Elevatable and Rotatable Antenna Array for Maritime Directional Modulation: Transformer Scheme Cuts Power by About 90% and Antennas by 87% While Raising Worst-Case Secrecy Rate

An elevatable and rotatable antenna array architecture is proposed for maritime communications with lower implementation complexity than 6D movable antennas, incorporating transceiver hardware impairments and jointly optimizing array height, array orientation, antenna orientation, and precoding vectors for worst-case secrecy rate maximization. Prior antenna-mobility studies mainly target terrestrial scenarios and typically involve small-scale element movement; this work brings height adjustability plus array and element rotation into maritime physical-layer security design and includes hardware impairments in the optimization model. The contribution is presented through the system model and problem formulation: position expressions for array rotation and elevation, transmit and receive signal models with hardware impairment coefficients, and optimization problem P1 with QoS, artificial-noise zero-interference, unit-norm, power allocation factor, and angle feasibility constraints.

For the single-user case, an alternating optimization algorithm decomposes the non-convex problem into information precoding, artificial-noise precoding, power allocation factor, and antenna configuration subproblems, where the power allocation factor admits a closed-form solution and the antenna configuration is set by orienting toward the legitimate user while suppressing the eavesdropper's achievable rate. The information precoding subproblem is solved via semidefinite relaxation and successive convex approximation, the artificial-noise subproblem becomes a generalized Rayleigh quotient solved by the largest eigenvector, and power allocation uses a first-order derivative yielding at most two interior stationary points compared against boundary points, giving the global optimum over a finite candidate set. The method section provides equivalent transformations and solution steps for each subproblem and states complexity orders: overall complexity arises from AO and SCA iteration counts, semidefinite program solving, eigenvalue decompositions, and the power allocation update.

For the multi-user case, a Transformer-based learning scheme models the coupling among channels, user locations, and antenna states through self-attention, using unsupervised training with online joint refinement to output precoding and antenna configurations. Directly extending the alternating optimization framework to multiple users incurs a considerably higher computational burden, and different users' optimal variables are strongly correlated through the shared propagation environment; the scheme approximates the worst-user rate with a soft minimum and adds QoS and artificial-noise penalties, while the online stage screens candidates with feasibility criteria. Training and inference are specified through the loss function, penalty terms, Adam update, feasibility conditions, and complexity orders; in simulations the scheme generally achieves higher worst-case user secrecy rates across different antenna counts, user counts, and eavesdropper counts.

Simulations show that optimizing array height, array orientation, and antenna orientations improves the secrecy rate, but the gain from raising the antenna alone is limited by path loss; taking a worst-case secrecy rate of 5 bps/Hz as reference, the proposed scheme reduces transmit power by about 10 dB versus isotropic antennas (about 90% power savings), cuts array size from about 128 elements to about 16 elements (about 87% antenna savings), and increases supported legitimate users from about 3 to about 5 (about 67% improvement) while guaranteeing about 2 bps/Hz worst-case secrecy rate. The results quantify the benefit of antenna physical degrees of freedom across power, antenna scale, and number of served users, and indicate that random antenna orientation, isotropic antennas, and fixed power allocation perform relatively lower. Conclusions come from simulations: carrier frequency 3 GHz, path loss exponent 2.2, shadow fading 2 dB, noise power -95 dBm, equal transceiver hardware impairment coefficients, and Bob minimum rate 1 bps/Hz, with convergence curves and secrecy-rate trends versus transmit power, antenna count, user count, hardware impairment coefficient, channel estimation error, and antenna directivity factor.

The results target maritime communication settings where the base station has an elevatable and rotatable planar array, users and the eavesdropper are single-antenna, and all lie within the maximum transmission distance; benefits grow with transmit power and antenna count and shrink as user count, hardware impairments, and channel estimation error increase. For a reader, this offers a way to bring antenna physical degrees of freedom into physical-layer security design: in long-range maritime links, jointly adjusting array and element orientations raises the secrecy rate more than raising the antenna alone, while the main value of elevation is extending the visible range and improving propagation accessibility.

The simulation conclusions depend on the adopted maritime channel model and parameter settings, and actual sea conditions, antenna mechanical precision, and channel estimation error distributions may change the magnitude of the gains; the hardware impairment coefficient is set equal at transmitter and receiver in simulations, so the effect of real device differences remains to be examined; the multi-user scheme involves offline training and online refinement, and how training cost and feasibility-criterion thresholds affect the final secrecy rate at different scenario scales is a direction a careful reader may continue to watch.