低信噪比无线通信信号非合作接收技术研究

【作者】 李静；

【导师】 葛临东；

【作者基本信息】 中国人民解放军信息工程大学 ， 军事情报学， 2005， 博士

【摘要】 无线信号的非合作接收是技术侦察中信息截获技术体系的一个重要而关键的组成部分。现代无线通信技术高速、低功率的发展趋势以及信息截获本身固有的特点,对信噪比门限低、抗信道衰落和失真能力强的非合作接收技术提出了迫切的需求。 均衡与同步是失真信道下进行有效接收必须解决的两大关键问题。非合作接收应用背景下的均衡与同步一般比常规通信接收中的相关问题有更高的技术难度,而低信噪比、严重失真的无线信道条件则使这两个技术问题更具挑战性。本文主要围绕这一应用背景下的均衡和同步两大技术问题进行了研究与讨论。 在绪论中我们阐述了本论文课题的应用背景。由于认清无线信道的特点有助于采取相应的信号处理措施从而提高接收系统的性能,因此在讨论具体的技术问题之前,绪论还概要地介绍了课题研究的重要基础之一——无线信道模型。 要实现恶劣条件下的有效接收,关键在于采取全局最优或接近最优的检测算法。基于软信息的迭代检测是一种以现实可容忍的运算代价实现逼近全局最优接收性能的信号处理思想。它强调接收机中各级处理模块之间软信息的传递,并打破了接收机中各级处理模块之间信息单向流动的传统做法,在信道译码与其它模块之间形成信息交流的良性循环,大大地提高了接收机抗严重信道失真和噪声的能力,是本文研究方法的重要基础。本文第二章从最优检测的角度介绍了迭代处理思想的基本原理,并引出这一思想的两个重要基础:软输入软输出(SISO)处理和分级迭代处理。在此基础上,针对近十年来发展起来的迭代处理在通信接收中的几个应用,简要介绍了turbo编译码、turbo均衡、迭代多用户检测以及迭代解调映射等的原理。 Turbo均衡是迭代处理思想在均衡领域中的成功应用。本文第三章首先描述了turbo均衡的通用结构,并就包括SISO映射、SISO逆映射及SISO译码在内的几个关键模块进行了深入详细的讨论。在SISO逆映射的讨论中,首次推导了针对Gray编码16QAM调制的均衡器外信息的简化计算公式。在关于SISO译码器外信息计算的讨论中,论文分析了常用计算方法的不合理性,提出一种新的译码器外信息计算方法。仿真结果表明,对于严重信道失真的高阶调制信号,采用该计算方法的turbo均衡具有十分明显的性能优势。 基于已经给出的turbo均衡的通用结构,本文第三章在信道响应已知的恒参信道条件下,研究讨论了分别基于最小均方误差(MMSE)线性均衡器(LE)和MMSE软判决反馈均衡器(SDFE)的turbo均衡算法及其性能。其中,turbo MMSE-LE采用了基于MMSE准则的最优均衡算法,在低信噪比下抗严重信道失真的能力比传统均衡有了极大的提高。不过,由于这种算法需要大量矩阵求逆运算,因而实现复杂度较高。针对这一特点,论文还介绍了简化的turbo MMSE-LE,减少了算法中矩阵求逆的次数。在基于SDFE的turbo均衡方面,论文讨论了按MMSE准则设计的前向滤波器和反馈滤波器的频域形式,在此基础上通过IDFT得到均衡滤波器的时域近似,并从方便实现的角度对均衡器结构进行了改进。这种turbo SDFE在计算滤波器系数时只涉及简单的标量计算和存在快速算法的IDFT,因而运算量低,易于实现。不过仿真结果表明,按上述思想实现更多还原

【Abstract】 Uncooperative receiving of wireless communication signals is a very important part of the information interception for technique based reconnaissance. Modern wireless communication, especially signal interception, which is characterized by low power and high transmission rate, urgently requires noncooperative receiving technologies with lower signal-to-noise-ratio threshold performance and better ability against serious channel distortion.Equalization and synchronization are key technologies for effective receiving of distorted signals. The implementations of equalization and synchronization in uncooperative receiver are usually more difficult than those in the conventional receivers, and the low SNR and severe distortion of wireless channel make them even more challenging. In this dissertation, we address the problems of equalization and synchronization under low SNR and harsh channel condition.In the exordium, the motivation of the thesis is presented. Since the knowledge of the channel is necessary for us to choose proper processing methods for effective detection and receiving, we also make a brief introduction to the frequently used models of wireless channel in the first chapter.Global optimum or near optimum detection is the only way to achieve effective receiving under harsh channel condition. The iterative detection method based on soft information is one of the ways to approach the global optimum performance at an acceptable cost of implementation complexity. In this thesis, we use iterative detection scheme to solve the problems of equalization.In iterative detection, the concatenated modules of the receiver make use of soft input and soft output (SISO) algorithms and exchange soft information of the transmitted bits. Other than the unilateral information transfer in a conventional receiver, the information transfer in an iterative detection is bidirectional, which makes it possible for the former module to benefit from the processing gain of the latter modules. As the iteration goes on, a benign circulation of soft information is set up and the performance of the receiver gradually approached the optimum bound. In chapter 2, the principle of iterative detection is discussed and several applications such as turbo coding and decoding, turbo equalization, iterative multi-user detection and iterative demodulation are introduced.In chapter 3, the universal structure of turbo equalization is presented. The modules including SISO mapping, SISO demapping and SISO decoding are discussed in detail, based on which we present our relevant contribution: a simplified algorithm for computation of extrinsic information of SISO equalization for Gray coded 16QAM and a new definition of the decoder’s extrinsic information.The rest of chapter 3 investigates two turbo equalization algorithms on condition that thechannel is static and the channel response is known. The first algorithm is based on minimum mean square error (MMSE) linear equalization (LE) and the other one is based on MMSE soft-decision-feedback-equalization (SDFE). Simulations show that both the algorithms outperform the conventional equalizers quite a lot. In turbo MMSE LE, the optimum equalization algorithm is exploited, involving lots of computation of matrix inversion, involved and thus a high complexity is resulted. In turbo SDFE, the feed-forward and feed-back filters are derived by the MMSE rule in frequency domain and then the filters in time domain are obtained by the use of IDFT. In the computation of filters in SDFE, only scalar computation and IFFT are needed, so it is very easy to be implemented. However, as shown in simulations results, the performance of turbo SDFE is inferior to that of turbo MMSE-LE by 2dB, which is resulted from the approximation processing in using filters with finite length instead of optimal filters with infinite length.In chapter 4, blind turbo equalization for time-invariant channel is investigated. Because training sequences are usually unavailable in uncooperative receiving, we have to perform turbo equalization without the help of training sequences. Blind equalization can be classified into two structures. The first uses adaptive filters to retrieve the transmitted symbols directly and the other performs channel response estimation and equalization in a relatively independent way. The second one is used in this thesis. Since the structure and SISO equalization algorithms have been fully discussed in chapter 3, the key problem is to fulfill blind channel estimation. Accordingly, we exploit iterative channel estimation that is based on recursive least square (RLS) algorithm with soft decisions as its reference signal. To obtain soft decisions for the first iteration, a blind equalization is proposed be used before the iterative channel estimation.The super exponential blind equalization is based on block data processing and has the ability to deal with severe channel distortion. Hence it is suitable for the initial equalization in the iterative channel estimation. After the initial blind equalization, SISO decoding and SISO mapping, the initial soft decision can be obtained and the channel estimation can be bootstrapped. In the succeeding iteration, RLS algorithm exploits the soft-decision of the previous iteration as the input signal. The channel parameters of SISO equalization are also refreshed in every iteration. As the reliability of the soft-decisions become better and better, the resultant channel estimation become more and more accurate. Simulation results show that the proposed blind turbo equalization can perform effectively.Chapter 5 addresses the equalization issues of frequency-selective fading channel.In conventional communications, training sequences are often necessary for channel probing, especially when the channel is severely fading. Generally, only blind technologies are effective in most of the uncooperative receiver. However, when the communication standard is known, it is possible and reasonable for the receiver to exploit training sequence to improve its performance. Hence we firstly discussed the training sequence aided channel estimation. There are two periods in channel estimation: the initial period and the iterative period. In the initial estimation, channel response is assumed to be static and the estimation obtained in the trainingperiod is used for the overall sequence. In the iterative period, the RLS based channel estimation takes the soft-decision as its input signal and the channel parameter of SISO equalization is updated symbol by symbol.Chapter 5 lays emphasis on the blind equalization of fading channel. A blind equalization algorithm based on Monte Carlo methods is used. In most Bayes detection problems, a multi-dimension integration or summation with extremely high complexity is usually involved, thus one has to resort to some numerical approaches for computation. Monte Carlo methods are one kind of such numerical approaches and Gibbs sampler is one of the most commonly used Monte Carlo methods. The Gibbs sampler in general form is firstly described and then the blind equalization based on the method is discussed. In Gibbs sampler equalization, the distribution of transmitted symbols and channel response are simultaneously computed in an iterative way. When the iteration converges, the estimations of the symbols and channel response can be obtained. This algorithm can only deal with time-invariant channel. However, being different from most of the blind algorithms, it is able to fulfill data recovery and channel estimation based on very short data segment. Bearing the fact in mind that the fading channel can be well approximated by a static channel in a very short period, we propose to divide a long faded data frame into a number of short segments and assume a static channel within each segment and then use Gibbs sampler equalization algorithm on them. What’s more, because Gibbs sampler is a typical SISO algorithm, it is easy to substitute it into the turbo equalization structure given in chapter 3 to fulfill turbo equalization for frequency-selective fading channel. Simulations results demonstrating performance of the equalization are also presented.Carrier synchronization is another key problem in uncooperative receiver. Though there have been many achievements about this issue, most of them are designed for high or medium SNR. When SNR is low, these methods often fail to achieve good synchronization. Therefore, chapter 6 is contributed to carrier synchronization for MPSK modulation signals with low SNR.Firstly, an NDA carrier frequency offset estimator based on auto-correlation of the signal’s nonlinear transformer is discussed. For a tinny frequency offset, this estimator can approach MCRB at low SNR. However, the frequency range within which it can perform effectively is limited because phase wrapping and error will occur when frequency offset is beyond the range. Towards to this problem, we propose an improved estimator, in which the frequency out of the range is coarsely measured and then is shifted so that the residual frequency is within the range and thus can be effectively estimated. Simulations show that the improved estimator performs much better than several other commonly used estimators when SNR is low.Secondly, we investigated carrier phase recovery method for MPSK signals with low SNR. In a transmitting system with an error control coding, the likelihood ratio output of the SISO decoder can reflect the carrier phase error and thus can be exploited to construct a performance function for carrier phase recovery. Based on this performance function, wepropose a phase searching method that can achieve high estimation accuracy at relatively low computation cost. As demonstrated by simulation results; this carrier phase recovery method performs well at low SNR and can overcome phase ambiguity problem that other methods often suffer from.SISO channel decoding is the topic of chapter 7. Channel decoding is very important for receiving signal with low SNR. It is also a key part of iterative detection. In iterative processing such as turbo equalization; iterative multi-use detection and iterative demodulation; the SISO decoder is different from the one used for error control. For example; the input of turbo equalization only consists of a priori information of the coded bits whereas the common decoding algorithm often takes both channel observation and a priori information as input. The output of decoder in turbo equalization consists of log-likelihood-ratio of all coded bits other than only LLR of information bits in the common decoder. According to these differences; we describe the log-MAX-MAP decoding algorithm for convolutional code; which is suitable for iterative processing. Because the description is not based on assumption of certain code; it is also suitable for decoding of non-systematic codes.更多还原