A DOA Estimation Method Based on an Improved Transformer Model for Uniform Linear Arrays with Low SNR

Figure 3

This paper proposes a model structure based on the Star-Transformer. After dimensional changes, the data must pass through the MHA-like layer with as the activation function and the LN layer with different association variables to obtain satellite nodes and relay nodes, respectively. After passing through the PFF layer, the output of the relay nodes must pass through the flatten layer to obtain one-dimensional output and pass through the FC layer. Sigmoid is used in the last FC layer to obtain the resolution angle probability. The MHA-like layer uses the dropout layer. The PFF layer and the updating of the satellite nodes use the residual structure.