CSI Feedback
How CSI Feedback Closes the MIMO Loop
Closed-loop MIMO only works if the transmitter knows the channel. In 5G NR the base station cannot measure its own downlink channel directly in FDD, so it transmits CSI-RS pilots and asks the user equipment (UE) to report back. The UE estimates the matrix channel H between every transmit and receive antenna port, then collapses that estimate into three or four small integers: CQI selects the modulation and coding scheme (MCS) the UE believes it can decode at roughly 10 percent block error rate, RI tells the gNB how many independent spatial layers the channel can support, and PMI points to the codebook entry whose precoding vector best matches the channel's dominant eigenmode. The gNB combines these with its own scheduling logic to pick a transmission for the next slot.
The reporting is configurable along several axes. CSI-RS and the associated report can be periodic (fixed cadence on PUCCH), semi-persistent (activated and deactivated by MAC control elements), or aperiodic (triggered on demand by a DCI field, with the report carried on PUSCH). Frequency granularity ranges from wideband (one report for the whole carrier) to subband (one report per group of resource blocks), trading overhead against precision. Separate CSI-IM resources let the UE measure interference and noise so the reported CQI reflects the post-equalization SINR rather than raw signal power.
The fundamental tension is overhead versus fidelity. A finer report tracks the channel more accurately and supports aggressive multi-user MIMO, but consumes scarce uplink capacity and arrives with latency that may already be stale by the time the gNB uses it. This is why 5G defines both a low-overhead Type I codebook for beam selection and a high-fidelity Type II codebook for multi-user precoding, and why TDD systems often bypass explicit feedback altogether by exploiting channel reciprocity from uplink sounding.
Indicators and the Reporting Equations
η ≈ (modulation order Qm) × (code rate R) bits/symbol
Rank (RI) upper bound:
RI ≤ min(Nt, Nr) (transmit / receive antenna count)
PMI precoding applied at the gNB:
y = W × x, W = codebook entry selected by PMI
Channel coherence time (feedback validity):
Tc ≈ 0.4 / fD, fD = (v × fc) / c
Example: v = 120 km/h, fc = 3.5 GHz → fD ≈ 389 Hz → Tc ≈ 1.0 ms, so a report older than ~1 ms is largely stale.
5G NR CSI Report Quantities and Codebooks
| Quantity | Full Name | Typical Bits | What It Controls | Range / Notes |
|---|---|---|---|---|
| CQI | Channel Quality Indicator | 4 per codeword | Modulation & coding scheme | Index 0 to 15 (QPSK to 256-QAM) |
| PMI | Precoding Matrix Indicator | 4 to 200+ | MIMO precoding weights | Type I coarse, Type II fine |
| RI | Rank Indicator | 1 to 3 | Number of spatial layers | 1 to 8 layers (DL) |
| LI | Layer Indicator | 1 to 2 | Strongest layer for phase ref | Used with Type II |
| Type I | Single-beam codebook | ~10 to 20 total | SU-MIMO, beam mgmt | Low overhead, coarse |
| Type II | Multi-beam codebook | ~50 to 200+ total | MU-MIMO precoding | High overhead, 2 to 4 DFT beams |
Frequently Asked Questions
What is the difference between Type I and Type II CSI codebooks in 5G NR?
Type I reports a single dominant beam per layer using a handful of bits, giving coarse spatial resolution suited to single-user MIMO and beam management. Type II reports a linear combination of 2 to 4 orthogonal DFT beams per polarization, each with its own amplitude and phase, reconstructing the channel finely enough for accurate multi-user MIMO and interference nulling. A rank-1 Type II report can exceed 200 bits versus roughly 10 to 20 bits for Type I, so Type II is reserved for cell-edge or high-load multi-user cases.
How does CSI-RS relate to CSI feedback?
CSI-RS is the downlink pilot the gNB transmits so the UE can measure the channel; the UE estimates the per-port channel, computes CQI, PMI, and RI, then sends those quantized indicators back as CSI feedback. CSI-RS supports up to 32 antenna ports and can be periodic, semi-persistent, or aperiodic (DCI-triggered). Separate CSI-IM resources let the UE measure interference so the reported CQI reflects true post-equalization SINR rather than just received signal power.
Why is CSI feedback less reliable in high-mobility or FDD systems?
The report describes the channel at the measurement instant, but the channel decorrelates as the user moves; coherence time is roughly 0.4 / fD, so at 3.5 GHz and 120 km/h (fD ≈ 389 Hz) it is near 1 ms, and a report delivered a few milliseconds late causes precoding mismatch. TDD systems can exploit reciprocity and derive the downlink channel from uplink SRS, but FDD uses different frequencies where reciprocity fails, so explicit CSI feedback is mandatory despite its latency and quantization limits.