In this work, we study LQG control systems where one of two feedback channels is discrete and incurs a communication cost, measured as time-averaged expected length of prefix-free codeword. This formulation to motivates a rate distortion problem, which we restrict to a particular policy space and express as a convex optimization. The optimization leads to a quantizer design and a subseqent achievability result.
翻译:在这项工作中,我们研究LQG控制系统,其中两个反馈渠道中的一个是离散的,并产生通信成本,以无前缀代码的预期时间平均长度来衡量。这种表述方式是为了引发一个汇率扭曲问题,我们把它限制在特定的政策空间,并把它表述成一个螺旋优化。 优化导致一个量化器设计和一个子等值实现结果。