无线通信中OFDM系统信道估计技术研究

【作者】 陈卫华；

【导师】 李少谦；

【作者基本信息】 电子科技大学 ， 通信与信息系统， 2010， 硕士

【摘要】 正交频分复用(Orthogonal Frequency Division Multiplexing, OFDM)由于具有频谱效率高和能有效抵抗多径干扰等优势,而成为新一代移动通信系统中解决高速率数据传输的新技术。然而多径效应带来的频率选择特性以及信道时变带来的子载波间干扰(Inter Carrier Interference, ICI)等给信道估计带来了新的挑战,因此,有效的信道估计算法成为OFDM技术成功应用的关键之一。本文主要研究了准静态信道环境下OFDM系统的参数化信道估计算法——旋转谱不变技术(Estimating Signal Parameters via Rotational Invariance Techniques, ESPRIT),以及快时变信道环境下OFDM系统分别基于线性模型(Linear Model, LM)和基扩展模型(Basis Expansion Model, BEM)的不同信道估计算法。第二章简要介绍了传统的准静态信道条件下的OFDM系统的非参数化信道估计技术,主要包括常值插值法、线性插值法、cubic插值法、DFT插值法等经典插值算法,以及在DFT插值法基础上改进的能够抑制部分噪声的算法。其中,DFT及抑制噪声的DFT插值算法因其优异的性能和较低的复杂度而受到青睐。然而,美中不足的是,DFT及抑制噪声的DFT插值算法是建模在信道多径时延为采样周期的整数倍上的,当信道多径时延为采样周期的非整数倍时,算法的信道估计性能会有很大的损失。第三章研究了准静态信道条件下的OFDM系统的参数化信道估计技术ESPRIT。该算法利用等距线性阵列的信号子空间旋转(或平移)不变特性,分离出信道子空间和噪声子空间,然后利用信道子空间估计出信道的多径时延信息,最后通过估计的多径时延信息重建信道。本章首先着重研究了ESPRIT算法的子空间分解技术——子空间跟踪和特征值分解两种方式,并利用跳频原理对子空间跟踪方式收敛速度进行了加速。通过性能分析与仿真对比,子空间跟踪方式性能更稳定,但需要信道条件(如多径数目和多径时延)在较长时间内保持不变,特征值分解方式则没有此限制,但性能较子空间跟踪方式差,尤其体现在较恶劣的多径环境下。接着将子空间跟踪法从多符号的应用中扩展到了单符号中。最后,本文还对导频数量、导频间隔和导频结构的优化设计进行了详细研究,研究发现,ESPRIT算法的信道估计性能主要取决于导频间隔,在满足特定导频间隔的前提下,可以用足够少的导频获得相应的性能,从而可以减小导频开销,提高频率利用率;而导频结构除了等间距结构外,还可以采用分组等间距结构,从而可以扩展ESPRIT算法的应用场景。第四章研究了快时变信道环境下的OFDM系统信道估计技术。信道的时变性在一段时期内,如一个OFDM符号长度,可以用LM模型或BEM模型来近似。其中,LM模型适用于时变性较温和的信道环境,即归一化多普勒频移小于0.2的环境,BEM模型则还适用于归一化多普勒频移大于0.2的环境,不过,现阶段LM模型可以针对非整数多径时延环境,而BEM模型只适用于整数多径时延环境。本章第一部分重点研究了LM模型下的参数化信道估计问题,并对现有算法作了改进,将整数时延模型推广到非整数时延模型,该算法基于ESPRIT算法,利用迭代的判决反馈方式来减小ICI影响,以获取对多径时延的精确估计,该算法克服了LM模型下传统整数时延模型的不足,且性能优良。本章第二部分重点研究了BEM模型下的信道估计问题,通过估计一组数量远少于快时变多径信道冲击响应数量的BEM加权系数来间接得到信道的估计值,大大简化了信道估计的难度。本文选取了DKL-BEM和GCE-BEM两种近似度高且稳定的模型,并根据模型特点分别给出了估计器的选型建议,即根据DKL-BEM模型需要利用信道统计特性的特点,采用线性最小均方误差估计器(Linear Minimum Mean Square Error, LMMSE),而GCE-BEM由于没有特殊要求,则可采用最小二乘估计器(Least Square, LS)或最佳线性无偏估计器(Best Linear Unbiased Estimator, BLUE)进行信道估计。更多还原

【Abstract】 Owe to the spectral efficiency and robustness against multipath channels, Orthogonal Frequency Division Multiplexing (OFDM) has become one of the major modulation techniques of high data rate communication systems. However, both the multipath channels and the inter carrier interference (ICI) have inceresed the difficulty of the channel estimation. Thus effictive channel estimation becomes one of the key factors that affect the application of OFDM. This paper is mainly about the study of parameter channel estimation of OFDM system in the condition of quasi-static channel environment which is Estimating Signal Parameters via Rotational Invariance Techniques (ESPRIT), and OFDM channel estimation of fast time-varying channel based on Linear Model (LM) or Basis Expansion Model (BEM).In chapter two, the technique of nonparameter channel estimation of quasi-static channel is introduced, including constant interpolation, linear interpolation, cubic interpolation, DFT interpolation, and the modified DFT interpolation method which can mitigation part of noise, etc.. And both DFT and modified DFT interpolation methods have attracted much more attention because of their outstanding performance among all of the interpolation method above, as well as low complexty. However, they are based on the channel model of integral path delays, performance of the two DFT interpolation methods will degrade rapidly due to the nonintgral path delays.In Chapter three, the technique of parameter channel estimation of quasi-static channel is studied. That is ESPRIT, which apart the signal or channel subspace and the nosie subspace by utilizing the shifting invariance character of channel subspace of the linear array. Then the multipath delays of the channel can be estimated from the channel subspace, and the channel information can be reconstructed by use of the mutipath delays. We mainly focus on two subspace decomposition method, subspace tracking and enginvalue decomposition, which are need by ESPRIT. And inspired by hopping frequency, the convergence speed of the subspace tracking method is accelerated. During the two subspace decomposition method, subspace tracking method has the robuster performance in case of the channel conditions, such as the multipath numbers and multipath delays, do not variant during a long time. The enginvalue decomposition method does not need such a precondition, but a poorer performance, especially in case of the poorer environment. Moreover, we introduce the subspace tracking method from multi-symbol into single-symbol. Finally, we investigate the optimization of pilot numbers, pilot space and pilot structure. From the study, we find that the pilot space is the key of the performance of ESPRIT. If the pilot space condition is satisfied, the pilot numbers can be as small as possible. As for the pilot structure, groupd equal space pilot structure can be used, as well as the conventional equal space pilot structure, which exposes the application scene of ESPRIT.Chapter four is mainly about the OFDM channel estimation on fast time-varying channel. During a time interval, such as the period of one OFDM symbol, the time-varying of the channel can be approximated by a LM model or BEM model. Both the two models can be used in the gentle time-varying channel, which means the normalized dopple shifting is less than 0.2, and the BEM model can also be used in the condition that the normalized dopple shifting is more than 0.2. In additional, the LM model can be used in nonintegral channel pathdelay condition, the BEM model but not. We firstly focus on the LM model, and a novel method which based on ESPRIT is proposed. Here, the iterative feedback method is needed to mitigate the ICI, so that the channel path delays can be acquired accurately. And because of this, we defend the flaw of conventional method which based on integral channel pathdelays. Secondly, we focus on the BEM model. Instead of estimating the channel impulse, much less numbers of coefficients are estimated, which degrade the channel estimation difficulty greatly. In the simulation, we choose DKL-BEM model and GCE-BEM model which two models have the higher and robust approximation, and Linear Minimum Mean Square Error (LMMSE) estimator is used in DKL-BEM model because both the model and the estimator need to know the statistic of the channel, As to GCE-BEM model Least Square estimator (LS) and Best Linear Unbiased Estimator (BLUE) is used without to know any information of the channel statistic.更多还原