Reference Signal Prediction Accuracy Indicator

Description

The Reference Signal Prediction Accuracy Indicator (RS-PAI) is a new physical layer metric introduced in 3GPP Release 19 to support advanced channel prediction techniques in 5G-Advanced and future 6G networks. As networks move toward higher frequencies and more dynamic environments, the ability to predict future channel states becomes crucial for maintaining link reliability, reducing overhead, and enabling proactive resource allocation. RS-PAI provides a standardized way to quantify how accurate those predictions are, allowing the network to make informed decisions about when to trust predicted reference signals and when to fall back to traditional measurement-based approaches.

In operation, RS-PAI is computed by comparing predicted reference signal characteristics—such as channel state information (CSI), beam quality, or path loss—against actual measured values. The indicator can be reported by the UE to the gNB or calculated internally by the network. It is expressed as a quantitative value or a set of levels that reflect the prediction error margin. This information is then used in procedures like beam failure recovery, handover optimization, and dynamic scheduling. For example, if RS-PAI indicates high accuracy, the network may reduce the frequency of reference signal transmissions, thereby saving energy and reducing overhead. Conversely, low accuracy triggers more frequent measurements or a switch to more robust transmission schemes.

The indicator is defined in 3GPP TS 38.212 (multiplexing and channel coding) and TS 38.214 (physical layer procedures for data), ensuring it integrates seamlessly with existing NR physical layer signaling. RS-PAI can be carried in uplink control information (UCI) or downlink control information (DCI), depending on the scenario. Its introduction reflects the growing importance of AI/ML-based prediction in radio access networks, where accurate forecasting of channel conditions can significantly enhance spectral efficiency and user experience.

Key technical aspects include the definition of prediction accuracy metrics, the mapping of these metrics to reporting values, and the procedures for triggering and using RS-PAI in various network operations. The indicator supports both periodic and aperiodic reporting, and can be configured with different granularities (e.g., per beam, per subband, or wideband). It also interacts with other Rel-19 features such as AI/ML-based beam management and mobility optimization, providing a feedback loop that helps train and validate prediction models.

Overall, RS-PAI is a foundational enabler for predictive radio resource management. By standardizing how prediction accuracy is measured and communicated, it allows multi-vendor interoperability and paves the way for autonomous network operations that rely on accurate channel forecasting.

Purpose & Motivation

RS-PAI was created to address the challenge of integrating channel prediction into 5G-Advanced networks. Traditional measurement-based approaches require frequent reference signal transmissions, which consume resources and energy, especially in high-frequency bands like mmWave. Prediction can reduce this overhead, but without a standardized accuracy metric, the network cannot reliably determine when predictions are trustworthy. RS-PAI solves this by providing a common language for prediction quality, enabling dynamic adaptation between predictive and measurement-based modes.

Historically, 3GPP releases up to Rel-18 focused on measurement and reporting of current channel conditions. With Rel-19, the industry began exploring AI/ML for air interface, including channel prediction. However, early implementations lacked a standardized way to assess prediction reliability, leading to potential performance degradation if predictions were inaccurate. RS-PAI fills this gap by defining a quantifiable indicator that can be used in link adaptation, beam management, and mobility procedures. It also supports the evolution toward energy-efficient networks by allowing reduced reference signal overhead when predictions are reliable.