基于信号结构稀疏性发掘的超大带宽MIMO系统预失真方法研究

In order to improve the robustness and the efficiency of linearization technology for the high power amplifier (HPA), a novel digital predistortion (DPD) technology suitable for the ultra-high wideband multiple-input multiple-output (MIMO) signal is developed to meet the scenario requirements and application demands for the wireless communication systems in future. Depending on the exploitation for the multi-dimensional and multi-transformed-domain sparsity features implied in the statistical structure of wireless signal, this project focuses on establishing a sparse representation framework between the HPA behavioral model and signal analysis perspective in the predistorter design. Based on this sparse framework, the corresponding equivalent models, theory and signal processing solutions for DPD system will be investigated. Three key aspects, including the HPA behavioral modeling, compression and reconstruction of sub-Nyquist sampling signal, and the stable DPD identification strategy, are mainly studied. Firstly, we will design multi-dimensional distributed concatenated Volterra-based behavioral model for concurrent multi-band and multi-antenna HPA to deal with a huge number of modeling coefficients, which may result in ill-conditioned model identification due to over-parameterization phenomenon. Secondly, the compression and reconstruction of second-order statistics i.e. covariance or power spectrum by using sub-Nyquist sampling signal will be developed to break through the bottleneck problem of Shannon sampling rate even in the absence of sparsity prior. Thirdly, in order to improve the numerical stability and computational complexity of traditional Least-square (LS) method for DPD model identification, an innovative DPD architecture and optimization methods based on both the frequency-domain measurement and the feedback of reconstructed power spectrum will be developed. Furthermore, the novelty of this project lies in improving the overall performance of adaptive predistortion in terms of the accuracy HPA modeling, high efficiency of identification, robustness of processing algorithm, and the feasibility of DPD implementation. By achieving technological breakthroughs for this novel DPD scheme, and completing the corresponding software & hardware verification prototyping platforms, the critical proposals have not only academic innovation and theoretical significance, but also have a broad prospect of actual applications in both the 5G mobile communication system and statistical signal processing fields.

针对信号超大带宽、海量吞吐、多路分支、动态复杂等基本特征，发展面向未来无线通信系统构架的预失真技术理论方法。研究方案立足于发掘无线信号自身在多维度、多变换域隐含的结构化稀疏性特征，在放大器行为模型与信号解析方式相匹配的全面稀疏性表达框架内，构建预失真处理各环节的等效模型与对应的解决方法。研究工作从三方面展开：多维度、分布式刻画的放大器行为建模方法，解决海量系数堆砌问题；低速率、弱稀疏性约束的信号统计量压缩与重构方法，解决信号可达采样速率瓶颈问题；基于频域测量与功率谱反馈构架的预失真辨识优化方法，解决传统时域辨识求解的稳定性与复杂度问题。在相关基础理论方法上取得创新性突破，获得预失真系统在模型准确性、辨识高效性、算法鲁棒性、实现可行性方面的全局优化；并将部分理论成果推广至应用技术层面，开发硬件验证平台与标准设计。研究不仅具有学术上的创新性，成果也将在通信统计信号处理领域具有广泛的应用前景。

针对现有预失真技术难以适应新一代无线通信高带宽需求提炼关键科学问题，开发适宜于极高宽带信号的新型自适应预失真技术方案，研究立足于发掘无线信号自身在多维度、多变换域隐含的结构化稀疏性特征，在放大器行为模型与信号解析方式相匹配的全面稀疏性表达框架下，围绕如何增强预失真技术鲁棒性和高效性的问题，从多维度分布式刻画的放大器行为建模方法、弱稀疏性约束的信号统计量压缩与重构方法、放大器模型辨识结构及算法鲁棒性等方面展开相关研究，构建预失真处理各环节的等效模型与对应解决方案。主要研究内容包含4方面：1）、基于协方差感知理论和信号稀疏分解，对稀疏表达Volterra模型做非结构先验性优化，将模型海量系数优化转化为具有收敛优化属性的模型稀疏表达估计问题，设计了三维分布式结构的多频带、多分支放大器等效模型；2）、通过建立具有多维分辨能力的宽带无线信号解析方法，开发了低速率、弱稀疏性约束的信号统计量压缩与重构方法，解决了信号可达采样速率瓶颈问题；进一步针对非静态信号条件下辨识的稳定性问题研究了时域预失真构架和频域测量预失真构架，提出了几类新型的混合DPD结构；考虑到学习稳定性和收敛性问题，提出了自适应灵活可变步长的学习算法，推导了选择准则的理论界；3）、针对具大子载波、高阶星座OFDM信号，研究了基于凸集拓展投影的高效主动星座扩展信号峰均比抑制方法，基于特别设计的迭代策略，仅利用一次迭代通过动态参数调整即可在峰平比和误码率之间获得良好的折衷，为系统参数设计提供了更高的灵活性；4）、开发了两个相关的软硬件实验平台：基于VC的宽带预失真关键算法软件仿真平台用以评估基础理论算法；以及基于FPGA兼容LTE-A系统参数的DPD硬件验证平台。本项目创新点在于获得预失真系统在模型准确性、辨识高效性、算法鲁棒性、实现可行性方面的全局优化。项目组已发表了多篇高质量论文和技术发明专利，研究不仅具有学术创新性，成果也在统计信号处理领域具有广泛的应用前景。