面向CPU/GPU异构环境下的GNSS协同并行解算方法

At present, GNSS is developing towards the era of multi-frequencies, multiple systems, large-scale stations, and widely needs of high accuracy and real-time applications. It is one of the bottleneck problem in the field of GNSS to improve the data processing efficiency of large GNSS networks. In response to the problem that current GNSS data processing theory and method cannot directly utilize the powerful computing capacity of the new heterogeneous platform of CPU/GPU, this project will surround the new target of coordination processing the GNSS large-scale data under the CPU/GPU hybrid high-performance computing environment, the collaborative parallel preprocessing method of large GNSS network data, the synergetic parallel computing method of high-order observation equation, and efficient parallel fixing method of high-dimensional ambiguity parameters will be studied under CPU/GPU platform. The load balancing and performance optimization strategy of CPU and GPU to deal with GNSS computing tasks will be realized. And a collaborative parallel processing software for large-scale GNSS network based on heterogeneous CPU/GPU platform will be developed and a series of parallel computations and analyses will be tested and verified. The key scientific problems concerning parallel block decomposition algorithms of the high-order observation equations and load balancing of computing tasks for large-scale GNSS network will be broken through. This project will propose a new models and methods about the collaborative parallel computing of huge GNSS network, and accomplish the innovation of synergistic rapid processing for GNSS data under the heterogeneous CPU/GPU system.

目前，GNSS正处在多频多系统、数据规模剧增、精确性与实时性应用需求广泛的时代，如何提高GNSS大网数据处理效率是GNSS领域的瓶颈问题。本项目针对当前GNSS数据处理无法直接利用新型异构平台强大计算能力的问题，围绕CPU/GPU异构计算环境下GNSS大网数据的协同处理模式新目标，研究异构平台下GNSS协同并行预处理、超高阶观测方程协同并行解算、高维模糊度参数协同并行固定的方法，实现CPU/GPU协同的GNSS数据处理的负载均衡与性能优化策略，并构建基于CPU/GPU异构平台的GNSS解算软件原型开展试验与分析，突破超高阶观测方程的并行分块分解和负载均衡分配方法等关键科学问题。本项目将提出全新的异构平台下协同处理GNSS数据的并行解算方法，实现GNSS大网数据在CPU/GPU异构平台下的协同快速处理的创新。

GNSS数据处理迎来上千地面观测站、多个导航卫星系统带来的超过120颗导航卫星和多频信号的时代。伴随而来的是GNSS网的观测数据急剧增长，使得数据处理更加庞大复杂，解算的参数更多、耗时更长，数据处理效率已是限制多频多系统GNSS大网数据处理与应用的瓶颈问题之一。.大数据时代背景下，并行计算、云计算等高性能计算技术成为大规模数据处理的首选方法，多个多核CPU和众核GPU组成的异构并行系统成为主流的新型高性能计算平台，本项目围绕高性能异构计算环境下GNSS多频多系统大网数据协同解算模式新目标，研究了与异构平台相匹配的GNSS数据协同并行预处理、大网观测方程并行解算、高维模糊度快速并行固定等问题。具体研究工作有：设计了多系统GNSS数据的并行预处理流程方法，并基于此开发并开源了可多平台应用的GNSS数据并行预处理软件；通过采用移动开窗法减弱电离层延迟影响，提出了精度与效率兼顾的单频载波相位平滑伪距算法；通过改进的多核任务并行与多节点服务并行的协同算法，实现了GNSS非差/差分模型的高效并行处理；构建了星间单差FCB并行估计方法，改进了非差模糊度并行解算策略；最后在异构处理平台通过任务分解与数据分块实现对GNSS大网密集型解算任务的负载均衡，升级改造了多用途、跨平台的GNSS数据并行处理软件GNSSer，开展了GNSS大网数据在异构平台下并行计算试验与验证。.本项目突破了GNSS超大规模观测方程的协同并行分块分解、异构多级框架下GNSS计算任务的负载均衡方法等关键科学问题，提出了全新的适合异构平台下协同处理GNSS大网数据的并行解算方法，实现了GNSS观测数据在多核多节点平台下协同调度方法的创新。将为提高GNSS大网数据的解算效率，建立协同高效、自动便捷的GNSS数据处理服务提供应用基础。已发表学术论文8篇，其中SCI/EI检索6篇；培养硕士研究生2名，申请国家发明专利2项，软件著作权1项，待出版中文著作1部。