| Broadband signal DOA estimation is important in the area of array signal processing. For the broadband DOA estimation algorithm basing on signal-subspace, the pre-estimation of incident signals's DOA is needed to achieve focusing matrix, but the closeness of pre-estimated results and real-direction of signal have great impact on the final estimation results. Rough estimates will make the performance drop significantly. So to analyse and search broadband DOA estimation algorithm, which the pre-estimation of incident signals's DOA is unnecessary, is worthy of attention. In this dissertation, it is in-depth analysed and researched that broadband DOA estimation algorithm which the pre-estimation of incident signals's DOA is unneccessary. And a lot of simulation and comparison have been done.In the second part of thesis, narrow-band signal model and wide-band signal model of DOA estimation are described, as well as some basic theory.In the third part, non-coherent method and the unnecessity of initial angle estimation method in coherent processing are described. For non-coherent method, TOPS and IMUSIC methods are described in detail. Experimental simulation results show that IMUSIC method has better estimation performance than TOPS method.For the unnecessity of initial angle estimation method in coherent processing, CSM, RSS, BI-CSM and R-CSM methods are described in detail. Comparative analysis is done for various methods. Analysis and simulation results show that RSS is of the most stable performance. 5 or above iterative steps are needed to achieve stable performance in low SNR circumstances; in the middle of the SNR or above circumstances, at least 3 iterative steps are needed. Besides, simulation of the application of RSS in broadband channelization receiver array is done. The simulation shows that RSS is a good choice for broadband channelization receiver array. A common DOA estimation is needed instead of once DOA estimation for every sub-channel in which signals exist when RSS is choosen as the DOA estimation algorithm.In the forth part, RBF neural network is brought in narrowband DOA estimation. Two methods which use different variables as input are proposed. These two methods are compared with MUSIC. The method which takes covariance matrix's elements as input has poor performance in low SNR circumstances; however, because of unneccesity for signal-subspace computation and spectral peak scan operation, the speed of the method which takes covariance matrix's elements as input is faster.RSS-RBFNN is proposed by combining RSS method and RBF neural network. Firstly, data from different frequencies are focused to reference frequency; secondly, the covariance matrix on reference frequency is computed; at last, take the element of covariance matrix into RBF neural network. At the same SNR, the RSS-RBFNN broadband DOA estimation method has better performance than the narrow RBFNN DOA estimation.Because the performance is better when the focusing region is smaller, a regional broadband DOA estimation method PNN-RSS is proposed based on probabilistic neural network (PNN),which is of highly fault-tolerant capability. When SNR < ?5 dB, the probability of successful judge is close to 100%. Estimation RMSE is reduced in PNN-RSS algorithm compared to RSS. At the same time, the identifying capacity is enhanced. For the parallel computing and regional scanning characteristics of PNN-RSS, the PNN-RSS needs less time than RSS to achieve one estimation result.All work is concluded at last. |
