Squared Generalized Cosine Similarity

Description

Squared Generalized Cosine Similarity (SGCS) is a metric defined in 3GPP NR specifications to evaluate the alignment or similarity between two complex vectors, often representing beamforming weights or channel state information. Mathematically, for vectors **a** and **b**, SGCS is computed as |**a**^H **b**|^2 / (||**a**||^2 ||**b**||^2), where ^H denotes the conjugate transpose. This yields a value between 0 and 1, indicating orthogonality (0) or perfect alignment (1). In NR, SGCS is applied primarily in beam management procedures to assess how similar candidate beams are, which informs decisions on beam switching, combining, or refinement.

Architecturally, SGCS is utilized within the gNB and UE physical layer algorithms, as detailed in specifications like 38.212 (multiplexing and channel coding) and 38.214 (physical layer procedures). It operates on beamforming vectors derived from channel measurements, such as those obtained via Channel State Information Reference Signals (CSI-RS) or Synchronization Signal Blocks (SSBs). The gNB may calculate SGCS between different beam pairs to identify redundant beams or to optimize multi-beam transmissions, reducing interference and improving spectral efficiency.

How it works: During beam management, the UE reports measurements like Reference Signal Received Power (RSRP) for multiple beams. The gNB can compute SGCS between the reported beams' spatial characteristics to determine if they are sufficiently distinct. For instance, in a multi-panel UE scenario, SGCS helps decide whether to use multiple simultaneous beams (spatial multiplexing) or switch to a single beam, based on similarity thresholds. This is critical for massive MIMO systems where numerous beams are available, and efficient selection is key to maintaining high throughput and coverage.

Key components include the beamforming codebook, from which vectors are selected, and the channel estimation module that provides the input vectors. SGCS is also referenced in 38.843 for non-terrestrial networks, where beam similarity assessment must account for dynamic satellite movements. By quantifying beam correlation, SGCS enables advanced features like beam failure recovery and coordinated multi-point transmission, ensuring robust performance in diverse deployment scenarios.

Purpose & Motivation

SGCS was introduced to address the complexity of beam management in NR, especially with massive MIMO and high-frequency bands (e.g., mmWave). Prior approaches relied heavily on RSRP-based beam selection, which could lead to suboptimal choices when beams have similar power but different spatial properties, causing interference or missed multiplexing opportunities. SGCS provides a standardized metric to evaluate beam similarity, enabling more intelligent beam coordination and resource allocation.

Historically, beam management in LTE was simpler due to limited MIMO layers. With NR's support for hundreds of antenna elements and flexible beamforming, a quantitative similarity measure became necessary to handle beam correlation and avoid redundant transmissions. SGCS, specified from Release 18, fills this gap by offering a mathematical foundation for comparing beam vectors, directly supporting features like multi-beam operation and enhanced mobility.

The motivation stems from the need to improve spectral efficiency and reduce overhead in dense beam environments. By using SGCS, networks can identify orthogonal beams for simultaneous transmission, enhance beam refinement accuracy, and adapt to rapid channel changes. This is particularly crucial for non-terrestrial networks in 38.843, where beam alignment must be dynamically adjusted due to satellite motion. Ultimately, SGCS contributes to higher data rates, better coverage, and reliable connectivity in advanced NR deployments.

Key Features

• Quantifies similarity between beamforming or channel vectors on a scale from 0 to 1
• Used in NR beam management for selection, switching, and refinement decisions
• Supports massive MIMO and multi-beam operations to reduce interference
• Integrates with physical layer procedures defined in 38.212 and 38.214
• Applicable to terrestrial and non-terrestrial networks (e.g., satellite)
• Enhances spectral efficiency by identifying orthogonal beams for multiplexing

Evolution Across Releases

Defining Specifications