Fig. 1. Mirror metasurface-based target sensing and beamforming system. Electromagnetic waves emitted or reflected from various targets in space are intercepted by the metasurface and converted into transmitted electric fields. These fields are then fed into the intelligent resolution framework of signal receiver module. This framework detects the number n, pitch angle θ, and azimuth angle φ of all targets and transmits the sensing results, along with the specified requirements, to beamforming module. Beamforming module then generates the four modes of metasurface topological patterns.

Fig. 2. A nine-beam metasurface design process. (a) The structure of a 1-bit polarized rotating unit cell. (b) Simulated transmission phase and transmission amplitude curves of the unit cell at 8–12 GHz. (c) The phase distribution of the array is optimized by GA based on expected beams in nine directions. (d) The process of optimizing the phase distribution of the array through the GA. (e)–(h) Simulated and theoretical far-field pattern in the direction of (e) φ=0°, (f) φ=60°, (g) φ=90°, and (h) φ=150°.

Fig. 3. (a) DNN comprises one input layer, five hidden layers, and one output layer, with the number of nodes in the hidden layers set at 15, 30, 50, 30, and 15. (b) After training, the network exhibits satisfactory convergence. (c) The outputs of DNN and the sampled electric field sequences are simultaneously utilized as input parameters for GA, which undergoes improvement through the selection of preferred individuals and random DNA recombination. (d) Comparative analysis of the perceptual accuracy between classical GA and the improved version DNN-GA indicates an improvement from 55% to 70% and to over 80%. (e) A confusion matrix for DNN prediction of the number of targets reveals an overall accuracy exceeding 98%. (f) Numerical simulation verification result.

Fig. 4. Optimization process, results, and simulations for beamforming module. (a), (d), (g), (j) The four modes undergo a GA optimization process: (a) tracking targets mode (mode T), (d) avoiding targets mode (mode A), (g) hybrid mode combining tracking and avoidance (mode H), and (j) reducing RCS mode (mode R). (b), (e), (h), (k) Resulting in the generation of (b) mode T, (e) mode A, (h) mode H, and (k) mode R with 2-bit phase arrangements. (c), (f), (i), (l) Simulations confirm that (c) mode T, (f) mode A, (i) mode H, and (l) mode R align with the design expectations.

Fig. 5. Experimental measurement. (a) The experimental system comprises signal receiver module, target sensing module, and beamforming module. Signal receiver module operates within a microwave anechoic chamber, where electromagnetic waves emitted from targets in different directions are converted into sequences of electric field signals by a vector network analyzer. These signals are then transmitted to a computer platform for validation of target sensing module and beamforming module. (b) Measured transmission electric field curves. (c) Target location perception results.