Recearch on stereophonic acoustic echo canceller with both high sound quality and fast convergence speed

1. 采用双立体声回波消除器的快速收敛算法，分析了不规则解在频域上的分布。通过估计两个独立的不规则解并求解一个含四个未知变量的联立方程，可以确定所有不规则解的分布；通过确定两个健谈者的不规则解并求解另一个含四个未知变量的联立方程，可以找到最优解。使用两个初始系数不同的saec可以估计出两个不规则解。为了估计独立解，周期性地评估独立性。当两个解不够独立时，saec被初始化。选择一个新的初始值，使其与另一个系数正交。实时信号处理系统介绍了采用第一种数字信号处理器的新型实时信号处理系统。对于基于pc的低成本电话会议，SAEC已经在Intel IA-32处理器上实现。为了有效地使用向量运算，进行了一些优化。优化后的程序几乎是传统代码的四倍基于输入信号白化的快速收敛算法研究了利用输入信号白化来提高收敛速度的算法。通过修改白化滤波器的位置，该算法无需像基于格点的白化方法那样同步主滤波器系数与白化滤波器系数。虽然这种方法在水龙头数量较少时是有效的，但对于大量水龙头的改进应该是未来的研究。

1. Fast convergence algorithm by using two stereophonic acoustic echo cancellers (SAECs)The distribution of the irregular solutions in the frequency domain has been analyzed. By estimating two independent irregular solutions and solving a simultaneous equation with four unknown variables, the distribution of all irregular solutions can be determined The optimum solution can be found by determining the irregular solutions for two talkers and solving another simultaneous equation with four unknown variables A fast convergence algorithm using two SAECs has been developed. Two irregular solutions can be estimated by using two SAECs with different initial coefficients. To estimate independent solutions, the independency is evaluated periodically. The SAECs are initialized when two solutions am not independent enough. A new initial value is so selected as to be orthogonal against another coefficient.2. Real-time signal-processing systemsNew real-time signal-processing systems with fist digital signal processors has been introduced.3. Implementation of SAEC on Intel IA-32 Processors For low-cost PC-based teleconferencing, an SAEC has been implemented on Intel IA-32 processors. Some optimization have been carried out to use vector operations effectively. The optimized program is almost four times as fast as a conventional code.4 Fast convergence algorithm by input-signal whiteningAn improvement of the convergence speed has been studied which utilizes input-signal whitening. By modifying the locations of the whitening filters, the proposed algorithm dips not have to synchronize the main filter coefficients with those for the whitening filter as in the Lattice-based whitening method. Though this approach is effective when the number of taps is small, an improvement for a large number of taps should be a future study.