RAN2 agrees on RSRP difference per sliding window for AI/ML performance monitoring

RAN2 is the radio layer 2 and 3 protocols working group. Here, reference signal received power (RSRP) difference is the absolute gap between predicted and actual values. In Release 20 artificial intelligence and machine learning (AI/ML) mobility, that gap serves prediction performance monitoring. The agreed calculation applies to each sliding prediction window (PW). The session chair said an earlier agreement left no overlapping-window issue. OPPO said overlapping windows posed no issue because "only the last instance of PW is calculated".

Source: R2-2605831, Vice Chairman (Samsung), 3GPP TSG RAN WG2 #135, meeting report, agenda item 8.3.3.

For one timing case, NTT DOCOMO proposed counting only prediction windows fully inside the monitoring window (MW). Apple, Ofinno, ASUSTeK, Nokia, MediaTek and Tejas Networks raised questions about using the metric for indirect prediction. Their questions concerned "averaging windows and its applicability to relative measurement events (e.g., Event A3)".

Source: R2-2605098, Apple (with Ofinno, ASUSTeK, Nokia, MediaTek, Tejas Networks), 3GPP TSG RAN WG2 #135. © Apple, Ofinno, ASUSTeK, Nokia, MediaTek, Tejas Networks.

InterDigital's first option averages absolute gaps between actual and predicted RSRP over N results. Its second option uses only the last result's absolute gap. Separately, ZTE called "two different definitions of RSRP difference for one PW" unnecessary. Our reading: the recorded agreement does not choose between InterDigital's averaged and last-result options.

Source: R2-2605766, ZTE, 3GPP TSG RAN WG2 #135. © ZTE.