【作者】 谢晖；

【导师】 冯穗力；

【作者基本信息】 华南理工大学 ， 信息与通信工程， 2014， 博士

【摘要】 信道估计是OFDM系统中最主要的挑战之一。作为21世纪最主要发现之一的压缩传感理论引领整个信息与信号处理领域的一场革命。它广泛地应用于图像、语音、音乐、无线通信、雷达以及天文数据等领域。本文主要关注压缩传感理论在OFDM系统下的稀疏信道估计方面的应用研究。专注于有效地解决OFDM系统下稀疏信道估计的挑战，本文构造了稀疏信道估计框架，基于这一框架，提出了新的稀疏信道估计方法。具体地，论文的创新性总结如下：1）OFDM系统下基于阈值的主要抽头检测的稀疏信道估计（M≥Lc p,M为导频的数目Lcp是循环前缀的长度）大量的传统信道估计算法都是在M≥Lcp情况下基于LS算法，LS算法在估计密集多径信道时性能达到最优。然而，如果信道时稀疏的，LS算法易受到噪声的干扰，导致估计的性能下降。为了克服LS算法的缺陷，提出了一种时域阈值来检测最主要抽头，此阈值由初始阈值滤除信道分量，保留噪声分量得到。由于提出的方法不需要信道的先验统计特性和噪声标准差，因此对于实际的无线通信系统带来很大的好处。理论分析与仿真结果表明提出的方法能在较大的稀疏率范围内获得更好的BER和NMSE估计性能，好的频谱利用效率以及适中的计算复杂度。2）OFDM系统下基于压缩传感的稀疏信道估计算法(M <Lcp)a)建立基于压缩传感的稀疏信道估计的框架。基于这一框架，提出全新的阈值估计方法。具体地，在最大迭代次数是K max情况下（也就是信道可能出现的主要径数的最大值），采用OMP算法估计信道冲击响应。为了提高估计的性能，估计具有m (m <M)个幅值的部分信道冲击响应。利用这m个信道抽头估计出噪声的标准差。最后，估计出有效地阈值。利用估计出的阈值实现有效的信道估计。b)在所提出的算法中， m个信道分量的索引对于阈值估计的精度至关重要。为了解决这一问题，引入基之间的相关性阈值来搜寻m个基的索引。3）高效的基于压缩传感的非采样间隔稀疏信道估计(M <Lcp)a)与采样间隔稀疏信道不同，非采样间隔稀疏信道会造成接收端的能量泄漏。本文推导了观测到的接收端具有不同的过采样因子R的信道冲击响应并发现如果考虑R>1的情况，相比较基带采样泄露情况将会改善。如果考虑R→∞,将不会出现泄露的情况。基于这一点，本文开发了具有高分辨率的测量矩阵来实现高分辨率的信道估计。采用具有次优导频排布与高分辨率的测量矩阵，在压缩传感的框架下实现了利用有限数目的导频(M <Lcp)来有效地估计高时间分辨率的信道冲击响应。仿真表明，与传统的信道估计算法相比，提出的基于压缩传感的方法能实现在保证好的估计性能的情况下减少导频的使用。b)对于基带采样来说，测量矩阵只有Lcp个基，然而，如果考虑过采样的情况，测量矩阵有(R-1) Lcp+1个基（R (R>1)为过采样因子）。过采样时测量矩阵所包含的基向量的个数比基带采样时测量矩阵所包含的基向量的个数高R-1倍。然而，对于K (K <<Lcp)稀疏信道来说，只有K个基对于压缩传感的重构是有用的。此外，对于稀疏信道来说，导频数目的减少时一个挑战而且是一个必须完成的工作，它对于提高频带资源的利用效率非常重要。因此，如何有效减少计算复杂度并且保持高的频带利用效率是关键。首先，粗略探测具有基带采样的测量矩阵中的“热点区域”（感兴趣的基所在的位置）。然后构造只在“热点区域”内具有过采样的智能测量矩阵。利用精心设计的测量矩阵以及有效地压缩传感重构算法，可以获得有效地信道估计性能。所提出的方法能够在M <Lcp且相对较低的计算复杂度情况下获得有效的信道估计。更多还原

【Abstract】 Channel estimation is one of the most important challenges in OFDM system. Asone of the major discoveries in the21th century, compressed sensing (CS) theoryleads a breakthrough in the whole information and signal processing societies. It canbe widely applied in images, audio, music, wireless communications, radar, andastronomical data etc. This thesis focuses on the research of applications ofcompressed sensing in sparse channel estimation in OFDM system. Aiming toeffectively solving the challenges of sparse channel estimation in OFDM system, thiswork constructs a sparse channel estimation framework, based on which, noveleffective sparse channel estimation methods are proposed. Specifically, the noveltiesof the thesis can be summarized as follows:1) Threshold based most significant taps detection for sparse channel estimationin OFDM system (M≥Lc p,Mis the number of pilots andLc pis the length of cyclicprefix)Numerous traditional channel estimation methods are initially based on LS in thecase of M≥Lcp,which is actually optimal when channel is rich multipath channel.However, if the channel is sparse, LS method is vulnerable to noise, which leads tothe degradations on the estimation performance. In order to overcome the drawbacksof LS method, a novel effective time domain threshold depending only on theeffective noise standard deviation estimated from the noise coefficients obtained byeliminating the channel coefficients with an initial estimated threshold is proposed todetect the most significant taps (MST). Since the proposed method requires neitherthe prior knowledge of channel statistics nor the noise standard deviation, which willsignificantly benefit the practical wireless communications. Both theoretical analysisand simulation results show that the proposed method can achieve better performancein both bit error rate (BER) and normalized mean square error (NMSE) thantraditional methods within a wide range of sparsity rate, has good spectral efficiencyand moderate computational complexity.2) A novel CS based sparse channel estimation in OFDM system (M <Lcp)a) The framework of CS based sparse channel estimation method is constructed.Based on the framework, a novel effective threshold is proposed. Specifically, channelimpulse response (CIR) is firstly estimated by OMP with the assumption of maximumiteration number ofK max,which is also the maximum possible number of significanttaps. Then, in order to improve the estimation performance by an effective threshold,partial CIR with m (m <M)coefficients is approximately estimated. With theestimated m channel taps, the noise standard deviation is estimated. Finally, aneffective threshold is obtained. With the estimated threshold, the effective channelestimation can be realized. b) In the proposed method, the m channel coefficients index is essential for theprecision of the threshold estimation. To solve this problem, the threshold ofcoherence between bases is introduced for searching the indices of the m bases.3) Efficient and effective CS based non-sample spaced sparse channel estimationin the case ofM <Lcpa) Unlike the sample spaced sparse channels, the non-sample spaced sparsechannel can cause power leakage at the receiver. We have derived the observed CIR atthe receiver with different oversampling factors R on the estimated CIR and foundthat if the oversampling R>1is considered, the leakage effect will be reducedcompared with the baseband sampling. If R→∞,there will be no leakage effect.Based on this fact, measurement matrix with finer time resolutions is developed forhigh resolution CIR estimation. By employing the measurement matrix with bothsuboptimal pilot arrangement and high resolution, CIR with finer time resolution canbe effectively estimated with limited number of pilots (M <Lcp)by CS. Simulationsshow that, compared with the traditional channel estimation method, the investigatedCS based method can realize superior channel estimation performance with asignificant reduction of pilots.b) For the baseband sampling, we only getLc pbases for the measurementmatrix, however, if the oversampling is considered, we get (R1) Lcp+1(In the casewhere R (R1),which is the oversampling factor) bases for the measurementmatrix. R1times higher. When we go back to a K (K <<Lcp)sparse channel,only K bases are useful for CS reconstruction. Additionally, for sparse channel,pilots reduction is a challenging and essential task, which can effectively promote thespectral efficiency. Therefore, how to effectively reduce computational complexitymeanwhile maintain high spectral efficiency is the key issue. Firstly, the "hot zones"(location of the interested bases) in the measurement matrix with baseband samplingare roughly detected. Then, smart measurement matrix can be constructed withoversampling only in those "hot zones". With a carefully designed measurementmatrix and by adopting the effective CS based channel reconstruction algorithm,effective channel estimation performances can be obtained. The proposed method canrealize effective channel estimation in the case ofM Lcpand comparatively lowcomputational complexity.更多还原