Bayesian Demodulation Algorithms for Digital Communications

9406073 FITZ贝叶斯滤波算法系列是最近开发的用于数字通信的解卷积(已知和未知信道)算法系列。贝叶斯滤波器是K-滞后最优逐符号估计器的近似，具有其独特的特征，即其产生的指数复杂度与滤波器滞后和信道长度两者解耦。该项目1)将对贝叶斯滤波算法进行重大改进，2)将对这些算法进行分析表征以更深入地了解其性能，3)将研究这些算法在数字通信中的一些应用，以及4)将在定点DSP架构中实现这些算法。贝叶斯滤波器可以被改进以包括明显更一般的判决反馈结构，以适用于更一般的信道模型，以更准确地对时变衰落建模，并实现最优组合算法。一般的判决反馈结构既类似于简化状态序列估计，又是其重要的推广。这种结构将在实现长解码延迟和保持合理的复杂性水平之间提供良好的折衷。本项目中考虑的信道模型将是自回归移动平均模型和部分观测马尔可夫链模型。为这些信道导出去卷积算法使得贝叶斯技术能够解决更广泛的问题。最后的概括是对时变信道的建模。如果能够在快速衰落的频率选择性信道中实现高性能的解卷积，那么数字通信系统(特别是使用TDMA的系统)的容量将有显著的改善。最后，将在贝叶斯框架中考虑对每个符号的多于一个观测进行最佳组合的方法。这种类型的典型应用包括天线分集系统和分数间隔均衡。贝叶斯滤波有很多潜在的应用领域，我们将对这些应用进行详细的探索。目标领域包括：利用记忆进行非线性调制的低复杂度解调；差错控制码的低复杂度解码；高速高频(HF)数字通信；磁记录；以及陆地移动无线电。***

9406073 Fitz The Bayesian filter family of algorithms is a recently developed family of deconvolution (both known and unknown channel) algorithms for digital communications. The Bayesian filters are approximations to the K-lag optimum symbol-by-symbol estimator with the unique characteristic that its resulting exponential complexity is decoupled from both the filter lag and the channel length. This project 1) will produce significant refinements to the Bayesian filter algorithms, 2) will produce an analytical characterization of these algorithms to provide greater insight into the performance, 3) will examine some applications of the algorithms in digital communications, and 4) will implement the algorithms in a fixed point DSP architecture. The Bayesian filters can be refined to include significantly more general decision feedback structures, to be applicable with more general channel models, to more accurately model time-varying fading, and to implement optimal combining algorithms. The general decision feedback structures are both similar to and a significant generalization of reduced state sequence estimation. This structure will provide an excellent tradeoff between achieving a long decoding lag and maintaining a reasonable level of complexity. The channel models considered in this project will be the autoregressive moving average model and partially observed Markov chain model. Deriving deconvolution algorithms for these channels enables a wider range of problems to be addressed by Bayesian techniques. The final generalization is in the modeling of time-varying channels. If high performance deconvolution can be accomplished in a rapidly fading frequency selective channel, then significant improvements are available in the capacity of digital communication systems (especially ones using TDMA). Finally, methods for optimally combining more than one observation per symbol will be considered in a Bayesian framework. Typical applications of this type include antenna diversity systems and fractional spaced equalization. The Bayesian filter has a great many potential applications areas and these applications will be explored in detail. The areas targeted are: reduced complexity demodulation of nonlinear modulations with memory; reduced complexity decoding of error control codes; high speed high frequency (HF) digital communications; magnetic recording; and land mobile radio. ***