Conditional generative adversarial networks for channel estimation in RIS-assisted ISAC systems
IEEE Transactions on Communications
Research
My research explores how new technologies can reshape wireless systems, spanning sensing, propagation control, learning, optimization, and emerging communication architectures.
Across intelligent surfaces, pinching-antenna systems, integrated sensing and communication, and quantum-enhanced approaches, I investigate ways to make wireless systems more adaptive and efficient.
Integrated Sensing & Communication
What if wireless could sense the world as it communicates?
Integrated sensing and communication (ISAC) brings sensing and data transmission into the same wireless system, using shared spectrum and hardware. This allows communication infrastructure to also perceive targets, motion, and the surrounding environment.
One system, two functions
One wireless base station carries data to users and receives echoes from a sensing target.
Communication: base station → users.
Sensing: base station → target → base station.
Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
Communication signalSensing signalReturning echo
One wireless base station carries data to users and receives echoes from a sensing target.
Multi-RIS, multi-target ISAC
A full-duplex ISAC base station senses targets directly, while a conventional RIS and a STAR-RIS assist communication. The STAR-RIS reflects uplink signals and transmits downlink signals through to users on its other side. Switch between the complete system, surfaces off for sensing, and surfaces on for communication and sensing.
The 3D scene requires JavaScript and WebGL.Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
Inspect the system
Hover or focus the base station, a surface, a user group, the targets or any channel to identify it. Select a channel to isolate it, or select an element to look closer.
Sensing round tripCascaded communication channelsSelf-interference at the BS
The research challenge
Combining sensing and communication introduces coupled channels, interference, and increasingly complex propagation conditions. Reliable estimation of the underlying channels is therefore essential to realizing these systems in practice.
My contribution
My work develops learning- and model-based approaches for estimating communication and sensing channels in emerging ISAC architectures, including RIS- and pinching-antenna-enabled systems.
Methods
- Deep learning
- Generative learning (CGAN)
- Extreme learning machines
Selected studies
Generative Channel Estimation
This work explores conditional generative learning for channel estimation in RIS-assisted ISAC systems, using adversarial training to improve the reconstruction of sensing and communication channels.
Feature Refinement-based Estimation
This work addresses channel estimation in RIS-ISAC systems through a two-stage learning framework. An extreme learning machine first refines the received features to extract more informative representations of the underlying channels, which are then used by dedicated neural networks to estimate the sensing and communication channels.
Extreme learning machine-based feature refinement for channel estimation in RIS-ISAC systems
IEEE Transactions on Vehicular Technology
Scalable Multi-RIS Multi-Target Estimation
This work combines physics-guided angle estimation with denoising-based channel reconstruction for multi-RIS, multi-target ISAC systems. The sensing model first learns target angles and uses them to reconstruct the channel, while dedicated denoising networks estimate the cascaded communication channels.
Hybrid angle-parametric and denoising-based channel estimation for multi-RIS multi-target ISAC systems
npj Wireless Technology
Reconfigurable Intelligent Surfaces
Can the wireless environment become part of the system design?
Reconfigurable intelligent surfaces (RIS) introduce programmable control over wireless propagation, allowing the environment itself to become an active design variable. This opens new possibilities for optimizing communication, sensing, and network operation.
Single or distributed RIS
A full-duplex base station and user communicate directly and through programmable reflecting surfaces. Each endpoint carries self-interference. Distributed deployment shares the same total element budget across two surfaces, without a link between them.
The 3D scene requires JavaScript and WebGL.Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
RIS-assisted link
The signal is reflected by the programmable surface before reaching the receiver. By adjusting the surface response, the propagation path can be shaped to improve the link.
Direct channelRIS-assisted linksObstructed pathSelf-interference
The research challenge
Making the wireless environment programmable introduces new design choices alongside new uncertainties. RIS systems have to account for imperfect channel knowledge, discrete phase control, interference, and the placement of surfaces - often while optimizing many coupled variables at once.
My contribution
My work develops learning- and optimization-based approaches for RIS-assisted wireless systems, from channel estimation and feature learning to discrete phase optimization and reinforcement learning for full-duplex and distributed RIS deployments.
Methods
- Reinforcement learning
- Alternating optimization
- Generative learning (CGAN)
- Extreme learning machines
Selected studies
Learning to configure RIS
This line of work develops learning-based approaches for configuring RIS-assisted systems, from optimizing continuous phase shifts to handling discrete phase constraints in full-duplex networks. The studies explore deep reinforcement learning as an alternative to conventional optimization for adapting RIS configurations to changing wireless conditions.
Related studies
Deep reinforcement learning for optimizing RIS-assisted HD-FD wireless systems
IEEE Communications Letters
Distributed RIS-assisted FD systems with discrete phase shifts: A reinforcement learning approach
IEEE GLOBECOM
On discrete phase shifts optimization of RIS-aided FD systems: Are all RIS elements needed?
IEEE ICC
Single or distributed?
This work investigates how the deployment of RISs affects full-duplex wireless systems, comparing a single surface with distributed surfaces under different propagation conditions. The goal is to understand when distributing the available RIS elements across multiple locations can provide an advantage over concentrating them in a single surface.
Related studies
Deep reinforcement learning for RIS-assisted FD systems: Single or distributed RIS?
IEEE Communications Letters
Distributed RIS-assisted FD systems with discrete phase shifts: A reinforcement learning approach
IEEE GLOBECOM
Synthesizing the field
These works provide broader perspectives on reinforcement learning for RIS-assisted wireless communications. The book develops a structured treatment of reinforcement learning methods for RIS systems, covering discrete and continuous optimization problems, while the magazine article surveys their application to full-duplex RIS systems, including resource allocation, sum-rate optimization, security, and large-scale deployments.
Related studies
Deep reinforcement learning for RIS-aided full-duplex systems: Advances and challenges
IEEE Communications Magazine
Reinforcement Learning for Reconfigurable Intelligent Surfaces: Assisted Wireless Communication Systems
Springer
Pinching-Antenna Systems
What happens when the antenna itself can move along the waveguide?
Pinching-antenna systems use dielectric waveguides with reconfigurable radiating elements that can be positioned along the waveguide to adapt how signals enter the wireless environment. This added spatial flexibility creates new opportunities for communication and sensing, while introducing new challenges in channel and parameter estimation.
Pinching antennas for ISAC
A dual-function base station feeds two dielectric waveguides mounted above a coverage region. Signals travel inside a guide from its feed point, and each pinching antenna radiates or collects a phase-shifted copy, the shift set by the guided distance from that feed. In the downlink configuration the transmit guide carries N antennas, activated at N of Q predefined candidate positions defined along that guide alone, serving the users and illuminating the target; a single antenna on the receive guide collects the echo, specified by its position rather than drawn from that candidate set. In the uplink configuration the roles are reversed: one antenna illuminates and N activated antennas collect, and there the candidate set applies to both guides. Scatterers add wanted multipath to the communication channel and clutter to the sensing one.
The 3D scene requires JavaScript and WebGL.Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
In-waveguide propagation: The PASS mechanism
The signal travels along the dielectric waveguide, with each activated pinching antenna radiating a phase-shifted copy. This guided phase progression couples the antennas through the shared feed.
Pinching antenna on a dielectric waveguideCommunication · directCommunication · scatteredSensing · targetSensing · clutter
The research challenge
Reconfigurable radiating points introduce a highly structured but difficult propagation environment. The shared waveguide couples the signals across PAs, while changing antenna positions alters the channel geometry, making reliable channel and parameter estimation challenging, especially with limited pilot observations.
My contribution
My work develops learning-based approaches for extracting channel and sensing information from these compressed, position-dependent observations. I investigate position-aware architectures for channel estimation and integrate model-based parameter extraction to recover quantities such as target range and angle.
Methods
- Classical estimation
- Deep learning
- Quantum machine learning
Selected studies
Estimating channels in a flexible architecture2026 · IEEE Transactions on Wireless Communications · Submitted2026 · IEEE GLOBECOM Workshops · Submitted2026 · IEEE Transactions on Vehicular Technology · Submitted
These studies develop learning-based approaches for channel estimation in pinching-antenna systems, from deep learning under compressed observations to joint channel estimation for PASS–ISAC. The work explores how position-aware models can recover communication and sensing channels despite the coupling and limited measurements inherent to the architecture.
Related work
Deep learning-based channel estimation for pinching antenna-enabled ISAC systems
IEEE Transactions on Wireless Communications
Joint channel estimation for pinching antenna-assisted ISAC systems
IEEE GLOBECOM Workshops
Channel estimation for pinching-antenna systems: Quantum machine learning approach
IEEE Transactions on Vehicular Technology
Quantum-Enhanced Communications
What new possibilities could quantum methods bring to wireless systems?
Quantum computing and quantum machine learning introduce new computational approaches for wireless communications. By combining quantum circuits with classical learning, these methods offer new ways to represent, process, and optimize complex wireless problems, including estimation and resource optimization.
Variational quantum circuit
Pilot observations and PA positions enter a hybrid estimator; quantum measurements become classical features used to reconstruct the channel.
Four qubits: angle encoding → trainable Ry and Rz rotations → a ring of controlled-NOT gates → measurement. Classical layers reconstruct the estimated channel.
Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
Inspect the circuit
Pilot observations and PA positions enter a hybrid estimator; quantum measurements become classical features used to reconstruct the channel.
The research challenge
Wireless channel estimation can become especially difficult when only limited pilot observations and training data are available. Learning-based methods can require substantial data to capture complex channel structure, creating a challenge when measurements are costly or scarce.
My contribution
My work investigates hybrid quantum-classical learning for channel estimation under limited data. By exploiting quantum recurrent processing, the work investigates how useful channel representations can be learned from limited observations while maintaining robustness under practical quantum execution conditions.
Methods
- Quantum machine learning
- Variational quantum circuits
Selected studies
Quantum learning for PASS channel estimation2026 · IEEE Transactions on Vehicular Technology · Submitted
This work investigates hybrid quantum-classical learning for PASS channel estimation, combining PA positional information with a quantum LSTM to recover channels from compressed pilot observations. The approach is evaluated under limited training data and practical quantum hardware constraints.
Related work
Channel estimation for pinching-antenna systems: Quantum machine learning approach
IEEE Transactions on Vehicular Technology
Looking across emerging wireless technologies and methods
Alongside my focused research, I have also contributed to broader perspectives on emerging wireless architectures, intelligent surfaces, and learning-based methods for wireless system design.
Ultra-massive MIMO array
An ultra-massive MIMO array built from many sub-arrays. This architecture combines the large array gain of each sub-array with spatial multiplexing across sub-arrays, enabling high-capacity links at terahertz frequencies.
The 3D scene requires JavaScript and WebGL.Help
- Drag
- Orbit the scene
- Click
- Look closer at an item
- Arrow keys
- Rotate the focused scene
Read the array
Hover or focus a sub-array or an antenna element to identify it, or click to look closer.
Active sub-arrayInactive sub-array · spatial modulationCarrier assignment · multi-carrier
Selected studies
This body of work brings together survey and magazine contributions on emerging wireless technologies and methods. It spans terahertz ultra-massive MIMO, intelligent surfaces for 6G, and learning-based optimization in communications and signal processing.
Ultramassive MIMO systems at terahertz bands: Prospects and challenges2020 · IEEE Vehicular Technology Magazine · Published
Ultramassive MIMO systems at terahertz bands: Prospects and challenges
IEEE Vehicular Technology Magazine
Intelligent surfaces for 6G wireless networks: A survey of optimization and performance analysis techniques2020 · IEEE Access · Published
Intelligent surfaces for 6G wireless networks: A survey of optimization and performance analysis techniques
IEEE Access
An overview of machine learning-based techniques for solving optimization problems in communications and signal processing2021 · IEEE Access · Published
An overview of machine learning-based techniques for solving optimization problems in communications and signal processing
IEEE Access