【作者】 钟华；

【导师】 郑林华；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2010， 博士

【摘要】 多载波调制技术具有抗多径能力和较高频谱利用率等优点,已经成为下一代宽带通信物理层核心技术之一。目前,对于多载波调制技术的研究主要集中在正交频分复用(OFDM,Orthogonal Frequency Division Multiplex)和滤波多音调制(FMT,Filtered Multitone Modulation)上。OFDM具有均衡简单等优点,但其子信道间的正交性容易受频偏等因素的影响而被破坏,从而引起信道间干扰(ICI,Inter-channel Interference),导致系统性能下降。FMT与OFDM的本质区别在于FMT子信道频谱互不重叠。正因为如此,FMT系统的ICI很小,抗频偏性能好,并且不需要循环前缀和虚载波等开销。但FMT系统子信道频谱不重叠是通过不满足理想重构条件的原型滤波器来实现,这将不可避免地引入符号间干扰(ISI,Inter-symbol Interference),尤其当FMT应用于无线多径环境时,ISI将进一步增大,导致误码率急剧增加,严重影响FMT系统性能。因此,在接收端必须采用均衡来消除码间干扰。论文就是针对FMT系统的均衡问题,对子信道均衡、Turbo迭代均衡以及盲均衡等技术进行了深入的研究和分析。论文首先分析了无线衰落信道和FMT系统模型,以滤波器组技术为出发点,介绍了FMT系统的基本原理,结合图形说明了FMT系统基于离散傅立叶变换(IDFT,Inversed Discrete Fourie Transform)和傅立叶变换(DFT,Discrete Fourie Transform)的快速算法和多相滤波的有效实现结构,推导了FMT信号的矩阵表示式,并重点分析了平坦衰落信道、频率选择性衰落信道以及时变频率选择性衰落信道条件下FMT均衡器所要处理的对象。FMT系统的一个显著特点是ICI很小,而接收信号中的ISI可以通过采用每个子信道独立均衡的方法来消除。因此,论文随后重点研究了FMT系统的子信道频域均衡算法。首先分析了基于判决反馈均衡(DFE,Decision Feedback Equalization)的四种FMT子信道均衡算法,根据最小均方误差(MMSE,Minimum Mean Square Error)准则推导了DFE均衡器系数的求解表达式,并对四种算法的可达比特率(ABR,Achievable Bit Rate)和误比特率(BER,Bit Error Ratio)性能进行了仿真分析。DFE均衡器系数通常是以MMSE准则来计算,但接收机的BER与信号的信干噪比(SINR,Signal-to-interference-plus-noise Ratio)有关,当均衡器系数满足MMSE准则时,并不能保证信号的SINR最大,相应也不能确保系统BER或ABR性能最优。基于此,论文随后通过推导严格采样FMT(CS-FMT,Critically Sampled FMT)系统的SINR和ABR的表达式,分析了基于MMSE准则并提出了最大信干噪比(MSINR,Maximum the Signal-to-interference-plus-noise Ratio)准则和最大子信道比特速率(MSB,Maximum Subchannel Bit-rate)准则的子信道线性均衡算法,仿真结果表明,基于MSINR准则和MSB准则均衡算法的SINR、ABR以及BER性能都优于基于MMSE准则的线性均衡算法和DFE算法。针对信道的非线性畸变问题,论文将函数连接型神经网络(FLANN,Functional Link Artificial Neural Networks)引入到FMT系统子信道均衡中,并提出了基于扩展卡尔曼滤波(EKF,Extended Kalman Filter)的FLANN网络学习算法,收敛速度和稳态误差性能相比较于误差反传(BP,Back Propagation)算法都得到了改善。Turbo均衡器将信道均衡和译码联合处理,在迭代过程中,均衡器充分利用了信道编码所带来的冗余信息,获得了更好的均衡效果。论文将Turbo迭代均衡思想引入FMT系统,并针对信道响应已知和信道响应未知两种情况,分别提出了一种FMT系统的Turbo均衡方法。当信道响应已知,并且子信道响应较为平坦时,论文参考了组合后置式DFE均衡器结构,信道衰落可以通过单抽头均衡器来补偿,而FMT系统自身所产生的ISI则通过Turbo迭代均衡来消除;当信道响应未知时,论文将基于软信息的迭代信道估计算法应用于FMT系统,提出了一种联合迭代信道估计和Turbo均衡的FMT接收方案。仿真结果表明,经过2次以上的迭代后,采用Turbo迭代均衡的FMT系统在信道响应已知和信道响应未知情况下的BER性能相比较与DFE均衡都能得到大幅度提高。盲均衡技术具有在整个通信过程中不需要训练序列的优点,论文最后研究了盲均衡算法在FMT系统中的应用。针对修正常系数模(MCMA,Modified Constant Modulus Algorithm)算法收敛速度与稳态误差相矛盾的问题,通过构造新的误差函数,并建立误差函数与步长因子之间的非线性函数关系,在MCMA算法基础上,提出了一种新的修正步长(MS-MCMA,Modified Step-size MCMA)盲均衡算法,仿真结果表明,MS-MCMA算法不仅改善了收敛速度,同时也降低了稳态误差;最后,论文将MS-MCMA算法应用于FMT系统,仿真结果证明了采用基于MS-MCMA算法的子信道盲均衡器能很好地消除ISI,并且收敛速度较MCMA算法提高了一倍,同时误码率性能也得到了改善。更多还原

【Abstract】 Multi-carrier modulation has been one of the physical layer key technologies of the next generation wideband communication due to its robustness to multi-path and high bandwidth efficiency, etc. The current research in multi-carrier modulation mainly focuses on OFDM (Orthogonal Frequency Division Multiplexing) and FMT (Filtered Multi-tone Modulation). OFDM has simple equalization, whereas the ICI (Inter-channel Interference) results in the performance degradation because the orthogonality is prone to be destroyed by the frequency offset. The essential difference between FMT and OFDM is the FMT’s non-overlapping sub-channel characteristics which bring a set of merits such as negligible ICI, good anti-frequency offset performance, no CP (Cyclic Prefix) and virtual carrier. However, the non-overlapping sub-channel characteristics are realized by the non-perfect reconstruction prototype filter, which inevitably introduce ISI (Inter-symbol Interference), and ISI will be future increased especially when FMT is applied in wireless multi-path environment. Thus, the receiver must eliminate ISI by equalization. This dissertation mainly makes a deep research on the equalization in the FMT systems, including subchannel equalization, Turbo iterative equalization and blind equalization.Several characteristics of the wireless fading channel and the system model of the FMT are firstly analyzed. The principle of FMT is introduced according to the filter technology, and the fast algorithm based on IDFT (Inversed Discrete Fourier Transform) and DFT (Discrete Fourier Transform), the efficient structure of poly-phase filter are presented by means of graphic illustration. And then the matrix representation of the FMT is deduced. Moreover, the objects of the FMT equalizer over flat fading channel, frequency selective fading channel and time varying frequency selective fading channel are especially analyzed.A notable feature of FMT system is the small ICI and the ISI in the received signal can be eliminated by independent per subchannel equalization. Therefore, the dissertation then focuses on the subchannel frequency domain equalization algorithms in FMT system. Firstly, four algorithms based on DFE (Decision Feedback Equalization) are analyzed and these ABR (Achievable Bit Rate) and BER (Bit Error Ratio) performance are simulated and compared, the coefficients of the four DFE equalizers are also deduced according to the MMSE (Minimum Mean Square Error) criterion. However, the equalizer coefficients satisfying the MMSE criterion cannot guarantee the maximum SINR (Signal-to-interference-plus-noise Ratio), corresponding the optimal BER or ABR performance since the BER performance is related to the received signal’s SINR. According to the above-mentioned analyses, the CS-FMT (Critically Sampled FMT) subchannel linear equalization algorithm based on MMSE criterion is analyzed, and the subchannel linear equalization algorithms based on MSINR (Maximum the Signal-to-interference-plus-noise Ratio) criterion and MSB (Maximum Subchannel Bit-rate) criterion are proposed respectively by deducing the expression of the CS-FMT’s SINR and ABR. Simulation results show that the SINR, ABR and BER performance of the equalization algorithms based on MSINR and MSB criterion and these superiority in comparison with those of the linear and DFE equalization algorithm based on MMSE criterion. Moreover, the FLANN (Functional Link Artificial Neural Networks) are introduced into the subchannel equalization in FMT system to combat the nonlinear distortion of the channel, and a novel learning algorithm based on EKF (Extended Kalman Filter) is proposed to train the FLANN networks. Simulation results indicate that the proposed algorithm can simultaneously improve the convergence and steady-state error performance by comparing the performance of BP (Back Propagation) learning algorithm.Making full use of the redundant information from channel coding in the iterative process, turbo equalization has better performance, in which the equalization and decoding are jointly processed. Introducing turbo equalization into FMT system, the dissertation respectively proposes a kind of turbo equalization algorithm for the known channel and the unknown channel. When the channel is known, and the subchannel is flat, the channel fading can be compensated by a one tap per subchannel equalizer, whereas the ISI caused by the prototype filers can be eliminated by the turbo equalizer. When the channel is unknown, a FMT receiver scheme using joint channel estimation and turbo equalization is presented by applying the iterative channel estimation algorithm based on soft information. Simulation results show that the proposed turbo equalization algorithms can yield great improvements in BER performance compared with the DFE equalizer whenever the channel is known or unknown.Blind equalizers do not require a training sequence in the whole communication process. Lastly, the dissertation researches the application of blind equalization algorithm in FMT systems. Constructing a kind of new error function and building a nonlinear function between the step-size factor and the error function, a novel MS-MCMA (modified step-size blind equalization algorithm) blind equalization algorithm is proposed to alleviate the contradiction between convergence speed and steady-state error of the MCMA (Modified Constant Modulus Algorithm). Simulation results show that the proposed algorithm can simultaneously improve convergence speed and reduce the steady-state error. And then, the proposed blind equalization algorithm is used in FMT system. Simulation results show that the subchannel blind equalizer adopting MS-MCMA can perfectly eliminate ISI and the MS-MCMA can simultaneously improve convergence speed by one time and reduce the steady-state error by comparing the performance of the MCMA.更多还原