A first UE may determine that a BPL with a second UE has degraded below a threshold and generate, based on the determination, a beam training frame for an on-demand beam training procedure. The beam training frame may include a first AGC symbol, SCI indicating that the beam training frame is for beam training, and one or more sets of an AGC symbol and beam sweeping reference signals. The first UE may transmit the beam training frame multiplexed with one or more data frames to the second UE. The second UE may receive the beam training frame for the on-demand beam training procedure based on a determination that the BPL with the first UE has degraded below the threshold.