An Improved Adaptive Fast Iterative Convergence Solution Method and System

Abstract

The invention belongs to the technical field of signal detection, and discloses an improved self-adaptive fast iterative convergence solution method and system, which approximates the parameter iteration process as Taylor's second-order expansion, so that the invention is not sensitive to the selection of the initial value, and is different from the traditional Compared with the method, it does not increase any computational complexity. The second and third derivatives of the logarithmic likelihood function are introduced in the correction item, that is, the slope and curvature characteristic mathematics of the logarithmic likelihood ratio function are added, so that it can be adaptive Adjusting the fluctuation of the correction term accurately, using the correlation between the second and third derivatives of the log-likelihood function, an adaptive parameter estimation threshold and an iterative convergence stopping condition are developed. The invention can not only ensure fast iteration, but also ensure convergence, and is not sensitive to the selection of the initial value.

Description

technical field [0001] The invention belongs to the technical field of number detection, and in particular relates to an improved self-adaptive fast iterative convergence solution method and system. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] The prior art close to this scheme is exactly the content described in the following paragraphs. [0004] Among the techniques related to iterative solutions, the techniques traditionally used include Newton-Raphson iterative method, scoring method, and EM method. The common feature of these methods is to select an initial value for the parameter to be estimated, and then follow the Taylor series Expand, and iteratively approach the parameters to be estimated. However, this method may cause the iteration to not converge. For example, due to the large influence of data noise, when the second-order derivative of the logarithmic likelihood function is small, t...

AI Technical Summary

Problems solved by technology

[0006] (1) In the prior art, the iterative convergence speed in signal data detection is slow, and the convergence threshold cannot be adaptively adjusted according to the actual situation, resulting in the limitation of a large number of parameter estimates in related physical applications

[0007] (2) Since the conventional iterative method is based on the principle of first-order Taylor series expansion for iterative estimation, this method only develops the second-order derivative of the log-likelihood function, and does not develop the higher-order derivative of the log-likelihood function. When the second-order derivative of the logarithmic likelihood function is small, the parameter iteration does not converge due to the large fluctuation of the correction item

The conventional method does not set an adaptive convergence threshold, and lacks a mechanism to suppress the fluctuation of the correction item. This problem seriously restricts the application of the iterative method.

[0008] (3) For example, in applications related to passive sensing, the receiver is a sensor array, and the transmitter is an enemy military radar. The purpose of passive sensing is to estimate the azimuth and elevation angles of local radars through passive sensor arrays. In the research of high-precision parameter estimation, such problems are generally realized by the ML (Maximum Likelihood) method. The biggest disadvantage of ML is the high computational complexity, which cannot meet the calculation requirements of real-time tasks under engineering conditions, so the numerical solution algorithm is often the first choice. , the conventional numerical iteration method usually faces the problem that the convergence of the iteration cannot be guaranteed, and the convergence value largely depends on the setting of the initial value. Sometimes even if it converges, the convergence speed is slow, and it is difficult to meet the requirements of this type of task. However, the present invention The proposed method not only guarantees convergence, but also has a fast convergence speed, and is not sensitive to the initial value setting

Images