors, the sound reaching the site boundary is generally lower than these source levels.

Technology Matters

Chiller acoustics have improved considerably over the past two decades.

Modern designs increasingly incorporate:

• magnetic-bearing centrifugal compressors;
• inverter-driven compressors;
• improved vibration isolation;
• flexible pipe connections;
• acoustic plant room design.

These technologies reduce both compressor vibration and structural transmission while maintaining high energy efficiency.

For water-cooled systems, the plant building itself often provides additional attenuation, making chillers a less significant contributor to environmental low-frequency noise than cooling towers.

MDCO Insight: In water-cooled hyperscale data centres, chillers generate measurable low-frequency sound, but enclosure within dedicated plant buildings usually makes them a secondary contributor compared with cooling towers.

Pumps

Why They Are Important

Pumps circulate chilled water and condenser water throughout the cooling system, enabling heat to be transferred continuously between the data halls, chillers and cooling towers.

For the representative 110 MW cooling load considered in this article, multiple pumps are required to circulate several thousand litres of water every second through the cooling system.

Unlike cooling towers, however, pumps are normally installed within enclosed mechanical plant rooms rather than outdoors.

Consequently, pumps are generally not major sources of airborne low-frequency noise experienced by nearby communities.

Their principal acoustic significance lies instead in the structural vibration they can transmit into connected pipework, floors and building structures if not properly isolated.

Representative Pump Installation

Large hyperscale facilities normally employ separate pump systems for the chilled-water and condenser-water circuits.

For the representative facility considered here, we assume:

• 11 chilled-water pumps; and
• 11 condenser-water pumps,

giving a total of 22 large pumps installed for the representative cooling system.

Each pump is assumed to be driven by a variable-speed electric motor, allowing the flow rate to adjust continuously in response to cooling demand.

Representative Low-Frequency Sound Power

Although pumps generate vibration across a broad frequency range, environmental acoustic assessments commonly evaluate the airborne sound radiated by the pump installation.

Published manufacturer data indicate that a large industrial pump typically produces a 63 Hz octave-band sound power level of approximately:

68–80 dB (L_w, 63 Hz octave band)

The range depends upon:

• pump size;
• rotational speed;
• hydraulic loading;
• motor design;
• vibration isolation.

Assuming 22 identical pumps, logarithmic addition gives:

L_total = L_single + 10 log₁₀(22)

Since:

10 log₁₀(22) ≈ 13.4 dB

the combined installation produces:

Individual Pump	22 Pumps Combined
68 dB	81.4 dB
80 dB	93.4 dB

For simplicity, these values are rounded to:

81–93 dB (L_w, 63 Hz octave band)

These figures represent the airborne sound emitted by the pumps before attenuation by the surrounding plant room.

In practice, the mechanical plant room usually provides substantial acoustic shielding, meaning that the airborne contribution from pumps at the site boundary is generally much smaller than that from cooling towers.

Technology Matters

For pumps, installation quality often has a greater influence on environmental performance than the pump itself.

Modern facilities commonly employ:

• resilient equipment mounts;
• flexible pipe connectors;
• inertia bases;
• vibration-isolated pipe supports;
• variable-speed drives.

These measures significantly reduce the transmission of vibration into the building structure and connected pipework.

Consequently, well-designed pump installations rarely become dominant contributors to community low-frequency noise.

MDCO Insight: Pumps generate relatively modest airborne low-frequency noise, but inadequate vibration isolation can allow structure-borne vibration to become a more significant engineering concern.

Transformers

Why They Are Important

Transformers differ fundamentally from cooling equipment because they contain no rotating machinery.

Instead, they generate sound through magnetostriction—the microscopic expansion and contraction of the laminated steel core as it is repeatedly magnetised and demagnetised by the alternating electrical supply.

For large hyperscale data centres in Malaysia, facilities exceeding approximately 100 MW are typically supplied from the 275 kV transmission network. The principal outdoor transformers are therefore commonly 275/33 kV power transformers located within the site’s electrical substation.

Other transformers within the facility, such as:

• 33/0.4 kV transformers;
• 33/11 kV transformers; and
• 11/0.4 kV transformers,

are generally installed indoors within electrical rooms or buildings and therefore contribute relatively little to community noise.

Although transformer sound power is generally lower than that of large cooling equipment, outdoor power transformers operate continuously whenever the data centre is energised. They therefore contribute a persistent tonal component to the overall acoustic environment.

Typical Low-Frequency Characteristics

Unlike cooling towers, chillers and pumps, transformer noise is not generated by rotating mechanical components.

For countries operating on a 50 Hz electrical supply, including Malaysia, transformer sound is dominated by:

• 100 Hz, caused by the magnetic forces acting on the transformer core;
• higher harmonic frequencies, including 200 Hz, 300 Hz and beyond.

The magnetic forces within the transformer core repeat twice during each 50 Hz electrical cycle, producing the characteristic 100 Hz acoustic hum.

As a result, transformer noise is highly tonal and remarkably stable because it is directly linked to the electrical supply frequency.

Unlike cooling equipment, transformer frequencies do not vary significantly with operating load or weather conditions.

Why the 63 Hz Octave Band Is Used

The actual dominant transformer tone occurs at approximately 100 Hz, rather than 63 Hz.

However, this article uses the 63 Hz octave band as a common reference point for comparing the major low-frequency noise sources within a data centre.

Environmental acoustic assessments generally examine a range of octave bands rather than isolated frequencies because different equipment types generate energy at different parts of the low-frequency spectrum.

The 63 Hz octave band (approximately 45–90 Hz) is particularly useful because it captures significant low-frequency energy from large cooling equipment such as cooling towers and compressors, which dominate the overall LFN profile of a typical data centre.

For transformers, the adjacent frequency bands—particularly around 100 Hz—would normally be examined in a complete acoustic assessment. The values below therefore represent the transformer contribution within the same comparative framework used for the other equipment categories, rather than suggesting that 63 Hz is the dominant transformer frequency.

Representative Transformer Installation

For the representative 100 MW data centre considered in this article, we assume the facility receives high-voltage supply through:

• four 275/33 kV outdoor power transformers.

This configuration represents a typical approach for a large hyperscale facility, where multiple transformers provide operational flexibility, maintenance capability and redundancy.

Because these transformers are installed outdoors, they represent the principal transformer-related noise source considered in this environmental assessment.

Representative Low-Frequency Sound Power

Although the dominant magnetic hum of a transformer operating on a 50 Hz electrical supply occurs at approximately 100 Hz, transformers also emit acoustic energy across neighbouring low-frequency octave bands. For consistency with the comparative methodology adopted throughout this article, we use the 63 Hz octave band as a common reference. Published manufacturer data indicate representative 63 Hz octave-band sound power levels in the range of::

65–78 dB (L_w, 63 Hz octave band)

The variation depends on:

• transformer rating;
• magnetic flux density;
• core construction;
• structural damping;
• low-noise transformer design.

Assuming four identical transformers, the combined sound power is calculated using logarithmic addition:

L_total = L_single + 10 log₁₀(N)

where:

• L_single = sound power of one cooling tower;
• N = number of identical cooling towers.

For four transformers:

10 log₁₀(4) ≈ 6.0 dB

Therefore:

Individual Transformer	Four Transformers Combined
65 dB	71 dB
78 dB	84 dB

This gives a combined transformer contribution of approximately:

71–84 dB (L_w, 63 Hz octave band)

This is significantly lower than the representative cooling tower contribution calculated earlier.

However, transformer noise remains important because it is:

• continuous;
• tonal;
• stable over time.

A persistent tonal sound may attract attention even when its overall sound energy is relatively modest.

Technology Matters

Transformer acoustic performance depends primarily on electromagnetic and structural design rather than cooling technology.

Modern low-noise transformer designs incorporate:

• grain-oriented electrical steel cores;
• step-lap core construction;
• reduced magnetic flux density;
• improved mechanical clamping;
• vibration-isolated foundations;
• acoustic enclosures where required.

Unlike cooling equipment, transformer noise is largely unaffected by:

• outdoor temperature;
• humidity;
• cooling demand;
• fan speed.

Instead, it remains closely linked to the electrical loading and physical design of the transformer.

MDCO Insight: Outdoor power transformers contribute a continuous tonal low-frequency component, but their overall acoustic contribution is generally smaller than the cooling systems required to reject data centre heat.

Equipment Normally Excluded from Continuous Noise Assessment

Not every electrical or mechanical system contributes equally to long-term community exposure.

Several important systems are normally excluded from continuous operational low-frequency noise assessments because they operate only intermittently or under specific conditions.

These include:

• standby diesel generators;
• fire pumps;
• emergency ventilation systems;
• maintenance equipment.

Although these systems may generate substantial noise when operating, they are assessed separately under emergency, testing or maintenance scenarios rather than normal data centre operation.

Similarly, construction activities fall outside the scope of operational low-frequency noise because they are temporary rather than continuous.

Accordingly, the worked example presented in this article focuses on the principal systems operating during normal data centre operation:

• cooling towers;
• chillers;
• pumps; and
• outdoor power transformers.

MDCO Insight: Not all equipment that can produce noise contributes equally to long-term community exposure; operational assessments focus on systems that operate continuously under normal conditions.

Bringing the Equipment Together

The representative equipment calculations above demonstrate an important engineering principle:

A data centre does not have a single characteristic noise source.

Instead, its acoustic environment emerges from the combined contribution of multiple systems operating simultaneously, including:

• cooling towers;
• chillers;
• circulating pumps;
• outdoor electrical transformers; and
• other supporting infrastructure.

However, these sources do not contribute equally.

For a typical water-cooled hyperscale data centre, the largest continuous low-frequency contribution generally comes from cooling towers, because they combine large axial fans with continuous heat rejection to the outdoor environment.

Chillers, pumps and transformers contribute additional low-frequency components, but their influence depends strongly on equipment technology, installation location and acoustic treatment.

This is why environmental acoustic assessments do not begin with sound pressure measurements at the site boundary.

Instead, engineers first determine the sound power level of each significant equipment item. These individual sources are then combined using logarithmic methods to establish the overall sound power emitted by the facility.

Only after the total source strength has been established do engineers calculate how the sound changes as it propagates through the environment, considering factors such as:

• distance;
• atmospheric conditions;
• terrain;
• buildings;
• ground effects; and
• temperature-related refraction.

The following section applies this approach to construct a representative 100 MW data centre model and demonstrates how the overall low-frequency noise profile changes as facility capacity increases towards 500 MW.

Constructing the Acoustic Model of a Representative 100 MW Data Centre

Having estimated the acoustic characteristics of the principal equipment categories, the next step is to combine these individual sources into a representative data centre model.

This follows the same methodology used in professional environmental noise assessments.

Rather than treating a data centre as a single generic sound source, engineers first identify the major noise-producing equipment, estimate the sound power emitted by each category, and then combine these contributions to determine the overall acoustic output of the facility.

For this worked example, the representative facility is based on:

• 100 MW IT load;
• approximately 110 MW cooling load;
• water-cooled cooling architecture;
• multiple cooling towers, chillers, pumps and outdoor transformers operating under full design conditions.

The equipment quantities and sound power values presented below are based on representative engineering assumptions derived from published manufacturer data and industry references.

They are intended to demonstrate the calculation methodology and the relative contribution of different equipment categories, rather than predict the exact acoustic performance of any specific data centre.

MDCO Insight: The acoustic footprint of a data centre emerges from the combined operation of many independent systems, with cooling infrastructure typically forming the dominant contribution.

Estimating the Number of Major Cooling Equipment

Unlike server racks, cooling and electrical equipment are rarely installed as a single large unit.

Instead, capacity is distributed among multiple machines to improve:

• operational resilience;
• maintenance flexibility;
• energy efficiency;
• redundancy.

For the representative 100 MW IT load hyperscale data centre considered in this article, we assume:

• approximately 110 MW cooling load;
• a water-cooled chilled-water cooling architecture;
• multiple cooling and electrical systems operating under full design conditions.

The representative equipment quantities are:

Equipment	Representative Quantity
Cooling towers	22 cells
Water-cooled chillers	11 units
Chilled-water and condenser-water pumps	22 units
Outdoor 275/33 kV transformers	4 units

These values represent an engineering model rather than a universal design standard.

Actual facilities may differ depending on:

• redundancy philosophy (N+1, 2N, etc.);
• cooling architecture;
• equipment capacity;
• site constraints;
• operator requirements.

The purpose of this model is not to predict one specific facility, but to demonstrate how multiple noise sources combine to form the overall low-frequency acoustic environment.

Why Noise Cannot Be Added Like Ordinary Numbers

One of the most common misconceptions is that ten identical cooling towers produce ten times the noise of one cooling tower.

This is not how sound behaves.

Sound energy combines according to logarithmic principles.

When several independent sources operate simultaneously, their acoustic energies are added together before being converted back into decibels.

For identical independent sources:

L_total = L_single + 10 log₁₀(N)

where:

• L_single = sound power of one cooling tower;
• N = number of identical cooling towers.

The practical consequences are:

• doubling the number of identical machines increases the level by approximately 3 dB;
• increasing the number tenfold increases the level by approximately 10 dB.

Therefore, increasing the number of cooling towers significantly increases total acoustic energy, but the increase is much smaller than the increase in equipment quantity.

MDCO Insight: Sound levels increase logarithmically, not proportionally. A fivefold increase in equipment quantity does not create five times the noise level.

Combining the Principal Low-Frequency Sources

The previous sections estimated the representative 63 Hz octave-band sound power generated by each major equipment category.

However, the acoustic contribution experienced by nearby communities depends not only on the sound generated by each machine, but also on whether the equipment is located indoors or outdoors and the attenuation provided by the surrounding structures.

For a typical large water-cooled hyperscale data centre:

• cooling towers are normally installed outdoors and radiate sound directly to the environment;
• chillers and pumps are commonly installed within enclosed mechanical plant buildings;
• outdoor transformers radiate sound continuously but contribute mainly through tonal components around 100 Hz and higher harmonics.

Therefore, the following calculation focuses on the effective outdoor 63 Hz octave-band contribution relevant to environmental low-frequency noise assessment.

Equipment	Quantity	Individual 63 Hz Octave-Band Sound Power	Effective Outdoor Contribution
Cooling towers	22	80–92 dB	93–105 dB
Water-cooled chillers	11	75–88 dB	Reduced by building attenuation; secondary contribution
Pumps	22	68–80 dB	Reduced by building attenuation; generally minor airborne contribution
Outdoor transformers	4	65–78 dB	71–84 dB

For chillers and pumps, the calculated equipment sound power represents the acoustic emission inside the mechanical plant area. The actual contribution reaching the surrounding environment depends on:

• plant room wall construction;
• ventilation openings;
• acoustic louvers;
• equipment mounting;
• vibration isolation.

In modern hyperscale facilities, these factors may provide substantial attenuation. Consequently, chillers and pumps are generally secondary contributors to outdoor low-frequency noise compared with cooling towers.

For the purpose of this illustrative model, we therefore consider:

• cooling towers as the dominant outdoor LFN source;
• outdoor transformers as a continuous but smaller contributor;
• indoor chillers and pumps as secondary contributors after enclosure attenuation.

The facility-level 63 Hz octave-band sound power is then estimated by logarithmically combining the effective outdoor sources:

L_facility = 10 log₁₀(10^L₁/10 + 10^L₂/10 + …)

Lower-bound Scenario

Assume:

• Cooling towers: 93 dB
• Outdoor transformers: 71 dB

The combined level is:

10 log₁₀(10^93/10 + 10^71/10 ) ≈ 93.0 dB

The transformer contribution is much smaller and the combined level is similar to that of the cooling towers.

Upper-bound Scenario

Assume:

• Cooling towers: 105 dB
• Outdoor transformers: 84 dB

The combined level is:

10 log₁₀(10^105/10 + 10^84/10 )

Giving:

≈ 105.0 dB

Therefore, for the representative water-cooled hyperscale facility:

Effective outdoor 63 Hz low-frequency sound power ≈93–105 dB (L_w, 63 Hz octave band)

This result is close to the cooling tower contribution alone, demonstrating an important engineering principle:

Although multiple systems generate low-frequency sound inside a data centre, the environmental acoustic footprint is usually dominated by the equipment that:

1. generates significant low-frequency energy; and
2. radiates directly to the outdoor environment.

For large water-cooled facilities, this is typically the cooling tower system.

MDCO Insight: The equipment that produces the most internal noise is not always the equipment that dominates community impact. Outdoor location and acoustic shielding are often equally important as sound generation itself.

Scaling from 100 MW to Larger Facilities

A larger data centre generally requires more cooling equipment, but the relationship between facility capacity and noise is not linear.

A 500 MW facility does not automatically produce five times the low-frequency sound power of a 100 MW facility.

This is because sound combines logarithmically rather than arithmetically.

When the number of similar sound sources increases, the additional acoustic energy must be combined using:

L_total = L_single + 10 log₁₀(N)

where N represents the change in the number of equivalent sound sources.

Example: Scaling from 100 MW to 150 MW

The representative 100 MW facility produces:

93–105 dB (L_w, 63 Hz octave band)

Assuming the cooling and electrical infrastructure scales proportionally with IT capacity:

150 MW / 100 MW = 1.5

The number of equivalent sound sources increases by 1.5 times.

The resulting increase in sound power is:

10 log₁₀(1.5) ≈ 1.76 dB

Therefore, the estimated facility sound power becomes:

Lower-bound case:

93 + 1.76 = 94.8 dB

Upper-bound case:

105+ 1.76 =106.8 dB

Rounded for practical interpretation:

150 MW facility:

approximately 95–107 dB (L_w, 63 Hz octave band)

The same logarithmic scaling approach is then applied to larger facilities.

IT Capacity	Approximate Cooling Load	Relative Equipment Quantity	Representative 63 Hz LFN Sound Power
100 MW	110 MW	1.0×	93–105 dB
150 MW	165 MW	1.5×	95–107 dB
200 MW	220 MW	2.0×	96–108 dB
300 MW	330 MW	3.0×	98–110 dB
500 MW	550 MW	5.0×	100–112 dB

These values are illustrative engineering estimates based on proportional scaling of equipment quantity.

Actual acoustic performance depends on:

• cooling technology;
• equipment selection;
• fan operating speed;
• acoustic treatment;
• site layout;
• operating strategy.

A larger but acoustically optimised facility may therefore produce lower environmental noise than a smaller facility using older or less efficient equipment.

MDCO Insight: Increasing data centre capacity increases the total acoustic energy, but the logarithmic nature of sound means that noise grows much more slowly than the physical scale of the facility.

From Sound Power to What Communities Hear

Up to this point, the calculations have considered sound power—the acoustic energy emitted by the facility.

Communities do not experience sound power.

They experience sound pressure, which is the actual sound level reaching a listener or measurement location.

The relationship between the two depends on many environmental factors, including:

• distance from the facility;
• atmospheric absorption;
• terrain;
• buildings;
• vegetation;
• wind;
• temperature gradients;
• atmospheric stability.

For low-frequency noise, these effects can become particularly important because long wavelengths allow sound to travel efficiently over large distances and interact strongly with atmospheric conditions.

The next section applies sound propagation calculations to estimate the sound pressure levels experienced at different distances from the representative facility:

• 100 metres;
• 500 metres;
• 1 kilometre;
• 1.5 kilometres;
• 2 kilometres.

The analysis will then examine how Malaysian tropical atmospheric conditions, including temperature stratification and inversion effects, may influence long-distance low-frequency sound propagation.

MDCO Insight: Communities experience sound after it has travelled through the surrounding environment, not the sound emitted directly by the equipment.

Distance Is the Most Important Factor

The most significant factor affecting environmental noise is usually distance.

Up to this point, the calculations have focused on sound power level, which describes the acoustic energy emitted by the data centre equipment.

Communities, however, experience sound pressure level—the sound energy that actually reaches a listener or measurement location.

Converting from sound power to sound pressure requires a propagation model.

For a simplified hemispherical free-field model, representative of a ground-mounted facility over acoustically reflective ground, the sound pressure level may be estimated using:

L_p = L_w -20 log₁₀(r) – 8

where:

• L_p = sound pressure level at distance r;
• L_w = sound power level of the source;
• r = distance from the source in meters.

This relationship represents geometric spreading only. It assumes sound radiates uniformly over a hemisphere from a source located on the ground. Actual environmental conditions may increase or decrease the received sound level due to:

• atmospheric absorption;
• ground effects;
• terrain;
• buildings;
• vegetation;
• wind;
• temperature gradients.

For the representative 100 MW data centre, the calculated low-frequency sound power is:

93–105 dB (L_w, 63 Hz octave band)

Using the free-field propagation relationship, the approximate sound pressure levels are:

Distance from Facility	Calculation	Estimated 63 Hz Sound Pressure Level (Lp)
100 m	93–105	45–57 dB
500 m	93–105	31–43 dB
1,000 m	93–105	25–37 dB
1,500 m	93–105	21–33 dB
2,000 m	93–105	19–31 dB

These values represent the theoretical sound pressure level under simplified free-field conditions.

The results illustrate an important engineering principle:

• The sound level decreases substantially with distance.
• However, because low-frequency sound experiences relatively low atmospheric absorption, it may remain measurable at distances where higher-frequency mechanical noise has already diminished.

Low-frequency sound also has longer wavelengths, allowing it to diffract around obstacles more effectively than higher-frequency sound.

Why Weather Matters

Unlike many environmental pollutants, sound does not travel through a fixed medium.

The atmosphere itself becomes part of the transmission path.

Two measurements taken at the same location on different days may produce noticeably different results even though the data centre equipment is operating identically.

This does not necessarily mean that the facility has become louder.

Instead, changes in atmospheric conditions may alter how efficiently sound energy travels from the source to the receiver.

Several meteorological factors influence environmental noise propagation, including:

• wind speed;
• wind direction;
• temperature;
• atmospheric stability;
• humidity;
• rainfall.

For low-frequency sound, atmospheric stability and wind conditions are generally the most important factors because they influence the direction and efficiency of sound propagation over long distances.

Stable Nighttime Conditions and Temperature Inversion

After sunset, particularly during clear and calm evenings, the ground loses heat through radiation and cools more rapidly than the air above it.

The cooler air remains near the ground while warmer air exists above, creating a stable atmospheric layer, commonly known as a temperature inversion.

Under normal daytime conditions, air temperature generally decreases with increasing height.

This causes sound waves to refract upwards, allowing some acoustic energy to move away from ground-level receivers.

During an inversion, the temperature profile reverses.

The warmer air above causes sound waves to bend downwards towards the ground rather than dispersing upwards.

This phenomenon is particularly relevant for low-frequency sound because the longer wavelengths are less affected by atmospheric absorption and can remain coherent over long distances.

Such stable nighttime conditions occur frequently in tropical climates, including Malaysia, especially during calm and clear periods.

Representative Effect

The magnitude of this effect depends on:

• inversion strength;
• source height;
• receiver height;
• terrain;
• distance.

For long-distance propagation, a moderate nighttime inversion may increase received low-frequency sound levels by approximately:

+3 to +10 dB

compared with neutral atmospheric conditions.

Under particularly favourable conditions, larger increases may occur.

Wind Direction and Propagation

Wind also influences sound propagation.

When wind blows from the data centre towards a residential area, sound waves are refracted downwards in the direction of travel.

This increases the amount of sound energy reaching the receiver.

Conversely, when the wind direction is from the receiver towards the facility, sound waves tend to refract upwards, reducing the received sound level.

Environmental noise assessments therefore commonly consider:

• downwind conditions;
• upwind conditions;
• calm conditions.

Representative Effect

For low-frequency sound over several hundred metres to kilometres:

• favourable downwind conditions may increase received levels by approximately +1 to +5 dB;
• upwind conditions may reduce received levels by a similar order of magnitude.

The exact effect depends on wind speed, atmospheric stability and terrain.

Humidity and Rainfall

Malaysia’s tropical climate is characterised by high humidity throughout the year.

However, humidity has relatively limited influence on low-frequency sound propagation.

Atmospheric absorption increases significantly with frequency.

Therefore:

• high-frequency mechanical noise is more strongly affected by atmospheric absorption;
• low-frequency components experience comparatively little attenuation.

This is one reason why low-frequency sound may remain detectable at distances where higher-frequency mechanical noise has largely disappeared.

Rainfall has a different effect.

Heavy rain generally does not significantly reduce low-frequency propagation.

Instead, it increases the background acoustic environment.

Consequently, residents may perceive less mechanical noise during heavy rainfall because:

• rainfall creates substantial background sound;
• windows may remain closed;
• attention is drawn away from distant mechanical sources.

The facility itself may not have become quieter.

Terrain, Buildings and Ground Effects

Atmospheric conditions are only one part of the propagation pathway.

The surrounding environment also influences received sound levels.

Important factors include:

• facility elevation;
• residential elevation;
• intervening buildings;
• earth bunds;
• acoustic barriers;
• ground surface characteristics.

Large buildings may:

• block direct sound paths;
• reflect sound towards other locations;
• create local amplification effects.

Vegetation generally provides limited attenuation for low-frequency sound because the wavelength is much larger than individual trees and shrubs.

Therefore, a dense tree belt that effectively reduces higher-frequency noise may provide little reduction for low-frequency components.

Combining Environmental Effects

The previous sections estimated the free-field propagation of a representative 100 MW data centre.

The baseline values were:

Distance	Neutral Atmospheric Condition (Lp)
100 m	45–57 dB
500 m	31–43 dB
1,000 m	25–37 dB
1,500 m	21–33 dB
2,000 m	19–31 dB

To illustrate the potential influence of Malaysian atmospheric conditions, three representative scenarios can be considered.

Scenario	Representative Conditions	Correction Applied
Neutral daytime	Well-mixed atmosphere	0 dB
Typical favourable night	Stable atmosphere + light downwind	+5 to +10 dB
Strong favourable propagation	Strong inversion + downwind + reflective environment	+10 to +15 dB

Applying these corrections gives the following illustrative range:

100 MW Data Centre

Distance	Neutral Condition (Lp)	Typical Favourable Night (Lp)	Strong Favourable Condition (Lp)
100 m	45–57 dB	50–67 dB	55–72 dB
500 m	31–43 dB	36–53 dB	41–58 dB
1,000 m	25–37 dB	30–47 dB	35–52 dB
1,500 m	21–33 dB	26–43 dB	31–48 dB
2,000 m	19–31 dB	24–41 dB	29–46 dB

Applying these same corrections for the 300 MW and 500 MW data centres gives the following estimated 63 Hz octave-band sound pressure levels:

300 MW Data Centre

Distance	Neutral Condition (Lp)	Typical Favourable Night (Lp)	Strong Favourable Propagation (Lp)
100 m	50–62 dB	55–72 dB	60–77 dB
500 m	36–48 dB	41–58 dB	46–63 dB
1,000 m	30–42 dB	35–52 dB	40–57 dB
1,500 m	26–38 dB	31–48 dB	36–53 dB
2,000 m	24–36 dB	29–46 dB	34–51 dB

500 MW Data Centre

Distance	Neutral Condition (Lp)	Typical Favourable Night (Lp)	Strong Favourable Propagation (Lp)
100 m	52–64 dB	57–74 dB	62–79 dB
500 m	38–50 dB	43–60 dB	48–65 dB
1,000 m	32–44 dB	37–54 dB	42–59 dB
1,500 m	28–40 dB	33–50 dB	38–55 dB
2,000 m	26–38 dB	31–48 dB	36–53 dB

These values are not predictions for any specific data centre.

They illustrate how the same physical facility can produce different community exposure depending on environmental conditions.

How Do These Levels Compare with Human Perception?

Estimating sound pressure levels is only the first step in understanding low-frequency noise.

The next question is:

Are these levels likely to be noticed by people living near the facility?

Unlike general environmental noise, there is no universally accepted sound pressure threshold at which low-frequency noise becomes noticeable, annoying or harmful to everyone. Human response depends not only on the measured sound level, but also on the frequency content, tonal characteristics, background environment, building construction, time of day and individual sensitivity.

Nevertheless, published research and engineering guidance provide useful reference points for interpreting the calculations presented in this article.

Several recognised assessment methods, including DIN 45680 and the UK DEFRA Procedure for the Assessment of Low Frequency Noise Complaints (NANR45), indicate that 63 Hz octave-band sound pressure levels of approximately 38–40 dB (L_p) under quiet residential conditions are generally where low-frequency noise begins to warrant further investigation. Above this range, the likelihood of perception and annoyance increases, particularly where the sound is continuous, tonal or occurs during the night.

This does not represent a health threshold, nor does it imply that everyone will notice the sound. Rather, it serves as an engineering benchmark for assessing the potential for community impact.

Comparing this benchmark with the representative calculations in this article provides several useful observations.

For the representative 100 MW data centre, predicted sound pressure levels under neutral atmospheric conditions generally fall below 40 dB beyond approximately 500–1,000 metres. However, under favourable nighttime propagation conditions—when stable atmospheric stratification and downwind conditions may increase sound levels by 5–15 dB—the predicted sound pressure may exceed 40 dB at distances approaching 2 kilometres.

For the representative 300 MW and 500 MW facilities, the potential area exceeding this indicative assessment level becomes progressively larger. Under strong favourable propagation conditions, representative calculations suggest that sound pressure levels exceeding 40 dB may remain possible at distances extending beyond 2 kilometres, depending on equipment design, site layout and local meteorological conditions.

These calculations do not imply that everyone living within these distances will experience annoyance or sleep disturbance. Indoor sound levels vary with building construction, windows, ventilation systems and background environmental noise, while individual sensitivity differs considerably between people.

However, the calculations do illustrate why some residents may report experiences such as:

• persistent humming;
• distant mechanical droning;
• vibration sensations;
• pressure or fullness in the ears; and
• difficulty sleeping during quiet nighttime periods.

These experiences have been reported in numerous international studies of environmental low-frequency noise and are broadly consistent with the principal outcomes identified in current research (→A09.03 Low Frequency Noise Effects: What Research Shows), namely annoyance and sleep disturbance. By contrast, evidence linking low-frequency noise directly to broader physiological health effects remains an active area of scientific investigation.

Interestingly, the representative propagation distances derived in this article are also broadly consistent with the distances reported in several recent community disputes and legal proceedings involving hyperscale data centres in the United States (→A09.01 Why Low-Frequency Noise Becomes a Community Conflict). Although each site differs in its cooling technology, terrain, meteorology and operating conditions, the engineering calculations demonstrate that low-frequency sound can remain within the range associated with increased perception or annoyance well beyond the immediate site boundary under favourable atmospheric conditions. This engineering observation is consistent with, but does not by itself verify, the experiences reported in those cases.

From a planning perspective, the calculations highlight an important consideration. While data centres occupy relatively compact sites, the potential area influenced by low-frequency sound may extend across several square kilometres under certain operating and meteorological conditions. Consequently, appropriate acoustic design, equipment selection, site layout, environmental assessment, long-term monitoring and community engagement all become important components of responsible data centre development.

MDCO Insight: Low-frequency noise is not solely a site-level issue; its potential to affect surrounding communities warrants careful engineering assessment, informed planning and ongoing monitoring.

The Observatory Perspective

Low-frequency noise illustrates how engineering, environmental science and community experience intersect in data centre development.

The engineering chain is straightforward:

Data processing generates heat. Heat requires cooling. Cooling equipment produces sound. The atmosphere influences how that sound propagates. Communities experience the final result.

The calculations presented in this article demonstrate that low-frequency noise is neither arbitrary nor impossible to estimate. Using established engineering principles, it is possible to approximate the sound produced by representative facilities, understand how it changes with distance and weather, and identify the conditions under which nearby communities may be more likely to perceive it.

The analysis also highlights that low-frequency noise should not be viewed simply as an operational issue within the site boundary. Under favourable atmospheric conditions, its potential influence may extend well beyond the immediate development, making it relevant to site selection, acoustic design, environmental assessment, operational management and long-term community engagement.

As Malaysia continues to develop larger and more concentrated data centre clusters, understanding low-frequency noise through evidence-based engineering provides a stronger foundation for constructive dialogue between developers, regulators, researchers and surrounding communities.

MDCO Insight: Understanding how low-frequency noise is generated, propagated and perceived provides a stronger foundation for responsible data centre planning and informed public discussion.

Limitations of This Analysis

This article presents an illustrative engineering analysis, not a site-specific environmental noise assessment.

The calculations are based on representative assumptions regarding facility capacity, cooling technology, equipment characteristics and sound propagation. Actual sound levels will vary depending on the detailed design of each facility, including equipment selection, operating strategy, site layout, terrain, building configuration and prevailing meteorological conditions.

For simplicity, the analysis focuses on the 63 Hz octave band, which provides a practical reference for comparing the principal sources of low-frequency noise from large data centres. A comprehensive acoustic assessment would normally examine multiple octave bands, tonal characteristics and site-specific propagation using recognised standards such as ISO 9613-2 together with field measurements where appropriate.

Accordingly, the values presented in this article should be interpreted as representative engineering estimates intended to explain the methodology and illustrate the relative influence of facility size, distance and atmospheric conditions, rather than predict the acoustic performance of any particular project.

MDCO Insight: Representative engineering models help explain how low-frequency noise behaves, but site-specific assessment remains essential for evaluating actual community exposure.

Selected References

• International Organization for Standardization (ISO). ISO 9613-2: Acoustics — Attenuation of Sound During Propagation Outdoors — Part 2: General Method of Calculation. Widely adopted engineering methodology for predicting outdoor sound propagation, including distance attenuation, atmospheric absorption and meteorological effects. https://www.iso.org/standard/74047.html
• International Organization for Standardization (ISO). ISO 3744: Acoustics — Determination of Sound Power Levels of Noise Sources Using Sound Pressure. International standard for determining the sound power levels of industrial equipment and mechanical systems. https://www.iso.org/standard/80866.html
• International Organization for Standardization (ISO). ISO 1996-2: Acoustics — Description, Measurement and Assessment of Environmental Noise — Part 2: Determination of Environmental Sound Pressure Levels. Provides guidance on environmental noise measurement, including frequency analysis and assessment of sound levels at receiving locations. https://www.iso.org/standard/59766.html
• Deutsches Institut für Normung (DIN). DIN 45680: Measurement and Assessment of Low-Frequency Noise Immissions. A widely referenced methodology for evaluating low-frequency noise, including frequency-band analysis and assessment of low-frequency components. https://www.din.de/en
• ASHRAE. ASHRAE Handbook — HVAC Applications. Provides engineering guidance on HVAC system design, cooling technologies, acoustics and noise control principles relevant to large-scale facilities, including data centres. https://www.ashrae.org/technical-resources/ashrae-handbook
• Cooling Technology Institute (CTI). ATC-128: Certification Standard for Measurement of Sound from Water-Cooling Towers. Provides recognised procedures for measuring and rating cooling tower acoustic performance. https://www.cti.org
• World Health Organization (WHO). Environmental Noise Guidelines for the European Region. Reviews scientific evidence relating environmental noise exposure to health outcomes and provides evidence-based recommendations for noise management, particularly regarding annoyance and sleep disturbance. https://www.who.int/europe/publications/i/item/9789289053563
• Moorhouse, A., Waddington, D., & Adams, M. (2005). Procedure for the Assessment of Low Frequency Noise Complaints. Prepared for the UK Department for Environment, Food and Rural Affairs (DEFRA). Provides practical guidance on measuring and assessing low-frequency noise complaints, including frequency analysis and subjective response considerations. https://www.gov.uk/government/organisations/department-for-environment-food-rural-affairs
• American National Standards Institute (ANSI). ANSI/ASA S12.9 Part 4: Noise Assessment and Prediction of Long-Term Community Response. Provides internationally recognised methods for relating environmental noise exposure to community response. https://webstore.ansi.org
• Chartered Institution of Building Services Engineers (CIBSE). CIBSE Guide B: Heating, Ventilating, Air Conditioning and Refrigeration. Provides engineering guidance on HVAC system design, plant selection and acoustic considerations for large buildings. https://www.cibse.org
• ASHRAE Technical Committee 9.9. Thermal Guidelines for Data Processing Environments. Provides guidance on thermal management, cooling technologies and heat rejection in data centres. https://www.ashrae.org/technical-resources/bookstore/datacom-series

Citation

Malaysia Data Centre Observatory (MDCO). A09.02 Estimating Low-Frequency Noise from a Typical Data Centre. MDCO Analyse Series.

MDCO Note

This article forms part of the Malaysia Data Centre Observatory (MDCO) Analyse Series, which aims to improve public understanding of data centre development through evidence-based, accessible and balanced analysis. It is intended for educational and informational purposes only and does not constitute legal, engineering, acoustic, environmental, medical or professional advice.

Low-frequency noise is a multidisciplinary subject spanning engineering, acoustics, environmental science, medicine, psychology, planning and public policy. MDCO does not advocate for or against any particular stakeholder, project, technology or regulatory position. Its role is to facilitate transparency, reduce information asymmetry and support informed discussion by explaining complex issues through publicly verifiable information and multiple complementary perspectives.