To have clear understanding of RI, you have to understand the detailed concept of MIMO, Channel Status Information Matrix and mathematical definition of Rank of a matrix. Depending on instruction from the network, UE may periodically or aperiodically measure RI and report it to Network. Refer to CQI,PMI,RI Report section for this.
Let's build RI up in three steps. First we look at what the number means physically, and why a UE with good antennas still reports rank 1 at times. Then we look at the report itself: the range of values, the number of bits and the transmission modes that carry it. Finally we look at when the UE sends RI, and how the eNB reads it together with PMI and CQI.
- What does the RI value tell the network ?
- Which RI values can the UE report, and how many bits does it take ?
- When does the UE report RI, and how does the eNB use it ?
- Reference
What does the RI value tell the network ?
RI is a single small number, but it answers a basic MIMO question for the next downlink transmission: how many layers can the channel carry right now. The paragraphs below give the intuitive answer first. After that we connect it to the definition in 36.213 and to the SINR the UE sees.
But I will explain the practical/intuitive meaning of RI here.
In simple words, RI is an indicator showing how well multiple Antenna work. What do you mean by "how well the multiple Antenna work ?". We usually say "Each of the multiple antenna (e.g, each antenna in MIMO configuration) works well if the signal from each antenna has NO correlation to each other". "No correlation" implies "no interference to each other".
Maxmum RI value is very closely related to the number of Antenna. Maximum RI is same as number of antenna on each side if the number of Tx antenna and Rx antenna is same. If the number of Tx and Rx are different, the one with less antenna is the same as Max achievable RI.
Max RI means "No Correlation between the antenna", "No interference to each other", "Best Performance".
For example, in case of 2x2 MIMO, the RI value can be 1 or 2. When the value 2 in this case means "No Correlation between the antenna", "No interference to each other", "Best Performance". If the value is 1, it implies that the signal from the two Tx antenna is percieved by UE to be like single signal from single Antenna, which means the worst performance.
36.213 states the rule in two sentences. For spatial multiplexing, the UE determines an RI that corresponds to the number of useful transmission layers. For transmit diversity, RI is equal to one. The word useful matters here. The UE does not report how many layers the channel matrix could carry in theory. In practice, it reports the rank that it expects to give the highest throughput, together with the CQI for that rank.
In matrix terms, the rank of the channel matrix H is only the upper bound. The singular values of H show how strong each possible layer is. A layer whose singular value is small compared with the noise adds little data, so the UE leaves it out of the RI. This is why RI depends on SINR as well as on antenna correlation.
Let's compare two positions in the same cell. Near the eNB, with high SINR and low correlation, each extra layer adds a full stream of data, and the UE reports rank 2. At the cell edge, one layer with all the transmit power and the precoding gain often carries more data than two weak layers. So the UE reports rank 1 there, even with two uncorrelated antennas. Rank 1 gives the lowest spatial multiplexing gain, but it is not always the lowest throughput.
RI is the number of useful layers : the UE reports how many layers it can use now, not how many antennas it has.RI 1 is not a fault : at low SINR, one strong layer can carry more data than two weak ones, so the UE reports rank 1 even with uncorrelated antennas.The antenna count sets the ceiling : the smaller of the Tx and Rx antenna counts bounds RI, and the configuration and the UE capability can lower that bound further.
Which RI values can the UE report, and how many bits does it take ?
The range of RI is set before the UE measures anything. The eNB antenna ports and the UE capability together fix the maximum number of layers, and that maximum fixes the bit width of the report. Let's see how 36.212 ties the two together.
For transmission modes 3 and 4, the maximum number of layers is the minimum of the number of PBCH antenna ports and the layer limit of the ue-Category. If the eNB configures maxLayersMIMO-r10 for the cell, that parameter sets the maximum instead. For transmission mode 9, the number of configured CSI-RS ports takes the place of the PBCH antenna ports. The downlink MIMO capability reported by the UE takes the place of the ue-Category when the UE reports it.
The table below is based on 36.212 Table 5.2.2.6.1-2, which covers the wideband reports of transmission modes 3, 4, 8, 9 and 10. The last column comes from 36.212 Table 5.2.2.6-5 to Table 5.2.2.6-7.
Antenna ports | Maximum layers | RI bits | Bits to RI |
2 | 1 or 2 | 1 | 0 = rank 1, 1 = rank 2 |
4 | 1 or 2 | 1 | 0 = rank 1, 1 = rank 2 |
4 | 4 | 2 | 00 = rank 1 up to 11 = rank 4 |
8 to 32 | 1 or 2 | 1 | 0 = rank 1, 1 = rank 2 |
8 to 32 | 4 | 2 | 00 = rank 1 up to 11 = rank 4 |
8 to 32 | 8 | 3 | 000 = rank 1 up to 111 = rank 8 |
Two consequences follow. A UE on a 2-port cell always sends a 1 bit RI, so the bit value 1 on the air means rank 2. And a UE category that supports 4 layers still sends 1 bit on a 2-port cell, because the cell is the tighter limit. When you decode a CSI report in a log, check the configured ports and maxLayersMIMO first. Otherwise the bit width will not match.
Only the spatial multiplexing modes carry RI. Transmission modes 3 and 4 always do, and transmission modes 8, 9 and 10 do when PMI/RI reporting is configured. In the other modes, the UE computes PMI and CQI assuming rank 1, and for transmit diversity 36.213 fixes RI at one.
The configuration fixes the bit width : 1 bit for up to 2 layers, 2 bits for 4 layers and 3 bits for 8 layers.The coded value is RI minus one : bit 0 means rank 1, and bits 11 mean rank 4.Only the spatial multiplexing modes carry RI : TM3 and TM4 report it, TM8, TM9 and TM10 report it when PMI/RI reporting is configured, and the other modes assume rank 1.
When does the UE report RI, and how does the eNB use it ?
Every RI report costs uplink resources, so the eNB configures when it comes. The RI also changes the meaning of the PMI and CQI that go with it. That is why the timing of the report and the content of the report are tied together.
On PUCCH, the eNB configures periodic RI reporting with ri-ConfigIndex. 36.213 Table 7.2.2-1B maps that index to a multiplier MRI of 1, 2, 4, 8, 16 or 32 and to a relative offset NOFFSET,RI. The RI period is MRI times the wideband CQI/PMI period Npd. So RI never comes more often than the wideband CQI. This keeps the RI overhead low, and it fits the usual case where the rank changes more slowly than the channel quality.
On PUSCH, an aperiodic CSI request makes the UE send RI in the same report as CQI and PMI. When RI goes on a PUSCH with UL-SCH data, 36.212 multiplexes the rank information in all layers of all transport blocks of that PUSCH.
The eNB reads the RI first, because the rest of the report depends on it. For transmission modes 4, 8, 9 and 10, the UE calculates the reported PMI and CQI conditioned on the reported RI. For transmission mode 3, the CQI is conditioned on the reported RI. So a CQI value reported with RI 2 describes a two-layer transmission, and the same CQI value with RI 1 describes a single layer.
The specification does not force the eNB to use the reported rank. The scheduler can choose fewer layers, for example when the UE has little data in its buffer. Scheduling more layers than reported leaves the eNB without a CQI for that transmission, so the link adaptation has to guess.
RI comes less often than CQI on PUCCH : its period is MRI times the CQI/PMI period, with MRI from 1 to 32.RI sets the context for PMI and CQI : the UE computes both conditioned on the reported rank, so they are read together.The eNB decides the final rank : RI is a recommendation, and the scheduler may choose fewer layers.