Method and device for transmitting and receiving wireless signal in wireless communication system
A method for transmitting data by a base station in a wireless communication system according to the present disclosure comprises: transmitting a reference signal (RS) to one or more user equipments (UEs); receiving, from the one or more UEs, channel quality indication (CQI) information based on the reception of the RS; determining a modulation and coding scheme (MCS) level on the basis of the CQI information; transmitting data to the one or more UEs in accordance with the MCS level; and receiving, from the one or more UEs, an acknowledgment/negative acknowledgment (ACK/NACK) with respect to the transmitted data, wherein the MCS level is determined on the basis of the CQI information and an offset determined according to a machine learning process, and the machine learning process for determining the offset is performed by configuring the selecting of one of a plurality of MCS offset values as an operation value of machine learning, configuring an error rate for the operation value as a state value of the machine learning, and configuring a processing rate at a level where the error rate satisfies a predetermined reference, as a compensation value of the machine learning.
1. A method of transmitting data by a base station in a wireless communication system, the method comprising:
transmitting a reference signal (RS) to one or more user equipments (UEs);
receiving channel quality indication (CQI) information based on reception of the RS from the one or more UEs;
determining a modulation and coding scheme (MCS) level based on the CQI information;
transmitting data to the one or more UEs in relation to the MCS level; and
receiving acknowledgement/negative acknowledgement (ACK/NACK) for the transmitted data,
wherein the MCS level is determined based on the CQI information and on an offset determined according to a machine learning process, and
wherein the machine learning process for determining the offset is performed by setting any one selected from among a plurality of MCS offset values to an operation value of machine learning, setting an error rate for the operation value to a state value of the machine learning, and setting a processing rate at a level at which the error rate satisfies a predetermined criterion to a reward value of the machine learning process.
2. The method of claim 1 , wherein the determining the offset comprises:
(1) obtaining a block error rate (BLER) as a current state of the machine learning;
(2) selecting any one of the MCS offset values as machine learning operations applied to the current state;
(3) repeating a predetermined number of times a process of calculating an expected reward value of each of the machine learning operations, updating a Q-table of the machine learning according to the calculated expected reward value, and recalculating the expected reward value based on the updated Q-table; and
determining one MCS offset value based on the updated Q-table according to the process of (3).
3. The method of claim 2 , wherein the current state of the machine learning is obtained based on a duration acquired by unequally quantizing a representation range of the BLER.
4. The method of claim 3 , wherein the unequally quantizing the representation range of the BLER comprises performing quantization by setting a first quantization unit of a first range near a target BLER in the representation range of the BLER to be smaller than a second quantization unit of a second range other than the first range.
5. The method of claim 3 , wherein the Q-table is a matrix determined based on the number of the MCS offset values and the quantized representation range of the BLER.
6. The method of claim 2 , wherein the determining the one MCS offset value comprises determining the one MCS offset value based on beta distribution having a first parameter and a second parameter, and wherein the first parameter increases based on reception of ACK and the second parameter increases based on reception of NACK.
7. The method of claim 1 , wherein the machine learning process for determining the offset comprises performing Q-learning.
8. The method of claim 1 , wherein the number of the MCS offset values is set to 2L+1, where L is the number of MCSs supported by the base station.
9. A base station for transmitting data in a wireless communication system, the base station comprising:
a radio frequency (RF) transceiver; and
a processor configured to:
control the RF transceiver to transmit a reference signal (RS) to one or more user equipments (UEs);
control the RF transceiver to receive channel quality indication (CQI) information based on reception of the RS from the one or more UEs;
determine a modulation and coding scheme (MCS) level based on the CQI information;
control the RF transceiver to transmit data to the one or more UEs in relation to the MCS level; and
control the RF transceiver to receive acknowledgment/negative acknowledgement (ACK/NACK) for the transmitted data,
wherein the MCS level is determined based on the CQI information and on an offset determined according to a machine learning process, and
wherein the machine learning process for determining the offset is performed by setting any one selected from among a plurality of MCS offset values to an operation value of machine learning, setting an error rate for the operation value to a state value of the machine learning, and setting a processing rate at a level at which the error rate satisfies a predetermined criterion to a reward value of the machine learning process.
10. The base station of claim 9 , wherein the determining the offset comprises:
(1) obtaining a block error rate (BLER) as a current state of the machine learning;
(2) selecting any one of the MCS offset values as machine learning operations applied to the current state;
(3) repeating a predetermined number of times a process of calculating an expected reward value of each of the machine learning operations, updating a Q-table of the machine learning according to the calculated expected reward value, and recalculating the expected reward value based on the updated Q-table; and
determining one MCS offset value based on the updated Q-table according to the process of (3).
11. The base station of claim 10 , wherein the current state of the machine learning is obtained based on a duration acquired by unequally quantizing a representation range of the BLER.
12. The base station of claim 11 , wherein the unequally quantizing the representation range of the BLER comprises performing quantization by setting a first quantization unit of a first range near a target BLER in the representation range of the BLER to be smaller than a second quantization unit of a second range other than the first range.
13. The base station of claim 11 , wherein the Q-table is a matrix determined based on the number of the MCS offset values and the quantized representation range of the BLER.
14. The base station of claim 10 , wherein the determining the one MCS offset value comprises determining the one MCS offset value based on beta distribution having a first parameter and a second parameter, and wherein the first parameter increases based on reception of ACK and the second parameter increases based on reception of NACK.
15. The base station of claim 10 , wherein the machine learning process for determining the offset comprises performing Q-learning.