Siyao Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Siyao Li has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Giuseppe Caire
Technische Universität Berlin
2 shared publications - Shuangyang Li
Faculty of Electrical Engineering and Computer Science, Technical University of Berlin, Berlin, Germany
2 shared publications - Defeng Zhou
University of Miami
1 shared publication - Dongwei Ren
College of Intelligence and Computing, Tianjin University, Tianjin, China
1 shared publication - Dongyu Wei
Department of Electrical and Computer Engineering, University of Miami, Coral Gables, FL, USA
1 shared publication - Fabian Göttsch
Faculty of EECS, Technische Universitt Berlin, Berlin, Germany
1 shared publication
Latest Publications
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 30 (10897798) , 627-631
Large Language Models for Beam Prediction Under the mmWave Massive MIMO Hybrid-Field Channels
Millimeter-wave (mmWave) technology is a key enabling technology for next-generation wireless communications. However, beam prediction (BP) in millimeter-wave massive multiple-input multiple-output (MIMO) systems is challenging in high mobility scenarios, especially in non-stationary time-varying hybrid-field channels. In this case, the number of multi paths, time delay, angle of departure (AoD), Doppler offset, and the propagation distance vary simultaneously at different time instants. In this paper, we introduce a BP framework leveraging generative large language models (LLM), specifically GPT-2. The proposed BP-GPT approach utilizes historical channel state information (CSI) to predict optimal beamforming vectors that maximize spectral efficiency in these non-stationary time-varying hybrid-field conditions. Extensive simulations demonstrate the superior performance and robustness of BP-GPT over conventional neural network-based methods.
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 30 (10897798) , 627-631
Large Language Models for Beam Prediction Under the mmWave Massive MIMO Hybrid-Field Channels
Millimeter-wave (mmWave) technology is a key enabling technology for next-generation wireless communications. However, beam prediction (BP) in millimeter-wave massive multiple-input multiple-output (MIMO) systems is challenging in high mobility scenarios, especially in non-stationary time-varying hybrid-field channels. In this case, the number of multi paths, time delay, angle of departure (AoD), Doppler offset, and the propagation distance vary simultaneously at different time instants. In this paper, we introduce a BP framework leveraging generative large language models (LLM), specifically GPT-2. The proposed BP-GPT approach utilizes historical channel state information (CSI) to predict optimal beamforming vectors that maximize spectral efficiency in these non-stationary time-varying hybrid-field conditions. Extensive simulations demonstrate the superior performance and robustness of BP-GPT over conventional neural network-based methods.
- Open Access
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 200431-200445
Robust Communication-Aware Jamming Detection and Avoidance for UAS Swarms
Unmanned autonomous system (UAS) swarms empower groups of robotic agents (such as drones) to collectively accomplish complex missions by navigating intricate environments, but face significant challenges when operating in contested areas with electromagnetic jamming that disrupts interagent communication. This paper introduces an innovative hierarchical control strategy that enables a drone swarm to traverse complex terrains, including regions compromised by electromagnetic jamming intended to disrupt communications. The swarm is initially unaware of the existence or location of these jamming regions, making autonomous detection and avoidance critical for mission success. The proposed strategy synergistically combines communication-aware formation control, particle swarm optimization (PSO) for reactive navigation, and a global path planner for deadlock resolution to ensure the swarm maintains optimal communication quality and mission continuity. A gradient controller maintains the desired swarm formation by maximizing communication quality among agents, while a PSO-based movement controller navigates toward a destination, avoiding newly discovered jamming areas. To address potential deadlocks, a path planning algorithm is invoked to generate a globally optimal route. The framework operates without prior knowledge of jamming locations, learning threat positions through real-time environmental interaction. A comparative analysis of various path planning algorithms, including A*, Greedy Best-First Search, Breadth-First Search, Dijkstra, Jump Point Search, and Theta*, is presented through simulation. The results demonstrate that A* provides optimal performance for our integrated framework, facilitating efficient and robust navigation for the UAS swarms, while maintaining high communication quality among individual drones.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 200431-200445
Open AccessRobust Communication-Aware Jamming Detection and Avoidance for UAS Swarms
Unmanned autonomous system (UAS) swarms empower groups of robotic agents (such as drones) to collectively accomplish complex missions by navigating intricate environments, but face significant challenges when operating in contested areas with electromagnetic jamming that disrupts interagent communication. This paper introduces an innovative hierarchical control strategy that enables a drone swarm to traverse complex terrains, including regions compromised by electromagnetic jamming intended to disrupt communications. The swarm is initially unaware of the existence or location of these jamming regions, making autonomous detection and avoidance critical for mission success. The proposed strategy synergistically combines communication-aware formation control, particle swarm optimization (PSO) for reactive navigation, and a global path planner for deadlock resolution to ensure the swarm maintains optimal communication quality and mission continuity. A gradient controller maintains the desired swarm formation by maximizing communication quality among agents, while a PSO-based movement controller navigates toward a destination, avoiding newly discovered jamming areas. To address potential deadlocks, a path planning algorithm is invoked to generate a globally optimal route. The framework operates without prior knowledge of jamming locations, learning threat positions through real-time environmental interaction. A comparative analysis of various path planning algorithms, including A*, Greedy Best-First Search, Breadth-First Search, Dijkstra, Jump Point Search, and Theta*, is presented through simulation. The results demonstrate that A* provides optimal performance for our integrated framework, facilitating efficient and robust navigation for the UAS swarms, while maintaining high communication quality among individual drones.
- Open Access
- 22 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Vehicular Technology
- Vol. 6, 1872-1880
Secondary Transmission and Channel Estimation for Symbiotic Radio With Orthogonal Time Frequency Space Modulation
Symbiotic radio (SR) holds promise for large-scale Internet of Things (IoT) networks by allowing both primary and secondary systems to share spectrum resources for mutually beneficial transmission. However, in high-mobility scenarios, rapidly changing channels pose significant challenges for reliable and low-latency communications. To enable SR transmission over high-mobility channels, this paper explores the secondary transmission and equivalent channel estimation problems in intelligent reflecting surface (IRS)-SR systems utilizing orthogonal time frequency space (OTFS) modulation. The IRS is divided into multiple groups, each attached with informative frequencies, for secondary transmission. Exploiting the spectrum structure induced by the secondary information, a deep learning (DL)-based secondary information detector is proposed. Moreover, after transforming the channel estimation problem into a compressed sensing (CS) problem, a DL-sparse Bayesian learning (SBL) network is proposed to improve the estimation accuracy and ease the computational afford by avoiding the large amount of iterations in conventional CS algorithms, e.g. the SBL algorithm. Finally, numerical results are provided to illustrate the superiority of the proposed detector and estimator over several benchmarks.
- 22 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Vehicular Technology
- Vol. 6, 1872-1880
Open AccessSecondary Transmission and Channel Estimation for Symbiotic Radio With Orthogonal Time Frequency Space Modulation
Symbiotic radio (SR) holds promise for large-scale Internet of Things (IoT) networks by allowing both primary and secondary systems to share spectrum resources for mutually beneficial transmission. However, in high-mobility scenarios, rapidly changing channels pose significant challenges for reliable and low-latency communications. To enable SR transmission over high-mobility channels, this paper explores the secondary transmission and equivalent channel estimation problems in intelligent reflecting surface (IRS)-SR systems utilizing orthogonal time frequency space (OTFS) modulation. The IRS is divided into multiple groups, each attached with informative frequencies, for secondary transmission. Exploiting the spectrum structure induced by the secondary information, a deep learning (DL)-based secondary information detector is proposed. Moreover, after transforming the channel estimation problem into a compressed sensing (CS) problem, a DL-sparse Bayesian learning (SBL) network is proposed to improve the estimation accuracy and ease the computational afford by avoiding the large amount of iterations in conventional CS algorithms, e.g. the SBL algorithm. Finally, numerical results are provided to illustrate the superiority of the proposed detector and estimator over several benchmarks.
- 14 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Wireless Communications
- Vol. 23 (10) , 14125-14139
Joint Fronthaul Load Balancing and Computation Resource Allocation in Cell-Free User-Centric Massive MIMO Networks
We consider scalable cell-free massive multiple-input multiple-output networks under an open radio access network paradigm comprising user equipments (UEs), radio units (RUs), and decentralized processing units (DUs). UEs are served by dynamically allocated user-centric clusters of RUs. The corresponding cluster processors (implementing the physical layer for each user) are hosted by the DUs as software-defined virtual network functions . Unlike the current literature, mainly focused on the characterization of the user rates under unrestricted fronthaul communication and computation, in this work we explicitly take into account the fronthaul topology, the limited fronthaul communication capacity, and computation constraints at the DUs. In particular, we systematically address the new problem of joint fronthaul load balancing and allocation of the computation resource. As a consequence of our new optimization framework, we present representative numerical results high-lighting the existence of an optimal number of quantization bits in the analog-to-digital conversion at the RUs.
- 14 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Wireless Communications
- Vol. 23 (10) , 14125-14139
Joint Fronthaul Load Balancing and Computation Resource Allocation in Cell-Free User-Centric Massive MIMO Networks
We consider scalable cell-free massive multiple-input multiple-output networks under an open radio access network paradigm comprising user equipments (UEs), radio units (RUs), and decentralized processing units (DUs). UEs are served by dynamically allocated user-centric clusters of RUs. The corresponding cluster processors (implementing the physical layer for each user) are hosted by the DUs as software-defined virtual network functions . Unlike the current literature, mainly focused on the characterization of the user rates under unrestricted fronthaul communication and computation, in this work we explicitly take into account the fronthaul topology, the limited fronthaul communication capacity, and computation constraints at the DUs. In particular, we systematically address the new problem of joint fronthaul load balancing and allocation of the computation resource. As a consequence of our new optimization framework, we present representative numerical results high-lighting the existence of an optimal number of quantization bits in the analog-to-digital conversion at the RUs.
- 6 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Theory
- Vol. 69 (9) , 5752-5770
On the Capacity and State Estimation Error of “Beam-Pointing” Channels: The Binary Case
Motivated by beamformed millimeter-wave (mmWave) communication, we consider the optimal tradeoff between reliable communication rate and state estimation error for a new state-dependent channel model with in-block memory referred to as the binary beam-pointing (BBP) channel. The multiantenna base station (BS) uses a finite beamforming codebook (i.e., discrete beam directions) and associates the target user to the most favorable beam, i.e., the beam directed along the strongest propagation path from BS to the user in 5G and IEEE 802.11ad mmWave communication systems. We model the target user’s Angle-of-Departure (AoD) as the state of a state-dependent channel. Since the AoD remains almost constant over several consecutive time slots, the channel has memory. In addition, we assume the BS receives implicit causal feedback (e.g., modeling a backscatter signal as in radar) and consider a joint communication and sensing (JCAS) problem where the BS is also interested in explicitly estimate the channel state (quantized AoD). We derive a closed-form solution to the capacity of the BBP channel model under the peak input cost constraint. Under the average constraint, we present an implicit capacity result, where the capacity is given as the solution of an optimization problem, and we provide an algorithm for its numerical computation. Finally, for the JCAS problem at hand, we provide the minimum distortion under the peak input constraint and show that this coincides with that obtained by the capacity-achieving strategy. This completely characterizes the capacity-distortion tradeoff for the BBP channel under peak input constraint. For the average constraint case, numerical results show that our capacity-achieving strategy yields a state estimation error very close to the theoretical minimum, showing the near-optimality of the proposed strategy for the JCAS problem under the average cost constraint.
- 6 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Theory
- Vol. 69 (9) , 5752-5770
On the Capacity and State Estimation Error of “Beam-Pointing” Channels: The Binary Case
Motivated by beamformed millimeter-wave (mmWave) communication, we consider the optimal tradeoff between reliable communication rate and state estimation error for a new state-dependent channel model with in-block memory referred to as the binary beam-pointing (BBP) channel. The multiantenna base station (BS) uses a finite beamforming codebook (i.e., discrete beam directions) and associates the target user to the most favorable beam, i.e., the beam directed along the strongest propagation path from BS to the user in 5G and IEEE 802.11ad mmWave communication systems. We model the target user’s Angle-of-Departure (AoD) as the state of a state-dependent channel. Since the AoD remains almost constant over several consecutive time slots, the channel has memory. In addition, we assume the BS receives implicit causal feedback (e.g., modeling a backscatter signal as in radar) and consider a joint communication and sensing (JCAS) problem where the BS is also interested in explicitly estimate the channel state (quantized AoD). We derive a closed-form solution to the capacity of the BBP channel model under the peak input cost constraint. Under the average constraint, we present an implicit capacity result, where the capacity is given as the solution of an optimization problem, and we provide an algorithm for its numerical computation. Finally, for the JCAS problem at hand, we provide the minimum distortion under the peak input constraint and show that this coincides with that obtained by the capacity-achieving strategy. This completely characterizes the capacity-distortion tradeoff for the BBP channel under peak input constraint. For the average constraint case, numerical results show that our capacity-achieving strategy yields a state estimation error very close to the theoretical minimum, showing the near-optimality of the proposed strategy for the JCAS problem under the average cost constraint.
- Open Access
- 1 August 2020
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 249 (2) , 25
ECoPANN: A Framework for Estimating Cosmological Parameters Using Artificial Neural Networks
In this work, we present a new method to estimate cosmological parameters accurately based on the artificial neural network (ANN), and a code called ECoPANN (Estimating Cosmological Parameters with ANN) is developed to achieve parameter inference. We test the ANN method by estimating the basic parameters of the concordance cosmological model using the simulated temperature power spectrum of the cosmic microwave background (CMB). The results show that the ANN performs excellently on best-fit values and errors of parameters, as well as correlations between parameters when compared with that of the Markov Chain Monte Carlo (MCMC) method. Besides, for a well-trained ANN model, it is capable of estimating parameters for multiple experiments that have different precisions, which can greatly reduce the consumption of time and computing resources for parameter inference. Furthermore, we extend the ANN to a multibranch network to achieve a joint constraint on parameters. We test the multibranch network using the simulated temperature and polarization power spectra of the CMB, Type Ia supernovae, and baryon acoustic oscillations and almost obtain the same results as the MCMC method. Therefore, we propose that the ANN can provide an alternative way to accurately and quickly estimate cosmological parameters, and ECoPANN can be applied to the research of cosmology and even other broader scientific fields.
- 1 August 2020
- Published by American Astronomical Society in The Astrophysical Journal Supplement Series
- Vol. 249 (2) , 25
Open AccessECoPANN: A Framework for Estimating Cosmological Parameters Using Artificial Neural Networks
In this work, we present a new method to estimate cosmological parameters accurately based on the artificial neural network (ANN), and a code called ECoPANN (Estimating Cosmological Parameters with ANN) is developed to achieve parameter inference. We test the ANN method by estimating the basic parameters of the concordance cosmological model using the simulated temperature power spectrum of the cosmic microwave background (CMB). The results show that the ANN performs excellently on best-fit values and errors of parameters, as well as correlations between parameters when compared with that of the Markov Chain Monte Carlo (MCMC) method. Besides, for a well-trained ANN model, it is capable of estimating parameters for multiple experiments that have different precisions, which can greatly reduce the consumption of time and computing resources for parameter inference. Furthermore, we extend the ANN to a multibranch network to achieve a joint constraint on parameters. We test the multibranch network using the simulated temperature and polarization power spectra of the CMB, Type Ia supernovae, and baryon acoustic oscillations and almost obtain the same results as the MCMC method. Therefore, we propose that the ANN can provide an alternative way to accurately and quickly estimate cosmological parameters, and ECoPANN can be applied to the research of cosmology and even other broader scientific fields.
- 1 September 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2756-2760
Dual Recursive Network for Fast Image Deraining
Recent years have witnessed the great progress on deep image deraining networks. On the one hand, deraining performance has been significantly improved by designing complex network architectures, yielding high computational cost. On the other hand, several lightweight networks try to improve computational efficiency, but at the cost of notable degrading deraining performance. In this paper, we propose a dual recursive network (DRN) for fast image deraining as well as comparable or superior deraining performance compared with state-of-the-art approaches. Specifically, our DRN utilizes a residual network (ResNet) with only 2 residual blocks (ResBlock), which is recursively unfolded to remove rain streaks in multiple stages. Meanwhile, the 2 ResBlocks can be recursively computed in one stage, forming the dual recursive network. Experimental results show that DRN is very computationally efficient and can achieve favorable deraining results on both synthetic and real rainy images. The source codes and pre-trained models are available at https://github.com/csdwren/DRN.
- 1 September 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2756-2760
Dual Recursive Network for Fast Image Deraining
Recent years have witnessed the great progress on deep image deraining networks. On the one hand, deraining performance has been significantly improved by designing complex network architectures, yielding high computational cost. On the other hand, several lightweight networks try to improve computational efficiency, but at the cost of notable degrading deraining performance. In this paper, we propose a dual recursive network (DRN) for fast image deraining as well as comparable or superior deraining performance compared with state-of-the-art approaches. Specifically, our DRN utilizes a residual network (ResNet) with only 2 residual blocks (ResBlock), which is recursively unfolded to remove rain streaks in multiple stages. Meanwhile, the 2 ResBlocks can be recursively computed in one stage, forming the dual recursive network. Experimental results show that DRN is very computationally efficient and can achieve favorable deraining results on both synthetic and real rainy images. The source codes and pre-trained models are available at https://github.com/csdwren/DRN.