Energy-aware latency minimization in multi-source spatiotemporal data fusion for IoT networks.

Energy-aware latency minimization in multi-source spatiotemporal data fusion for IoT networks - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Apr 13;16:17206. doi: 10.1038/s41598-026-48152-8 Energy-aware latency minimization in multi-source spatiotemporal data fusion for IoT networks Mohammed H Alsharif 1, ✉ , Arun Kumar 2 , Saibal Manna 3 , Vivek Saxena 3 , Shakir Khan 4 , Mohsin Ali Koondhar 5 , Sabri Saeed 6, ✉ , Nour H M Alsharif 7 Author information Article notes Copyright and License information 1 Department of AI Convergence Electronic Engineering, Sejong University, Seoul, 05006 Republic of Korea 2 Department of Electronics and Communication Engineering, Sikkim Manipal Institute of Technology, Sikkim Manipal University, Majitar, Rangpo India 3 Department of Electrical and Computer Engineering, ABES Engineering College, Ghaziabad, UP India 4 College of Computer and Information Sciences, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia 5 Department of Electrical Engineering, Quaid-e-Awam University of Engineering, Science and Technology Nawabshah, Sindh, 67480 Pakistan 6 Department of Networks and Cybersecurity, Al Janad University for Science and Technology, Taiz, Yemen 7 Department of Mathematics, Faculty of Engineering and Natural Sciences, Hitit University, Çorum, Turkey ✉ Corresponding author. Received 2025 Aug 25; Accepted 2026 Apr 6; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13233870  PMID: 41974815 Abstract As smart city infrastructures evolve, they generate massive volumes of spatiotemporal data from diverse sources such as surveillance cameras, wearable health monitors, drones, and environmental sensors. Efficient fusion of this data is crucial for applications that demand ultra-low latency (typically between 2 ms and 5 ms), to support time-sensitive operations. However, the heterogeneity of data sources, coupled with constraints on communication bandwidth and computational resources, poses a significant challenge to maintaining such stringent latency and energy efficiency standards. This study proposes a joint optimization framework for multi-source spatiotemporal data fusion that dynamically allocates bandwidth and CPU cycles to minimize a weighted objective of latency and energy consumption under realistic wireless channel conditions and strict resource constraints. The results demonstrate that the proposed method consistently delivers low latency and energy-efficient performance across all data sources. It outperforms traditional equal and delay-tolerant strategies by significantly reducing both latency and energy consumption. This efficiency, combined with the framework’s robustness andscalability, makes it highly suitable for smart city applications. Keywords: Smart cities, Energy efficiency, Energy-aware latency, Internet of Things, Energy optimization, Spatiotemporal data fusion, Resource allocation Subject terms: Engineering, Mathematics and computing Introduction The proliferation of smart city infrastructures has led to a dramatic increase in the deployment of Internet of Things (IoT) devices, including high-definition surveillance cameras, wearable health monitors, GPS modules, drones, and environmental sensors 1 , 2 . These heterogeneous sources continuously generate vast volumes of spatiotemporal data that must be fused in real-time to enable mission-critical applications such as autonomous vehicle coordination, emergency response, and industrial automation 3 , 4 . Many of these applications demand ultra-low-latency communication and processing (typically within 2–5 ms), to ensure safe and timely responses 5 , 6 . For instance, coordinated drone navigation requires latency under 3 ms 7 , robot arm collaboration in manufacturing operates effectively below 2.5 ms 8 , and adaptive traffic signal control systems react optimally within 2.5 ms 9 . These strict latency constraints underscore the importance of real-time, energy-efficient data fusion systems in smart urban environments. To address these challenges, this study introduces a novel joint optimization framework that intelligently and dynamically allocates both communication bandwidth and computational processing power (CPU cycles) across multiple heterogeneous data sources in a smart city environment. Unlike traditional approaches that statically divide resources or rely on simple heuristics, this framework uses a holistic, data-driven optimization strategy to ensure efficient and fair resource utilization based on each source’s specific requirements. The core objective of the framework is to minimize a weighted aggregate of communication latency, processing latency, and total energy consumption, reflecting the critical trade-offs in real-time smart city operations. Importantly, the model incorporates both per-source and global system constraints, including strict latency thresholds (< 5 ms per source, < 20 ms overall), bandwidth limitations, CPU availability, and energy budgets. What sets this approach apart is its source-aware resource allocation mechanism. Each data stream is treated uniquely based on key characteristics such as its data volume, wireless channel quality (e.g., SNR, fading conditions). The model leverages this information to compute optimal bandwidth allocations and CPU cycles that minimize latency and energy use without compromising synchronization across sources. For example, high-data, latency-sensitive sources like surveillance cameras are granted more bandwidth, while computation-heavy sources like wearable devices are allocated higher CPU resources to reduce processing delays. This tailored allocation strategy enables the system to avoid both under-provisioning (which leads to delay violations) and over-provisioning (which causes energy inefficiencies), ensuring that each source receives exactly the resources it needs to operate effectively in a dynamic, constrained smart city infrastructure. The contributions of this study are threefold. First, we develop a joint energy-aware latency minimization framework that integrates communication and processing delays with their corresponding energy costs into a single optimization model, ensuring ultra-low latency and sustainable energy use under realistic resource constraints. Second, we introduce a source-aware dynamic resource allocation strategy that tailors bandwidth and CPU assignments to each source’s data volume, computational load, and wireless channel quality, while explicitly incorporating fading effects and synchronization constraints achieving fine-grained efficiency beyond traditional heuristic or equal-allocation methods. Third, we demonstrate the robustness and scalability of the proposed framework through extensive simulations and sensitivity analyses, showing that it consistently outperforms equal and delay-tolerant baselines by significantly reducing both latency and energy consumption. The remainder of this paper is organized as follows. Section " Related works " reviews related work on data fusion and resource allocation. Section " System model and optimization formulation " is divided into three subsections: the system model, the optimization problem formulation, and the simulation setup and algorithm implementation. Section " Results and Discussion " presents the simulation results, including performance comparisons with traditional methods and a detailed sensitivity analysis. Finally, Section " Conclusion " concludes the paper. Related works Table 1 presents a comparative overview of the most closely related research on data fusion and resource allocation in smart cities, highlighting the unique contributions of this study. It is noteworthy that recent studies 10 – 14 have investigated UAV-enabled multi-source sensing and collaborative computing to support flexible and on-demand data collection in dynamic environments. For example, multi-UAV–assisted on-demand charging frameworks have been proposed for dense wireless rechargeable sensor networks, where UAVs jointly manage energy delivery, sensing coverage, and communication efficiency. In addition, lightweight blockchain-based architectures have been explored to enable secure and decentralized collaborative computing in UAV ad-hoc networks, addressing challenges related to trust, data integrity, and coordination among multiple aerial nodes. While these works primarily focus on UAV mobility, charging strategies, or security mechanisms, the proposed framework in this paper is complementary, as it addresses the joint bandwidth and computation resource allocation problem under strict latency and energy constraints, which is a critical requirement for real-time multi-source sensing systems, including UAV-enabled networks. The authors in 15 proposed a reputation-driven framework to enhance trust, security, and collaboration in cross-chain IoT data sharing environments. By integrating blockchain interoperability with a dynamic reputation evaluation mechanism, the approach mitigates malicious behavior and improves data reliability. The results demonstrate improved transparency, secure data exchange, and incentivized cooperation among heterogeneous IoT networks. In addition, 16 investigated link reliability challenges in mobile LoRa networks and proposes adaptive techniques to enhance communication performance under mobility conditions. The study analyzes channel dynamics, packet delivery performance, and signal variations in realistic deployment scenarios. Experimental results show significant improvements in link stability and data transmission reliability for mobile IoT applications. Table 1. Comparative summary of related works in data fusion and resource allocation. References Multi-source support Sync. constraint Realistic channel modeling Remarks 17 × × × CPU/BW allocation, Edge offloading and task assignment, Single-user focus, no sync. or source heterogeneity 18 × × × Mobile edge computing resource assignment, Single task/user; sync. not considered 19 ✓ × × Sensor fusion for intelligent vehicles, Fusion quality affected by temporal misalignment 20 ✓ × × Distributed sensing, Static allocation, Non-dynamic adaptation or sync. constraints 21 ✓ × × IoT-aware smart city data analysis, Focus on applications 22 ✓ ✓ × RGB–thermal multimodal crowd counting Bidirectional gated & multiscale dynamic fusion No wireless channel modeling 23 ✓ × ✓ Multi-source crowd sensing (pedestrian + navigation data) Meta-learning & multi-view fusion 24 ✓ × × Multisource fusion with energy-efficient AIoT VMD + KPCA + Informer model Sustainability focus, no sync. Constraints 25 ✓ × ✓ Multi-source heterogeneous IoT data fusion Mixed info gain strategy Proposed Method ✓ ✓ ✓ Latency-aware multi-source data fusion, Non-Convex formulation, Joint, BW&CPU optimization Open in a new tab System model and optimization formulation The proposed system (Fig. 1 ) is designed for operation within a smart city environment, where numerous heterogeneous data sources including, IoT sensors, surveillance cameras, GPS modules, and edge devices generate continuous and diverse spatiotemporal data streams. Each source is defined by a specific data size D i (in bits), a data transmission rate b i (bps), and a computational workload C i (in CPU cycles), and it transmits its data to a centralized fusion center through a wireless channel with a given gain and transmit power to each source. To process the incoming data, the fusion center allocates a communication bandwidth B i (Hz) and a processing frequency f i (cycles/second) to each source. The overarching goal is to minimize the total cost composed of weighted latency and energy consumption across all sources, while respecting system-wide constraints on communication and computational resources. Fig. 1. Open in a new tab System model of the proposed a multi-source spatiotemporal data fusion in smart cities for energy-aware latency minimization. The total latency incurred by each source comprises two components: communication latency and processing latency, details as follows: Communication latency ( ) represents the time required to transmit data over a wireless channel and is modeled using Shannon’s capacity formula: 1 where N o denotes the noise power spectral density, capturing the impact of bandwidth, channel quality, and noise on transmission delay. Processing latency ( ) is the time required to process the data once received at the fusion center and is calculated as: 2 The total latency ( ) for each source is the sum of communication and processing latencies: 3 To ensure energy efficiency, the model also incorporates two energy components, details as follows: Communication energy ( ) is the amount of energy used by source i to send its data to the fusion center. It depends on how much power the device uses to transmit and how long the transmission takes. 4 Processing energy ( ) is required when IoT devices handle data, particularly for tasks like data fusion and signal processing. This energy cost depends on the computational load and the processing power of the device. The energy required for processing can be expressed as: 5 The power consumption of a CMOS-based processor is given by, , where a is the Hardware-dependent coefficient (typically ~ 10 –21 for modern CPUs), and f i is CPU frequency assigned to source i . The total energy ( ) for each source is the sum of communication and processing energies: 6 The optimization problem aims to minimize a weighted sum of latency and energy consumption in a multi-source IoT data fusion system while adhering to strict resource constraints (Bandwidth, CPU, Energy, and Latency). The optimization formula becomes: 7 Subject to (Constraints): 8 9 10 11 12 13 14 15 16 17 The optimization problem for multi-source spatiotemporal data fusion in IoT networks is subject to several critical constraints, each designed to ensure that the system operates efficiently while maintaining performance and adhering to the available physical resources. Bandwidth, as a shared resource in wireless networks, must be carefully managed to avoid congestion and reduced throughput. Constraint (8) ensures that the total bandwidth used by all sources does not exceed the maximum available bandwidth B max of the network. Meanwhile, constraint (9) defines the lower and upper bounds for the bandwidth allocated to each source, ensuring that each source operates within these bounds based on its communication needs and the physical limitations of the network. This helps ensure fair allocation, preventing sources from either monopolizing the bandwidth or underusing it. Similarly, each source requires a certain amount of computational capacity to process its data, and constraint (10) limits the total CPU processing power used by all sources to ensure that the system does not exceed the available computational resources F max . This prevents system overload, ensuring that the computational resources are sufficient to handle the data processing tasks without delays or failures. In addition, constraint (11) places limits on the minimum and maximum CPU frequencies allocated to each source, ensuring that each device has enough computational resources to process its data while preventing over-allocation, which could lead to unnecessary energy consumption. This constraint balances computational performance with energy efficiency, which is crucial for prolonging the lifetime of battery-operated devices and minimizing environmental impact. Furthermore, constraint (12) ensures that the energy consumed by communication and processing for each sources does not exceed the maximum allowable energy budget , safeguarding the system from energy overload and ensuring that the system operates sustainably. Constraint (13) ensures that the aggregate energy of all sources combined does not exceed the overall system energy , which guarantees that the system as a whole meets its energy requirements. Constraint (14) and (15) ensures that the communication and process energy for each source remains within predefined minimum and maximum bounds, preventing both underutilization and excessive energy use. This balance guarantees fair energy distribution across sources. Latency is another critical consideration for real-time data fusion, and constraint (16) ensures that the latency for each source, comprising both communication and processing latency, does not exceed its maximum allowable latency . Finally, constraint (17) ensures that the aggregate latency of all sources combined does not exceed the overall system latency , which guarantees that the system as a whole meets its latency requirements. These constraints ensure that resources such as bandwidth, CPU power, and energy are allocated efficiently, preventing overuse or underuse of resources. By setting limits on energy consumption and latency, the system ensures that it operates sustainably while still meeting the performance requirements of time-sensitive applications. The constraints also provide fair resource distribution, preventing any one source from monopolizing the available bandwidth or processing capacity, which is crucial for scalability in large IoT networks. Additionally, these constraints help to optimize system performance while maintaining energy efficiency, which is vital for battery-powered devices and for minimizing the environmental impact of IoT systems. The optimization problem presented in this context is a non-linear constrained optimization problem, where the objective is to minimize a weighted sum of latency and energy consumption across multiple data sources in a multi-source spatiotemporal data fusion system. The decision variables, such as the bandwidth and CPU frequency allocated to each source, are subject to several constraints, including bandwidth limits, computational capacity, energy consumption bounds, and latency requirements. The problem involves non-linear relationships due to the logarithmic nature of communication latency, the quadratic dependence of processing energy on CPU frequency, and the complex interaction between communication and processing components. This non-linearity, combined with multiple constraints, makes the problem inherently challenging and suitable for optimization techniques capable of handling complex, constrained, and non-convex objective functions. MATLAB is an ideal choice for solving this type of optimization problem due to its extensive built-in libraries, flexibility, and ease of use for handling both mathematical optimization and numerical simulations. The Optimization Toolbox in MATLAB offers powerful functions like ‘ fmincon’ that can efficiently solve non-linear constrained problems, making it suitable for the given scenario. ‘ fmincon’ uses advanced optimization algorithms, such as Sequential Quadratic Programming (SQP), which is highly effective for handling non-linear objective functions and constraints. This algorithm iteratively approximates the problem using a quadratic model, making it suitable for problems where the objective function and constraints are smooth but non-linear. A simulation environment is established with parameters summarized in Table 2 . The optimization algorithm (Algorithm 1) iteratively minimizes the weighted sum of communication and processing Energies and latencies under resource and QoS constraints. Starting from a random feasible allocation of bandwidth and CPU, the algorithm precomputes the initial signal-to-noise ratios (SNRs), communication delays, and processing delays. Using MATLAB’s fmincon solver, it searches for the optimal resource allocation that minimizes the total weighted latency while satisfying bandwidth, CPU, and individual latency constraints. The loop continues until convergence is achieved based on a threshold ε = 10 –4 . The final output includes the optimal per-source bandwidth and CPU allocation, along with the total system latency and energy. Table 2. Simulation setup. Parameter Symbol Value/range Number of Sources S 5 Total Bandwidth (Uplink) B max 20 MHz Min Bandwidth/Source B i min 1 MHz Total CPU Power F max 10 GHz Min CPU/Source f i min 0.5 GHz Data Size D i Rand(10, 100) Mbps Computation Load C i Rand (1, 10) Giga-cycles Priority Weights w i Rand (1, 5); Higher = more critical Max. Latency L i max 5 ms System’s Max. Latency 20 ms Max. Energy 0.5 J System’s Max. Energy 2 J Channel Model h i Rayleigh fading Transmit Power (Uplink) P i tx 0.1 W PSD of noise N o 1e −6 W/Hz Open in a new tab Five heterogeneous data sources are considered as a representative yet interpretable smart city scenario, allowing clear visualization of per-source allocation behavior while maintaining tractable optimization complexity. The framework itself is scalable and not restricted to this number of sources. It is important to note that the adopted simulation framework represents a system-level abstraction of multi-source IoT data fusion. The model focuses on joint bandwidth and CPU allocation under latency and energy constraints, while lower-layer protocol overhead, queuing dynamics, packet retransmissions, and detailed fusion-processing pipeline delays are abstracted into the latency formulation. This abstraction enables clear evaluation of the core optimization behavior without conflating it with protocol-specific implementation details. Algorithm 1. Open in a new tab Weighted Latency Minimization for Spatiotemporal Fusion. Results and discussion Figure 2 , from top to bottom, illustrates the relationship between data size, optimal bandwidth allocation, and optimal CPU allocation, providing strong evidence of the proposed framework’s capacity to minimize both latency and energy consumption across diverse data sources in smart city environments. The system’s resource allocation clearly correlates with each source’s data intensity and urgency, ensuring that communication and computation resources are not wasted but strategically deployed. Source 4, which transmits the largest data volume (100 Mbits), characteristic of a high-definition surveillance camera, receives the highest bandwidth (5.5 MHz) and CPU allocation (2.0 GHz). This decision is justified by the critical nature of surveillance data, which requires real-time processing and uninterrupted video streaming. Without this allocation, latency would breach the 5 ms threshold, compromising safety applications. Source 3, a drone-mounted camera or mobile sensing platform generating 65 Mbits, is granted 4.8 MHz of bandwidth and 1.7 GHz of CPU, enabling responsive aerial data analysis while maintaining system-wide efficiency. Source 2, representing wearable health monitors transmitting 40 Mbits of physiological signals, is assigned 4.1 MHz bandwidth and 1.5 GHz CPU. This allocation supports time-critical biometric processing, where delays could affect emergency response decisions. Source 1, with the smallest data volume (25 Mbits), presumably a GPS tracker or environmental sensor, receives 3.0 MHz and 1.2 GHz, which is sufficient given its lower real-time demands. Interestingly, Source 5 also transmits 40 Mbits, but due to its classification as a non-critical application (e.g., infotainment or social media), it is assigned only 2.6 MHz and 1.3 GHz, reflecting a prioritization of latency-sensitive sources over delay-tolerant ones. These results underscore the contribution of the proposed joint optimization approach, which does not treat all sources equally but tailors resource allocations according to data volume, urgency, and processing complexity. Unlike traditional fixed or equal allocation strategies, this dynamic and adaptive model ensures that all sources remain within latency constraints, energy usage is minimized, and synchronization is maintained critical requirements for efficient and responsive smart city operations. Fig. 2. Open in a new tab Joint Optimization of Bandwidth and CPU Allocation Based on Source Data. Figure 3 illustrates the communication latency, processing latency, and resulting total latency for five distinct data sources. Source 4, previously identified in Fig. 2 as a high-data-volume surveillance camera (100 Mbits), exhibits the highest communication latency (2.88 ms) and processing latency (1.92 ms), resulting in a total latency of approximately 4.8 ms; however, this remains below the critical 5 ms threshold (indicated by the red dashed line). This outcome demonstrates the framework’s ability to manage bandwidth- and CPU-intensive sources without breaching stringent latency constraints. Similarly, Source 3, representing a drone-based mobile sensing unit with 65 Mbits of data, shows a communication latency of 2.58 ms and processing latency of 1.72 ms, yielding a total latency of 4.3 ms. Sources 2 and 5, corresponding to a wearable health device and an infotainment/social media upload, experience moderate communication latencies of 1.98 ms and 1.8 ms, and processing latencies of 1.32 ms and 1.2 ms, resulting in total latencies of 3.3 ms and 3.0 ms, respectively, reflecting balanced resource provisioning. Source 1, associated with a lightweight GPS or environmental sensor transmitting 25 Mbits of data, maintains the lowest latencies: 1.32 ms for communication and 0.88 ms for processing, totaling just 2.2 ms due to its minimal communication and computational demands. These results, together with the resource allocations shown in Fig. 2 , underscore the effectiveness of the proposed framework’s joint optimization strategy, wherein bandwidth and CPU resources are dynamically allocated based on data size and application priority. This guarantees ultra-low-latency fusion across diverse data sources, maintaining sub-5 ms performance and precise synchronization crucial for smart city applications like coordinated drone control, emergency response, and industrial robot collaboration. Fig. 3. Open in a new tab Optimal communication, processing, and total latency. Figure 4 illustrates the energy consumption profile across five sources, closely aligning with the resource allocation and latency trends shown in Figs. 2 and 3 , thereby validating the efficiency of the proposed multi-source spatiotemporal data fusion framework. Source 4, previously identified as a high-definition surveillance camera transmitting 100 Mbits, exhibits the highest energy consumption 0.34 J for communication and 0.144 J for processing, totaling approximately 0.48 J, just below the maximum threshold (indicated by the red dashed line). This is attributed to its significant bandwidth allocation (5.5 MHz) and high CPU usage (2 GHz) necessary for real-time operation. Similarly, Source 3 (a drone-based or mobile sensing platform) consumes 0.28 J for communication and 0.12 J for processing, totaling about 0.4 J, reflecting its moderate-to-high data load (65 Mbits) and balanced resource allocation (4.8 MHz bandwidth and 1.7 GHz CPU). Source 2, associated with wearable health monitoring, consumes 0.25 J in communication and 0.105 J in processing, totaling approximately 0.35 J, consistent with its mid-level data and computational requirements. In contrast, Sources 1 and 5, representing low-complexity data types such as GPS and infotainment streams, exhibit the lowest total energy consumption 0.3 J and 0.31 J, respectively, corresponding to their minimal resource needs (bandwidth: 2.6–3.0 MHz, CPU: 1.2–1.3 GHz). Importantly, the optimized energy consumption across all sources demonstrates that the proposed framework not only satisfies stringent latency constraints (as shown in Fig. 3 ), but also effectively adheres to energy budgets, supporting sustainable and scalable smart city operations. The strong correlation between bandwidth and CPU allocation (Fig. 2 ), latency minimization (Fig. 3 ), and energy efficiency (Fig. 4 ) underscores the robustness of the joint optimization strategy. Fig. 4. Open in a new tab Optimal communication, processing, and total energy. The sensitivity analysis in the Fig. 5 provides crucial insights into the robustness of the proposed optimization framework described by the objective function Eq. ( 7 ), which jointly minimizes total energy consumption across multiple data sources by balancing communication energy, processing energy by trade-off weights γ i and δ i . Figure 5 shown how variations in these trade-off weights affect the objective function value, where the value remains virtually unchanged (around 93.9897907) across several orders of magnitude for both parameters. Despite changing log 10 ( γ ) ∈ [− 6, − 12] and log 10 ( δ )∈ [− 7, − 12], the surface is flat, with variations occurring only at the seventh decimal place of the objective value. This is a strong indicator that the optimization solution is not sensitive to exact tuning of energy weighting parameters. The communication latency (first term) is a logarithmic function of bandwidth and SNR, and the processing latency depends linearly on CPU frequency. These latency terms are dominant in delay-sensitive applications. Since the energy weights γ and δ are small-scale tuning factors (typically used for regularization or balancing), their impact is bounded and does not override the core latency-driven components unless made very large. Thus, robust performance whether in a drone swarm (2–3 ms), robot control (~ 2.5 ms), or traffic signal adjustments, the system guarantees consistent latency minimization even if γ and δ are roughly estimated. Fig. 5. Open in a new tab Robustness of the optimization objective to communication ( γ ) and processing ( δ ) energy weights. Figure 6 provides a comparative evaluation of the proposed joint optimization framework against two representative practical allocation strategies, namely equal resource allocation and delay-tolerant (minimal resource) allocation, across five heterogeneous sources in terms of latency and energy consumption. The results indicate that the proposed approach achieves a well-balanced latency–energy tradeoff while consistently satisfying both per-source and system-level constraints. Fig. 6. Open in a new tab Comparison of the joint optimization technique vs. traditional methods per source. It is important to emphasize that many recent MEC-based and joint optimization studies are developed under substantially different system architectures, objective functions, and modeling assumptions, such as task offloading, single-objective latency minimization, or partially coupled resource allocation frameworks. These differences make direct one-to-one numerical comparisons methodologically inappropriate, as the underlying optimization structures, constraint couplings, and performance metrics are not equivalent. Therefore, the comparisons presented here focus on practical allocation strategies operating under the same system model and constraints, ensuring fairness and consistency in performance evaluation. To further position the proposed framework relative to established optimization paradigms, convex relaxation or decomposition-based heuristics could also be applied to the same coupled nonlinear formulation. However, such methods would ultimately target the identical constraint structure and objective function. Consequently, the performance improvements observed in Fig. 6 primarily reflect the benefit of the unified joint latency–energy optimization formulation rather than solver-specific implementation choices. In the latency comparison, the proposed method (blue bars) maintains total latency below the 5 ms threshold for each sources. Specifically, Source 1 achieves 2.2 ms, Source 2 (3.3 ms), Source 3 (4.3 ms), Source 4 (4.8 ms), and Source 5 (3 ms). This performance is achieved by dynamically assigning bandwidth and CPU based on each source’s data size and processing demand. In contrast, the traditional equal allocation strategy (orange bars), where each source receives 4 MHz bandwidth and 1.54 GHz CPU (from total 20 MHz and 7.7 GHz divided equally among 5 sources), leads to a noticeable latency increase: Source 1 (2.43 ms), Source 2 (4.04 ms), Source 3 (6.59 ms), Source 4 (9.72 ms), and Source 5 (3.9 ms). This leads to an optimal total latency of approximately 26.68 ms under equal allocation. Most notably, Sources 3 and 4 exceed the latency threshold, indicating resource mismatch under uniform allocation. On the other hand, the delay-tolerant approach (yellow bars), which minimizes resource usage by assigning only 1 MHz bandwidth and 0.5 GHz CPU per source regardless of demand, results in severe latency violations: Source 1 (7.78 ms), Source 2 (9.4 ms), Source 3 (22.4 ms), Source 4 (16.92 ms), and Source 5 (16.34 ms). This leads to an optimal total latency of approximately 72.84 ms. On the energy comparison side, the proposed method again proves efficient. It keeps total energy consumption well under the 0.5 J threshold for each sources, with Source 1 (0.30 J), Source 2 (0.35 J), Source 3 (0.4 J), Source 4 (0.48 J), and Source 5 (0.31 J). This is achieved through careful balancing of communication and computation energy based on demand. In the traditional case, equal CPU and bandwidth lead to slight inefficiencies, with Sources 3 and 4 reaching ~ 0.7 J and slightly exceeding the energy threshold. The delay-tolerant system performs the worst, with Source 4 consuming nearly 10 J, and other sources ranging from ~ 2.5 to 6.5 J. This is due to prolonged transmission and processing times under severely limited resources, which not only violates energy constraints but also amplifies latency. These results validate that the proposed method intelligently matches resource allocation to the specific characteristics of each source, considering data size, channel conditions, and computational demands resulting in a superior latency-energy trade-off. In contrast, when resources are equally distributed without accounting for individual source requirements (traditional method), some sources become over-provisioned while others are under-provisioned, leading to performance inefficiencies. Moreover, when resources are minimized without regard to application-critical requirements (delay-tolerant approach), both latency and energy performance degrade significantly, making it unsuitable for smart city applications. Figure 7 provides a comprehensive system-level comparison of total latency and total energy consumption across three resource allocation strategies: the proposed optimization-based approach, the traditional equal allocation method, and the delay-tolerant minimal allocation. The results clearly demonstrate the effectiveness of the proposed framework in smart city scenarios where latency and energy efficiency are critical. Fig. 7. Open in a new tab System-level comparison of latency and energy for different resource allocation strategies. In terms of system-wide latency, the proposed method achieves a total latency of approximately 18.6 ms, remaining well below the predefined system threshold of 20 ms. This performance is attributed to the dynamic allocation of bandwidth and CPU resources based on each source’s data volume, processing demand, and wireless channel conditions. In contrast, the traditional equal allocation strategy assigns 4 MHz bandwidth and 1.54 GHz CPU to each of the five sources, resulting in a higher total latency of 26.68 ms due to resource mismatch. Some sources are over-provisioned while others are under-provisioned, leading to inefficiencies. The delay-tolerant strategy, which allocates only 1 MHz bandwidth and 0.5 GHz CPU per source, performs the worst, with a total latency of 72.84 ms, indicating that limited resources significantly degrade performance in time-sensitive applications. In the system-wide energy comparison, the proposed method again proves most efficient, consuming only 1.84 J in total, which is well below the 2 J system threshold. This is achieved by optimizing resource use and minimizing unnecessary energy expenditure. The traditional method results in a total energy consumption of 2.25 J due to uniform allocation inefficiencies. The delay-tolerant strategy incurs the highest energy cost at 27.25 J, as prolonged transmission and processing under limited resources lead to substantial energy waste. These results highlight the key contribution of the proposed framework. By dynamically tailoring resource allocation to the specific needs of each source, it achieves a superior trade-off between latency and energy. This ensures scalable and efficient performance across heterogeneous smart city systems. Figure 8 presents a comparison of computation time as a function of the number of sources (S), contrasting the proposed optimization approach with the traditional equal allocation method. The proposed method demonstrates relatively stable computation times around 0.016 s. As the number of sources increases, the computation time increases slightly to 0.027 s, reflecting the increasing complexity of the multi-source optimization problem. This small increase is due to the system’s need to dynamically allocate resources across a larger number of data sources while balancing both bandwidth and CPU power to maintain performance. Despite the additional complexity, the proposed method’s computation time remains manageable, showcasing its efficiency in handling multiple sources without significant performance degradation. On the other hand, the traditional method with equal resource allocation shows consistent, low computation times. However, the lack of adaptive resource allocation in the traditional approach means that it struggles to optimize resource distribution as the number of sources grows, resulting in less efficient use of system resources. The proposed method stands out by ensuring that even as the system scales to handle more sources, the computation time remains stable, demonstrating its ability to effectively manage resource allocation. Fig. 8. Open in a new tab Computational complexity analysis of the proposed optimization vs. traditional allocation method. Table 3 provides a comparison between the proposed multi-source spatiotemporal fusion method and a traditional (single-source or fixed-allocation) system across several key features relevant to smart city applications. These include total latency, synchronization accuracy, flexibility, and runtime complexity. The proposed method clearly outperforms the traditional approach in terms of latency, achieving significantly lower end-to-end delays due to its ability to jointly optimize bandwidth and CPU allocations while adapting to heterogeneous data characteristics. In terms of synchronization accuracy, the proposed method enforces inter-source latency alignment, which ensures that all incoming data streams arrive within a tightly controlled window. This is vital in applications such as cooperative vehicular systems, augmented reality, or real-time surveillance, where delayed or misaligned inputs can lead to degraded situational awareness or decision errors. In contrast, the traditional system lacks such synchronization enforcement, often resulting in temporal misalignment and reduced fusion quality. On the flexibility front, the proposed system supports multi-source configurations, making it inherently more suitable for modern smart city environments that rely on data from a variety of sensors, devices, and consumer electronics. The traditional approach, however, is typically single-source oriented or relies on fixed/static allocation strategies, limiting its adaptability to complex scenarios. Table 3. Comparative evaluation of proposed multi-source fusion vs. traditional methods. Feature Proposed Fusion Traditional (Single/Fixed) Latency ✓ Lower × Higher (unsynchronized) Synchronization accuracy ✓ Enforced × Not guaranteed Flexibility ✓ Multi-source × Often single-source Runtime Slightly higher Very fast Open in a new tab Lastly, while the runtime complexity of the proposed method is slightly higher, this is a reasonable and acceptable trade-off considering the notable improvements in latency and synchronization. Conclusion This paper introduced a robust optimization-based framework for energy-aware, low-latency spatiotemporal data fusion in multi-source IoT systems operating within smart cities. The proposed joint resource allocation method effectively allocates communication and computational resources according to each data source’s unique characteristics, including data volume, urgency, and processing demand. Simulation results validate the effectiveness of the framework, latency per source is kept under the 5 ms critical threshold (ranging from 2.2 to 4.8 ms), and total system latency remains at 17.6 ms well below the 20 ms requirement for real-time operation. Energy consumption is equally well managed, with per-source values ranging from 0.3 to 0.48 J and total energy capped at 1.84 J, outperforming traditional equal allocation (2.25 J) and delay-tolerant methods (27.25 J). Additionally, sensitivity analysis confirms the framework’s robustness to variations in energy-weighting hyperparameters ( γ , δ ), demonstrating stable objective values across several orders of magnitude. Integrating the proposed optimization framework with UAV mobility control, on-demand charging, or secure collaborative computing mechanisms represents an interesting direction for future research. Moreover, future work will extend the proposed framework to large-scale IoT deployments involving tens or hundreds of heterogeneous data sources to further evaluate scalability and resource contention effects. Extensive Monte Carlo simulations under diverse wireless channel models, including Rician and shadowed fading, will also be conducted to capture statistical performance variations and robustness under realistic propagation conditions. These extended evaluations will provide deeper insights into the trade-offs between latency, energy efficiency, and computational complexity in dense smart city environments. Author contributions Concepts: Mohammed H. Alsharif and Arun Kumar; Methodology: Mohammed H. Alsharif, Saibal Manna, Vivek Saxena, and Sabri Saeed; Software: Mohammed H. Alsharif and Nour H. M. Alsharif; Validation: Nour H. M. Alsharif; Writing-Original Draft Preparation: Mohammed H. Alsharif, Arun Kumar, and Shakir Khan; Writing-Final Review and Editing: Shakir Khan, Mohsin Ali Koondhar and Sabri Saeed; Project Administration: Mohsin Ali Koondhar and Sabri Saeed. Funding No fund. Data availability The datasets used and analyzed during the current research are available from the corresponding author on the reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Mohammed H. Alsharif, Email: malsharif@sejong.ac.kr Sabri Saeed, Email: Sabri.alshibany@just.edu.ye. References 1. Kanellopoulos, D., Sharma, V. K., Panagiotakopoulos, T. & Kameas, A. Networking architectures and protocols for IoT applications in smart cities: Recent developments and perspectives. Electronics 12 , 2490 (2023). [ Google Scholar ] 2. Jahid, A., Alsharif, M. H. & Hall, T. J. The convergence of Blockchain, IoT and 6G: Potential, opportunities, challenges and research roadmap. J. Netw. Comput. Appl. 103677 (2023). 3. Trigka, M. & Dritsas, E. Edge and cloud computing in smart cities. Future Internet 17 , 118 (2025). [ Google Scholar ] 4. Alsharif, M. H., Jahid, A., Kannadasan, R. & Kim, M.-K. Unleashing the potential of sixth generation (6G) wireless networks in smart energy grid management: A comprehensive review. Energy Rep. 11 , 1376–1398 (2024). [ Google Scholar ] 5. Feng, D. et al. Ultra-reliable and low-latency communications: Applications, opportunities and challenges. Sci. China Inf. Sci. 64 , 120301 (2021). [ Google Scholar ] 6. Hazra, A., Munusamy, A., Adhikari, M. & Awasthi, L. K. 6G-Enabled ultra-reliable low latency communication for Industry 5.0: Challenges and future directions. IEEE Commun. Stand. Mag. 8 , 36–42 (2024). [ Google Scholar ] 7. Qu, C., Sorbelli, F. B., Singh, R., Calyam, P. & Das, S. K. Environmentally-aware and energy-efficient multi-drone coordination and networking for disaster response. IEEE Trans. Netw. Serv. Manag. 20 , 1093–1109 (2023). [ Google Scholar ] 8. Kumar, S. S., Kumar, S. R. & Ramesh, G. From industry 4.0 to 5.0: Enriching manufacturing excellence through human–robot interaction and technological empowerment. In Intelligent Systems and Industrial Internet of Things for Sustainable Development 24–51 (Chapman and Hall/CRC, 2024). [ Google Scholar ] 9. Hu, C., Wang, Y., Na, J., Guo, G., Zuo, Z., Gao , H. et al., Optimal adaptive cruise control in mixed traffic with communication latence and driver reaction. IEEE Trans. Intell. Transp. Syst. (2024). 10. Hussain, A. et al. Computing challenges of UAV networks: A comprehensive survey. Comput. Mater. Contin. 10.32604/cmc.2024.056183 (2024). [ Google Scholar ] 11. Wang, Y., Li, J., Yang, X. & Peng, Q. UAV–ground vehicle collaborative delivery in emergency response: A review of key technologies and future trends. Appl. Sci. 15 , 9803 (2025). [ Google Scholar ] 12. Liu, Y., Xie, X., Kong, Y., Zhang, B. & Kang, Y. Survey on the integration of communication, sensing, and computing power for low-altitude UAVs. In 2025 7th International Conference on Electronics and Communication, Network and Computer Technology (ECNCT) 196–207 (2025). 13. Zhang, Y., Fan, R., Mei, H., Du, S. & Zhou, Z. Federated learning-driven trajectory and resource optimization in UAV-enabled integrated sensing and communication systems. IEEE Trans. Veh. Technol. (2025). 14. He, B., Ji, X., Li, G. & Cheng, B. Key technologies and applications of UAVs in underground space: A review. IEEE Trans. Cogn. Commun. Netw. 10 , 1026–1049 (2024). [ Google Scholar ] 15. Xiong, R. et al. Enhancing trust and collaboration: A reputation-driven mechanism for cross-chain IoT data sharing. Comput. Netw. 264 , 111266 (2025). [ Google Scholar ] 16. Chen, C. et al. Enhancing link performance for mobile LoRa networks. IEEE Trans. Mob. Comput. 10.1109/tmc.2025.3588923 (2025). [ Google Scholar ] 17. Qian, Y. et al. Learning to optimize resource assignment for task offloading in mobile edge computing. IEEE Commun. Lett. 26 , 1303–1307 (2022). [ Google Scholar ] 18. Yan, J., Bi, S., Zhang, Y. J. & Tao, M. Optimal task offloading and resource allocation in mobile-edge computing with inter-user task dependency. IEEE Trans. Wirel. Commun. 19 , 235–250 (2019). [ Google Scholar ] 19. Tao, X. et al. A multi-sensor fusion positioning strategy for intelligent vehicles using global pose graph optimization. IEEE Trans. Veh. Technol. 71 , 2614–2627 (2021). [ Google Scholar ] 20. Tkach, I. & Edan, Y. Distributed Heterogeneous Multi Sensor Task Allocation Systems (Springer, 2020). [ Google Scholar ] 21. Rafique, Y., Leivadeas, A.& Ibnkahla, M. An IoT-aware VNF placement proof of concept in a hybrid edge-cloud smart city environment. In 2022 IEEE Wireless Communications and Networking Conference (WCNC) 1395–1400 (2022). 22. Xie, Z. et al. Bgdfnet: Bidirectional gated and dynamic fusion network for RGB-t crowd counting in smart city system. IEEE Trans. Instrum. Meas. 73 , 1–16 (2024). [ Google Scholar ] 23. Zhang, Y., Li, C., Li, K., He, S. & Chen, J. High-resolution large-scale urban traffic speed estimation with multi-source crowd sensing data. IEEE Trans. Veh. Technol. 73 , 12345–12357 (2024). [ Google Scholar ] 24. Cai, J. et al. Multi-source fusion enhanced power-efficient sustainable computing for air quality monitoring. IEEE Internet Things J. 10.1109/JIOT.2024.3420956 (2024). [ Google Scholar ] 25. Sun, R. & Ren, Y. A multi-source heterogeneous data fusion method for intelligent systems in the Internet of Things. Intell. Syst. Appl. 23 , 200424 (2024). [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The datasets used and analyzed during the current research are available from the corresponding author on the reasonable request. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (4.1 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Minimum Energy Information Fusion in Sensor Networks#639139
• MLDCSS: Energy-Aware Multi-Level Cooperative Spectrum Sensing with Hierarchical Fusion and Non-Coherent MIMO in CR-IoT Networks#235257
• Structural impact of non-IID heterogeneity on federated behavioral anomaly detection in IoT and IoMT systems.#338459
• Overlapping Covariance Intersection: Fusion with Partial Structural Knowledge of Correlation from Multiple Sources#1156304
• A Methodology to Assess Power Modeling in Energy-Aware Federated Learning on Heterogeneous Mobile Devices#238585