Research On Blind Receiver Based On Deep Learning And Software Defined Radio

Adaptive modulation and coding systems are able to dynamically adjust transmission parameters based on channel states to improve spectrum utilization efficiency.In such a system,the receiver has to know in advance which modulation and coding types the transmitter is using in order to demodulate and decode the received signal correctly.In cognitive radio and electronic warfare,a blind receiver that can perform signal demodulation and decoding without modulation and coding information is of great value.Existing blind receivers usually suffer from demodulation and decoding errors resulting from low modulation and coding identification accuracy and bad extensibility due to the limitation of hardware platform.The thesis designs and realizes a blind receiver based on deep learning and software defined radio.The designed receiver uses deep learning algorithm instead of traditional feature based algorithms to achieve better accuracy of modulation and coding identification.Moreover,it adopts software defined radio instead of traditional circuits to facilitate the upgradation and extension of receiver functionsFirstly,this thesis introduces the research background and purpose.Moreover,the recent results of adaptive modulation and coding technology,deep learning and software defined radio are also presented.Then,the thesis summarizes the software defined radio platform and deep learning platform for later use.After that,the detailed design and realization of blind demodulator is introduced.The blind demodulator uses a modulation identifier to recognize the modulation type of the received signal,according to which a variable demodulator is reconfigured to complete the signal demodulation.The modulation identifier is realized based on deep learning with constellation representation and a convolutional neural network named Alex Net.The variable demodulator is realized based on Universal Software Radio Peripheral(USRP)and GNU Radio platforms.Moreover,the detailed design and realization of blind decoder is introduced.The blind decoder uses a code identifier to recognize the code of the received signal,according to which a variable decoder is reconfigured to complete the signal decoding.The code recognizer is realized based on deep learning with word vector representation and a convolutional neural network called TextCNN.The variable decoder is also realized based on USRP and GNU Radio platforms.Finally,the performance of blind receiver in real environment is introduced.The results show that the blind receiver achieves high recognition accuracy of modulation and coding types,exhibits stable and reliable receiving performance,and is easy to expand and upgrade.In addition,it has good performance in terms of recognition speed and switching speed among different modulation and coding types.