一种易于电路实现的鲁棒盲神经网络波束形成器

绝大多数通讯信号都具有周期平稳信号特性。利用信号的周期平稳特性可以进行真正的盲自适应波束形成，因而受到了广泛关注。鲁棒ＣＡＢ（Ｒ－ＣＡＢ）算法就是其中的一种，但其中涉及矩阵求逆，运算量很大。波束形成属于多维信号处理问题，阵列中阵元数的增加导致运算量成几何级数增加。文中提出了一种盲神经网络波束形成算法。由于神经网络具有网状计算结构的特点，故该方法不仅避免了矩阵求逆运算，同时更便于实时实现。该方法还利用对角加载技术，从而加快了收敛速度，保证了算法的鲁棒性。仿真实验表明其性能优越，易于电路实现

Neural Network Based Blind Beamformer Suitable for Circuit Realization

Most of man made communication signals are cyclostationary. Because the blind beamforming algorithms using cylcostationary signal properties are the genuine blind algorithms, they attract much attention. R CAB algorithm is one of these algorithms. However, the matrix inversion process in CAB leads to the complicated computation. Beamforming is a kind of multidimensional signal processing problems. With the increase of the array sensors, the burdens of computation rise faster. A new blind beamformer with improved Hopfield network and the corresponding circuit are presented. With the parallel structure of neural network, it not only avoids the computation of the matrix inversion, but also makes easy to work in real time. Because the diagonal loading technique is employed, the converging speed is faster and the performance is robust. Simulations demonstrate that the proposed approaches have the better performance and can be easily realized by circuits.