2026· IEEE Transactions on Wireless Communications· Vol 25, pp. 21056-21072· 0 citations· 56 references
Computer Science
Abstract
With the growing number of antennas in massive multiple-input multiple-output (MIMO) systems, robust and fast channel estimation becomes increasingly critical yet remains highly challenging. In this work, we propose a lightweight zero-shot self-supervised (ZS-SS) learning framework. It leverages non-local self-similarity in wireless channels to construct a channel-coefficient bank and generate training pairs via randomized and non-contiguous spatial permutations to decorrelate noise. These pairs then train a compact convolutional neural network (CNN) with a specially designed composite loss for robust channel estimation. To further improve adaptability and efficiency, we incorporate a meta-learning approach for fast inference time to dynamic channel environments. Simulations under Gaussian and representative non-Gaussian scenarios show that our method achieves up to 90% gains over traditional estimators and consistent improvements over state-of-the-art baselines, while running nearly 100 times faster. This demonstrates its practicality and suitability for real-time deployment in resource-limited massive MIMO systems.
Massive multiple-input multiple-output (MIMO) technology underpins the spectral efficiency targets of fifth-generation (5G) New Radio (NR) networks, and its performance depends critically on accurate channel state information at the receiver. Classical pilot-based channel estimators face three fundamental constraints: pilot overhead that scales with the antenna count, matrix inversion complexity that grows cubically with the array size, and severe noise sensitivity at low signal-to-noise ratio (SNR) or under high-mobility conditions. The objective of this study is to develop a channel estimator that accurately reconstructs the full time-frequency channel response from a sparse pilot grid while remaining computationally feasible for real-time operation. A two-stage hybrid deep learning estimator is proposed in which least-squares estimates at pilots placed at every twelfth subcarrier are first expanded by two-dimensional bilinear interpolation and then refined by a time-distributed convolutional neural network (CNN) coupled with a long short-term memory (LSTM) recurrent stage; the model is trained and evaluated on time-varying 3GPP TR 38.901 tapped delay line channels with Jakes Doppler fading at speeds of up to 120 km/h. Evaluated on a 4×4 MIMO-OFDM link with 624 subcarriers over an SNR range of 0-20 dB, the proposed estimator reduces the normalized mean-squared error by up to 88%, approximately halves the bit-error rate at mid-range SNRs, and raises the spectral efficiency from approximately 0.13 to 4.3-5.0 bits/s/Hz relative to a conventional two-dimensional interpolation baseline, while consuming only half the pilot overhead of a dense-pilot configuration; inference latency on a graphics processing unit is below 1 ms per frame. These results indicate that hybrid CNN-LSTM processing offers a practical route to accurate, low-overhead, and latency-compliant channel estimation for 5G massive MIMO deployments.
Chirag Pradhan· Journal of Intelligent Decis...· 0 citations
The scalability of modern Massive MIMO systems and prospective 6G networks is fundamentally constrained by the “pilot contamination” effect and the prohibitive overhead of time-frequency resources required for orthogonal pilot transmission. In ultra-dense deployment scenarios, traditional pilot-aided channel estimation methods exhibit a critical degradation in spectral efficiency. The study aims to develop a resource-efficient method for blind channel estimation that is invariant to antenna array topology, with the goal of minimizing signaling overhead and maximizing throughput capacity under conditions of complex spatial correlation. The proposed approach is based on the statistical processing of the sample covariance matrix of received signals utilizing a deep convolutional neural network. In contrast to direct reconstruction techniques, the algorithm employs a residual learning strategy to isolate and mitigate estimation noise arising from finite sample sizes. To resolve the phase ambiguity of the signal subspace, a “virtual pilot” concept (a single reference symbol) is introduced, ensuring that resource overhead approaches zero asymptotically. The study is validated across a wide spectrum of configurations, including linear, rectangular, and circular arrays, as well as distributed antenna systems. Simulation results confirm that the proposed method yields a significant gain in the system’s aggregate spectral efficiency by liberating resources previously allocated to pilot sequences. Despite a marginal degradation in estimation accuracy compared to conventional methods, the algorithm demonstrates high robustness to various spatial correlation profiles and antenna geometries. The proposed method facilitates the realization of massive connectivity scenarios on existing base station hardware architectures, effectively overcoming throughput limitations imposed by the coherence interval length.
Cong Quyen Pham, E. Glushankov· Infokommunikacionnye tehnolo...· 0 citations
Accurate channel state information (CSI) is essential for multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems, yet fast time-varying channels pose significant prediction challenges. Traditional approaches fail under high mobility, while deep learning methods rely heavily on large labeled datasets, limiting generalization with scarce training data. Although large language models (LLMs) show promise, their massive parameter count hinders deployment on resource-constrained edge devices. This paper proposes a lightweight, end-to-end CSI prediction framework built upon a general LLM. A time-frequency dual-domain feature extraction module captures subcarrier correlations and temporal dynamics from historical CSI, overcoming single-domain limitations. The end-to-end design maps historical CSI directly to future states, avoiding error propagation inherent in explicit channel estimation. Parameter efficiency is achieved through low-rank adaptation (LoRA) combined with knowledge distillation from a pre-trained LLM, enabling effective few-shot learning at low computational cost. Simulations demonstrate that the proposed scheme delivers robust prediction accuracy across diverse mobility scenar-ios, maintains strong performance under limited training data, and exhibits zero-shot cross-scenario generalization, significantly outperforming both conventional and deep learning baselines in TDD and FDD modes.
Keywords: Channel prediction, multiple-input multiple-output (MIMO), orthogonal frequency division multiplexing (OFDM), large language model (LLM), knowledge distillation, low-rank adaptation (LoRA)
Efficient and reliable transmission of channel state information (CSI) is crucial for maintaining optimal performance in massive multiple-input multiple-output (MIMO) wireless communication systems. However, the feedback overhead remains a significant challenge, particularly in frequency division duplex (FDD) systems with large antenna arrays. In this paper, we propose a novel CSI feedback compression scheme based on sparse residual component selection. The proposed method builds a training-set statistical mean baseline from the training dataset and identifies the K positions whose expected residual magnitude is largest, yielding a fixed importance mask shared between the user equipment (UE) and the base station (BS). Only the residual values at selected positions are encoded and transmitted using a CsiNet-style convolutional autoencoder, while the remaining positions are approximated by the statistical baseline at the BS without any additional feedback. As compression becomes more aggressive, the encoder and decoder operate on a smaller spatial grid at high compression, directly reducing computational cost. Experiments on the COST 2100 dataset show that the proposed method achieves competitive reconstruction quality compared with CsiNet, CsiNet+, and Sobel selective processing, while significantly reducing computational overhead at intermediate and high compression levels.
Prakash Heramil, Hong-Yu Wu, Jie-Lun Zhang et al.· National Aerospace and Elect...· 0 citations
This paper proposes a novel hybrid Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) architecture for accurate channel state estimation in 2-user Non-Orthogonal Multiple Access (NOMA) systems operating under diverse wireless propagation environments. The proposed model integrates convolutional layers for effective spatial feature extraction from received pilot-assisted signals with gated recurrent units to capture temporal dependencies, enabling robust offline training and online joint symbol detection. The continuous channel state information (CSI) is subsequently recovered via a Least Squares (LS) back-calculation step, ensuring accurate continuous channel estimation. Comprehensive evaluations are conducted across four representative channel models—Multipath (frequency-selective fading), Rician (line-of-sight dominant), Nakagami-m (variable fading severity), and millimeter-wave (mmWave, high path loss and blockage-prone)—under varying cyclic prefix (CP) lengths and pilot densities. Simulation results demonstrate that the hybrid CNN-GRU significantly outperforms conventional Least Squares (LS) and Minimum Mean Square Error (MMSE) estimators in terms of Mean Square Error (MSE), Symbol Error Rate (SER), and derived classification metrics (accuracy, precision, recall, and F1 score). Substantial performance gains are achieved, particularly with increased pilot density (up to 64 pilots) and extended CP (20), yielding MSE reductions of 40–84% and near-perfect classification accuracy (>0.99 at SNR=20 dB in optimal configurations). The proposed approach exhibits remarkable robustness across all channels, with the most pronounced improvements in challenging Nakagami-m and mmWave environments, where conventional methods struggle due to severe fading and propagation impairments. These findings highlight the efficacy of the hybrid deep learning framework in enhancing channel estimation reliability for practical NOMA deployments in next-generation wireless systems.
S. Heshmat, Sara Khaled, Ahmed Ezzat et al.· IEEE Access· 0 citations
Efficient Channel State Information (CSI) feedback is indispensable for frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) systems. Existing compressed sensing (CS) algorithms exploit delay-domain sparsity but suffer from prohibitive iterative latency and discrete grid mismatch. Conversely, deep learning (DL) approaches achieve rapid inference but lack spatial scalability and domain adaptability, failing to generalize to unseen propagation environments, and demand computationally heavy encoders and decoder. In this paper, we propose TAP, a Tap-Assisted Parametric CSI Compression. TAP is a one-shot neural framework that unifies the speed of DL with the mathematical interpretability of CS. TAP replaces iterative pursuit with a lightweight 1D neural network that extracts dominant continuous propagation delays from temporal channel sequences via a differentiable sub-grid interpolation operator. TAP achieves true architecture independence, enabling zero-shot generalization across diverse array geometries and unseen propagation environments. Furthermore, TAP yields a completely decoder-free payload, allowing the BS to reconstruct the channel via a simple inverse fast Fourier transform (IFFT). Extensive evaluations across five 3GPP environments demonstrate that TAP achieves a 3.13 to 12.22 dB channel frequency response normalized mean square error (CFR-NMSE) improvement over CsiNet while shrinking the model footprint by 660 times to under 1 MB. Operating with sub-millisecond latencies, TAP accelerates inference by 2700 times over classical iterative OMP, providing a scalable and deployment-ready solution for next-generation networks.
Minwoo Kim, Hyeonsu Lyu, Sehyun Ryu et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
