To maximize the throughput of frequency-selective multicast channel, the minimum-phase Finite Impulse Response (FIR) precoder design is investigated in this paper. This problem can be solved in two steps. Firstly, we focus on designing a nonminimum-phase FIR precoder under the criterion of maximizing the throughput, and develop two efficient algorithms for the FIR precoder design from perspectives of frequency domain and time domain. In the second step, based on the theory of spectral factorization, the nonminimum-phase FIR precoder is transformed into the corresponding minimum-phase FIR precoder by a classic iterative algorithm without affecting the throughput. Numerical results indicate that the achievable rate of the proposed design has remarkable improvement over that of existing schemes, moreover, the group delay introduced by the FIR precoder is minimized.
对于非协作通信场景下辐射源识别(SEI)问题,基于人工射频指纹特征(Radio Frequency Fingerprints,RFF)的识别方式准确率不高,基于深度学习的方法又对训练数据量有过高的要求。为了克服该问题,提出一种结合了人工射频指纹特征的基于贝叶斯卷积神经网络(CNN)的半监督SEI算法,将一个回归拟合信号双谱的直方图特征的CNN嵌入一个SEI的贝叶斯CNN中,并通过基于模糊度的半监督学习方法进一步降低算法对标签训练集的依赖性。在模拟数据集和真实数据集中的实验结果表明,在标签训练集规模为500~4 500条数据时,提出的方法比端到端的卷积神经网络识别方法的识别率提高了5%~20%。
