arXiv CSopen access

Real-Time State Estimation in Smart Grids over 5G Networks: Experimental Validation Using Raspberry Pis and Typhoon HIL

1

Real-Time State Estimation in Smart Grids over 5G Networks: Experimental Validation Using Raspberry Pis and Typhoon HIL Biswajit Kumar Dash, Luis Herrera, and Filippo Malandra

arXiv:2606.27642v1 [eess.SY] 26 Jun 2026

Department of Electrical Engineering, State University of New York at Buffalo, NY, USA Email: {biswajit, lcherrer, filippom}@buffalo.edu

Abstract—Reliable, low-latency communication is critical for real-time monitoring and control in modern Smart Grids (SGs). The emergence of 5G networks, with enhanced reliability, significantly lower latency, and native support for massive machine-type communication, offers strong potential to enable advanced grid applications such as state estimation (SE) and fault detection. While existing studies investigate 5G for SG use cases, most rely on simulations or analytical models; experimental validation using real hardware and SG data remains limited. This paper fills this gap by presenting a fully experimental validation of real-time SE over a commercial 5G network using a 5G-based multi-node testbed built with Raspberry Pi (RPi)-based SG nodes and a Typhoon Hardware-in-the-Loop (HIL) real-time simulator. We first characterize 5G communication performance using simulated SG data under varying reporting rates and deployment environments by evaluating Key Performance Indicators (KPIs) such as end-to-end delay, jitter, and frame loss. Experimental results show that the worst-case mean delay observed for the 5G is approximately 6.5× lower than that of our previous LTE cat-M study at the corresponding reporting rate. We then stream realtime voltage, current, and phase-angle measurements—generated by an IEEE 4-node feeder model in Typhoon HIL simulator— to a remote Phasor Data Concentrator (PDC) for SE and fault detection. Results demonstrate that 5G-enabled measurements support accurate SE under both steady-state and dynamic load variations. Furthermore, fault-detection experiments confirm reliable and prompt fault detection, with detection delays as low as 0.80 s. Index Terms—Smart Grid communication, 5G Networks, Multi-node testbed, Hardware-in-the-Loop (HIL), Network performance, State Estimation, Synchrophasor Measurements

I. I NTRODUCTION MERGING and future Smart Grid (SG) systems increasingly rely on advanced communication and networking technologies to meet the growing complexity and diversity of modern power systems. With global shifts towards sustainable energy sources and the adoption of diverse SG applications, there is a pressing need for enhanced reliability, reduced latency, and scalable connectivity. These improvements are crucial for enabling efficient, bidirectional data exchange across the distributed power grid, thereby supporting intelligent monitoring, estimation, and decision-making capabilities. The growing demand for reliable, high-performance, and scalable communication solutions in SG applications has driven the adoption of 5G technology, built upon its three foundational pillars: enhanced mobile broadband (eMBB), ultra-

E

reliable low-latency communications (URLLC), and machinetype communication (mMTC) [1]. These capabilities promise to meet the stringent requirements of SG functions such as real-time monitoring, distributed control, and state estimation (SE), which are critical for ensuring grid stability and resilience. Despite these promises, the practical deployment of 5G in SG environments requires rigorous performance analysis to validate its ability to deliver low latency, high reliability, and scalability under realistic conditions. Existing studies have primarily relied on three approaches: mathematical analysis (e.g., [2]), which offers theoretical insights but often oversimplifies network behavior; network simulation [1], [3], which enables large-scale scenario modeling but lacks hardware-level fidelity; and field trials (e.g., [4]), which provide real-world validation but are costly and difficult to scale. Moreover, these methods rarely capture the tight coupling between communication performance and power system dynamics, leaving a critical gap in understanding the end-to-end behavior of SG applications over 5G networks. To address this gap, experimental performance analysis has emerged as a promising alternative [5]–[7]. Unlike purely simulated environments, experimental setups allow direct measurement of latency, throughput, and reliability while accounting for hardware constraints, protocol overheads, and environmental factors. More importantly, by integrating communication hardware with real-time power system simulators such as Typhoon Hardware-in-the-Loop (HIL) or OPAL-RT, researchers can evaluate how network performance affects essential grid functions, such as SE and control [8]. This holistic approach provides actionable insights for system operators and technology developers. However, experimental evaluation introduces its own challenges, including the complexity of integrating heterogeneous hardware, ensuring synchronization between communication and power domains, and maintaining reproducibility across different test conditions. In this work, we present a multi-node experimental testbed that leverages commercial off-the-shelf (COTS) devices, such as Raspberry Pi (RPi) units equipped with 5G connectivity, integrated with a Typhoon HIL real-time simulator. This architecture enables the execution of SE algorithms over a commercial U.S.-based 5G communication network while maintaining tight coupling with power system dynamics. By combining cost-effective hardware with high-fidelity simu-

2

lation, our approach delivers a scalable and reproducible platform for evaluating 5G performance in SG applications. The results demonstrate the feasibility of deploying 5G for critical SG functions and provide valuable insights into latency, reliability, and integration challenges. The main contributions of this work are as follows: • Design and deployment of a multi-node testbed that integrates 5G connectivity with high-fidelity real-time power system simulation with Typhoon HIL. • Experimental evaluation of 5G network performance to study its feasibility for a variety of SG applications. • Validation of real-time SE over a commercial 5G network using synchrophasor data from an IEEE 4-node feeder. • Demonstration of fault-detection capability by injecting faults into the test feeder and showing that a global residual-based metric can reliably and quickly identify fault events. The remainder of this paper is organized as follows. Section II reviews the state of the art. Section III describes the architecture of our 5G-enabled SG testbed. Section IV presents experimental results on the 5G communication performance. Section V provides the real-time SE framework and experimental validation over 5G networks. Finally, Section VI concludes the paper. II. S TATE OF THE A RT This section reviews the existing literature on communication technologies for SGs in Section II-A, discusses how communication performance influences key SG applications in Section II-B, and identifies the research gaps that motivate this work. A. Communication Technologies for Smart Grids Communication networks play a critical role in supporting monitoring, control, and automation in modern SGs [9], [10]. Although both wired and wireless technologies have been explored to meet the stringent latency, reliability, and scalability demands of emerging grid applications, wireless networks become attractive due to their flexibility, scalability, and lower installation cost [11]. A variety of wireless technologies, including WLANs and Zigbee, has been examined for distribution-level monitoring and control [12], [13]. Although these systems demonstrate useful capabilities, their reliability, coexistence performance, and coverage limitations restrict large-scale deployment in SG environments. Zigbee, for instance, is particularly vulnerable in harsh or interference-heavy environments [14]. Broadband cellular networks such as 3G and LTE have also been evaluated for grid telemetry [15], with studies considering both machineto-machine (M2M) and mixed traffic scenarios [16], [17]. However, LTE is not optimized for transmitting small, periodic sensor measurements due to its heavy protocol overhead [18]. Studies have therefore explored cellular IoT technologies such as LTE cat-M and NB-IoT, which offer reduced device costs, improved coverage, and lower energy consumption [19], [20]. Our previous work experimentally evaluated LTE catM for SG communications using a single Arduino-based

measurement node [5] and later extended the study to a multinode configuration [6]. These studies highlighted two practical limitations. First, the measured delay (approximately 176– 200 ms) exceeded the latency requirements of several key applications, including SE. Second, these studies primarily evaluated communication performance using simulated measurement data, leaving open questions about system performance with actual grid signals. These limitations motivate the exploration of emerging 5G networks, which promise significantly lower latency, higher reliability, and improved support for massive deployments of lightweight SG devices. B. Impact of Communications on Smart Grid Applications The high reliability and low latency of 5G communications make them ideal for time-critical SG applications, including SE, fault detection, and control that require timely and accurate measurement data. In [21], the authors examine the use of 5G technologies for fault identification, estimation, monitoring, and fault distributed generation control. The impact of 5G communication failures on SE techniques is investigated in [2]. Leveraging the low-latency and high-reliability properties of 5G, [1] proposes distributed SE methods using the Alternating Direction Method of Multipliers (ADMM) and Belief Propagation (BP). Similarly, [3] develops a Gaussian Belief Propagation (GBP) method for linear SE in 5G cloud radio access networks. Beyond 5G-focused studies, several works analyze how communication disruptions affect SE regardless of the underlying technology. venda et al. [22] analyze the impact of delays and packet loss on SE accuracy and show that such irregularities can significantly increase the mean square error in voltage angle estimations. Cokic et al. [23] highlight how network congestion degrades real-time application performance by increasing latency and delivering outdated information. Tsitsimelis et al. [24] examine the reliability of LTE’s random-access channel (RACH) and demonstrate how increased device contention and cell coverage variations can impair Wide Area Monitoring Systems (WAMS). Aminifar et al. [25] investigate how WAMS network failures may bring the system to an unobservable state and cause severe cascading events. In addition, Gu et al. [26] propose a dynamic SE method that integrates Kalman filtering with Kriging-based forecasting to mitigate communication failures. Despite extensive research on wireless communication technologies for SGs and increasing interest in 5G-enabled monitoring and estimation, several important gaps remain. First, most existing studies rely on analytical models or offline simulations, thereby lacking experimental validation. While simulations and analytical methods can be cost-effective for predicting communication network behavior during the design phase, experimental studies provide tangible insights into the complexities of SG communication systems. Second, existing work rarely provides real-world experimental validation of 5G performance using actual grid measurements. Finally, there is a lack of integrated experimental platforms that combine real hardware-based SG nodes, commercial 5G networks, HIL

3

Power System Emulation ADC

Control Messages Data Frames

DC

A

SG Node Components

Typhoon HIL 602+ Simulator

C RPi

5G HAT UDP Server (PDC)

RPi LCD

5G BS

Antennas

SG Nodes

Internet

5G-based Communication Network

SG Application

Fig. 1. 5G-based multi-node Smart Grid (SG) communication testbed using Typhoon HIL simulator.

power-system simulators, and application-level evaluations, such as SE, load tracking, and fault detection. These gaps are addressed in this paper by conducting an experimental campaign on a custom-built testbed over a commercial 5G network. III. S YSTEM D ESCRIPTION Fig. 1 illustrates the architecture of our multi-node testbed, which is designed as an end-to-end cyber-physical platform consisting of four tightly coupled layers: i) power system emulation, ii) SG node, iii) 5G-based communication network, and iv) SG application. The main function of the power system emulation layer is to provide realistic grid dynamics. To do so, we used a Typhoon HIL 602+ real-time simulator, with the Typhoon HIL Control Center software running on a host computer. The reference scenario considered in this work is an IEEE 4node distribution feeder model that generates real-time analog outputs corresponding to bus voltages, currents, and power measurements. These analog signals are routed through a breakout interface to the SG nodes. Each SG node performs computation tasks, using a RPi 4 Model B device, and communication tasks, using an external 5G modem (i.e., Quectel RM520N-GL 5G HAT) and four cellular antennas, as shown in Fig. 1. Each SG node also features an LCD screen for local visualization and operation. The SG nodes interface directly with the Typhoon HIL simulator to acquire measurement signals. Since the RPi does not include a built-in Analog-to-Digital Converter (ADC), each node employs an external 16-bit ADS1115 ADC to digitize the analog outputs produced by the HIL simulator. To accommodate both positive and negative signal values, measurements are appropriately scaled within Typhoon to match the ADC input range, and inverse scaling is applied at the SG node to reconstruct the original physical quantities. Each SG node locally processes the digitized measurements and packages them into synchrophasor-like data frames compliant with the IEEE C37.118.2 Standard [27]. These frames are transmitted at configurable reporting rates over the commercial 5G network.

The 5G-based communication network layer provides bidirectional connectivity between the SG nodes and the SG application layer. Measurement frames generated by the SG nodes traverse the 5G Base Station (BS), the operator’s core network, and the public Internet before reaching the server. Bidirectional connectivity is also used to transmit control messages from the server, enabling dynamic adjustments to parameters such as reporting rate, payload size, and experiment configuration. The SG application layer consists of a UDP server running on a remotely connected virtual machine, which is in charge of exchanging data packets with the SG nodes as well as aggregating the received data and executing the tasks associated with the selected SG application. In the considered synchrophasor-like scenario, the UDP server acts as a Phasor Data Concentrator (PDC). In this work, we focus on an SE application; however, the proposed testbed can be used to emulate a variety of SG applications, including Phasor Measurement Unit (PMU), distributed optimization of the power distribution system, and smart metering [7]: this aspect is presented in Section IV. Then, we formulate the real-time SE problem and experimentally validate its performance under diverse operating scenarios (Section V). IV. E XPERIMENTAL P ERFORMANCE E VALUATION OF 5G FOR S MART G RID C OMMUNICATIONS To assess the suitability of the commercial 5G network and our custom-built SG nodes for SG communications, we conducted a series of experiments designed to emulate realistic SG traffic patterns. These experiments evaluate communication performance and are conducted using SG nodes without the Typhoon HIL simulator, thereby isolating communication from power system dynamics. Specifically, we study the communication performance of the commercial 5G network over a range of reporting rates (Section IV-A) and compare its performance under indoor and outdoor deployment scenarios (Section IV-B). Each SG node was configured to transmit frames periodically at a rate of 𝜆 frames per second (fps), a setup flexible enough to represent various SG data types (e.g.,

4

synchrophasors, smart metering, and monitoring data). We conducted an end-to-end performance evaluation focusing on three network Key Performance Indicators (KPIs), including delay, jitter, and frame loss. End-to-end delay was computed at frame level by subtracting the transmission time from the reception time of each frame and was then used1 . Here, the delay accounts for the 5G transmission delay and internetinduced latency. A. 5G Communication Performance Under Different Reporting Rates To evaluate the suitability of the commercial 5G network for supporting a broad range of SG applications, we conducted an experimental study over reporting rates representative of those applications. In particular, we focused on the IEEE Standard C37.118.2 [27], analyzing power systems operating at two possible frequency, i.e., 50 Hz (with reporting rate 𝜆 = 0.5, 1, 10, 25, 50 and 100 frames per second (fps) [27]) and 60 Hz (with 𝜆 = 12, 15, 20, 30, 60 and 120 fps [27]). For every value of 𝜆, three experiments were run: each experiment featured the transmission of 30000 frames, except for smaller reporting rates such as 10–15 fps (15000 frames), and 0.5– 1 fps (7500 frames). All experiments were conducted using a single SG node deployed in an indoor environment (on the University at Buffalo campus). These data were then aggregated for all experiments with the same reporting rate and results were summarized in Table I. For each reporting rate, we considered delay information of all the frames transmitted in the three experiments and computed minimum, maximum, mean, standard deviation (St.d.), first and third quartiles (Q1, Q3), jitter, 95% Confidence Interval (95% CI), and frame loss percentage.

Delay stats (ms) Max

Delay stats (ms)

95% CI

Indoor

Max

Mean St.d.

Q1

Q3

Jitter

95% CI

Frame Loss (%)

21.30 85.14 25.50 3.02 24.38 24.80 2.15 25.50 ± 0.32 2.22e-03

Outdoor 15.41 124.00 22.35 5.05 18.09 24.50 4.05 22.35 ± 1.82 1.11e-03

Both indoor and outdoor experiments exhibit very similar, highly reliable, and low-latency 5G performance. The mean delay ranges from 22.35 ms (outdoor) to 25.50 ms (indoor), a substantial improvement over our previous LTE cat-M system, which reported delays of approximately 176 ms outdoors and 200 ms indoors [6]. Jitter is also low in both settings, ranging from 2.15 ms (indoor) to 4.05 ms (outdoor), indicating stable transmission. The St.d. and 95% CI remain close, and the frame loss rate is in the order of 10−3 (which is considered very low for most of the envisioned SG applications). Although the performance is similar, we observe two subtle differences: (1) indoor experiments show a slightly higher mean delay than the outdoor experiments, and (2) outdoor experiments exhibit a broader delay spread (higher St.d. and jitter) than the indoor experiments. These differences become more apparent in the delay histograms in Fig. 2 across all three experiments. For visualization clarity, the delay distributions are truncated at the 99.9th percentile to remove rare extreme outliers (<0.1% of samples). All reported metrics in Table II are computed using the full dataset.

Q3

Jitter

12.59 325.73 32.66 7.77 29.59 36.85

6.11

32.66 ± 5.37 0.00e+00

1

16.57 170.64 32.14 3.54 31.00 31.48

2.00

32.14 ± 3.33 0.00e+00

10

19.05 151.31 29.86 3.94 27.76 31.25

1.78

29.86 ± 5.74 1.11e-02

12

13.59 271.97 23.58 4.72 19.61 26.08

5.12

23.58 ± 1.47 0.00e+00

15

13.79 192.85 22.31 5.18 17.81 25.02

6.83

22.31 ± 2.41 4.43e-03

20

21.80 160.62 26.46 3.87 24.36 27.91

2.44

26.46 ± 0.78 0.00e+00

25

21.00 163.46 25.87 3.95 23.92 25.82

2.19

25.87 ± 1.08 0.00e+00

30

16.97 144.82 23.55 3.92 21.12 24.76

3.46

23.55 ± 1.70 0.00e+00

50

21.30

25.50 3.02 24.38 24.80

2.15

25.50 ± 0.32 0.00e+00

60

16.50 122.15 21.99 4.33 18.66 24.03

4.21

21.99 ± 0.89 0.00e+00

100

16.99

22.09 4.34 19.34 24.36

4.04

22.09 ± 0.88 0.00e+00

120

16.59 156.50 20.44 5.11 18.22 21.09

3.56

20.44 ± 0.44 0.00e+00

92.92

Q1

TABLE II D ELAY S TATISTICS AND F RAME L OSS IN I NDOOR AND O UTDOOR E XPERIMENTS .

Frame Loss (%)

0.5

85.14

Mean St.d.

Depending on the chosen SG application, nodes may be installed indoor or outdoor, we further examined the network communication performance by comparing the differences in these two scenarios: we conducted three indoor and three outdoor experiments using a single SG node. As in Table I, we computed statistics at frame level and presented comparative results indoor vs outdoor in Table II.

Min

D ELAY S TATISTICS VS . R EPORTING R ATE .

Min

B. Indoor vs. Outdoor Communication Performance

Setup

TABLE I Reporting rate 𝜆 fps

observed at very low reporting rates (0.5–10 fps), where the delay is slightly higher (29–32 ms) and more variable. This behavior can be attributed to larger inter-frame intervals, which allow channel conditions to drift and may cause the modem to re-initiate scheduling or random-access procedures. However, these effects are modest, and the performance remains well within acceptable limits for SG applications. Once the reporting rate reaches the IEEE-recommended typical operational range (12–60 fps) and beyond, the system stabilizes. The mean delay converges to 20–26 ms and 95% CIs become narrow, indicating that frequent transmissions keep the radio link consistently engaged and support smoother scheduling at the cellular BS. Even at extreme rates of 100–120 fps—well above conventional PMU practice–the 5G link maintains low-latency, low-variability performance with no observable congestion or frame loss. These results reveal that 5G network performance is largely insensitive to the reporting rates we used, thus making 5G (and our testbed) suitable for a wide variety of SG applications.

Across all reporting rates, the 5G-based communication network consistently demonstrates stable and robust performance with a low mean delay (20-33 ms), a low jitter, and frame loss nearing zero across the full range. However, a mild trend is 1 To use this method, the clocks of both transmitter and receiver need to be synchronized with high accuracy: this is made possible thanks to the use of the Network Time Protocol (NTP), which ensures a synchronization error in the order of a few ms.

5

Indoor Exp. 2

Exp. 1

Exp. 3 20000

10000 0

15

20

25 30 35 Delay (ms)

40

45

Count

Count

Count

20000 10000

0

15

20

25 30 35 Delay (ms)

40

45

10000

0

15

20

25 30 35 Delay (ms)

40

45

15

20

25 30 35 Delay (ms)

40

45

Outdoor

0

15

20

25 30 35 Delay (ms)

40

45

Count

Count

Count

10000 10000

5000 0

15

20

25 30 35 Delay (ms)

40

45

10000 5000 0

Fig. 2. Comparison of delay distributions in indoor and outdoor environments.

Fig. 3. IEEE 4-node power system model.

In indoor experiments, the majority of delay samples cluster very tightly around 24–28 ms, with only small tails extending beyond 30 ms. In contrast, outdoor delays span a wider range: a large number of samples fall between 16–20 ms, and smaller clusters appear around 24–30 ms. This pattern explains why outdoor experiments have a slightly lower mean delay (because of the strong concentration of delays below 20 ms) but a larger overall variation. This pattern can be explained by the scenarios under which the experiments were conducted. The outdoor experiments took place during winter in Buffalo, NY, at approximately 0◦ C, near a UB parking lot during late afternoon, coinciding with typical end-ofoffice hours, when there was continuous movement of cars and pedestrians. Such mobility may introduce time-varying multipath reflections and intermittent micro-blockages, while low ambient temperatures may affect hardware performance. Collectively, these factors result in links that are generally faster (benefiting from line-of-sight propagation) but more dynamic, leading to increased variability in delay. By contrast, the indoor environment provided controlled temperature and minimal movement, resulting in a more static propagation setting. Although indoor propagation experiences additional penetration losses, which slightly elevate the average delay, the channel conditions are more stable over time, leading to lower jitter, smaller St.d., and a tighter Q1-Q3 range. Thus, indoor performance is slightly slower but more stable, whereas outdoor performance is slightly faster but more variable. Across all experimental results presented in Section IV, the highest mean delay observed for the 5G was 32.66 ms, occurring at a reporting rate of 0.5 fps (Section IV-A). At the same reporting rate, our previous LTE cat-M study reported a mean delay of 212.52 ms [6]. Thus, even under the worstcase delay observed for the 5G system, the 5G achieves approximately 6.5× lower mean delay than LTE cat-M.

streams that emulate synchrophasor traffic. While those experiments demonstrated the suitability of 5G for low-latency and lightweight SG communications, they did not incorporate realistic power system dynamics. In this section, we extend the analysis to a closed-loop cyber-physical setting by experimentally evaluating real-time SE using synchrophasor measurements obtained from an IEEE 4-node distribution feeder model implemented on a Typhoon HIL real-time simulator. This section first describes the real-time SE framework implemented using the proposed testbed, including the power system model and the mathematical formulation of the SE problem (Section V-A). It then presents an experimental validation of the SE framework under diverse operating conditions, including steady-state operation, dynamic load variations, and fault events (Section V-B).

V. R EAL - TIME S TATE E STIMATION F RAMEWORK AND E XPERIMENTAL VALIDATION OVER 5G N ETWORKS SE is a crucial analytical process in modern power systems that reconstructs the grid’s operational state by estimating voltage magnitudes and angles from available measurements. Accurate SE enhances system observability, supports load forecasting, improves stability assessment, and is frequently used as a foundation for event detection and control actions. In Section IV, we evaluated the communication performance of the proposed 5G-enabled SG nodes using simulated data

A. Real-Time SE Framework 1) Power System Model: The experiments were conducted using the testbed shown in Fig. 1, which is based on an IEEE 4-node distribution feeder model implemented on a Typhoon HIL 602+ real-time simulator. The feeder, shown in Fig. 3, consists of four buses interconnected through resistiveinductive (RL) line segments, capturing realistic voltage drops and phase-angle variations along the feeder. Power grid Node 1 interfaces with an external grid through a reference voltage source, setting the system voltage and frequency. Power flows from Node 1 to Node 2 via an RL line, from Node 2 to Node 3 through a distribution transformer, and from Node 3 to Node 4 through another RL segment. For this study, measurements were collected at Nodes 1, 2, and 4 using three RPi-based SG nodes as shown in Fig. 3. At Node 1, the measured quantities included voltage magnitude, current magnitude, and active power. At Node 2, voltage magnitude, voltage angle, and current magnitude were measured. At Node 4, voltage magnitude, voltage angle, and active power were acquired. All remaining electrical states in the system were treated as unknown and were estimated by the SE algorithm. 2) Mathematical SE Framework: We adopted the standard weighted least squares (WLS) formulation for the SE [28]. Let 𝑥 = [𝑉1 , 𝑉2 , 𝑉3 , 𝑉4 , 𝜃 2 , 𝜃 3 , 𝜃 4 ] ⊤ ,

(1)

denotes the system state vector, where 𝑉𝑖 and 𝜃 𝑖 represent the voltage magnitude and phase angle, respectively, at Node 𝑖.

6

Because Node 1 is the reference, its angle is fixed at 𝜃 1 = 0 and is omitted from the optimization variables. At each reporting instant 𝑘, the measurement vector 𝑧 consists of all synchrophasor quantities provided by the SG nodes: 𝑧 = [𝑉ˆ1 , 𝐼ˆ1 , 𝑆ˆ1 , 𝑉ˆ2 , 𝜃ˆ2 , 𝐼ˆ2 , 𝑉ˆ4 , 𝜃ˆ4 , 𝑆ˆ4 ] ⊤ ,

(2)

ˆ denotes a real-time measurement. All quantities were where (·) converted to per-unit before processing. Each measurement 𝑧 𝑖 is related to the system states through a nonlinear function ℎ𝑖 (𝑥) [28]: 𝑧𝑖 = ℎ𝑖 (𝑥) + 𝜀 𝑖 ,

(3)

where 𝜀 𝑖 represents measurement noise. The complete measurement model is written compactly as 𝑧 = ℎ(𝑥) + 𝜀.

(4)

WLS SE Problem: The estimator computes the state 𝑥 by minimizing the weighted squared residual [28]: Í𝑚 min 𝑥 𝐹 (𝑥) = 𝑖=1 𝑤 𝑖 [𝑧𝑖 − ℎ𝑖 (𝑥)] 2 = ∥𝑟 (𝑥)∥ 2𝑊 , s.t. (5) 𝑐 𝑖 (𝑥) = 0, for 𝑖 ∈ N𝑖𝑛𝑡 𝑥∈X where 𝑟 (𝑥) = 𝑧−ℎ(𝑥) is the residual vector, 𝑐 𝑖 (𝑥) are the power flow functions at internal nodes N𝑖𝑛𝑡 (e.g. Nodes 2 and 3 in Fig. 3), X is a convex set used to bound the states (typically upper and lower bounds), and 𝑊 = diag(𝑤 1 , . . . , 𝑤 𝑚 ) is the diagonal weighting matrix. The optimization enforces the AC power-balance equations at the interior nodes (Node 2 and Node 3), together with reasonable operating bounds on voltage magnitudes (0.5– 1.0 p.u.) and phase angles (−70◦ to 0◦ ). The problem is solved using MATLAB’s fmincon optimization solver with the Sequential Quadratic Programming (SQP) algorithm, which iteratively updates 𝑥 using the nonlinear measurement model and its Jacobian. Residual Norm and Fault Indicator: After each SE update at reporting instant 𝑘, the measurement residual is computed as 𝑟 (𝑘) = 𝑧meas (𝑘) − 𝑧 est (𝑘), (6) where 𝑧 est (𝑘) = ℎ(𝑥 ∗ (𝑘)) is the predicted measurement at the converged estimate 𝑥 ∗ (𝑘) and 𝑧meas (𝑘) is the measured value. The corresponding WLS cost function value is 𝐹 (𝑘) = ∥𝑟 (𝑘)∥ 2𝑊 .

(7)

The scalar value 𝐹 (𝑘) provides a global measure of mismatch between the system model and the measured data and is used as a fault-detection metric in this study. Under normal operating conditions, 𝐹 (𝑘) remains below a threshold value, while abrupt disturbances cause sharp increases. A fault is declared whenever 𝐹 (𝑘) > 𝑇,

(8)

where 𝑇 is a detection threshold selected based on the statistical behavior of 𝐹 (𝑘) under normal operating conditions.

TABLE III C OMMUNICATION P ERFORMANCE OF SG N ODES OVER THE BASELINE SE E XPERIMENT. SG node location per bus Node 1 Node 2 Node 4

Sent Received Lost Loss Mean frames frames frames rate (%) delay (ms) 30000 30000 0 0.00 25.38 30000 30000 0 0.00 31.93 30000 30000 0 0.00 27.54

Jitter (ms) 3.17 3.02 3.36

B. Experimental Validation of SE Under Diverse Operating Conditions To evaluate the SE performance, the measured synchrophasor signals are compared with their corresponding modelbased estimates at each reporting instant 𝑘. The frame ID serves as the reporting instant 𝑘. For each frame, we compute the absolute residual from (6), which captures the instantaneous mismatch between measured and estimated quantities. To quantify overall estimation accuracy across all frames, we use the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), defined as 1Õ |𝑧 meas (𝑘) − 𝑧est (𝑘)|, 𝑛 𝑘=1 v t 𝑛 1Õ (𝑧meas (𝑘) − 𝑧 est (𝑘)) 2 , RMSE = 𝑛 𝑘=1 𝑛

MAE =

(9)

where 𝑧meas (𝑘) and 𝑧est (𝑘) denote the measured and estimated values of a given signal at frame ID 𝑘, and 𝑛 is the total number of frames. To evaluate the performance of the SE framework, three experimental scenarios are considered: • SE under steady-state conditions, which assesses estimation accuracy during stable operation; • SE under dynamic load conditions, which evaluates SE performance under controlled load variations; • Fault detection under stable load conditions, which examines the ability of the SE framework to detect abrupt disturbances. All experiments were conducted indoors using the proposed testbed. Each SG node transmits 30000 synchrophasor frames over the commercial 5G network. A reporting rate of 50 fps was used for the steady-state condition and fault-detection experiments, while 10 fps was used for dynamic load scenarios. 1) SE Under Steady-State Conditions: This experiment serves as the baseline case for evaluating the intrinsic accuracy of the SE framework under stable operating conditions. No load variations or faults were introduced in this experiment. Table III summarizes the end-to-end communication performance during data collection. All three nodes experienced zero frame loss, indicating a reliable 5G link during the experiment. The average delay ranged from 25 to 32 ms, and jitter remained low (approximately 3 ms across all nodes), consistent with the results reported in Section IV. These stable network conditions ensure that the observed SE performance reflects the electrical behavior of the power system model rather than communication-induced effects. For each subfigure, the measured and estimated results are shown at the top, and the corresponding absolute residual

7

TABLE IV BASELINE SE ACCURACY ACROSS THE N ODES .

Node 1 Vmag (V)

Signal type Node 1 𝑉 𝑚𝑎𝑔 (V) Node 2 𝑉 𝑚𝑎𝑔 (V) Node 2 𝑉 𝑎𝑛𝑔 (deg) Node 4 𝑉 𝑚𝑎𝑔 (V) Node 4 𝑉 𝑎𝑛𝑔 (deg)

MAE 0.68 0.65 0.07 0.58 0.56

RMSE 0.78 0.69 0.07 0.74 0.69

7220

Vmag-meas

Vmag-est

7200 7180

7200 7198

7160

(a) 𝑉𝑚𝑎𝑔: meas vs. est

10000 0.5

10010 1

1.5

2

2.5

3 104

Frame ID

Error (V)

6

2 1 0 10000

4

10010

C OMMUNICATION P ERFORMANCE OF SG N ODES D URING SE E XPERIMENT U NDER DYNAMIC L OAD C ONDITIONS .

0 1

1.5

2

2.5

3 104

Frame ID

Vmag-est

7115 7100 7090

7110 10000 10010 0.5 1 1.5

10020 2 2.5

Error (V)

Frame ID 2 1 0 10000

4 2

10010

3 10

Node 2 Vang (deg)

Vmag-meas

7110

Vang-meas

0 -1

-0.4 -0.45 -0.5 10000

-2 -3

0

10020

0

0.4

1

2

Frame ID

(a) 𝑉𝑚𝑎𝑔: meas vs. est

3 104

0.2 0.1 0 10000

0.2

10020 2

0

10010

1

3 104

10020

2

Frame ID

SG node location per bus Node 1 Node 2 Node 4

Sent Received Lost Loss Mean frames frames frames rate (%) delay (ms) 30000 29999 1 0.01 37.13 30000 30000 0 0.00 33.57 30000 30000 0 0.00 30.69

Jitter (ms) 2.17 1.60 0.69

Vang-est

Frame ID

0 0

10010 1

4

Error (deg)

Node 2 Vmag (V)

Fig. 4. Baseline SE at Node 1: measured vs. estimated 𝑉𝑚𝑎𝑔 with residual errors. 7120

Fig. 6. Baseline SE at Node 4: measured vs. estimated (a) 𝑉𝑚𝑎𝑔 and (b) 𝑉𝑎𝑛𝑔 with residual errors.

TABLE V

10020

2

0.5

(b) 𝑉 𝑎𝑛𝑔: meas vs. est

10020

(b) 𝑉 𝑎𝑛𝑔: meas vs. est

3 104

Fig. 5. Baseline SE at Node 2: measured vs. estimated (a) 𝑉𝑚𝑎𝑔 and (b) 𝑉𝑎𝑛𝑔 with residual errors.

errors are shown at the bottom. The associated MAE and RMSE values are reported in Table IV. Fig. 4 shows the measured voltage magnitude (𝑉𝑚𝑎𝑔-𝑚𝑒𝑎𝑠) and estimated voltage magnitude (𝑉 𝑚𝑎𝑔-𝑒𝑠𝑡) over the total number of frames for Node 1. The results indicate that 𝑉𝑚𝑎𝑔-𝑒𝑠𝑡 closely aligns with 𝑉 𝑚𝑎𝑔-𝑚𝑒𝑎𝑠, as observed in both the estimation plot and the error plot. The residual error remains consistently small across all frames, with no noticeable deviations. The MAE and RMSE are approximately 0.68 V and 0.78 V, respectively, confirming strong estimation accuracy. This shows that the SE algorithm effectively estimated the measured voltage magnitude for Node 1. Fig. 5 shows the voltage magnitude (𝑉𝑚𝑎𝑔) and angle (𝑉 𝑎𝑛𝑔) estimation results for Node 2. As shown in Fig. 5(a), 𝑉𝑚𝑎𝑔-𝑒𝑠𝑡 closely follows 𝑉𝑚𝑎𝑔-𝑚𝑒𝑎𝑠. The error subplot shows that the residuals remain low and stable across the frames. The resulting MAE and RMSE for 𝑉 𝑚𝑎𝑔 are 0.65 V and 0.69 V, respectively. Similarly, the voltage angle results in Fig. 5(b) shows strong agreement between estimated (𝑉 𝑎𝑛𝑔-𝑒𝑠𝑡) and the measured values (𝑉 𝑎𝑛𝑔-𝑚𝑒𝑎𝑠). The residuals remain confined within a narrow band, and both the MAE

and RMSE values are 0.07 deg. Similar trends are observed for Node 4 in Fig. 6, where both 𝑉 𝑚𝑎𝑔-𝑒𝑠𝑡 and 𝑉 𝑎𝑛𝑔-𝑒𝑠𝑡 closely track their measured counterparts. The corresponding MAE and RMSE values are 0.58 V and 0.74 V for the 𝑉 𝑚𝑎𝑔, and 0.56 deg and 0.69 deg for 𝑉 𝑎𝑛𝑔, confirming consistently accurate estimation performance. 2) SE Under Dynamic Load Conditions: This experiment evaluates the ability of the SE framework to track time-varying system states under controlled load variations. A periodic square-wave load disturbance was applied at Node 4, switching every 100 s with a 50% duty cycle. Table V reports the communication performance during this experiment. Only a single lost frame was observed at Node 1 (0.01%), while the mean delay ranged from 30.69 ms to 37.13 ms. Jitter remained below 2.2 ms across all nodes. These results indicate stable 5G communication, ensuring that the observed variations in the estimated states are driven by applied load changes rather than by communication effects. The Node 1 results in Fig. 7 show that 𝑉 𝑚𝑎𝑔-𝑒𝑠𝑡 generally follows 𝑉𝑚𝑎𝑔-𝑚𝑒𝑎𝑠. The periodic load change every 100 s is not clearly visible at this node, likely because Node 1 is electrically close to the source in the test feeder and therefore less sensitive to downstream load variations. Although the load change is not directly observed in the measured voltage trajectory, slightly elevated estimation errors appear around the transition instants, as shown by the absolute percentage residuals in Fig. 7. This behavior is expected, as the SE algorithm jointly computes the system state using all network measurements, causing transient mismatches at nodes unaffected by the load change itself. This effect is reflected in the MAE (2.89 V) and RMSE (21.42 V) in Table VI, where a few large deviations near the switching instants disproportionately raise the RMSE. The influence of load variations is more pronounced at Nodes 2 and 4. The Node 2 results in Fig. 8(a) show that 𝑉𝑚𝑎𝑔-𝑒𝑠𝑡

MAE 2.89 2.99 0.06 2.81 0.13

RMSE 21.42 20.75 0.06 17.40 0.72

7202

Vmag-est

0.5

1

0.5 0 -0.5 10000

Error (%)

6 4

10010

2

2.5

3 104

0 1

1.5

2

2.5

3 104

Frame ID

7100 7000 6900

7400 7200

10400

10600

Vmag-est

10800

7000 0.5

1

1.5

2

2.5

Error (%)

Vang-meas

10400

10600

10800

Vang-est

-0.4 -0.6

0

10400

10600

10800

-0.5 0

1

2

3 104

Frame ID 100

4 2 0

5

0.5

104

Frame ID 10

3

Node 2 Vang (deg)

Vmag-meas

7600

Error (%)

Node 2 Vmag (V)

Fig. 7. SE performance at Node 1 during dynamic load conditions: measured vs. estimated 𝑉𝑚𝑎𝑔 with residual errors. 7800

40 20 0

50

10400

10600

10800

0

0 0

1

2

Frame ID

(a) 𝑉𝑚𝑎𝑔: meas vs. est

3 104

3

-30

10400 10600 10800

-40 0

1

2

3 104

Frame ID 40

10400 10600 10800

20

Vang-est

-40 -45

104

20 10 0

40

Vang-meas

-20

10 5 0

20

10400 10600 10800

0 1

2

3 104

0

1

2

Frame ID

(b) 𝑉 𝑎𝑛𝑔: meas vs. est

3 104

Fig. 9. SE performance at Node 4 during dynamic load conditions: measured vs. estimated (a) 𝑉𝑚𝑎𝑔 and (b) 𝑉𝑎𝑛𝑔 with residual errors.

10020

2 0.5

2

(a) 𝑉𝑚𝑎𝑔: meas vs. est

Frame ID 8

1

Frame ID

10020

1.5

10400 10600 10800

0

7200 7199

Vmag-est

Frame ID

0

10010

Vmag-meas

2000 1800 1600

2000

0

Vmag-meas

7200 7199 10000

7201

2500

Error (%)

Node 1 Vmag (V)

Signal type Node 1 𝑉 𝑚𝑎𝑔 (V) Node 2 𝑉 𝑚𝑎𝑔 (V) Node 2 𝑉 𝑎𝑛𝑔 (deg) Node 4 𝑉 𝑚𝑎𝑔 (V) Node 4 𝑉 𝑎𝑛𝑔 (deg)

3000

Error (%)

Node 4 Vmag (V)

TABLE VI SE ACCURACY D URING L OAD VARIATIONS .

Node 4 Vang (deg)

8

0

1

2

Frame ID

(b) 𝑉 𝑎𝑛𝑔: meas vs. est

3 104

Fig. 8. SE performance at Node 2 during dynamic load conditions: measured vs. estimated (a) 𝑉𝑚𝑎𝑔 and (b) 𝑉𝑎𝑛𝑔 with residual errors.

follows the general trajectory of 𝑉 𝑚𝑎𝑔-𝑚𝑒𝑎𝑠 and successfully captures the periodic load variations occurring every 100 s. However, several notable deviations occur around the loadswitching intervals, as highlighted in the residual error subplot. These points exhibit error levels of approximately 2–4%, corresponding to brief misalignment between the measured and estimated values during the state changes. The zoomed-in region of Fig. 8(a) confirms that the algorithm momentarily misestimates a small number of samples immediately following each load transition but quickly converges back to the correct trend. The resulting MAE and RMSE for 𝑉𝑚𝑎𝑔 are 2.92 V and 20.41 V, respectively; the elevated RMSE again reflects the influence of a small number of high-error samples during switching events, consistent with the behavior observed at Node 1. The 𝑉 𝑎𝑛𝑔 results in Fig. 8(b) show a similar trend. The 𝑉 𝑎𝑛𝑔-𝑒𝑠𝑡 closely follows the 𝑉 𝑎𝑛𝑔-𝑚𝑒𝑎𝑠 across the experiment. The residual error plot (bottom) of Fig.8(b) shows visually noticeable deviations, but this is primarily due to the small range of measured values (approximately -0.65 deg to -0.45 deg). As a result, even a minor estimation offset—for

example, estimating -0.64 deg instead of -0.65 deg— appears visually amplified when plotted over such a narrow range. The quantitative results confirm high accuracy, with an MAE of 0.06 deg and an RMSE of 0.06 deg, demonstrating the algorithm’s ability to reliably estimate 𝑉 𝑎𝑛𝑔 at Node 2 even under dynamic load variations. Similar behavior is observed for Node 4, as shown in Fig. 9. The 𝑉𝑚𝑎𝑔-𝑒𝑠𝑡 generally follows 𝑉𝑚𝑎𝑔-𝑚𝑒𝑎𝑠, with deviations occurring primarily at the load-switching points. The zoomedin view shows a few estimated samples that briefly overshoot or anticipate the transition before the estimator recovers. The corresponding MAE and RMSE values for 𝑉 𝑚𝑎𝑔 are 2.34 V and 17.57 V, respectively, again reflecting the influence of a small number of high-error samples during abrupt state changes. The 𝑉 𝑎𝑛𝑔 results exhibit a similar pattern, with small deviations during transitions and overall low error (MAE = 0.13 deg, RMSE = 0.70 deg). These results confirm that the estimator is generally accurate but exhibits brief misalignment during abrupt load transitions. As the downstream-most bus, Node 4 experiences the most pronounced effects of the switching events; nevertheless, the algorithm successfully tracks the system state, with only minor misestimations at the transition points. 3) Fault Detection Under Stable Load Conditions: This experiment evaluates the robustness of the SE framework in detecting abrupt electrical disturbances. Fault events were introduced into the IEEE-4 node feeder model by inserting a fault path between Nodes 3 and 4 via a controllable threephase switch. Activating the switch connects an additional impedance load to the feeder, producing fault-like behavior characterized by sharp current increases, voltage magnitude drops, and abrupt phase-angle deviations. Fault detection is performed using the WLS cost function value 𝐹 (𝑘), which provides a global measure of mismatch between the system model and the measured data. A fault is declared whenever 𝐹 (𝑘) exceeds a predefined threshold 𝑇, as defined in (7) and (8). The threshold is selected empirically based on the distribution of 𝐹 (𝑘) under normal operating conditions and is approximately 6 × 10−3 . Table VII summarizes the communication performance during the fault-detection experiment. All nodes maintained reliable connectivity with zero frame loss. The average delay

9

TABLE VII

VI. C ONCLUSION

C OMMUNICATION P ERFORMANCE OF SG N ODES D URING FAULT D ETECTION E XPERIMENT.

Node 2 Vmag (V)

SG node location per bus Node 1 Node 2 Node 4

Sent Received Lost Loss Mean frames frames frames rate (%) delay (ms) 30000 30000 0 0.00 31.28 30000 30000 0 0.00 30.70 30000 30000 0 0.00 28.86

Jitter (ms) 3.39 2.82 2.94

Vmag-meas

Vmag-est

7115 7110 7105 7100

F(k) value

×10-3

(a) Fault 2

Fault 1

0.01

15

0 5550

10

Fault 3

∆ID=40

5600

5 0.5

1

1.5

2

2.5

Frame ID

(b)

3 ×104

Fig. 10. Fault-detection SE results: (a) measured vs. estimated 𝑉𝑚𝑎𝑔 at Node 2 and (b) cost function values 𝐹 (𝑘 ) over the frame ID.

remained close to 30 ms, and jitter was approximately 3 ms, indicating a stable communication throughout the experiment. Three fault events are injected at frame IDs 5557, 14558, and 23556 during the experiment. Fig. 10(a) shows the voltage magnitude at Node 2 with three injected faults highlighted in magenta. Once a fault is inserted, clear deviations appear between 𝑉𝑚𝑎𝑔-𝑚𝑒𝑎𝑠 and 𝑉 𝑚𝑎𝑔-𝑒𝑠𝑡 at Node 2, where the faults are applied. Fig. 10(b) shows the cost function values 𝐹 (𝑘) over the frame IDs with annotation of the injected faults. Each injected fault produces a sharp increase in 𝐹 (𝑘) that clearly exceeds the threshold 𝑇. As illustrated in the zoomed inset of Fig. 10(b), Fault 1 is inserted at frame ID 5557 and detected at frame ID 5597, corresponding to a difference of 40 frames. Given a reporting rate of 50 fps (0.02 s per frame), this yields a detection delay of 0.80 s. The detection delays for all three faults are summarized in Table VIII. It is important to note that the phasor data were measured and transmitted at 50 fps, corresponding to a time resolution of 0.02 s (20 ms) per measurement. As discussed earlier, we used an external ADC and, for the fault-detection experiments, employed all four channels—three to read the signals from each node and one to record the fault-trigger signal. Because reading multiple ADC channels introduces processing delays, and due to our system implementation, we could not operate reliably above 50 fps while maintaining data transmission timing accuracy. As a result, detection delays are quantized in 20 ms intervals, and any faster changes occurring between consecutive samples cannot be observed in the recorded data. TABLE VIII FAULT D ETECTION D ELAY ACROSS THE FAULTS . Fault no. Fault insertion frame ID Fault detection frame ID ΔID Delay (s) 1 5557 5597 40 0.80 2 14558 14598 40 0.80 3 23556 23600 44 0.88

This paper presents a multi-node 5G-enabled SG testbed integrated with a Typhoon HIL real-time simulator, capable of transmitting real-time power system measurements over a commercial cellular network. The SG nodes are implemented using RPi devices equipped with 5G HAT modules to enable 5G connectivity. An extensive network performance evaluation was conducted on a U.S.-based commercial cellular network using KPIs such as delay, jitter, and frame loss to assess the suitability of 5G of SG communications. The results demonstrate consistent performance across both indoor and outdoor environments and show that 5G can reliably accommodate a wide range of reporting rates, from as low as 0.5 fps to as high as 120 fps, corresponding to a broad spectrum of SG applications. The experimental results further indicate that 5G provides significantly lower delay, reduced jitter, and improved reliability compared to LTE catM. In particular, the highest mean delay observed for the 5G system was 32.66 ms at a reporting rate of 0.5 fps, which is approximately 6.5× lower than the corresponding LTE cat-M result at the same reporting rate. We further experimentally validated real-time SE over a commercial 5G link using synchrophasor data from an IEEE 4-node feeder, evaluating estimation accuracy under both steady-state and dynamic load conditions. In addition, the proposed system demonstrates effective fault-detection capability, achieving detection delays as low as 0.80 s. These results confirm the suitability of commercial 5G networks for realworld SG monitoring, estimation, and protection tasks. The current testbed design relies on an external 16-bit ADC to convert the Typhoon HIL analog outputs into digital inputs for the RPi-based SG nodes. The accuracy of this conversion could be further improved by employing higherprecision ADC modules available on the market, which would likely enhance both load-tracking accuracy and fault-detection sensitivity.

R EFERENCES [1] M. Cosovic, A. Tsitsimelis, D. Vukobratovic, J. Matamoros, and C. Anton-Haro, “5G Mobile Cellular Networks: Enabling Distributed State Estimation for Smart Grids,” IEEE Communications Magazine, vol. 55, no. 10, pp. 62–69, 2017. [2] T. A. Zerihun, M. Garau, and B. E. Helvik, “Effect of Communication Failures on State Estimation of 5G-Enabled Smart Grid,” IEEE Access, vol. 8, pp. 112 642–112 658, 2020. [3] M. Cosovic, D. Vukobratovic, and V. Stankovic, “Linear state estimation via 5G C-RAN cellular networks using Gaussian belief propagation,” in 2018 IEEE Wireless Communications and Networking Conference (WCNC), 2018, pp. 1–6. [4] B. Holfeld, S. Jaeckel, L. Thiele, T. Wirth, and K. Scheppelmann, “Smart Grid Communications: LTE Outdoor Field Trials at 450 MHz,” in 2015 IEEE 81st Vehicular Technology Conference (VTC Spring). IEEE, 2015, pp. 1–5. [5] S. Shah, S. Koley, and F. Malandra, “Experimental End-To-End Delay Analysis of LTE cat-M With High-Rate Synchrophasor Communications,” IEEE Internet of Things Journal, pp. 1–1, 2023. [6] B. K. Dash, S. Shah, and F. Malandra, “Network Performance Analysis of Smart Grid Communications Over LTE cat-M,” in 2023 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), 2023, pp. 1–6.

10

[7] B. K. Dash, G. Thomas, A. Inaolaji, and F. Malandra, “ExTODS: Experimental 5G-Enabled Testbed for Distributed Optimization of Distribution Systems,” in 2025 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), 2025, pp. 1–7. [8] I. Kharchouf and O. A. Mohammed, “Controller Hardware-in-the-Loop Testbed of a Distributed Consensus Multi-Agent System Control under Deception and Disruption Cyber-Attacks,” Energies, vol. 17, no. 7, p. 1669, 2024. [9] I. Srivastava, S. Bhat, and A. R. Singh, “Smart Grid Communication: Recent Trends and Challenges,” Next Generation Smart Grids: Modeling, Control and Optimization, pp. 49–75, 2022. [10] N. Suhaimy, N. A. M. Radzi, W. S. H. M. W. Ahmad, K. H. M. Azmi, and M. A. Hannan, “Current and Future Communication Solutions for Smart Grids: A Review,” IEEE Access, vol. 10, pp. 43 639–43 668, 2022. [11] B. Appasani and D. K. Mohanta, “A Review on Synchrophasor Communication System: Communication Technologies, Standards and Applications,” Protection and Control of Modern Power Systems, vol. 3, no. 4, pp. 1–17, 2018. [12] H. Gharavi and B. Hu, “Scalable Synchrophasors Communication Network Design and Implementation for Real-Time Distributed Generation Grid,” IEEE Transactions on Smart Grid, vol. 6, no. 5, pp. 2539–2550, 2015. [13] M.-S. Kang, Y.-L. Ke, and J.-S. Li, “Implementation of smart loading monitoring and control system with ZigBee wireless network,” in 6th IEEE Conf. on Industrial Electronics and Applic., 2011, pp. 907–912. [14] B. K. Dash and J. Peng, “Zigbee Wireless Sensor Networks: Performance Study in an Apartment-Based Indoor Environment,” Journal of Computer Networks and Communications, vol. 2022, 2022. [15] F. Malandra, R. Pourramezan, H. Karimi, and B. Sansò, “Impact of PMU and Smart Meter Applications on the Performance of LTE-based Smart City Communications,” in 2018 IEEE 29th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), 2018, pp. 1–6. [16] F. Malandra, S. Rochefort, P. Potvin, and B. Sansò, “A case study for M2M traffic characterization in a smart city environment,” in Proceedings of the 1st International Conference on Internet of Things and Machine Learning, 2017, pp. 1–9. [17] F. Malandra, L.-O. Chiquette, L.-P. Lafontaine-Bédard, and B. Sansò, “Traffic characterization and LTE performance analysis for M2M communications in smart cities,” Pervasive and Mobile Computing, vol. 48, pp. 59–68, 2018. [18] R. Samoilenko, N. Accurso, and F. Malandra, “A Simulation Study on the Impact of IoT Traffic in a Smart-city LTE Network,” in 2020 IEEE 31st Annual International Symposium on Personal, Indoor and Mobile Radio Communications. IEEE, 2020, pp. 1–6. [19] P.-C. Hsieh, Y. Jia, D. Parra, and P. Aithal, “An Experimental Study on Coverage Enhancement of LTE Cat-M1 for Machine-Type Communication,” in 2018 IEEE International Conference on Communications (ICC), 2018, pp. 1–5. [20] M. Elsaadany, A. Ali, and W. Hamouda, “Cellular LTE-A Technologies for the Future Internet-of-Things: Physical Layer Features and Challenges,” IEEE Communications Surveys & Tutorials, vol. 19, no. 4, pp. 2544–2572, 2017. [21] D. Porcu, I. P. Chochliouros, S. Castro, G. Fiorentino, R. Costa, D. Nodaros, V. Koumaras, F. Brasca, N. Di Pietro, G. Papaioannou et al., “5G Communications as Enabler for Smart Power Grids: The Case of the Smart5Grid Project,” in IFIP International Conference on Artificial Intelligence Applications and Innovations. Springer, 2021, pp. 7–20. [22] V. G. Švenda, A. M. Stanković, A. T. Sarić, and M. K. Transtrum, “Influence of Communication Irregularities and Co-simulation on Hybrid Power System State Estimation,” in 2018 IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT-Europe). IEEE, 2018, pp. 1–6. [23] M. Cokic, I. Seskar, and D. Popovic, “Communication Congestion Impact on State Estimation for Smart Grid Infrastructure,” in 2021 29th Telecommunications Forum (TELFOR). IEEE, 2021, pp. 1–4. [24] A. Tsitsimelis, C. Kalalas, J. Alonso-Zarate, and C. Antón-Haro, “On the impact of LTE RACH reliability on state estimation in widearea monitoring systems,” in 2018 IEEE Wireless Communications and Networking Conference (WCNC). IEEE, 2018, pp. 1–6. [25] F. Aminifar, M. Fotuhi-Firuzabad, M. Shahidehpour, and A. Safdarian, “Impact of WAMS Malfunction on Power System Reliability Assessment,” IEEE Trans. on Smart Grid, vol. 3, no. 3, pp. 1302–1309, 2012. [26] C. Gu and P. Jirutitijaroen, “Dynamic State Estimation Under Communication Failure Using Kriging Based Bus Load Forecasting,” IEEE Transactions on Power Systems, vol. 30, no. 6, pp. 2831–2840, 2015.

[27] “IEEE Standard for Synchrophasor Data Transfer for Power Systems,” IEEE Std C37.118.2-2024 (Revision of IEEE Std C37.118.2-2011), pp. 1–79, 2024. [28] A. Abur and A. G. Exposito, Power System State Estimation: Theory and Implementation. CRC press, 2004.