Array direction finding method based on deep capsule network under multipath effect

Abstract

The invention provides an array direction finding method based on a deep capsule network. According to the technical scheme, multi-path signal information obtained by an array is input into a deep capsule network, and all incoming wave directions of multi-path signals are obtained through calculation of the deep capsule network. The structure of the deep capsule network comprises two one-dimensional convolution layers, a main capsule layer, a digital capsule layer and a full connection layer which are connected in sequence; and the main capsule layer comprises four capsules which are connected in sequence. When the deep capsule network is trained, performing frequency division on the array observation data, and selecting a part, corresponding to each specific frequency point, of the array observation data; and dividing each part into a plurality of sub-regions in a spatial domain, extracting a feature vector of sub-region data corresponding to each specific frequency point, and sending the feature vector to the deep capsule network for parallel training. The method has the characteristics of high real-time performance and high applicability while realizing high-precision direction finding.

Description

technical field [0001] The invention belongs to the field of array signal processing, relates to an intelligent array direction finding method, and is especially suitable for complex multipath environments. Background technique [0002] In shore-to-ship and ship-to-ship direction finding positioning, due to the influence of ocean waves and the curvature of the earth, the signal may directly reach the receiving antenna of the array direction finding system, or may reach the array direction finding system after being reflected by obstacles such as the sea surface and clouds. to the receiving antenna of the system. Therefore, in addition to the direct signal of the target radiation source, the signal received by the array direction finding system also has a reflected signal, which is the multipath effect. Since the direct signal and its reflected signal can be considered to be sent by the same signal source, the two signals are highly correlated or coherent, which seriously af...

AI Technical Summary

Problems solved by technology

However, this sound signal DOA estimation method based on deep learning takes advantage of the time-domain characteristics of the sound signal, and it needs to accumulate long-term observation data to achieve direction finding, and it is only for single-source sound signal direction finding scenarios, or At the same time, different frequency sound signal direction finding scenarios cannot adapt to electromagnetic signal direction finding scenarios under multipath effects.

Images