Hybrid BEAMfoRming and deep leArnIng techniqueS for mm-wave wirEless networks
The main goal of the BEAM-RAISE project is the design, the implementation and the assessment of a full level experimentation tool in millimeter wave (mm-wave) frequencies, for hybrid beamforming and nonconventional Multiple Input Multiple Output (MIMO), supported by deep learning and data analytics. This is achieved by combining results and know-how from various research fields. The implemented engine aims to extend and improve the capabilities of an indoor mm-wave mobile Software Defined Radio (SDR) testbed for 5G and beyond systems.
Implementation of hybrid beamforming techniques over SDR based environment, and development and evaluation of MIMO-like spatial focusing.
Design and implementation of deep learning techniques for channel estimation and beamforming weights selection over the experimentation engine. The task of the beamformer engine will be carried out on the Base Station (BS) host at the mobile edge. Channel estimation will be executed on mobile user equipment (UE). The performance and the feasibility of the suggested techniques will be demonstrated by executing evaluation tests for different set of parameters and mobile scenarios.
Hardware implementation of an SDR transceiver containing a mm-wave antenna array system, an RF-box containing all the microwave RF components, (phase shifters, amplifiers, up/down-converters) and two SDR USRP units.
A novel mm-wave antenna array system will be designed and fabricated. Two basic types of antennas can be considered as the best candidates for the antenna system implementation: (a) Electronically Switched Parasitic Array Radiators (ESPARs) and (b) Dielectric Resonator (DR) antennas, or even a combination of them. Experimental verification tests and measurements will be carried out.
Use of robotic devices to implement a sandbox for link optimization tests and development of a Radio Environment Map (REM) service. For this purpose, mobile robots (distributed in an indoor experimentation testbed room) will accommodate lightweight USRPs in order to execute radio monitoring measurements. MIMO-like configurations will be generated according to the REM measurements and the routes of the mobile robotic devices. As a result, the antenna control policy can be proactively determined.
A measurement collection service, a real-time monitoring application and Generative Adversarial Network (GAN) compose the tool for the implementation of the proactive beamformer. Big data analytics tools will be utilized, due to extremely large amount of acquired measured data, in order to enhance the development activities.
Although the mm-wave band has been assigned to 5G and beyond networks there are still open issues and many concerns regarding the transmission characteristics of the mm-wave channels for mobile networks. Some of which are:
High path loss – In the mm-wave band, the path loss becomes very large and the communication range decreases. Moreover, raindrops are roughly the same size as the radio wavelengths and, therefore, cause scattering of the radio signal and energy absorption.
Low signal penetration and alternative paths – While signals at low frequencies can penetrate easily through buildings, mm-wave signals do not penetrate most solid materials. As a result, the movement of obstacles and reflectors, or even changes in the orientation of a handset relative to a body or a hand, can cause the channel to rapidly appear or disappear.
Beam directionality – In current communication technologies, the trend requires transmissions to be mostly directional, through beamforming. Directional links require precise alignment of Tx and Rx beams, and hence there is a high risk of link loss due to beam misalignment in mm-waves. Therefore, the mm-wave-related limitations pose new challenges, as summarized in Table I, for proper mobile protocol design, and require that the next generation of mm-wave mobile communication protocols support mechanisms by which the BSs and the end users can quickly determine the best directions to establish the mmwave link.
The researcher’s attention shifts towards mm-wave technology to meet the challenging objectives and requirements dictated by the growing use of mobile services and applications. As a result, the mm-wave band is necessary for the development of future mobile networks, including 6G. The 6G wireless network must also offer customizable and flexible features and capabilities. As a result of this demand, the new radio access techniques are already integrating Radio Access Network (RAN) functions. The integration of deep learning networks to enable MIMO systems on the road to 6G has already been foreseen by cutting-edge research in radio access techniques. While Deep Learning Networks (DLN) supports unconventional MIMO techniques, more sophisticated learning systems can be employed to forecast RAN behavior based on UE trajectories and REM measurements. Advanced beamformers include learning mechanisms that are able to run on mobile phones. Innovative research is focused on developing hybrid beamforming algorithms that is also hosted at the mobile edge, allowing the UE to offload measurement and delay sensitive tasks to the Mobile Edge Computing (MEC) server. Deep learning is also used in a multitude of channel estimation techniques to minimize the reference signal and feedback overhead. The deep learning models used by current technologies can either forecast the beamforming vectors at the BSs or learn how to acquire channel state information (CSI).
The proposed series of experiments builds on the expertise and experience the Telecommunication Systems Lab (University of Piraeus) has acquired through its involvement in several R&D projects, with the following goals: a) the development of SDR-based modems for mobile standards and proprietary waveforms; b) the development of adaptive and advanced transmission and reception algorithms, c) the design, implementation, and measurement of adaptive mm-wave antennas (ESPAR and DR antennas), d) the development of machine learning algorithms for signal processing tasks in radio and networking functionalities, and e) the performance analysis and evaluation of SDR platforms.
Through deep learning and big data analytics, BEAM-RAISE presents an integrated experimentation engine that can incorporate research findings from multiple ICT disciplines and enable hybrid beamforming and nonconventional MIMO. Within this context, portable robotic equipment will carry onboard lightweight SDR devices acting as UEs. The transceiver will consist of two X310 USRPs. The X310 USRPs will be equipped with a mm-wave antenna-RF module connected to them, designed and delivered by the Lab, in order to enable hybrid beamforming. A mm-wave multiport ESPAR (MuPAR) antenna is an ideal candidate antenna type in order to be designed, implemented, and integrated into the BEAM-RAISE indoor testbed since it offers more degrees of freedom and provides MIMO capabilities. While the learning-supported beamformer engine runs on the MEC entity, following the gathering of CSI feedback from the UEs, the learning techniques for joint channel estimation will be implemented assuming uplink transmission. Additionally, the researchers will develop and assess a powerful monitoring service that can track and gather all measurement data. Tools for big data analytics will be employed during this stage due to the enormous amount of processing data. The testbed that will be developed for the project will provide a variety of top-tier SDR hardware to guarantee the viability of the BEAM-RAISE goal. This range and diversity of SDR resources (particularly the premium X310s SDR platform) is available from the Lab. Moreover, BEAM-RAISE has access to mobile robots operating in the Lab facilities, allowing it to conduct controlled, tractable, and reproducible mobile experiments. The research team will then assess the hybrid beamforming techniques using the SDR equipment and create a test environment for link optimization. In the end, a REM service will be developed.
It is projected that wireless data traffic will increase more than 10,000-fold by the year 2030. These requirements have been considered for the development of 6G wireless communications networks. These networks will have to support high capacity and massive connectivity with an increasingly diverse set of services, applications, and users, such as Machine-to-Machine (M2M), Internet of Things (IoT) etc.
The conventional old standards for mobile communications operate at the sub-6GHz band and are characterized by limited bandwidth and unbounded or large latency. It is apparent that the 6G systems’ requirements cannot be met by these standards and will require data transmission capacities that go well beyond the capabilities of the conventional communication technologies, calling for new solutions. Since the sub-6GHz band is overcrowded, a viable solution is the exploitation of higher frequency bands that offer much larger available bandwidths. In this regard, the mm-wave band (>25 GHz), is very appealing due to the tremendous amount of available spectrum. Until recently, mm-wave signals had not been used for mobile communications due to the many technical and design challenges they pose, which impact nearly every aspect of device engineering, including materials, form factor, industrial design, thermals, and regulatory requirements for radiated power. However, the advancement of semiconductor technology made recently mmwave systems feasible and hence they have been introduced in 5G systems for Fixed Wireless Access applications. An ideal candidate frequency band which would facilitate the use of mm-wave for future 6G mobile communications scenarios, is the 27 GHz band.
There are several promising features in the use of mm-wave frequencies for mobile communications. Besides the huge amount of available bandwidth that allows for high capacities and low latencies, the small size of antennas makes it possible to build complex antenna arrays, that can be used for MIMO multiplexing systems or for beamforming applications to obtain very high antenna gains, thus further increasing the transmission rates. In addition, the inherent security of mm-wave transmissions may also be exploited due to the limited transmission range and the relatively narrow beamwidth that can be achieved. Nevertheless, these potentials are hindered by the very difficult propagation characteristics of the mm-wave channel, creating a number of challenges that have to be encountered, especially when considering scenarios with high mobility.
The propagation conditions at mm-waves are susceptible to severe free space loss, losses due to weather conditions, high penetration losses, shadowing, and severe scattering from local objects. It is thus vital to understand the mm-wave propagation channel characteristics in order to perform a precise and consistent system design of indoor mm-wave mobile communication networks.
Furthermore, 6G mobile communication systems combine a variety of key radio access technologies, involving massive MIMO, non-conventional hybrid beamforming schemes and MEC. Meanwhile, the implementation of deep learning techniques in 6G mobile networks is a rising research topic. The upcoming services and applications announce the evolution from conventional network deployments to intelligent heterogenous learning. Key feature of these new, non-conventional deployments is the ability to dynamically adapt to the new peculiarities of the radio environment and envision the RAN behavior. State-of-the-art research has already considered solutions which integrate massive MIMO and hybrid beamforming schemes supported by DLN. These solutions involve learning at the MEC server for minimizing the overhead and the maximum system delay. At the same time, more advanced innovative deep learning techniques integrate GAN to proactively determine the antenna control strategy and other beyond 5G-related parameters. Unfortunately, few research works have focused on the theory of GANs. Meanwhile, a DLN structure is both powerful and flexible enough to predict changing mobility route and contribute in the extraction and prediction of useful mobility patterns from rapidly changing networks.
The mm-wave antenna is an additional substantial aspect of the proposed experimental platform. A MuPAR antenna system can be considered as an ideal candidate. A novel design will be designed and developed before its performance evaluation. Consequently, it will be installed on the X310 USRPs and serve as the BS antenna of the BEAM-RAISE testbed. Pattern reconfigurability is a critical feature in 5G mobile communications since it allows the ability to utilize various diversity schemes, spatial multiplexing, beamforming (or beam selection) or hybrid beam / antenna selection, in order to improve system performance in terms of capacity and reliability. Furthermore, the new generation mobile communication industry imposes compact size, complexity and cost limitations. Fortunately, these constraints, can be overcome by a reconfigurable parasitic antenna array that is able to provide multiple radiation patterns. The ESPAR antenna is a special type of parasitic antenna array that requires only one RF chain, since it contains a single active element. The second main asset of the ESPAR antenna is the passive parasitic nature of the rest of the array radiators, which are distributed around the active element in several geometrical arrangements and in small distances (λ/4 - λ/16).
The ESPAR antenna preserves several features of a MIMO system, such as beamforming and spatial multiplexing capabilities. However, the beamforming capabilities offered by the ESPAR antenna are limited. Evaluation results reveal that the number of parasitic elements must be particularly increased in order to realize full 360° azimuthal beamforming. This is because the variable reactors (or the PIN diodes) of the passive elements cannot offer the same degrees of freedom at the beam formation, compared to the degrees of freedom that are provided by multi-feed-port antennas (i.e. phased array antennas) when adjusting the excitation amplitudes and phases. Parasitic array antennas with multiple feed ports (MuPARs) have been proposed in order to decrease the number of antenna elements, while maintaining the beamforming ability. A key feature of the MuPAR antennas is that they can demonstrate higher-resolution beamforming capabilities while maintaining the main ESPAR advantages, such as low fabrication cost, compact size and significantly reduced complexity. Moreover, MuPARs can support various advanced hybrid beamforming techniques that result in a substantial enhancement of link reliability and availability, as well as increase the capacity of the provided radio channel. MuPAR also achieves an increased number of degrees of freedom – allowing massive MIMO capabilities while preserving a compact size and a limited number of active elements.
In addition, DRAs are also promising candidates to replace traditional radiating elements at high frequencies, especially for applications at mm-waves and beyond. This is mainly attributed to the fact that DRAs do not suffer from conduction losses and are characterized by high radiation efficiency when excited properly. DRAs rely on radiating resonators that can transform guided waves into unguided waves (RF signals). In the past, these antennas have been mainly realized by making use of ceramic materials characterized by high permittivity and high Q factor (between 20 and 2000). Currently, DRAs made from plastic material (PolyVinyl Chloride - PVC) are being realized. The main advantages of DRAs are summarized as follows: (a) as compared to traditional metallic antennas whose size is proportional to λο, DRAs are characterized by a smaller form factor especially when a material with high dielectric constant εr is selected for the design. (b) due to the absence of conducting material, the DRAs are characterized by high radiation efficiency when a low-loss dielectric material is chosen. This characteristic makes them very suitable for applications at very high frequencies, such as in the range from 30GHz to 300GHz. As a matter of fact, at these frequencies, traditional metallic antennas suffer from higher conductor losses, (c) DRAs can be characterized by a large impedance bandwidth if the dimensions of the resonator and the material dielectric constant are chosen properly, (d) DRAs can be excited using different techniques which is helpful in different applications and for array integration. (e) the gain, bandwidth, and polarization characteristics of a DRA can be easily controlled using different design techniques.
BEAM-RAISE's research team consists of one professor, two postdocs and two PhD students showcasing an unquestionably excellent academic profile. BEAM-RAISE brings together a balanced mix of research expertise, with each research team member contributing complementary skills, competencies, and knowledge.
However, all selected researchers also have large-scale industrial experience and exhibit a strong background in many European and national applied technology projects and companies. This can be considered as a key asset for giving BEAM-RAISE a business perspective and adapting that insight into a useful industrial product. Successful results of the experiment could even lead towards a spin-off or creation of a promising start-up focused on wireless communications beyond 5G, more specifically beamforming and deep learning schemes for wireless communications and antenna development. For example, deploying a mm-wave MuPAR antenna (or Dielectric Resonator - DR) at the BS can be viewed as a low-cost and compact size solution compared to the existing traditional expensive antenna systems integrated at the BSs. Spatial focusing capability and channel orthogonality can be maintained while reducing the number of RF chains required in MIMO systems. These MuPAR antenna advanced characteristics and capabilities can provide a significant economic impact, offering a fresh perspective for the future generation of BSs (particularly the indoor Base Stations used in micro/pico-cells). In addition, the beamforming technology being developed for BEAM-RAISE also has a positive effect and has a significant impact on environmental and health safety. In particular, beamforming can be used to reduce BS transmit power levels to prevent unwanted transmissions due to omnidirectional fixed patterns and minimize exposure to EM radiation. The above effects can be easily used by the BEAM-RAISE research team to evolve the proposed experimental testbed into a powerful business tool.
BEAM-RAISE investigates hybrid beamforming and MIMO technologies for 6G by conducting a series of indoor experiments. A Deep Learning scheme and data analysis based on REM measurements and UE routes are also used to attain a detailed user behavior prediction and subsequently define the beamforming policy employed to the BS's mm-wave antenna system. BEAM-RAISE not only offers a concise theoretical background, but also provides a practical real-world SDR tool for 6G experiments. In real-life situations, the intention is to create an experimentation sandbox that enables connection optimization and brings a whole new perspective to next-generation cellular networks. All these features can be utilized to constitute new transmission techniques for mobile communication channels. In addition, real-time dynamic antenna control and beamforming strategy changes also facilitate the idea of system reconfiguration in the general 6G context.
Furthermore, mm-wave communications demonstrate the ability to offer a considerable impact in science, society, environment, and economy. The scientific significance of mm-wave communications involves greater availability in spectrum providing higher data rates (GBps) and low end-to-end latency. This is necessary in order to upgrade connectivity between the demanding next generation BSs and the increased number of users. Future integrated mobile communication systems will exploit smart end user devices provided with sensorial, cognitive, decision and communication functionalities, able to perceive the surrounding environment, collecting both public-interest information or obtaining data for the autonomous real-time management of the device itself. Moreover, connectivity is a key enabler for the provision of value-added services relating to the different types of users. In the global context of mobile communications, connectivity will be a critical enabler to support the take-off of new business opportunities.
Τhe academic and scientific exploitation of the project is an additional important impact of BEAM-RAISE. The exploitation strategy is based on incorporating innovations, insights and results of the BEAM-RAISE project into its academic profile and 6G research community. This can be accomplished by improving the scope and quality of teaching, incorporating new knowledge and techniques into the taught curriculum, and demonstrating project results to students and faculty members via seminars and tutorials. Additionally, BEAM RAISE's knowledge and experience can be used to introduce new PhD topics in wireless communications, making adjustments to and steering Lab's current research direction, as well as establishing better collaboration with industry partners. Moreover, working with reputable industry partners and device vendors strengthens the relationship between the University and applied research and attracts new collaborations. Lab's involvement in various National and European projects, including real-life experiments, is already of proven value in establishing the Lab as Europe's leading Research and Development organization due to its increased visibility, recognition, and significant publication record. Technically, the Telecommunication Systems Lab (TSL) already supports and it will further enhance its portfolio on 5G networks, mm-wave technology, SDR implementations, and in tools that already include a) channel sounding SDR framework, b) ITSG5, and c) LTE in-house developed protocol stacks. Finally, the BEAM-RAISE implementation will be provided as open source. Future experimenters are more than encouraged to use and improve upon the implemented experimentation engine.
The USRP hardware restrictions (2 Tx and 2 Rx ports per unit) set a limit in the number of RF chains that can utilized for transmission-reception in the BEAM-RAISE experimental testbed (up to 4). Nevertheless, an increased number of degrees of freedom is provided by combining digital beamforming with beamforming in the analog domain realized by the phase shifters and pattern diversity of the mm-wave antenna array. As a result, this hybrid type of beamforming can provide substantial spatial focusing capability and consequently, channel orthogonality similar or even better to conventional MIMO. Therefore, when a MIMO-like transmission in noted for the BEAM-RAISE experiment, we address a scaled-down system with limited number of active elements but extreme “MIMO-like” capabilities of beamforming and interference control due to parasitic antenna elements. Fig. 1 presents a block diagram that illustrates a possible configuration of the BEAM-RAISER receiver, operating as the BS of the experimentation engine.
Fig. 1: Block diagram of the BEAM-RAISE SDR receiver, operating as the base station of the experimentation engine.
Related to the novel mm-wave antennas that will be designed and developed, one of the most basic steps of our methodology is the employment of an array configuration in order to maximize the effective aperture of the antenna and achieve a sufficiently high radiation pattern gain that is able to resolve the high path loss issue, encountered in the mm-wave band. In addition, array antennas provide the feature of pattern reconfigurability and achieve a full 360° azimuthal coverage with a high gain antenna beam. This is necessary in order to satisfy the broadcasting requirement as well as the highly dynamic environment of the mobile communications. Pattern reconfigurability multiplies the digital capabilities of the antenna and increase the reliability of the wireless link. In particular, a pattern reconfigurable array offers the feature of many discrete antenna radiation beams and the ability of an optimal selection among them, allowing the use of pattern diversity and MIMO techniques. The potential of an array to reconfigure its radiation pattern according to the noise environment and steer it towards a desired direction, can be considered highly preferable in vehicular communications. Various mm-wave antenna models will be designed and simulated in a 3D EM solver. The solver that will be employed is the CST Microwave Studio and its operation is mainly based on the FDTD (Finite Difference Time Domain) method. The antenna design will involve the following: (a) design of a fully parameterized antenna geometrical model, (b) assignment of the appropriate material properties, (c) employment of an adequate mesh during simulation, (d) antenna model optimization through a special parameterized investigation of the basic design parameters and (e) generation of simulation results (i.e. S11 vs frequency, radiation pattern, directivity, gain).