Research And Implementation Of The Key Technology Of Software Framework In Heterogeneous Signal Processing Platform

As wireless communication is becoming more complex and flexible,the problem how technical developers quickly make response to market demand and submit solutions,has been the focus for wireless communications research and development.Software Defined Radio is considered as a better solution,which has received extensive attention and also has made great progress until now.Heterogeneous platforms,such as ATCA,CPCI and VPX,widely applied in the field of high-performance information and signal processing,carry an enormous number of heterogeneous processors,including FPGA,DSP,GPU,GPP,etc.How to implement the unified software processing framework which is consistent with the idea of SDR for the heterogeneous processors of these platforms has become a hot research topic.Due to lightweight and high performance signal processing platform application requirements,this thesis carries out the research on the key technology of software processing framework,containing virtualization,resource allocation and real-time scheduling.The main work is as follows:1.Aimed at solving the tightly coupled problems between software and hardware in heterogeneous platform,this paper presents a hardware-modeling-based hardware virtualization method.In this method,heterogeneous platform is abstracted and represented as a collection of the platform,board,processor,communication node and bus connection elements.Meanwhile the method also composes of operating system virtualization,device management service,automatic data routing scheme.Then through the typical multi-platform hardware examples,the effectiveness of this method is verified.2.For analyzing the consequences to executing scheduling performance of framework which is brought by synchronous data flow model,a improved Topology-Matrix-based buffer allocation policy is put forward to augment the open-source software framework GNU Radio,by drawing lessons from the existing model-based implementation.And DQPSK transmitting and receiving application experiments show that the new framework realizes a drop in the amounts of latency and allocated buffer for running applications,where the duration of latency is shortened by more than 60%,only at the expense of some degree of reduction in throughput,the maximum amount of which is estimated about 50%.3.According to requirements of automatically deploying applications to hardware resources in platform,a new windowed dynamic algorithm is proposed on the background of computing resource management service provided by ALOE(Abstraction Layer and Operating Environment).Lower complexity can be achieved by adjusting the preserved number of nodes during the node selection diagram of the algorithm.Simulation results show that this algorithm can cut down the complexity of algorithm but for a certain degree of loss in reliability of mapping solution,and the reliability of mapping solution is still user-tolerable.