Kalman-aided uplink channel estimation for time-varying multi-user vehicular STAR-RIS systems Authors: Aswiniya Ambigapathy, Sriharipriya KC Published: 2026 ABSTRACT Accurate uplink channel estimation in high-mobility vehicular networks remains challenging due to rapid channel variations and Doppler effects, especially in systems assisted by simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS). Most existing studies focus on limited user scenarios and static or low-mobility conditions, which restricts their applicability to realistic vehicular environments. This work proposes a Kalman-aided channel estimation framework for multi-user uplink vehicular STAR-RIS systems operating under high mobility. The proposed approach integrates least squares (LS) estimation with discrete Fourier transform (DFT)-based orthogonal pilot design to obtain initial channel estimates, followed by a Kalman filter to continuously track time-varying Rician fading channels modeled with Jakes temporal correlation. Path-loss scaling is incorporated to improve composite channel tracking accuracy. The framework effectively mitigates Doppler-induced estimation degradation and supports scalable multi-user operation. Simulation results show that the proposed estimator consistently outperforms conventional methods in high-mobility scenarios. In particular, the normalized mean square error is reduced by up to 25 dB for the time-switching protocol and approximately 22 dB for the energy-splitting protocol compared with the corresponding baseline estimation schemes. These results indicate that the proposed estimation and tracking framework maintains stable accuracy across a wide range of vehicular operating conditions. Introduction The growth of vehicle-to-everything (V2X) communications aims to reduce accidents and make everyday travel safer and more convenient. The evolution of sixth-generation (6G) wireless networks and intelligent transportation systems (ITS) has significantly increased the demand for ultra-reliable, high-throughput, and low-latency communication in vehicular environments. These systems form the backbone of emerging applications such as autonomous driving, real-time traffic control, and cooperative vehicular safety. However, vehicular communication links are severely degraded by complicated propagation environments, particularly due to building blockages in metropolitan areas and rapidly time-varying channels caused by high vehicle mobility1. Reconfigurable Intelligent Surfaces (RIS), also referred to as Intelligent Reflecting Surfaces (IRS), have recently emerged as a promising technology to enhance wireless coverage, spectral efficiency, and energy efficiency2. By intelligently adjusting signal propagation using a large number of low-cost passive elements, RIS enables the creation of a reconfigurable radio environment. However, conventional RIS architectures can only reflect signals within a single half-space, thereby serving users located on one side of the surface only3. To overcome this limitation, a brand-new idea called Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surfaces (STAR-RIS) has been introduced4. It can transmit and reflect the incident signals, allowing users on both sides of the RIS to achieve a full-space smart radio environment. The benefits of implementing STAR-RIS in automotive networks, such as increased system throughput, reduced power consumption, and extended network coverage, have sparked growing research interest5. To enable full-space signal manipulation, STAR-RIS operates under three practical protocols, namely energy splitting (ES), mode switching (MS), and time switching (TS), each offering distinct trade-offs between design flexibility, implementation complexity, and transmission-reflection control: ES allows simultaneous transmission and reflection with fine-grained amplitude and phase tuning; MS partitions elements into either transmission or reflection mode for simpler hardware; and TS alternates all elements between transmission and reflection over time slots, enabling independent optimization at the cost of synchronization overhead4. In practical wireless systems, hardware imperfections are inevitable and may noticeably influence system performance, particularly in high-mobility and high-frequency scenarios. Typical impairments include I/Q imbalance, limited phase resolution, and mutual coupling among antenna and surface elements. In reconfigurable intelligent surface systems, such non-idealities lead to deviations between the designed and actual phase and amplitude responses, resulting in configuration mismatches. Previous studies have reported that hardware impairments at both the transceiver and RIS sides can degrade beamforming performance, channel estimation accuracy, and achievable data rates, especially when a large number of surface elements is employed6–8. These effects become more pronounced in vehicular environments, where Doppler shifts and rapid channel variations further aggravate the impact of hardware non-idealities. For this reason, incorporating hardware impairments is necessary to develop realistic and reliable STAR-RIS-assisted channel estimation frameworks for high-mobility vehicular communication systems. The fundamental protocols of STAR-RIS can be efficiently integrated with several emerging technologies. For example, combining STAR-RIS with MIMO or massive MIMO systems improves spatial multiplexing and beamforming performance, which supports reliable multi-user communication in high-mobility vehicular scenarios5. The use of millimeter-wave (mmWave) or terahertz (THz) communications helps overcome line-of-sight blockages that commonly occur in urban vehicular environments, thereby extending coverage and improving link reliability9. Combining STAR-RIS with non-orthogonal multiple access (NOMA) can improve spectral efficiency and allow multiple vehicles to be served simultaneously in dense traffic conditions10. In addition, STAR-RIS has been investigated within integrated sensing and communication (ISAC) frameworks, where vehicular nodes carry out radar sensing and data transmission in parallel5,10. These works indicate that STAR-RIS can support joint communication sensing operations and improve coverage and link quality in next-generation V2X systems. Although STAR-RIS has demonstrated promising gains in coverage and throughput, most existing works still examine it under static or slowly changing conditions. Its performance and effectiveness in fast vehicular environments have not yet been sufficiently examined. In particular, the challenges related to time-varying channel estimation, multi-user interference, and protocol optimization between the TS and ES modes under Doppler effects remain largely unaddressed. To address these gaps, this paper investigates a STAR-RIS-assisted multi-user vehicular communication system and develops a channel estimation framework based on Least Squares (LS), discrete Fourier transform-based orthogonal pilot design, and Kalman filtering. Both the TS and ES protocols are considered under Rician fading channels with Jakes temporal correlation. Closed-form LS estimators and normalized mean square error (NMSE) expressions are derived, and a Kalman filter is employed to track the temporal variation of the composite direct and cascaded channels. The performance of the proposed framework is evaluated in terms of NMSE and achievable channel capacity under different system parameters, including pilot length, transmit power, and vehicle speed. The main objective of this work is to improve channel estimation accuracy for STAR-RIS-assisted vehicular communication systems operating in realistic high-mobility environments. This study provides a practically implementable channel estimation framework for STAR-RIS-assisted vehicular communication systems under high mobility, with explicit computational complexity and pilot overhead considerations. We note that this work considers a narrowband channel model, primarily to keep the analysis tractable and focus on the core estimation and tracking problem in STAR-RIS-assisted vehicular systems. We recognize that real-world vehicular standards like 5G NR-V2X use wideband OFDM, where the channel must be handled across multiple subcarriers. The proposed LS-Kalman framework could potentially be applied to such systems by running the estimator on each subcarrier separately, though this would naturally increase the computational load depending on the number of subcarriers involved. How to handle this efficiently in resource-limited vehicular platforms is a practical challenge that we leave open for future work. Motivated by the above observations, this paper investigates uplink channel estimation for STAR-RIS-assisted multi-user vehicular communication systems under realistic high-mobility conditions. The main contributions of this work are summarized as follows. First, we develop a unified LS-Kalman channel estimation framework for both TS and ES protocols by combining DFT-based orthogonal pilot design with Kalman filtering to track time-varying Rician channels modeled using Jakes temporal correlation. Second, we extend existing STAR-RIS channel estimation studies from static or two-user scenarios to a general multi-user uplink vehicular setting and analyze the impact of the number of users, vehicle speed, and pilot length on estimation accuracy. Third, closed-form NMSE expressions are derived for both TS and ES protocols, and the achievable uplink channel capacity under imperfect CSI is evaluated by explicitly incorporating the residual estimation error. Extensive simulation results show that the proposed framework performs well under Doppler effects and multi-user interference, and that the TS protocol provides lower NMSE and higher spectral efficiency than the ES protocol in high-mobility vehicular scenarios. Related work This section reviews the related literature on STAR-RIS in vehicular and high-mobility communications scenarios. In5, Li et al. explored STAR-RIS-aided Integrated Sensing and Communication (ISAC) for high-mobility vehicles, demonstrating channel parameter extraction and vehicle localization using multidimensional orthogonal matching pursuit (MOMP) algorithms. In11, Li et al. proposed an energy-harvesting relay-assisted STAR-RIS-enhanced vehicular NOMA network. The authors derive outage probabilities for near and far vehicles using Gamma-distributed channel approximations and demonstrate that adjusting power allocation and STAR-RIS coefficients under the energy splitting (ES) protocol improves fairness and reliability without increasing surface complexity. In12, Li et al. deployed STAR-RIS in a high-mobility ISAC framework to simultaneously serve indoor users and track outdoor dynamic scatterers in the environment. A GM-PHD filter is used for multi-target tracking, and beam prediction strategies are developed to enhance communication performance, showcasing STAR-RIS potential for mobility-aware sensing and beamforming. In13, Wu et al. investigated channel estimation schemes for STAR-RIS under the two fundamental protocols: Time Switching (TS) and Energy Splitting (ES). Results demonstrated that TS yields lower estimation error compared to ES due to the absence of power leakage, and they proposed a near-optimal training pattern design under practical phase-shift constraints. However, their analysis was restricted to generic wireless scenarios, without considering the challenges of vehicular communication where high mobility, Doppler effects, and multi-user interference significantly complicate CSI acquisition. Different from the STAR-RIS channel estimation method in13, which considers a generic static or slowly varying scenario and mainly focuses on training pattern optimization under practical phase-shift constraints, this work explicitly addresses high-mobility vehicular environments with time-varying Rician fading and Jakes temporal correlation. In addition, while13 mainly investigates two-user setups without temporal tracking, the proposed framework extends the analysis to a general multi-user uplink scenario and incorporates a Kalman filter to continuously track Doppler-induced channel variations. Furthermore, this paper provides a unified LS-Kalman estimation framework for both TS and ES protocols and evaluates their performance in terms of NMSE and achievable channel capacity under realistic vehicular mobility, which is not considered in13. In14, Aung et al. optimized a STAR-RIS-assisted V2X using a deep reinforcement learning (DRL) framework. The authors jointly optimize spectrum allocation, STAR-RIS configuration, and beamforming to maximize V2I data rates while satisfying V2V latency and reliability constraints. Their DDQN-attention based solution demonstrates STAR-RIS’s adaptability in complicated vehicular networks by outperforming traditional RIS and vanilla DRL techniques. In15, Shang et al. proposed an AGSM-assisted STAR-RIS scheme that integrates generalized spatial modulation (GSM) with Alamouti space-time block coding (STBC). In the first design, STAR-RIS is partitioned into reflective and transmissive blocks, to enable simultaneous multi-user communication, thereby improving BER performance. In a second design, STAR-RIS itself functions as an access point (AP), which simplifies hardware implementation and enhances spectral efficiency. Likewise, in16, Ghadi et al. evaluated a dual RIS/STAR-IOS-assisted Vehicle-to-Vehicle NOMA/OMA system under composite Fisher-Snedecor F fading and demonstrated that STAR-IOS greatly enhances performance, primarily under NOMA schemes, by obtaining closed-form formulas for outage probability, ergodic capacity, underscoring its promise for vehicular communication. Although the above works demonstrate the potential of STAR-RIS in vehicular, ISAC, NOMA, and multi-user communication scenarios, most of them focus on system-level performance optimization, sensing-communication integration, or outage and capacity analysis under quasi-static or slowly varying channels. Moreover, the majority of existing studies either consider single-user or two-user setups, or do not explicitly address the challenges of channel estimation in high-mobility vehicular environments. In particular, the joint impact of Doppler effects, Jakes temporal correlation, and multi-user interference of STAR-RIS-assisted uplink channel estimation under both TS and ES protocols remains largely unexplored. The main symbols and notations used throughout this paper are summarized in Table 1. Table 1List of main symbols.SymbolDescriptionmNumber of antennas at the base stationNTotal number of passive elements at the STAR-RISMNumber of STAR-RIS sub-surfaces (element groups)KNumber of vehicular users \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} Pilot length under the Time Switching (TS) protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:$$\end{document} Pilot length under the Energy Splitting (ES) protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document} Composite channel vector of user k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}$$\end{document} Direct channel from user k to the BS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{r}_{k}$$\end{document} Channel from user k to the STAR-RISGChannel from STAR-RIS to the BSΘSTAR-RIS phase-shift matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document} DFT-based pilot matrix for TS protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}$$\end{document} DFT-based pilot matrix for ES protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:P}_{k}$$\end{document} Transmit power of user k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\beta\:}_{m}^{t},{\:\beta\:}_{m}^{r}$$\end{document} Energy splitting ratios (ES protocol) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\sigma\:}^{2}$$\end{document} Noise power \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} Maximum Doppler frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{T}_{s}$$\end{document} Sampling interval \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{F}_{k}$$\end{document} Kalman state transition matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{k}$$\end{document} Process noise covariance matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{err,\:k}$$\end{document} Channel estimation error covariance matrix Table 1List of main symbols.SymbolDescriptionmNumber of antennas at the base stationNTotal number of passive elements at the STAR-RISMNumber of STAR-RIS sub-surfaces (element groups)KNumber of vehicular users \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} Pilot length under the Time Switching (TS) protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:$$\end{document} Pilot length under the Energy Splitting (ES) protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document} Composite channel vector of user k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}$$\end{document} Direct channel from user k to the BS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{r}_{k}$$\end{document} Channel from user k to the STAR-RISGChannel from STAR-RIS to the BSΘSTAR-RIS phase-shift matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document} DFT-based pilot matrix for TS protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}$$\end{document} DFT-based pilot matrix for ES protocol \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:P}_{k}$$\end{document} Transmit power of user k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\beta\:}_{m}^{t},{\:\beta\:}_{m}^{r}$$\end{document} Energy splitting ratios (ES protocol) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\sigma\:}^{2}$$\end{document} Noise power \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} Maximum Doppler frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{T}_{s}$$\end{document} Sampling interval \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{F}_{k}$$\end{document} Kalman state transition matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{k}$$\end{document} Process noise covariance matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{err,\:k}$$\end{document} Channel estimation error covariance matrix List of main symbols. System model As shown in Fig. 1, we consider a narrow-band wireless channel in a vehicular communication scenario, where a STAR-RIS with N passive elements is installed to assist uplink data transmission from K single-antenna vehicular users to a base station (BS) equipped with m antennas. To minimize the channel estimation cost, we recall the element-grouping technique in17,18, which divides the STAR-RIS elements into M sub-surfaces (M < < N), each consisting of N/M neighboring elements that share a common transmission/reflection coefficient. This grouping strategy reduces the channel estimation overhead while preserving adequate beamforming flexibility. The Jakes temporal correlation model is employed to describe the time variation of the wireless channels caused by vehicular motion. To account for both line-of-sight (LOS) and non-line-of-sight (NLOS) components, the complex channel amplitudes are modeled according to a Rician fading distribution, while their temporal variations are governed by the Jakes temporal correlation model. The direct channels from vehicular user k to the base station (BS) at time t and to the STAR-RIS are represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}\left(t\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:\:{C}^{(m\:\times\:\:1)}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{r}_{k}\left(t\right)\in\:\:{C}^{(M\:\times\:\:1)}$$\end{document}, respectively, where k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:$$\end{document} K = {1, 2,…, K}. The channel from STAR-RIS to the BS is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:G\left(t\right)\in\:\:{C}^{(m\:\times\:\:M)}$$\end{document}. The effective cascaded channel from user k via the STAR-RIS is then given by,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Q}_{k}\left(t\right)=G\left(t\right)\:{\Theta\:}\:\left(\mathrm{t}\right)\:{r}_{k}\left(t\right)\:\in\:\:{C}^{(m\:\times\:\:1)}$$\end{document} Depending on the employed protocol, Θ (t) corresponds to the TS or ES phase-shift matrix defined in Sections III and IV. Where, Θ (t) = diag (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{j{\theta\:}_{1}\left(t\right)},\:{e}^{j{\theta\:}_{2}\left(t\right)},\:\dots\:..,\:{e}^{j{\theta\:}_{M}\left(t\right)})$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:\:{C}^{\left(M\:\times\:\:M\right)}\:$$\end{document}represents the STAR-RIS phase-shift matrix under the considered protocol. Fig. 1A STAR-RIS-assisted multi-user uplink vehicular communication system. A STAR-RIS-assisted multi-user uplink vehicular communication system. The channels follow the Jakes spectrum and are time-varying due to vehicular mobility. Each channel coefficient is modeled as a superposition of multiple sinusoids according to the Jakes fading model, given by2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k\:}\left({t}_{i}\right)=\:\sum\:_{n=1}^{{N}_{s}}{\alpha\:}_{k,n}{e}^{j(2\pi\:{f}_{d,n}{t}_{i}+{\varPhi\:}_{k,n})}$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{r}_{k\:}\left({t}_{i}\right)=\:\sum\:_{n=1}^{{N}_{s}}{\beta\:}_{k,n}{e}^{j(2\pi\:{f}_{d,n}{t}_{i}+{\psi\:}_{k,n})}$$\end{document}4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:G\left({t}_{i}\right)=\:\sum\:_{n=1}^{{N}_{s}}{\gamma\:}_{n}{e}^{j(2\pi\:{f}_{d,n}{t}_{i}+{\chi\:}_{n})}$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\gamma\:}_{n}\in\:\:{C}^{\left(m\:\times\:\:M\right)}\:$$\end{document}denotes the complex amplitude matrix of the n-th path, Ns denotes the number of sinusoids, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{d,n}\:$$\end{document} is the Doppler frequency shift of the n-th path, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{k,n},\;{\beta\:}_{k,n},$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\gamma\:}_{n}$$\end{document} are complex amplitudes following a Rician distribution, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{k,n},$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\psi\:}_{k,n},$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\chi\:}_{n}$$\end{document} represent independent random phases uniformly distributed over [0,2π]. At time instant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:t}_{i}$$\end{document}, the received signal at the BS from user k is expressed as,5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k}\left({t}_{i}\right)=\sqrt{{\mathrm{P}}_{\mathrm{k}}}{(h}_{k}\left({t}_{i}\right)+{Q}_{k}\left({t}_{i}\right)\left)\:{s}_{k}\right({t}_{i})+{n}_{k}({t}_{i})$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{s}_{k}\left({t}_{i}\right)$$\end{document} is the pilot or data symbol transmitted by user k with transmit power \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:P}_{k}$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{n}_{k}\left({t}_{i}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\sim\:CN\:(0,{\sigma\:}^{2}{I}_{m})$$\end{document} represents additive white Gaussian noise. Due to the high mobility of vehicular users, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}\left({t}_{i}\right)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{r}_{k}\left({t}_{i}\right)$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:G\left({t}_{i}\right)$$\end{document} vary according to the Jakes fading model, but are approximately constant within each short transmission block. The system operates under time-switching (TS) and energy-splitting (ES) protocols. Under the TS protocol, the STAR-RIS alternates between transmission and reflection modes in separate time intervals, while under the ES protocol, each sub-surface simultaneously performs both transmission and reflection by splitting the impinging energy between the two sides. STAR-RIS channel estimation under time switching for multi-user vehicular communication In this part, we suggest an efficient scheme to separately estimate the channels of the K users under the TS protocol. Transmission and reflection model For TS, the STAR-RIS alternates all components between the transmission and reflection modes over two distinct time intervals known as the T-period and R-period. The number of time slots allocated to the T-period and R-period, respectively, is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\tau\:}_{R}$$\end{document}. Additionally, we define \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document}iTS \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\triangleq\:$$\end{document} [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{{j\theta\:}_{1,\mathrm{i}\:}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{{j\theta\:}_{2,\mathrm{i}}}$$\end{document} ………\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:,{e}^{{j\theta\:}_{\mathrm{M},\mathrm{i}}}$$\end{document}], \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\phi\:$$\end{document}iTS\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\triangleq\:$$\end{document} [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{{j\phi\:}_{1,\mathrm{i}\:}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}^{{j\phi\:}_{2,\mathrm{i}}}$$\end{document} ………\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:,{e}^{{j\phi\:}_{\mathrm{M},\mathrm{i}}}$$\end{document}] as the transmission and reflection coefficient vectors, respectively, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\theta\:}_{m,i}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\phi\:}_{m,i}\in\:$$\end{document} [0,2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\pi\:$$\end{document}), m \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{M}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\triangleq\:\:$$\end{document}{1, ……, M} is the phase-shift of the m-th sub-surface in time slot i. Signal model The channel estimation design for the k-th user during the transmission T-period is described in detail below. In the case of R-period, the same approach can be applied directly. Every user k \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:$$\end{document} K = {1, 2,…, K} sends pilot symbols to the BS in succession during the T-period. The maximum transmit power over \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} time intervals is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\mathrm{P}}_{\mathrm{k}}$$\end{document}. In time slot i, i = 1, 2,……,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\tau\:}_{T}$$\end{document}, the received signal at the BS is as follows:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k,i}^{\left(T\right)}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}\left({\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\mathrm{G}\left({t}_{i}\right)diag\left({\theta\:}_{i}^{TS}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right){\mathrm{s}}_{\mathrm{k},\mathrm{i}}+{\mathrm{n}}_{\mathrm{k},\mathrm{i}}^{\left(\mathrm{T}\right)}$$\end{document} Where, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{n}_{k,i}^{\left(T\right)} \sim CN\left(0,{\sigma\:}^{2}{I}_{m}\right)$$\end{document} is the additive white Gaussian noise (AWGN) at the BS; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{s}_{k,i}\:$$\end{document} is the pilot symbol in time slot i, which can be set as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{s}_{k,i}=1$$\end{document} for simplicity. Similarly, for the R-period, the STAR-RIS operates in reflection mode for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{R}$$\end{document} time slots, and the received signal at the BS from the same k-th user in time slot i, i = 1, 2,……,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{R}$$\end{document}, is written as,7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k,i}^{\left(R\right)}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}\left({\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\mathrm{G}\left({t}_{i}\right)diag\left({\phi\:}_{i}^{TS}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right){\mathrm{s}}_{\mathrm{k},\mathrm{i}}+{\mathrm{n}}_{\mathrm{k},\mathrm{i}}^{\left(\mathrm{R}\right)}$$\end{document} Stacking all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} slots for the T-period into vector form:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k}^{\left(T\right)}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}\:{\varPhi\:}_{T}{x}_{k}+{n}_{k}^{\left(T\right)}$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document} = [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}^{T},\:{r}_{k}^{T}{]}^{T}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:\:{C}^{\left(m\:+\:M\right)\:\times\:\:1}\:$$\end{document}is the composite channel vector, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}\in\:\:{C}^{{\tau\:}_{T}\:\times\:\:(m+M)}$$\end{document} is the pilot matrix incorporating STAR-RIS phase shifts. Similarly, stacking all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{R}$$\end{document} slots for the R-period into vector form:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k}^{\left(R\right)}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}\:{\varPhi\:}_{R}{z}_{k}+{n}_{k}^{\left(R\right)}$$\end{document} With \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{z}_{k}$$\end{document} = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{[}{h}_{k}^{T},\:{r}_{k}^{T}{]}^{T}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{R\:}\in\:\:{C}^{{\tau\:}_{R}\:\times\:\:(m+M)}$$\end{document} as the reflection pilot matrix. LS Estimator for TS (T-period) According to Eq. (5), we start from the stacked pilot observation model for the T-period (the R-period is identical with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\varPhi\:}_{R}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{z}_{k}\:$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k}^{\left(R\right)}$$\end{document} replacing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T,}\;{x}_{k}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{y}_{k}^{\left(T\right)}$$\end{document}) with the goal of estimating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:x}_{k}$$\end{document}. The LS estimator minimizes the squared error10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{J}\left({x}_{k}\right)=\Vert\:{y}_{k}^{\left(T\right)}\:-\:\sqrt{{\mathrm{P}}_{\mathrm{k}}}\:{\varPhi\:}_{T}{{x}_{k}\left|\right|}_{2}^{2}$$\end{document} Assuming \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document} has full column rank, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}^{H}{\varPhi\:}_{T}$$\end{document} is invertible, and the LS estimate is,11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\widehat{\mathrm{x}}}_{\mathrm{k}}}^{(LS,\:T)}=\:\frac{1}{\sqrt{{\mathrm{P}}_{\mathrm{k}}}\:}\:({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}{)}^{-1}{\varPhi\:}_{T}^{H}{y}_{k}^{\left(T\right)}$$\end{document} If the pilot matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document} is scaled to include the transmit power (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\stackrel{\sim}{\varPhi\:}}_{T}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document}), the LS estimator simplifies to:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\widehat{\mathrm{x}}}_{\mathrm{k}}}^{(LS,\:T)}=\:({\stackrel{\sim}{\varPhi\:}}_{T}^{H}\:{{\stackrel{\sim}{\varPhi\:}}_{T})}^{-1}{\stackrel{\sim}{\varPhi\:}}_{T}^{H}{y}_{k}^{\left(T\right)}$$\end{document} Estimation error and error covariance (TS) Define the estimation error13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}_{k}^{\left(T\right)}=\:{{\widehat{\mathrm{x}}}_{\mathrm{k}}}^{(LS,\:T)}-\:{x}_{k}\:=\:\frac{1}{\sqrt{{\mathrm{P}}_{\mathrm{k}}}\:}\:({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}{)}^{-1}{\varPhi\:}_{T}^{H}{n}_{k}^{\left(T\right)}$$\end{document} Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{n}_{k}^{\left(T\right)}$$\end{document}~\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:CN\left(0,{\sigma\:}^{2}{I}_{{\tau\:}_{T}}\right)$$\end{document}, the error covariance matrix is14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\boldsymbol{C}}_{err,k}^{\left(T\right)}=\:E\left[{\mathrm{e}}_{k}^{\left(T\right)}{(e}_{k}^{\left(T\right)}{)}^{H}\right]\:=\:\:\frac{{\sigma\:}^{2}}{{P}_{k}}\:({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}{)}^{-1}$$\end{document} MSE and normalized MSE (NMSE) for TS The MSE of the LS estimator in the T-period is defined as,15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{\left(T\right)}=E|\left|{\widehat{\mathrm{x}}}_{\mathrm{k}}-\:{x}_{k}\right|{|}_{2}^{2}\:=\:E\left\|{e}_{k}^{\left(T\right)}\right\|_{2}^{2}$$\end{document} Using the trace form,16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{\left(T\right)}=\:Tr\:\left\{{\boldsymbol{C}}_{err,k}^{\left(T\right)}\right\}=\frac{{\sigma\:}^{2}}{{P}_{k}}Tr\left[\right({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}{)}^{-1}]$$\end{document} The normalized MSE (NMSE) is defined by normalizing the MSE w.r.t. the average channel energy,17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{\left(T\right)}=\frac{\:{\sigma\:}^{2}Tr\left[\right({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}{)}^{-1}]}{{P}_{k}E\left[\left\|x_k\right\|_2^2\right]}$$\end{document} Orthogonal DFT pilots (TS-optimal design) Assume we design \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{T}$$\end{document} using orthogonal DFT-based pilots such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\varPhi\:}_{T}^{H}{\varPhi\:}_{T}=\:{\tau\:}_{T\:}{I}_{m+M}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\:\tau\:}_{T\:}\ge\:\:m+M,$$\end{document} then18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{\left(T\right)}=\frac{\:{\sigma\:}^{2}}{{P}_{k}}\:\frac{m+M}{\:{\tau\:}_{T\:}}$$\end{document} This shows explicitly: (i) MSE decreases as 1/\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T\:}$$\end{document} (longer pilot = better estimation), (ii) MSE increases with the dimension (m+M) (more unknowns = harder estimation). If we choose the minimum TS overhead \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\tau\:}_{T\:}=m+M$$\end{document}, then19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{\left(T\right)}=\:\frac{{\sigma\:}^{2}}{{P}_{k}}\:$$\end{document} The corresponding NMSE is20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{\left(T\right)}=\frac{\:\:\:{\sigma\:}^{2}\:}{{P}_{k}E\left[\left\|x_k\right\|_2^2\right]}$$\end{document} This expression holds under orthogonal DFT-based pilot design with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\tau\:}_{T\:}=m+M$$\end{document}. Which is the minimum achievable LS, MSE and NMSE under the unit-modulus/DFT pilot choice. The R-period derivations are identical, replacing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\varPhi\:}_{T},\;\:{y}_{k}^{\left(T\right)},\;{\:MSE}_{k}^{\left(T\right)}\;and\;{NMSE}_{k}^{\left(T\right)}$$\end{document} by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\:\varPhi\:}_{R},\;\:{y}_{k}^{\left(R\right)},\;{\:MSE}_{k}^{\left(R\right)}\;and\;{NMSE}_{k}^{\left(R\right)}$$\end{document}. The pilot sequences assigned to different users are mutually orthogonal, which suppresses inter-user pilot interference during channel estimation. In the LS derivation section, the channel is assumed constant within each pilot block; hence the time index (t) is omitted. In the Kalman filter section, the index (t) is introduced to model the temporal evolution of the channel across consecutive blocks. Since the ES protocol follows a similar linear observation structure as the TS protocol but with power-splitting and coupled phase-shift constraints, we briefly summarize the LS estimation and error analysis for completeness and highlight only the key differences specific to the ES signal model. STAR-RIS channel estimation under energy splitting for multi-user vehicular communication In this part, we introduce the channel estimation scheme for the ES protocol. Under ES, every STAR-RIS element concurrently forwards part of the incident energy to the transmission side and part to the reflection side; all K vehicular users transmit pilots simultaneously, and the BS (with m antennas) receives a superposition of direct and cascaded signals. We present the per-slot signal model, the stacked LS estimator, the MSE/error covariance, and the Kalman filter used to track Jakes time variations. Transmission and reflection model With an energy splitting ratio of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\beta\:}_{m}^{t},\:{\beta\:}_{m}^{r}\:\in\:\left[\mathrm{0,1}\right]$$\end{document}, the signals impinged on each subsurface are divided into transmitted and reflected ones for the ES protocols. The number of time periods for channel estimate is represented by τ. The transmission and reflection vectors at time slot i are therefore defined as follows: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\theta\:$$\end{document}iES \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\triangleq\:$$\end{document} [ \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\beta\:}_{1}^{t}e}^{j\theta_{1,i}\:}$$\end{document}, ………\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:,{\:\beta\:}_{M}^{t}{e}^{j\theta_{M,i}\:}$$\end{document} ], and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\phi\:$$\end{document}iES\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\triangleq\:$$\end{document} [\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{\:\beta\:}_{1}^{r}e}^{j\phi_{1,i}\:}$$\end{document}, ………\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:,{\:\beta\:}_{M}^{r}{e}^{j\phi_{M,i}\:}$$\end{document}], respectively. The law of energy conservation states that we have \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\beta\:}_{m}^{t}+\:{\:\beta\:}_{m}^{r}=1,\;m=1,\dots\:..,M$$\end{document} [4]. In this paper, we examine the practical coupled phase-shift model, in which the transmission and reflection phase shifts of each element, denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\theta\:}_{m,i}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\phi\:}_{m,i}$$\end{document}, are inherently coupled with each other. The electric and magnetic impedances of the equivalent circuits of a fully passive STAR-RIS are purely imaginary, which leads to the following constraint on the phase-shifts of each sub-surface19.21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\mathrm{cos}({\theta\:}_{m,i}-\:{\phi\:}_{m,i})\:=\:0,\;\;\;\;\;\;\;\;\forall\:m,i,$$\end{document} Signal model During the channel estimation stage, all k vehicular users continuously send pilot symbols to the BS, while the STAR-RIS training patterns are properly designed to assist the channel estimation of both direct and cascaded channels. For a fair comparison with the TS protocol, the total uplink transmit power per time slot is fixed to p. Since all users participate in the pilot transmission, the power is equally allocated by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}=\frac{p}{K}$$\end{document}, which satisfies the condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\sum\:_{k=1}^{K}{P}_{k}=p.$$\end{document} The baseband received signal at the BS at time slot i is expressed as:22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {y}_{k,i}^{ES} & =\sqrt{{P}_{k}}\:{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\:{\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\sqrt{\frac{{P}_{k}}{2}}\:{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\:\left(\mathrm{G}\left({t}_{i}\right)\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}\left({\theta\:}_{i}^{\left(ES,T\right)}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right) \\ & \quad +\sqrt{\frac{{P}_{k}}{2}}{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\left(\mathrm{G}\left({t}_{i}\right)\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}\left({\phi\:}_{i}^{(ES,R)}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right)+{n}_{k,i} \end{aligned}$$\end{document} Here p denotes the total uplink pilot power per slot, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}$$\end{document} is the per-user pilot power. Collect observations over τ pilot slots and stack per user unknowns.23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Y}_{k}^{ES}=[{y}_{(k,1)}^{ES},\:{y}_{(k,2)}^{ES},\:\dots\:..\:,\:{y}_{(k,\tau\:)}^{ES}{]}^{T}\in\:\:{C}^{\:(\tau\:\:\times\:\:1)}$$\end{document} Construct the combined ES pilot matrix. 24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}=[{\varPhi\:}_{d\:}\frac{1}{\sqrt{2}}{\varPhi\:}_{c}]$$\end{document} The construction of the DFT-based pilot matrix and the stacking procedure follow the same principles as in the TS protocol, and are summarized here only to highlight the ES-specific power-splitting structure. Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{d\:}$$\end{document} contains DFT row for the direct channel, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{c}$$\end{document} contains DFT rows multiplied by RIS patterns for the cascaded channel, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{1}{\sqrt{2}}$$\end{document} for ES power split, build the pilot matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}\in\:\:{C}^{\tau\:\:\times\:\:\left(m+M\right)}$$\end{document} from the known pilot symbols \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{s}_{k,i}$$\end{document} and the RIS patterns \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\{{\theta\:}_{i}^{\left(ES,T\right)},\:{\phi\:}_{i}^{\left(ES,R\right)}{\}}_{i=1}^{{\uptau\:}}$$\end{document}(DFT-based rows recommended). Then the stacked model for user k (or the per-user columns of the multi-user stacking) is25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Y}_{k}^{ES}=\:\sqrt{\frac{{P}_{k}}{2}}\:{\varPhi\:}_{ES}{x}_{k}+\:{N}_{k}$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Y}_{k}^{ES}\:\in\:\:{C}^{\:(\tau\:\:\times\:\:1)}$$\end{document} stacks the BS observations associated (after pilot separation/matching) with user k, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{N}_{k}$$\end{document} is the stacked noise vector. For simultaneous multi-user pilots the full stacking can be arranged as a block form as, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{Y}^{ES}=\:\sqrt{P/2}\:{\varPhi\:}_{ES}{X}^{ES}+\:N$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}\in\:{C}^{\tau\:\:\times\:\:\left(m+M\right)}$$\end{document} is the pilot/RIS design matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{X}^{ES}=[$$\end{document} x1,….,xK] \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:{C}^{\:\left(m+M\right)\:\times\:\:K}$$\end{document} stacks the per-user composite channel vectors as columns, N is the additive noise matrix. Least-squares (LS) estimator for ES Since the ES observation model in Eq. (25) has the same linear structure as the TS model, the LS estimation follows directly from the derivations in Section III, and only the ES-specific closed-form solution is summarized below. Assume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}$$\end{document} (or the user-specific columns \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES,k})$$\end{document} has full column rank and τ ≥ dim (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document}) = m + M. The closed-form LS estimator is given by26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{x}}_{k}^{(LS,\:\:ES)}=\sqrt{\frac{2}{{P}_{k}}}\:\left({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}\right)^{-1}\:\left({\varPhi\:}_{ES}^{H}\right)\:{Y}_{k}^{ES}$$\end{document} Estimation error and error covariance (ES) Following the same error analysis as in the TS case, the LS estimation error and the corresponding error covariance for the ES protocol are obtained as27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{e}_{k}^{\left(ES\right)}={\widehat{x}}_{k}^{(LS,\:ES)}-\:{x}_{k}=\sqrt{\frac{2}{{P}_{k}}}\:\left({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}\right)^{-1}\left({\varPhi\:}_{ES}^{H}\right)\:{N}_{k}$$\end{document} and28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{err,k}^{ES}=E\left\{{e}_{k}^{\left(ES\right)}{(e}_{k}^{\left(ES\right)}{)}^{H}\right\}=\frac{{2\sigma\:}^{2}}{{P}_{k}}\:({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}$$\end{document} MSE and normalized MSE (NMSE) for ES Using the same definition as in the TS case, the MSE of the LS estimator under ES is given by the trace of the error covariance as29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{ES}=Tr\left({C}_{err,k}^{ES}\right)=\:\:\:\frac{{2\sigma\:}^{2}}{{P}_{k}}\:Tr\left[\right({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}\:]$$\end{document} The Normalized MSE is obtained by normalizing with respect to the average channel energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:E\left[\left\|{x}_{k}\right\|_{2}^{2}\right]$$\end{document}:30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{ES}=\frac{\:\:\:\:{2\sigma\:}^{2}Tr\left[\right({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}\:]}{{P}_{k}E\left[\left\|{x}_{k}\right\|_{2}^{2}\right]}$$\end{document} Orthogonal DFT pilots (ES-optimal design) Similar to the TS case, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}_{ES}$$\end{document} is orthogonal with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}=\:\tau\:{I}_{m+M}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:\tau\:\ge\:m+M$$\end{document}, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:(\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}=\frac{1}{\tau\:}{I}_{m+M}$$\end{document}, and the MSE simplifies as31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{ES}=\:\frac{{2\sigma\:}^{2}}{{P}_{k}}\:\frac{m+M}{\tau\:}$$\end{document} This corresponds to the minimum LS-MSE under orthogonal pilots when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:= m+M,$$\end{document} 32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{MSE}_{k}^{ES}=\:\frac{{2\sigma\:}^{2}}{{P}_{k}}$$\end{document} The corresponding NMSE is33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{ES}=\frac{\:\:\:2{\sigma\:}^{2}}{{\:P}_{k}E\left[\left\|{x}_{k}\right\|_{2}^{2}\right]}$$\end{document} This expression holds under orthogonal DFT-based pilots with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:=m+M$$\end{document}. Kalman filter for time-varying (Jakes) channels To track the Jakes time variations across pilot frames, the channel dynamics and measurements are cast into a state-space form and processed using a Kalman filter. State model (common for TS and ES) For the composite channel vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}\left(\mathrm{t}\right)$$\end{document} of user k at discrete pilot frame index t, we use a first-order auto-regressive approximation consistent with the Jakes temporal correlation:34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}\left(\mathrm{t}+1\right)=\:{F}_{k}{x}_{k}\left(\mathrm{t}\right)+\:{w}_{k}\left(\mathrm{t}\right)$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{F}_{k}=\:{\alpha\:}_{k}{I}_{m+M}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\alpha\:}_{k}={J}_{0}\left(2\pi\:{f}_{D,k}\varDelta\:t\right)$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{w}_{k}\left(t\right)\sim\:CN(0,{Q}_{k})$$\end{document} is the process noise, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} is the Doppler frequency associated with user k and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\varDelta\:t$$\end{document} is the interval between pilot frames. It should be noted that the state transition matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{F}_{k}$$\end{document} is constructed using Jakes temporal correlation, which treats the channel statistics as stationary over consecutive pilot frames. In practice, as vehicles move along their paths, the geometry of the scattering environment changes, causing the AoA and AoD to shift gradually over time. For short to moderate observation windows, this stationarity assumption holds reasonably well, as the angular changes between consecutive frames are small relative to the Doppler-induced temporal variations. However, in prolonged high-speed trajectories, the gradual evolution of spatial angles could eventually cause the Kalman tracker to diverge if the state model is not updated. Incorporating geometry-based AoA/AoD tracking into the state transition model is an interesting extension that we leave for future work. Measurement model (LS per-frame \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\to\:$$\end{document} Kalman measurement) We use the frame-wise LS estimate as the measurement for the Kalman filter (two-stage approach). The LS outputs differ between TS and ES because of the transmit power splitting and pilot structure. Time switching (TS) measurement Per frame t the LS measurement (stacked over \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} slots) for user k is,35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{z}}_{k}^{\left(T\right)}\left(t\right)\triangleq\:\:{\widehat{\mathrm{x}}}_{\mathrm{k}}^{(\mathrm{L}\mathrm{S},\mathrm{T})}\left(t\right)=\:{x}_{k}\left(\mathrm{t}\right)+\:{\mathrm{v}}_{k}^{\left(T\right)}\left(t\right)\:$$\end{document} With measurement noise36\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{v}}_{k}^{\left(T\right)}\left(t\right)\sim(0,{R}_{k}^{\left(T\right)}\left(t\right)),\;\;\;\;\;\;\;\;{\mathrm{R}}_{k}^{\left(T\right)}\left(t\right)=\:\frac{{\sigma\:}^{2}}{{P}_{k}}\:\left({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}\right)^{-1}$$\end{document} Energy splitting (ES) measurement Per frame t the LS measurement for user k is,37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{z}}_{k}^{ES}\left(t\right)\triangleq\:\:{\widehat{x}}_{k}^{(LS,ES)}\left(t\right)=\:{x}_{k}\left(\mathrm{t}\right)+\:{v}_{k}^{ES}\left(t\right)\:\:\:\:\:$$\end{document}38\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{v}_{k}^{ES}\left(t\right)\sim\:CN(0,{R}_{k}^{ES}(t\left)\right),\;\;\;\;\;\;{\mathrm{R}}_{k}^{ES}\left(t\right)=\:\frac{{2\sigma\:}^{2}}{{P}_{k}}\:({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}$$\end{document} Kalman recursions Predict (common) 39\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t-1\right)=\:{F}_{k}{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t-1|t-1\right)$$\end{document} Predict (common) Predict (common) In this work, the Doppler frequency \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} is assumed to be known or accurately estimated at the BS based on vehicle speed and carrier frequency, which is a common assumption in vehicular communication systems. Small Doppler estimation errors are absorbed into the process noise covariance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{Q}_{k}$$\end{document}. It is worth discussing, however, what happens when there is a more significant mismatch between the assumed and actual Doppler frequency. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} is overestimated, the state transition coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{k}$$\end{document} becomes smaller than it should be, causing the filter to treat the channel as changing faster than it actually is, which leads to a higher process noise assumption and slightly degraded NMSE. Conversely, if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} is underestimated, the filter assumes stronger temporal correlation than actually exists, which can cause it to trust its predictions too much and respond slowly to actual channel changes. In both cases, the Kalman filter does not diverge immediately - the process noise covariance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{Q}_{k}$$\end{document} acts as a regularization buffer that absorbs moderate mismatches. For larger mismatches at very high vehicle speeds, adaptive Doppler tracking or joint Doppler-channel estimation would be a more robust solution, which we identify as a useful direction for future work.40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}\left(t|t-1\right)=\:{F}_{k}{P}_{k}\left(t-1|t-1\right){F}_{k}^{H}+\:{Q}_{k}$$\end{document} In practice, Doppler can be estimated from vehicle speed and carrier frequency or tracked using standard frequency-offset estimation methods, and small mismatch errors are absorbed into the process noise covariance. b) Update - Time Switching (TS) b) Update - Time Switching (TS) Update - Time Switching (TS) Measurement covariance: 41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{R}}_{k}^{\left(T\right)}\left(t\right)=\:\frac{{\sigma\:}^{2}}{{P}_{k}}\:\left({\varPhi\:}_{T}^{H}{\varPhi\:}_{T}\right)^{-1}$$\end{document} Kalman gain:42\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{K}_{k}^{\left(T\right)}\left(t\right)=P}_{k}\left(t|t-1\right)\:({P}_{k}\left(t|t-1\right)+\:{\mathrm{R}}_{k}^{\left(T\right)}\left(t\right){)}^{-1}$$\end{document} State update:43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t\right)=\:{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t-1\right)+\:{K}_{k}^{\left(T\right)}\left(t\right)({\mathrm{z}}_{k}^{\left(T\right)}\left(t\right)-{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t-1\right))$$\end{document} Covariance update (Joseph form – numerically stable and includes measurement noise):44\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}\left(t|t\right)=(I-\:{{K}_{k}^{\left(T\right)}\left(t\right))\:P}_{k}\left(t|t-1\right)(I-{K}_{k}^{\left(T\right)}\left(t\right))^{H}\:+\:{K}_{k}^{\left(T\right)}\left(t\right){R}_{k}^{\left(T\right)}\left(t\right){(K}_{k}^{\left(T\right)}\left(t\right))^{H}$$\end{document} c) Update - Energy Splitting (ES) c) Update - Energy Splitting (ES) Update - Energy Splitting (ES) Measurement covariance:45\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{R}}_{k}^{ES}\left(t\right)=\:\frac{{2\sigma\:}^{2}}{{P}_{k}}\:({\varPhi\:}_{ES}^{H}{\varPhi\:}_{ES}{)}^{-1}$$\end{document} Kalman gain:46\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{K}_{k}^{ES}\left(t\right)=P}_{k}\left(t|t-1\right)\:({P}_{k}\left(t|t-1\right)+{\mathrm{R}}_{k}^{ES}\left(t\right))^{-1}$$\end{document} State update:47\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t\right)=\:{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t-1\right)+\:{K}_{k}^{ES}\left(t\right)({\mathrm{z}}_{k}^{ES}\left(t\right)-{\widehat{\mathrm{x}}}_{\mathrm{k}}\left(t|t-1\right))$$\end{document} Covariance update:48\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}\left(t|t\right)=(I-\:{{K}_{k}^{ES}\left(t\right))\:P}_{k}\left(t|t-1\right)(I-{K}_{k}^{ES}\left(t\right))^{H}\:+\:{K}_{k}^{ES}\left(t\right){\mathrm{R}}_{k}^{ES}\left(t\right)({K}_{k}^{ES}\left(t\right))^{H}$$\end{document} Here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}(.)$$\end{document} is the state estimation covariance and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{K}_{k}^{ES}\left(t\right),\;{K}_{k}^{\left(T\right)}\left(t\right)$$\end{document} are the Kalman gain. Because the measurement covariance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{k}^{\left(T\right)}\left(t\right)$$\end{document} is known from the LS error expression, the Kalman update is optimal (minimum MSE) under the Gaussian linear assumptions. In summary, the proposed STAR-RIS-assisted framework employs both TS and ES protocols for uplink channel estimation in multi-user vehicular systems. Under TS, the STAR-RIS alternates between transmission and reflection modes, whereas under ES each sub-surface simultaneously forwards energy to both sides. In both cases, the frame-wise LS estimator provides unbiased measurements of the composite channels, and the Kalman filter exploits the Jakes temporal correlation to track their evolution across pilot frames. This two-stage LS-Kalman approach enables accurate tracking of fast time-varying channels in STAR-RIS-aided multi-user vehicular environments. Computational complexity analysis This section examines the computational complexity of the proposed LS-Kalman channel estimation framework for STAR-RIS-assisted multi-user vehicular communication systems. The complexity is evaluated in terms of the number of vehicular users K, the number of BS antennas m, the number of STAR-RIS sub-surfaces M, and the pilot length. Least squares (LS) estimation complexity For both TS and ES protocols, the LS estimator is applied to estimate the composite channel vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\in\:\:{C}^{\left(m\:+\:M\right)\:\times\:1}$$\end{document}. The dominant operations in LS estimation include: Computation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}^{H}\varPhi\:$$\end{document}: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!\left(\tau (m+M)^2\right)$$\end{document}.Matrix inversion of size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(m+M)\times(m+M) : O\!\left((m+M)^3\right)$$\end{document}.Multiplication with the received pilot vector: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!(\tau\:\left(m+M)\right) $$\end{document}. Computation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}^{H}\varPhi\:$$\end{document}: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!\left(\tau (m+M)^2\right)$$\end{document}. Computation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}^{H}\varPhi\:$$\end{document}: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!\left(\tau (m+M)^2\right)$$\end{document}. Matrix inversion of size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(m+M)\times(m+M) : O\!\left((m+M)^3\right)$$\end{document}. Matrix inversion of size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(m+M)\times(m+M) : O\!\left((m+M)^3\right)$$\end{document}. Multiplication with the received pilot vector: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!(\tau\:\left(m+M)\right) $$\end{document}. Multiplication with the received pilot vector: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\!(\tau\:\left(m+M)\right) $$\end{document}. Hence, the overall LS estimation complexity per user is dominated by the matrix inversion and can be expressed as49\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^3\right)$$\end{document} For a multi-user scenario with K vehicular users, the total LS estimation complexity scales as50\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(K(m+M)^3\right)$$\end{document} It is worth noting that, due to the use of fixed DFT-based pilot matrices, the Gram matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\varPhi\:}^{H}\varPhi\:$$\end{document} can be precomputed offline, which significantly reduces the real-time computational burden at the base station. Kalman filter complexity To track the time-varying channels induced by vehicular mobility, a Kalman filter is employed for each user based on the Jakes temporal correlation model. The Kalman filter operates on the same composite channel vector of dimension (m + M). The dominant operations per Kalman update include: State prediction and covariance propagation: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}.Kalman gain computation (involving matrix inversion): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^3\right)$$\end{document}.State update and covariance update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}. State prediction and covariance propagation: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}. State prediction and covariance propagation: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}. Kalman gain computation (involving matrix inversion): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^3\right)$$\end{document}. Kalman gain computation (involving matrix inversion): \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^3\right)$$\end{document}. State update and covariance update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}. State update and covariance update: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^2\right)$$\end{document}. Therefore, the computational complexity of the Kalman filter per user per frame is51\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left((m+M)^3\right)$$\end{document} For K users, the overall Kalman tracking complexity becomes52\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(K(m+M)^3\right)$$\end{document} Overall computational complexity and practical feasibility By combining LS estimation and Kalman-based channel tracking, the total computational complexity per frame of the proposed framework scales as53\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:O\left(K(m+M)^3\right)$$\end{document} Compared to conventional LS-only channel estimation, the proposed LS-Kalman framework introduces additional computational overhead due to the Kalman filtering stage. This additional overhead is justified by the significant NMSE reduction and the improved tracking capability for time-varying channels observed in the simulation results. Furthermore, the element-grouping strategy (M < < N) significantly reduces the effective channel dimension, making the proposed approach computationally feasible for real-time implementation. Since channel estimation is performed at the BS, which typically has sufficient processing capability, the proposed framework remains scalable for moderate numbers of users and STAR-RIS sub-surfaces in practical vehicular communication systems. Complexity comparison between TS and ES protocols The LS and Kalman filtering operations under the TS and ES protocols share the same computational structure. The primary difference between the two protocols lies in the measurement noise covariance matrices due to power splitting in the ES protocol, which does not affect the asymptotic computational complexity. As a result, both TS and ES protocols exhibit identical computational complexity orders, while their performance differences stem from protocol-specific power allocation and signal coupling effects rather than computational cost. Results and discussion Simulation setup In this portion, we present numerical results to verify the effectiveness of the suggested channel estimation schemes for both Time Switching (TS) and Energy Splitting (ES) protocols in a vehicular STAR-RIS-assisted communication system. The system consists of K = 4 single-antenna vehicular users as the baseline configuration, and additional simulations are conducted by varying the number of users from K = 2 to K = 16 to evaluate the impact of multi-user interference on channel estimation performance. A STAR-RIS is equipped with N=80 passive elements, which are grouped into M = 20 sub-surfaces, each sharing a common transmission and reflection coefficient. All wireless channels are modeled as Rician fading to incorporate both line-of-sight (LOS) and non-line-of-sight (NLOS) components. The Rician factor is set to 10 dB, which indicates a moderately strong LOS component. The NLOS temporal variation of each channel coefficient follows the Jakes temporal correlation, represented as a sum of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{N}_{s}=8$$\end{document} sinusoids with independent random phases. This combined Rician-Jakes channel model captures both the deterministic LOS paths and the time-correlated multipath fading due to vehicular mobility. In a 2 GHz carrier band, the maximum Doppler frequency is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D}=300$$\end{document} Hz, which corresponds to typical highway vehicle speeds, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D}$$\end{document} denotes the maximum Doppler frequency and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D,k}$$\end{document} denotes the user-specific Doppler frequency. The sampling interval is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{T}_{s}=1$$\end{document} ms, and all results are averaged over 2000 Monte Carlo iterations for statistical accuracy. The noise power (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\sigma\:}^{2})$$\end{document} is -110 dBm. The maximum transmit power of each user is set to P = 30 dBm. To examine NMSE trends under various power settings, the transmit power is varied from 0 dBm to 40 dBm. The TS protocol’s pilot length, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}=m+M=24,$$\end{document} guaranteeing orthogonal DFT-based pilot sequences. The distance-dependent path loss is modeled as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:L(d)={\beta\:}_{0}(d/{d}_{0}{)}^{-\alpha\:}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{0}=\:-30$$\end{document} dB represents the path loss at the reference distance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{d}_{0}=1\;m,\:$$\end{document} d represents the link distance, and α is the path-loss exponent. We consider a two-dimensional coordinate system in which the STAR-RIS center is located at (50 m, 0), and the Base Station (BS) is located at the origin (0, 0). Two representative vehicular users, one on the transmission side (T-user) and one on the reflection side (R-user), are located at (54 m, 3 m) and (46 m, -3 m), respectively. The user-BS, user-STAR-RIS, and BS-STAR-RIS channels have path-loss exponents of 3.5, 2.8, and 2.2 respectively, which correspond to normal urban vehicle conditions. At the considered Doppler frequency and sampling interval, the channel coherence time is on the order of several milliseconds, and the chosen pilot length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} occupies only a small fraction of the coherence interval, ensuring low training overhead even at high vehicular speeds. All simulation parameters and estimation procedures were implemented in MATLAB, and the results were obtained using Monte Carlo averaging over independent channel realizations. We compare the performance of the proposed channel estimation schemes against two benchmark methods for fair evaluation: ON/OFF Scheme for TS: The direct channels are estimated by turning off all STAR-RIS sub-surfaces, following the training procedure described in20. Subsequently, the cascaded channels are obtained by sequentially activating one sub-surface at a time in either transmission or reflection mode.Two-Phase Estimation Schemes for ES: In this method, the direct and cascaded links are estimated in two separate phases. In the first phase, the users transmit orthogonal pilot sequences while all STAR-RIS sub-surfaces remain inactive to estimate the direct channels. In the second phase, the transmission and reflection phase-shift patterns are configured according to the ES protocol to estimate the cascaded channels. ON/OFF Scheme for TS: The direct channels are estimated by turning off all STAR-RIS sub-surfaces, following the training procedure described in20. Subsequently, the cascaded channels are obtained by sequentially activating one sub-surface at a time in either transmission or reflection mode. ON/OFF Scheme for TS: The direct channels are estimated by turning off all STAR-RIS sub-surfaces, following the training procedure described in20. Subsequently, the cascaded channels are obtained by sequentially activating one sub-surface at a time in either transmission or reflection mode. Two-Phase Estimation Schemes for ES: In this method, the direct and cascaded links are estimated in two separate phases. In the first phase, the users transmit orthogonal pilot sequences while all STAR-RIS sub-surfaces remain inactive to estimate the direct channels. In the second phase, the transmission and reflection phase-shift patterns are configured according to the ES protocol to estimate the cascaded channels. Two-Phase Estimation Schemes for ES: In this method, the direct and cascaded links are estimated in two separate phases. In the first phase, the users transmit orthogonal pilot sequences while all STAR-RIS sub-surfaces remain inactive to estimate the direct channels. In the second phase, the transmission and reflection phase-shift patterns are configured according to the ES protocol to estimate the cascaded channels. It is important to clarify that the ON/OFF scheme in20 and the two-phase ES estimator adopted in this work are not newly proposed methods, but representative baseline schemes originally introduced for STAR-RIS channel estimation under the TS and ES protocols, respectively, as reported in13 and related works. In this paper, these benchmark schemes are employed to provide a fair and consistent reference for performance comparison under the same training protocols and system assumptions. Different from13, which focuses on static or slowly varying scenarios with limited user configurations, this work extends these baseline estimators to a high-mobility multi-user uplink vehicular framework with time-varying Rician fading and Jakes temporal correlation, and integrates Kalman filtering to track Doppler-induced channel variations. In addition, this paper evaluates their impact not only on NMSE but also on achievable uplink channel capacity under realistic vehicular mobility, which is not investigated in13. For fair benchmarking, the numerical results of the reference TS ON/OFF and ES two-phase estimation schemes are reproduced from the base paper under the same simulation parameters. The proposed LS-Kalman framework is simulated under identical settings for direct performance comparison. It is worth noting that most existing compressed-sensing-based and learning-based channel estimation methods for RIS or STAR-RIS are developed for downlink transmission, static or slowly varying channels, or single-user scenarios, and typically rely on sparse angular-domain channel models and long coherence times. In high-mobility vehicular uplink systems with fast time variations and short coherence intervals, these approaches require prohibitive training overhead and frequent retraining, and are not directly applicable to the considered multi-user LS-Kalman tracking framework. Therefore, in this work we focus on representative training-based STAR-RIS baseline schemes under the TS and ES protocols to ensure a fair and practically meaningful comparison. Performance measures NMSE analysis To evaluate the estimation accuracy of the proposed STAR-RIS-aided vehicular communication framework, the normalized mean square error (NMSE) is used as the performance metric, defined as54\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}=\frac{E\left[{\left\| {\widehat{\mathrm{x}}}_{\mathrm{k}}-{x}_{k}\right\| }^{2}\right]}{E\left[{\left\| {x}_{k}\right\| }^{2}\right]}$$\end{document} Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{\mathrm{x}}}_{\mathrm{k}}$$\end{document} represents the estimated composite channel vector obtained from the LS or Kalman filter, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document} denotes the true channel vector for user k. For the Time Switching (TS) protocol, the NMSE is55\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{\left(T\right)}=\frac{{\sigma\:}^{2}}{{P}_{k}E\left[{\left\| {x}_{k}\right\| }^{2}\right]}$$\end{document} For the Energy Splitting (ES) protocol, the NMSE is56\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{NMSE}_{k}^{ES}=\frac{2{\sigma\:}^{2}}{{P}_{k}E\left[{ \left\| {x}_{k}\right\| }^{2}\right]}$$\end{document} The factor of two indicates that ES suffers approximately a 3 dB degradation compared to TS due to power splitting between transmission and reflection. By exploiting the Jakes temporal correlation through Kalman filtering, the NMSE is further reduced, especially at low Doppler frequencies, where the temporal correlation parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{k}=$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{J}_{0}\left(2\pi\:{f}_{D,k}\varDelta\:t\right)$$\end{document} is high. Channel capacity analysis In this section, we evaluate the achievable uplink channel capacity of the proposed STAR-RIS-aided vehicular communication system under imperfect channel estimation. The estimated composite channel obtained from the LS-Kalman framework is utilized to compute the ergodic achievable rate, while accounting for the residual estimation error characterized by the NMSE results derived in Section IV. Using the worst-case uncorrelated estimation error approximation, the instantaneous achievable uplink rate is expressed as57\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{k}=E\left[\log_{2}\left(1+\frac{{P}_{k}\left|\widehat{h}_{k}\right|^{2}}{{\sigma}^{2}+{P}_{k}E\left[\left|h_{k}-\widehat{h}_{k}\right|^{2}\right]}\right)\right]$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\widehat{h}}_{\mathrm{k}}$$\end{document} is the estimated composite channel (direct and STAR-RIS-assisted), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}$$\end{document} is the transmit power of user k, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\sigma\:}^{2}$$\end{document} denotes the noise variance. Both thermal noise and the residual estimation error power, which is directly related to the NMSE, are included in the denominator. Hence, more accurate channel estimation (lower NMSE) results in a higher effective SNR and improved system throughput. Capacity under the time switching (TS) protocol From system model, the received signal during TS is given by58\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{y}_{k,i}^{\left(T\right)}=\sqrt{{\mathrm{P}}_{\mathrm{k}}}\left({\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\mathrm{G}\left({t}_{i}\right)diag\left({\theta\:}_{i}^{TS}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right){\mathrm{s}}_{\mathrm{k},\mathrm{i}}+{\mathrm{n}}_{\mathrm{k},\mathrm{i}}^{\left(\mathrm{T}\right)}$$\end{document} The effective composite channel is59\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{h}_{k}^{TS}\left({t}_{i}\right)={\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\mathrm{G}\left({t}_{i}\right)diag\left({\theta\:}_{i}^{TS}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)$$\end{document} Using the estimated channel and the corresponding NMSE of the LS-Kalman estimator in TS mode, the achievable capacity is given by60\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{k}^{TS}=E\left[\log_{2}\left(1+\frac{(1-NMSE_{k}^{TS})P_{k}}{\sigma^{2}+NMSE_{k}^{TS}P_{k}}\right)\right]$$\end{document} If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{T}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\tau\:}_{R}$$\end{document} denote the normalized time durations for transmission and reflection periods, respectively, the average TS capacity per frame is written as61\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{avg}^{TS}=\:\frac{{\tau\:}_{T}{C}_{T}\:+\:{\tau\:}_{R}{C}_{R}}{{\tau\:}_{T}+{\tau\:}_{R}}$$\end{document} Capacity under the energy splitting (ES) protocol Similarly, from system model, the received signal during ES is62\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {y}_{k,i}^{ES} & =\sqrt{{P}_{k}}\:{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\:{\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\sqrt{\frac{{P}_{k}}{2}}\:{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\:\left(\mathrm{G}\left({t}_{i}\right)\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}\left({\theta\:}_{i}^{\left(ES,T\right)}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right) \\ & \quad +\sqrt{\frac{{P}_{k}}{2}}{\mathrm{s}}_{\mathrm{k},\mathrm{i}}\left(\mathrm{G}\left({t}_{i}\right)\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{g}\left({\phi\:}_{i}^{(ES,R)}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right)+{n}_{k,i} \end{aligned}$$\end{document} The effective channel is,63\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {h}_{k}^{ES}\left({t}_{i}\right) & ={\mathrm{h}}_{\mathrm{k}}\left({t}_{i}\right)+\frac{1}{\sqrt{2}}\left(\mathrm{G}\left({t}_{i}\right)diag\left({\theta\:}_{i}^{\left(ES,T\right)}\right){\mathrm{r}}_{k}\left({t}_{i}\right)\right) \\ & \quad +\frac{1}{\sqrt{2}}\left(\mathrm{G}\left({t}_{i}\right)diag\left({\phi\:}_{i}^{(ES,R)}\right){\mathrm{r}}_{\mathrm{k}}\left({t}_{i}\right)\right) \end{aligned}$$\end{document} In general, using the composite channel average energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:E}_{h}^{ES}$$\end{document}, the achievable capacity is given by64\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{k}^{ES}=E\left[{\mathrm{log}}_{2} \left (1+\frac{(1-{NMSE}_{k}^{ES}){P}_{k}{E}_{h}^{ES}}{{\sigma\:}^{2}+{NMSE}_{k}^{ES}{P}_{k}{E}_{h}^{ES}}\right) \right]$$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{E}_{h}^{ES} = 1$$\end{document}, then the simplified form under unit energy normalization is written as65\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{C}_{k}^{ES}=E\left[{\mathrm{log}}_{2} \left (1+\frac{(1-{NMSE}_{k}^{ES})\frac{{P}_{k}}{2}}{{\sigma\:}^{2}+{NMSE}_{k}^{ES}\frac{{P}_{k}}{2}}\right) \right]$$\end{document} Due to power splitting between transmission and reflection paths, the effective received signal power in ES is lower than in TS for the same transmit power \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{P}_{k}$$\end{document}, which results in lower achievable capacity. Comparison analysis In this section, the NMSE performance under different energy splitting ratios \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}$$\end{document} is evaluated for the TS and ES Protocols in the considered STAR-RIS-assisted uplink system. The ES protocol is analyzed for the overall NMSE, as well as for the transmission side (T-user) and reflection side (R-user) users separately. Figure 2 illustrates the NMSE performance as the energy splitting ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}$$\end{document} varies from 0.2 to 0.8. For the ES protocol, the overall NMSE attains its lowest value of 3.18645 × 10⁻⁸ at the balanced splitting point \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}$$\end{document} = 0.5. At this operating point, the NMSE values of the transmission-side user and the reflection-side user are nearly identical, indicating that both links contribute almost equally to the overall estimation error. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}$$\end{document} increases, the ES T-user decreases due to higher transmission energy, whereas the ES R-user NMSE increases because of reduced reflection energy. At higher levels of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\:\beta\:}_{t}$$\end{document}, this trade-off becomes more noticeable , with the T-user NMSE dropping to 1.9945 × 10− 8 and the R-user NMSE reaches 7.9785 × 10− 8 at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.8.$$\end{document} The TS protocol consistently outperforms the ES protocol over the entire range of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}.$$\end{document} At \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.5,$$\end{document} TS attains an NMSE of 7.97296 × 10− 9, which is significantly lower than the ES performance. This enhancement results from TS use of orthogonal time allocation for both transmission and reflection, avoiding the interference coupling present in ES. TS consistently provides better estimation accuracy, although ES performs best at the balanced split \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.5.$$\end{document} The selected benchmark schemes represent the standard training-based channel estimation methods for STAR-RIS under TS and ES protocols, enabling a direct and fair performance comparison with the proposed LS-Kalman framework. Although the balanced split \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}\:$$\end{document}= 0.5 is optimal in the considered symmetric geometry, in asymmetric deployments with different path losses on the transmission and reflection sides, the optimal splitting ratio is expected to shift toward the stronger link, which can be addressed by adaptive \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}\:$$\end{document} optimization in future work. Fig. 2NMSE under different energy splitting ratios. NMSE under different energy splitting ratios. Figure 3 displays the NMSE performance of several TS and ES systems as the total transmit power increases from 0 to 40 dBm. The TS protocol outperforms the TS ON/OFF method in terms of estimation accuracy. TS reaches an NMSE of 1.097 × 10− 11 at 20 dBm, whereas TS ON/OFF reaches 2.3743 × 10− 11, respectively. The ES ideal phase shift and two-phase methods overlap with each other and produce the same NMSE values (e.g., 3.7566 × 10− 8 at 20 dBm), representing the best performance among ES-based schemes. Phase quantization and non-ideal hardware effects cause the ES Practical Phase Shift configuration to perform poorly, with a larger NMSE of 1.3251 × 10− 7 at 20 dBm. While ES ideal and ES Two-phase provide the best performance among ES variants, TS and TS ON/OFF consistently outperform all ES schemes over the whole power range. Fig. 3NMSE performance versus the total transmit power under different channel estimation schemes. NMSE performance versus the total transmit power under different channel estimation schemes. Figure 4 shows the NMSE performance as the number of STAR-RIS sub-surfaces M increases from 10 to 100. For all values of M, the TS method consistently produces the lowest estimation error. TS reaches an NMSE of 9.3558 × 10− 9 at M = 60, however TS ON/OFF retains a little greater error at 1.9103 × 10− 8. Both TS-based methods maintain almost constant NMSE as M increases. This behavior occurs because the pilot length is scaled with the composite channel dimension under orthogonal DFT pilot design, which keeps the LS estimation error variance approximately unchanged as the number of STAR-RIS sub-surfaces increases. The ES Ideal Phase Shift and Two-Phase curves totally overlap with each other, resulting in same NMSE values 1.5654 × 10− 5 at M = 60. Due to hardware imperfections, the ES Practical Phase Shift scheme shows notably higher error, resulting in an NMSE of 4.5212 × 10− 5 at M = 60. In contrast to TS, ES schemes only exhibit a slight reduction in NMSE for the Two-Phase/Ideal configuration, while the practical ES model remains almost flat with high error. For all values of M, TS and TS ON/OFF perform better than all ES versions overall, proving their superior estimation accuracy and robustness even as the number of STAR-RIS sub-surfaces rises. Fig. 4NMSE performance versus the number of STAR-RIS sub-surfaces M under different channel estimation schemes. NMSE performance versus the number of STAR-RIS sub-surfaces M under different channel estimation schemes. Figure 5 shows the TS and ES methods NMSE performance at different vehicle speeds ranging from 0 to 160 km/h. The NMSE values of all schemes remain almost unchanged, showing that the estimator is resilient to changes in mobility. At 80 km/h, TS-based schemes achieve an NMSE of 1.1004 × 10− 9, whereas the TS ON/OFF schemes attain an NMSE of 2.3536 × 10− 9. The Ideal Phase Shift and Two-Phase designs for ES-based approaches overlap, achieving 3.5373 × 10− 6; however, the Practical Phase Shift model has a greater inaccuracy at 1.3498 × 10− 5. This result shows that the proposed estimation schemes, particularly the TS protocol, are highly robust to Doppler changes. This behavior mainly results from incorporating the Jakes temporal correlation model into the state transition matrix of the Kalman filter, which allows accurate tracking of time-varying channels. It is also worth noting the specific contribution of the Kalman tracking step compared to plain frame-by-frame LS estimation. At \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{D}$$\end{document} = 300 Hz and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{T}_{s}$$\end{document} = 1 ms, the temporal correlation coefficient is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\alpha\:}_{k}\:={J}_{0}(2\pi\:\:\times\:\:300\:\times\:\:0.001)\:\approx\:\:0.27,$$\end{document} meaning the channel changes considerably between consecutive pilot frames. At such high Doppler, a plain LS estimator applied independently at each frame would struggle to maintain stable accuracy as vehicle speed increases, since it has no mechanism to track or compensate for the rapid channel variations. The Kalman filter addresses this by exploiting the residual temporal structure through the process noise covariance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{Q}_{k}$$\end{document}, which regularizes the estimate across frames and prevents error accumulation. The consistently flat NMSE curves observed across all vehicle speeds in Fig. 5 are a direct result of this tracking mechanism, and would not be achievable with plain LS estimation alone. The Kalman filter exploits the temporal correlation between successive channel states to track Doppler variations and preserve estimation accuracy at high vehicle speeds. Hence, the estimation accuracy remains stable even under high vehicular mobility. Fig. 5NMSE performance versus vehicle speed under different channel estimation schemes. NMSE performance versus vehicle speed under different channel estimation schemes. Figure 6 shows the TS and ES schemes NMSE performance as the pilot length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:$$\end{document} increases from 20 to 90. The TS protocol attains the lowest estimation error across all pilot lengths, with an NMSE of 6.5786 × 10− 10 at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:=60.$$\end{document} TS ON/OFF also attains low error but remains slightly higher than TS, recording 1.3889 × 10− 9 at the same pilot length. Both TS-based schemes show gradual NMSE improvement as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:\:$$\end{document} increases. For the ES schemes, the Ideal Phase Shift and Two-Phase curves overlap each other and show identical performance, attaining 3.7707 × 10− 6 at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\tau\:=60.$$\end{document} The ES Practical Phase Shift configuration demonstrates the highest NMSE, with a value of 1.479 × 10− 5 due to phase errors caused by hardware imperfections. Increasing the pilot length results in only a small progress for ES schemes, and their NMSE remains significantly higher than TS-based methods. For all pilot lengths, TS and TS ON/OFF provide superior estimation accuracy for all pilot lengths, highlighting the efficiency of TS-based training compared to ES schemes. Fig. 6NMSE performance versus pilot length under different channel estimation schemes. NMSE performance versus pilot length under different channel estimation schemes. Figure 7 presents the NMSE performance of the TS and ES schemes as the number of users K increases from 2 to 16. The TS protocol achieves the lowest estimation error over the entire range of user loads, indicating strong resistance to multi-user interference. At K = 10, the TS scheme attains an NMSE of 1.1066 × 10⁻¹¹, which is the lowest among all considered methods. The TS ON/OFF benchmark also provides low estimation error but remains slightly higher than TS, with an NMSE of 2.37156 × 10⁻¹¹ at K = 10. Both TS-based schemes exhibit nearly constant NMSE values as K increases, showing that their estimation performance is only weakly affected by the number of users. This trend results from the use of DFT-based orthogonal pilot sequences, which suppress inter-user pilot interference during channel estimation. For the ES schemes, the Ideal Phase Shift and Two-Phase configurations coincide and achieve an NMSE of 3.6448 × 10⁻⁸ at K = 10. This result reflects the performance limitation imposed by power splitting, which reduces the effective pilot energy and limits further improvement even under ideal phase-shift design. The ES Practical Phase Shift scheme records the highest estimation error, with an NMSE of 1.3763 × 10⁻⁷ at K = 10, which is mainly caused by phase quantization effects and hardware non-idealities at the STAR-RIS. Overall, the NMSE varies only slightly with K for all schemes, indicating stable behavior in multi-user conditions. Nevertheless, for all values of K, the TS and TS ON/OFF schemes consistently outperform the ES-based methods, confirming the advantage of TS-based training in multi-user vehicular STAR-RIS channel estimation. Fig. 7NMSE performance versus the number of users K under different channel estimation schemes. NMSE performance versus the number of users K under different channel estimation schemes. Figure 8 shows the average channel capacity of the TS and ES schemes as the energy splitting ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}$$\end{document} varies from 0.2 to 0.8. The TS protocol achieves the highest and most stable capacity across the entire range, reaching 16.309 bps/Hz at the balanced point \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.5.$$\end{document} The total capacity of the ES protocol peaks at 14.31 bps/Hz at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.5.$$\end{document} The ES T-user capacity increases with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:{\beta\:}_{t}$$\end{document} when transmission energy increases, while the ES R-user capacity drops when reflection energy decreases. The combined ES curve reflects this trade-off, achieving its maximum when the energy is equally split. Overall, TS provides significantly higher channel capacity than all ES configurations, and ES achieves its best performance at the balanced split \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\beta\:}_{t}=0.5$$\end{document}. Fig. 8Average channel capacity versus energy splitting ratio. Average channel capacity versus energy splitting ratio. Figure 9 shows the average channel capacity performance of the TS and ES schemes as the total transmit power increases from 0 to 40 dBm. The channel capacity increases for all schemes as the transmit power increases. The TS scheme achieves the highest channel capacity over the entire transmit power range. At 20 dBm, the TS scheme achieves a channel capacity of 36.3946 bps/Hz, while the TS ON/OFF scheme achieves 35.2876 bps/Hz. The ES Ideal Phase Shift and ES Two-Phase schemes overlap and achieve a capacity of 24.666 bps/Hz. The ES Practical Phase Shift scheme achieves a capacity of 22.8474 bps/Hz at 20 dBm. For all transmit power values, the TS and TS ON/OFF schemes show higher channel capacity than the ES schemes. Fig. 9Average channel capacity versus Total transmit power. Average channel capacity versus Total transmit power. The average channel capacity performance as the number of STAR-RIS sub-surfaces M increases from 10 to 100 is shown in Fig. 10. The TS and TS ON/OFF schemes achieve the highest and nearly identical channel capacity for all values of M. At M = 60, the TS scheme achieves a capacity of 16.3078 bps/Hz. The ES Ideal Phase Shift and ES Two-Phase schemes overlap and achieve a capacity of about 14.6 bps/Hz at M = 60. The ES Practical Phase Shift scheme achieves the lowest capacity and remains nearly constant as M increases. The TS and TS ON/OFF schemes remain almost unchanged as M increases. The ES Ideal Phase Shift and ES Two-Phase schemes show a gradual increase in capacity with increasing M. For all values of M, the TS and TS ON/OFF schemes achieve higher channel capacity than the ES schemes. Fig. 10Average channel capacity versus the number of STAR-RIS sub-surfaces M. Average channel capacity versus the number of STAR-RIS sub-surfaces M. Figure 11 illustrates the average channel capacity performance of the TS and ES schemes for different vehicle speeds from 0 to 160 km/h. The capacity remains almost constant for all schemes, indicating that mobility does not significantly affect the achievable rate in the considered system. This trend is expected because the Kalman filter exploits Jakes temporal correlation to track Doppler-induced channel variations across frames, thereby keeping the residual channel estimation error nearly constant over the considered speed range. At 80 km/h, the TS scheme achieves the highest capacity of 16.3098 bps/Hz, while the TS ON/OFF scheme achieves a slightly lower but nearly identical capacity, followed by ES Ideal Phase Shift/Two-Phase with 15.1164 bps/Hz and ES Practical Phase Shift with 14.6794 bps/Hz. Overall, TS-based schemes consistently outperform all ES variants, and the nearly flat trend across vehicle speeds confirms robustness against Doppler effects in typical urban vehicular environments. Fig. 11Average channel capacity versus vehicle speed. Average channel capacity versus vehicle speed. Figure 12 shows the average channel capacity of the TS and ES schemes as the number of users K increases from 2 to 16. The TS and TS ON/OFF schemes achieve the highest capacity for all user loads and exhibit nearly identical performance. At K = 10, both the TS and TS ON/OFF schemes achieve a capacity of 23.4963 bps/Hz, which is the highest among all the considered schemes. As K increases, their capacity curves remain nearly unchanged, showing that the achievable rate is only weakly influenced by the number of users. This result shows that the TS-based schemes are able to handle multi-user interference effectively. Among the ES-based schemes, the Ideal Phase Shift and ES Two-Phase schemes overlap and achieve a capacity of 22.2146 bps/Hz at K = 10. The ES Practical Phase Shift scheme yields the lowest capacity, with 21.6371 bps/Hz at K = 10, due to phase quantization and hardware non-idealities at the STAR-RIS that reduce coherent combining efficiency. In general, the capacity changes with respect to K are small for all schemes, indicating stable operation in multi-user scenarios. For all values of K, the TS and TS ON/OFF schemes provide higher capacity than the ES-based methods, confirming the advantage of TS-based training in multi-user vehicular STAR-RIS systems. Fig. 12Average channel capacity versus the number of users K. Average channel capacity versus the number of users K. Conclusion In this work, we proposed efficient channel estimation schemes for a STAR-RIS-assisted multi-user vehicular communication system under the TS and ES protocols. Closed-form LS estimators and NMSE expressions were derived for each protocol, and a Kalman-based tracking model exploiting Jakes’ temporal correlation was incorporated to handle high-mobility channel variations. A practical and two-phase ES training design and a full-power TS training strategy were analyzed using the unified composite channel model. Numerical results demonstrated that TS achieves lower NMSE and higher uplink capacity than ES under the considered system settings, while ES performance depends strongly on the energy splitting ratio. The Kalman filter improves estimation quality in dynamic vehicular environments. Moreover, the proposed framework exhibits robustness against multi-user interference and maintains stable performance as the number of vehicular users increases. The computational complexity analysis confirms that the LS-Kalman framework remains feasible for practical implementation at the base station, even in moderate multi-user and high-mobility scenarios. By considering practical phase-shift imperfections, the proposed framework partially reflects hardware non-idealities in realistic STAR-RIS deployments. This work assumes perfect synchronization between the base station and STAR-RIS and considers narrowband channel estimation under fixed vehicular geometry. One limitation of the current Kalman model is that the state transition matrix relies on Jakes temporal correlation, which assumes the channel statistics remain roughly stable between frames. This works well when the vehicle travels over short distances between pilot transmissions, but over longer trajectories the AoA and AoD can shift noticeably as the surrounding scattering geometry changes. Handling this kind of slow geometric drift in the state model is something we plan to address in future work. Hardware impairments such as phase noise, mutual coupling, and synchronization errors are not fully modeled. In addition, the study focuses on uplink channel estimation and does not explicitly consider downlink estimation and beamforming optimization. The developed method therefore provides an effective channel estimation solution for STAR-RIS-enabled vehicular systems. Future work will investigate joint channel estimation and robust beamforming, as well as downlink training design for STAR-RIS-assisted vehicular networks under hardware impairments and imperfect synchronization. The sensitivity of the Kalman tracker to Doppler frequency mismatch is another practical concern, and joint Doppler estimation and channel tracking remains an open problem for highly dynamic scenarios, which we identify as a direction for future work. Looking ahead, the Kalman-aided estimation approach developed here could also be relevant beyond terrestrial vehicular settings. LEO satellites move at very high speeds, which creates severe Doppler conditions that share some similarity with the high-mobility vehicular problem studied in this work. At the same time, STAR-RIS is well suited for satellite scenarios because it can serve ground users on both sides of the surface simultaneously, which is difficult to achieve with conventional RIS. That said, the scale of the problem is much larger - coordinating massive numbers of ground terminals and designing satellite constellations under heavy Doppler dispersion are challenges that go well beyond what is addressed here21,22. We see this as a worthwhile direction to pursue in future work.