Apparatus and method for controlling transmission power based on reinforcement learning
A method of controlling transmission power for wireless communication includes obtaining detected transmission power; generating a state variable and a reward variable of a reinforcement learning model based on the detected transmission power, threshold transmission power, and a channel state; and training a reinforced learning agent based on the state variable and the reward variable to output an action variable of the reinforcement learning model representing the transmission power.
1. A method of controlling transmission power for wireless communication, the method comprising:
obtaining detected transmission power;
generating a state variable and a reward variable based on the detected transmission power, a threshold transmission power, and a channel state; and
training a reinforced learning agent based on the state variable and the reward variable to output an action variable representing the transmission power,
wherein the training of the reinforced learning agent comprises generating, by the reinforced learning agent, the action variable based on the state variable and the reward variable, and
wherein the generating of the action variable comprises randomly generating the action variable with a probability ε, and greedily generating the action variable with a probability (1−ε).
2. The method of claim 1 , wherein the generating the state variable and the reward variable comprises:
calculating a transmission power residual rate of a unit period based on the threshold transmission power and the detected transmission power.
3. The method of claim 2 , wherein the generating of the state variable and the reward variable comprises:
obtaining an environment variable based on at least one communication parameter indicating the channel state; and
calculating the state variable based on the transmission power residual rate and the environment variable.
4. The method of claim 2 , wherein the generating the state variable and the reward variable comprises:
calculating the reward variable as a positive value based on the transmission power residual rate and the channel state when the transmission power residual rate is positive.
5. The method of claim 4 , wherein the calculating of the reward variable comprises:
calculating an average error rate during the unit period; and
calculating the reward variable based on the transmission power residual rate and the average error rate.
6. The method of claim 1 , wherein the training of the reinforced learning agent further comprises:
gradually reducing the probability ε.
7. The method of claim 1 , wherein the greedily generating of the action variable comprises:
setting a range of transmission power based on a transmission power of a previous unit period;
calculating a plurality of Q-values of Q-learning respectively corresponding to a plurality of transmission power candidates included in the range of transmission power;
selecting one transmission power candidate from among the transmission power candidates based on the plurality of Q-values; and
generating the action variable and updating a Q-table based on the selected transmission power candidate.
8. The method of claim 7 , wherein the range of transmission power includes the transmission power of the previous unit period.
9. The method of claim 7 , wherein the selecting the transmission power candidate comprises:
applying a weight to at least one of the plurality of transmission power candidates, the weight equal to or less than the threshold transmission power; and
selecting, as the selected transmission power candidate, a transmission power candidate corresponding to the largest sum of a weight and a Q-value from among the transmission power candidates.
10. The method of claim 1 , wherein the threshold transmission power is defined based on a specific absorption rate (SAR).
11. The method of claim 1 , further comprising:
adjusting the transmission power based on the action variable.
12. An apparatus comprising:
a memory configured to store instructions; and
at least one processor configured to communicate with the memory and, by executing the instructions, control transmission power for wireless communication,
wherein, to control the transmission power, the at least one processor is configured to
obtain detected transmission power;
generate a state variable and a reward variable based on the detected transmission power, a threshold transmission power, and a channel state; and
train a reinforced learning agent based on the state variable and the reward variable to output an action variable representing the transmission power,
wherein the at least one processor is configured to calculate a transmission power residual rate of a unit period based on the threshold transmission power and the detected transmission power to generate the state variable and the reward variable.
13. The apparatus of claim 12 , wherein, to train the reinforced learning agent, the at least one processor is further configured to:
set a range of transmission power based on a transmission power of a previous unit period,
calculate a plurality of Q-values of Q-learning respectively corresponding to a plurality of transmission power candidates included in the range of transmission power,
select one transmission power candidate from among the transmission power candidates based on the plurality of Q-values, and
generate the action variable and update a Q-table based on the selected transmission power candidate.
14. A method of controlling transmission power for wireless communication, the method comprising:
obtaining detected transmission power; and
training a reinforced learning agent, based on the detected transmission power, a threshold transmission power, and a channel state, to output an action variable representing the transmission power,
wherein the training of the reinforced learning agent comprises
setting a range of transmission power based on a transmission power of a previous unit period;
calculating a plurality of Q-values of Q-learning respectively corresponding to a plurality of transmission power candidates included in the range of transmission power;
selecting one transmission power candidate from among the plurality of transmission power candidates based on the plurality of Q-values; and
generating the action variable and updating a Q-table based on the selected transmission power candidate.
15. The method of claim 14 , wherein the range of transmission power comprises the transmission power of the previous unit period.
16. The method of claim 15 , wherein the selecting of the transmission power candidate comprises:
applying weights to at least one of the plurality transmission power candidates, the weights equal to or less than the threshold transmission power; and
selecting, as the selected transmission power candidate, a transmission power candidate corresponding to the largest sum of a weight and a Q-value from among the transmission power candidates.
17. The method of claim 14 , wherein the threshold transmission power is defined based on a specific absorption rate (SAR).
18. The method of claim 14 , further comprising:
adjusting the transmission power based on the action variable.