Notes on infrasonic data center noise

A lot of the concern over health effects of data centers involves infrasound (0-20 Hz) and low frequency noise (10-250 Hz); see Infrasound in Biology and Medicine: Insights into Mechanisms, Health Outcomes and Research Perspectives - A Narrative Review¹.

Unfortunately, Minnesota’s noise pollution regulations seem to specify that only audible sound needs to be limited. The Minnesota Pollution Control Agency (MPCA) document A Guide to Noise Control in Minnesota² states that for residential housing areas (NAC 1), the legal limit for noise is 50 dB(A) at night time. This value is taken from the Minnesota Administrative Rules, chapter 7030.³

The Alternative Urban Areawide Review (AUAR) for the Hermantown data center project⁴ states:

The Hermantown Zoning Code regulates mechanical noise associated with building operation by the standards set by the MPCA. All future development will be required to comply with these requirements.

The unit dB(A) in the MPCA document refers to A-weighted decibels. A-weighting is the process of adjusting measured values to account for the sensitivity of the human ear. If you look at the graph on page 9 of the MPCA document, you can see that A-weighting reduces the measured sound level significantly below 100 Hz.

Here’s the same graph from Wikipedia:

You can see that below 50 Hz (generally agreed on as the lower limit of human hearing), noise is recorded as 30 dB or more below its true level. This is a problem when it comes to low frequency and infrasonic data center noise pollution.

According to Review of evaluation criteria for infrasound and low frequency noise in the general environment⁵, multiple cited studies have shown that this A-weighted curve is unsuitable for evaluating effects of low frequency noise (LFN).

One cited study is Low Frequency Noise and Annoyance⁶, which states:

Conventional methods of assessing annoyance, typically based on A-weighted equivalent level, are inadequate for low frequency noise and lead to incorrect decisions by regulatory authorities.

That study mentions the World Health Organization Guidelines for Community Noise ⁷, the result of an expert task force. A quote from that document:

When prominent low-frequency components are present, noise measures based on A-weighting are inappropriate. The difference between dB(C) and dB(A) will give crude information about the presence of low-frequency components in noise, but if the difference is more than 10 dB, it is recommended that a frequency analysis of the noise be performed. It should be noted that a large proportion of low-frequency components in noise may increase considerably the adverse effects on health.

Also, according to the WHO document, 50-55 dB(A) outdoors in a living area constitutes a “moderate annoyance” or “serious annoyance”.

Another study is Low frequency noise enhances cortisol among noise sensitive subjects during work performance (Waye, Bengtsson, Rylander, Hucklebridge, Evans and Clow, 2002)⁸. In this study, subjects were exposed to just over 2 hours of LFN at 40 dB(A) in an office-like situation. Subjects susceptible to noise were found to have disrupted circadian cortisol levels (which might lead to disrupted sleep), higher reported stress levels, and lower task performance. The data center noise will of course be continuous, rather than for two hours at a time.

The international standard for how to measure infrasound, ISO 7196:1995, specifies a G-weighted curve rather than A-weighted. The G curve includes measurement of frequencies between 0.2 and 20 Hz, and doesn’t reduce measured values by 30 dB until the frequency is as low as 2 Hz.

My overall conclusion is that the proposed sound limits for the data center are based on inappropriate measurement methods. It’s entirely plausible, given available experimental results, that infrasound from the data center could meet proposed dB(A) limits, but still be loud enough at low frequencies to result in disrupted sleep, increased stress, and other negative health effects in the surrounding neighborhood.

I strongly suggest that the project should be held to a contractual requirement to hold sound levels below 50-55 dB(C), not dB(A), in accordance with WHO guidelines.

Ideally, Minnesota administrative rules would also be updated to mandate appropriate measurement of low frequency noise and infrasound.

For a less formal look at the problem, musician and science enthusiast Benn Jordan has been carrying out investigations into data center noise. As part of his investigation into infrasound he tested its effect on groups of (consenting) people — without telling them if they were in the exposed group, of course — and found that it seemed to lead to tingling sensations, nausea, dizziness, anxiety, eye irritation, and other negative health effects.