Type-II Codebook

Type-II codebooks extend the Grid of Beams concept by selecting a set of L orthogonal beams from a Discrete Fourier Transform (DFT) basis. Unlike Type-I, which selects a single beam, Type-II linearly combines 2 to 4 beams per layer. The base station oversamples a 2D DFT grid to create a dense set of candidate beams, ensuring high spatial resolution for precise beamforming in both azimuth and elevation domains.

A defining feature is the reporting of sub-band amplitude and phase coefficients. The wideband amplitude is quantized with higher resolution, while sub-band coefficients capture frequency-selective fading:

• Wideband (WB) amplitude: Reported for each beam across the entire bandwidth part.
• Sub-band (SB) amplitude and phase: Reported per sub-band to track frequency selectivity.
• Differential encoding: Sub-band phase is encoded relative to the wideband phase to reduce feedback overhead.

3GPP Release 16 introduced the port selection codebook variant for Frequency Division Duplex (FDD) systems without channel reciprocity. Instead of selecting DFT beams, the user equipment selects CSI-RS ports from a beamformed reference signal. This approach leverages angle-of-arrival reciprocity and reduces the codebook search complexity while maintaining the linear combination structure of amplitude and phase weighting across sub-bands.

Type-II codebooks achieve high precision at the cost of significant uplink feedback overhead. A typical configuration with L=4 beams, 2 layers, and 13 sub-bands can generate thousands of bits per report. Mitigation strategies include:

• Frequency-domain compression using Discrete Fourier Transform (DFT) basis in R16 enhanced Type-II
• Tap-based time-domain compression exploiting channel sparsity
• Machine learning autoencoders replacing explicit coefficient quantization

The high spatial granularity of Type-II codebooks directly translates to multi-user MIMO (MU-MIMO) gains. By accurately reporting the channel eigenstructure, the base station can compute near-optimal Zero-Forcing (ZF) or Minimum Mean Square Error (MMSE) precoders. This enables:

• Serving up to 12 layers simultaneously in massive MIMO
• Inter-user interference suppression through precise null steering
• Spectral efficiency improvements of 30-50% over Type-I codebooks in dense urban deployments

The 3GPP standard defines specific quantization resolutions for Type-II coefficients:

• Wideband amplitude: 3-bit quantization with non-uniform step sizes
• Sub-band amplitude: 1-bit (on/off) per sub-band
• Sub-band phase: Quadrature Phase Shift Keying (QPSK) or 8-PSK This hierarchical quantization prioritizes the most impactful coefficients while constraining total feedback payload. The strongest coefficient per layer is normalized to unity to serve as a phase reference.

The feedback index sent from the user equipment (UE) to the base station (gNB) that recommends a specific precoding matrix from the Type-II codebook. This index represents the UE's calculation of the optimal combination of beams, amplitudes, and phase shifts to maximize its downlink signal-to-noise ratio. The PMI report includes both wideband and sub-band components to capture frequency-selective fading.

The process of reducing the massive feedback overhead generated by a high-resolution Type-II codebook. Because reporting complex coefficients for multiple beams across many sub-bands consumes significant uplink resources, techniques like frequency-domain compression and neural network autoencoders (e.g., CsiNet) are applied to represent the precoding matrix with fewer bits before transmission.

The multi-antenna technology that drives the need for high-resolution codebooks. A base station equipped with a large array of active elements (e.g., 64T64R) uses Type-II codebook feedback to perform multi-user MIMO (MU-MIMO) precoding. The granular spatial selectivity provided by the codebook allows the gNB to form narrow beams that serve multiple users simultaneously on the same time-frequency resource with minimal inter-user interference.

The downlink pilot signal specifically designed for UE measurement and CSI reporting. The gNB transmits beamformed CSI-RS resources, and the UE uses these to estimate the downlink channel and select the optimal Type-II codebook parameters. The density and periodicity of CSI-RS transmission directly impact the freshness and accuracy of the reported PMI, creating a trade-off between overhead and beamforming precision.

The phenomenon where CSI becomes outdated between the measurement instant and the actual data transmission due to node mobility. In high-Doppler scenarios, a Type-II codebook report may no longer match the current channel state, degrading MU-MIMO performance. Mitigation strategies include CSI prediction using recurrent neural networks and increasing the CSI reporting periodicity at the cost of higher uplink overhead.

An architecture that splits precoding between a low-dimensional digital baseband processor and a high-dimensional analog phase-shifter network. In millimeter wave systems, the Type-II codebook is adapted to account for the constraints of the analog beamforming network, where beams are selected from a discrete set of spatial directions. The codebook must jointly optimize digital precoding weights and analog beam selection.