3GPP Release 19: 5G-Advanced Phase 2 — Ambient IoT and 6G Foundations
Release 19 completes the 5G-Advanced picture while simultaneously laying the groundwork for 6G. Its most novel feature — Ambient IoT — represents a fundamental paradigm shift: devices that operate without any battery, harvesting energy from ambient RF and communicating by reflecting rather than transmitting. Beyond Ambient IoT, Rel-19 deepens AI/ML integration, adds inter-satellite links to the NTN framework, and formally opens 6G pre-standardisation study items feeding into ITU-R IMT-2030.
Overview
3GPP Rel-19 represents a deliberate dual focus. On one side, it extends the 5G-Advanced enhancements that operators need to deploy now — better AI/ML tooling, improved satellite connectivity, and more capable V2X. On the other, it opens study items that will feed directly into the Rel-21 timeline targeting 6G commercial deployments between 2028 and 2030. The two tracks are intentionally parallel: standardisation of 6G cannot wait until 5G-Advanced is fully deployed.
Ambient IoT is the most consequential new technology in Rel-19. The cellular industry has spent twenty years trying to reduce the cost and power consumption of connected devices — from UMTS modules at $50 to NB-IoT at $2. Ambient IoT removes the floor entirely: a device with no battery and no transmitter costs a fraction of a cent in materials and lasts indefinitely. If the technical challenges are solved, the addressable market grows from billions of devices to trillions.
Ambient IoT — Battery-Free Devices
An Ambient IoT device — called a tag in the Rel-19 specification — has no battery and no active radio transmitter. It operates through two physical mechanisms working together:
The tag captures energy from ambient RF signals using a rectenna— a rectifying antenna circuit that converts the incoming RF wave to direct current. The RF source can be a nearby gNB, a Wi-Fi access point, or a dedicated Ambient IoT reader transmitting a continuous-wave carrier at a known frequency. Harvested power is typically tens to hundreds of microwatts — enough to power a low-power sensor IC, a small memory, and the modulator circuit. The tag accumulates charge in a micro-capacitor and transmits in short bursts when sufficient energy has been stored.
Instead of generating a new RF signal, the tag encodes data by modulating the reflection of the incident wave. By rapidly switching its antenna between an impedance-matched state (absorbing the signal) and a mismatched state (reflecting it), the tag scatters the incoming signal differently — encoding a bit stream in the reflected amplitude and phase. This requires only the switching circuit, not a power amplifier, oscillator, or DAC. A dedicated Reader co-located with or adjacent to the gNB transmits the carrier and receives the backscattered response.
Rel-19 standardises two link directions between the Reader and the tag:
Carries commands and configuration from the Reader to the tag: wake-up sequences, query commands, configuration updates, and inventory commands (to select which tags should respond in a dense tag environment). Uses simple modulation schemes like on-off keying or amplitude-shift keying that a low-power tag demodulator can decode using harvested energy alone.
Carries sensor readings, identifiers, and status from the tag back to the Reader via backscatter. Data rates are typically 10–100 kbps — sufficient for temperature, humidity, pressure, and inventory data but not for high-rate sensor streams. The Reader is the computationally intensive side; the tag's contribution is the modulated reflection.
Target applications include: retail price labels that update automatically via the store's 5G network; supply-chain shipping labels that report location and condition throughout a journey; implanted biomedical sensors powered by body-heat RF; and environmental monitoring nodes too numerous to battery-power or maintain.
AI/ML Rel-19 Enhancements
Building on the Rel-18 AI/ML framework and the three standardised use cases, Rel-19 deepens the AI/ML integration across the full 5G stack:
Rel-18 defined how models are transferred to UEs. Rel-19 defines the full operational lifecycle: the gNB continuously monitors the inference quality of deployed models (comparing model-predicted beam selections against measured outcomes, for example). When quality drops below a threshold — because the UE has moved to a new propagation environment — the network triggers a model re-training or update procedure, pushing a new model version to the UE without manual intervention. Models that are no longer needed are retired to free the UE's storage and processing resources.
Rel-18's three use cases each used separate neural networks, requiring a UE to store and execute multiple models. Rel-19 investigates and standardises interfaces for shared model architectures where a single trained network serves multiple use cases — for example, a shared encoder that produces a representation used for both CSI compression and beam prediction simultaneously. This reduces the total model storage and inference compute requirements on UEs, making AI/ML practical on lower-capability devices such as RedCap and eRedCap.
At the network level, Rel-19 tightens the integration between the 5GC NWDAF (Network Data Analytics Function) and the O-RAN RIC (RAN Intelligent Controller). Standardised data collection interfaces allow the RIC to receive per-UE, per-cell, and per-slice performance data from the 5GC analytics engine, enabling AI-driven optimisation decisions that span both the RAN and the core simultaneously — for example, jointly optimising load balancing and slice admission control.
NTN Phase 2 — Enhanced Satellite Connectivity
Rel-19 extends the Rel-17 NTN foundation with three significant additions:
Rel-17 NTN always routes traffic through a ground station gateway — from the UE to the satellite, then down to the ground, then back up to the destination satellite if needed. ISLs allow a satellite to forward traffic directly to an adjacent satellite using a laser or mmWave link, without returning to the ground. This reduces latency for users far from ground gateways and enables truly global coverage without dense ground infrastructure. Rel-19 standardises the intra-plane ISL (between satellites in the same orbital plane) and inter-plane ISL (between satellites in adjacent planes).
A LEO satellite moving at 7.5 km/s is overhead for roughly 10 minutes. As it moves below the horizon, the UE must transfer to the next satellite — analogous to an inter-gNB handover. Rel-17 defined the basic NTN access procedures but left footprint handover as an incomplete area. Rel-19 defines the full handover procedure: measurement reporting of candidate satellites, the Xn-equivalent signalling between the source and target satellite (or their ground segments), and the UE state transfer — making the transition seamless without dropping the PDU session.
Rel-19 also extends NTN to IoT devices. RedCap NTN allows Rel-17 RedCap devices — smartwatches, industrial sensors, HD cameras — to connect directly to LEO satellites, enabling global 5G IoT coverage in areas without any terrestrial 5G network. Combined with ISLs, a sensor in a remote mining site, a ship in the middle of the Pacific, or a reindeer tracker in northern Siberia connects to the same 5G slice as an urban device.
V2X Phase 3 and Sidelink Evolution
V2X Phase 3 targets cooperative driving scenarios where vehicles need to share rich sensor data — not just safety messages. The key application is collective perception: each vehicle continuously shares compressed versions of its camera frames, radar returns, and LiDAR point clouds with surrounding vehicles. A following vehicle can effectively see around the truck in front by receiving the leading vehicle's forward-facing camera feed and combining it with its own sensors. Requirements:
- Data rate per vehicle: ~10 Mbps for compressed sensor streams — far beyond Phase 1/2 V2X safety messages (which were kilobits per second). NR-V2X unicast with HARQ feedback and up to 100 MHz bandwidth in FR1 handles this; LTE-V2X could not.
- Latency: under 10 ms end-to-end — sensor data that is more than 100 ms old is useless for real-time driving decisions. Configured Grant sidelink and URLLC-capable PC5 resources from Rel-17 are extended with Phase 3 scheduling priorities.
- Dense vehicle environments: resource selection algorithms are refined for scenarios with 200+ vehicles in a 500 m radius (highway traffic jams, city centre intersections), where the chance of resource collision without coordination is significant.
Enhanced ProSe multi-hop relay extends the Rel-17 UE-to-network relay to support chained hops: a remote UE connects to a relay UE which connects to a second relay UE which connects to the network. Each hop adds latency but extends coverage significantly — enabling deep indoor coverage in large multi-storey buildings where only the outer-facing floors have direct gNB coverage.
Integrated Sensing and Communication (ISAC)
Integrated Sensing and Communication (ISAC) uses the 5G NR radio signal simultaneously for both communication and radar-style sensing. The gNB transmits a standard NR downlink waveform; objects in the environment — vehicles, pedestrians, drones, rainfall — reflect a portion of the signal back. The gNB (or a dedicated sensing receiver) processes the reflected signals to estimate the range, velocity, and angle of the reflecting objects.
NR's OFDM waveform is particularly well-suited to sensing. Because the subcarrier spacing and timing are precisely known, the same channel estimation hardware used to equalise the communication channel can be adapted to process reflected-signal timing and Doppler shifts. A gNB that already performs channel estimation for hundreds of UEs can, with software changes, also build a radar-style map of its coverage area. Rel-19 formalises:
- ISAC architecture: the roles of the sensing transmitter, sensing receiver, and sensing data consumer within the 5G system. The sensing receiver may be co-located with the gNB or at a separate TRP; the sensing data is reported to the 5GC where it can be consumed by applications.
- Reference signal design: new NR reference signals optimised for sensing (wider bandwidth, higher density in time) that can be multiplexed with communication reference signals without reducing spectral efficiency.
- Signalling framework: how sensing configurations are requested, authorised, and reported — including privacy safeguards so sensing cannot be used to track individual people without consent.
Full standardisation of ISAC procedures is expected in Rel-20 and Rel-21. Applications include drone detection in controlled airspace, vehicle speed measurement, gesture recognition for human-computer interaction, and supplementary indoor mapping for emergency services.
6G Pre-Standardisation — IMT-2030 Foundations
Rel-19 formally opens study items that feed directly into ITU-R's IMT-2030 framework — the international definition of what 6G must achieve. The ITU-R targets include 1 Tbps peak rates, sub-100 µs latency, 10× better energy efficiency than IMT-2020, and native AI integration throughout. 3GPP study items opened in Rel-19:
Frequencies from 100 GHz to 300 GHz have never been used in cellular systems. Before a standard can be written, 3GPP needs validated channel models — how signals propagate, reflect, diffract, and are absorbed at these frequencies. The Rel-19 study covers measurement campaigns in indoor, outdoor, and vehicular environments, producing the statistical channel models that future air interface design will be based on. Hardware feasibility (amplifier efficiency, phase noise, ADC power) at these frequencies is studied alongside the propagation models.
Rather than adding ML to a fixed-waveform standard (as Rel-18/19 do), this study item investigates a clean-slate AI-native air interface where all layers of the PHY and MAC are end-to-end learned neural networks. Research questions include: how to train an air interface that must work across billions of different radio environments; how to guarantee safety and reliability properties for a neural-network-defined physical layer; and how standards bodies certify interoperability for learned systems.
Rather than a single gNB with many antennas, D-MIMO distributes hundreds of antenna panels across a wide geographic area, all connected by a dense fronthaul network and serving each UE cooperatively. From the UE's perspective there is no cell boundary — all panels serve all UEs. The network capacity and coverage gains over co-located massive MIMO are substantial but the fronthaul data rate and synchronisation requirements are extreme. Rel-19 studies the fronthaul architecture and protocol requirements.
Terahertz-band candidates — 141 GHz, 182 GHz, 252 GHz — are studied for regulatory feasibility, antenna design, and use cases. At these frequencies, atmospheric absorption (particularly water vapour) severely limits outdoor range. The most promising initial use cases are ultra-short-range indoor applications: chip-to-chip wireless interconnect replacing PCB traces, rack- to-rack data centre wireless backplane, and ultra-dense indoor hotspots in auditoriums or transport hubs where distances are measured in metres.
Why Rel-19 Matters
- Ambient IoT could connect trillions of devices — a market impossible to address with battery-powered technologies. If backscatter chips reach sub-cent prices at scale, every product on a retail shelf, every component in a supply chain, and every surface in a smart building could carry a connected identifier that costs less than a printed label.
- ISLs enable LEO constellations to route traffic globally without dense ground infrastructure — SpaceX Starlink's second-generation satellites already carry ISLs as a proprietary feature. Rel-19 standardises them for 5G NR, meaning any NTN-compatible device can benefit from inter-satellite routing without vendor lock-in.
- ISAC standardisation will enable gNBs to double as sensing nodes— supplementing or replacing dedicated radar infrastructure for applications like drone detection in controlled airspace and vehicle speed enforcement, using the radio assets that operators have already deployed for communications.
- The 6G study items feed directly into ITU-R IMT-2030 and the 3GPP Rel-21 schedule targeting 2028–2030 commercial deployments. Sub-THz channel models from Rel-19 will become the foundation of Rel-21's 6G PHY specification, following the same pattern that IMT-2000 channel models from the late 1990s became the foundation of UMTS.
- V2X Phase 3 collective perception sets the stage for Level 4 automated driving — the ability for a following vehicle to see around obstacles using a leading vehicle's sensor data is qualitatively different from the safety-message beaconing of Phase 1/2 V2X, enabling genuine cooperative driving decisions rather than just collision warnings.