Method and apparatus for assigning frequency resource in non-terrestrial network
An operation method of a satellite in a non-terrestrial network may comprise: determining a frequency resource allocation order for allocating frequency resources to beams in consideration of a frequency band of a terrestrial system and degrees of interference to the terrestrial system; configuring a minimum performance condition for maintaining a service of the non-terrestrial network, the minimum performance condition being applied to each of the beams; configuring an operating condition of a multi-agent deep reinforcement learning for each beam controller of the satellite; and controlling the each beam controller to sequentially allocate the frequency resources to a managed beam according to the frequency resource allocation order while considering the minimum performance condition.
1 . An operation method of a satellite in a non-terrestrial network, the operation method comprising:
determining a frequency resource allocation order for allocating frequency resources to beams in consideration of a frequency band of a terrestrial system and degrees of interference to the terrestrial system;
configuring a minimum performance condition for maintaining a service of the non-terrestrial network, the minimum performance condition being applied to each of the beams;
configuring an operating condition of a multi-agent deep reinforcement learning for each beam controller of the satellite; and
controlling the each beam controller to sequentially allocate the frequency resources to a managed beam according to the frequency resource allocation order while considering the minimum performance condition,
wherein the operating condition includes a state and a learning objective, the state is set based on an interference power observed between terrestrial cells for the managed beam by the each beam controller, and the learning objective is configured such that a higher reward is obtained as a cumulative interference to the terrestrial system is reduced according to an action.
2 . The operation method according to claim 1 , wherein the determining of the frequency resource allocation order comprises:
calculating a distance between a cell center of the terrestrial system and a cell center of a terrestrial cell formed by each of the beams; and
determining the frequency resource allocation order in consideration of a level at which the distance affects the degree of interference.
3 . The operation method according to claim 2 , wherein in the determining of the frequency resource allocation order, as the distance decreases, the satellite gives a priority in the frequency resource allocation order to a beam forming a terrestrial cell corresponding to the distance.
4 . The operation method according to claim 1 , wherein the minimum performance condition is a minimum signal-to-interference-plus-noise-ratio (SINR) condition of uplink signals.
5 . The operation method according to claim 1 , wherein the configuring of the operating condition for the multi-agent deep reinforcement learning comprises:
configuring the each beam controller of the satellite to use the interference power between terrestrial cells, which is observed for the beam managed by the each beam controller, as the state constituting the operating condition;
allowing the each beam controller of the satellite to perform an operation of allocating one frequency resource among frequency resources allowed for the satellite as the action constituting the operating condition, so as to achieve the learning objective constituting the operating condition under the state; and
configuring the learning objective constituting the operating condition so that the higher reward is obtained as the cumulative interference to the terrestrial system is reduced according to the action.
6 . The operation method according to claim 5 , further comprising configuring a penalty to be given when the action does not satisfy a minimum performance condition of the non-terrestrial network as the learning objective constituting the operating condition.
7 . The operation method according to claim 1 , further comprising modifying, by the each beam controller of the satellite, a policy by updating weights of a neural network of the each beam controller.
8 . A satellite of a non-terrestrial network, comprising a processor, wherein the processor causes the satellite to perform:
determining a frequency resource allocation order for allocating frequency resources to beams in consideration of a frequency band of a terrestrial system and degrees of interference to the terrestrial system;
configuring a minimum performance condition for maintaining a service of the non-terrestrial network, the minimum performance condition being applied to each of the beams;
configuring an operating condition of a multi-agent deep reinforcement learning for each beam controller of the satellite; and
controlling the each beam controller to sequentially allocate the frequency resources to a managed beam according to the frequency resource allocation order while considering the minimum performance condition,
wherein the operating condition includes a state and a learning objective, the state is set based on an interference power observed between terrestrial cells for the managed beam by the each beam controller, and the learning objective is configured such that a higher reward is obtained as a cumulative interference to the terrestrial system is reduced according to an action.
9 . The satellite according to claim 8 , wherein in the determining of the frequency resource allocation order, the processor further causes the satellite to perform:
calculating a distance between a cell center of the terrestrial system and a cell center of a terrestrial cell formed by each of the beams; and
determining the frequency resource allocation order in consideration of a level at which the distance affects the degree of interference.
10 . The satellite according to claim 8 , wherein in the configuring of the operating condition for the multi-agent deep reinforcement learning, the processor further causes the satellite to perform:
configuring the each beam controller of the satellite to use the interference power between terrestrial cells, which is observed for the beam managed by the each beam controller, as the state constituting the operating condition;
allowing the each beam controller of the satellite to perform an operation of allocating one frequency resource among frequency resources allowed for the satellite as the action constituting the operating condition, so as to achieve the learning objective constituting the operating condition under the state; and
configuring the learning objective constituting the operating condition so that the higher reward is obtained as the cumulative interference to the terrestrial system is reduced according to the action.
11 . The satellite according to claim 10 , wherein the processor further causes the satellite to perform: configuring a penalty to be given when the action does not satisfy a minimum performance condition of the non-terrestrial network as the learning objective constituting the operating condition.
12 . The satellite according to claim 8 , wherein the processor further causes the satellite to perform: modifying, by the each beam controller of the satellite, a policy by updating weights of a neural network of the each beam controller.