Channel prediction for MIMO systems based on reciprocity

An article of manufacture including a non-transitory computer-readable medium, having stored thereon program instructions that, upon execution by a computing system, cause the computing system to perform operations comprising:

• receiving, at a node of a radio access network, input signals on a plurality of wireless channels, wherein uplink channels in the plurality of wireless channels are respectively associated with downlink channels in the plurality of wireless channels, and wherein instances of the uplink channels and the downlink channels that are associated operate on different frequencies;
• determining, from the input signals, uplink characteristics for the uplink channels;
• providing, to a trained neural network, the uplink characteristics, wherein the trained neural network was iteratively trained, with a training data set of recorded uplink characteristics and recorded downlink characteristics, to form predictions of the recorded downlink characteristics from the recorded uplink characteristics, and wherein the trained neural network includes an encoder that maps the recorded uplink characteristics to latent vector representations, and a decoder that maps the latent vector representations to the predictions of the recorded downlink characteristics;
• receiving, from the trained neural network, a prediction of downlink characteristics for the downlink channels;
• and
• transmitting, on the downlink channels, output signals based on the prediction of downlink characteristics.

A computer-implemented method comprising:

• receiving, at a node of a radio access network, input signals on a plurality of wireless uplink channels associated with a first subset of antennas;
• determining, from the input signals, uplink characteristics for the wireless uplink channels;
• providing, to a trained neural network, the uplink characteristics, wherein the trained neural network was iteratively trained, with a training data set of recorded wireless channel characteristics, to form predictions of further wireless channel characteristics from the recorded wireless channel characteristics, and wherein the trained neural network includes an encoder that maps the recorded wireless channel characteristics to latent vector representations, and a decoder that maps the latent vector representations to the predictions of the further wireless channel characteristics;
• receiving, from the trained neural network, a prediction of wireless channel characteristics associated with a second subset of antennas;
• and
• communicating, by way of the second subset of antennas, in accordance with the wireless channel characteristics associated with the second subset of antennas.