Method, apparatus and computer program

1 . An apparatus comprising:

at least one processor; and

at least one memory storing instructions that, when executed by the at least one processor, causes the apparatus to perform:

receiving, from the access node, beam usage information, wherein the beam usage information comprises a number of times a beam has been scheduled;

determining a first beam and a second beam for receiving at least one data packet from an access node of a network,

wherein determining the first beam and the second beam comprises:

obtaining one or more channel state information measurements, wherein the one or more channel state information measurements include signal strength information about N strongest beams, Reference Signal Received Power (RSRP) Reference Signal Received Quality (RSRQ), and Signal-to-Interference-plus-Noise Ratio (SINR); and

determining the first beam and the second beam based on the obtained one or more channel state information measurements and the beam usage information by:

determining a plurality of candidate second beams based on the obtained one or more channel state information measurements;

sending, to the access node, a list comprising the determined plurality of candidate second beams; and

receiving, from the access node, information identifying the second beam selected from the list comprising the determined plurality of candidate second beams,

wherein a machine learning model is utilized to determine at least the first beam, the machine learning model comprising a deep Q-network configured to output Q-values for each beam, and configured to select the first beam based on a highest Q-value subject to epsilon-greedy exploration in which a random beam is selected with an exploration probability,

wherein the second beam comprises an anchor beam configured as a safety net beam for retransmissions, wherein the anchor beam is selected as a strongest beam based on RSRP, RSRQ, or SINR indicated in the one or more channel state information measurements;

sending, to the access node, information identifying the first beam and information identifying the second beam;

receiving, from the access node, a re-transmission of the at least one data packet using the second beam when reception of the at least one data packet from the access node has failed using the first beam; and

perform model re-training of the machine learning model in response to receiving the re-transmission of the at least one data packet using the second beam, wherein re-training of the machine learning model includes providing feedback to the machine learning model based on the retransmission received via the anchor beam.