Model-Driven Deep Neural Network for Enhancing Direction Finding with Commodity 5G gNodeB
Shengheng Liu, Zihuan Mao, Xingkang Li, Mengguan Pan, Peng Liu, Yongming Huang, Xiaohu You · 2024
Read Model-Driven Deep Neural Network for Enhancing Direction Finding with Commodity 5G gNodeB by Shengheng Liu, Zihuan Mao, Xingkang Li, Mengguan Pan, Peng Liu, Yongming Huang, Xiaohu You (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-10644.txt.
Introduction
Shengheng Liu and co-authors improve how a 5G base station finds the direction of a signal, in a 2024 paper. Finding direction well is a base skill for smart linked devices, but cheap hardware warps signals and hurts accuracy. Pure maths methods often fail when those warps are present in real sites.
The team recast the task as restoring a picture of the signal spread, and built a neural network guided by the maths of the problem. One part forms beams, one builds a wider view of the signal, and one rebuilds a clean spread map. The design corrects phase errors on its own. Tests in simulation and in the field showed accurate direction finding with standard 5G gear, a first for this mixed approach, the authors say.
Catalog record
| CREATOR | Shengheng Liu, Zihuan Mao, Xingkang Li, Mengguan Pan, Peng Liu, Yongming Huang, Xiaohu You |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-10644
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Model-Driven Deep Neural Network for Enhancing Direction Finding with Commodity 5G gNodeB Shengheng Liu, Zihuan Mao, Xingkang Li, Mengguan Pan, Peng Liu, Yongming Huang, Xiaohu You Abstract . Pervasive and high-accuracy positioning has become increasingly important as a fundamental enabler for intelligent connected devices in mobile networks. Nevertheless, current wireless networks heavily rely on pure model-driven techniques to achieve positioning functionality, often succumbing to performance deterioration due to hardware impairments in practical scenarios. Here we reformulate the direction finding or angle-of-arrival (AoA) estimation problem as an image recovery task of the spatial spectrum and propose a new model-driven deep neural network (MoD-DNN) framework. The proposed MoD-DNN scheme comprises three modules: a multi-task autoencoder-based beamformer, a coarray spectrum generation module, and a model-driven deep learning-based spatial spectrum reconstruction module. Our technique enables automatic calibration of angular-dependent phase error thereby enhancing the resilience of direction-finding precision against realistic system non-idealities. We validate the proposed scheme both using numerical simulations and field tests. The results show that the proposed MoD-DNN framework enables effective spectrum calibration and accurate AoA estimation. To the best of our knowledge, this study marks the first successful demonstration of hybrid data-and-model-driven direction finding utilizing readily available commodity 5G gNodeB. DOI: XXXXXXX.XXXXXXX Journal: TOSN Volume: 20 4 1 1 CCS: Hardware Wireless integrated network sensors CCS: Networks Location based services CCS: Theory of computation Machine learning theory 1. Introduction High-accuracy positioning holds promise in supporting a diverse range of services within the forthcoming-generation of wireless communications [ You et al., 2023 ] . The accurate position aids mobile devices in optimizing resource allocation and network configuration, thereby reducing communication latency and energy consumption [ Wang et al., 2023 ] . Additionally, location awareness assumes a critical stance in bolstering security and privacy protection, facilitating identity authentication and safeguarding sensitive data in the sphere of cyber-physical systems. The advent of high-accuracy positioning services will transcend the conventional boundaries of mobile user lifestyles and usher in transformative prospects across diverse niche markets [ Dang et al., 2020 ] . While a variety of techniques are available to accomplish this goal, including the global positioning system (GPS) [ Nguyen et al., 2015 ] , ultra wide-band (UWB) [ Pannuto et al., 2018 ] , and vision-inertial fusion [ Teng et al., 2019 ] , each manifests distinct advantages and disadvantages. Nevertheless, none singularly achieves comprehensive coverage across both indoor and outdoor scenarios without the need for supplementary device installation. Recent developments have witnessed a notable surge in the deployment of fifth-generation (5G) base stations (a.k.a. gNodeBs or gNBs) in both indoor and outdoor environments to meet the demands for service quality and extensive coverage [ Li et al., 2022 ] , with their numbers projected to continue growing in the foreseeable future. This presents a viable opportunity to realize high-precision positioning by leveraging the 5G infrastructure itself, thereby obviating the necessity for additional hardware [ Pan et al., 2022 ] . Consequently, the integration of high data-rate communication and high accuracy positioning assumes an unprecedentedly heightened significance [ Liu et al., 2022 ] . The positioning methods has been extensively explored and can be broadly categorized into two main groups: fingerprint-based methods and geometric-based methods. The fingerprint-based methods capitalizes on the uniqueness of the wireless channel that interconnects users and access point (AP), which is shaped by the intricate scattering environments [ Lin et al., 2021 ] . However, the dynamic nature of the propagation medium introduces a discrepancy between the receive signal strength indicator (RSSI) and the outdated fingerprint database, thereby compromising the robustness of fingerprint-based methods [ Zhu et al., 2021 ] . To overcome this issue, various strategies have been proposed that involve fusing derivative fingerprints to extract the resilient component of RSSI and enhance the accuracy of positioning [ Guo et al., 2019 ] . However, the computational complexity of these methods grows exponentially with the expansion of the database, thus posing a formidable challenge in terms of the requisite computation resources [ Huang et al., 2020 ] . In contrast to the fingerprint-based methods, geometric-based methods impose minimal requirements concerning environmental dynamics [ Chen et al., 2022 ] . The 3rd Generation Partnership Project (3GPP) Release 16 [ 3rd Generation Partnership Project, 3GPP ] , which reached its finalization in 2021, introduced several geometric-based schemes tailored for 5G. These encompass downlink-time-difference-of-arrival (DL-TDoA), uplink-time-difference-of-arrival (UL-TDoA), downlink-angle-of-departure (DL-AoD), uplink-angle-of-arrival (UL-AoA), and multi-round trip time (Multi-RTT) positioning techniques. Direction finding is a frequently visited problem in the realm of multichannel signal processing. When assuming an idealized mathematical model with perfect system response, maximum-likelihood (ML) [ Stoica and Nehorai, 1989 ] and subspace algorithms [ Schmidt, 1986 ] can be harnessed to estimate geometric parameters with super-resolution. However, the true system response invariably deviates from the highly idealized mathematical model. For instance, the accuracy of array response for angle-of-arrival (AoA) estimation is affected by mutual coupling effects, sensor location perturbations, gain-phase mismatches in multichannel receivers, and other unpredictable factors [ Liu et al., 2018 , Wang et al., 2018 ] . To tackle this challenge, the nonlinear fitting capabilities of artificial intelligence (AI) are envisioned to usher in a new paradigm for positioning [ Xiao et al., 2019 ] . The concept of AI-based direction finding was initially introduced by Rastogi et al. [ Rastogi et al., 1987 ] , where a Hopfield neural network was employed to establish a relationship between an energy function and signal parameters for direction estimation. More recent advancements have focused on deep neural network (DNN) approaches for AoA estimation in wireless systems [ Barthelme and Utschick, 2021 ] . One such approach involves a modified DNN with multiple hidden layers designed to extract sparse angular domain features, thereby enhancing the accuracy of estimation [ Huang et al., 2018 ] . In parallel, feature-to-feature methodologies have been explored, offering incremental improvements in the physical interpretability of these models [ Xiang et al., 2020 ] . By leveraging neural networks, specific input parameter features [ Naseri et al., 2022 ] have been used for AoA estimation, which reinforces the linkage between input data and AoA estimation with clearer physical significance. Nevertheless, their accuracy relies heavily on the availability of extensive datasets, curtailing their ability to generalize effectively across diverse scenarios. A two-stage multi-layer perceptron was later proposed to improve AoA resolution and strengthen generalization capability [ Zhang et al., 2024 ] . Still, these methods lack explicit designs to address hardware impairments, resulting in degraded estimation accuracy as such impairments become more influential. To tackle this challenge, super-resolution AoA estimation was achieved by using time-domain signals as input to train the DNN [ Chen et al., 2024 ] . Nonetheless, purely data-driven approaches like this focus on training a single parameter, often neglecting valuable information linked to other critical feature parameters [ Jiang et al., 2019 ] . In contrast to neural networks that directly output AoA values, some methods solve the problem indirectly by reconstructing the spatial spectrum [ Wu et al., 2019 , Ren et al., 2022 ] . Adaptations of the multiple signal classification (MUSIC) algorithm through deep learning have demonstrated superior spectrum recovery, particularly under low signal-to-noise ratio (SNR) conditions and limited snapshots [