Codebook Design

The foundational codebook for standard-resolution spatial feedback in 5G NR. It uses a grid of beams (GoB) structure based on a 2D Discrete Fourier Transform (DFT) to quantize the azimuth and zenith angles of departure.

• W1 (Wideband): Selects a beam group from the DFT grid, capturing long-term spatial correlation.
• W2 (Subband): Performs co-phasing and beam selection within the chosen group for short-term frequency-selective precoding.
• Rank 1-8 Support: Enables up to 8-layer spatial multiplexing.
• Low Overhead: Optimized for single-user MIMO with minimal CSI feedback payload.

Designed for multi-user MIMO (MU-MIMO) with significantly higher spatial resolution. It employs a linear combination of multiple DFT beams with amplitude and phase quantization to approximate the dominant eigenvectors of the channel.

• L Beam Selection: Selects L=2-4 orthogonal DFT beams per polarization per layer.
• Amplitude Quantization: Applies both wideband and subband amplitude scaling with non-uniform quantization (3-bit and 1-bit).
• Phase Quantization: Uses QPSK or 8-PSK co-phasing between beams.
• Overhead Trade-off: Delivers ~30% throughput gain over Type-I but with substantially higher feedback payload.

A variant of the Type-II codebook that reduces complexity by operating in the beamformed spatial domain rather than the full antenna domain. The UE selects beams from a set of pre-beamformed CSI-RS ports instead of the full DFT grid.

• Reduced UE Complexity: Limits the search space to K pre-coded ports, lowering computational burden.
• Angular Reciprocity: Leverages uplink SRS at the gNB to form the beamformed ports for FDD operation.
• Overhead Reduction: Significantly cuts the number of linear combination coefficients compared to regular Type-II.
• eType-II Enhancement: Further enhanced in Rel-17 with frequency domain compression.

The Rel-17 eType-II codebook introduces frequency domain (FD) compression to address the prohibitive feedback overhead of subband reporting in high-bandwidth systems. It exploits the sparsity of the channel in the delay domain.

• Wf Basis: The UE reports a compressed set of FD basis vectors (DFT columns) instead of per-subband coefficients.
• Sparsity Exploitation: Only a small number of non-zero linear combination coefficients are reported, identified via a bitmap.
• Two-Step Feedback: Part 1 carries the number of non-zero coefficients; Part 2 carries the quantized amplitudes and phases.
• Massive Overhead Reduction: Reduces payload by up to 50% compared to Rel-16 Type-II without sacrificing MU-MIMO performance.

The standardized mechanism by which the UE selects and reports the optimal precoding matrix indicator (PMI) to the gNB. This closed-loop process is fundamental to FDD massive MIMO operation.

• CSI-RS Measurement: gNB transmits beamformed or non-precoded CSI-RS across the configured ports.
• PMI Selection: UE searches the codebook for the matrix maximizing a metric like mutual information or SINR.
• RI/CQI/PMI Report: UE feeds back the Rank Indicator, Channel Quality Indicator, and Precoding Matrix Indicator.
• Reciprocity Gap: In FDD, the UE must perform this exhaustive search because the gNB cannot infer the downlink channel from uplink measurements.

The 3GPP codebook structure is parameterized by the antenna array geometry and oversampling factors, defining the DFT grid resolution.

• N1, N2: Number of antenna ports in the horizontal and vertical dimensions of the panel.
• O1, O2: Oversampling factors that control the angular granularity of the DFT beams. Higher values yield finer spatial quantization.
• i1,1, i1,2: Indices selecting the beam group in the horizontal and vertical dimensions.
• i2: Index selecting the specific beam and co-phasing within the group.
• Configuration Flexibility: The gNB configures these parameters via RRC signaling based on the deployed antenna topology.

Precoding is the multi-antenna transmission technique that applies a complex weight matrix to data streams at the transmitter, optimizing the signal for specific spatial channel conditions. The codebook provides the finite set of precoding matrices from which the user equipment selects its preferred configuration.

• Enables spatial multiplexing by steering independent data layers
• Implements interference nulling to suppress crosstalk between users
• Relies on accurate Channel State Information (CSI) for weight calculation
• Linear precoding schemes include Zero-Forcing (ZF) and Minimum Mean Square Error (MMSE)

CSI Feedback is the mechanism by which user equipment quantizes and reports its estimated downlink channel state back to the base station. The codebook defines the quantization resolution and the specific indices the UE transmits.

• Type-I codebooks provide low-resolution spatial information with minimal overhead
• Type-II codebooks report high-resolution linear combinations of beams for multi-user MIMO
• Feedback payload size is a critical constraint driving CSI compression research
• Precoding Matrix Indicator (PMI) is the index into the codebook selected by the UE

Beamforming uses an array of antennas to direct signal transmission or reception in a specific angular direction. Codebook-based beamforming restricts the beam selection to a discrete set of steering vectors defined by the codebook.

• Grid of Beams (GoB) is the foundational concept for 3GPP codebook design
• Each codeword corresponds to a specific beam direction and width
• Discrete Fourier Transform (DFT) vectors form the basis of many standardized codebooks
• Beam management procedures in 5G NR rely on codebook-based beam sweeping

Massive MIMO equips base stations with a large number of antenna elements to simultaneously serve multiple users through spatial multiplexing. Codebook design becomes increasingly complex as the antenna count grows, requiring efficient quantization of high-dimensional spatial channels.

• Angular domain sparsity is exploited by codebooks using DFT basis vectors
• Dual-stage codebooks (W1 + W2) separate long-term wideband and short-term subband precoding
• Port selection codebooks reduce complexity by selecting from virtualized antenna ports
• The codebook must scale gracefully with 64T64R and higher antenna configurations

The 3GPP Clustered Delay Line (CDL) models are standardized geometric channel representations used for link-level simulation and codebook performance evaluation in 5G NR. They define clusters of multipath components with specific angles of arrival, departure, and delay profiles.

• CDL-A, CDL-B, CDL-C represent non-line-of-sight scenarios with varying delay spreads
• CDL-D, CDL-E model line-of-sight conditions
• Each cluster has defined Angle of Departure (AoD) and Angle of Arrival (AoA) statistics
• Essential for benchmarking codebook selection accuracy under realistic propagation conditions