| 摘要: |
| 随着信号采集设备的带宽越来越宽，大量感兴趣或者不感兴趣信号被捕捉，多信号的盲识别问题是一个难题，更是一个亟需解决的问题。传统的识别大都基于功率、频谱或相位等诸多先验知识进行模板匹配，但在全盲条件下对多信号进行自适应识别是一个更加复杂的问题。为此，提出了一种基于通用软件无线电外设（Universal Software Radio Peripheral,USRP）、片上射频网络（RF Network on Chips,RFNOC）和Keras的自适应信号盲识别算法。首先构造基于深度学习的神经网络，然后使用初始IQ数据、初始功率谱密度数据和快速傅里叶变换(Fast Fourier Transform,FFT)累积算法处理后的谱相关密度数据等三种不同的初始数据去训练它，利用其自适应性实现多信号的盲识别，最后通过基于USRP、RFNOC和Keras的软硬件验证了该算法的有效性和鲁棒性。 |
| 关键词: 盲信号识别 通用软件无线电外设 片上射频网络 神经网络 深度学习 |
| DOI： |
|
| 基金项目: |
|
| Adaptive Blind Signal Recognition with USRP, RFNOC and Keras |
| XU Sheng,LU Guangkuo |
| (1.Project Management Center,Navy Equipment Department,Beijing 100841,China;Southwest China Institute of Electronic Technology,Chengdu 610036,China) |
| Abstract: |
| The adaptive blind classification and recognition of the received communication signals in real time is a challenging but necessary task.Without prior knowledge of the received data,such as power,frequency,and phase,coupled with real-world concerns such as signal degradation and interference,the task of reconstructing the sent information is made even more complex.The proposed approach for blind classifying the live over-the-air signals received by the electronic warfare(EW) equipment,involves using universal software radio peripheral（USRP） plus RF network on chips(RFNOC) to train a series of neural networks by deep learning over a set of IQ,power spectrum density(PSD) and spectrum correlation density(SCD) data types.Finally,simulation results demonstrate the effectiveness and superiority of the proposed algorithm. |
| Key words: blind signal recognition universal software radio peripheral(USRP) RF network on chips(RFNOC) neural network deep learning |
《电讯技术》编辑部 