B. Lall

HALO: Hybrid Adaptive Load Offloading of Grant-Based Traffic to Grant-Free Bandwidth for Spectrum Efficiency in Next Generation Networks

Efficient spectrum utilization is critical for next-generation cellular networks such as 5G-Advanced and 6G, which must support diverse services with heterogeneous quality-of-service (QoS) requirements. Grant-free (GF) communication in 5G new radio enables low-latency unscheduled transmissions but often suffers from underutilized resources due to sporadic traffic patterns. In contrast, grant-based (GB) communication offers reliable scheduled access but can experience congestion and resource exhaustion under high load. This paper proposes hybrid adaptive load offloading (HALO), a context-oriented deep reinforcement learning (DRL)-based framework that dynamically offloads low-priority GB traffic to underutilized GF resources during congestion periods. HALO effectively implements soft access class barring, redirecting rather than blocking traffic, without requiring complex bandwidth reconfiguration. The framework adapts to real-time network conditions to balance spectrum efficiency, information freshness of GF traffic, and QoS satisfaction across multiple service classes. Simulation results demonstrate significant bandwidth utilization gains of up to 40% compared to baseline approaches, highlighting the effectiveness of HALO for hybrid traffic management and its applicability to future 5G-Advanced and emerging 6G standards.

DeepQ-Charge: A Decentralized Model-Free Framework for Real-Time Mobile EV Charger Dispatch Using Deep Reinforcement Learning

Can Experts Adapt Without Training? On Test-Time Modality Generalization in MVLMs

Medical vision-language models (MVLMs) promise broad zero-shot generalization, yet their reliability collapses when confronted with unseen modalities and domains, precisely where clinical robustness matters most. To address this gap, we revisit test-time modality generalization from the perspective of Mixture-of-Experts (MoE) and ask: can experts route-and-adapt without any optimization during inference? We identify a fundamental specialization-generalization dilemma at test time, where blindly aggregating modality experts dilutes modality-specific knowledge, while selecting one highly confident expert risks mismatch under shift. To address this, we propose MoBE: a fully optimization-free framework that performs dynamic expert selection and adaptation at test time. MoBE combines entropy-guided dynamic routing in MoE settings with expert-wise Bayesian adaptation, enabling experts to update their confidence and adapt online without gradient updates. Without parametric updates, MoBE augments a static MVLM with test-time routing and online statistics, achieving average accuracy gains of +4.72, +7.17, and +4.3 over state-of-the-art TTA methods across seen, unseen, and heterogeneous medical benchmarks, highlighting the effectiveness of training-free expert adaptation for robust modality generalization.

Performance Analysis of FD C-NOMA With Intelligent TS and Battery Management in Non-Linear Energy Harvesting Networks

Driven by the demands of 5G/6G and internet of things (IoT) for extensive connectivity and enhanced spectral efficiency, this paper presents a full-duplex (FD) cooperative non-orthogonal multiple access (C-NOMA) framework that incorporates a battery-assisted practical non-linear energy harvesting (NL-EH) model. The three-node downlink network features a multi-antenna base station serving a near user (NU) (functioning as an FD relay) and a far user (FU) that employs maximal ratio combining (MRC) to process direct and relayed signals. We introduce an intelligent dynamic battery energy (DBE) management scheme that calculates the precise energy deficit per symbol required to exactly achieve target transmit power per symbol interval, ensuring stable QoS while minimizing the long-term ergodic battery energy consumption per symbol ( $\bar {Q}_{b}$ ) compared to fixed battery energy (FBE) methods. Concurrently, an adaptive dynamic time-switching (TS) protocol utilizes real-time channel state information (CSI) to optimize the EH time fraction, mitigating self-loop interference outages and adhering to energy causality constraints. A comprehensive mathematical framework derives exact closed-form expressions for outage probability, delay-limited throughput, ergodic capacity, and $\bar {Q}_{b}$ across these schemes. High-SNR asymptotic analyses demonstrate that, while an NL-EH architecture results in a zero-diversity error floor, integrating the direct link restores spatial diversity. Additionally, while perfect successive interference cancellation (SIC) enables continuous capacity growth, practical imperfect SIC leads to a zero high-SNR slope due to the dominance of interference. We found critical throughput starvation in the dynamic TS DBE scheme at high SNRs and propose enforcing a minimum IT fraction constraint or adopting a continuous non-linear mapping to ensure effective communication. We demonstrate selecting quantum of battery energy per symbol and TS parameter is crucial for achieving maximum FU throughput while ensuring a target throughput for the NU. Extensive simulations show the proposed dynamic TS and DBE frameworks outperform static benchmarks.