基于深度学习的泊松散弹噪声受限光子计数通信系统速率优化研究

Rate Optimization of Poisson Shot Noise Limited Photon Counting System Based on Deep Learning

泊松散弹噪声具有非线性、非加性的性质,因而受到研究者的广泛关注。20世纪80年代以来,大量关于泊松信道速率和容量的成果被提出。然而,由于泊松噪声的特殊性质,导致速率公式复杂,过去的研究成果大多集中在理论极限性能的推导上下界的问题上。提出了一种基于深度学习的方法,针对复杂的速率公式,给出了非理想条件下的最优解,对于实际系统的性能有一定的参考价值。并且,深度学习较传统梯度下降搜索算法有更高的泛用性和更快的速度。

Poisson shot noise has the characteristics of non-linearity and non-additiveness,and has attracted widespread attention of researchers.Since the 1980s,a large number of results on Poisson channel rate and capacity have been proposed.However,due to the special nature of Poisson noise,the rate formula is complicated.Most of the previous research results have focused on the derivation of the theoretical limit performance.In this paper,a method based on deep learning is proposed.For the complex rate formula,the optimal solution under non-ideal conditions is given,which has certain reference value for the performance of the actual system.Moreover,deep learning has higher versatility and faster speed than traditional gradient descent search algorithms.