Types of beamforming in 5G:

Following is my understanding of types of beamforming in 5G:

1st one is Eigen based beamforming. Basically, UE (user equipment) sends SRSs (Sounding Reference Signals) in uplink, gNodeB measures them and computes the DL beam weighting factors. These are called dynamic beams because beams are following the UE in a very accurate way. This requires significant baseband processing on gNodeB side and it works only for TDD ( because it needs reciprocity between UL and DL). Here’s a picture of how those dynamic beams track the UE meaning that a beam can be positioned with precision anywhere inside the sector:

2nd one is Grid of Beams based on SSBs and csi-rs beams. gNodeB sends those beams periodically in downlink and UE measures and report them. Those beams are called static beams as their weights are predefined and stored by gNodeB. UE measures all beams and reports the strongest beams and so gNodeB will send PDSCH data on one or several strongest beams. In most cases one SSB is covered by 4 csi-rs beams (called refined beams) like in the picture below:

3rd one is Grid of Beams based on PMI (Precoding Matrix Indicator). gNodeB sends in downlink CSI-RS reference signals (one per each antenna port) and UE measures them with RX antennas and computes the PMI matrix (from the air interface matrix and with the help of SVD matrix decomposition UE is able to compute PMI matrix, diagonal matrix and equalizer matrix from the air interface matrix). This PMI is sent to gNodeB that uses as weights for beamforming. Keep in mind that this is from a codebook meaning that PMI options are limited (not like 1st type of beamforming) ie weights are available only in a limited combination (limited options as per codebook). PMI sent by UE is not exactly a matrix but some coefficients (i11; i12; i13 and i2). From those coefficients, gNodeB will constitute the PMI matrix. An example of PMI based grid of beams is in the next picture (each circle represents a beam so all circles together cover the whole sector). For more details on what N1, N2, O1 and O2 represent please check sharetechnote's page. Please note that, contrary to the 1st option (EBB), in this 3rd case beams can be positioned only in the circles and not elsewhere (number of circles/beams locations depend on csi-rs ports and logical antenna ports and their position in H and V plane depends on AAU type and configuration, tilt and azimuth of sector).

Now one question you would have is how to identify in UE logs which type of beamforming was used? This is not an easy question but there are some answers. First of all, you will notice that UEs are configured with srs and nzp-csi-rs. From csi reports, go and look for reportquantity. If the value of reportquantity is only “cri-RI-PMI-CQI” (like in the following example) and there is only one reportconfigtoaddmodlist (like 0 in this case) then 3rd option is used (Grid of Beams based on PMI) like in this example:

 However, even in this case, it could be a dynamic switch between 1st option (Eigen based Beamforming using srs) and 3rd option used (Grid of Beams based on PMI) depending on UL SINR of SRS measured by gNodeB for TDD cases. That means in good RF conditions SRS based weights are used and in weak RF conditions PMI based weights are used. For example: for SRS UL SINR> -15 dB then SRS based weights are used; for SRS UL SINR < -15 dB then PMI based weights are used. This is one real scenario used by gNodeB to choose between SRS based weights or PMI based weights, in the future other scenarios may occur. This switching between SRS based and PMI based works only for TDDbecause it needs reciprocity between UL and DL. For FDD the SRS based method cannot be used.

Another case: if reportquantity is both of “cri-RSRP” and “ssb-lndex-RSRP” and there are several reportconfigtoaddmodlist configured then gNodeB is also prepared for the 2nd case (Grid of Beams based on SSBs and csi-rs beams) like in this example:

What you also need to keep in mind is that there are also 3 types of beamforming: digital beamforming (meaning that beam weights are processed in baseband), analog beamforming (meaning that weights are processed in analog/RF domain and cannot be changed with different weights per RB or RE for same antenna port like in digital domain) and hybrid beamforming (a mix of both analog and digital). Most AAUs (Active Antenna Units) are a mix of digital and analog beamforming more exactly a hybrid beamforming solution.

Thank you, again, for your time.