optimize power allocation

This work addresses the significant performance degradation of conventional water-filling power allocation in the power amplifier (PA) saturation region, which stems from its neglect of PA nonlinearities. Departing from the common linear assumption, this paper is the first to explicitly model the hard-limiter nonlinearity of PAs within the power allocation framework. Leveraging Bussgang’s theorem for statistical linearization, it characterizes the trade-off between signal gain and distortion-induced noise, and jointly optimizes power allocation with a spatial back-off strategy. By introducing a closed-form threshold on thermal noise variance that distinguishes between noise-limited and distortion-limited regimes, and combining projected gradient descent with channel Frobenius norm analysis, the proposed amplifier-aware method substantially enhances system capacity in the PA saturation region, as confirmed by simulations.

This work addresses adversarial resource allocation under spectrum sharing in multi-operator low Earth orbit satellite systems by formulating it as a minimax game between transmit power and worst-case interference. It introduces, for the first time, the adversarial water-filling (AWF) framework to this setting, integrating optimization theory with learning-based methods to effectively tackle non-convex resource allocation problems under both Gaussian and discrete constellations. The authors propose learnable projected hypergradient dynamics and develop a wireless foundation model featuring permutation invariance, constraint-aware graph neural networks, sparse message passing, and a global water-level latent variable. This model demonstrates strong generalization across unseen problem scales, constraints, and constellation types, achieving solution speeds over an order of magnitude faster than conventional iterative methods.

This work addresses energy efficiency and latency optimization in integrated sensing and communication (ISAC) systems under imperfect information. The authors jointly optimize time-slot allocation, beamforming adaptation, functionality selection, and user–target pairing to minimize energy consumption while prioritizing time savings, accounting for uncertainties arising from target dynamics, quantization errors, feedback delays, and hardware constraints. The problem is innovatively formulated as a semi-infinite nonconvex mixed-integer nonlinear program. By exploiting hidden convexity, the authors develop a structure-aware exact reformulation that equivalently transforms the problem into a globally solvable mixed-integer semidefinite program (MISDP). Simulations demonstrate that the proposed approach achieves up to 88% resource savings compared to baseline schemes and reveals strong coupling among the various resource management components.

This work addresses power allocation over parallel Gaussian channels—such as OFDM subcarriers—under a total power constraint, aiming to minimize the sum of squared deviations between achieved spectral efficiencies and prescribed targets. By analyzing the KKT conditions, the study reveals a novel structural property: the optimal solution never overshoots the target spectral efficiencies and may leave part of the total power unused, thereby departing from the classical water-filling paradigm. A closed-form solution is derived using the Lambert W function, and the associated dual variable is efficiently computed via a one-dimensional monotonic bisection method with complexity O(N log(1/ε)). Numerical experiments demonstrate that, for N = 1024, the proposed algorithm achieves machine-precision accuracy and is up to 1890× faster than generic numerical solvers, while significantly outperforming water-filling and other baselines in target tracking performance.

This work addresses the dual-timescale optimization challenge of joint power control and beamforming in cell-free massive MIMO—specifically, the scalability bottleneck in uplink power minimization under max-min fairness constraints. We propose a long-term joint optimization framework leveraging statistical channel state information (CSI). Innovatively, beamforming is modeled as a parameterized mapping function, co-optimized with power coefficients, thereby overcoming the high computational and signaling overhead of conventional short-term iterative algorithms and the suboptimality of existing long-term methods that decouple beamforming from power control. By integrating stochastic optimization, functional-space parameterization, and distributed implementation, the framework significantly reduces computational complexity and inter-access-point signaling load. Simulation results demonstrate that the proposed method consistently outperforms state-of-the-art short-term and long-term baseline algorithms in terms of both power efficiency and fairness performance.

This work addresses the performance degradation in user-centric cell-free massive MIMO systems caused by inaccurate channel state information. To mitigate the adverse effects of channel estimation errors while maintaining computational efficiency, the authors propose a robust power allocation method based on a Tikhonov-regularized least-squares framework, integrated with zero-forcing precoding. This study is the first to introduce Tikhonov regularization into power optimization for this specific setting, achieving a favorable balance between robustness and low complexity. Simulation results demonstrate that, under channel uncertainty, the proposed scheme significantly outperforms existing non-robust approaches in terms of system performance, while incurring low computational overhead—making it well-suited for large-scale deployment.

This work addresses the vulnerability of deep learning–based power control in massive MIMO systems to input perturbations—such as user location errors—and the absence of formal robustness guarantees. It presents the first formal verification framework for deep neural networks in regression tasks with nonlinear output constraints, combining DeepPoly abstract bound propagation with constraint programming. Adversarial perturbations are modeled as hyperrectangular sets, and a constrained numerical feasibility problem is formulated to ensure a provable lower bound on performance. Experimental results demonstrate that well-trained models exhibit local robustness under position perturbations of up to ±1 meter, with an optimality gap no greater than 1%.

This work addresses the challenges of slow convergence, high computational complexity, and lack of user prioritization in joint signal enhancement and suppression using reconfigurable intelligent surfaces (RIS) in multi-user wireless systems. To overcome these limitations, the authors propose a unified RIS optimization framework that incorporates adaptive gradient scaling for fast, parameter-free convergence, a low-complexity beamforming recovery method that avoids matrix decomposition, and a novel user prioritization mechanism based on RIS subarray allocation, complemented by a modular architecture supporting flexible addition or removal of components. Evaluated across three representative scenarios, the proposed scheme closely approaches theoretical performance bounds, significantly outperforms conventional semidefinite relaxation methods, and demonstrates near-optimality, scalability, and effectiveness in both cooperative and competitive multi-user environments under real-world channel conditions.

This work addresses the challenge of guaranteeing statistical end-to-end latency and accuracy quality-of-service in multi-cell edge intelligence systems under spatiotemporal uncertainties. To this end, the authors propose a joint wireless and computational resource pre-deployment optimization framework. By integrating Poisson point processes, queueing theory, and empirical AI inference workload measurements, they establish a unified stochastic modeling framework and, for the first time, derive an analytically tractable expression for end-to-end offloading latency. The resulting non-convex joint optimization problem is decomposed into convex subproblems, enabling the attainment of a globally optimal solution. The study further uncovers fundamental trade-offs among base station density, cell size, transmission latency, computational cost, and user fairness, and identifies a cost-efficient design regime in interference-limited scenarios.

This work proposes a dual-process architecture integrating large language models (LLMs) and numerical optimization to enable policy-driven power reconfiguration in communication systems under safety constraints. The LLM acts as a high-level policy interpreter, translating natural language instructions into adjustments of channel weights and power budgets, while a fast optimizer performs constrained power allocation via projected gradient ascent with a weighted mutual information objective. Inspired by System 1/System 2 cognitive mechanisms, the framework incorporates multiple reliability safeguards—including exponential smoothing, normalization, and fallback strategies—to ensure robust operation. Experimental results in an 8-channel setting demonstrate the system’s ability to flexibly satisfy diverse policy requirements and autonomously reconfigure upon abrupt channel changes, reducing the dispersion of mutual information distribution by 60%.