The 915 MHz ISM band is a reasonable choice for underground sensor mesh networks — it's unlicensed, well-supported by commercial radio hardware (LoRa modules, sub-GHz mesh chipsets), and propagates better through solid material than 2.4 GHz. But "better than 2.4 GHz" is a low bar, and the practical path loss numbers in real underground geology vary enough across substrate types that designing a node spacing without actual site measurement is a significant gamble.
This article goes through the path loss measurements reported in open literature and our own field characterization data for 915 MHz propagation in three common underground substrates: sandstone, granite, and reinforced concrete corridors. The numbers matter because they directly determine node spacing, battery sizing, and antenna design — all of which affect system cost and deployment logistics.
Propagation Modes in Underground Environments
Before diving into substrate-specific numbers, it's worth clarifying the two distinct propagation regimes that underground sensor networks operate in, because they have very different link budget characteristics:
Through-earth propagation: the signal passes through solid material (rock, soil, concrete) between source and destination nodes. Attenuation is dominated by the material's conductivity and permittivity at the operating frequency. Path loss in this mode is roughly exponential with distance, described by a path loss exponent in the material and a total link budget that includes both the free-space component and the additional material loss. For 915 MHz through competent rock, this is the limiting factor at distances beyond about 10 meters.
In-tunnel air propagation: both nodes are in an air-filled tunnel corridor, and the signal propagates through air in the tunnel with wall reflections creating a waveguide-like environment. This mode is much more favorable than through-earth propagation — the tunnel acts as an imperfect waveguide that can support 915 MHz signals over hundreds of meters with total path loss in the 40–80 dB range, depending on tunnel geometry and wall material. The dominant impairments are multipath fading and standing wave patterns from wall reflections, not bulk material attenuation.
For most underground sensor mesh applications — where nodes are placed in a tunnel corridor and need to communicate along the corridor — the in-tunnel air propagation mode applies. The through-earth mode is relevant when a node needs to transmit through the tunnel wall to a receiver outside the tunnel, or in configurations where nodes are embedded in solid material rather than mounted in air-filled passages.
Path Loss in Sandstone: ~4 dB/m
Sandstone is among the more favorable substrates for through-material 915 MHz propagation. Its conductivity is low (0.001–0.01 mS/m for dry sandstone), and its relative permittivity (4–8) is moderate. Path loss through dry sandstone at 915 MHz is reported in published mining and underground communications literature at approximately 3–6 dB/m, with a central estimate around 4 dB/m for typical southwestern US desert sandstone formations.
At 4 dB/m, a transmitter with +20 dBm output power and a receiver with -110 dBm sensitivity (a LoRa module using SF12/BW125 configuration is approximately this sensitive) has a link budget of 130 dB. Subtracting cable/antenna losses (~3 dB each end), the remaining 124 dB of link budget against 4 dB/m gives maximum through-material range of approximately 31 meters. In practice, additional margin for fading and unexpected material variation means a design range of 20–25 meters through sandstone.
For in-tunnel propagation in a sandstone-walled tunnel, the constraint is looser. Path loss along a 2-meter diameter circular tunnel at 915 MHz in the waveguide regime is on the order of 0.1–0.3 dB/m for the propagating mode, meaning the same link budget supports node-to-node distances of 100–200 meters along a straight corridor before accounting for bends and discontinuities. This is a substantial contrast with through-material propagation, and it explains why surface-to-underground (through-earth) links are much harder than along-corridor links at the same frequency.
Path Loss in Granite: ~8 dB/m
Granite presents a more challenging substrate. While its electrical conductivity is also low in dry conditions (0.001–0.01 mS/m), its higher density and greater permittivity (7–12) increase path loss. Published measurements in granite mine tunnels report 915 MHz through-material path loss of 6–10 dB/m, with a central estimate around 8 dB/m.
At 8 dB/m, the same LoRa link budget gives maximum through-material range of approximately 15 meters, with practical design range around 10–12 meters. For a sensor network requiring coverage across multiple parallel tunnels in a granite formation, this constrains the cross-tunnel communication architecture significantly — you either need relay nodes in every tunnel, or you route all communication through a primary conduit that maintains line-of-sight air propagation back to a surface gateway.
Moisture content in granite fractures significantly worsens propagation. Granite with saturated fractures can exhibit effective conductivity an order of magnitude higher than dry granite, pushing path loss toward 15–20 dB/m and reducing practical through-material range to 6–8 meters. Deep granite mines with water ingress are among the most challenging RF environments for underground mesh networks at any frequency.
Path Loss in Reinforced Concrete: ~12 dB/m
Reinforced concrete is the worst-case common substrate for 915 MHz propagation, for two reasons: the concrete aggregate itself is moderately lossy (conductivity 0.01–0.1 mS/m depending on aggregate type, moisture, and admixtures), and the embedded steel rebar acts as a Faraday cage at the 915 MHz scale when rebar spacing (typically 20–30 cm) is comparable to or smaller than the half-wavelength (16.4 cm at 915 MHz). This creates additional eddy-current losses in the steel that conventional bulk-conductivity models underestimate.
Measured path loss in reinforced concrete at 915 MHz is typically 10–15 dB/m, with some published measurements in very dense reinforced concrete exceeding 20 dB/m. A practical central estimate for standard-construction reinforced concrete (20 MPa, 25 cm rebar spacing) is ~12 dB/m.
At 12 dB/m through-material, the LoRa link budget gives maximum range of approximately 10 meters through concrete, with practical design range of 7–8 meters accounting for margin. This constrains sensor placement in concrete tunnel environments to nodes visible along the air-filled corridor, not through walls. Long-range communication in reinforced-concrete tunnels must use corridor propagation, and cross-wall links require either very short distances, much lower frequencies, or physical conductor runs.
The in-tunnel air propagation in a reinforced-concrete corridor is also noisier than in rock tunnels, due to the rebar-induced standing-wave pattern mentioned in our earlier article on GPS-denied sensor assumptions. At 915 MHz in a 2-meter diameter concrete tunnel, rebar reflection creates spatial fading with 20–30 cm periodicity — meaning a node at a fading null can have 15–20 dB worse received signal than a node 10 cm away at a constructive peak.
LoRa Advantages for Low-SNR Underground Links
LoRa (Long Range) modulation — a chirp spread-spectrum PHY from Semtech, implemented in the SX127x/SX126x chipset family widely used in the LoRaWAN ecosystem — is well-suited for underground sensor mesh applications for several reasons beyond raw link budget:
- Multipath resistance: chirp spread-spectrum is inherently resilient to multipath delay spread up to the chirp period, which at typical LoRa symbol rates (125 kHz BW) is approximately 8 microseconds — corresponding to 2.4 km of differential path delay. Tunnel multipath delays are typically much shorter (nanoseconds to microseconds for corridor reflections), so LoRa handles tunnel multipath well without equalization.
- Receive sensitivity: SF12 / BW125 configuration achieves approximately -137 dBm sensitivity in theory, with practical receiver performance around -130 to -136 dBm depending on implementation. This is 15–25 dB better than 802.15.4/Zigbee at the same transmit power, which translates directly to increased range or reduced transmit power (battery life).
- Duty cycle and power: LoRa transmissions are short (a 100-byte packet at SF7/BW125 takes about 130 ms; at SF12/BW125 about 2 seconds). With infrequent sensor reporting (every 5–30 minutes for steady-state monitoring), average current draw can be kept in the tens of microampere range, enabling multi-year battery life from small lithium cells.
The tradeoff is data rate — SF12/BW125 achieves about 250 bps, which limits throughput. For sensor data (temperature, seismic event packets, status) this is typically adequate; for streaming seismic waveforms or high-rate acoustic data it is not. High-rate data acquisition nodes need either wired backhaul or a higher-data-rate radio link (which will sacrifice the LoRa link budget advantage).
Mesh vs. Star Topology in Tunnel Corridors
Star topology — all nodes communicate directly with a single gateway — is simpler to implement and reason about. It's also fragile for tunnel corridor deployments because: (1) gateway placement at one end of the tunnel creates a link budget problem for nodes far from the gateway, (2) the gateway is a single point of failure, and (3) a long straight tunnel acts as a radio dead-end for any gateway not placed midway or with nodes that can relay.
Mesh topology with multi-hop routing distributes these problems. Nodes route packets through their nearest neighbors, so the maximum hop distance is node-to-adjacent-node rather than node-to-gateway. This allows longer corridors to be covered with fewer high-power relay nodes, and the network continues to function if any single node fails — traffic routes around it.
The practical constraint on mesh topology is routing protocol overhead and timing. Underground sensor networks often run time-slotted channel access (TSCA) protocols to avoid collision, which requires time synchronization between nodes — which brings us back to the GPS-denied timing problem discussed in our earlier article. Mesh routing in timing-sensitive networks requires a synchronization protocol that works without GPS, and most off-the-shelf LoRa mesh stacks assume either GPS or a wired clock reference. This is a solvable engineering problem, not an intractable one, but it is one that requires explicit design attention.
For corridor deployments, our working approach is to treat each straight corridor segment between bends as a waveguide section, size node spacing for the corridor geometry and expected substrate (with measured rather than assumed path loss), and use a tree-topology mesh with the gateway at the surface access point. Multi-hop depths of 3–5 hops are manageable with existing LoRa mesh implementations at acceptable latency for threat detection applications (under 10 seconds end-to-end for a 5-hop alert). Deeper networks need careful latency characterization to confirm they meet the operational alert-time requirement.
A Note on 2.4 GHz Comparison
The through-material path loss difference between 915 MHz and 2.4 GHz is substantial: 2.4 GHz exhibits approximately 2.5× higher path loss per meter in most geological substrates compared to 915 MHz (the ratio of conductivity-dominated attenuation scales roughly with √f). In reinforced concrete, 2.4 GHz through-material range is typically 3–5 meters — barely across a 0.5-meter wall. For any application requiring penetration of construction materials, 915 MHz (or lower sub-GHz frequencies) is the clear choice. For in-tunnel air propagation in very long straight corridors, 2.4 GHz is not categorically worse (multipath is the dominant impairment, not material loss), but the link budget advantage of LoRa at 915 MHz still favors sub-GHz for nodes that need to tolerate fading nulls.
Where 2.4 GHz remains relevant is for high-data-rate backhaul over short distances within the tunnel — bridging a dense sensor array to a local edge compute node, for example. For that use case, Wi-Fi 6 or 802.11ad can deliver the bandwidth needed for seismic waveform streaming, and the range requirement (10–20 meters) is manageable. The correct architecture is often not a choice between 915 MHz and 2.4 GHz but both: a 915 MHz mesh backbone for long-corridor coverage and sensor reporting, with 2.4 GHz or wired backhaul for local high-rate data collection at access nodes.