| In the third generation mobile communication system, there is a key technique named smart antenna. One of the crucial techniques of the smart antenna is the direction of arrival estimation, which has an widely application involves the communication, radar, sonar, earthquake survey and so on. In this paper, the main work is to do some research on the DOA estimation algorithms based on particle filter.First, this paper introduce the sub-space algorithm of DOA estimattion detailedly. Multiple Signal Classification (MUSIC) algorithm was first published by Schmidt in 1979. The put forward of this algorithm founded the space spectral estimation algorithms, and it also makes a promotion of the characteristic structure algorithms which has been made a symbol of the space spectral estimation theory system.Another important kind of sub-space algorithm is ESPRIT algorithm which was put forward by Roy and Kailath in 1986. ESPRIT is a typical algorithm of modern DOA estimation and it also need to process the received data by making characteristic decomposition of the covariance matrix like MUSIC algorithm. Compared with the MUSIC algorithm, the advantages of the ESPRIT algorithm are low calculation complexity and no need to search the spectral kurtosis.The DOA matrix algorithm is an effective two dimensional DOA estimation algorithm, it has advantages such as no need to search the spectral kurtosis, has low calculation complexity and two dimensional parameter couple together automaticly by the corresponding relationship of the eigenvalue and eigenvector. At the same time, the DOA matrix algorithm has the disadvantages of angle annex when arbitrary incidence angle has the same value, but the modified time-space DOA matrix algorithm solves this problem effectively.The sub-space algorithm of DOA estimation can achieve a very high precision and resolution, but it needs to observe the received signal for a while to obtain its DOA. Specially with the lower SNR, the sub-space algorithm needs to observe so long a time, therefore it is not ideal to estimate the movement DOA for real-time demand. The Bayesian algorithm may estimate the movement DOA effectively, moreover, it has very good DOA tracking capacity.The extend kalman filter is a classical algorithm in nonlinear estimation field. It has simple procedure and easy to come true, but it only uses the first order expansions of the Taylor series, the linearization will bring great error when the high order item can not be ignore. Further more, it is difficult to obtain the Jacobian matrix of the nonlinear functions in many actual problems. Hereby we must make improve on this algorithm for actual use.Because it is more easily to close to a nonlinear function distribution than close to a nonlinear function itself, the methods solve nonlinear problems by close the nonlinear distribution using the sampling way has been got the researchers extensive concern recently. The unscent kalman filter is a kind of this methods. The UKF sample way is a kind of determinately sample, which has a few of sigma points and the exactly number of sigma points is determined by the chosen sample policy. The UKF has the same calculation complication with EKF, whereas it has better performance and not need to calculate the Jacobian matrix.In the communication system, DOA estimation is a strong non-Gaussan non-linear problem. The sub-space algorithm and the non-linear kalman filter all needs the Gaussan noise supposition, which has limited the algorithm in actual use. Particle filter is a kind of recursive Bayesian filter algorithm based on Monte Carlo method. It has been a pop and effective algorithm in the research of non-Gaussan non-linear system and is wildly used in target tracking, information processing and digital communication areas.The basic thought of the sequential importance sampling algorithm (SIS) is to construct the required posterior distuibution by a series of weighted particles. It can be gone into particulars as follows: first, we use the system model to predict the state prior distuibution. Then modify it by the new observation to obtain the posterior distuibution and the state estimation. The main problem of SIS is degeneracy phenomenon. The simplest method to slove this problem is to increase the sample numbers, but this method is limited by the high calculation complexity. So the main measures we chose is to select the importance distuibution propriety or resamping.SIR particle filter is the combination of SIS particle filter and resamping method. The resamping method can avoid the degeneracy phenomenon, but it brings new problems which named sample impoverishment. The resamping method causes reduplicate particles which is harmful to the particle diversity. The particles with high weights have been sampled many times and other particles disappear gradually.In the SIR particle filter, the sampled particles are far different from the real particles sampled from the posterior distuibution because the importance distuibution has not taken into account the new observation. UPF update the particles with UKF on the frame of SIR particle filter. Because of it takes into account the new observation, the UPF can obtain better importance distuibution and better performance.Particle filter is a kind of effective DOA estimate algorithm which can estimate the movement DOA for real-time demand in the non Gaussan entironment and it is easy to come true. The SIR particle filter which use resamping method can effectively avoid the degeneracy phenomenon, but it has a high calculation complexity for the great sample numbers The unscent particle filter can obtain better performance than the SIR particle filter. Specially, it can reduce the sample numbers and enhance the calculation speed, thus has an important practical significance.The real-time DOA estimation based on Bayesian algorithm which has been introduced before needs to know the number of sources. Moreover, it needs the supposition that the source number is invariable in the estimation process. But in communication system, the number of sources is usually unknown and variable.Based on the SIR particle filter, this paper put forward a DOA tracking algorithm with unknown number of sources by design a importance distuibution close to optimal and combine with the RJMCMC method. By adding a MCMC move on each particle, we can estimate the system order effectively, which is the same meaning as we can estimate the number of sources on-line. At the same time, the RJMCMC method increases the diversity of the particles and avoids the sample impoverishment effectively. The simulation approve that the combination of SIR particle filter and RJMCMC method is availably to track the signal DOA with unknown number of sources.Compared with other Bayesian algorithm, the algorithm put forward by this paper which do not need to know the number of sources and do not need the Gaussan noise supposition is more easier to come ture. Compared with the sub-space algorithm, the algorithm put forward by this paper can estimate the movement DOA effectively with no Gaussan noise supposition. It has obvious advantage in the strong non-Gaussan and non-linear entironment. |
