3GPP Quarterly — September 2025

5G-Advanced CSI feedback will be built on precoding matrices, but debate on channel matrix feedback continues

The topic is enhancing Channel State Information (CSI) feedback using AI models where both the phone (UE) and the network (gNB) have parts of the model. This aims to improve accuracy or reduce overhead compared to current 5G methods. The key question was what information the AI model should target: the 'precoding matrix' (how signals should be combined at the antennas) or the raw 'channel matrix' (the measured radio channel itself).

A group of companies, led by the moderator (Qualcomm), proposed supporting at least the precoding matrix as the target CSI type. The proposal also included a 'FFS' (For Further Study) clause on whether to consider the raw channel matrix for further improvement, starting with evaluations on aspects like fusion with uplink reference signals.

The meeting noted this proposal. A large coalition of 17 companies, including Samsung, Nokia, Apple, and MediaTek, supported prioritizing the precoding matrix, arguing it is well-studied from 5G and offers a good performance-complexity trade-off. Huawei and ZTE argued for also considering the channel matrix, citing potential gains from combining it with uplink measurements. The proposal was updated to incorporate Huawei's wording but was not adopted as a final decision.

3GPP agrees to study multiple formats for AI training data in 5G Advanced

Channel State Information (CSI) tells the network the best way to send data to a phone over the current radio conditions. In 5G Advanced Release 20, AI models will be used to compress this CSI, reducing the data phones need to send back. To train these AI models, high-quality 'ground truth' CSI data must be collected from phones. A key debate is how phones should quantize (digitally represent) this data before sending it to the network for training, balancing accuracy against phone processing complexity and battery drain.

At this RAN1#122 meeting, the working group responsible for the physical layer decided to study several quantization methods for this training data. The agreed options include: reusing the legacy 5G 'eType II' codebook (Option 0), using simple scalar quantization of the CSI matrix (Option 1), using enhanced codebook methods with reduced computational complexity (Option 2), reporting dominant eigenvalues and eigenvectors (Option 3), and using an enhanced 'eType II' codebook with new, more precise parameters like a larger number of spatial beams (Option 4). The study will evaluate these options based on overhead, performance loss, and complexity, using unquantized data (Float32) and the legacy eType II codebook as benchmarks.

Standardizing the time window for on-demand SIB1 scheduling

When a phone sends a request for SIB1, the network schedules the transmission within a specific time window. The parameter 'od-sib1-windowStartOffset' defines when this window starts after the phone's PRACH transmission. The meeting had previously agreed the value should be zero or positive, measured in slots, but the exact set of allowable values was undecided.

The meeting agreed on the value range for od-sib1-windowStartOffset. It will use the set of values {sl0, sl1, sl2, sl4, sl8, sl10, sl20, sl40, sl80}, where 'sl' stands for slots. This range is identical to the existing 'ra-ResponseWindow' parameter used in 5G random-access procedures, with the addition of a zero-slot option (sl0). This agreement finalizes a key timing parameter for the on-demand SIB1 feature.

Proposal on RedCap indicator for UL-WUS configuration deferred to RAN2

A question was raised about whether Reduced Capability (RedCap) devices should be allowed to use the UL-WUS configuration for on-demand SIB1, as they might be barred from camping on the cell providing that configuration.

Samsung proposed adding a 'RedCapAllowed' parameter to the UL-WUS configuration to explicitly control this. Other companies, including Google and Apple, argued that this is not a physical layer ambiguity and should be determined by RAN2 (protocols — the messages between handset and network). The moderator noted that this topic has been discussed in RAN2 for multiple meetings, and the RAN1 group agreed to leave the decision to RAN2.

Release 20 Ambient IoT outdoor study will explore basic positioning measurements

The discussion concerns whether to include an objective on positioning in the Release 20 study for Ambient Internet of Things (IoT) devices operating outdoors. Ambient IoT refers to a category of low-power IoT devices that harvest energy from their surroundings, such as from radio waves, and communicate with a network reader. A group of companies pushed for a minimal-scope addition to the study.

The moderator's proposal (Proposal 1v2) suggested adding an objective to the study item (SID) to 'Identify D2R measurement(s) by one Reader, and the involved A-IoT signal(s)/channel(s), which are feasible for Device 2b/Device C for more accurate outdoor Device localization than based on Reader-ID [RAN1].' It explicitly stated that RAN1, the working group responsible for the physical layer, is not expected to conduct accuracy evaluations or conclude on positioning accuracy as part of the study. The proposal assumes no new positioning architecture will be specified in Release 20.

The meeting did not reach a decision on this proposal. The document was 'revised', indicating further work is needed. The proposal will be revisited at a future meeting, with RAN#111 (March 2026) slated to decide whether to include positioning in the normative work.

RAN1 establishes evaluation framework for 6G Ambient IoT in outdoor active devices

Ambient IoT refers to a network of very low-power, energy-harvesting devices designed for applications like industrial sensors and tracking. For Release 20, 3GPP is studying how to extend this technology, previously focused on indoor environments, to work for outdoor devices with small batteries or energy harvesters (known as Device 2b and Device C). The working group RAN1, responsible for the physical layer, reached several agreements on how companies should evaluate these devices' performance.

The agreed link budget methodology uses 'Budget-Alt2' to derive receiver sensitivity. For link-level simulations, the TDL-C channel model with a 300 ns delay spread will be used. The coverage evaluation follows a three-step process from Release 19: first, obtain the required signal-to-interference-plus-noise ratio (SINR) via simulation; second, calculate receiver sensitivity based on that SINR; third, determine the maximum path loss and coverage distance using a link budget calculation.

Pathloss models for evaluation include Urban Macro (UMa) Non-Line-of-Sight (NLOS), Rural Macro (RMa) NLOS, and Urban Micro (UMi) NLOS, all defined in specification TR 38.901. Companies may also report results using RMa Line-of-Sight (LOS).

The agreements specify detailed assumptions for the evaluation. Base station (reader) transmit power is set at 43 dBm for standalone deployments, or 38/33 dBm for in-band deployments. Device transmit powers are set at {-10, -20} dBm for Device 2b, and {-3, 0, 5} dBm for Device C. Key link budget parameters include a base station antenna gain of 17 dBi, device antenna gain of 0 dBi, a base station noise figure of 5 dB, and device noise figures of 15 dB for Device 2b and 13 dB for Device C. Transmission bandwidths start at 180 kHz for the reader-to-device link and 15 kHz for the device-to-reader link.

Clock inaccuracies are modeled differently for the two device types. For Device 2b, the initial carrier frequency offset can be randomly selected from 1,000 to 10,000 parts per million (ppm), improving to 100 ppm after calibration. For the more capable Device C, the initial offset is 50 ppm, improving to 10 ppm after calibration.

A new 'penetration margin' item is added to the link budget template, with example values of 0 dB (no obstruction), 20 dB (device inside a car), and 80 dB (device inside a metallic container). Companies must also report their assumptions on the energy source (e.g., solar) and storage capacity used for their data rate evaluations.

No decision on whether to specify UE expectation for inconsistent DCI gap skipping commands

In 5G, a UE can be configured with two different interpretations for a '0' value in the measurement gap cancellation DCI field. Under 'Option 2', a '0' tells the UE to follow the standard measurement gap behavior, effectively ignoring the skip command.

At a previous meeting (RAN1#121), it was concluded that "UE does not expect different values in 'Measurement gap cancellation' fields in DCI formats that are associated with the same measurement gap occasion and are received in same PDCCH monitoring occasion." However, it was unclear if a specification update was needed.

At RAN1#122, the moderator asked companies whether to adopt a proposed text change (TP-5.2v1) to formally state this expectation in the standard, or whether existing rules for handling inconsistent DCI information were sufficient. Opinions were split: Qualcomm, OPPO, vivo, and DOCOMO supported the clarification (Option A), while Samsung, ZTE, and Sanechips opposed it (Option B). The moderator stated, "As this should effectively be a very corner case, moderator considers not bringing this for online discussion." No vote was taken on the proposal.

The reason for the disagreement is not stated in the document.

No decision on defining 'last DCI format' for measurement gap skipping

When a UE receives multiple DCI commands ('Option 2' type) related to the same measurement gap, the standard says it follows the command in the 'last DCI format.' The definition of 'last' was ambiguous.

At RAN1#122, the moderator presented three options: define 'last' based on the index of the PDCCH monitoring occasion (Option A), define it based on the end time of the PDCCH reception (Option B), or deem no clarification necessary (Option C). Companies were split: Qualcomm, vivo, and DOCOMO preferred Option A, OPPO also preferred clarification, while Samsung, ZTE, and Sanechips preferred Option C. The moderator stated, "While some companies see the merit, other companies consider that no clarification is needed" and suggested a brief online discussion. No decision was made.

The reason for the disagreement is not stated in the document.

Network Digital Twins definition sharpened to clarify role in high-risk operation evaluation

This topic concerns the use of Network Digital Twins (NDTs) – virtual, software-based models of a mobile network used for simulation – in managing network changes that could cause major outages. A change was made to the standard 3GPP 28.561 to provide a clearer definition of "high-risk operations" and how an NDT is used to evaluate them.

The updated text in clause 5.2.2.2 now defines high-risk operations as "operations which have the potential of causing major service impact, for instance... potentially dangerous configuration or modification, policy modification... software version upgrade, and board switching, or optimization procedures on a live network."

It clarifies that "Operators should not carry out the high-risk network operations or perform direct optimization in the physical network." Instead, "The NDT can be utilized to evaluate both known... and those suspected to pose high-risk consequences to the network."

The text gives an example: "the NDT can predict SLA degradation and single node failures in mobile networks. When the prediction indicates insufficient network resources... the NDT results help operators to take appropriate preventive actions to avoid network outages."