PMI — Precoding matrix indicator
Space — layers, ports, reporting
Where it sits
What it is
What it is. An index into a codebook of precoding matrices — the device's compressed suggestion for how the transmitter should aim. TS 38.214 clause 5.2.2.2, which is by some margin the longest sub-clause in that document.
| Codebook | Clause | Idea | Report size |
|---|---|---|---|
| Type I single-panel | 5.2.2.2.1 | Pick one beam from a grid, plus a co-phasing term | Small |
| Type I multi-panel | 5.2.2.2.2 | The same, across several panels | Small |
| Type II | 5.2.2.2.3 | Combine several beams with amplitudes and phases — a much better approximation | Large |
| Type II port selection | 5.2.2.2.4 | Type II over a reduced port set, when reciprocity gives a head start | Medium |
| Enhanced Type II | 5.2.2.2.5–.7 | Compressed in frequency, so wideband reports get cheaper | Medium |
| Type II for CJT | 5.2.2.2.8–.9 | One report describing several transmission points coherently — typeII-CJT-r18 | Large |
| Predicted PMI | 5.2.2.2.10–.11 | Precoders for slots that have not happened — csi-InferencePrediction-r19 | Large |
The whole table is one argument getting more expensive. A precoder that is slightly wrong wastes a little power with four antennas and a great deal with sixty-four, so each generation spends more uplink bits to describe the channel more exactly — until Release 19, where the answer stops being describe it better and becomes describe it sooner than it happens (PDSCH §13.2).
And PMI is only ever a suggestion. Nothing obliges the base station to use the matrix the device named — see precoding. The codebook is a shared vocabulary for describing a channel, not a contract about what will be transmitted.
Read on
This concept was first written up in ref-mimo, which reads the whole group as one argument.
Before this concept, the hierarchy says to learn the following — the full chain, in order: