Deep Learning-Driven Hybrid Beamforming, Channel Estimation, and Resource Allocation for Enhanced mmWave MIMO-OFDM Systems
Year: 2026
Type: article
Abstract: New wireless networks use millimeter-wave large MIMO-OFDM systems, however beam squint effects, pilot overhead, and low power and spectrum allocation under dynamic channel conditions limit their effectiveness. These challenges are addressed by an integrated deep learning-driven system that improves hybrid beamforming, adaptive channel prediction, and dynamic resource allocation in wideband mmWave contexts Framework includes multi-scale graph attention-based hybrid beamforming, transformer-based adaptive channel estimation with frequency-aware memory, meta-learning-enhanced deep reinforcement learning for resource allocation, sparse coding-based beamforming enhancement, and multi-agent reinforcement learning for coordinated beam alignment and powers. Real-world urban mmWave studies indicate considerable improvements in all performance metrics. The framework boosts spectral efficiency by 30-40% across antenna and user configurations, peaking at 28.5 bits/s/Hz in large-scale MIMO. Improved channel estimate accuracy by 35%, lowering normalized mean square error to -22.1 dB and pilot overhead by 35%. Wideband beam misalignment is considerably minimized by 96.5% beamforming accuracy. Dynamic resource allocation efficiency surpasses 97%, saving 25-35% power without compromising throughput. The framework is appropriate for 5G and 6G ultra-reliable and high-capacity wireless networks because closely integrated deep learning architectures can enable scalable, low-latency, and energy-efficient mmWave MIMO-OFDM communication sets.
DOI: 10.54392/irjmt2638
Authors: Anuprita Linge, Rahul Pethe, Abhay Kasetwar
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7160837161