Authors: Ajay Damor, Dr Nidhi Tiwari, Professor Madhavi S Bhanwar
Abstract: With the surge of data demands, ultra-reliable low-latency communications (URLLC), and massive connectivity envisioned in 6G networks, accurate and efficient channel state information (CSI) acquisition becomes critically important. Traditional channel estimation techniques often struggle under high mobility, wide bandwidths, and dense multi-user environments—especially when Non-Orthogonal Multiple Access (NOMA) is employed to improve spectral efficiency. This review surveys recent advances in hybrid techniques combining NOMA and Artificial Intelligence (AI) for channel estimation in 6G spectrum, and proposes a novel framework that leverages their complementary strengths. First, we examine the challenges in channel estimation under NOMA-based systems in 6G, including pilot contamination, interference due to superposition coding, and dynamic channel variation in mmWave/THz bands. Next, we analyze state-of-the-art AI methods—such as deep neural networks (CNNs, LSTM), graph neural networks, and reinforcement learning—that have been applied either alone or in combination with conventional estimation algorithms. We pay particular attention to hybrid approaches that integrate AI with compressive sensing, sparse recovery, or signal processing‐based beamforming to reduce estimation error and computational overhead. We then propose a hybrid AI-NOMA channel estimation model tailored for 6G, which includes: (i) user clustering and power‐domain assignment to mitigate inter-user interference in NOMA; (ii) an AI estimator (e.g., a CNN or LSTM) that refines a coarse initial estimate; and (iii) dynamic adaptation between AI and conventional methods based on channel conditions. Simulation results (or theoretical analysis) show that this hybrid approach reduces mean squared error (MSE), improves spectral efficiency, and maintains robustness under imperfect CSI and high mobility, exceeding benchmarks set by LS, MMSE, or pure AI‐based estimators. Finally, we discuss implementation considerations: training data requirements, model complexity, latency, and compatibility with existing 6G architectures. Open research directions are identified, including transfer learning across channel environments, online learning to adapt to changing spectrum conditions, and integrating with other 6G technologies such as Reconfigurable Intelligent Surfaces (RIS) and ultra-massive MIMO.
