Channel Quality Indicator (CQI)

CQI is not a raw signal measurement but an indexed recommendation reported by the UE to the gNB. Each CQI index (1-15 in 5G NR) maps directly to a specific modulation order (QPSK, 16QAM, 64QAM, 256QAM) and target code rate.

• CQI 1: QPSK, very low code rate (most robust, lowest throughput)
• CQI 7: 16QAM, moderate code rate
• CQI 15: 256QAM, high code rate (least robust, highest throughput)

The mapping assumes a Block Error Rate (BLER) target of 0.1, meaning the UE selects the highest MCS it can decode with 90% reliability.

CQI can be reported at different frequency granularities to balance feedback overhead against scheduling precision:

• Wideband CQI: A single value averaged over the entire system bandwidth. Low overhead but ignores frequency-selective fading.
• Sub-band CQI: The bandwidth is partitioned into multiple sub-bands, and a separate CQI is reported for each. This enables frequency-selective scheduling, where the scheduler allocates resource blocks in the sub-bands with the best channel quality.
• Best-M Reporting: The UE reports CQI only for the top-M sub-bands to further reduce uplink control information payload.

The network configures the UE to report CQI either periodically on PUCCH or aperiodically on PUSCH, triggered by a DCI grant:

• Periodic Reporting: Configured via RRC with a fixed interval (e.g., every 5ms). Suitable for stable, low-mobility scenarios.
• Aperiodic Reporting: Triggered on-demand by the gNB when a large data burst arrives. Allows the scheduler to obtain a fresh CQI just before a high-throughput transmission.
• Semi-Persistent Reporting: A hybrid mode activated/deactivated by MAC CE, balancing the low latency of periodic reporting with the flexibility of aperiodic triggers.

In Deep Reinforcement Learning for RAN optimization, the CQI vector across users is a critical component of the state space. It provides the agent with a compressed representation of the instantaneous channel conditions without requiring raw IQ samples.

• A DRL agent observes the CQI distribution across UEs to infer cell-edge vs. cell-center conditions.
• The agent learns to correlate historical CQI sequences with future throughput to perform predictive link adaptation.
• CQI variance can indicate mobility: a rapidly fluctuating CQI suggests a fast-moving UE, prompting the agent to prioritize robust MCS selection over spectral efficiency.

CQI reporting consumes precious uplink control channel resources. To manage this, 5G NR employs differential compression:

• A wideband CQI is reported as an absolute index.
• Sub-band CQIs are reported as 2-bit differential offsets relative to the wideband value, indicating whether the sub-band is slightly better (+1), equal (0), or slightly worse (-1 or -2).
• This hierarchical encoding reduces the payload size from hundreds of bits to tens of bits per reporting instance, preserving uplink capacity for user data.

The UE internally estimates the Signal-to-Interference-plus-Noise Ratio (SINR) on reference signals (CSI-RS) and then maps this to a CQI index using a standardized lookup table. This mapping is non-linear and vendor-specific in its implementation.

• Effective SINR Mapping: The UE combines per-subcarrier SINR estimates into a single effective SINR using methods like Exponential Effective SINR Mapping (EESM) or Mutual Information Effective SINR Mapping (MIESM).
• The mapping accounts for the UE's receiver capabilities, such as MMSE-IRC interference rejection, meaning two UEs with the same raw SINR may report different CQI values based on their receiver sophistication.

The process of dynamically selecting the Modulation and Coding Scheme (MCS) based on real-time CQI reports. When a UE reports a high CQI index (e.g., 15 in LTE), the scheduler assigns 64QAM with a high code rate; a low CQI triggers QPSK with robust coding.

• Outer Loop Link Adaptation (OLLA): Corrects CQI estimation errors by tracking HARQ ACK/NACK statistics
• Inner Loop: Selects MCS directly from the reported wideband or sub-band CQI
• Target BLER: Typically 10% for initial transmissions, balancing throughput and retransmission overhead

The fundamental physical-layer metric that CQI attempts to quantify. SINR measures the power of the desired signal divided by the sum of interference and thermal noise. CQI reporting effectively compresses the continuous SINR distribution into a discrete index.

• SINR-to-CQI Mapping: Vendor-specific lookup tables translate measured SINR to a 4-bit CQI value
• Effective SINR: Accounts for frequency selectivity and receiver processing gains
• Reference Signals: CSI-RS or CRS provide the basis for SINR estimation at the UE

The direct output of the CQI-to-MCS mapping process. The MCS table defines the combination of modulation order (QPSK, 16QAM, 64QAM, 256QAM) and code rate (ratio of information bits to total transmitted bits) applied to each resource block allocation.

• 5G NR MCS Table: Supports up to 256QAM (MCS index 28) with spectral efficiency exceeding 7.4 bps/Hz
• Low-SE MCS: Reserved for ultra-reliable low-latency communication (URLLC) with target BLER of 10^-5
• Adaptive MCS Switching: Occurs every transmission time interval (TTI) based on the latest CQI report

AI-driven approaches that forecast future CQI values to enable proactive link adaptation rather than reactive MCS selection. DRL agents learn to predict channel degradation before it occurs, reducing the latency penalty of waiting for explicit CQI reports.

• State Space: Historical CQI sequences, HARQ statistics, and UE mobility vectors
• Action Space: Preemptive MCS selection or scheduling decisions
• Reward Function: Maximizes throughput while constraining BLER below target
• Advantage: Compensates for CQI reporting delay (4–8 ms in LTE) in high-mobility scenarios

Hybrid Automatic Repeat Request (HARQ) provides a closed-loop correction mechanism that compensates for CQI estimation errors. When the initial transmission fails despite an optimistic CQI report, HARQ retransmissions with incremental redundancy recover the block.

• CQI Overestimation: Leads to aggressive MCS selection and increased first-transmission BLER
• CQI Underestimation: Results in conservative MCS and underutilized spectral efficiency
• HARQ Feedback: Used by OLLA to apply a dynamic back-off offset to reported CQI values, converging to the target BLER over multiple transmissions