Power allocation in spectrum sharing systems is challenging due to excessive interference that the secondary system could impose on the primary system. Therefore, an interference threshold constraint is considered to regulate the secondary system's activity. However, the primary receivers should measure the interference and inform the secondary users accordingly. These cause design complexities, e.g., due to transceiver's hardware impairments, and impose a substantial signaling overhead. We set our main goal to mitigate these requirements in order to make the spectrum sharing systems practically feasible. To cope with the lack of a model we develop a coexisting deep reinforcement learning approach for continuous power allocation in both systems. Importantly, via our solution, the two systems allocate power merely based on geographical location of their users. Moreover, the inter-system signaling requirement is reduced to exchanging only the number of primary users that their QoS requirements are violated. We observe that compared to a centralized agent that allocates power based on full (accurate) channel information, our solution is more robust and strictly guarantees QoS requirements of the primary users. This implies that both systems can operate simultaneously with almost-zero inter-system signaling overhead.
翻译:由于二级系统可能对初级系统造成过度干扰,在频谱共享系统中的权力分配具有挑战性,因此,在监管二级系统的活动时,可以考虑干预门槛限制,但初级接收器应测量干扰,并相应告知二级用户。这造成了设计的复杂性,例如,由于收发报机硬件受损,造成设计的复杂性,并造成大量的间接信号。我们设定了减少这些要求的主要目标,以便使频谱共享系统切实可行。为了应对缺乏一种模式,我们为两个系统的持续电力分配开发了一种共存的深层强化学习方法。重要的是,通过我们的解决方案,两个系统仅根据用户的地理位置分配权力。此外,系统间信号要求被缩减为仅交换其QOS要求被违反的初级用户数目。我们注意到,与基于充分(准确的)频道信息分配权力的中央代理相比,我们的解决办法更加有力和严格地保证了主要用户的QOS要求。这意味着,两个系统可以同时运行,与几乎为零的系统间间接信号。