Why Don't Our Two-Way Radios Work Inside the Data Hall, and What Does a Data Centre DAS Need?

Two-way radios fail inside a data hall for a different reason than they fail in a basement carpark. It is not the concrete — it is the metal. Rows of populated server racks, aisle containment panels and welded mesh colocation cages block signal horizontally across the hall, so a radio that works fine in the aisle you are standing in can be unusable two rows away. A data centre DAS therefore has to be laid out per aisle rather than by coverage radius, and it has to survive an elevated noise floor from the UPS, busway and CRAH drives on the uplink path.

Why is a data hall worse than a normal office floor?

In an office, the dominant loss is vertical — reinforced-concrete slabs at roughly 15 to 20 dB per floor. A data hall is usually one large volume with 4.5 to 6 m slab-to-slab height, so the vertical problem largely disappears. What replaces it is horizontal clutter no office floor has.

A populated 42U or 48U rack is a perforated steel box packed with metal chassis, and a row of them is close to a wall as far as 450 MHz is concerned. In the halls I have surveyed, each populated rack row costs somewhere in the region of 10 to 15 dB at UHF — measure it rather than assume it, because it varies enormously with fill ratio and door type. Two rows between the antenna and the technician gives away 20 to 30 dB, which is the whole design margin.

Aisle containment sharpens this, and in one respect helps. A contained cold aisle is typically 1200 mm wide with end doors and a ceiling lid. At 450 MHz the wavelength is about 667 mm, so a 1200 mm aisle sits comfortably above waveguide cutoff — signal propagates well along the aisle and very badly across the rows. The practical consequence: an antenna inside or at the end of a contained aisle covers that aisle properly, and an antenna above the containment lid does not cover it at all.

What about mesh cages in colocation halls?

Welded mesh cage partitions are the most commonly missed dead zone in a colocation build. Typical cage mesh has apertures around 50 mm. For an aperture to leak RF meaningfully it needs to be a reasonable fraction of a wavelength — roughly one tenth, so about 67 mm at 450 MHz. A 50 mm mesh is below that, so electrically the cage is a Faraday cage and a customer's engineer locked inside it has no radio.

No antenna placement outside the cage fixes this. Either you put an antenna inside each cage or you formally exclude cages from the coverage scope in writing. Deciding that at handover rather than at design stage is how acceptance tests fail.

The numbers worth designing to

• Design threshold for reliable DMR voice: −95 dBm at the portable

• Fade margin: 10 dB commercial, 15 dB mission-critical — design a Tier III/IV facility at 15 dB, so deliver −80 dBm

• Populated rack row loss at 450 MHz: ~10–15 dB per row, verify by measurement

• Welded mesh cage: effectively opaque below ~67 mm aperture at 450 MHz

• Typical contained cold aisle: 1200 mm wide, above waveguide cutoff at UHF

• Portable 4 W (36 dBm) against repeater 40 W (46 dBm) — a 10 dB asymmetry that makes uplink the limiting path

Why is the uplink usually the limiting path here?

Every server power supply, every UPS inverter and every variable-frequency drive on the CRAH fans contributes broadband noise. In a data hall the noise floor at a ceiling antenna is routinely several dB above thermal, and that penalty lands entirely on the uplink — the 4 W portable trying to reach the system, not the 40 W repeater trying to reach the technician.

This is why a downlink-only prediction passes and the site still fails on handover. The uplink budget has to be computed separately and the design validated against whichever path is worse. It is the same mechanism as the LED and VFD interference problem that degrades radios in malls and carparks, concentrated into one room.

What does the Singapore regulatory position require?

The repeater is a licensed radio station: it needs an IMDA station licence and type-approved equipment, with the exact model and power named in the equipment schedule. Building-side obligations sit under IMDA's COPIF for info-communication facilities and space, and the SCDF Fire Code governs coverage for emergency responders. None of these is satisfied by a coverage prediction — they are satisfied by a measured grid survey at handover.

This matters more than it used to. Singapore's data centre capacity is now allocated selectively — DC-CFA2 opened on 1 December 2025 for at least 200 MW at a 1.25 PUE requirement and closed on 31 March 2026 — so new halls are being fitted out under compressed programmes where an RF survey is easy to defer and expensive to retrofit once the white space is live.

Related questions

Can a data centre share the building's telco DAS for two-way radio? Rarely without modification. Telco DAS is built for cellular bands, and its couplers and antennas may not be specified down to 400–470 MHz. Verify band support component by component before assuming reuse.

When should the survey be done — shell, fit-out, or live? Survey at shell and re-test at fit-out. Rack fill changes the loss picture by 20 dB or more, so a shell-stage survey alone over-predicts coverage on a hall that is later filled.

Do battery rooms and gas-suppression rooms need their own antennas? Usually yes — they are enclosed, often metal-lined, and exactly where a lone technician needs to be reachable. Treat them as named coverage points, not incidental spill from the main hall.

How Suneast approaches this

Suneast designs and installs in-building two-way radio coverage systems for Singapore buildings, including data centres. We measure rack-row and cage attenuation on site rather than assuming it, compute both downlink and uplink budgets, and state which path limits the design. Antennas are planned per aisle and per cage, excluded areas are named explicitly in the design document, and coverage is proven at handover by measured grid survey rather than prediction. Our explainer on what a distributed antenna system is covers the fundamentals.