Design And Verification Of Small Intelligent Signal Processing Platform Based On Software Defined Radio

The development of Software Defined Radio(SDR)has led to a more flexible and scalable wireless communication solution,which has been widely used in the civilian and military field.In recent years,the rapid development of artificial intelligence and the outstanding problems about radio spectrum tense and more complicated communication environment,the smart radio technology has gradually become a hot research topic.In power and volume limited occasions,small intelligent Software Defined Radio equipment has been taken seriously.Especially high integration embedded system ZYNQ on a chip,it provides a new solution for small intelligent Software Defined Radio applications.The design methods of intelligent,high integration,miniaturized platform are described in detail in this paper without reducing real-time performance and universality of platform.A high-speed miniaturization design scheme of intelligent signal processing structure is proposed in this paper,which is mainly based on the high performance chip ZYNQ(ARM + FPGA)with Software Defined Radio thought.The key support modules are decomposed.The hardware and software co-design and implementation of high-speed sampling and multi-rate transformation,high-speed transmission interface and deep learning accelerator are studied in depth,and integrated experiments are designed to carry out platform verification.Firstly,the basic knowledge involved in the platform design,such as high-speed sampling and transmission,deep learning hardware acceleration,is briefly described,which lays a theoretical foundation for the design and implementation of platform.Secondly,in view of the sampling rate of current small Software Defined Radio platform which is not high,limited transmission rate and other problems,high-speed sampling AD/DA and high-speed transmission interface is used and a complete hardware and software drive to meet the higher signal bandwidth as far as possible is designed to make the radio frequency wider.The JESD204 B transmission module is designed on FPGA to ensure the high-speed and stable transmission of sampled data,the multi-rate transformation module and the Aurora transmission module are designed to meet the demand of the post-stage transmission bandwidth of various signal processing tasks.Thirdly,in recent years,there has been a lot of theoretical verification of deep learning in traditional signal processing field,but there is no reconfigurable,flexible and extensible deep learning processor in the market that can meet the needs of Software Defined Radio.In view of this requirement,the deployment of deep learning applications that consume more computing resources are considered in the design of platform.The reconfigurable deep learning accelerator(DPU)is introduced into the FPGA of the platform to meet the real-time intelligent processing tasks of signal,and the upper Linux driver and compilation tools are provided to facilitate the deployment of custom deep learning models.Fourthly,the open source operating system Linux is transplanted on the ARM terminal according to the ease of use of the platform,and various Linux drivers corresponding to the hardware are realized.The flexibility and ease of use of Linux greatly facilitate the development and transplantation of Software Defined Radio applications.Finally,the application of modulation signal recognition based on deep learning is deployed on the platform to verify the platform function.The recognition model is trained on the PC side and the model after training is compressed.The recognition accuracy reaches 90.02% after deployment on the platform,which only loses about 2%compared with that before deployment.At the same time,a pair of communication transceivers based on DPSK modulation are designed to verify functional completeness of the high-speed transmission interface,the AD/DA acquisition module and the multirate conversion module of the platform.The transmission bandwidth is up to 12.5Gbps and the performance of the platform meets the requirements of the project.New drivers and algorithms can be transplanted to this platform in different signal processing occasions quickly through Linux,which conforms to the Software Defined Radio idea of flexibility and expandability.