Sensing Assisted CSI Acquisition for ISAC

Introduction

Channel State Information (CSI) acquisition in 5G NR involves a well-defined pipeline: the gNB transmits CSI-RS, the UE measures the channel, compresses the information using a codebook (e.g., Type I or Type II), and feeds it back via UCI on PUCCH/PUSCH. This feedback loop introduces latency (typically several milliseconds) and overhead (CSI-RS resources + feedback bits), both of which limit system throughput — especially for FDD massive MIMO.

Sensing-assisted CSI acquisition aims to bypass or augment parts of this feedback loop by deriving channel state information directly from sensing measurements at the gNB side. Since the gNB performs sensing in a monostatic or bistatic configuration, it can extract spatial and temporal channel characteristics without relying on UE feedback.

 CSI Acquisition: Conventional vs. Sensing-Assisted
┌──────────────────────────────────────────────────────────────────────┐
│                                                                      │
│  Conventional CSI Acquisition:                                       │
│  ┌──────┐  CSI-RS  ┌──────┐  Measure  ┌──────┐  Feedback  ┌──────┐  │
│  │ gNB  │─────────▶│  UE  │──────────▶│  UE  │───────────▶│ gNB  │  │
│  │      │          │      │           │Encode│            │Decode│  │
│  └──────┘          └──────┘           └──────┘            └──┬───┘  │
│                                                              │      │
│                                               CSI (PMI/RI/CQI)      │
│                                                                      │
│  Sensing-Assisted CSI Acquisition:                                   │
│  ┌──────┐  Sense   ┌──────────────┐  Extract  ┌──────────────────┐  │
│  │ gNB  │─────────▶│ Echo Signal  │──────────▶│ Spatial Channel  │  │
│  │      │  Echo    │ Processing   │           │ Parameters       │  │
│  └──────┘          └──────────────┘           └────────┬─────────┘  │
│                                                        │            │
│                                                        ▼            │
│                                              ┌──────────────────┐   │
│                                              │  Predicted CSI   │   │
│                                              │  (No UE Feedback)│   │
│                                              └──────────────────┘   │
│                                                                      │
└──────────────────────────────────────────────────────────────────────┘

Key Concepts

Sensing-Derived Spatial Covariance

The spatial covariance matrix of the downlink channel is a key input for precoding and rank adaptation. In FDD systems, the UE must measure and report this information. With sensing:

The gNB estimates the angular power spectrum from echo signals.
The dominant AoDs identified by sensing correspond to the dominant eigenvectors of the spatial covariance.
This information can be used to select or refine the precoding matrix without UE feedback.

CSI Prediction and Extrapolation

CSI becomes outdated between measurement and application due to processing and feedback delay. Sensing can mitigate this by:

Doppler-based prediction: Sensing-estimated Doppler shifts enable prediction of the channel evolution over the feedback delay interval.
Trajectory-based prediction: If the UE’s position and velocity are tracked via sensing, the gNB can predict the future channel state geometrically.

 CSI Prediction via Sensing
┌──────────────────────────────────────────────────────────────────┐
│                                                                  │
│   CSI measured          Feedback delay           CSI applied     │
│   at time t₀            Δt (stale!)              at time t₁     │
│   ────┬──────────────────────────────────────────────┬────       │
│       │                                              │           │
│       │   Without sensing:  CSI(t₀) used at t₁       │           │
│       │   ────────────────▶ ✗ Mismatch              │           │
│       │                                              │           │
│       │   With sensing:     CSI(t₀) + Doppler info   │           │
│       │   ────────────────▶ CSI(t₁) predicted ✓     │           │
│       │                                              │           │
└──────────────────────────────────────────────────────────────────┘

Reduced Feedback Overhead

By providing channel knowledge at the gNB side, sensing reduces the UE’s feedback burden:

Reduced codebook search: The UE can be instructed to search only a subset of the codebook if the gNB provides a hint (e.g., a restricted beam set) based on sensing.
Fewer CSI-RS ports: If dominant spatial directions are known from sensing, fewer CSI-RS ports are needed for the UE to estimate the residual channel.
Longer CSI-RS periodicity: With sensing filling the gaps, CSI-RS can be transmitted less frequently.

Benefits

Table 6 Benefits of Sensing-Assisted CSI Acquisition
Benefit	Description
Reduced feedback latency impact	Sensing-based prediction compensates for CSI aging.
Lower uplink overhead	Less frequent or smaller CSI reports needed from the UE.
FDD enablement for massive MIMO	Partial CSI obtained at gNB without full downlink channel sounding.
Better MU-MIMO scheduling	Sensing-derived spatial information enables more accurate user pairing.

Challenges

Monostatic-to-bistatic mapping: The sensing echo represents the round-trip channel (gNB → scatterer → gNB), which differs from the one-way communication channel (gNB → UE). Mapping between these requires knowledge of the scatterer geometry.
Limited path resolution: Sensing may not resolve all multipath components, especially weak scattered paths that contribute to the communication channel.
Codebook compatibility: Existing 3GPP codebook structures (Type I/II) may not be directly amenable to sensing-derived CSI injection; new interfaces or codebook designs may be needed.