HHO-SFC An Energy-Aware Service Function Chaining Framework for ECG Monitoring in Fog–Cloud IoT Systems

Authors
Mohammad Saleh Arvari 1 iD
Sam Jabbehdari 1 iD
Seyyed Hamid Haj Seyyed Javadi 3 iD
Affiliations
1Department of Computer Engineering, NT.C., Islamic Azad University, Tehran, Iran
2Department of Computer Engineering, CT.C., Islamic Azad University, Tehran, Iran
3Department of Computer Engineering, Shahed University, Tehran, Iran
Overview

Abstract

Distributed electrocardiogram (ECG) monitoring requires coordination of processing and communication resources across wearable devices, fog servers, and cloud infrastructure. This study presents HHO-SFC, a service function chaining framework that uses Harris Hawks Optimization to organize ECG preprocessing, feature extraction, alert generation, secure transmission, and cloud analytics. The framework considers latency, energy consumption, and resource utilization. A simulation-based evaluation uses ECG traces from the MIT-BIH Arrhythmia Database and a network configuration comprising 1,000 sensors and 50 fog servers. Three operating scenarios are considered: normal monitoring, emergency surge, and rural deployment. Under normal conditions, the tabulated HHO-SFC latency is 62.3 ms and per-sensor energy consumption is 10.2 J, corresponding to reductions of 16.7% and 18.4%, respectively, relative to the MO-DRL comparator. Under emergency conditions, HHO-SFC has lower energy consumption and higher reported reliability than the three comparators, but its latency of 116.1 ms exceeds both the 100-ms design target and the 112.1-ms value of the Chaotic-LFL-DQL comparator. Rural latency is also slightly higher than that comparator, although energy consumption is lower. These descriptive results indicate an energy–latency trade-off rather than uniform superiority. The evidence is limited to aggregate simulation outcomes and does not establish clinical effectiveness, statistical significance, or guaranteed real-time performance.

ECG monitoring; fog–cloud computing; service function chaining; Harris Hawks Optimization; energy consumption; Internet of Things

Introduction

Wearable ECG monitoring produces continuous data streams that must be processed and transmitted under limited computational and communication resources. Cloud-based architectures support centralized storage and analysis, but transmission to remote infrastructure adds communication delay and makes performance dependent on network conditions. Fog computing distributes processing closer to data sources and provides an intermediate layer between wearable devices and cloud services [, , ].

Service function chaining (SFC) represents processing as an ordered sequence of virtual network functions (VNFs), consistent with the SFC architectural framework []. In an ECG monitoring application, the chain can include signal preprocessing, feature extraction, anomaly screening, encryption, and analytics. The placement of these functions affects communication volume, processing delay, and energy use. Placement decisions must therefore consider the order of the functions and the resources available at each execution node [, ].

Learning-based methods have addressed dynamic placement and orchestration. Multi-objective deep reinforcement learning supports SFC placement [], while federated reinforcement learning distributes orchestration-related learning []. These approaches provide relevant methodological comparators, although training costs, deployment assumptions, and adaptation mechanisms differ from population-based optimization. Meaningful comparisons require equivalent workloads and clearly defined computational budgets.

Harris Hawks Optimization (HHO) is a population-based metaheuristic that combines exploration and exploitation []. The present framework applies HHO to ECG-oriented service placement in a fog–cloud architecture. Its contribution is the application-level integration of the service chain and a descriptive evaluation across three network scenarios. Lévy-flight search is treated as part of the HHO family rather than, by itself, a novel algorithmic contribution. The analysis focuses on infrastructure performance and does not evaluate a clinical diagnostic system.

Proposed Framework

Architecture and service chain

The architecture comprises wearable ECG sensors, nearby fog nodes, and a cloud data center. Sensors supply ECG data to fog nodes, where early processing and alert screening can be performed. Cloud resources support storage and computationally intensive analysis. The logical chain contains five functions: preprocessing, feature extraction, alert generation, secure transmission, and cloud analytics. HHO acts as the placement controller; it is not an additional ECG processing function.

Logical service-chain functions

Function

Purpose

Execution role

VNF1

Signal preprocessing

Decode sampled ECG data and reduce noise

VNF2

Feature extraction

Estimate R–R intervals and other ECG features

VNF3

Alert screening

Apply configured abnormality rules

VNF4

Secure transmission

Protect fog–cloud communication

VNF5

Storage and analytics

Store data and support subsequent analysis

Placement determines the nodes used by successive functions and the links carrying intermediate data. Processing at fog nodes can reduce cloud traffic, but local processing also consumes energy and finite capacity. The model therefore considers both execution and communication costs. Immediate alert screening and complete-chain completion are separate logical events; the available results do not provide separate measurements of their latency.

Analytical placement model

Let G = (V, E) denote the infrastructure graph, where V contains eligible execution nodes and E contains communication links. A request r contains an ordered set of functions k = 1, …, Kᵣ. Let xᵣₖᵢ be a binary variable equal to one when function k of request r is assigned to node i. Each function requires one eligible execution node:

Σᵢ xᵣₖᵢ = 1 for every request r and function k.

For resource type m, let aᵣₖₘ denote a function’s resource demand and Cᵢₘ the available capacity at node i. Resource feasibility is expressed as:

Σᵣ Σₖ aᵣₖₘ xᵣₖᵢ ≤ Cᵢₘ for every node i and resource type m.

Let yᵣₖₑ indicate whether link e carries traffic between successive functions k and k + 1, with data rate Dᵣₖ. The aggregate traffic assigned to a link must not exceed its bandwidth capacity:

Σᵣ Σₖ Dᵣₖ yᵣₖₑ ≤ BWₑ for every link e.

Routes must connect the assigned nodes in chain order. At a node i, the outgoing-minus-incoming flow for each successive function pair equals xᵣₖᵢ − xᵣ,ₖ₊₁,ᵢ. This condition links routing to placement. End-to-end delay includes processing, transmission, propagation, and queueing components:

Lᵣ = Σₖ Σᵢ xᵣₖᵢ tᵣₖᵢ + Σₖ Σₑ yᵣₖₑ (t_tx,ᵣₖₑ + t_prop,ₑ + t_queue,ᵣₖₑ).

A 100-ms latency value is used as an engineering target rather than a universal clinical threshold or an achieved guarantee. A hard constraint Lᵣ ≤ 100 ms would exclude any placement that violates it. Since the emergency-scenario mean exceeds this value, the reported outcomes are interpreted as target-based performance observations rather than proof of hard-constraint feasibility.

A dimensionless weighted objective can express the intended energy–latency–utilization trade-off:

J = 0.5(E_total / E_ref) + 0.3(L / L_ref) + 0.2U.

Here, E_ref and L_ref are positive reference scales and U is a dimensionless utilization measure. The weights correspond to the controller specification. The normalized expression and resource constraints define the analytical formulation. The evaluation does not isolate the effects of normalization, constraint handling, or objective weights.

Energy accounting

Energy accounting distinguishes computation, communication, and idle operation over a common observation interval. A dimensionally consistent model is:

E_total = Σᵢ [P_cpu,ᵢ t_proc,ᵢ + P_idle,ᵢ t_idle,ᵢ] + Σₑ P_tx,ₑ t_tx,ₑ.

For a transmitted data volume Bₑ in bits and effective data rate Rₑ in bits per second, t_tx,ₑ = Bₑ / Rₑ. Multiplying power by transmission duration yields joules. Propagation delay is not substituted for radio transmission duration. Receive energy, retransmissions, encryption, orchestration, and migration also contribute when those operations occur and must be assigned consistently to avoid double counting.

Per-sensor energy is an accounting quantity that depends on the system boundary and observation interval. The aggregate results report joules per sensor without specifying the duration or allocation of infrastructure energy. Accordingly, energy values support within-scenario descriptive comparisons but cannot be converted to wearable battery life.

ECG preprocessing and feature extraction

The simulation describes ECG streams resampled to 250 Hz and a nominal sensor resolution of 12 bits. These are simulation settings, distinct from the native MIT-BIH recordings. The preprocessing chain specifies a 30-tap FIR low-pass filter with a 40-Hz cutoff followed by four-level Daubechies-4 wavelet denoising. Quantitative claims about signal-to-noise improvement or diagnostic fidelity are not evaluated by the infrastructure results.

For a universal wavelet threshold, T = σ√(2 ln N), where σ denotes the noise standard deviation and N is the number of samples []. Filter design and boundary handling affect timing and waveform morphology. Buffering and processing windows also contribute to total response time and must be distinguished from network transfer delay.

R-peak detection can use a Pan–Tompkins-type procedure []. For an R–R interval measured in milliseconds, the heart-rate calculation is:

HR (beats/min) = 60,000 / RR_ms.

A Bazett-type QT correction can be written as QTc_ms = QT_ms / √(RR_ms / 1000). This expression specifies units but does not validate QT measurement or a clinical decision threshold. ST-segment analysis requires an appropriate ST measurement point, a reference baseline, and calibrated amplitude units; the T-wave offset is not used as a substitute for the ST measurement point. The present evaluation does not establish diagnostic performance for these features.

Alert screening and orchestration

The proposed alert function applies configured abnormality rules to the extracted features and can route notifications to a patient-facing application or another authorized destination. Synthetic bradycardia and ventricular-tachycardia patterns represent workload events in the simulation. They do not establish sensitivity, specificity, false-alarm frequency, or the suitability of automatic emergency-service notification.

The controller specification uses 50 HHO candidates and a maximum of 100 iterations. Candidate solutions represent alternative service placements. HHO combines exploration with exploitation through its established search strategies []. Discrete placement requires a mapping from candidate vectors to eligible nodes and a mechanism for rejecting or repairing infeasible assignments. Those implementation details, migration triggers, and controller execution overhead are not identifiable from the aggregate performance tables.

Dynamic orchestration can affect an ongoing stream through reconfiguration delay and state transfer. The reported infrastructure results do not demonstrate that optimization completes within the latency target, nor do they establish a global optimum. The results provide no convergence-time or global-optimality guarantee.

Secure transmission and cloud services

The architecture specifies AES-128 encryption and TLS 1.3 for fog–cloud communication. TLS 1.3 protects records using authenticated encryption with associated data; its use of hash-based functions in key derivation and handshake processing should not be described as a separate HMAC-SHA256 record-integrity mechanism []. Cipher-suite selection, authentication, key management, and endpoint protection determine the properties of an actual deployment.

Encryption is a component of data protection and does not independently demonstrate regulatory compliance. The simulation does not establish compliance certification or comprehensive security assurance. Similarly, cloud-based CNN–LSTM analysis and longitudinal risk prediction are architectural extensions rather than validated diagnostic outcomes in this study. No classification-accuracy or mortality-reduction claim is made.

Simulation Evaluation

Configuration and scenarios

The evaluation uses trace-driven simulation using iFogSim for fog–cloud resource management [] and ns-3 for network modeling []. The configuration includes 1,000 wearable sensors, 50 fog servers, and a centralized cloud data center. MIT-BIH provides the ECG traces []. The database contains 48 approximately half-hour recordings from 47 subjects, originally sampled at 360 Hz; the simulated streams are resampled to 250 Hz []. Replicated streams represent concurrent workload, not 1,000 independent patients.

Reported infrastructure and workload settings

Component

Configuration

Sensors

1,000 devices; 250-Hz streams; nominal 12-bit resolution; 100–500 mAh battery capacity

Fog layer

50 servers; CPU clock ranges of 1–5 GHz; 4–16 GB RAM

Cloud layer

100 vCPUs; 384 GB RAM

Backhaul

4G/5G; 10–100 Mbps; 20–50 ms round-trip time

Local links

Nominal Wi-Fi 6 rate of 1 Gbps and BLE rate of 2 Mbps

HHO controller

Population size 50; maximum 100 iterations

Normal monitoring

10–50 packets/s; 95% connectivity

Emergency surge

100–200 packets/s; 85% connectivity

Rural deployment

20–30% packet loss; 50–100 ms latency

Link rates are nominal configuration values, not measured application throughput. The configuration does not identify whether packet rates are per sensor or aggregate, and it does not fully specify packet payloads, simulator coupling, queue disciplines, execution demands, or simulation duration. These limits constrain reproducibility and interpretation of the numerical results.

Comparators and analysis

The comparator configurations are labeled MO-DRL, FRL, and Chaotic-LFL-DQL. The first two labels refer to the methodological families described in [, ]. Chaotic-LFL-DQL has no verified bibliographic attribution in this study. Results are therefore interpreted as comparisons among simulation configurations, without claiming faithful reproduction of a published Chaotic-LFL-DQL algorithm.

The analysis uses the aggregate values in Tables 3–5. Percentage reduction for latency or energy is calculated as 100 × (comparator − HHO-SFC) / comparator. Positive values indicate a reduction, and negative values indicate a higher HHO-SFC value. Independent run-level observations are unavailable, so confidence intervals, hypothesis tests, and effect sizes are not reported. Reliability percentages are retained as reported metrics; their event definition and denominator are not supplied and must not be interpreted as clinical safety probabilities.

Normal monitoring

Performance under normal monitoring

Metric

MO-DRL

FRL

Chaotic-LFL-DQL

HHO-SFC

Latency (ms)

74.8

87.9

68.2

62.3

Energy per sensor (J)

12.5

14.6

11.8

10.2

Throughput (Mbps)

3.8

3.1

4.2

4.8

Latency reduction (%)ᵃ

16.7

29.1

8.7

—

Energy reduction (%)ᵃ

18.4

30.1

13.6

—

ᵃ Reduction achieved by HHO-SFC relative to the comparator named in each column; a negative value denotes a higher HHO-SFC value.

Under normal monitoring, HHO-SFC has the lowest latency (62.3 ms) and energy consumption (10.2 J) among the tabulated configurations. Relative to MO-DRL, latency and energy are lower by 16.7% and 18.4%, respectively. Relative to Chaotic-LFL-DQL, the corresponding reductions are 8.7% and 13.6%. Throughput is 4.8 Mbps, compared with 3.1–4.2 Mbps for the comparators. These values are descriptive point estimates.

Emergency surge

Performance under emergency surge

Metric

MO-DRL

FRL

Chaotic-LFL-DQL

HHO-SFC

Latency (ms)

152.4

188.2

112.1

116.1

Energy per sensor (J)

16.8

19.5

16.1

13.2

Reported reliability (%)

78.4

68.3

82.1

88.5

Throughput (Mbps)

2.1

1.5

2.4

2.5

Latency reduction (%)ᵃ

23.8

38.3

-3.6

—

Energy reduction (%)ᵃ

21.4

32.3

18.0

—

ᵃ Reduction achieved by HHO-SFC relative to the comparator named in each column; a negative value denotes a higher HHO-SFC value.

Under emergency surge, HHO-SFC latency increases to 116.1 ms. This value is 23.8% lower than MO-DRL and 38.3% lower than FRL, but 3.6% higher than Chaotic-LFL-DQL. Energy consumption is 13.2 J, representing reductions of 21.4%, 32.3%, and 18.0% relative to the three comparators. Reported reliability is 88.5%, and throughput is 2.5 Mbps. The average latency exceeds the 100-ms engineering target; a separate sub-100-ms alert-path result cannot be inferred from this table.

Rural deployment

Performance under rural deployment

Metric

MO-DRL

FRL

Chaotic-LFL-DQL

HHO-SFC

Latency (ms)

108.4

132.5

84.1

84.4

Energy per sensor (J)

13.8

15.2

13.1

10.9

Reported reliability (%)

81.2

71.5

84.3

86.2

Throughput (Mbps)

2.7

2.0

3.0

3.4

Latency reduction (%)ᵃ

22.1

36.3

-0.4

—

Energy reduction (%)ᵃ

21.0

28.3

16.8

—

ᵃ Reduction achieved by HHO-SFC relative to the comparator named in each column; a negative value denotes a higher HHO-SFC value.

In the rural scenario, HHO-SFC latency is 84.4 ms, compared with 84.1 ms for Chaotic-LFL-DQL. This 0.3-ms difference represents a 0.4% increase rather than an improvement. Relative to MO-DRL and FRL, latency reductions are 22.1% and 36.3%. Energy consumption is 10.9 J, lower than all three comparators; reported reliability is 86.2%, and throughput is 3.4 Mbps. These are descriptive latency differences; statistical equivalence and optimality were not tested.

Discussion

The consistent descriptive advantage of HHO-SFC is lower reported energy consumption across the three scenarios. Latency performance is more context dependent: the framework has the lowest normal-scenario latency but does not minimize latency in the emergency or rural scenarios. These patterns are compatible with a weighted objective that trades energy against delay, although the aggregate outcomes cannot identify which algorithmic component caused the observed differences.

Emergency traffic increases HHO-SFC latency from 62.3 to 116.1 ms while throughput decreases from 4.8 to 2.5 Mbps. Thus, the scenario comparison identifies a practical limitation for time-sensitive processing under load. Lower mean latency than some comparators is insufficient to establish a deadline guarantee. Such a guarantee requires request-level timing, including high-percentile delay, missed deadlines, unsuccessful deliveries, and the distinct alert and full-chain paths.

The principal limitations concern reproducibility, attribution, and validation. Aggregate tables do not provide the independent observations required for statistical inference. Baseline configurations, the Chaotic-LFL-DQL implementation, energy accounting intervals, and reliability definitions are incompletely documented. The analytical formulation cannot be independently reconciled with simulator execution from the aggregate outputs. There is no ablation analysis isolating HHO modifications, no demonstrated optimality bound, and no reported scalability series beyond the 1,000-sensor configuration.

The ECG records supply representative input traces, but repeated sensor streams are not independent patient observations. Synthetic abnormal patterns test workload behavior rather than clinical discrimination. Diagnostic accuracy, false-alarm burden, real-world wearable battery life, and regulatory compliance remain outside the evaluated outcomes. These limitations restrict the findings to an exploratory infrastructure simulation.

Further evaluation should use reproducible simulator configurations, independently seeded runs, disclosed processing and network costs, and matched comparator budgets. Standard HHO and component-ablation comparisons would clarify the value of the orchestration design. Hardware experiments and separately designed clinical validation would be necessary before deployment claims could be supported.

Conclusion

HHO-SFC organizes ECG-related services across fog and cloud resources using HHO-based placement. In the simulation results, it achieves lower per-sensor energy consumption than the three comparators in normal, emergency, and rural scenarios. Normal-scenario latency is 62.3 ms, but emergency latency rises to 116.1 ms and exceeds the 100-ms target. The framework therefore demonstrates a descriptive energy–latency trade-off rather than universal performance dominance. The results support further investigation of the architecture, while reproducible implementation, statistical validation, and clinical evaluation remain necessary to establish broader effectiveness.

Declarations

Ethics and participant protection

This study used publicly available ECG recordings from the MIT-BIH Arrhythmia Database for secondary analysis and simulation of computing and communication infrastructure. No participants were recruited or contacted, no clinical interventions were performed, and no new patient data were collected. The work was confined to existing public recordings and simulated infrastructure.

Author contributions

Mohammad Saleh Arvari: Simulation implementation and writing of the original draft. Parvaneh Asghari: Methodology design and supervision. Sam Jabbehdari: Methodological evaluation and critical review of the manuscript. Seyyed Hamid Haj Seyyed Javadi: Methodological evaluation and critical review of the manuscript.

Data availability

The ECG source data are publicly available from the MIT-BIH Arrhythmia Database, version 1.0.0, through PhysioNet, under the Open Data Commons Attribution License v1.0. Aggregate simulation results are presented in Tables 3–5. The public ECG repository contains the source recordings, not the study-specific simulator implementation or independent run-level outputs.

Use of artificial intelligence

OpenAI ChatGPT assisted with language editing, bibliographic verification, consistency checks, and manuscript formatting. It was not used to generate the simulation measurements, conduct new simulations, or create ethics documentation.

References

  • [1] P. Keikhosrokiani and N. S. A. B. Kamaruddin, "IoT-Based In-Hospital-In-Home Heart Disease Remote Monitoring System with Machine Learning Features for Decision Making," in Connected e-Health, S. Mishra, A. González-Briones, A. K. Bhoi, P. K. Mallick, and J. M. Corchado, Eds. Cham, Switzerland: Springer, 2022, pp. 349–369, vol. 1021, doi: 10.1007/978-3-030-97929-4_16.
  • [2] M. L. Sahu, M. Atulkar, M. K. Ahirwal, and A. Ahamad, "IoT-enabled cloud-based real-time remote ECG monitoring system," J. Med. Eng. Technol., vol. 45, no. 6, pp. 473–485, Aug. 2021, doi: 10.1080/03091902.2021.1921870.
  • [3] B. Farahani, M. Barzegari, F. Shams Aliee, and K. A. Shaik, "Towards collaborative intelligent IoT eHealth: From device to fog, and cloud," Microprocessors Microsyst., vol. 72, p. 102938, 2020, doi: 10.1016/j.micpro.2019.102938.
  • [4] J. Halpern and C. Pignataro, Eds., “Service Function Chaining (SFC) Architecture,” RFC 7665, Oct. 2015, doi: 10.17487/RFC7665.
  • [5] Y. Zhang, F. Zhang, S. Tong, and A. Rezaeipanah, "A dynamic planning model for deploying service functions chain in fog-cloud computing," J. King Saud Univ. Comput. Inf. Sci., vol. 34, no. 10, pp. 7948–7960, Nov. 2022, doi: 10.1016/j.jksuci.2022.07.012.
  • [6] Y. Liu, H. Lu, X. Li, Y. Zhang, L. Xi and D. Zhao, "Dynamic Service Function Chain Orchestration for NFV/MEC-Enabled IoT Networks: A Deep Reinforcement Learning Approach," in IEEE Internet of Things Journal, vol. 8, no. 9, pp. 7450-7465, May 2021, doi: 10.1109/JIOT.2020.3038793.
  • [7] Y. Bi, C. C. Meixner, M. Bunyakitanon, X. Vasilakos, R. Nejabati and D. Simeonidou, "Multi-Objective Deep Reinforcement Learning Assisted Service Function Chains Placement," in IEEE Transactions on Network and Service Management, vol. 18, no. 4, pp. 4134-4150, Dec. 2021, doi: 10.1109/TNSM.2021.3127685.
  • [8] H. Huang et al., "Scalable Orchestration of Service Function Chains in NFV-Enabled Networks: A Federated Reinforcement Learning Approach," in IEEE Journal on Selected Areas in Communications, vol. 39, no. 8, pp. 2558-2571, Aug. 2021, doi: 10.1109/JSAC.2021.3087227.
  • [9] A. A. Heidari, S. Mirjalili, H. Faris, I. Aljarah, M. Mafarja, and H. Chen, "Harris hawks optimization: Algorithm and applications," Future Gener. Comput. Syst., vol. 97, pp. 849–872, Aug. 2019, doi: 10.1016/j.future.2019.02.028.
  • [10] M. Nawaz and J. Ahmed, "Cloud-based healthcare framework for real-time anomaly detection and classification of 1-D ECG signals," PLOS One, vol. 17, no. 12, p. e0279305, Dec. 2022, doi: 10.1371/journal.pone.0279305.
  • [11] J. Kabdjou and N. Shinomiya, "A Method for Service Function Chain Migration Based on Server Failure Prediction in Mobile Edge Computing Environment," in IEEE Access, vol. 13, pp. 78664-78678, 2025, doi: 10.1109/ACCESS.2025.3564967.
  • [12] D. Alsadie, "Advancements in heuristic task scheduling for IoT applications in fog-cloud computing: challenges and prospects," PeerJ Comput. Sci., vol. 10, p. e2128, 2024, doi: 10.7717/peerj-cs.2128.
  • [13] H. Gupta, A. V. Dastjerdi, S. K. Ghosh, and R. Buyya, "iFogSim: A toolkit for modeling and simulation of resource management techniques in the Internet of Things, Edge and Fog computing environments," Softw. Pract. Exp., vol. 47, no. 9, pp. 1275–1296, 2017, doi: 10.1002/spe.2509.
  • [14] G. B. Moody and R. G. Mark, "The impact of the MIT-BIH Arrhythmia Database," in IEEE Engineering in Medicine and Biology Magazine, vol. 20, no. 3, pp. 45-50, May-June 2001, doi: 10.1109/51.932724.
  • [15] G. Moody and R. Mark, “MIT-BIH Arrhythmia Database,” PhysioNet, version 1.0.0, 2005. doi: 10.13026/C2F305. https://physionet.org/content/mitdb/1.0.0/
  • [16] A. L. Goldberger et al., “PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals,” Circulation, vol. 101, no. 23, pp. e215–e220, 2000, doi: 10.1161/01.CIR.101.23.e215.
  • [17] D. L. Donoho and I. M. Johnstone, “Ideal spatial adaptation by wavelet shrinkage,” Biometrika, vol. 81, no. 3, pp. 425–455, 1994, doi: 10.1093/biomet/81.3.425.
  • [18] J. Pan and W. J. Tompkins, "A Real-Time QRS Detection Algorithm," IEEE Transactions on Biomedical Engineering, vol. BME-32, no. 3, pp. 230-236, March 1985, doi: 10.1109/TBME.1985.325532.
  • [19] E. Rescorla, “The Transport Layer Security (TLS) Protocol Version 1.3,” RFC 8446, 2018. doi: 10.17487/RFC8446.
  • [20] G. F. Riley and T. R. Henderson, “The ns-3 Network Simulator,” in Modeling and Tools for Network Simulation, K. Wehrle, M. Güneş, and J. Gross, Eds. Berlin, Heidelberg: Springer, 2010, pp. 15–34, doi: 10.1007/978-3-642-12331-3_2.