Fair Spectrum Sharing

Through spectrum and interference management, wireless communications can coexist with radar. If a high degree of coordination between the two systems can be achieved, the joint design of radar waveforms and communication codebooks can be used to prevent the two systems from interfering with each other. If time synchronization is not possible, the radar system can design transmit power and receive filters while the MIMO communication system can design space-time codebooks to reduce mutual interference. The design strategy is to maximize the information rate of the communication system while ensuring that the received signal-to-disturbance ratio of each radar resolution bin exceeds a specified level. A control center with coordination capabilities is required or, under certain security restrictions, a direct exchange of information between radar and communication systems is required.

Project Work At Baylor University

The project work aims to investigate deep reinforcement learning (DRL) methods for self-adaption and self-optimization of wireless communication signal transmission under complex channel conditions. Successful methods can enhance heterogeneous wireless systems to efficiently use spectrum and reduce mutual interference. The approach is different from the current state-of-the-art methods for spectrum and interference management and should overcome current obstacles. States in DRL are constructed according to the communication environment, which may include channel usage conflict and receiver interference. However, we will not include channel state information in the system state. It is difficult to estimate user channels that change over time, and a lot of feedback is required. It is almost impossible to estimate interference channels between heterogeneous systems. Multiple cognitive transmitters are agents that learn to adopt good policies to take advantageous actions while observing the state. The actions taken should determine what state the system will be in next. However, in most current reinforcement learning methods for dynamic spectrum access, the state only reflects the channel occupancy of the primary users (incumbent users of the spectrum) regardless of the transmission of the cognitive users. Therefore, the state transition is controlled by the incumbent users’ spectrum usage and switching, and has nothing to do with the operation of the cognitive users. We will design the system state of DRL to reflect the effect of transmission decisions. This can be done by including (historical) mutual interference in the state, which is directly affected by the (historical) transmission power levels of the cognitive users. Currently, most reinforcement learning methods for spectrum management have rewards that are imminent and myopic. The project work will consider time-related factors in the design of the reward function. This will ensure that the full benefits of DRL can be realized.