RF attenuation through building materials

Why there is no single loss value

Wall-loss tables usually give one number per material. The measurements behind them do not agree on one, and the disagreement comes from what was tested, not from measurement noise. Each point below is computed from this dataset.

• Thickness. Plain concrete at 900 MHz lost 10.5–13.6 dB at 102 mm, 21.6–27.6 dB at 203 mm, and 32.9–45.2 dB at 305 mm, across eight mixes. At 5.5 GHz the dataset's concrete spans 15.1–58.6 dB over 40–203 mm.
• A panel is not a wall. At 6.5 GHz, single drywall panels measured 0–1.5 dB, while in-place partitions measured 2.1–3.0 dB. A built partition has two faces, a cavity, and a frame; its loss is its own, not the sheet's.
• Coatings change the answer. Clear and ordinary glass measured 0.5–7.8 dB at 6.5 GHz. Low-E and coated glass walls and windows measured 15.4–38.4 dB. One Low-E window lost 29.7 dB with co-polarised antennas and 15.4 dB with cross-polarised ones at 6.75 GHz.
• Where you look. One study measured the same stack of cinder blocks twice: 3.9 dB with the horns on the hollow part and 11.2 dB on the solid part (fitted values at 5.5 GHz).
• Moisture. A 13 mm plywood panel lost <0.1 dB dry and 2.5 dB soaked at 5 GHz. Soaked specimens usually, though not always, lost more than their dry counterparts.
• Test method and band. Even inside one report, bands can disagree: NIST's 89 mm brick wall measured 5.4 dB at 2 GHz and 15.1 dB at 3 GHz, where NIST switched to different antennas for the second band.

Use a range to decide what to measure and how much margin to carry, not to replace measuring. The guide to indoor path loss and wall attenuation covers how SiteRF treats walls explicitly rather than hiding them in one coefficient.

How to read a result

• Loss is the signal power without the material in the path minus the power through it, in dB, at normal incidence. Higher is worse for coverage.
• Panel is one sheet or slab of material. Assembly is a built wall, door, window, or stack of units with its frame, studs, cores, or coatings. Compare like with like.
• Evidence is one of four kinds: measured values; a fit to measurements (reported only inside the frequencies it was measured at); a standard's formula; or a physics model. Models are listed apart from measurements.
• Calculated estimates fill a frequency nothing was measured at. Interpolated means a specimen's own measurements on either side of your frequency were joined (linear in log-frequency). Scaled means one nearby measurement was moved to your frequency by the change a physics model predicts at that thickness. Each estimate carries a confidence tier (high, moderate, low, very low) and its distance from the measurement, and says why. Built assemblies, bands that disagree beyond the physics, and long distances all lower the tier. You set how far from a measurement to accept with the tolerance control, or max_distance_pct in the API; estimates are never mixed into the measured range.
• Nearby measurements are within 35% of your frequency but not at it. They are shown at the frequency they were measured, never extrapolated. A single-frequency measurement counts as in band within 15%.
• A range is not an uncertainty interval. It is the minimum and maximum across different specimens, with the count beside it.

Where the numbers come from

9 sources, covering 500 MHz to 100 GHz. Each record in the explorer links to its source and says which table it came from. Values were read against the original documents, and each source's file is pinned by hash so a silently revised document cannot change a published number.

• Electromagnetic Signal Attenuation in Construction Materials (NISTIR 6055)William C. Stone, 1997 · Fit to measurements · 500 MHz to 8 GHzLaboratory transmission through specimens between horn antennas 2 m apart with a metal multipath shield; spread-spectrum swept CW at 2 MHz steps, time-gated, averaged over ten positions. Reported as polynomial fits of received signal relative to free space.Known gaps: Two bands only: 0.5-2.0 GHz and 3.0-8.0 GHz, so 2.4 GHz falls between them. The instrument's stated dynamic range is 80 dB.
• 4-40 GHz Transmission Measurement of Indoor Building Materials at Normal IncidenceJon W. Wallace, Rashid Mehmood, Michael A. Jensen, 2018 · Fit to measurements · 4 GHz to 40 GHzNormal-incidence transmission in a compact anechoic chamber, double-ridged horns 50 cm apart, vector network analyzer. Reported as least-squares straight-line fits of transmission (dB) against frequency (GHz).Known gaps: Straight-line fits over 4-40 GHz smooth over the periodic behaviour the authors describe for glass, wood and drywall. Samples are small (at least 40 cm x 40 cm) panels, not installed walls.
• Wideband Penetration Loss through Building Materials and Partitions at 6.75 GHz in FR1(C) and 16.95 GHz in the FR3 Upper Mid-band spectrumDipankar Shakya, Mingjun Ying, Theodore S. Rappaport, Hitesh Poddar, Peijie Ma, Yanbo Wang, Idris Al-Wazani, 2024 · Measured values · 6.75 GHz to 16.95 GHz1 GHz-wideband sliding-correlation channel sounder, horn antennas on boresight (normal incidence), first-arriving multipath only. Co- and cross-polarised. Panels measured in a lab; walls and doors measured in place in a university building.Known gaps: Two frequencies. The paper's text and Table II disagree on the clear-glass thickness (0.5 cm in the text, 1 cm in the table).
• Experimental Analysis and Modeling of Penetration Loss for Building Materials in FR1 and FR3 bandsEnrui Liu, Pan Tang, Tao Jiang, Jianhua Zhang, 2024 · Fit to measurements · 4.25 GHz to 15.75 GHzVector-network-analyzer penetration measurement with horn antennas at 2.4 m, 1 GHz steps from 4.25 to 15.75 GHz. Reported as linear fits PL = k*f + b (f in GHz) per specimen.Known gaps: Thin boards and a 4 cm concrete slab, not building-scale walls. The authors note glass does not follow a linear trend.
• Attenuation of Several Common Building Materials in Millimeter-Wave Frequency Bands: 28, 73 and 91 GHzNozhan Hosseini, Mahfuza Khatun, Changyu Guo, Kairui Du, Ozgur Ozdemir, David W. Matolak, Ismail Guvenc, Hani Mehrpouyan, 2020 · Measured values · 28 GHz to 91 GHzDirectional antennas, narrowband and wideband, multiple locations averaged. Reported as mean specific attenuation in dB/cm; the loss here is that rate multiplied by the specimen thickness.Known gaps: Millimetre-wave only. The paper's text gives 5.05 dB/cm for plywood at 28 GHz where its Table III gives 5.09; the table value is used.
• Characterization of Ultra Wideband Communication Channels (Ph.D. dissertation, Virginia Tech)Ali Hussein Muqaibel, 2003 · Fit to measurements · 500 MHz to 14.6 GHzUltra-wideband free-space transmission measurements of installed-style samples and walls built on a moving platform. Table 4.3 gives each sample's insertion loss at 5 GHz and a straight-line fit over its measured frequency range; a fit is used here only where it reproduces the 5 GHz value within 0.15 dB, otherwise the 5 GHz point is used.Known gaps: Straight-line fits smooth over the ripple of thin panels. Two rows are entered as 5 GHz points only because their printed fit is unusable (glass, structural wood). Brick and concrete block are measurable only over 1-7 and 2-6.8 GHz because of their loss.
• Study on channel model for frequencies from 0.5 to 100 GHz (TR 38.901), clause 7.4.3.1, Table 7.4.3-1: Material penetration losses3GPP · Standard's formula · 500 MHz to 100 GHzStandardised link-level model: penetration loss L = a + b*f with f in GHz, for a stated reference thickness. A modelling convention for system simulation, not a measurement.Known gaps: The IRR-glass row was revised in Release 19 (earlier papers quote 23 + 0.3f). The papers in this dataset that compare against it report large mismatches for concrete (on thinner or interior specimens, so not like-for-like) and for IRR glass.
• Recommendation ITU-R P.2040-4 (09/2025): Effects of building materials and structures on radiowave propagation in the range of 1 MHz to 450 GHzInternational Telecommunication Union, Radiocommunication Sector, 2025 · Physics model · 1 GHz to 100 GHzTable 3 gives relative permittivity a*f^b and conductivity c*f^d per material class. SiteRF computes the normal-incidence power transmission of one homogeneous slab of the requested thickness, including both surface reflections and internal multiple reflections.Known gaps: A homogeneous slab: no studs, rebar, voids, coatings, moisture gradient, or oblique incidence. Computed at the thickness you give, so it is only as good as that thickness and the material class.
• Reflection and Transmission Properties of Common Construction Materials at 2.4GHz FrequencyTarmo Koppel, Andrejs Šiškins, Heldur Haldre, Nikolajs Toropovs, Inese Vilcāne, Piia Tint, 2017 · Measured values · 2.4 GHzFree-space reflection and transmission at 2.4 GHz (0 dBm) with three standard-gain horn antennas about 1 m from the sample at a 30 degree incident angle; 300 x 300 mm samples conditioned outdoors for several days. Transmission and reflection are reported as electric-field ratios against empty-holder and metal-plate references.Known gaps: The only source at exactly 2.4 GHz, but one value per sample with no repeats, at oblique incidence, on small samples near the horns; the authors report that sample size and shape matter (a round plate lost 3.0 dB where the same square plate lost 9.4 dB). Its 16 mm gypsum board (6.9 dB) is far above other sources for drywall.

What is missing

• 2.4 GHz. Measured coverage is thin and mixed. Ultra-wideband straight-line fits (Muqaibel, 2003) cover drywall, plywood, wooden doors, foam, brick, concrete block, and cloth partitions. Fifteen small panels were measured at exactly 2.4 GHz (Koppel et al., 2017), but at 30 degrees on 300 mm samples, and the same concrete plate lost 3.0 dB round and 9.4 dB square. NIST's bands stop at 2.0 GHz and restart at 3.0 GHz. Everything else is calculated, with its confidence shown.
• Thin sourcing. Only one source covers reinforced concrete, aerated / lightweight concrete, brick-faced concrete or block, low-e / coated glass, glass whiteboard, office cloth partition, metal (steel door), steel rebar mesh (no concrete). Treat a narrow range as a small sample, not as agreement.
• Not covered: oblique incidence, several walls in a row, reflections, and whole-building entry loss (see ITU-R P.2109 and P.2346 for that).
• Vendor rule-of-thumb tables (a dB figure per material) are not included. They rarely state a method, thickness, or moisture, so they cannot be cited the way a measurement can.

How the dataset is built

NIST's 1997 report publishes polynomial fits for each specimen. A script extracts them from the report's PDF, then keeps a curve only if it passes consistency checks: it must agree with its own plus and minus one standard deviation curves, agree with the report's separately printed amplitude fit, and stay inside the instrument's 80 dB dynamic range. 99 of 115 curves passed; the rest are rejected with their reasons recorded, never silently adjusted. Values from the other papers are transcribed by hand with a table locator and checked against the rendered pages. The ultra-wideband fits from Muqaibel's Table 4.3 are accepted only where they reproduce the table's own 5 GHz value. Estimates are calculated at request time from the committed records. Nothing is fetched at request time: the API serves one versioned file (dataset 2026-10-20).

For a clear-path baseline, use the free-space path loss calculator; to combine a wall loss with power and gain, use the RF link-budget calculator.