Delay Sensitive Services In Multichannel Random Access Networks With Selfish Users

Sammanfattning: Because of the scarcity of spectrum, it is becoming more and more important to use the spectrum efficiently. Dynamic spectrum allocation schemes are proposed to increase the efficiency of spectrum usage. Meanwhile, delay-sensitive services like voice service are still a strong contributor to operator revenues and constitute a significant portion of user demand. Therefore, it is valuable to study whether delay-sensitive services like voice service can be provided in multichannel random access network, in which multiple operators use the unlicensed spectrum and decentralized dynamic spectrum allocation scheme is adopted. In this study, two User Reinforcement Learning schemes are proposed to model the behavior of selfish users in the multichannel random access network. We analyze the performance of these two schemes and compare them with Operator Reinforcement Learning scheme and Operator Non-Reinforcement Learning scheme in terms of delay, jitter, packet loss rate, consecutive packet loss rate, throughput per channel and utility per user. We find that both the "User Reinforcement Learning Non-Retransmission" scheme and the "Operator Reinforcement Learning" scheme can provide a satisfactory quality service under a moderately high load.