A reconfigurable intelligent surface (RIS) is a promising technology for beyond-fifth-generation (B5G) and sixth-generation (6G) wireless communications, but its passive reflection and two-hop double-fading structure make cascaded channel estimation challenging. Conventional convolutional neural network (CNN) estimators require many trainable parameters, while a single shallow parameterized quantum circuit (PQC) may have limited feature representation. Deep quantum circuits can also suffer from noise and barren-plateau effects on noisy intermediate-scale quantum (NISQ) devices. To address these issues, this paper proposes a multi-filter quantum convolutional neural network (MF-QCNN) for cascaded channel estimation in RIS-assisted multi-user uplink systems. The proposed model uses multiple independent shallow PQC filters in parallel, concatenates their measured features, and estimates the cascaded channel through a compact classical dense head, with the total trainable-parameter count scaling as 182F+696 for F parallel filters. Simulation results, compared with a single-filter quantum convolutional neural network (QCNN), CNN, and multilayer perceptron (MLP) baselines, show that at a signal-to-noise ratio (SNR) of 20 dB, the 3-filter MF-QCNN reduces the normalized mean squared error (NMSE) by approximately 22.9, 8.1, and 4.6 dB relative to the single-filter QCNN, CNN, and MLP baselines, respectively, while using only about 19.3% of the CNN trainable parameters. Under zero-forcing (ZF) precoding, it achieves the highest achievable sum rate among the learning-based estimators; at SNR = 30 dB, it improves the achievable sum rate by approximately 17.4% and 12.8% over the CNN and MLP baselines, respectively. These simulation results suggest that the parallel shallow-PQC design can serve as a compact quantum-aided estimator for RIS channel estimation and may provide a useful basis for future studies on AI-native transceiver design in B5G/6G networks.
A reconfigurable intelligent surface (RIS) is a promising technology for beyond-fifth-generation (B5G) and sixth-generation (6G) wireless communications, but its passive reflection and two-hop double-fading structure make cascaded channel estimation challenging. Conventional convolutional neural network (CNN) estimators require many trainable parameters, while a single shallow parameterized quantum circuit (PQC) may have limited feature representation. Deep quantum circuits can also suffer from noise and barren-plateau effects on noisy intermediate-scale quantum (NISQ) devices. To address these issues, this paper proposes a multi-filter quantum convolutional neural network (MF-QCNN) for cascaded channel estimation in RIS-assisted multi-user uplink systems. The proposed model uses multiple independent shallow PQC filters in parallel, concatenates their measured features, and estimates the cascaded channel through a compact classical dense head, with the total trainable-parameter count scaling as 182F+696 for F parallel filters. Simulation results, compared with a single-filter quantum convolutional neural network (QCNN), CNN, and multilayer perceptron (MLP) baselines, show that at a signal-to-noise ratio (SNR) of 20 dB, the 3-filter MF-QCNN reduces the normalized mean squared error (NMSE) by approximately 22.9, 8.1, and 4.6 dB relative to the single-filter QCNN, CNN, and MLP baselines, respectively, while using only about 19.3% of the CNN trainable parameters. Under zero-forcing (ZF) precoding, it achieves the highest achievable sum rate among the learning-based estimators; at SNR = 30 dB, it improves the achievable sum rate by approximately 17.4% and 12.8% over the CNN and MLP baselines, respectively. These simulation results suggest that the parallel shallow-PQC design can serve as a compact quantum-aided estimator for RIS channel estimation and may provide a useful basis for future studies on AI-native transceiver design in B5G/6G networks.
Sixth-generation (6G) wireless networks are expected to deliver transformative capabilities including ultra-reliable low-latency communications and massive connectivity. Reconfigurable intelligent surfaces (RISs) have emerged as a key enabling technology by dynamically manipulating wireless propagation through programmable reflecting elements. However, accurate channel estimation in RIS-assisted networks faces significant challenges from cascaded channel structures, passive elements, and high-dimensional parameter spaces. Quantum machine learning (QML) offers a fundamentally different computational paradigm that may address these challenges through quantum properties. This article provides a tutorial overview of QML for channel estimation in RIS-assisted 6G networks. We present RIS-assisted system architectures and channel estimation challenges, introduce quantum computing fundamentals, and survey QML architectures. As a use case, we consider a hybrid quantum-classical framework, e.g., convolutional neural network (CNN)-quantum long short-term memory (QLSTM), for channel estimation in RIS-assisted non-orthogonal multiple access (NOMA) systems. This case study shows consistent accuracy improvements compared to quantum and classical benchmarks. Finally, we discuss implementation challenges and future research directions for practical deployment of quantum-enhanced wireless systems.
Accurate channel estimation is a fundamental requirement for high-performance wireless communication systems, where conventional deep learning approaches often suffer from excessive trainable parameters and high memory cost as antenna and subcarrier dimensions scale up. This paper proposes a quantum machine learning (QML)-based channel estimation framework that addresses this scalability challenge through a compact hybrid quantum-classical architecture. An 8-qubit quantum neural network (QNN) based on a parameterized quantum circuit (PQC) encodes received pilot-domain features using both Cartesian and polar components through Hadamard initialization and rotation-based quantum encoding, and learns channel representations using trainable variational rotations and controlled-Z (CZ) entanglement. The Pauli-Z measurement outputs are then processed by a lightweight classical dense head to produce the final channel estimate. The proposed QNN is compared with a classical deep neural network (DNN) and a QNN without hidden dense layers over a 2 × 1 Rayleigh fading multiple-input single-output (MISO) channel across an SNR range of 10 dB to 30 dB. Simulation results show that the proposed hybrid QNN achieves the lowest normalized mean square error (NMSE) across the entire evaluated SNR range while using only 1,012 trainable parameters, compared with 3,356 for the DNN. At 30 dB SNR, the proposed QNN achieves the lowest NMSE of approximately − 20.24 dB, compared with approximately − 17.54 dB for the DNN baseline. An ablation study further confirms that the dense post-processing head is essential for accurate regression from raw quantum measurement outputs. These results suggest that compact hybrid quantum-classical estimators offer a promising direction for accurate and parameter-efficient channel estimation in next-generation wireless communication systems.
Quantum Convolutional Neural Networks (QCNN) migrate the convolution and pooling concepts of classical Convolutional Neural Networks (CNN) to variational quantum circuits, showing potential on Noisy Intermediate-Scale Quantum (NISQ) devices. However, traditional QCNN adopts a single-depth circuit structure, which has limitations in multi-scale feature extraction, and is prone to the barren plateau phenomenon during training, making it difficult to adapt to high-dimensional real-world scenarios such as network intrusion detection. This paper proposes a QCNN architecture based on a multi-path parallel structure. For the network intrusion detection task, it draws on the Inception module to construct three parallel quantum convolution paths: shallow, medium, and deep, achieving hierarchical extraction of multi-scale features; residual connections are introduced to preserve encoded state information and provide gradient shortcuts; a hybrid optimization strategy combining classical Adam and Quantum Natural Gradient (QNG) is adopted to adapt to the heterogeneous parameter space. Experimental validation shows that, on the UNSW-NB15 network intrusion detection dataset, the proposed Multi-Path Parallel Quantum Convolutional Neural Network (MP-QCNN) improves detection accuracy by 3% compared to traditional single-path QCNN and effectively mitigates the barren plateau phenomenon. This design strictly maintains NISQ depth constraints and quantum parallelism advantages, providing an efficient and stable feature extraction framework for high-dimensional tasks such as network intrusion detection.
This paper investigates advanced signal processing techniques to address the critical challenge of Multi-User Interference (MUI) within the demanding framework of 6th Generation (6G) wireless systems. The study presents a direct comparative analysis between classical deep learning and emerging quantum machine learning paradigms. We first establish the capabilities of Convolutional Neural Networks (CNNs), a state-of-the-art deep learning model, in handling the complex, non-ideal channel environments of modern wireless communications. Subsequently, we explore the theoretical advantages and practical performance of Quantum Convolutional Neural Networks (QCNNs), positing them as a next-generation solution for processing unprecedented data complexity. Through rigorous simulation in a multi-user SIMO uplink scenario, this paper demonstrates that QCNN consistently outperforms its classical CNN counterpart in both signal detection accuracy and overall system sum rate, particularly in low signal-to-noise ratio (SNR) regimes, all while utilizing significantly fewer model parameters. The core contribution of this work is to provide direct empirical evidence of the potential quantum advantage offered by QCNNs for the complex signal processing tasks inherent to future 6G systems.
Channel estimation is one of the most fundamental and performance-limiting operations in fifth-generation (5G) and emerging sixth-generation (6G) wireless networks, where ultra-massive MIMO arrays, millimetre-wave and terahertz carrier frequencies, and non-Gaussian hardware-induced noise combine to render classical least-squares and minimum mean squared error estimators inadequate. Deep learning has narrowed the performance gap, yet it offers no information-theoretic guarantee on how close any particular network architecture is to the fundamental noise floor. This paper introduces the AI-Quantum Channel Modeling and Estimation (AQCME) framework, which integrates variational quantum circuits, quantum signal processing, and a quantum eigenspace denoising operator into a unified trainable architecture for simultaneous channel estimation and noise suppression in MIMO-OFDM systems. The defining theoretical result is a quantum Cramer-Rao lower bound on normalised mean squared error, grounded in the quantum Fisher information matrix of the encoded channel state, together with a rigorous proof that the AQCME estimator approaches this bound as training data grows – to the best of our knowledge the first constructive achievability result of its kind in the wireless channel estimation literature. The framework is supported by formal analysis of quantum channel state preparation, quantum kernel universality with an explicit barren-plateau trainability discussion, ergodic spectral efficiency, Nakagami-m outage probability, and convergence of the hybrid quantum-classical gradient descent. Monte Carlo simulations over 50,000 channel realisations, reported with 90 percent confidence intervals throughout, validate every theoretical prediction. Ablation studies across circuit depths and qubit counts, a hardware-noise sweep over gate error rates from 1e-4 to 1e-2, and hyperparameter sensitivity analysis confirm robustness. At 20 dB signal-to-noise ratio, AQCME reduces normalised mean squared error by 4.8 dB over MMSE and by 2.1 dB over the best deep learning baseline; under emulated near-term noisy quantum hardware conditions it still outperforms MMSE by 2.6 dB. All quantum results in this work are obtained by exact classical statevector emulation of the variational circuit; no physical quantum processor was used, and reported hardware-noise behaviour is modelled by a depolarising channel. Spectral efficiency improves by up to 22.4 percent in massive MIMO scenarios, the URLLC outage target of 1e-5 is met at 6.2 dB lower signal-to-noise ratio than with MMSE, and inference complexity scales logarithmically with antenna count rather than quartically as with MMSE matrix inversion.
Abstract Quantum machine learning is one of the most promising applications of quantum computing in the noisy intermediate-scale quantum (NISQ) era. We propose a quantum convolutional neural network(QCNN) inspired by convolutional neural networks (CNN), which greatly reduces the computing complexity compared with its classical counterparts, with O (( log 2 M ) 6 ) basic gates and O ( m 2 + e ) variational parameters, where M is the input data size, m is the filter mask size, and e is the number of parameters in a Hamiltonian. Our model is robust to certain noise for image recognition tasks and the parameters are independent on the input sizes, making it friendly to near-term quantum devices. We demonstrate QCNN with two explicit examples. First, QCNN is applied to image processing, and numerical simulation of three types of spatial filtering, image smoothing, sharpening, and edge detection is performed. Secondly, we demonstrate QCNN in recognizing image, namely, the recognition of handwritten numbers. Compared with previous work, this machine learning model can provide implementable quantum circuits that accurately corresponds to a specific classical convolutional kernel. It provides an efficient avenue to transform CNN to QCNN directly and opens up the prospect of exploiting quantum power to process information in the era of big data.