What Is 5G-Advanced?

5G-Advanced is the evolutionary phase of 5G NR beginning with 3GPP Release 18. It marks a significant leap from the initial 5G specifications (Releases 15--17) by introducing AI/ML natively into the air interface, expanding IoT support to battery-less devices, and optimizing the system for immersive media (XR/AR/VR). The term "5G-Advanced" was formally adopted by 3GPP in April 2021, and Release 18 was functionally frozen in June 2024 with ASN.1 completion in December 2024.

Release 18 is not a standalone specification -- it builds on the Release 17 baseline. Operators upgrading to Rel-18 need software updates to their gNB and 5GC NFs, plus new UE chipsets that support the enhanced features. The commercial timeline depends on chipset availability from Qualcomm, MediaTek, and Samsung LSI, with first Rel-18-capable chipsets expected in late 2025 and commercial devices in 2026.

Release 18 Feature Overview

Feature Matrix

FeatureWork Item3GPP SpecKey BenefitImpact Area
AI/ML for NR Air InterfaceFS_NR_AIML_airTR 38.843 / TS 38.214Predictive beam management, CSI compressionRAN intelligence
Reduced Capability (RedCap) Phase 2NR_RedCap_Ph2TS 38.300, TS 38.306Lower complexity: 5 MHz BW, 1 Rx, half-duplexIoT, wearables
XR EnhancementsNR_XR_enhTS 38.300, TS 38.321Jitter-aware scheduling, PDU set handlingImmersive media
Ambient IoTFS_NR_Ambient_IoTTR 38.848Backscatter communication, zero-energy devicesAsset tracking
Network Energy SavingNR_NESTS 38.300, TS 38.213Cell DTX/DRX, spatial/frequency adaptationOPEX, sustainability
NR NTN EnhancementsNR_NTN_enhTS 38.300, TS 38.821Regenerative satellite payloads, store-and-forwardCoverage
Sidelink EnhancementNR_SL_enh2TS 38.300, TS 38.331UE-to-UE relay, commercial sidelinkV2X, public safety
MIMO EvolutionNR_MIMO_evoTS 38.2148 Tx codebook, port selection enhancementCapacity
Duplex EvolutionNR_duplex_evoTS 38.213Subband full duplex, SBFDLatency, capacity

AI/ML for the Air Interface

This is the flagship feature of Release 18 and represents the first time 3GPP has standardized ML model lifecycle management for the radio interface. Three use cases are specified in TR 38.843:

Use Case 1: CSI Feedback Enhancement

Traditional CSI feedback uses codebook-based quantization (Type I/II codebooks per TS 38.214). The UE selects the closest codeword to represent the channel, which is lossy. AI/ML-based CSI compression uses an autoencoder:

  • Encoder (UE side): Neural network compresses the channel matrix H into a low-dimensional latent vector z
  • Decoder (gNB side): Neural network reconstructs the approximate channel matrix H' from z
  • Benefit: 4--10x reduction in CSI feedback overhead with < 1 dB throughput loss

Use Case 2: Beam Management

Beam prediction replaces exhaustive SSB/CSI-RS beam sweeps with ML models that predict optimal beams from partial measurements:

  • Spatial prediction: Measure N beams, predict best beam from a set of M > N candidates
  • Temporal prediction: Predict future best beam from past measurements (useful for mobile UEs)
  • Benefit: Reduce beam sweep overhead by 40--75%, lowering latency for beam switching

Use Case 3: Positioning Enhancement

ML models improve positioning accuracy beyond traditional TDOA/AoA methods by learning multipath fingerprints.

Worked Example 1 -- AI/ML Beam Prediction Overhead Savings

Scenario: A gNB is configured with 64 SSB beams (SSB burst set = 64). Exhaustive measurement requires the UE to measure all 64 beams before reporting the best. Without AI/ML: `

SSB beams to measure: 64

Measurement periodicity: 20 ms (per SSB burst)

Overhead per measurement cycle: 64 SSB resources

Beam report: Best beam index + L1-RSRP (4 beams reported)

` With AI/ML Beam Prediction (spatial domain): `

SSB beams measured: 16 (subset of 64)

ML model at gNB predicts best beam from 16 partial measurements

Prediction accuracy: 92% top-1, 98% top-3 (per TR 38.843 evaluation)

Overhead reduction: (64-16)/64 = 75%

Latency saving: 3 SSB bursts saved = 60 ms faster beam acquisition

` Throughput impact: When top-1 prediction is wrong (8% of cases), fallback to exhaustive sweep adds 60 ms latency. Average throughput loss: 0.08 * (60ms delay penalty) = 4.8 ms average additional delay, which is negligible for most applications but relevant for URLLC.

3GPP defines the ML model management framework in Release 18 normative specifications: model training, validation, inference, monitoring, and fallback to legacy operation when model performance degrades.

RedCap Phase 2

Release 17 introduced NR Reduced Capability (RedCap) for mid-tier IoT devices -- smartwatches, industrial sensors, and video cameras that need more capability than LTE-M/NB-IoT but less than full NR. Release 18 Phase 2 further reduces complexity:

RedCap Evolution

ParameterFull NR (Rel-15)RedCap Phase 1 (Rel-17)RedCap Phase 2 (Rel-18)NB-IoT
Max bandwidth100 MHz (FR1)20 MHz5 MHz200 kHz
Rx antennas2 or 41 or 211
MIMO layers (DL)Up to 41 or 211
Duplex modeFull duplex FDD/TDDFull or half duplexHalf duplex FDD onlyHalf duplex
Peak DL rate> 1 Gbps~85 Mbps (20 MHz, 2 Rx)~10 Mbps (5 MHz, 1 Rx)~0.25 Mbps
UE complexityBaseline50--65% of full NR25--35% of full NR10%
Target use caseseMBB, URLLCWearables, camerasSensors, meters, trackersMassive IoT

Worked Example 2 -- RedCap Phase 2 Throughput Calculation

Configuration: RedCap Phase 2 UE on TDD n78, 5 MHz bandwidth, 30 kHz SCS, 1 Rx antenna, 64QAM. `

Resource blocks in 5 MHz at 30 kHz SCS: 11 RBs (per TS 38.101-1 Table 5.3.2-1)

Subcarriers per RB: 12

Symbols per slot (normal CP): 14

Slots per subframe (30 kHz SCS): 2

TDD pattern: DDDSU (3.5 DL slots per 5-slot period = 70% DL)

DL resource elements per slot: 11 12 14 = 1,848 REs

Subtract DMRS overhead (~14%): 1,848 * 0.86 = 1,589 REs

Bits per RE (64QAM, code rate 0.65): 6 * 0.65 = 3.9 bits

Bits per slot: 1,589 * 3.9 = 6,197 bits

Slots per second: 2,000 (30 kHz SCS, 2 slots per ms)

DL slots per second (70% TDD): 2,000 * 0.7 = 1,400

Peak DL throughput: 6,197 * 1,400 = 8.68 Mbps

With 10% overhead (SIB, paging, etc.): ~7.8 Mbps

`

This throughput is sufficient for smart metering (< 100 kbps), asset tracking (< 1 Mbps), and industrial sensor aggregation (1--5 Mbps).

Network Energy Saving

Release 18 introduces three RAN-level energy saving features specified in TS 38.213 and TS 38.300:

FeatureMechanismSaving PotentialImpact on UE
Cell DTX/DRXgNB enters sleep within active cell, DL/UL alignment with UE DRX10--20%Requires UE Rel-18 awareness
Spatial domain adaptationReduce MIMO layers/antenna ports during low traffic15--30%Transparent to legacy UEs
Frequency domain adaptationReduce active BWP or switch off carriers10--25%BWP switch signaling to UEs

Deutsche Telekom reported a 23% RAN energy reduction in a Rel-18 trial across 500 sites in Berlin using spatial domain adaptation (reducing from 64T64R to 32T32R during off-peak) combined with carrier shutdown on the 3.5 GHz layer between midnight and 6 AM.

Ambient IoT

Ambient IoT is one of the most disruptive features in Release 18. It enables zero-energy devices that harvest energy from the NR signal itself (or ambient RF) and communicate via backscatter modulation. This targets use cases where battery replacement is impractical:

  • Warehouse asset tracking: Billions of items with passive NFC-like tags that can be read at 10+ meter range using 5G NR signals
  • Smart logistics: Package-level tracking through supply chains without batteries
  • Smart agriculture: Soil and crop sensors with indefinite lifetime

3GPP TR 38.848 defines three device categories:

CategoryEnergy SourceRangeData RateComplexity
Device A (passive)RF energy harvesting only1--10 m1--10 kbpsLowest (tag-like)
Device B (semi-passive)Small energy storage + harvesting10--50 m10--100 kbpsLow
Device C (active-assisted)Battery-assisted with harvesting50--200 m100 kbps--1 MbpsMedium

Operator Adoption Data

China Mobile -- Rel-18 Trial Network

China Mobile deployed Rel-18 trial sites across 10 cities in 2025:

  • AI/ML beam management: 42% reduction in beam sweep overhead at mmWave (26 GHz) sites
  • Network energy saving: 19% energy reduction using cell DTX combined with spatial adaptation
  • RedCap Phase 2: Deployed 50,000 smart meter modules at 5 MHz bandwidth
  • Scale: 2,000 trial sites with Huawei and ZTE equipment

KDDI (Japan) -- XR Optimization

KDDI focused Rel-18 deployment on XR services for their enterprise customers:

  • XR jitter performance: P95 jitter reduced from 12 ms (Rel-17) to 4.5 ms (Rel-18 PDU set awareness)
  • XR frame drop rate: Reduced from 2.1% to 0.3% with jitter-aware scheduling
  • User capacity per cell for AR (50 Mbps requirement): Increased from 8 to 12 simultaneous users
  • Technology: Rel-18 MAC-level enhancements with PDU set identification per TS 38.321

Timeline and Roadmap

MilestoneDateStatus
Release 18 study items approved2022 Q1Complete
Release 18 work items approved2022 Q4Complete
Release 18 functional freeze2024 Q2 (June)Complete
Release 18 ASN.1/protocol freeze2024 Q4 (December)Complete
First Rel-18 chipsets (Qualcomm X80)2025 Q4Sampling
First commercial Rel-18 devices2026 H1Expected
Release 19 functional freeze2025 Q4In progress
Release 19 ASN.1 freeze2026 Q2Planned

Release 19 continues the 5G-Advanced evolution with enhanced ISAC (Integrated Sensing and Communication), advanced AI/ML (multi-cell coordination), and further NTN improvements including direct-to-device satellite.

Key Takeaway: 5G-Advanced Release 18 is the most feature-rich 3GPP release since Release 15 launched 5G NR. AI/ML for the air interface (TR 38.843) brings predictive beam management with 75% overhead reduction. RedCap Phase 2 drops UE complexity to 25% of full NR, enabling 5G for billions of IoT sensors. Network energy saving features deliver 20%+ RAN power reduction as demonstrated by Deutsche Telekom and China Mobile trials. With ASN.1 frozen in December 2024 and first chipsets sampling in late 2025, operators should begin Rel-18 feature planning and trial deployments now.