Radio Propagation Models:
5G NR to 6G Sub-THz

The Physics of Radio Propagation

Every mobile network ever built faces the same fundamental enemy: physics. The moment a radio signal leaves an antenna, it begins to weaken. Understanding exactly how and why it weakens — and building mathematical models that predict that weakening — is the foundation upon which every cell plan, every coverage map, and every link budget is constructed. Get the propagation model wrong, and you build a network that either wastes money on unnecessary sites or leaves customers without coverage.

Why Signals Weaken: The Inverse Square Law

In the simplest possible case — empty space with no obstacles — a radio signal spreads outward from the transmitter as an expanding sphere. The surface area of a sphere is 4πr², so the power density (watts per square metre) at distance d decreases as 1/d². Double the distance, and the received power drops to one quarter. This is the inverse square law, and it sets the absolute minimum signal loss for any wireless link.

In decibels, this relationship is captured by the Free Space Path Loss (FSPL) equation:

Notice the 20·log₁₀(f) term: doubling the frequency adds 6 dB of loss. This is why mmWave signals at 28 GHz suffer about 22 dB more free-space loss than a 700 MHz signal at the same distance. It is not that the wave “weakens faster” at higher frequency — the spreading loss is the same. The difference is that a higher-frequency signal has a shorter wavelength, which means a fixed-size receive antenna captures a smaller fraction of the expanding sphere.

Three Mechanisms: Reflection, Diffraction, Scattering

In the real world, signals rarely travel in a straight line from transmitter to receiver. Three physical mechanisms alter the signal path:

• Reflection occurs when a wave encounters a surface much larger than its wavelength (buildings, the ground, water). The wave bounces in a predictable direction according to Snell’s law. This is the dominant mechanism in urban areas and creates strong multipath components.
• Diffraction occurs when a wave encounters an edge or corner of an obstacle (rooftop, hilltop, building corner). The wave “bends” around the obstacle, enabling NLOS (Non-Line-of-Sight) coverage. Diffraction loss increases sharply with frequency, which is why mmWave signals penetrate poorly behind obstacles while sub-1 GHz signals wrap around buildings.
• Scattering occurs when a wave encounters objects comparable in size to its wavelength (lamp posts, leaves, raindrops, surface roughness). The wave is dispersed in many directions. At mmWave frequencies, even small objects become effective scatterers.

Additional Propagation Phenomena

Beyond the three primary mechanisms, several additional phenomena affect radio propagation in real-world deployments:

Atmospheric Refraction

The refractive index of the atmosphere decreases with altitude, causing radio waves to bend slightly downward. This “standard atmosphere” refraction extends the radio horizon beyond the geometric horizon by approximately 15%. RF engineers use an “effective Earth radius” of K · R_earth where K = 4/3 = 1.333 to account for this bending. Under anomalous atmospheric conditions (temperature inversions, ducting), K can increase to 2 or more, enabling signals to propagate hundreds of kilometres beyond the normal range — this is the well-known “tropospheric ducting” phenomenon that can cause interference between distant cells.

Ground Reflections and Fresnel Zones

For terrestrial LOS links, the ground-reflected ray interferes with the direct ray. The result depends on the phase difference between the two paths, which varies with distance, antenna heights, and frequency. The Fresnel zone — an ellipsoidal region around the direct path — defines the volume of space that must be kept clear of obstructions for efficient propagation. The first Fresnel zone radius at the midpoint of a link of length d at frequency f is:

At mmWave, the Fresnel zone is remarkably narrow (just over 1 metre at midpoint for a 500 m link). This means that even a single tree branch or street lamp entering the first Fresnel zone can measurably affect signal quality. Conversely, it means that narrow beams from massive MIMO arrays can thread between obstacles that would block lower-frequency signals.

Diffraction Loss: The Knife-Edge Model

When a radio wave encounters a sharp edge (rooftop, hilltop, building corner), it diffracts around the obstacle according to Huygen’s principle. The simplest model for single-edge diffraction is the knife-edge diffraction model, characterized by the Fresnel-Kirchhoff diffraction parameter ν:

The critical frequency dependence is in the √(1/λ) term: at higher frequencies (shorter λ), the diffraction parameter ν increases for the same geometry, meaning more diffraction loss. A building corner that causes 10 dB of diffraction loss at 700 MHz may cause 20+ dB at 3.5 GHz and 30+ dB at 28 GHz. This is the physical mechanism behind the dramatic LOS/NLOS path loss difference at mmWave.

In urban environments, signals typically diffract over multiple building rows (multi-screen diffraction). Models like the Deygout method (using the dominant edge) and the Bullington equivalent-edge method are used to handle multiple diffracting edges. TR 38.901 does not model individual diffracting edges — their cumulative effect is captured statistically in the NLOS PLE and shadow fading.

Foliage Loss

Vegetation attenuation is a significant but often overlooked factor, especially in suburban and rural environments. Trees and dense foliage absorb and scatter radio energy. The ITU-R P.833 recommendation provides models for vegetation loss:

• Sub-1 GHz: 0.2–0.5 dB/m of foliage depth. A 10 m deep tree canopy adds 2–5 dB.
• 3.5 GHz: 0.5–1.5 dB/m. The same 10 m canopy adds 5–15 dB. Significant for street-level small cells on tree-lined avenues.
• 28 GHz: 3–5 dB/m. Even a single tree (2–3 m depth) adds 6–15 dB. Dense foliage can completely block mmWave signals.
• Seasonal variation: Deciduous trees in temperate climates can show 10–15 dB less foliage loss in winter (bare branches) compared to summer (full canopy). Network planning must consider worst-case seasonal conditions.

Path Loss vs. Fading: The Critical Distinction

Engineers must distinguish between two types of signal variation:

• Path loss (large-scale) is the predictable, distance-dependent reduction in signal power. It changes slowly as the user moves — on the order of tens of metres. This is what propagation models predict.
• Fading is the random fluctuation around the predicted path loss. Shadow fading (large-scale fading) is caused by buildings and terrain blocking the signal and follows a log-normal distribution with standard deviation typically 4–8 dB. Small-scale fading (multipath fading) varies on the scale of half a wavelength (about 5 cm at 3.5 GHz) and follows Rayleigh or Rician distributions.

A propagation model predicts the median path loss. Shadow fading is added as a random variable with a specified standard deviation. Small-scale fading is handled separately in channel models (e.g., the spatial channel model SCM or the cluster delay line CDL model).

The Link Budget Equation

Every wireless link can be expressed as a link budget — a simple accounting of gains and losses from transmitter to receiver:

The propagation model is the heart of this equation. Everything else — transmit power, antenna gain, receiver sensitivity — is known hardware specification. Path loss is the unknown that determines whether a link works or fails. This is why propagation modeling matters more than any other aspect of network planning.

Why Frequency Matters: The Wavelength Perspective

The interaction between radio waves and the physical environment is fundamentally governed by the ratio of wavelength to object size. This ratio determines which of the three mechanisms (reflection, diffraction, scattering) dominates:

• Object >> λ: Strong reflection and shadowing. A building (10 m) at 3.5 GHz (λ = 8.6 cm) is ~116 wavelengths across — it creates a sharp shadow with clear reflection.
• Object ≈ λ: Strong scattering. A lamp post (15 cm diameter) at 28 GHz (λ = 10.7 mm) is ~14 wavelengths across — it scatters energy in many directions.
• Object << λ: Wave passes through with minimal interaction. A 1 cm pebble at 700 MHz (λ = 43 cm) is invisible to the wave.

This is why the same urban environment produces fundamentally different propagation at different frequencies. At 700 MHz, radio waves diffract around buildings and penetrate walls easily. At 28 GHz, they bounce off surfaces like light, cannot penetrate solid walls, and scatter from objects that were invisible at lower frequencies. At 140 GHz, even rough surfaces scatter energy, and atmospheric molecules absorb it.

Polarization Effects

Radio waves are electromagnetic waves with electric and magnetic field components that oscillate perpendicular to the direction of propagation. The orientation of the electric field defines the polarization. Polarization effects become increasingly important at higher frequencies:

• Vertical polarization (V-pol): The electric field oscillates vertically. V-pol experiences less reflection loss from the ground at typical elevation angles, making it preferred for land mobile communications.
• Horizontal polarization (H-pol): The electric field oscillates horizontally. H-pol may experience less foliage attenuation in some configurations.
• Cross-polarization ratio (XPR): In LOS conditions, the XPR can exceed 25 dB (strong polarization preservation). In NLOS, scattering and reflection depolarize the signal, reducing XPR to 5–10 dB. TR 38.901 specifies XPR for each scenario.
• Dual-polarized MIMO: Modern 5G NR systems use ±45° slant-polarized antenna elements. The cross-polar isolation between these polarizations provides a second MIMO layer without requiring spatial separation, effectively doubling spectral efficiency when the channel supports it.

At mmWave frequencies, polarization purity is higher in LOS (XPR > 20 dB) but rapidly degrades after reflections (XPR drops to 5–8 dB after a single reflection). This means that dual-polarized massive MIMO systems at mmWave can reliably achieve 2x spatial multiplexing in LOS but must fall back to single-layer operation in rich NLOS environments.

Temporal Variability

Propagation conditions are not static — they vary over multiple timescales:

• Milliseconds: Small-scale fading from multipath. Varies at half-wavelength intervals (5 mm at 28 GHz). Managed by OFDM and HARQ.
• Seconds: Shadowing changes from vehicle and pedestrian movement. Decorrelation distance 10–50 m.
• Minutes to hours: Traffic density changes, weather variations (rain onset/cessation), atmospheric pressure changes.
• Seasonal: Foliage changes in deciduous trees (10+ dB variation between summer and winter at 3.5 GHz). Building construction and demolition.
• Diurnal: Temperature and humidity cycles affect atmospheric refraction, particularly for long-range links (RMa, NTN).

TR 38.901 captures only the spatial variability (path loss + shadow fading at a snapshot in time). Temporal variability is handled by the small-scale fading model (CDL/TDL) for fast variations and by operational margins in the link budget for slow variations. For NTN links, temporal variability from atmospheric conditions is a primary design concern.

The Pioneers — Classic Empirical Models

Before 5G, before 3GPP TR 38.901, RF engineers relied on empirical models derived from extensive measurement campaigns. These models remain in use for 2G/3G/4G planning and serve as the conceptual foundation for modern models. Every RF engineer must understand them.

Okumura-Hata Model (1968/1980)

Yoshihisa Okumura conducted the seminal measurement campaign in Tokyo in 1968, measuring signal strength across a wide range of frequencies (200–1920 MHz), distances (1–100 km), and antenna heights. His 1968 paper, published in the journal of the Institute of Electronics and Communication Engineers of Japan, presented results as correction factor curves plotted on graph paper — precise enough for engineering use but cumbersome to implement in computer simulations.

In 1980, Masaharu Hata of Okumura’s laboratory converted these graphical results into closed-form equations, creating the most widely used propagation model in the history of mobile communications. The genius of Hata’s formulation was its simplicity: the entire urban propagation environment was captured in a single logarithmic equation with correction factors for antenna height and city size.

Applicability and Validity Range

The Okumura-Hata model is valid under these specific conditions:

• Frequency: 150–1500 MHz
• Distance: 1–20 km (not valid below 1 km due to near-field effects of urban clutter)
• BS antenna height: 30–200 m (above rooftop level)
• UE antenna height: 1–10 m
• Terrain: Quasi-smooth terrain (not mountainous)

Using the model outside these bounds — particularly at frequencies above 1500 MHz or distances below 1 km — can produce significant errors. The model was never intended for small cell or micro cell planning.

COST 231 Hata Model

The European COST 231 project extended the Hata model to cover 1500–2000 MHz, enabling planning for GSM 1800 and early UMTS networks. The formula adds a metropolitan correction factor C_m:

Suburban and Open Area Corrections

The base Okumura-Hata formula predicts urban path loss. For suburban and open (rural) areas, correction factors are applied:

The open area correction of nearly 29 dB at 900 MHz illustrates the enormous impact of the urban environment on path loss. A rural open field has path loss close to free space, while a dense city adds 29 dB of additional clutter loss from buildings, vehicles, and other obstructions.

COST 231 Walfisch-Ikegami (W-I) Model

The W-I model adds physical parameters — building heights, street widths, street orientation angle — to capture the “street canyon” waveguide effect in urban environments. It was developed under the European COST 231 project (1989–1996) and represents a transition from purely empirical models toward semi-deterministic approaches. It provides separate LOS and NLOS formulas:

W-I NLOS Components

The W-I NLOS formula decomposes the total path loss into three physically meaningful components:

The W-I model’s major innovation was incorporating street orientation. A street aligned directly toward the BS (incident angle φ = 0°) has 10–15 dB less loss than a street perpendicular to the BS direction (φ = 90°). This anisotropic coverage pattern is well-documented in urban measurement campaigns and is physically intuitive — a street pointing toward the BS acts as a waveguide, while a perpendicular street forces the signal to diffract around a corner.

SUI (Stanford University Interim) Model

The SUI model, developed for IEEE 802.16 (WiMAX) in 2001 by Stanford University and AT&T Wireless, classifies terrain into three categories and uses a modified path loss exponent. It was designed for fixed broadband wireless access at frequencies from 2 to 11.5 GHz:

• Category A: Maximum path loss (hilly terrain, heavy tree density). γ ranges from 4.6 to 6.4.
• Category B: Moderate path loss (hilly with light vegetation, flat with moderate-to-heavy). γ ranges from 4.0 to 5.2.
• Category C: Minimum path loss (flat terrain, light tree density). γ ranges from 3.6 to 4.4.

ECC-33 Model

The ECC-33 model, developed by the Electronic Communications Committee (a body within CEPT), extends empirical models to modern cellular bands up to 3.5 GHz. It was developed in 2003 to support spectrum planning for 3G/4G services in European countries. The model extrapolates Okumura measurements using a combination of free space attenuation, basic median path loss, and BS/UE height gain factors.

Comparison of Classic Models

Parameter	Okumura-Hata	COST 231 Hata	W-I	SUI
Freq Range	150–1500 MHz	1500–2000 MHz	800–2000 MHz	2–11.5 GHz
Distance	1–20 km	1–20 km	0.02–5 km	0.1–8 km
BS Height	30–200 m	30–200 m	4–50 m	10–80 m
Environment	Urban/Suburban	Urban/Suburban	Urban streets	Suburban/Rural
Based on	Tokyo measurements	European cities	Building geometry	IEEE 802.16 data
PLE (typical)	3.5–4.5	3.5–4.5	2.6 (LOS), 3.8+ (NLOS)	3.6–6.4
Best for	2G/3G macro	GSM 1800, UMTS	Dense urban micro	WiMAX, fixed wireless

Limitations of Classic Models

Despite their widespread use, classic empirical models have fundamental limitations that made them inadequate for 5G NR planning:

• Frequency ceiling: Okumura-Hata stops at 1500 MHz; COST 231 Hata at 2000 MHz. Neither can model the 3.5 GHz (n78) or 28 GHz (n257/n258) bands that define 5G NR.
• No LOS/NLOS distinction: Classic models produce a single path loss value without explicitly separating LOS and NLOS conditions. This matters enormously at mmWave frequencies where the LOS/NLOS difference exceeds 30 dB.
• No 3D distance: Classic models use horizontal distance only. At mmWave frequencies with small cell radii, the vertical component (BS height minus UE height) becomes a significant fraction of the total distance.
• No indoor scenarios: Classic models are exclusively outdoor macro. They cannot model indoor hotspot or factory environments.
• Measurement-specific bias: Okumura’s data was from 1960s Tokyo. Urban environments have changed significantly in building materials, heights, and density.

These limitations drove 3GPP to develop an entirely new channel model framework for 5G NR, which we explore beginning in Chapter 3.

The 3GPP Revolution — TR 38.901 Framework

When 3GPP began standardizing 5G NR in Release 14, the existing propagation models were fundamentally inadequate. Okumura-Hata could not handle frequencies above 2 GHz. COST 231 stopped at 2 GHz. WiMAX-era models lacked the scenario diversity needed for 5G’s vastly different deployment types. 3GPP needed a single, unified channel model framework that could span 0.5 to 100 GHz, cover every environment from rural fields to factory floors, and be validated against the most extensive measurement campaign in wireless history.

The result was 3GPP TR 38.901 (“Study on channel model for frequencies from 0.5 to 100 GHz”), which built upon the earlier 3GPP TR 36.873 (for sub-6 GHz LTE-Advanced) and the ITU-R M.2412 calibration requirements. Measurement data came from research labs worldwide: NYU Wireless, Ericsson, Nokia, Samsung, Qualcomm, Huawei, NIST, and dozens of universities across Europe, Asia, and North America.

Model Structure: The Three Layers

Every TR 38.901 scenario follows the same structure:

Key Parameters

• d_3D: 3D distance between BS and UE = √(d_2D² + (h_BS - h_UT)²)
• d_2D: 2D horizontal distance between BS and UE
• h_BS: Base station antenna height above ground
• h_UT: User terminal antenna height above ground (default 1.5 m outdoor)
• f_c: Carrier frequency in GHz
• d_BP: Breakpoint distance = 4 · h′_BS · h′_UT · f_c / c, where h′ = h - h_E (effective heights)

Deployment Scenarios at a Glance

Scenario	BS Height	Freq Range	Max Distance	Key Use Case
UMa	25 m	0.5–100 GHz	5 km	Urban macro cells
UMi-SC	10 m	0.5–100 GHz	5 km	Street-level small cells
RMa	35 m (typ.)	0.5–30 GHz	21 km	Rural coverage
InH-Office	3 m (ceiling)	0.5–100 GHz	150 m	Office/commercial indoor
InF-SL/DL/SH/DH	1–25 m	0.5–100 GHz	600 m	Factory / Industry 4.0

Measurement Campaign Foundation

TR 38.901 is not a theoretical model — it is grounded in the largest coordinated measurement campaign in wireless history. Key contributors and their measurement bands include:

• NYU Wireless (NYU, USA): Extensive campaigns at 28 GHz, 38 GHz, 73 GHz, and 140 GHz in New York City. Pioneered the CI model methodology and provided critical outdoor urban data.
• Ericsson: Measurements at 2 GHz, 15 GHz, 28 GHz, and 60 GHz across Stockholm and other European cities. Focus on UMa and UMi scenarios with calibrated channel sounders.
• Nokia Bell Labs: Indoor and outdoor campaigns at 10 GHz, 28 GHz, and 60 GHz. Factory environment measurements that informed the InF model.
• Samsung: Dense urban measurements at 28 GHz and 39 GHz in Seoul and Daejeon. Key data for high-frequency UMi validation.
• Qualcomm: Measurements at 28 GHz and 60 GHz in San Diego. Contributed to O2I penetration loss characterization.
• Huawei: Campaigns at 3.5 GHz, 6 GHz, and 28 GHz in Beijing and Shanghai. Large dataset for UMa validation in Asian urban environments.
• NIST (USA): Factory environment measurements at 28 GHz that were essential for the InF model development.

This global measurement foundation ensures that TR 38.901 is not biased toward any single city, building style, or equipment vendor. The model has been cross-validated against independent datasets from multiple continents.

How TR 38.901 Relates to Other Standards

TR 38.901 does not exist in isolation. It builds upon and relates to several other channel models:

• 3GPP TR 36.873 (Rel-12): The predecessor model for LTE-Advanced, covering 2–6 GHz with 3D geometry. TR 38.901 extends this to 0.5–100 GHz and adds mmWave-specific features.
• ITU-R M.2412 (IMT-2020 evaluation): The ITU-R evaluation methodology for 5G (IMT-2020) candidate technologies. TR 38.901 was designed to be compatible with M.2412 calibration scenarios.
• 3GPP TR 38.900 (Rel-14): The initial study on channel models above 6 GHz that preceded TR 38.901. Some early 5G simulations reference 38.900, but 38.901 supersedes it.
• WINNER II / WINNER+ models: European research project models that influenced 3GPP’s approach, particularly the scenario classification (UMa, UMi, RMa, InH) and the stochastic channel generation method.
• mmMAGIC model: EU Horizon 2020 project channel model for 6–100 GHz. Its measurements directly contributed to TR 38.901 validation.

Applying the Complete Channel Model

It is important to understand that path loss (which this article focuses on) is only one component of the complete TR 38.901 channel model. The full model also generates:

For network planning (coverage prediction), only the path loss formulas are needed. For system-level capacity simulations (throughput, latency, reliability), the complete model with all LSPs and SSPs must be generated. For link-level simulations (BLER curves, HARQ performance), the cluster delay line (CDL) or tapped delay line (TDL) simplifications are used.

Common Simulation Parameters

For engineers setting up 3GPP-compliant system-level simulations, TR 38.901 Section 7.2 specifies default parameter values. Here are the most commonly used ones for each scenario:

Parameter	UMa	UMi-SC	RMa	InH-Office
hBS	25 m	10 m	35 m	3 m (ceiling)
hUT	1.5–22.5 m	1.5–22.5 m	1.5 m	1–2.5 m
hUT default	1.5 m	1.5 m	1.5 m	1.5 m
ISD	500 m (dense), 1732 m	200 m	1732–5000 m	—
Min 2D distance	35 m	10 m	35 m	1 m
Max 2D distance	5000 m	5000 m	10000 m	150 m
Building height	20 m (avg)	Building < BS	5 m	3 m (room)
Street width	20 m	20 m	20 m	—

Correlation Between LSPs

An important but often overlooked aspect of TR 38.901 is the cross-correlation between large-scale parameters. For example, in UMa NLOS:

• DS vs. ASA: Correlation = 0.4 (longer delay spread correlates with wider angular spread)
• DS vs. K-factor: Correlation = -0.4 (longer delay spread in non-LOS-dominant conditions)
• SF vs. DS: Correlation = -0.5 (more shadowing = less multipath delay spread)
• SF vs. K-factor: Correlation = 0.5 (less shadowing = stronger direct path = higher K-factor)

These cross-correlations ensure that the generated channel parameters are physically consistent. Without them, the simulation could produce impossible combinations (e.g., high K-factor with long delay spread, which would require both a strong direct path and strong scattering simultaneously).

The Role of TR 38.901 in 5G NR Evaluation

When 3GPP evaluated candidate 5G NR technologies against the IMT-2020 requirements, TR 38.901 was the mandatory channel model for all simulations. This means that every 5G NR feature — from the OFDM numerology to massive MIMO to LDPC/Polar coding — was designed and optimized against TR 38.901 channel conditions. The model is not just a planning tool; it is embedded in the DNA of 5G NR itself.

For this reason, using a different propagation model for network planning (e.g., reverting to Okumura-Hata at sub-6 GHz) creates a systematic mismatch between the channel the system was designed for and the channel the planner assumes. While the difference may be manageable at low frequencies, it becomes significant at mmWave where the 3GPP model captures phenomena (LOS probability, breakpoint distance, shadow fading correlation) that classic models do not.

O2I Model Integration in TR 38.901

A commonly misunderstood aspect of TR 38.901 is how outdoor-to-indoor users are handled in system-level simulations. The specification defines that for each user, the simulator must:

1. Determine indoor/outdoor status: Based on the simulation parameters, a fraction of users are designated as “indoor” (typically 80% in urban scenarios per ITU-R M.2412).
2. For indoor users, assign building type: Low-loss or high-loss, based on the deployment scenario (e.g., 70% high-loss in dense urban, 30% high-loss in suburban).
3. Calculate outdoor path loss: Use the UMa or UMi model to calculate path loss from BS to the building facade.
4. Add O2I loss: Apply the building penetration loss (PL_tw) from Table 7.4.3-1/2 based on building type, frequency, and material composition.
5. Add indoor distance loss: Draw a random indoor depth d_in uniformly between 0 and 25 m, then add 0.5·d_in dB.
6. Add O2I shadow fading: Draw from N(0, σ_p²) where σ_p = 4.4 dB (low-loss) or 6.5 dB (high-loss).

This procedure ensures that system-level simulation results include the impact of indoor users, which is critical for capacity evaluation. Without O2I modeling, simulations would overestimate the SINR of indoor users (by 15–30 dB at 3.5 GHz) and produce unrealistically optimistic throughput predictions.

High-Rise Building Effects

TR 38.901 includes a special provision for high-rise buildings in the UMa scenario. For UE heights above 13 m (approximately the 4th floor), the LOS probability function includes a height-dependent correction term C′(d, h_UT) that increases LOS probability for elevated users. This is physically correct: a user on the 20th floor of a building has a better chance of LOS to a macro cell than a ground-level user because the signal passes above many intervening buildings.

However, the effective antenna height h′_UT in the breakpoint distance calculation also changes for elevated users. The effective environment height h_E is randomly selected from a discrete distribution:

In practice, high-rise modeling is important for cities with significant vertical development (Mumbai, Delhi NCR, Bangalore, Hyderabad). Users on upper floors may receive stronger signals from distant cells (due to elevated LOS) while suffering worse SINR from increased inter-cell interference (visible to more cells). This trade-off between coverage improvement and interference increase at height is a key consideration for dense urban network optimization.

UMa — Urban Macro

The Urban Macro (UMa) scenario models conventional macro cells with base station antennas mounted above rooftop level. It is the workhorse model for 5G NR coverage planning in cities. The base station height is assumed to be 25 m, and the UE height defaults to 1.5 m outdoor (but can range from 1.5 to 22.5 m for high-rise users).

UMa LOS Path Loss

The breakpoint distance is the transition point where the path loss exponent (PLE) changes from 2.2 (below breakpoint) to 4.0 (above breakpoint). Below the breakpoint, there is constructive interference between the direct ray and the ground-reflected ray. Above the breakpoint, the two rays begin to interfere destructively, causing faster power decay.

Breakpoint Distance at Key Frequencies

The breakpoint distance is directly proportional to frequency, which has major implications for cell planning:

Frequency	d′BP (UMa)	d′BP (UMi)	Implication
0.7 GHz	112 m	42 m	Most UMa users beyond breakpoint (PLE=4)
2.1 GHz	336 m	126 m	Moderate breakpoint, balanced coverage
3.5 GHz	560 m	210 m	Many UMa users within breakpoint (PLE=2.2)
28 GHz	4,480 m	1,680 m	All practical users within breakpoint
39 GHz	6,240 m	2,340 m	Breakpoint irrelevant (cells too small)

At mmWave frequencies, the breakpoint distance exceeds the maximum cell range, meaning that all users are in the PLE=2.2 regime for LOS. This is a favourable property — mmWave LOS path loss grows relatively slowly with distance. The problem at mmWave is not the LOS path loss (which is manageable); it is the transition to NLOS (which is catastrophic).

UMa NLOS Path Loss

The difference between LOS and NLOS at 500 m is a staggering 31.6 dB — a factor of over 1,400 in received power. This demonstrates why LOS probability is critical for coverage planning.

UMa at mmWave: 28 GHz Example

The same UMa formulas apply at mmWave frequencies, but the numbers become dramatically different:

At 28 GHz, even the LOS path loss at just 200 m exceeds 107 dB. Combine this with the fact that LOS probability at 200 m in UMa is only ~18%, and you understand why mmWave 5G NR has inter-site distances of 100–200 m (compared to 500–1500 m for sub-6 GHz).

UMa Path Loss: Complete Frequency Comparison

To provide a complete reference for network planning across all 5G NR frequency bands, here is UMa NLOS path loss at key distances for every major band:

Distance	700 MHz	2100 MHz	3500 MHz	28 GHz	39 GHz
50 m (NLOS)	83.5 dB	93.1 dB	97.6 dB	115.5 dB	118.4 dB
100 m (NLOS)	95.3 dB	104.8 dB	109.3 dB	127.2 dB	130.1 dB
200 m (NLOS)	107.1 dB	116.6 dB	121.1 dB	139.0 dB	141.9 dB
500 m (NLOS)	122.7 dB	132.3 dB	136.8 dB	154.7 dB	157.6 dB
1000 m (NLOS)	134.5 dB	144.0 dB	148.5 dB	166.4 dB	169.3 dB

This table makes several planning realities immediately clear:

• At 700 MHz (n28): NLOS coverage at 1 km is viable (134.5 dB vs. 155 dB link budget = 20.5 dB margin)
• At 3.5 GHz (n78): NLOS coverage beyond 500 m is marginal (136.8 dB, only ~18 dB margin with massive MIMO)
• At 28 GHz (n257): NLOS coverage beyond 200 m is essentially impossible (139 dB, only ~16 dB margin, consumed by shadow fading + body loss)
• At 39 GHz (n260): NLOS coverage beyond 100 m requires extraordinary beamforming gain

UMa Optional NLOS Formula

TR 38.901 also provides an optional UMa NLOS formula based on the 3D distance model with different coefficients. This alternative was derived from a different subset of measurement data and can be used for cross-validation:

Shadow Fading Correlation

Shadow fading is not independent between nearby locations. TR 38.901 specifies spatial correlation using an exponential autocorrelation model:

This correlation matters for system-level simulations: without it, interference between users would be unrealistically variable, leading to optimistic throughput predictions. Decorrelation distances of 37–50 m mean that shadow fading is essentially constant within a typical cell sector but varies between sectors.

UMi — Urban Micro (Street Canyon)

The Urban Micro – Street Canyon (UMi-SC) scenario models small cells deployed below rooftop level. The base station height is 10 m — typically a lamp post or low building facade mount. This model captures the waveguide effect of street canyons, where signal energy is confined between parallel rows of buildings, resulting in lower path loss along the street and higher loss perpendicular to it.

Key Differences from UMa

• Lower BS height: 10 m (vs. 25 m for UMa), meaning the BS is embedded within the urban clutter rather than above it.
• PLE (LOS): 2.1 (vs. 2.2 for UMa) — the street canyon waveguide effect slightly reduces loss along the street axis.
• PLE (NLOS): 3.53 (vs. 3.91 for UMa) — shorter inter-site distances mean fewer obstructions.
• Faster LOS probability decay: The lower BS height means buildings block the LOS path sooner. At 200 m, UMi LOS probability is only about 18%, compared to about 18% for UMa (they converge at this distance).
• Shadow fading (NLOS): 7.82 dB (vs. 6 dB for UMa) — more variation because the signal path is more sensitive to individual building geometries.

Frequency Scaling in UMi

One of the most important practical aspects of the UMi model is how path loss scales with frequency. The 20·log₁₀(f_c) term in the LOS formula means that every octave of frequency adds 6 dB of loss. The 21.3·log₁₀(f_c) term in the NLOS formula means slightly more than 6 dB per octave. To illustrate:

Frequency	UMi-LOS @ 100m	UMi-NLOS @ 100m	Δ from 3.5 GHz
0.7 GHz (n28)	74.7 dB	90.3 dB	-14 / -17 dB
2.1 GHz (n1)	83.6 dB	101.2 dB	-5 / -6 dB
3.5 GHz (n78)	88.6 dB	107.1 dB	Reference
28 GHz (n257)	106.5 dB	127.7 dB	+18 / +21 dB
39 GHz (n260)	109.4 dB	131.0 dB	+21 / +24 dB
60 GHz (unlicensed)	113.1 dB	135.4 dB	+25 / +28 dB

The 18 dB LOS increase from 3.5 GHz to 28 GHz must be compensated by antenna gain. A 3.5 GHz antenna element has an area of λ²/(4π) ≈ 5.8 cm². At 28 GHz, the element area shrinks to 0.09 cm² — but you can fit 64 times as many elements in the same physical aperture. A 64T64R massive MIMO panel at 28 GHz provides approximately 18 dBi of beamforming gain, which exactly compensates the increased free-space loss. This is the fundamental reason why massive MIMO is not optional at mmWave — it is a physical necessity.

Street Canyon Waveguide Effect

The UMi-SC model captures a phenomenon unique to urban street canyons: the street itself acts as a waveguide. Parallel rows of buildings confine the radio energy within the street, reducing path loss along the street axis compared to what would be expected from free-space spreading. This is why the UMi-SC LOS PLE (2.1) is lower than free-space (2.0) in some measurement campaigns — the walls actively help by reflecting energy back into the propagation corridor.

However, turning a corner into a perpendicular street immediately transitions the channel from LOS to NLOS, with a sudden 15–25 dB increase in path loss. This discontinuity is a major challenge for mmWave mobility: a UE moving at walking speed (5 km/h) can experience a 30 dB signal change in less than a second when rounding a corner. 5G NR beam management (SSB beam sweeping, CSI-RS based beam refinement) is specifically designed to handle these rapid channel transitions.

RMa — Rural Macro

The Rural Macro (RMa) scenario models base stations in open rural environments with large inter-site distances. Buildings are sparse and low (typically 5 m), and the terrain is relatively flat. This model supports ranges up to 21 km and is limited to frequencies up to 30 GHz — rural mmWave deployment is not yet a standard use case.

RMa is the only scenario in TR 38.901 with an explicit frequency ceiling of 30 GHz. The reasoning is straightforward: rural deployments with large cell radii and low subscriber density have no economic justification for mmWave, and no measurement campaigns have validated rural propagation at frequencies above 30 GHz.

At 700 MHz with a 35 m tower, a rural cell can maintain an NLOS path loss below 141 dB at 5 km. Given a typical rural link budget margin of 150 dB (high-power BS, low noise figure, receiver sensitivity of -110 dBm), this allows comfortable coverage with roughly 9 dB of margin for shadow fading and body loss.

RMa vs. UMa: When Rural Meets Urban

In practice, many real-world deployments blur the line between rural and urban. Suburban areas, peri-urban regions, and small towns may have characteristics of both. The key differentiators are:

• Building height: RMa assumes 5 m (1–2 storey), UMa assumes buildings are above 10 m
• Street width: RMa assumes 20 m (wider roads), UMa does not explicitly include street width
• BS height: RMa allows up to 150 m (tall lattice towers), UMa is typically 25 m (rooftop)
• Clutter density: RMa assumes sparse buildings with open fields between them

For suburban environments that fall between these extremes, many operators use UMa with a reduced shadow fading standard deviation (5 dB instead of 6 dB) as a pragmatic compromise.

RMa for 5G FWA (Fixed Wireless Access)

One of the most important applications of the RMa model is Fixed Wireless Access (FWA) planning. FWA uses 5G NR to replace or supplement wired broadband connections, particularly in rural areas where fibre deployment is economically infeasible. Key planning considerations:

• CPE (Customer Premises Equipment) height: Unlike mobile UEs at 1.5 m, FWA CPEs are mounted on rooftops or walls at 5–10 m. This significantly improves LOS probability and reduces path loss by 5–10 dB compared to mobile use.
• High-gain CPE antenna: FWA CPEs typically use directional antennas with 8–14 dBi gain, compared to 0 dBi for mobile UE. This additional gain extends the viable cell radius.
• Target throughput: FWA requires sustained 50–100 Mbps, which demands a minimum SNR of approximately 10–15 dB. This translates to a maximum path loss approximately 10 dB less than the voice coverage limit.
• Frequency selection: n28 (700 MHz) for maximum range in rural areas, n78 (3.5 GHz) for capacity in suburban FWA deployments.

This example illustrates why 5G FWA at 700 MHz is a commercially viable alternative to fibre in rural India. With Airtel’s n28 band and rooftop CPEs, a single rural macro cell can provide broadband service to a village 8 km away with data rates exceeding 100 Mbps.

RMa Limitations and Alternatives

While the RMa model serves well for flat rural terrain, India’s diverse geography includes mountains (Himalayas, Western Ghats), dense forests (Northeast, central India), and desert (Rajasthan), none of which are well-characterized by the default RMa model. For these special environments:

• Mountainous terrain: Use the ITU-R P.526 diffraction method with digital terrain elevation data (SRTM 30m or higher resolution). Mountain ridges act as knife-edge diffracting obstacles. The path loss can vary by 30+ dB depending on whether the terrain blocks or clears the Fresnel zone. RMa does not capture terrain-specific diffraction.
• Dense forest: Add ITU-R P.833 vegetation loss (0.5–2 dB/m at 3.5 GHz through forest canopy). A 100 m forest strip can add 50–200 dB at mmWave, effectively creating an impenetrable barrier. At sub-1 GHz, the same forest adds 5–15 dB — challenging but manageable.
• Desert/semi-arid: Hot desert conditions create super-refractive atmospheric layers that can cause multipath fading on long links (ducting). The RMa model assumes standard refractivity. For desert links exceeding 5 km, an additional 3–5 dB ducting margin is advisable.
• Coastal: Sea surface reflections create strong two-ray multipath. Over-water links experience periodic deep fades as the ground-reflected ray alternates between constructive and destructive interference with varying distance. RMa is not validated for over-water propagation; use the ITU-R P.1546 model instead.

RMa Path Loss at Candidate Rural Frequencies

For rural India planning, the following table compares path loss at the three most relevant frequencies across the RMa range. All values assume h_BS = 35 m, h_UT = 1.5 m, LOS condition:

Distance	700 MHz (n28)	2100 MHz (n1)	3500 MHz (n78)
1 km	91 dB	101 dB	106 dB
3 km	101 dB	111 dB	116 dB
5 km	106 dB	116 dB	121 dB
10 km	114 dB	124 dB	129 dB
15 km	119 dB	129 dB	134 dB
21 km	123 dB	133 dB	138 dB

At 700 MHz (n28), a rural cell achieves 123 dB path loss at the maximum 21 km range. With a typical rural macro link budget of 155+ dB (65 dBm EIRP + 0 dBi UE gain vs. -90 dBm sensitivity), there remains 32 dB of margin for shadow fading, body loss, and O2I penetration into rural homes. This is why n28 is the foundation of rural 5G coverage in India and globally.

InH & InF — Indoor Models

Indoor propagation behaves fundamentally differently from outdoor. Walls, floors, ceilings, furniture, and people all interact with radio waves at close range. 3GPP TR 38.901 defines two indoor scenarios: InH (Indoor Hotspot – Office) and InF (Indoor Factory), the latter added in Release 16 to support Industry 4.0 and URLLC requirements.

InH-Office

The InH-Office LOS PLE of 1.73 is remarkable — it is lower than free space! This occurs because office corridors act as waveguides, with reflections from walls, ceiling, and floor constructively reinforcing the direct signal. This phenomenon is well-documented in measurements and is physically accurate, not a model artifact.

InF — Indoor Factory

The factory environment introduces a new parameter: clutter density, which accounts for metallic machinery, storage racks, conveyor systems, and other industrial equipment. The InF model defines four sub-scenarios:

An indoor LOS path loss of only 68.8 dB at 30 m is remarkably low. For comparison, a sub-6 GHz small cell with 24 dBm EIRP and an InH AP with receive sensitivity of -100 dBm has a link budget of 124 dB — leaving over 55 dB of margin. This explains why a single Wi-Fi 6 access point or 5G NR small cell can cover an entire floor of an open-plan office.

InF NLOS Formulas by Sub-Scenario

The InF-DL sub-scenario has the highest NLOS PLE at 3.57, reflecting the severe obstruction from dense metallic machinery at low BS heights. In contrast, InF-DH has a lower PLE (2.19) because the high ceiling-mounted BS can “see over” the machinery, maintaining quasi-LOS conditions to many locations. This counterintuitive result has been confirmed by NIST measurements in automotive assembly plants and metal fabrication facilities.

O2I Penetration & Material Losses

Outdoor-to-Indoor (O2I) penetration loss is one of the most critical factors in 5G planning. Modern building materials — particularly low-emissivity (low-E) glass and reinforced concrete — create severe signal barriers at 5G frequencies. The total O2I loss is the sum of building penetration loss and indoor propagation loss.

Material Penetration Losses

Material	Loss Formula (dB)	@ 3.5 GHz	@ 28 GHz	@ 39 GHz
Standard Glass	2 + 0.2·f	2.7 dB	7.6 dB	9.8 dB
IRR Glass (Low-E)	23 + 0.3·f	24.1 dB	31.4 dB	34.7 dB
Concrete	5 + 4·f	19.0 dB	117.0 dB	161.0 dB
Wood	4.85 + 0.12·f	5.3 dB	8.2 dB	9.5 dB

Low-Loss vs. High-Loss Buildings

TR 38.901 classifies buildings into two categories:

• Low-loss buildings: Older construction with standard glass, thin walls, and wooden frames. O2I loss at 3.5 GHz is typically 10–15 dB. These buildings are well-served by outdoor macro cells.
• High-loss buildings: Modern construction with IRR glass (energy-efficient), thick concrete walls, and metal cladding. O2I loss at 3.5 GHz can exceed 25 dB, and at 28 GHz can exceed 40 dB. These buildings require dedicated indoor small cells.

O2I Loss at Key 5G NR Frequencies

To provide practical reference values for network planners, the following table summarizes total O2I building penetration losses for typical low-loss and high-loss buildings:

Frequency	Low-Loss Building	High-Loss Building	Planning Impact
700 MHz (n28)	11 dB	17 dB	Indoor coverage possible from macro
1800 MHz (n3)	14 dB	22 dB	Indoor coverage marginal from macro
2100 MHz (n1)	15 dB	24 dB	Indoor requires strong outdoor signal
3500 MHz (n78)	18 dB	30 dB	Indoor small cells often needed
28 GHz (n257)	28 dB	42+ dB	Indoor small cells mandatory
39 GHz (n260)	32 dB	48+ dB	No outdoor-to-indoor penetration

The table reveals a stark reality: at mmWave frequencies, even low-loss buildings impose nearly 30 dB of penetration loss. For a 28 GHz macro cell with 65 dBm EIRP (typical for a 64T64R massive MIMO panel), the additional 30–42 dB of O2I loss reduces the effective indoor range to just tens of metres. This is the fundamental reason why 5G NR mmWave is primarily an outdoor or dedicated indoor technology.

Indoor Distance Loss

Once the signal penetrates the building exterior, it continues to lose power as it propagates through the indoor environment. TR 38.901 models this as a simple linear loss:

The total O2I path loss is therefore: outdoor path loss (UMa or UMi to the building exterior) + building penetration loss (PL_tw) + indoor distance loss (0.5·d_in) + indoor shadow fading (4.4 dB standard deviation for low-loss, 6.5 dB for high-loss).

Complete O2I Worked Example

The Modern Building Problem

The proliferation of energy-efficient building design has created a growing challenge for mobile network operators. Modern buildings increasingly use:

• Low-E (low-emissivity) glass: Metal oxide or metallic coatings on window glass that reflect infrared radiation to improve thermal insulation. These coatings are highly reflective at radio frequencies, adding 20–35 dB of loss per pane.
• Metal-clad facades: Aluminum composite panels (ACPs) and similar metal cladding used for architectural aesthetics and weather protection. These can add 30+ dB of penetration loss.
• Reinforced concrete: Steel rebar in concrete walls creates a Faraday cage effect, blocking RF signals. The effect increases with frequency.
• Multi-pane windows: Double and triple-pane windows with metal-coated glass and argon gas filling add cumulative loss from each pane and the gas layers.

In India’s rapidly modernizing cities, new commercial buildings (IT parks, corporate offices, shopping malls) are predominantly high-loss constructions. This has led operators to deploy extensive in-building DAS (Distributed Antenna Systems) and small cell networks in premium buildings, adding significantly to CAPEX. The emergence of transparent RIS technology (Chapter 13) may eventually provide a less expensive solution by selectively allowing radio signals to pass through otherwise opaque building envelopes.

O2I Planning Strategies for Indian Operators

For Indian operators like Airtel, Jio, and Vi, the O2I challenge is particularly acute due to the rapid proliferation of modern buildings in metro and tier-1 cities. Here are practical strategies informed by the propagation models in this guide:

Floor Loss and Multi-Floor Coverage

When an indoor small cell serves users on multiple floors, inter-floor penetration loss must be considered. TR 38.901 does not explicitly define inter-floor loss, but measurements from various studies provide practical values:

Floors Penetrated	3.5 GHz Loss	28 GHz Loss	Construction Type
1 floor	15–20 dB	25–35 dB	Reinforced concrete slab
2 floors	25–30 dB	40–50 dB	Cumulative
3+ floors	30–40 dB	50+ dB	Signal effectively blocked
1 floor (wooden)	8–12 dB	15–20 dB	Wooden joist/plywood floor

The implication is clear: in modern concrete buildings, each indoor small cell effectively serves only its own floor. Multi-floor coverage from a single AP is not viable at 3.5 GHz (except in wooden/lightweight construction) and impossible at mmWave. Network planners must provision one set of indoor APs per floor in concrete buildings.

LOS Probability Models

The LOS probability function determines the likelihood that a direct, unobstructed path exists between the base station and the user at a given 2D distance. This is arguably the most impactful part of the channel model, because LOS and NLOS path loss can differ by 20–40 dB. A network planned assuming optimistic LOS conditions will dramatically under-provision coverage.

Practical Impact

Distance (d2D)	UMa PLOS	UMi PLOS	RMa PLOS
18 m	100%	100%	99.2%
50 m	63%	61%	96.1%
100 m	39%	30%	91.4%
200 m	18%	18%	82.7%
500 m	4%	3.6%	61.3%
1000 m	1.8%	1.8%	37.0%

In an urban macro deployment at 200 m, only about 18% of user locations are LOS. At 500 m, it drops to 4%. This means that in practical 5G NR coverage planning, the NLOS formula dominates the coverage prediction for the vast majority of the cell area.

InF LOS Probability

For factory environments, the LOS probability depends on the sub-scenario (clutter density and BS height). The InF models use a unique formulation based on clutter height and density:

Why LOS Probability Matters More at mmWave

The importance of LOS probability increases dramatically with frequency. At 700 MHz, diffraction around buildings is effective, so even NLOS users receive usable signal. The LOS/NLOS path loss difference at 700 MHz is about 15–20 dB. At 28 GHz, diffraction is negligible (wavelength is 10.7 mm, far smaller than building dimensions), so the LOS/NLOS difference can exceed 30–40 dB. This means:

• At 700 MHz (n28): A user in NLOS at 300 m is still within coverage. LOS probability is a secondary planning factor.
• At 3.5 GHz (n78): NLOS coverage requires careful link budget analysis. LOS probability determines the coverage confidence level.
• At 28 GHz (n257): Only LOS users have reliable coverage. NLOS users are effectively out of range. LOS probability is the primary planning metric.

Stochastic vs. Map-Based LOS

The TR 38.901 LOS probability functions are stochastic: they predict the probability of LOS at a given distance without reference to the actual building layout. In real network planning, tools like Atoll, Planet, and Ranplan use 3D building databases to determine LOS/NLOS deterministically for each pixel. The stochastic model is used for system-level simulations where per-pixel building data is not available or would be computationally prohibitive for Monte Carlo runs involving thousands of user drops.

Applying LOS Probability in System Simulations

In a 3GPP-compliant system-level simulation (such as those required for Release evaluation), LOS probability is applied as follows:

1. Drop users randomly within the simulation area according to the defined user distribution (typically uniform).
2. For each user, calculate the 2D distance to the serving BS.
3. Draw a random number U ~ Uniform(0,1). If U < P_LOS(d_2D), assign LOS condition; otherwise assign NLOS.
4. Apply the corresponding path loss formula (LOS or NLOS) to calculate the median path loss.
5. Add shadow fading as a zero-mean Gaussian with the appropriate standard deviation (LOS or NLOS).
6. Ensure spatial consistency: Nearby users should have the same LOS/NLOS status (within the correlation distance). This prevents physically impossible situations where adjacent users see different LOS conditions.

The combined effect of LOS probability and the LOS/NLOS path loss difference creates a bimodal received signal distribution at any given distance. Instead of a smooth Gaussian (as implied by a single path loss formula + shadow fading), the actual distribution has two peaks: one for LOS users (high RSRP) and one for NLOS users (low RSRP). This bimodal distribution is critical for accurate coverage and capacity predictions.

CI and ABG Research Models

Alongside the 3GPP standardized models, academic research has produced two important alternative path loss models. The Close-In (CI) model and the Alpha-Beta-Gamma (ABG) model were championed by NYU Wireless (Prof. Theodore Rappaport’s group) and have been extensively validated through measurement campaigns from 0.5 to 150 GHz.

CI (Close-In Free Space Reference Distance) Model

The CI model has a single free parameter: the PLE n. The 1-metre reference point is physically anchored to the free-space path loss at that distance, making the model inherently consistent across frequencies. This physical grounding is the CI model’s greatest strength — it cannot predict less loss than free space at 1 metre.

CIF (CI with Frequency-Dependent PLE) Model

NYU Wireless also proposed an extension called the CIF model, which adds a frequency-dependent term to the PLE:

The CIF model was proposed to capture the observation that the PLE sometimes varies slightly with frequency (e.g., NLOS PLE may increase at mmWave due to reduced diffraction). However, analysis by Rappaport et al. showed that the improvement from adding the b parameter is statistically marginal in most environments, and the simpler CI model is preferred for its parsimony and physical interpretability.

ABG (Alpha-Beta-Gamma) Model

The ABG model has three free parameters, giving it more flexibility to fit measurement data. However, this flexibility comes at a cost: the model can predict physically impossible results (e.g., less loss than free space at certain distances and frequencies) because the floating intercept β is not anchored to physics.

The CI vs. ABG Debate

The choice between CI and ABG has been one of the most debated topics in the propagation modeling community. The debate centers on a fundamental trade-off between physical interpretability and statistical fit:

• CI advocates (led by NYU Wireless) argue that the physically anchored 1m reference point ensures the model behaves correctly at all distances and frequencies, even when extrapolating beyond measurement data. The single-parameter simplicity makes it easy to compare environments and frequencies. And the PLE is physically meaningful — it describes the rate of power decay with distance.
• ABG advocates (various European groups) argue that the additional parameters allow better fit to measurement data, especially when the frequency-dependent propagation characteristics deviate from the simple log-distance relationship. The floating intercept can capture scenario-specific effects (like wall penetration in indoor environments) that the CI model lumps into the PLE.

In practice, the difference in prediction accuracy between CI and ABG is usually less than 1 dB when both are fitted to the same measurement data. The CI model is preferred for cross-frequency analysis and initial system design, while ABG is sometimes used for site-specific optimization when detailed measurement data is available. 3GPP TR 38.901 uses neither CI nor ABG directly but employs scenario-specific formulas that share structural elements with both.

CI vs. ABG vs. 3GPP: When to Use Each

Criterion	CI Model	ABG Model	3GPP TR 38.901
Free parameters	1 (PLE)	3 (α, β, γ)	Scenario-specific
Physical anchor	Yes (FSPL at 1m)	No (floating)	Varies
Cross-frequency	Inherently consistent	Can be inconsistent	Validated per scenario
Best for	Research, quick estimates	Curve-fitting to data	Standards-compliant planning
Industry adoption	Academic (NYU)	Academic	Industry standard
Freq range tested	0.5–150 GHz	0.5–150 GHz	0.5–100 GHz

NYU Wireless Measurement Campaigns

The most extensive validation of the CI model comes from NYU Wireless, led by Professor Theodore S. Rappaport. Their measurement campaigns have produced foundational datasets at frequencies from 2 GHz to 150 GHz across multiple environments:

• 28 GHz outdoor (2013): 75 TX-RX location pairs in New York City. Found PLE of 2.55 (LOS) and 5.76 (NLOS) using omnidirectional antennas, or 2.0 (LOS) and 3.4 (NLOS) using directional best-beam measurements.
• 73 GHz outdoor (2014): 36 TX-RX pairs. PLE of 2.0 (LOS) and 3.2 (NLOS) with directional antennas. Confirmed that mmWave propagation is viable for cellular when directional antennas are used.
• 28/73 GHz indoor (2014–2015): Office and shopping mall environments. Found PLE of 1.6 (LOS, office) confirming the below-free-space indoor waveguide effect.
• 140 GHz outdoor (2019–2021): Pioneer sub-THz measurements in downtown Brooklyn. PLE of 2.0 (LOS) and 3.5 (NLOS). Molecular absorption adds only 0.25 dB at 100 m.

The CI model’s key finding: the PLE is remarkably consistent across frequencies for a given environment. Urban LOS PLE is approximately 2.0–2.1 from 2 GHz to 140 GHz. This frequency independence of the PLE (when FSPL is properly anchored at 1 m) is the strongest argument for the CI model’s physical basis.

Multi-Frequency Measurement Analysis

NYU Wireless published a landmark analysis in 2017 comparing PLEs across frequencies at the same locations in New York City. The results demonstrated the frequency-invariance of the CI PLE:

Frequency	UMi LOS PLE	UMi NLOS PLE	σSF (LOS)	σSF (NLOS)
2.5 GHz	2.07	3.30	3.6 dB	8.0 dB
5.8 GHz	1.98	3.22	3.2 dB	8.3 dB
10 GHz	2.01	3.25	3.4 dB	8.1 dB
28 GHz	1.98	3.19	3.1 dB	8.2 dB
38 GHz	2.02	3.26	3.5 dB	8.5 dB
73 GHz	2.00	3.21	3.0 dB	7.9 dB
140 GHz	2.02	3.45	2.8 dB	9.1 dB

The LOS PLE varies from 1.98 to 2.07 across frequencies spanning 2.5 to 140 GHz — a remarkable consistency that supports the CI model’s physical foundation. The NLOS PLE shows slightly more variation (3.19–3.45), with a slight increase at 140 GHz likely due to reduced diffraction at shorter wavelengths.

This table also reveals an important practical insight: shadow fading is remarkably stable across frequencies, ranging from 2.8–3.6 dB in LOS and 7.9–9.1 dB in NLOS. This means that the randomness of propagation does not change significantly with frequency — only the deterministic path loss changes. Network planning margins for shadow fading can therefore be applied consistently across frequency bands.

Typical PLE Values (CI Model)

Environment	PLE (LOS)	PLE (NLOS)	σSF LOS	σSF NLOS
Urban Micro	1.98	3.19	3.1 dB	8.2 dB
Urban Macro	2.00	3.52	4.0 dB	7.8 dB
Indoor Office	1.60	3.08	2.4 dB	8.6 dB
Indoor Factory	2.01	3.28	3.5 dB	7.4 dB
Rural	2.16	3.07	2.7 dB	6.7 dB

The Sub-THz Challenge (100–300 GHz)

The journey to 6G pushes wireless communications into the sub-terahertz (sub-THz) spectrum: 100–300 GHz. At these frequencies, a new physical phenomenon dominates the propagation channel — molecular absorption. Water vapor (H₂O) and oxygen (O₂) molecules resonate at specific frequencies, absorbing electromagnetic energy and converting it to heat. This absorption is negligible below 100 GHz but becomes a defining factor above it.

Molecular Absorption

The absorption coefficient k(f) determines how much additional loss per unit distance the atmosphere introduces at frequency f. At sea level with moderate humidity (50% RH, 20°C), the major absorption peaks are:

Between these peaks, transmission windows exist where absorption is relatively low:

• ~140 GHz window: k ≈ 0.002–0.01 dB/m. The most promising 6G band with 10+ GHz bandwidth. Candidate for 3GPP FR3.
• ~220 GHz window: k ≈ 0.01–0.05 dB/m. Wider bandwidth but higher absorption. Suitable for short-range high-rate.
• ~300 GHz window: k ≈ 0.02–0.08 dB/m. Extreme bandwidth (20+ GHz) but limited to very short distances.

The critical insight is that absorption loss scales linearly with distance (in dB). At 140 GHz with k = 0.005 dB/m, a 100-metre link adds only 0.5 dB of absorption. But at 183 GHz (peak), k can exceed 10 dB/m, making communication impossible beyond a few metres. System designers must carefully select frequencies within transmission windows.

Humidity and Temperature Dependence

Unlike terrestrial propagation models where environmental parameters are absorbed into statistical shadowing, sub-THz absorption is highly sensitive to atmospheric conditions:

• Humidity: Water vapour is the dominant absorber. At 140 GHz, absorption increases from ~0.003 dB/m at 20% RH to ~0.01 dB/m at 80% RH. Tropical climates face 2–3 times more absorption than arid regions.
• Temperature: Higher temperatures increase water vapour capacity of air. A 10°C increase at constant relative humidity roughly doubles the absolute humidity and correspondingly increases absorption.
• Altitude: Lower atmospheric pressure at altitude reduces molecular density, decreasing absorption. A hilltop deployment at 2000 m altitude sees roughly 20% less absorption than sea level.
• Rain: Raindrop scattering becomes significant above 100 GHz. At 200 GHz in heavy rain (50 mm/hr), rain attenuation can reach 20+ dB/km — comparable to or exceeding molecular absorption.

These dependencies mean that 6G sub-THz systems will need real-time atmospheric sensing to adapt their link budgets. A system designed for a sunny day in Delhi will not perform the same way during monsoon season. This represents a fundamental departure from current 5G NR systems, which treat atmospheric conditions as a static margin.

The HITRAN Database

The absorption coefficient k(f) is not a simple formula — it is computed from the HITRAN (High-Resolution Transmission Molecular Absorption) spectroscopic database, maintained by Harvard-Smithsonian Center for Astrophysics. HITRAN contains precise line-by-line absorption parameters for over 50 molecular species across the electromagnetic spectrum. For 6G channel modeling, the relevant species are H₂O (water vapour) and O₂ (molecular oxygen). The ITU-R Recommendation P.676 provides a simplified computational method based on HITRAN data for frequencies up to 1000 GHz.

Candidate 6G Frequency Bands

Based on the absorption spectrum analysis, several candidate frequency bands have emerged for 6G standardization. The selection balances available bandwidth against atmospheric loss:

Band	Frequency Range	Available BW	Absorption @ 100m	Target Use Case
D-Band (W1)	130–175 GHz	~45 GHz	0.3–0.5 dB	Urban outdoor, backhaul, access
G-Band (W2)	200–255 GHz	~55 GHz	1–5 dB	Short-range outdoor, indoor
W3	275–325 GHz	~50 GHz	2–8 dB	Indoor hotspot, kiosk
Upper mid-band	7–24 GHz	Variable	Negligible	Coverage + capacity balance

The D-Band (130–175 GHz) is widely considered the most promising 6G frequency range. It offers enormous bandwidth (45+ GHz, enabling peak data rates exceeding 100 Gbps) while keeping atmospheric absorption manageable (less than 0.5 dB at 100 m in the window around 140 GHz). WRC-27 agenda item 1.7 is expected to consider identification of bands above 100 GHz for IMT.

Link Budget Reality Check at Sub-THz

To understand whether sub-THz communication is actually viable, let us construct a complete link budget at 140 GHz and compare it to a 28 GHz mmWave link:

This link budget shows that sub-THz communication is not inherently impossible — the additional free-space loss (~14 dB from 28 to 140 GHz) is partially compensated by higher UE antenna gain (more elements fit in the same physical area at shorter wavelengths) and much wider bandwidth. The key technology challenges are the PA power (currently limited to ~10 dBm at 140 GHz in CMOS) and the noise figure (higher at sub-THz due to transistor limitations).

THz Channel Measurements & Proposed Models

The race to characterize the sub-THz channel has produced a burst of measurement campaigns and model proposals from leading research institutions worldwide. While no single model has achieved the consensus status of TR 38.901, several frameworks are converging toward what will become the 6G channel standard.

State of the Art: Where We Stand in 2026

As of early 2026, sub-THz channel characterization is progressing rapidly but remains far behind the maturity of TR 38.901. Key milestones:

• Measurement campaigns: Over 30 institutions worldwide have published sub-THz measurement results. Coverage ranges from indoor office to outdoor urban to industrial environments, but most campaigns are limited to LOS or simple NLOS geometries with small numbers of TX-RX positions.
• Channel sounders: Custom channel sounders at 140 GHz are now available from Keysight, Rohde & Schwarz, and several university groups. Bandwidth capabilities range from 2 GHz to 20 GHz, enabling sub-nanosecond delay resolution.
• Model proposals: At least 5 competing model frameworks exist (CI-based, ABG-based, ray-tracing hybrid, stochastic geometry, and neural network). No consensus has emerged on the optimal structure.
• Standardization: 3GPP Rel-20 (2026–2027) is expected to include a study item on channel models above 100 GHz. ITU-R WP 3K/3M is collecting national contributions. IEEE 802.15.3d has already standardized a model for 252–325 GHz.

Key Research Efforts

Sub-THz Multipath: Fewer but Specular

One of the most significant differences between sub-6 GHz and sub-THz channels is the nature of multipath. At sub-6 GHz, dense scattering creates rich multipath with many clusters spanning wide angular ranges. At sub-THz, scattering from rough surfaces is suppressed (surface features must be comparable to the wavelength to scatter effectively), leaving only specular reflections from smooth surfaces (glass, metal, polished stone). The result is a channel with:

• Fewer multipath clusters: Typically 3–8 clusters at 140 GHz vs. 10–25 at 3.5 GHz
• Narrower angular spread: Clusters are tightly focused in angular domain
• Sparser power delay profile: Fewer resolvable delay taps, easier equalization
• Higher Rician K-factor: In LOS, the direct path dominates much more strongly

This sparsity is actually advantageous for massive MIMO beamforming: with fewer dominant paths, beam alignment is simpler and beamforming gain is higher. However, it means that beam diversity (using multiple reflected paths as backup beams) is more limited, increasing the impact of blockage events.

Human Body Blockage at Sub-THz

Human body blockage is a critical concern at sub-THz frequencies. At 140 GHz, the wavelength (2.1 mm) is far smaller than the human body, making diffraction negligible. A single person walking through the beam path can cause 15–30 dB of sudden signal loss lasting 200–500 ms. In crowded environments (concert halls, stadiums, dense urban sidewalks), simultaneous multi-person blockage is a realistic scenario.

NYU Wireless measurements at 140 GHz with a single human blocker found average blockage loss of 20 dB with a 350 ms average duration. IEEE 802.11ay at 60 GHz uses beam tracking and multi-beam redundancy to mitigate blockage. Future 6G sub-THz systems will likely require even more aggressive multi-connectivity and RIS-assisted bypass paths to maintain the ultra-reliable links demanded by 6G URLLC use cases.