DRL-based Channel Access in NR Unlicensed Spectrum for Downlink URLLC

TY - CONF

T1 - DRL-based Channel Access in NR Unlicensed Spectrum for Downlink URLLC

AU - Liu, Yan

AU - Zhou, Hui

AU - Deng, Yansha

AU - Allanathan, Arumugam N.

N1 - Publisher Copyright: © 2022 IEEE.

PY - 2022

Y1 - 2022

N2 - To improve the capacity of cellular systems without additional expenses on licensed frequency bands, the 3rd Gen-eration Partnership Project (3GPP) has proposed New Radio Unlicensed (NR-U). It should be noted that each node in NR-U has to perform the Listen-Before- Talk (LBT) operation before transmission to avoid collisions by other unlicensed radio access technologies (e.g., WiFi). Thus, packets transmissions are prone to delay due to the LBT channel access mechanism. How to achieve Ultra-Reliable and Low-Latency Communications (URLLC) requirements in NR-U networks under the coexistence with WiFi networks is of importance and extremely challenging. In this paper, we develop a novel deep reinforcement learning (DRL) framework to optimize the downlink URLLC trans-mission in the NR-U and WiFi coexistence system through dynamically adjusting the energy detection (ED) thresholds. Our results have shown that the NR-U system reliability has been improved significantly via the DRL compared to that without learning approaches, but with the sacrifice of WiFi system reliability. To address this, we redesigned the reward to take fairness into account, which guarantees the WiFi system reliability while improvina the NR- U system reliability.

AB - To improve the capacity of cellular systems without additional expenses on licensed frequency bands, the 3rd Gen-eration Partnership Project (3GPP) has proposed New Radio Unlicensed (NR-U). It should be noted that each node in NR-U has to perform the Listen-Before- Talk (LBT) operation before transmission to avoid collisions by other unlicensed radio access technologies (e.g., WiFi). Thus, packets transmissions are prone to delay due to the LBT channel access mechanism. How to achieve Ultra-Reliable and Low-Latency Communications (URLLC) requirements in NR-U networks under the coexistence with WiFi networks is of importance and extremely challenging. In this paper, we develop a novel deep reinforcement learning (DRL) framework to optimize the downlink URLLC trans-mission in the NR-U and WiFi coexistence system through dynamically adjusting the energy detection (ED) thresholds. Our results have shown that the NR-U system reliability has been improved significantly via the DRL compared to that without learning approaches, but with the sacrifice of WiFi system reliability. To address this, we redesigned the reward to take fairness into account, which guarantees the WiFi system reliability while improvina the NR- U system reliability.

KW - 5G NR-U

KW - Channel Access

KW - Deep Reinforcement Learning

KW - URLLC

KW - WiFi

UR - http://www.scopus.com/inward/record.url?scp=85146954480&partnerID=8YFLogxK

U2 - 10.1109/GLOBECOM48099.2022.10000882

DO - 10.1109/GLOBECOM48099.2022.10000882

M3 - Paper

AN - SCOPUS:85146954480

SP - 591

EP - 596

T2 - 2022 IEEE Global Communications Conference, GLOBECOM 2022

Y2 - 4 December 2022 through 8 December 2022

ER -