Performance Analysis and Optimization of 5G-Based Wireless Sensor Networks for Industrial Applications

The phrase “Industrial WSN” indicates WSNs built 'to suit' in industrial plants and applications. These WSNs have different challenges and requirements compared to WSN applications in other fields (e.g., environment monitoring, smart home). Class of IWSN type of IWSNs can be categorized as [25]:

• Severe Operating Conditions: Many IWSNs work in some severe circumstances such as high/low temperature, humidity, vibration and electromagnetism interference or the existence of a variety of chemical substances or other dangerous media. This in turn needs to have strong and stable sensor nodes and communication protocols.
• Real-Time Application: A number of industrial applications require the real-time accesses to monitoring and control data. For time-critical operations, data transmission in IWSN networks must be low-latency and highly reliable.
• High Reliability and Availability: Downtime due to network failure in industrial environment may cost greatly, involving production standstill, safety risks and financial losses. IWSNs should be very dependable and available with fault-tolerant and redundant mechanisms.
• Safety: Cyber threats are increasingly making industrial control systems more vulnerable. The IWSNs must be secure to safeguard sensitive information and unauthorized access.
• Deterministic Communication: Some types of industrial applications, especially in the area of motion control and automation, require deterministic communication with guaranteed latency and jitter. This makes it necessary to employ technologies such as TSN in many cases.
• Integration with Industrial Protocols: IWSN may have to integrate with the already existing industrial communication protocols, for example, Modbus, Profibus, or EtherCAT.

Some IWSN Applications include [28]:

• Process Monitoring and Control: Temperature, pressure and the flow rate or other minor process parameters in manufacturing plants, chemical plants, or power stations.
• Machine Condition Monitoring: To capture vibration, temperature data of the machine so as to anticipate failures and schedule maintenance in advance.
• Industrial Environmental Monitoring: Scrap and recycle operations, air quality, noise and other environmental monitoring in factories or industrial installations.
• Asset tracking and management: Depicting the location and physical condition of machinery and assets in a plant or warehouse.
• Safety & Security: The use of wireless sensors for fire detection, gas leakage detection, intruder alarms and similar safety-critical applications.

Industrial WSNs are an active and significant research area focusing on the particular challenges and benefits of deploying WSNs in industrial environments. The deployment of IWSNs before and after integration with 5G is compared in Table 1 [22].

Table 1. Deployment of IWSNs before and after 5G integration

Figure 1(a) shows the internal structure and data flow of a WSN node in 5G Case. In the input layer, sensors inputs transduce environmental features; a signal conditioning stage may convert these signals into machine-readable form. The conditioned signals are A/D converted by the A/D Converter. This node is running on a battery that provides power to the essential components. Processed digital signals and power meet at the Low Power Microcontroller, which is the WSN node CPU. Here is where data management, control functions and low-power processing will happen. The microcontroller sends data to two major devices: 5G Transceiver and Flash Storage. The transceiver supports real-time data transfer to a 5G RAN gNB (Next Generation Node B), allowing quick communication between the other network entities. On the other, we store data locally in flash for further use or processing. This feature shows the inter-operation of classical WSN elements with 5G communication services, ensuring rapid and effective data relays in IoT and sensor network scenarios [8].

Figure 1(b) shows the deployment of a node in a small industrial network as one of the nodes. The other 5G WSN node, located outside the indicated industrial network, is connected to an industrial gateway through the 5G network. This is the gateway through which data first enters your internal network and flows out to a few critical elements. The data obtained for monitoring and control are transmitted to a SCADA system. The data is used in the automated control of industrial processes by a Programmable Logic Controller (PLC). A factory edge server can locally process data, resulting in quicker time to respond for real-time applications. The factory edge server is also linked to a cloud interface for data exchange with cloud-based services such as storage, analysis and remote access. This architecture describes how 5G may improve industrial networks, by the wireless evolution of sensor nodes and coexistence with current industrial (SCADA, PLC) systems and offloading computing to both the edge and cloud [12].

(a) The internal structure of an industrial WSN node

(b) A typical architecture for a 5G-based IWSN Infrastructure

Figure 1. Architectural components of Industrial Wireless Sensor Networks (IWSN)

A data packet generated by a WSN experiences multiple processing and transmission stages before reaching the AS. Each segment of the 5G communication architecture introduces specific delay components that collectively determine the overall latency.

4.1 Radio Access Network latency

The RAN enables wireless connectivity between WSNs and the 5G infrastructure via the next-generation Node B (gNB). The primary contributors to RAN latency are as follows.

Numerology and Symbol Duration: The duration of a transmission symbol is governed by the selected numerology and the associated SCS, and is given by:

Symbol_Duration = 1/(2^Numerology*15 kHz)     (1)

Scheduling and Retransmission Delays: Additional latency is introduced by radio resource scheduling mechanisms and HARQ retransmissions used for error recovery.

Accordingly, the total latency incurred within the RAN is expressed as:

RAN_Latency = Symbol_Duration+Scheduling_Delay + Retransmission_Delay     (2)

4.2 Transport Network latency

The TN interconnects the gNB with the CN using high-capacity wired or optical links. The latency within the TN consists of two main components.

Propagation Delay: This delay corresponds to the time required for signals to traverse the physical distance between network nodes and is defined as:

Propagation_Delay_TN = Distance_TN/Propagation_Speed_TN     (3)

where, Distance_TN represents the link length and Propagation_Speed_TN denotes the signal propagation speed.

Transit Delay: Transit delay accounts for processing, queuing, and transmission operations at intermediate routers and switches along the path, and is modeled as:

Transit_Delay_TN = Number_of_Nodes_TN * (Processing_Delay_TN + Queueing_Delay_TN+Transmission_Time_TN)     (4)

The aggregate latency introduced by the TN is therefore:

Transport_Latency = Propagation_Delay_TN+Transit_Delay_TN     (5)

4.3 Core network latency

The CN is responsible for user data routing and session management. Its latency contribution depends on both propagation effects and processing delays.

Propagation Delay: The propagation delay within the CN is expressed as:

Propagation_Delay_CN = Distance_CN/Propagation_Speed_CN     (6)

Transit Delay under MEC Deployment: When MEC is employed, the transit delay is primarily associated with the UPF and is given by:

Transit_Delay_CN_MEC = Processing_Time_UPF + Queueing_Delay_UPF + Transmission_Time_UPF     (7)

Transit Delay under Centralized Deployment: For centralized cloud-based deployments, additional traversal through intermediate UPF nodes increases the transit delay, which is modeled as:

Transit_Delay_CN_Cloud = 2 * (Transit_Delay_CN_MEC + Sum_of_Intermediate_UPF_Delays)     (8)

The total latency contributed by the CN can thus be written as:

Core_Latency = Propagation_Delay_CN + Transit_Delay_CN     (9)

4.4 Application servers latency

The AS processes incoming sensor data and generates corresponding responses. The latency at this stage depends on the computational workload and available processing capacity, and is expressed as:

Application_Server_Latency = (Processing_Cycles_Per_Bit * Packet_Size * Number_of_Packets) / Processing_Capacity_AS     (10)

4.5 Data aggregation at edge nodes latency

While data aggregation reduces the number of transmitted packets, it introduces additional processing and transmission delay at edge nodes. The aggregation latency is modeled as:

Aggregation_Latency = Aggregation_Delay + (Aggregated_Packet_Size/Link_Capacity)     (11)

4.6 Total end-to-end latency

The overall end-to-end latency experienced by a data packet is obtained by summing the latency contributions from all network segments and processing stages:

Total_End_to_End_Latency = RAN_Latency+Transport_Latency+Core_Latency + UPF_to_AS_Latency + Application_Server_Latency + Peering_Point_Latency+Aggregation_Latency     (12)

4.7 Energy model

To enhance the proposed latency model with energy considerations, we introduce a set of analytical expressions that estimate energy consumption in 5G-based industrial IWSNs. The energy required to transmit a data packet is given by:

E_tx = P_tx × (Packet_Size / Data_Rate)      (13)

where, P_tx is the transmission power (in watts).

In environments with interference, the retransmission energy overhead can be estimated as:

E_re = P_tx × (Packet_Size / Data_Rate) × P_retx     (14)

where, P_retx is the probability of retransmission.

For periodic sensing applications, the energy per sensing interval considering the duty cycle is modeled as:

E_interval = DC × P_active × T + (1 – DC) × P_sleep × T     (15)

where, DC is the duty cycle, P_active and P_sleep are the active and sleep power levels respectively, and T is the sensing interval.

4.8 Security mechanisms

The impact of security mechanisms on energy is included via:

E_sec = (P_enc + P_auth) × t_sec × Num_layers     (16)

where, P_enc and P_auth are the power consumed during encryption and authentication, t_sec is the processing time per operation, and Num_layers is the number of applied security layers.

The total energy per successfully transmitted packet becomes:

E_total = E_tx + E_re + E_sec + E_interval     (17)

These equations formalize the trade-offs discussed earlier between transmission power, duty cycle, security processing, and overall energy efficiency. They provide a foundation for future optimization studies that aim to minimize latency and energy consumption simultaneously in time-sensitive industrial scenarios.

The energy consumption model proposed here deals only with communication and processing costs, so it abstracts away some physical and environmental effects that drive the long-term operation of sensor nodes. Specifically, battery ageing and efficiency changes with temperature are not modeled explicitly. As opposed to the idealized assumption in this paper, in real its industrial deployments, battery capacity gradually degrades as charge–discharge cycles proceed, and energy efficiency also varies with ambient temperature, particularly under severe thermal conditions. Such effects can result in an increased effective energy consumption and a decreased node lifetime compared to what the model predicts. Detailed electrochemical battery degradation models provide additional improvement with prediction accuracy while additional thermal dynamics can also provide an overall improvement in prediction accuracy but this modeling would drastically increase complexity and require device-specific parameters that may not be available during system design [25-31]. As a result, this model is better seen as an upper-bound idealized estimate of energy performance, useful principally for comparative assessment and optimization at the system level but not for accurate lifetime prediction. In the future, we plan to expand the model to include battery ageing and influence of temperature employing calibration factor based on empirical or measurement data.

This sub-section presents an optimization for reducing end-to-end delay by optimizing SCS. Employing a constant traffic load, packet size and MEC deployment, the method analyzes several SCS configurations with respect to their effect on radio access and transport delays. We search for the best SCS by simulating each scenario and selecting the corresponding end-to-end latency, that is, as low as it can be. The simulation results show end-to-end latency is decreased as SCS value increases due to the very short symbol duration, leading lower radio access and transport delays (Figure 3).

Figure 3. subcarrier spacing (SCS) effect on delay components

6.1 Optimization of network deployment strategy for tot E2E reduction

The optimization example treats various network deployment options to minimize the end-to-end delay under fixed traffic load, sub-carrier spacing and packet size. The method determines all affecting latencies for each deployment alternative and chooses the policy that minimizes total latency. Where in the numerical results indicate that when MEC is deployed at gNB, the end-to-end latency decreases to the greatest extent as transport and CN latencies are minimized as shown in Figure 4.

Figure 4. Deployment strategy effect on delay components

6.2 Optimization of link capacity allocation (α) for tot E2E reduction

This optimization scenario aims to minimize the end-to-end latency subject to fixed SCS, packet size and traffic load by tuning the link capacity allocation parameter (α). This technique examines the impact on queuing and transport delays of varying values for α, and it chooses that value which minimizes the total delay.

Results show that by allocating larger capacity, queuing delay and transport delay is decreased which results in reducing the end-to-end latency as shown in Figure 5.

Figure 5. Link capacity allocation (α) effect on delay components

6.3 Optimization of duty cycle for maximizing battery life

In this example, we consider a low-power sensor node and the challenge of tuning its duty cycle to maximize its battery life. One way is to evaluate multiple duty-cycle options for a given packet size, battery capacity, and transmission power, and estimate the energy and battery life consumption. The results suggest that the battery lifetime is significantly enhanced for lower duty cycle, which shows the energy saving versus sensor activations compromise (Figure 6) [28-34].

Figure 6. Duty cycling effect

6.4 Optimization of transmission power for energy efficiency

This optimization example examines transmission power control to reduce energy consumption while preserving reliable communication. Using fixed packet size, transmission distance, and duty cycle, the approach evaluates multiple power levels and selects the configuration that minimizes energy usage while ensuring a packet success rate above 95%. The results show that a moderate transmission power provides the best trade-off between energy efficiency and communication reliability, see Figure 7 [29-34].

Figure 7. Transmission power effect

In response to the dynamic wireless network and to shed lights on learning-based optimization, we present a future-preparatory adaptive optimization framework for 5G-enabled industrial scenarios. The structure of the framework is an opportunistic base discretized in time, which starts with assigning a set of baseline system parameters (e.g., network load, network architecture, slice setup and radio resource allocation). The basic latency terms, i.e., RAN latency, transport latency and CN latency are calculated based on the analytical models introduced in this section.

At each time instance, the system state is described by observations on a number of measurable variables such as traffic load now, measured channel quality, length of queue and slice utilization and remaining energy. From the value of such a state, the optimization agent determines control decisions like SCS, link capacity assignment, transmission power and slice priority. We measure the effect of these actions through a refinement of the end-to-end latency and energy consumption. A reward function is then designed to penalize large latency, packet loss and energy consumption to drive the optimization process towards achieving low-latency and energy-efficient operating points.

Reinforcement learning methods, like Q-learning or deep Q-networks, may be used to approximate a mapping from states to control actions by interacting with the environmental model in an iterative way. The learning process itself adheres to the standard observe–act–reward–update loop and under typical assumptions of bounded rewards, sufficient exploration, and diminishing learning rates it hone in. The environment dynamics are characterized by stochastic traffic arrivals, interference and channel variation modeled as probabilistic processes using discrete event simulation for numerical evaluation.

A simulation environment built using SimPy, MATLAB, or python scientific libraries can be introduced for simulating the time evolution of a network and to evaluate how well we learned our policies. Furthermore, queuing-theoretic models are employed to verify the simulated latency behavior analytically under some simplifying assumptions. Such a joint analytical-and-learning-based framework offers a solid foundation for dynamic optimization while ensuring tractability and clarity, as well as supports systematic characterization of adaptation policies to industrial 5G setups.

One important advantage of the 5G latency model we derive is its ability to satisfy different environmental and network conditions. In this section, the basic model is enhanced by incorporating such parameters as mobility, interference, attenuation, number of IWSNs, congestion and security. Each factor incurs a delay, which causes additional delays on the network performance. The accuracy of the model is sufficiently adaptable to support accurate latency estimation in 5G practical settings. The final end-to-end latency with additional considerations is given by:

TotE2E = RANLat + TPProp + TPTran + CNProp + CNTran + UPFASLat + ASLat + PPLat + I_Delay + M_Delay + A_Delay + IWSN_Delay + C_Delay + S_Delay     (18)

where,

• I_Delay = Interference-induced delay
• M_Delay = Mobility-induced delay
• A_Delay = Attenuation-induced delay
• IWSN_Delay = Delay due to the number of IWSNs
• C_Delay = Congestion delay
• S_Delay = Security processing delay

Mobility introduces handover delays as users move between cells. The handover delay is given by:

M_Delay = Handover_Rate * Handover_Time     (19)

where,

• Handover_Rate = Number of handovers per second
• Handover_Time = Delay introduced by each handover

Interference causes retransmissions, increasing total latency. The delay due to interference is modeled as:

I_Delay = P_retx * (TransTime_TN + TransTime_CN)     (20)

where, P_retx = Probability of retransmission due to interference

In order to enhance the realism of the interference and attenuation models, we further modify the retransmission and path loss formulations to include statistical channel effects. In short-range industrial wireless with rich multipath, the instantaneous signal-to-noise ratio (SNR) is often modeled by Rayleigh fading assumption. Under this model, the instantaneous SNR (i.e. SNR_inst) can be expressed as a random variable having an exponential distribution:

f_SNR_inst(SNR_inst) = (1 / SNR_avg) * exp(-SNR_inst / SNR_avg)     (21)

where, SNR_avg is the average signal to noise ratio. Where ρ is the probability that the instantaneous SNR does not go below the minimum decoding threshold SNR_th, which corresponds to the packet error probability:

P_error = 1 - exp(-SNR_th / SNR_avg)     (22)

Assuming that packet errors result in retransmissions, the expected retransmission probability becomes:

P_retx = 1 - exp(-SNR_th / SNR_avg)     (23)

Signal attenuation affects transmission power and requires retransmissions or error correction, adding delay:

A_Delay = Path_Loss_Factor * (TransTime_TN + TransTime_CN)     (24)

where, Path_Loss_Factor = Signal degradation factor due to path loss.

To account for large-scale attenuation and shadowing effects, the path loss is modeled using the log-distance path loss model with shadow fading:

PathLoss = PathLoss_ref + 10 * PathLoss_Exp * log10(Distance / Distance_ref) + Shadowing      (25)

where, PathLoss_ref is the reference path loss at distance Distance_ref, PathLoss_Exp is the path loss exponent, and Shadowing is a zero-mean Gaussian random variable representing large-scale shadow fading.

The presence of multiple IWSNs influences latency due to contention and routing overhead:

IWSN_Delay = Num_IWSN * Routing_Overhead     (26)

where,

• Num_IWSN = Number of IWSNs in the network
• Routing_Overhead = Average delay per IWSN hop

Network congestion impacts queuing delays at nodes, increasing overall latency:

C_Delay = (ρ_CN) / (µ_CN * (1 - ρCN))     (27)

where,

• ρ_CN = λ_CN / µ_CN (Traffic intensity in the CN)
• λ_CN = Packet arrival rate in the CN
• µ_CN = Service rate of the CN

Security mechanisms (e.g., encryption, authentication) introduce additional processing overhead:

S_Delay = (Enc_Time + Auth_Time) * Num_Security_Layers      (28)

where,

• Enc_Time = Encryption processing time
• Auth_Time = Authentication processing time
• Num_Security_Layers = Number of applied security mechanisms

Table 7 presents the impact of each factor on total latency.

Table 7. Impact of environmental and network factors on total 5G latency

Table 8. Security protocol overhead calculated using the proposed latency model

The extended 5G latency model is successfully combines multiple factors, indicating its flexibility and adaptability. The results suggest that among the investigated latency factors, mobility, interference and congestion contribute most to total latency, security mechanisms bring a noticeable processing overhead for high security levels. It will be more realistic considering the attenuation, and IWSNs also make network performance prediction more complete. This model can be further extended with more parameters to achieve even better accuracy in practical 5G systems.

In order to study the impact of the security mechanisms on the network attributes, latency and energy efficiency, Table 8 shows a summary of tailored and deployed protocols at each protocol layer. Therefore, in order to make the security latency analysis more realistic and analysis-friendly, security-induced delay is modeled in a phased manner, which clearly separates the handshake overhead (session initialization) from the data-plane again (per-packet) overhead [20]. As observed in practical deployments of 5G-based industrial WSNs, handshake operations (key exchange, authentication, and security context establishment) are performed once per session, while encryption and integrity protection in the data plane are performed continuously for each packet being transmitted [28]. For long-lived industrial flows or short-lived control transactions where the impact is high, aggregating these two phases into a single total delays substantially over- or under-estimates their contribution to latency. Thus, the overall security delay is defined as the summation of one-time handshake latency and data-dependent processing latency, which is parameterized using the experimental 5G, IoT and industrial security measures available in the study [34]. The layered approach facilitates the precise incorporation of security overhead into the end-to-end latency model while allowing for meaningful analysis of security strength vs. real-time responsiveness.

Industrial environments are inherently dynamic, characterized by fluctuating network loads, interference, mobility, and rapidly changing service demands. To validate the reliability and generalizability of the proposed end-to-end latency model under such realistic conditions, we extended our analytical and simulation framework to incorporate traffic burstiness, user mobility, and interference-induced retransmissions.

We adopted an M/M/1 queuing model to simulate the impact of dynamic traffic patterns on queuing delay. Packet arrivals were modeled as a Poisson process with exponentially distributed inter-arrival times, while service times were assumed to follow an exponential distribution. The average queuing delay (Q_delay) is calculated using the formula:

Q_delay = λ / [μ * (μ - λ)]     (29)

where, λ denotes the average packet arrival rate and μ represents the service rate.

Simulations were conducted for λ values ranging from 100 to 1500 packets per second, while the service rate μ was held constant at 1600 packets per second. As illustrated in Figure 8, queuing delay remained negligible at low traffic loads but increased nonlinearly as λ approached μ, with a sharp rise observed beyond 1300 packets per second.

Figure 8. Queuing delay vs. traffic load based on M/M/1 model

To model mobility scenarios like as sensors on automated guided vehicles (AGVs), we added handover delay entries based on 3GPP specified (3GPP, 2013d) transition times 3GPP TS 38.331 [5] specifies handover execution times in the order of 40 ms - 50 ms for example one handover every 20 seconds (Handover_Rate = 0.05) and Handover_Time = 50 milliseconds give a contribution of 2.5 milliseconds per second of communication, M_delay.

Industrial wireless channels usually undergo degrading effects from wiring in metallic industrial machine systems and buildings. Under such conditions, packet loss is increased, requiring retransmissions. We consider the case of P_retx = 0.2 and T_TN + T_CN = 12 milliseconds, the extra latency is 2.4 milliseconds.

Table 9. The estimated latency overheads introduced by each dynamic factor

A composite simulation with respect to the above dynamic factors was conducted, where traffic was of Poisson nature having λ values varying between 200 and 1600 packets/sec, the Handover was at one every 30 seconds and interference induced retransmission probability was varied between 10% and 20%. Dynamic factor Estimated latency overheads (ms) Table 9.

Despite the presence of these dynamic influences, the overall deviation in latency predictions remained within ±10% when benchmarked against results reported in recent empirical studies [24, 27], affirming the robustness and adaptability of our model in real-time industrial deployments.

In this sense, we enhance the latency model by considering both resource isolation and priority scheduling introduced by 5G network slicing to quantify their impact on the performance of an industrial WSN. Resource isolation at the system level is reserved for part of the network resources for the industrial slice as:

Slice_Resource = Slice_Fraction * Total_Resource     (30)

Priority scheduling is abstracted by a scheduling gain factor that reduces effective queuing and access delay:

Effective_Delay = (Base_Delay / Sched_Gain) + (1 - Slice_Fraction) * Contention_Delay      (31)

where, Base_Delay is the base-line end-to-end delay without sliceing, Contention_Delay is background traffic contend delay, Slice_Fraction is the reserved resource fraction, and Sched_Gain > 1 is the industrial slice priority.

Table 10. Impact of network slicing on IWSN latency

Three representative slicing configurations are evaluated using Base_Delay = 18 ms and Contention_Delay = 10 ms. As shown in Table 10, the results demonstrate the effective latency decrease in response to our increase in Slice_Fraction and Sched_Gain, which reduces the effective latency from about 21 ms to 12 ms (a decrease of ≈ 43%). This shows that 5G NS can improve latency determinism and performance isolation of industrial traffic [20].

This section makes a comparative analysis of our work with other recent contributions in the area of 5G-IWSNs to underline the uniqueness and benefits of the proposed framework. Although many papers investigated the latency, energy saving and MEC deployment, the existing works address these topics independently. Table 16 shows a structured comparison between six representative studies [37-41], as well as us. We compare each of works for their modelling of latency, integration of energy efficiency, support of MEC, real-time capability, and application domains. In contrast to previous works, our study establishes a comprehensive end-to-end latency model side-by-side with a power-aware analytical framework and deployment flexibility (MEC@Edge, Core and Cloud) validated through simulations and case-studies across a variety of industrial scenarios.

This comparison illustrates that our work uniquely provides:

• A complete analytical latency breakdown across the 5G architecture (including RAN, TN, CN, and AS).
• An integrated energy model capturing transmission, sleep cycles, retransmission, and security.
• Flexibility in deployment architecture, supporting MEC at multiple levels and centralized cloud.
• Validation through simulation, theorem-driven trade-off analysis, and multiple industrial case studies (e.g., predictive maintenance, AGVs, environmental monitoring, and smart grids).

Compared to the cited literature, our framework is among the first to present a holistic performance optimization strategy for 5G-IWSNs, aligning with the operational requirements of Industry 4.0 and enabling systematic deployment planning.

Table 16. Comparative analysis of recent 5G-IWSN studies