Subterranean Telemetry: The Radio Model Says Your Meter Is Covered. The Basement Disagrees. [2026]
The water meter is in a concrete pit under a sidewalk. The gas meter is in a basement two floors below grade. The pressure sensor is on a pipe in a utility tunnel that runs under half a campus. The coverage map says all three are inside the cell. The deployment engineer drives out, powers up the device, and gets nothing.
This is the single most common failure in LPWAN deployment, and it is not a hardware problem. It is a modeling problem. The radio planning tools that say “covered” are running path-loss equations that were fitted to outdoor and shallow-indoor measurements. Below grade, the physics changes, the equations do not, and the gap between the two is where the deployment budget goes to die.
What NB-IoT Actually Promises
Narrowband IoT was standardized by 3GPP in Release 13 specifically to solve the deep-coverage problem that LTE could not. The design choices are deliberate and they work. The technology achieves a Maximum Coupling Loss of 164 dB, which translates to a 20 dB coverage improvement over traditional cellular — enough to penetrate thick concrete, metal, and soil, connecting devices from basements, tunnels, sewage networks, and remote rural areas where standard cellular struggles. [Trafalgar Wireless, 2026]
Three mechanisms do the work. The narrow 180 kHz bandwidth concentrates transmission power into a dense signal. Signal repetition retransmits small data packets multiple times, increasing successful reception for devices in tough locations. [Trafalgar Wireless, 2026] And the deployment mode is flexible: standalone mode uses dedicated spectrum, guard-band mode leverages unused space within LTE carriers, and in-band mode operates within normal LTE carriers — the most common approach at 57.6% of the market because it maximizes existing infrastructure. [Norvi, Feb 2026]
The market believes the promise. Narrowband IoT was a $13.62 billion market in 2026, up from $10.43 billion in 2025, with projections of $51.82 billion by 2031 — a 30.62% CAGR. [Trafalgar Wireless, 2026] Water utilities are the tip of the spear: one utility reduced costs by 95 cents per meter after automation, saving $181,000 monthly. [Norvi, Feb 2026]
None of this is wrong. The technology reaches places nothing else reaches. The problem is what happens between “can reach” and “will reach at this specific meter pit,” and that problem lives in the planning model.
Where the Models Break
Radio coverage planning runs on empirical path-loss models — Okumura-Hata, COST-231, the 3GPP TR 38.901 family. They were fitted to measurement campaigns conducted outdoors and at shallow indoor depth. They are good at what they were fitted to. Below grade, they are not.
A research team at the Technical University of Denmark ran a measurement campaign through the underground tunnels and basements spanning their entire campus, collecting NB-IoT received-signal-strength samples at ground level, first floor, and underground levels -1 and -2. [arXiv 2006.00880] Their framing of the problem is direct: existing outdoor-to-indoor path-loss models lack accuracy in underground situations, so IoT coverage planning in those areas cannot rely on robust tools and becomes a process of trial and error. [ResearchGate, 2019]
The headline result: the existing models produce prediction errors of 2 to 12 dB in underground tunnels and deep-indoor environments. [ZYIoT, Mar 2026] Twelve decibels is not a rounding error. In link-budget terms it is the difference between a meter that reports every hour for ten years on one battery and a meter that burns through its power budget on retransmissions and goes dark in eighteen months.
The finding that matters more than the error magnitude is why the models miss. The DTU team observed that received power does not decrease with the transmitter-receiver separation distance. [arXiv 2006.00880] Every empirical path-loss model in use has distance as its primary independent variable. Underground, distance stops predicting signal.
What Actually Predicts Coverage Underground
When distance failed as a predictor, the DTU team derived new parameters — indoor depth, indoor distance, and average distance to the closest corridor — and tested their significance to signal attenuation. [arXiv 2006.00880] The result inverts the textbook: indoor distance and penetration depth do not explain the signal attenuation well and actually increase the error of the prediction, while average distance to the nearest corridor emerged as a promising feature. [ResearchGate, 2019]
The physical explanation is that a corridor or tunnel is a waveguide. Radio energy that enters it propagates along it with far less loss than the free-space or through-wall models predict, because the walls are reflecting rather than absorbing. A meter at the end of a long straight utility tunnel can see a stronger signal than a meter ten meters closer to the base station but sitting in a closed concrete room. The geometry of the space is the variable. Depth and distance are noise.
| Variable | What the Standard Models Assume | What the Tunnel Measurements Show |
|---|---|---|
| TX–RX distance | Primary predictor; loss increases log-linearly with distance | Received power does not decrease with distance below grade |
| Penetration depth | Fixed dB penalty per floor or per meter of soil | Adding depth to the model increases prediction error |
| Indoor distance | Additional loss per meter traveled inside the structure | Does not explain attenuation; increases error |
| Distance to nearest corridor | Not a parameter in any standard model | The most promising predictor of signal strength underground |
Read that last row again. The variable that best predicts whether your meter will connect is one that the industry-standard planning tools do not have a field for.
What This Means at the Meter Pit
The practical guidance from deployment vendors already reflects this, even if they do not cite the physics. Even on cellular networks, not every meter location has adequate signal; the recommendation is to test a representative sample of sites for RSSI and SNR before full rollout. [Black Ant, Jul 2026] That is the vendor’s polite way of saying the coverage map cannot be trusted below grade and you need to go measure.
Four things follow for anyone planning an underground deployment:
Sample by geometry, not by distance. A representative sample is not “every 500 meters.” It is one measurement per distinct spatial condition: closed room, room adjacent to corridor, inside corridor, corridor junction, pit under open sidewalk, pit under structure. If the DTU finding holds, two meters in identical rooms 400 meters apart will look more alike than two meters 10 meters apart where one is in a corridor.
Budget the repetitions. NB-IoT’s signal repetition is what rescues marginal links, but every repetition costs battery. A device that the model says is at 144 dB coupling loss and is actually at 156 dB will connect — and will do it by repeating transmissions enough times to drain a ten-year battery in a fraction of that. The 2–12 dB model error is a battery-life error before it is a connectivity error.
Plan for the device to outlive your carrier contract. Devices stored in the field for over 10 years need eSIM/eUICC to reduce carrier lock-in and safely perform firmware updates over the air. [ZYIoT, Mar 2026] A meter in a pit is not getting a truck roll for a SIM swap. The provisioning decision is a ten-year decision made on day one.
Know what the newer releases give you. 3GPP Release 18 adds enhanced positioning capabilities, which in mine and confined-space applications means rescuers can confirm personnel locations through base station positioning even dozens of meters underground. [ZYIoT, Mar 2026] That is a coverage capability that did not exist in the Release 13 devices most fleets are still running. Deployment plans written against 2019 hardware are leaving capability on the table.
Carl Sagan’s baloney detection kit has one rule that applies here more than any other: wherever possible, there must be independent confirmation of the facts. The coverage map is a fact. The RSSI reading at the meter pit is an independent confirmation. When they disagree, the map is not more true because it came from a more expensive tool.
This is the site’s thesis in a single radio problem. The propagation model is the model. The tunnel is the pipe. Everyone who has ever planned a deployment from a coverage map and then stood in a basement with a device that will not connect has learned, at some cost, that the model gets the credit and the pipe does the work. The DTU team did the harder thing — they went underground with a meter and measured — and what they found is that the variable everyone plans around does not predict the outcome, and the variable that does is not in the tool.
That is not an indictment of the models. Okumura-Hata was never fitted to a tunnel. It is an indictment of using a model outside the conditions it was built for and trusting the output because it came out of a computer. Deming would call it a process running outside its control limits and nobody checking the chart. Measure the corridor. Then trust the map.
Sources
Technical University of Denmark, “Experimental Evaluation of Empirical NB-IoT Propagation Modelling in a Deep-Indoor Scenario,” arXiv 2006.00880 (campus-wide tunnel and basement measurement campaign; received power independent of TX–RX distance; corridor-distance parameter) · “Investigation of Deep Indoor NB-IoT Propagation Attenuation,” ResearchGate 337232028, 2019 (2–12 dB model error; depth and indoor distance increase error; corridor distance as promising feature) · Trafalgar Wireless, “What Is NB-IoT Connectivity,” June 2026 (164 dB MCL; 20 dB over cellular; 180 kHz bandwidth; repetition; market $13.62B 2026 → $51.82B 2031) · Norvi, “NB-IoT Telemetry Device for Water, Gas, and Utility Monitoring,” February 2026 (deployment modes; in-band 57.6% share; $0.95/meter, $181K/month utility case) · ZYIoT, “How is NB-IoT Redefining Industrial Connectivity in 2026,” March 2026 (2–12 dB error citation; Release 18 positioning; eSIM/eUICC and FOTA lifecycle) · Black Ant, “NB-IoT Water Meter Smart Metering: A Utility Deployment Guide,” July 2026 (RSSI/SNR site sampling recommendation) · 3GPP Release 13 NB-IoT specification; IEEE, “NB-IoT system deployment for smart metering: Evaluation of coverage and capacity performances.”
The DTU campaign is the primary source for the model-error and corridor-distance findings. Vendor guidance is cited for deployment practice, not for the physics. Corrections from engineers who have run their own underground measurement campaigns are welcome at Scott@IoTunderground.com.
