3GPP Release 15: 5G NR — A New Radio From Scratch
Release 15 is the most consequential mobile standard since Rel-8. It defines both 5G New Radio — a completely new air interface — and the 5G Core — a reimagined network built on web-scale software principles. Unlike the LTE era where successive releases evolved a single architecture, 5G was designed from the start with a flexibility principle: one standard, one protocol stack, spanning every conceivable use case from 700 MHz rural coverage to 28 GHz millimetre-wave hotspots.
Overview
3GPP began the 5G standardisation process in 2015, driven by ITU-R's IMT-2020 requirements: at least 20 Gbps peak downlink, 10 ms end-to-end latency for eMBB, and 1 ms for URLLC. LTE-Advanced Pro could meet none of these targets even with every enhancement stacked on top. A new air interface was the only path forward.
Rel-15 was delivered in two drops: the first NSA specification was frozen in December 2017 to allow rapid commercial deployment; the full SA specification froze in June 2018. The result is a standard that simultaneously covers three service categories that had never before been addressed by a single cellular technology: eMBB (enhanced Mobile Broadband), URLLC (Ultra-Reliable Low-Latency Communications), and mMTC (massive Machine-Type Communications). The air interface and core network were redesigned together so that the same physical infrastructure could serve all three — separated logically through network slicing.
5G NR — Flexible Numerology
The single most important design decision in NR is flexible numerology. Instead of fixing subcarrier spacing at 15 kHz as LTE does, NR defines a scalable set — 15, 30, 60, 120, and 240 kHz — referred to as µ = 0, 1, 2, 3, 4. The relationship is simple: subcarrier spacing doubles with each step. Wider spacing means a shorter OFDM symbol duration, a shorter slot, and lower scheduling latency. This one design choice lets the same NR specification serve two fundamentally different frequency ranges:
Uses µ=0 (15 kHz) or µ=1 (30 kHz) subcarrier spacing. Channel bandwidths up to 100 MHz. Covers the traditional macro and small-cell deployment bands: 700 MHz rural coverage, 2.1 GHz urban, 3.5 GHz mid-band. FR1 is where the vast majority of the world's 5G subscribers connect today. The 3.5 GHz n78 band is the most widely deployed globally, offering a good balance between coverage and capacity.
Uses µ=3 (120 kHz) spacing. Channel bandwidths up to 400 MHz; up to 800 MHz with carrier aggregation. This is millimetre-wave territory: wavelengths of 5–12 mm, antenna arrays with 64, 128, or 256 elements fitting within a hand-sized panel. Deployed in dense urban hotspots — stadiums, train stations, convention centres — where massive bandwidth compensates for limited range. The physics are entirely different from FR1, yet the same NR standard drives both.
Beam Management at mmWave
At mmWave frequencies, signals do not diffract around obstacles the way sub-6 GHz signals do. Line-of-sight dominates — a body blocking the signal path can cause a 20 dB drop. The solution is massive beamforming: concentrating transmit energy into a narrow beam aimed precisely at the receiving device. NR defines a hierarchical beam management procedure with three phases:
- P1 — Beam sweeping: the gNB periodically sweeps SSB (Synchronisation Signal Block) beams across a wide angular range. The UE listens and reports which SSB beam index produced the strongest received signal. P1 gives a coarse best-beam estimate across the full angular space.
- P2 — Beam refinement: having identified the P1 winner, the gNB sweeps finer CSI-RS beams around that angular region. The UE reports the best narrow-beam index. P2 narrows to the optimal beam without re-scanning the full space.
- P3 — Beam tracking: within an ongoing transmission, the UE's preferred beam is monitored slot-by-slot. As the UE or surrounding objects move, the beam is adjusted within a slot without triggering a full P1/P2 sweep again.
Because each beam carries data to a single UE in a specific direction, the gNB can point multiple independent beams at different UEs simultaneously — spatial reuse that multiplies effective cell capacity proportionally to the number of simultaneous beams.
NSA and SA Deployment Options
Operators faced a practical dilemma in 2018: a full 5G deployment requires both a new radio access network and a new core network. Rel-15 defined two paths to manage this transition commercially:
The UE maintains dual connectivity to an LTE eNB (the master, anchoring the control plane to the existing EPC) and an NR gNB (the secondary, adding user-plane capacity). The 4G Evolved Packet Core handles all core functions — no 5G Core is needed. Operators could launch "5G" NR radio relatively quickly, reusing their existing 4G infrastructure. Limitation: maximum latency and slice capability are bounded by the EPC. All the first wave of commercial 5G deployments in 2019 used NSA Option 3x.
The UE connects only to an NR gNB, which connects directly to the 5G Core via the NG interface. No LTE anchor. Full 5G capabilities are available: network slicing, all 5GC network functions, sub-millisecond latency on the air interface, and native support for URLLC services. SA requires deploying the 5GC first — a significant capital investment — but unlocks everything Rel-15 designed the 5GC to deliver.
5G Core — Service-Based Architecture
The 5GC replaced 4G's point-to-point interfaces (S11, S1-MME, S6a, and a dozen others) with a Service-Based Architecture (SBA). Every network function exposes its capabilities as HTTP/2 REST API services on an internal service bus. Any authorised NF can call any other — there are no fixed bilateral interface definitions. Key network functions:
The UE's control-plane anchor. Handles registration, authentication (delegating credential checking to the AUSF), mobility management, and handover signalling. Functionally equivalent to the 4G MME but decoupled from session management.
Manages PDU sessions: IP address allocation, QoS flow establishment, and data path selection. Controls the UPF via the N4 interface. Combining parts of the old MME and P-GW into one logical function removes the session state duplication that existed in 4G.
The data-plane forwarding element. Routes user IP packets, applies QoS marking, enforces per-flow policing, and acts as the session anchor for continuity during handovers. The UPF is the only NF that touches actual user data. Multiple UPFs can be chained for local breakout or edge computing scenarios.
A service registry. When an NF starts up, it registers its capabilities with the NRF. Any other NF that needs to call a service queries the NRF first to discover which instances are available. This eliminates hard-coded network topology — NFs can be added, removed, or scaled horizontally without reconfiguring everything else.
Network Slicing
For the first time in cellular history, the network itself can be partitioned into multiple isolated logical networks — slices — running on shared physical infrastructure. Each slice is identified by an S-NSSAI (Single Network Slice Selection Assistance Information), a compact identifier the UE includes in its registration request.
A UE can simultaneously connect to multiple slices — for example, one eMBB slice for consumer broadband and one private slice for corporate VPN traffic with strict QoS guarantees. The NSSF selects the appropriate slice based on the S-NSSAI and the UE's subscription; the AMF registers the UE to it; the SMF and UPF instances serving that slice are dedicated to its QoS policy. Key slice use cases:
- Factory automation slice: dedicated SMF and UPF with sub-millisecond latency configuration and a maximum bit-rate floor for machine control traffic — guaranteed even when the shared consumer network is saturated.
- Public eMBB slice: optimised for throughput and wide coverage, accepting higher latency in exchange for maximum spectral efficiency.
- Emergency services slice: available with priority access during disaster events, overriding normal admission control so first-responder communications are never dropped due to congestion.
Why Rel-15 Mattered
- Flexible numerology was the key that bridged 700 MHz macro to 28 GHz mmWave — one standard, one chipset family, one protocol stack covering a 40:1 frequency range. No previous cellular generation achieved this.
- Network slicing enabled the industrial 5G use case that commercially separates 5G from LTE — private network contracts with factories, logistics operators, and mining companies became possible because a guaranteed slice could be sold alongside the consumer service on the same hardware.
- NSA deployment let operators launch 5G quickly while SA deployment was engineered — a deliberate two-phase commercial strategy that allowed revenue from 5G to help fund 5GC investment.
- The SBA architecture enables cloud-native network functions running as microservices in Kubernetes containers. Every major operator's 5GC is now containerised — a direct consequence of the HTTP/2 REST API design in the SBA.
- Beam management at mmWave made FR2 commercially viable — the P1/P2/P3 hierarchy handles the most challenging aspect of mmWave (beam pointing and tracking) within the standard protocol, so any conformant gNB and UE interoperate.