On the Viability of Using a Subset of Transmitter-Observation Receivers for Training a Common DPD in Fully Digital MIMO Transmitters

This paper investigates the viability of applying the same, common set of DPD coefficients, trained using a reduced number of TORs, to linearize all RF chains in fully digital massive MIMO transmitters. It is shown that if all RF chains are operated at same output power, the common set of DPD coefficients can be found by averaging coefficients found by training each chain on its own. This suggests that only a few chains may be needed for training, provided they are a representative sample. Experimental results are conducted where 1 and 3 chains are used for training. It is found that training for 1 chain, significant variations in NMSE and ACPR up to 9 dB across the chains are realized. For training with 3 chains, the common set of DPD coefficients can reduce the variation to 1-2 dB. Finally, after over-the-air combining, excellent linearization performance is found for 3 chains.