Research

MINT Research Overview

i-Dris: Intelligent Rapid Deployment for Reconfigurable Intelligent Surface in Millimeter Wave Communication System (Future Tech Award 2022)

i-Dris integrates 5G mmWave base stationreconfigurable intelligent surface (RIS) for improving signal coverage/throughput. We design multi-time scaled temporal reinforcement learning for RIS deployment equipped on auto-guided vehicle, which determines future actions from both historical and current rates observed from the edge server. Under blockages with non-line-of-sight, i-Dris can achieve 800~900 Mbps rate (1.2~3.6x more compared to non-RIS). i-Dris only requires 10-min of deployment time (1 hr for other methods). Compared to existing prototypes in academia/industry, i-Dris does not need fiber, huge human resources, re-measurement offline training. It learns from historical data guaranteering robustness to changing environments.

遠傳與網通大廠光寶科技、國際材料大廠與頂尖學府國立陽明交通大學共同研發的「B5G毫米波整合智慧反射板」同樣在MWC首次登場，為全台首發的5G毫米波室內佈建解決方案，可以強化被環境遮蔽區域的5G涵蓋範圍、包含大樓林立的深層室內，並可以根據環境需求，客製化調整無線電波角度，精準解決真實情境中複雜的盲區覆蓋問題。

6G Research Challenges and Opportunities

Whilst the fifth-generation (5G) systems are being rolled out across the globe, researchers have turned their attention to the exploration of radical next-generation solutions. At this early evolutionary stage we survey five main research facets of this field, namely Facet 1: next-generation architectures, spectrum and services, Facet 2: next-generation networking, Facet 3: Internet of Things (IoT), Facet 4: wireless positioning and sensing, as well as Facet 5: applications of deep learning in 6G networks. In this work, we have provided a critical appraisal of the literature of promising techniques ranging from the associated architectures, networking, applications as well as designs. We have portrayed a plethora of heterogeneous architectures relying on cooperative hybrid networks supported by diverse access and transmission mechanisms. The vulnerabilities of these techniques are also addressed and carefully considered for highlighting the most of promising future research directions. Additionally, we have listed a rich suite of learning-driven optimization techniques. We conclude by observing the evolutionary paradigm-shift that has taken place from pure single-component bandwidth-efficiency, power-efficiency or delay-optimization towards multi-component designs, as exemplified by the twin-component ultra-reliable low-latency mode of the 5G system. We advocate a further evolutionary step towards multi-component Pareto optimization, which requires the exploration of the entire Pareto front of all optiomal solutions, where none of the components of the objective function may be improved without degrading at least one of the other components.


Resource Management and Deployment for Reconfigurable Intelligent Surface (RIS)

Substantially increasing wireless traffic and extending serving coverage is required with the advent of sixthgeneration (6G) wireless communication networks. Reconfigurable intelligent surface (RIS) is widely considered as a promising technique which is capable of improving the system sum rate and energy efficiency. Moreover, full-duplex (FD) multiinput-multi-output (MIMO) transmission provides simultaneous transmit and received signals, which theoretically provides twice of spectrum efficiency. However, the self-interference (SI) in FD system is a challenging task requiring high-overhead cancellation, which can be resolved by configuring appropriate phase shifts of RIS. This work has proposed an RIS-empowered full-duplex interference cancellation (RFIC) scheme in order to alleviate the severe interference in an RIS-FD system. We consider the interference minimization of RIS-FD MIMO while guaranteeing quality-of-service (QoS) of whole system. The closed-form solution of RIS phase shifts is theoretically derived with the discussion of different numbers of RIS elements and receiving antennas. Simulation results reveal that the proposed RFIC scheme outperforms existing benchmarks with more than 50% of performance gain of sum rate.


An impediment of the conventional RIS solutions is that the transmitter and the user have to be within the same 180-degree half-plane, rather than roaming across the entire 360-degree full plane. By contrast, the STAR-RIS architecture, or termed as intelligent omni-surface (IOS), allows full-plane coverage by potentially harnessing full reflection, full transmission, as well as simultaneous transmission and reflection. These modes were discussedwith special emphasis on NTT DOCOMO's prototype. There are three different principles governing their operations, namely the so-called energy-splitting, partitioning and time-switching types, which have their different pros and cons. In the energy-splitting mode the signal impinging upon an element is partially reflected and transmitted. By contrast, the partitioning type may be viewed as having a reflection-only and transmission-only segment of reduced sizes. Finally, the time-switching type is capable of switching the reflective elements between the transmit and reflect modes. There is a huge variety of compelling applications scenarios, such as STAR-RIS-NOMA, STAR-RIS-CoMP and multi-STAR-RIS deployment as well as AI-assisted STAR-RIS, which require further exploration by the research community.


Advanced Beam Seaerch and Handover Scheme for LEO Networks

Conventional Handover Scheme for LEO

Advanced Beam Search and Handover Scheme for LEO

Device-free Wireless Sensing for Human Presence Detection

Human presence detection is a promising technology for detecting a human body’s presence in an area of interest. With the combination of presence detection and smart home concept, we can adopt an intelligence appliance to execute numerous interesting events in the room. Channel state information (CSI) extracted from commercial Wi-Fi router can offer high precision environment detection based on its abundant signal characteristics, which can be applied for wireless sensing, including presence, fall, and respiration rate detections. We propose a recurrent-based environmental adaptive device-free presence detection (REAP) for Wi-Fi CSI-based networks. It extracts meaningful information from CSI, including motion and static features, and applies a recurrent-based machine learning method to identify whether a human is present or absent in a certain room. Moreover, we design an adaptive structure of named REAP-T by using the micro and macro transfer learning, increase the model robustness, and reduce the data and time consumed for training additional new environments. Finally, a robust real-time system and demonstration are presented to show the performance of proposed REAP and REAP-T approaches. The proposed system can be real-time implemented on commercial Wi-Fi routers with high accuracy of human presence detection.


Scenarios for Human Presence Detection

Multiple People Presence Detection

Device-free Wireless Presence Detection for Different Spots

Device-free Wireless Presence Detection for Different Users

Wireless Sensing for User Leaving

Wireless Sensing for Intruder Detection

Applications of Deep Learning in 6G Networks

Given the rapid development of AI-empowered deep learning, both supervised learning as well as unsupervised learning and reinforcement learning have found favour in solving challenging communications and networking problems. Specific examples are constituted by radio interference management, resource allocation, multiple parameter optimization, network traffic prediction, computing resource assignment, and flexible configuration of network functions. In supervised learning, ground truth labels and fixed-size inputs constitute a deep layered neural network (NN). However, labeling is not required in unsupervised learning, which exploits the correlation between samples of the dataset. In reinforcement learning, an agent will interact with the environment and then updates the model based on the corresponding rewards. Note that deep learning can deal with comparably complex problems in a non-linear and non-convex manner than that utilizing machine learning. Therefore, we can efficiently manage both vertical and horizontal networks with the aid of deep neural networks.


Deep Learning-based Beamforming for mmWave Networks

Millimeter wave (mmWave) technology achieving multi-gigabits speed plays a significant role in beyond 5G and the next 6G wireless communication networks thanks to its huge spectrum utilization and beam-based directional transmissions. To tackle temporary ultra-high data demands of hotspot areas, three-dimensional (3D) heterogeneous network (HetNet) is designed with the integration of mmWave unmanned aerial vehicles (UAV) to provide resilient instantaneous control and data transmissions. However, some critical beam-related issues for mmWave implementation of UAVs/drones are addressed including robust initial beam alignment, mobility-aware beam tracking and beam refinement. In this research, we aim at developing robust and efficient beam control mechanisms by implementing a prototype of 3D flying heterogeneous communications. The backhaul connections operate at mmWave frequency between airship and UAV/drone, while fronthaul links adopt lower frequency bands such as Wi-Fi for multiuser data transmissions. We evaluate system performances for our proposed beam control schemes and provide a real-time prototype of 3D on-demand flying mobile communication for mmWave HetNets.


NS-3 Implementation of Fast Beam Training (FBT) Algorithm for mmWave Networks

High-Definition Video Streaming using Drone-based mmWave Communications

Multiuser Transmission for mmWave Networks

Fast Beam Training for mmWave UAV Communications

Resource Allocation for Multi-Numerology and Full-duplex Wireless Networks 

We study joint interference cancellation and resource allocation for full-duplex (FD) cloud radio access networks (C-RANs) with antenna correlation. The target is to maximize the capacity of downlink (DL) user equipments (UEs) with guaranteed quality of service (QoS) of uplink (UL) UEs under the constraints of DL and UL transmit power. Intractability of the considered problem involves non-convexity and coupling between postcoding and remote radio head (RRH) selection. To deal with the high coupling between system variables, an estimation-free self-interference (SI) cancellation (EFSC) scheme with merit of significantly reducing signaling overhead for channel estimation is proposed. The reduced signaling overhead is also derived in closed-form expression. Furthermore, we propose a generalized Bender's decomposition-based resource allocation (GRA) algorithm, which separates the continuous and discrete variables to solve the optimization problem. With the design of a flexible utility function, the tradeoff between DL capacity and co-channel interference (CCI) can be achieved. Moreover, we identify the scenario in which antenna correlation will be beneficial for both the UL and DL communications in FD C-RAN with the implementation of EFSC scheme. The effectiveness of the proposed methods and theorems are verified via simulation results.


IEEE VTC Fall 2020 Presentation

Skeleton-based Location Estimation and Tracking for Wireless Networks

Map information can assist indoor localization to avoid improbable cases and achieve accurate location estimation. We proposed an automatic method to extract useful information from indoor map as spatial skeleton database (SSD). Based on conventional probabilistic fingerprinting technique and particle filter tracking algorithm, we also proposed spatial skeleton-based dynamic probabilistic fingerprinting database (S-DFD) to filter out reference points (RPs) in fingerprinting database according to the previous target location and the walking distance between RPs. Finally, we proposed a spatial skeleton-based particle filter tracking (S-PT) which use SSD to construct realistic transition model. According to the experiment result, the whole system consists of SSD, S-DFD and S-PT called spatial skeleton-enhanced location tracking for indoor localization (SELT) can achieve accurate location estimation.


Skeleton-based Location Tracking

You can observe both the Spatial Skeleton and Geospatial Skeleton from the following two demo sites by inserting a map image.

Device-free Human Breath Rate Detection

We are developing device-free breath detection system using commercial Wi-Fi routers. Our target is to detect user's breath rate with hassle-free and user-centric focuses for potential Apnea in healthcare applications. Sleep apnea syndrome is a common sleep disorder that can lead to a variety of diseases. The traditional diagnostic method, polysomnography (PSG), is time-consuming, expensive, and inconvenient for patients. In this research, we proposed the transfer learning based apnea detection (TLAD) system as a non-contact based method utilizing the channel state information (CSI) from commercial Wi-Fi devices. In order to reduce the overhead of collecting CSI data and improving efficiency during training process, the transfer learning technique is applied to establish pre-trained model by utilizing open source contact-based thoracic movement data. Moreover, existing research works detect apnea based on breathing pauses and shallow breathing periods, which are not effective to identify complex apnea characteristics. This potential drawback is overcome in proposed TLAD system since both CSI amplitude and frequency features are extracted for apnea classification. Our experimental results showed that the TLAD system achieves an F1-score of 90.1, which is superior to other existing methods.


Device-free Respiration Rate Detection

Indoor Device-free Respiration Rate Detection

Advanced Angle of Arrival (AoA) Estimation Techniques

In recent years, indoor positioning system employing Bluetooth has attracted tremendous attention. However, it is investigated that its received channel information is significantly affected by hardware configuration and wireless environments, especially the received dataset of angle of arrival (AOA), which leads to inaccurate channel estimation and positioning. Furthermore, AOA estimation at larger angle direction is severely influenced by the variant wireless environment and signal distortion, which has not been resolve in existing literature. We propose an advanced I/Q density-based AOA estimation (IQDAE) to deal with the above-mentioned problem, which consists of two sub-schemes. We firstly employ the designed phase difference (PD) filter to convert I/Q signals to phase information and then select the candidate sets by eliminating outliers. Afterward, we conceive a PD density-based classification algorithm to estimate AOA. The experimental results show that the mean absolute error of proposed IQDAE algorithm is comparably smaller than that from the other schemes, including commercial solutions, especially at larger angles. The results indicate that we can effectively increase the service range for the Bluetooth positioning system by adopting the proposed algorithm.


Federated Edge Learning for Next Generation Hybrid mmWave/THz Wireless Communications

In the next sixth generation (6G) wireless communications, there emerge exponentially high data demands for diverse services and applications. Therefore, terahertz (THz) is considered as a promising technology using wider hundreds-of-GHz bandwidth compared to millimeter wave (mmWave) transmissions, which is capable of supporting Tbps high speed data traffic. However, THz transmission possesses more severe path loss due to higher operating frequencies at 0.1-10 THz. Accordingly, new hybrid beamforming techniques should be enhanced by array-of-subarray (AoSA) equipped with tens-of-thousands of antenna elements. Due to short and medium distance-based wireless transmissions, a large number of THz-enabled base stations (BSs) should be deployed to support seamless coverage and high throughput performance. When using ultra-thin THz beams, beam training becomes compellingly imperative due to enormous number of beam training steps, which potentially induces high training latency overhead. In addition to beams, there are power, time, and subbands resources providing abundant but complex wireless environments, which the optimal solutions cannot be readily obtained by using conventional optimization methods. Empowered by state-of-the-art machine learning and deep learning in artificial intelligence (AI) fields, the system is capable of supporting resilient, automatic and intelligent THz/mmWave wireless networks. In a multi-BS and multi-network scenarios, federated edge learning (FEL) can perform resource assignment collaboratively among local BSs and devices. Aided by FEL-empowered central server, hybrid TH/mmWave beamforming and AI-enhanced radio resource management (RRM) have the capability to provide flawless services by guaranteeing globally-optimal data rate, latency, utilization, fairness, security, and energy/spectrum efficiencies.

IEEE VTC Fall 2022 Presentation

Reinforcement Learning for Channel Allocations in WiFi Networks

IEEE 802.11ax system has been adopted to provide enhanced throughput performance for next-generation wireless local area networks. Its orthogonal frequency division multiple access (OFDMA) allows massive users to concurrently utilize different subbands for data transmission from their corresponding access points (APs). However, severe adjacent channel interference (ACI) incurs overlapping channels under the scenarios of dense users with multiple APs, which should be properly alleviated to provide adequate system throughput. In this work, we propose a long-/short-term reinforcement learning channel allocation (LSRCA) scheme to effectively mitigate ACI for multi- AP scenarios in IEEE 802.11ax systems. With the considerations of signal features from both long and short time durations, the LSRCA algorithm can maximize effective sum rate through online adaptation and learning via the updates of two Q-tables for weighting adjustments and action execution. Experimental results in realistic fields have demonstrated the effectiveness of LSRCA scheme by providing higher system throughput compared to existing benchmark methods.


Research Tools (only accessible by lab members)