Research On Multilevel Redundancy Protection And Reliability Analysis Of Software

With the development of Chinese aerospace science,the tasks of spacecrafts are becoming more and more complex.Almost all activities on the spacecraft need the support of software and the reliability of the software has become one of the guarantees for the stability of the spacecraft.However,because of the existence of cosmic rays in space,hardware processors such as DSP and FPGA which are widely used in aerospace are bombarded when running.Then the single event upset is caused.The direct results are data errors and execution stop.All of them could lead a bad impact on the spacecraft.So,how to improve the reliability of the software system in the processor chip and ensure the execution of them has become a hotspot and difficulty in the research of the field.In recent years,scholars have proposed a series of software and hardware protection methods for single event upset on processor chips.The redundancy protection is most widely used because of its low implementation difficulty and significant protection effect.However,the hardware resources on the chip are limited and the cost of redundancy is also an important problem.In this paper,we studied the optimal redundancy allocation of modules combined with the theory of multi-level redundancy,modules’ reliability and costs.And some improvements on relevant algorithms are performed.At the same time,the voter of the redundant modules is studied and the fault-tolerance ability of it is strengthened.The main work of this article is as follows:(1)Combined with the characteristics of DSP projects,a set of rules is proposed to the multi-level division of DSP codes.Through the analysis and sub-division of the main()function,the code structure can be divided by the loop structures and the branch structures.And then,the DSP projects can be transformed into several modules in different levels which is called Multi-level structure.The relationship between the upper and lower modules is cascade.(2)The solution of the multi-level structure is completed.Besides,a reliability model and a cost evaluation model for the multi-level structure are established.By analyzing the reliability and cost of it,the problem of improving the reliability is transformed into a single-object and multi-constrained optimization problem.The reliability is the objective function and the cost is the constraint.The optimal redundancy allocation of modules maximizes the reliability of the code.(3)To improve the Memetic Algorithm(MA),an Optimized Memetic Algorithm(OMA)is proposed.By analyzing the defects of the traditional MA,the crossover operator and the local search operator are improved in a targeted manner to make it more efficient and reasonable in cross-individual selecting and local searching.Finally,the performance of the algorithm is analyzed by an example experiment.Through the comparison of different algorithms in this field,the superiority of the OMA is verified.(4)To study the voter of the discrete data,a new s*k voter is proposed.Because the redundancy is used to modules,the voter for the redundant output is a key link.Through performance verification of the word voter and the bit voter,the s*k voter is proposed.With the bit-width as the entry point and achieving a balance between the error rate and the success rate.The s*k voter could provide more flexible voting ways and broader options.According to the results of experiments,the multi-level redundancy based on Optimized Memetic Algorithm(OMA)and the s*k voter proposed in this paper can improve the reliability of DSP projects significantly.The method proposed here can provide effective reference and guidance for engineers when designing.