Frequency Range 3 (FR3) in 5G/6G Systems
- FR3 is defined as the upper mid-band between sub-6 GHz and mmWave bands, offering contiguous hundreds-of-MHz bandwidth and moderate propagation losses.
- FR3 supports both coverage-centric and high-resolution sensing services for 5G/6G networks by enabling dense beamforming and efficient multi-band processing.
- Research in FR3 addresses large aperture arrays, near-field channel effects, and spectrum-sharing strategies to optimize capacity and manage interference.
Frequency Range 3 (FR3) is the upper mid-band studied for advanced 5G evolution and 6G systems, positioned between sub-6 GHz Frequency Range 1 (FR1) and millimeter-wave Frequency Range 2 (FR2). In the cited literature, FR3 is most commonly defined as approximately GHz or GHz, although some channel-modeling and measurement works discuss GHz implementations or subranges such as GHz, GHz, and GHz in order to capture emerging 6G operating regimes and standardization extensions (Bazzi et al., 25 Feb 2025, Tang et al., 7 Feb 2026, Abbasi et al., 2024). FR3 is treated as a coverage-capacity compromise: it offers contiguous bandwidths from hundreds of MHz to GHz per sub-band, supports dense beamforming and high-resolution sensing, and exhibits propagation that is materially more favorable than FR2 while being less robust than FR1 (Baduge et al., 23 Jun 2025, Xu et al., 22 Jun 2025).
1. Definition, spectrum position, and 6G role
FR3 is formalized in the surveyed works as the “upper mid-band,” filling the spectral gap between legacy FR1 and FR2. Several papers use GHz or GHz as the operative range, while the 3GPP Rel-19 channel-modeling tutorial discusses a GHz scope for simulation and implementation studies (Bazzi et al., 25 Feb 2025, Tang et al., 7 Feb 2026). This definitional variation does not alter the main systems view: FR3 is consistently treated as the band where multi-hundred-MHz carriers, massive MIMO apertures, and moderate propagation losses can coexist in a single deployment regime.
The importance of FR3 in 6G derives from that intermediate position. The band is repeatedly characterized as offering larger bandwidth than FR1, but less severe path loss, blockage, and penetration impairment than FR2. In vehicular, urban, and integrated sensing and communication (ISAC) studies, this is presented as the reason FR3 can support both coverage-oriented and resolution-oriented services, including eMBB, positioning, environment sensing, and beam-aware mobility management (Aghaei et al., 5 Apr 2026, Baduge et al., 23 Jun 2025).
Candidate non-contiguous allocations further shape the FR3 research agenda. One multiband sensing study lists typical 6G candidate segments 0 GHz, 1 GHz, 2 GHz, 3 GHz, and 4 GHz, illustrating that FR3 is not only broad but operationally fragmented by incumbent services and regulatory constraints (Pegoraro et al., 4 Oct 2025). That fragmentation is central to later work on multiband aggregation, sensing artifacts, and coexistence.
2. Propagation characteristics and measured channel behavior
A basic FR3 propagation reference used across the literature is free-space path loss,
5
supplemented in practice by shadow fading, clutter, diffraction, foliage, and penetration terms (Sung et al., 8 Nov 2025, Bazzi et al., 25 Feb 2025). Surveyed close-in and scenario-specific models show that FR3 path-loss exponents remain substantially below the severe attenuation of FR2, but increase with frequency and with LoS-to-NLoS transition. For example, an overview paper reports urban microcell LoS exponents of 6 at 7 GHz and 8 at 9 GHz, with NLoS 0; an equal-aperture measurement campaign reports 1 and 2 for 3 and 4 GHz LoS, and 5 and 6 for NLoS (Bazzi et al., 25 Feb 2025, Liu et al., 17 Apr 2026).
Material and atmospheric losses rise with frequency across FR3. One survey reports insulating-glass loss of approximately 7 dB at 8 GHz, 9 dB at 0 GHz, and more than 1 dB at 2 GHz per single pane; heavy-rain attenuation is given as 3 dB/km at 4 GHz and 5 dB/km at 6 GHz, while foliage loss increases from 7 dB/100 m at 8 GHz to 9 dB/100 m at 0 GHz (Bazzi et al., 25 Feb 2025). These numbers explain why FR3 is often described as a compromise rather than a uniformly favorable extension of FR1.
Delay and angular dispersion in FR3 are frequency dependent but scenario dependent as well. Measurement-based studies at 1 and 2 GHz report that RMS delay spread remains nearly constant in LoS but decreases in NLoS when moving upward in frequency, with one campaign giving NLoS means of about 3 ns at 4 GHz and 5 ns at 6 GHz, alongside decreases in azimuthal and elevation spreads (Liu et al., 10 Jun 2026). Another equal-aperture study reports RMS delay spreads of about 7 ns and 8 ns in LoS, and 9 ns and 0 ns in NLoS, for 1 and 2 GHz respectively, together with a collapse of azimuth spread from about 3 to 4 in LoS (Liu et al., 17 Apr 2026). By contrast, ultra-wideband double-directional measurements over 5 GHz in an urban microcell vegetation-blocked street canyon report omnidirectional mean delay spread around 6 ns and modest angular spreads near 7 rad, emphasizing that directional soundings and street-canyon geometry can yield much more compact channels (Abbasi et al., 2024).
A common misconception is that higher frequency in FR3 simply produces uniformly worse channels. The measured literature is more specific. Higher FR3 sub-bands often suppress weaker late multipath and narrow the angular domain, which can improve dominant singular modes and raise MIMO capacity in some settings, even when coverage margins decline (Liu et al., 10 Jun 2026, Liu et al., 17 Apr 2026).
3. Large-aperture arrays, near-field propagation, and standardized channel models
FR3 is unusually important for extremely large aperture arrays because wavelength is short enough to permit dense apertures, yet link distances remain in the tens to hundreds of meters. In this regime the Rayleigh distance,
8
can become comparable to ordinary cell ranges. One FR3 ISAC survey notes that for an aperture of about 9 m at 0 GHz, 1 m, and with larger ELAAs it can extend to several hundreds of meters (Baduge et al., 23 Jun 2025). This invalidates the planar-wave and spatial-stationarity assumptions that underpinned earlier sub-6 GHz channel models.
The 3GPP-adopted XL-MIMO framework addresses this by replacing far-field phase tapers with element-wise spherical-wave distances and by modeling spatial non-stationarity (SNS) through visibility-region or blocker-based attenuation factors. In that framework, nonlinear phase curvature, element-dependent angles, and element-dependent delays appear explicitly, while incomplete scattering and blockage illuminate only subregions of the array (Xu et al., 22 Jun 2025). Simulation results reported for that model show near-field capacity gains over far-field modeling of 2 bps/Hz in urban micro at radii 3 m and 4 bps/Hz in indoor hotspot at 5 m; SNS also increases average coupling loss by about 6 dB in UMi and 7 dB in InH relative to spatially stationary modeling (Xu et al., 22 Jun 2025).
These ideas are carried into 3GPP Release 19. The Rel-19 FR3 tutorial describes four major upgrades to the legacy TR 38.901 framework: a new Suburban Macro scenario, explicit spherical-wave near-field modeling for ELAA, SNS across large apertures, and recalibrated large-scale parameters from new 8 GHz measurements (Tang et al., 7 Feb 2026). The same tutorial emphasizes that Rel-19 uses a unified spherical-wave formulation rather than an explicit near-/far-field switch, which is important because FR3 deployments may continuously cross that boundary as aperture, carrier frequency, and link range vary.
Measurement campaigns at 9 GHz corroborate the model shift. A UMi ultra-massive MIMO study reports that near-field LoS phase curvature over a 128-element transmit array is better fit by a quadratic term than by a linear one, with RMSE 0 rad versus 1 rad, and that channel correlation, SNS, and hardening depend strongly on whether the link is near-field LoS, foliage-shaded, or far-field street canyon (Shi et al., 9 Apr 2026). This suggests that FR3 should not be treated as a simple frequency translation of FR1 macro-cell models.
4. Coverage, capacity, architecture, and network planning
Across comparative studies, FR3 is not uniformly dominant; it is advantageous because its weaknesses and strengths are balanced. Under equal physical aperture, ray-tracing results for high-rise urban vehicle-to-base-station links show that FR3 yields higher SNR than 2 GHz in interference-free cell-edge conditions and higher SINR for cell-edge users under full interference, even though mmWave arrays contain more elements. The cited explanation is that mmWave array gain does not fully compensate for the severe path loss experienced at the cell edge (Aghaei et al., 5 Apr 2026). A broader FR1/FR3/FR2 study reports high-rise urban cell-edge rates of 3 Gbps at 4 GHz and 5 Gbps at 6 GHz in the interference-free case, versus 7 Gbps at 8 GHz; under full interference the corresponding cell-edge rates are 9, 0, and 1 Gbps, respectively (Aghaei et al., 8 Apr 2026).
Equal-aperture measurements sharpen the trade-off. At fixed physical area, 2 GHz can host four times as many elements as 3 GHz, and one campaign instantiated this as 4 and 5. Even with that aperture-driven gain, the 6 GHz band retains a residual cell-edge deficit of about 7 dB relative to 8 GHz, but it achieves higher median spectral efficiency under equal aperture, 9 versus 0 bit/s/Hz (Liu et al., 17 Apr 2026). The same study finds that, for fixed element count, array topology differences such as 1, 2, and 3 have negligible impact on coverage and spectral efficiency in its UMa setting.
Architectural work therefore focuses on how to exploit FR3 without overpaying in RF power. One study distinguishes frequency-integrated (FI) and frequency-partitioned (FP) architectures, each with fully digital (FD) and hybrid analog-digital (HAD) realizations, and shared-RF-chain (SRF) or dedicated-RF-chain (DRF) variants. In the evaluated uplink multi-user setup, FI yields more than 4 the sum rate of FP, but at a 5 dB RF-front-end power penalty; SRF and DRF provide comparable spectral efficiency, while SRF consumes about half the static power of DRF; FD provides about 6 higher spectral efficiency than HAD, but with 7 higher power consumption (Mizmizi et al., 3 Mar 2025). A separate parametric base-station power model for FR3 concludes that with 8 antennas at 9 load, the power amplifier dominates when 00 or fewer RF chains are used, whereas digital and analog processing dominate when 01 or more RF chains are used; in that model, hybrid beamforming reaches 02 Gbit/s per user in downlink and improves energy efficiency by 03 relative to fully digital beamforming (Peschiera et al., 12 Oct 2025).
Planning and measurement overhead are also active topics because exhaustive FR3 ray tracing is expensive and measurement gaps reduce throughput. The CommUNext framework addresses cross-band FR3 coverage prediction by using low-frequency maps and sparse target-band observations. Its authors report that generating about 04 single-band, single-beam maps at 05 GHz with NVIDIA Sionna on an RTX 2080 Ti can take about one month of GPU time, motivating data-driven substitutes; for full 06 GHz directional prediction, Full CommUNext with segmentation reaches median 07 dB and 08 dB, while Partial CommUNext with block coverage sampling reaches 09 dB and 10 dB (Sung et al., 8 Nov 2025).
5. Integrated sensing, multiband processing, and FR3-specific ISAC issues
FR3 is central to ISAC because bandwidth directly determines ranging resolution,
11
and FR3 makes 12 MHz to 13 GHz-and-beyond sensing bandwidths practically available without moving into FR2 (Bazzi et al., 25 Feb 2025, Baduge et al., 23 Jun 2025). Combined with large arrays, this enables fine delay and angular estimation, multi-target sensing, and near-field range-angle coupling.
A representative FR3 sensing framework is ADMM-assisted compressed multiband sensing (ADMM-CMS), proposed for joint delay-angle recovery across multiple FR3 sub-bands using uplink QAM-modulated pilots. The method models each sub-band with Kronecker dictionaries, stacks per-band sparse coefficients into a jointly sparse matrix, and solves a mixed 14-regularized problem with adaptive ADMM. In simulations over 15 and 16 GHz sub-bands with 17 antennas and 18 pilots, ADMM-CMS attains a 19 dB gain in per-antenna transmit power over Bartlett-type beamforming for achieving 20 successful recovery probability, and at 21 dBm per-antenna transmit power it reduces delay RMSE by 22 and 23 relative to performing compressed sensing separately on the constituent 24 and 25 GHz sub-bands (Wang et al., 3 Oct 2025).
A second line of work emphasizes that FR3 sensing is not only wideband but frequency selective in its clutter structure. A DMC-aware multiband estimator models specular components jointly with background dense multipath components and studies Cramér-Rao bounds across 26 and 27 GHz. The reported result is that lower sub-bands can enjoy stronger specular gains but also stronger and more spread DMC, whereas higher sub-bands may become preferable at high SNR because DMC decays faster. In the representative case, multiband estimation reduces delay RMSE by 28 relative to 29 GHz-only sensing and by 30 relative to 31 GHz-only sensing, while also reducing false alarms by up to 32 in the CRB-achieving regime (Wang et al., 26 Feb 2026). This directly contradicts the simplistic view that “higher frequency always gives worse sensing.”
FR3 multiband sensing is additionally complicated by frequency anisotropy and non-contiguous allocations. A coherent ranging study over 33 GHz introduces phase-coherence metrics such as MPC and shows that extended targets, especially humans, may lose coherence as aggregated bandwidth grows. The same work proposes the Subsets-Product Backprojection (SPBP) algorithm to suppress grating lobes induced by non-contiguous sub-bands, reporting more than 34 dB grating-lobe suppression in moderate-gap cases and improved multitarget OSPA relative to direct backprojection and OMP (Pegoraro et al., 4 Oct 2025). An earlier experimental study at 35 and 36 GHz similarly found that a lower FR3 band can reveal additional target-induced multipath while the higher band exhibits stronger blockage, implying that FR3 sensing may need to detect both emergent and missing paths depending on sub-band (Bomfin et al., 2024).
From a systems viewpoint, later work pushes sensing beyond dedicated pilots. One FR3 ISAC perspective argues that sensing cannot rely solely on pilot resources, discusses payload-aided sensing, and proposes hierarchical beam alignment in which coarse access occurs in lower FR3 and refinement in upper FR3, while also identifying intra- and inter-beam squint as specific wideband FR3 design issues (Bazzi et al., 18 May 2026).
6. Coexistence with incumbents and spectrum-sharing constraints
FR3 is not a greenfield band. The surveyed literature identifies satellite downlinks, Earth exploration satellite systems, radio astronomy, radiolocation, and military radars as incumbent services already present in substantial portions of the band (Bazzi et al., 25 Feb 2025, Testolina et al., 11 Jun 2026). This makes coexistence a first-order design variable rather than a secondary deployment detail.
The interference mechanisms are not limited to direct main-beam leakage. A Boston digital-twin study using Sionna ray tracing finds that sidelobes and NLoS paths significantly contribute to radio-frequency interference toward satellites. In that model, aggregated path gain at the satellite input peaks around 37 dB for a 38 GHz omni gNB configuration, 39 dB for a 40 GHz 4-element case, 41 dB for a 42 GHz 16-element case, and 43 dB for a 44 GHz 36-element case, with reflections peaking around 45 elevation and sometimes exceeding LoS when the LoS is nulled by the beam pattern (Testolina et al., 11 Jun 2026). The conclusion is explicit: directionality alone is insufficient to guarantee incumbent protection.
Mitigation strategies therefore combine beam-space and power-space control. A QoS-aware spectrum-sharing study formulates terrestrial utility maximization under rate, INR, and power constraints, and pairs it with regularized spatial nulling. Its numerical results show that nulling alone can achieve median INR around 46 dB but creates severe user-RSS outliers and poor fairness, while standalone power control keeps worst-case RSS loss below 47 dB and achieves near-perfect Jain’s fairness index. Their joint application, for example with 48 and 49, achieves INR median 50 dB, worst-case RSS loss below 51 dB, and gNB power savings of up to 52 in some settings (Tsampazi et al., 23 Mar 2026). Earlier overview work had already identified null-steering conditions of the form 53 as a practical coexistence mechanism for satellite protection (Bazzi et al., 25 Feb 2025).
The planning implication is that FR3 operation depends jointly on beamforming, site geometry, activation patterns, and regulation. This suggests that spectrum sharing in FR3 is fundamentally spatial and dynamic: the same band can be usable or unusable depending on elevation angle, building clutter, sidelobe structure, and whether the network can exploit natural blockage, nulling, or adaptive power control (Testolina et al., 11 Jun 2026, Tsampazi et al., 23 Mar 2026).
