Why Propagation Models Matter for 5G Planning
Every cell radius calculation, every coverage map, and every interference analysis starts with a propagation model -- the mathematical equation that predicts how much signal strength is lost between transmitter and receiver. Use the wrong model and your planned coverage is fiction: too optimistic means coverage holes in deployment, too conservative means over-spending on unnecessary sites.
3GPP standardized the propagation models for 5G NR evaluation in TR 38.901 (Study on channel model for frequencies from 0.5 GHz to 100 GHz). This technical report defines four deployment scenarios, each with LOS and NLOS path loss equations, shadow fading distributions, and applicability ranges. All major planning tools (Atoll, ASSET, Planet) implement TR 38.901 as their primary 5G model.
Ericsson reported that calibrating TR 38.901 models with local measurement data improved their coverage prediction accuracy from ±8 dB to ±3 dB for STC (Saudi Telecom Company) during their 3.5 GHz network rollout across Riyadh. Nokia published similar findings for KDDI Japan, where tuned UMi models achieved ±2.5 dB accuracy in dense Tokyo districts.The Four TR 38.901 Scenarios
| Scenario | Full Name | Typical Deployment | BS Height | UE Height | Freq Range | ISD Range |
|---|---|---|---|---|---|---|
| UMa | Urban Macro | Urban macro cells, BS above rooftop | 25 m | 1.5--22.5 m | 0.5--100 GHz | 200--5000 m |
| UMi-Street Canyon | Urban Micro | Small cells, BS below rooftop, street level | 10 m | 1.5--22.5 m | 0.5--100 GHz | 10--5000 m |
| RMa | Rural Macro | Rural wide-area macro cells | 10--150 m | 1--10 m | 0.5--30 GHz | 35--10000 m |
| InH-Office | Indoor Hotspot | Indoor office environment | 1--3 m (ceiling) | 1--2.5 m | 0.5--100 GHz | 1--150 m |
All models are valid for frequencies from 0.5 GHz to 100 GHz (except RMa: 0.5--30 GHz) per TR 38.901 clause 7.4.1.
Path Loss Equations
UMa (Urban Macro)
LOS (TR 38.901 Table 7.4.1-1):`
PL_UMa_LOS = 28.0 + 22·log10(d_3D) + 20·log10(f_c)
for 10 m ≤ d_2D ≤ d_BP
PL_UMa_LOS = 28.0 + 40·log10(d_3D) + 20·log10(f_c)
- 9·log10((d_BP)² + (h_BS - h_UE)²)
for d_BP ≤ d_2D ≤ 5000 m
`
Where breakpoint distance: d_BP = 4·h_BS·h_UE·f_c / c
`
PL_UMa_NLOS = 13.54 + 39.08·log10(d_3D) + 20·log10(f_c) - 0.6·(h_UE - 1.5)
PL_UMa = max(PL_UMa_LOS, PL_UMa_NLOS)
`
Shadow fading: σ_SF = 4 dB (LOS), 6 dB (NLOS)
UMi-Street Canyon
LOS:`
PL_UMi_LOS = 32.4 + 21·log10(d_3D) + 20·log10(f_c)
for 10 m ≤ d_2D ≤ d_BP
PL_UMi_LOS = 32.4 + 40·log10(d_3D) + 20·log10(f_c)
- 9.5·log10((d_BP)² + (h_BS - h_UE)²)
for d_BP ≤ d_2D ≤ 5000 m
`
NLOS:
`
PL_UMi_NLOS = 22.4 + 35.3·log10(d_3D) + 21.3·log10(f_c) - 0.3·(h_UE - 1.5)
PL_UMi = max(PL_UMi_LOS, PL_UMi_NLOS)
`
Shadow fading: σ_SF = 4 dB (LOS), 7.82 dB (NLOS)
Shadow Fading Comparison
| Scenario | LOS σ_SF (dB) | NLOS σ_SF (dB) | LOS Probability Model | Decorrelation Distance (m) |
|---|---|---|---|---|
| UMa | 4.0 | 6.0 | Clause 7.4.2, Table 7.4.2-1 | 37 (LOS), 50 (NLOS) |
| UMi | 4.0 | 7.82 | Clause 7.4.2, Table 7.4.2-1 | 10 (LOS), 13 (NLOS) |
| RMa | 4.0 | 8.0 | Clause 7.4.2, Table 7.4.2-1 | 37 (LOS), 120 (NLOS) |
| InH | 3.0 | 8.03 | Clause 7.4.2, Table 7.4.2-1 | 10 (LOS), 6 (NLOS) |
Shadow fading follows a log-normal distribution. Network planning typically uses the 75th percentile (mean + 0.67·σ) for coverage probability design, meaning you add 0.67 × σ_SF dB as a shadow fading margin.
Worked Example 1: UMa Path Loss at 3.5 GHz
Given:- Frequency: f_c = 3.5 GHz
- BS height: h_BS = 25 m
- UE height: h_UE = 1.5 m
- 2D distance: d_2D = 500 m
- 3D distance: d_3D = sqrt(500² + (25-1.5)²) = sqrt(250000 + 552.25) = 500.55 m
`
d_BP = 4 × h_BS × h_UE × f_c / c
= 4 × 25 × 1.5 × 3.5×10⁹ / (3×10⁸)
= 525,000,000,000 / 300,000,000
= 1,750 m
`
Since d_2D = 500 m < d_BP = 1,750 m, we use the first LOS equation.
Step 2: LOS path loss`
PL_UMa_LOS = 28.0 + 22·log10(500.55) + 20·log10(3.5)
= 28.0 + 22 × 2.6994 + 20 × 0.5441
= 28.0 + 59.39 + 10.88
= 98.3 dB
`
Step 3: NLOS path loss
`
PL_UMa_NLOS = 13.54 + 39.08·log10(500.55) + 20·log10(3.5) - 0.6·(1.5 - 1.5)
= 13.54 + 39.08 × 2.6994 + 10.88 - 0
= 13.54 + 105.50 + 10.88
= 129.9 dB
`
Step 4: Apply shadow fading margin (75th percentile)
`
LOS total: 98.3 + 0.67 × 4.0 = 98.3 + 2.7 = 101.0 dB
NLOS total: 129.9 + 0.67 × 6.0 = 129.9 + 4.0 = 133.9 dB
`
Interpretation: At 500 m from a 25 m macro site on 3.5 GHz, path loss ranges from 101 dB (LOS with margin) to 134 dB (NLOS with margin). Given a typical downlink MAPL of 142 dB, both LOS and NLOS are within coverage -- but at 1 km, NLOS path loss would exceed 150 dB, creating a coverage hole.
Worked Example 2: UMi vs InH at 28 GHz (mmWave)
Given:- Frequency: f_c = 28 GHz
- Scenario A (UMi): h_BS = 10 m, h_UE = 1.5 m, d_3D = 100 m
- Scenario B (InH): h_BS = 3 m, h_UE = 1.5 m, d_3D = 30 m
`
PL_UMi_NLOS = 22.4 + 35.3·log10(100) + 21.3·log10(28) - 0.3·(1.5 - 1.5)
= 22.4 + 35.3 × 2.0 + 21.3 × 1.4472
= 22.4 + 70.6 + 30.8
= 123.8 dB
With shadow fading margin (75th %ile): 123.8 + 0.67 × 7.82 = 129.0 dB
`
InH NLOS at 30 m (TR 38.901 Table 7.4.1-1):
`
PL_InH_NLOS = 17.3 + 38.3·log10(30) + 24.9·log10(28)
= 17.3 + 38.3 × 1.4771 + 24.9 × 1.4472
= 17.3 + 56.6 + 36.0
= 109.9 dB
With shadow fading margin (75th %ile): 109.9 + 0.67 × 8.03 = 115.3 dB
`
Comparison:
| Metric | UMi (100 m, outdoor) | InH (30 m, indoor) |
|---|---|---|
| NLOS path loss | 123.8 dB | 109.9 dB |
| With SF margin | 129.0 dB | 115.3 dB |
| Typical EIRP (28 GHz) | 55 dBm | 33 dBm |
| Rx sensitivity | -95 dBm | -95 dBm |
| Available MAPL | 150 dB | 128 dB |
| Link margin | 21.0 dB | 12.7 dB |
Both scenarios have positive link margin, but the outdoor UMi deployment benefits from higher EIRP (larger phased array at the BS), while the indoor InH deployment operates with reduced EIRP but shorter distances.
LOS Probability Models
TR 38.901 clause 7.4.2 defines the probability of LOS as a function of 2D distance. This is critical because the LOS/NLOS path loss difference can exceed 30 dB.
| Scenario | LOS Probability Formula (simplified) | 50% LOS Distance |
|---|---|---|
| UMa | P_LOS = min(18/d_2D, 1)·(1-exp(-d_2D/63)) + exp(-d_2D/63) | ~65 m |
| UMi | P_LOS = min(18/d_2D, 1)·(1-exp(-d_2D/36)) + exp(-d_2D/36) | ~40 m |
| RMa | P_LOS = exp(-(d_2D-10)/1000) for d_2D > 10 m | ~700 m |
| InH-Office | P_LOS = exp(-(d_2D-1.2)/4.7)·(1-exp(-d_2D/6.5)) + exp(-d_2D/6.5) for d_2D > 1.2 m | ~5 m |
The dramatically different LOS probabilities explain why RMa cells can be 10--30 km while UMi cells are typically 100--300 m.
Model Calibration with Drive Test Data
Ericsson published calibration results for three major operators:| Operator | Market | Frequency | Model | Uncalibrated RMSE | Calibrated RMSE | Calibration Offset |
|---|---|---|---|---|---|---|
| STC (Saudi Arabia) | Riyadh urban | 3.5 GHz | UMa NLOS | 8.2 dB | 3.1 dB | +4.5 dB (buildings more lossy) |
| KDDI (Japan) | Tokyo dense urban | 3.7 GHz | UMi NLOS | 6.8 dB | 2.5 dB | -2.1 dB (narrower streets, waveguide) |
| Telstra (Australia) | Melbourne suburban | 3.5 GHz | UMa NLOS | 7.5 dB | 2.8 dB | +3.2 dB (wide streets, fewer reflections) |
Calibration typically involves adding a constant offset and adjusting the distance-dependent slope. The uncalibrated models provide reasonable first-pass estimates, but production planning always requires drive-test calibration.
3GPP Specification References
- TR 38.901: Study on channel model for frequencies from 0.5 GHz to 100 GHz -- the primary source for all path loss equations, shadow fading parameters, LOS probabilities, and fast fading channel models
- TR 38.900: Channel model for frequency spectrum above 6 GHz -- predecessor to TR 38.901, focused on mmWave; largely superseded but still referenced
- TS 38.104: Base station radio transmission and reception -- defines reference propagation conditions for conformance testing that align with TR 38.901
- ITU-R M.2412: Guidelines for evaluation of radio interface technologies for IMT-2020 -- specifies the evaluation methodology using TR 38.901 models
Choosing the Right Model for Your Deployment
| Deployment Type | Recommended Model | Key Considerations |
|---|---|---|
| Macro overlay, urban | UMa | BS must be above average rooftop; use NLOS equation for >80% of coverage area |
| Small cell, street level | UMi-Street Canyon | BS below rooftop; strong street canyon waveguide effects at mmWave |
| Rural wide area | RMa | Valid only up to 30 GHz; very high LOS probability → large cell radii |
| Indoor enterprise | InH-Office | Short distances, ceiling-mounted APs; model assumes open-plan office |
| Outdoor-to-indoor | UMa + O2I penetration loss | Add building penetration loss per TR 38.901 clause 7.4.3 (15--25 dB for modern buildings at 3.5 GHz) |
Conclusion
TR 38.901 provides a rigorous, frequency-scalable propagation framework that underpins all 5G NR planning. The four scenarios -- UMa, UMi, RMa, and InH -- cover the full range of 5G deployments from rural macro to indoor hotspot. Engineers must select the correct scenario based on BS height relative to surrounding buildings, apply appropriate shadow fading margins for target coverage probability, and calibrate against local drive test data to achieve the ±3 dB prediction accuracy required for efficient site planning.