Precoding Matrix Indicator (PMI)

PMI operates by indexing into a predefined codebook known to both the transmitter and receiver. The UE evaluates all candidate matrices and selects the one that maximizes a specific metric, such as post-processing Signal-to-Interference-plus-Noise Ratio (SINR) or mutual information. This closed-loop mechanism avoids the overhead of sending full, unquantized Channel State Information (CSI) back to the transmitter.

• Codebooks are standardized by 3GPP for LTE and NR
• Common codebook types include Type-I (single-user) and Type-II (multi-user, higher resolution)
• Selection is typically based on maximizing throughput or minimizing error rate

The PMI is intrinsically linked to the Rank Indicator (RI) . The UE first determines the number of usable spatial layers (the rank) and then selects the PMI corresponding to that specific rank. A PMI reported for a rank-2 transmission indexes a different set of matrices than a PMI for a rank-1 transmission.

• The codebook structure is hierarchical, with distinct subsets for each allowed rank
• A change in RI typically invalidates the previously reported PMI
• Joint RI/PMI reporting is common in practical systems to reduce latency

PMI feedback can be configured for different frequency granularities. Wideband PMI provides a single recommendation for the entire system bandwidth, offering low overhead but poor performance in frequency-selective channels. Subband PMI reports separate indices for different frequency sub-blocks, enabling frequency-selective precoding that adapts to the channel's varying response across the spectrum.

• Subband size is configurable by the network (e.g., 4-8 Physical Resource Blocks in LTE)
• A trade-off exists between feedback accuracy and uplink control overhead
• Best-M reporting can further compress feedback by only reporting PMI for the strongest subbands

Many standardized codebooks, particularly for Massive MIMO systems, are built on the Grid of Beams (GoB) principle. The codebook consists of a set of Discrete Fourier Transform (DFT) vectors that form a grid of fixed, narrow beams pointing in different directions. The PMI selects one or more of these beams, and the precoder is constructed by combining them with appropriate co-phasing coefficients to form the final transmission beam.

• Provides a structured, efficient way to sample the angular domain
• Enables beam-based rather than antenna-element-based feedback
• Scales efficiently with the number of antenna ports

The reporting of PMI consumes precious uplink control channel resources. The network configures the reporting periodicity and offset to balance the freshness of CSI against overhead. In rapidly changing channels, a high periodicity is required to prevent the precoder from becoming stale, while in static environments, infrequent reporting suffices.

• Periodic reporting uses PUCCH (Physical Uplink Control Channel)
• Aperiodic reporting can be triggered via DCI (Downlink Control Information) on PUSCH (Physical Uplink Shared Channel) for on-demand, high-resolution feedback
• Semi-persistent reporting combines aspects of both for periodic traffic with low latency

In Multi-User MIMO (MU-MIMO) , the PMI plays a crucial role in managing inter-user interference. The base station collects PMI reports from multiple UEs and uses them to construct a composite precoder. A well-designed codebook (like Type-II) allows the UE to report not just its preferred beam but also the strongest interfering beams and their associated amplitude/phase, enabling the base station to perform sophisticated null-steering to other users.

• Type-II codebooks enable high-resolution, multi-beam feedback
• Linear combination of beams allows for precise channel representation
• Critical for achieving the high spectral efficiency promised by MU-MIMO

The Rank Indicator is the foundational feedback parameter that precedes PMI selection. It specifies the number of independent spatial layers the channel can support.

• Determines PMI dimensionality: A rank-2 channel requires a precoding matrix with 2 columns.
• Calculated via channel matrix analysis: The UE estimates the channel's condition number and spatial correlation to determine viable eigenmodes.
• RI overrides PMI: If the rank changes, the entire precoding codebook subset changes. The gNB cannot apply a rank-4 PMI if the UE reports RI=1.

In 5G NR, the RI is reported either wideband or per sub-band, directly constraining the PMI search space.

The Channel Quality Indicator quantifies the signal-to-interference-plus-noise ratio (SINR) the UE expects after applying the recommended PMI and RI.

• Post-processing metric: CQI assumes the gNB will use the reported PMI. It reflects the effective SINR per codeword.
• Modulation and Coding Scheme (MCS) mapping: A CQI index of 15 maps to 64QAM with a coding rate of 948/1024, while index 1 maps to QPSK.
• PMI-CQI coupling: A sub-optimal PMI selection leads to a lower reported CQI, reducing throughput even if the raw channel conditions are good.

The CQI provides the gNB with the rate control feedback necessary to select a transport block size that meets the target Block Error Rate (BLER), typically 10%.

The codebook is the finite set of precoding matrices known to both the UE and gNB. The PMI is simply an index into this pre-defined table.

• Type I Single-Panel: Standard codebook for up to 32 ports, using a grid of beams with co-phasing. Suitable for low to medium spatial resolution.
• Type II Port Selection: High-resolution codebook using linear combination of beams with amplitude and phase quantization per sub-band. Provides significant MU-MIMO gains.
• DFT-based structure: Most codebooks are based on Discrete Fourier Transform vectors to match uniform linear array steering vectors.

The 3GPP TS 38.214 specification defines the exact codebook generation formulas. The UE searches this codebook to find the matrix that maximizes a metric like mutual information or SINR.

The CSI-RS is the downlink reference signal that enables the UE to measure the channel and derive PMI, RI, and CQI.

• Configurable density: CSI-RS can be sparse (e.g., every 5 ms) for low mobility or dense for high-resolution feedback.
• Port mapping: Each CSI-RS port corresponds to a virtualized antenna element. A 32-port CSI-RS allows the UE to estimate a 32xN channel matrix.
• Beamformed vs. non-beamformed: In 5G, CSI-RS can be precoded with a coarse beam to improve measurement SNR, or transmitted wide-beam for full cell coverage.

The accuracy of PMI feedback is fundamentally limited by the channel estimation quality derived from these reference signals. Pilot contamination in multi-cell scenarios degrades this estimate.

The CSI Report is the aggregate feedback message containing PMI, RI, CQI, and optional Layer Indicator (LI). It is transmitted on the PUCCH or PUSCH.

• Periodic reporting: Configured via RRC, sent on PUCCH. Low overhead but limited payload size.
• Aperiodic reporting: Triggered by DCI, sent on PUSCH. Allows large, detailed reports including sub-band PMI.
• Semi-persistent reporting: A hybrid mode activated/deactivated by MAC CE.
• Wideband vs. sub-band: PMI can be reported for the entire bandwidth or per sub-band for frequency-selective precoding.

The CSI report timing and content are critical for Massive MIMO and MU-MIMO performance, where stale or coarse PMI leads to inter-user interference.

Codebook Subset Restriction is a gNB mechanism that limits the PMI indices a UE is allowed to report, effectively blacklisting certain precoding matrices.

• Interference management: In MU-MIMO, the gNB can restrict PMIs that would cause excessive interference to co-scheduled users.
• Hardware limitations: Some antenna configurations cannot generate certain beam patterns. Restriction prevents the UE from requesting impossible precoders.
• Signaled via RRC: A bitmap indicates which PMIs are valid. The UE's codebook search is constrained to this subset.
• Impact on CQI: Restricting the optimal PMI forces the UE to report a sub-optimal matrix, which lowers the reported CQI and thus the achievable rate.

This mechanism gives the scheduler final control over the spatial transmission strategy.