We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Information Theory

Title: Deep Reinforcement Learning in mmW-NOMA: Joint Power Allocation and Hybrid Beamforming

Abstract: High demand of data rate in the next generation of wireless communication could be ensured by Non-Orthogonal Multiple Access (NOMA) approach in the millimetre-wave (mmW) frequency band. Decreasing the interference on the other users while maintaining the bit rate via joint power allocation and beamforming is mandatory to guarantee the high demand of bit-rate. Furthermore, mmW frequency bands dictates the hybrid structure for beamforming because of the trade-off in implementation and performance, simultaneously. In this paper, joint power allocation and hybrid beamforming of mmW-NOMA systems is brought up via recent advances in machine learning and control theory approaches called Deep Reinforcement Learning (DRL). Actor-critic phenomena is exploited to measure the immediate reward and providing the new action to maximize the overall Q-value of the network. Additionally, to improve the stability of the approach, we have utilized Soft Actor-Critic (SAC) approach where overall reward and action entropy is maximized, simultaneously. The immediate reward has been defined based on the soft weighted summation of the rate of all the users. The soft weighting is based on the achieved rate and allocated power of each user. Furthermore, the channel responses between the users and base station (BS) is defined as the state of environment, while action space is involved of the digital and analog beamforming weights and allocated power to each user. The simulation results represent the superiority of the proposed approach rather than the Time-Division Multiple Access (TDMA) and Non-Line of Sight (NLOS)-NOMA in terms of sum-rate of the users. It's outperformance is caused by the joint optimization and independency of the proposed approach to the channel responses.
Comments: 20 pages (single Column), 9 figures. arXiv admin note: text overlap with arXiv:2205.06489
Subjects: Information Theory (cs.IT); Artificial Intelligence (cs.AI); Signal Processing (eess.SP)
Cite as: arXiv:2205.06814 [cs.IT]
  (or arXiv:2205.06814v1 [cs.IT] for this version)

Submission history

From: Abbas Akbarpour-Kasgari [view email]
[v1] Fri, 13 May 2022 07:55:48 GMT (666kb,D)

Link back to: arXiv, form interface, contact.