QoS-based management of biomedical wireless sensor networks for patient monitoring
© Abreu et al.; licensee Springer. 2014
Received: 15 November 2013
Accepted: 28 April 2014
Published: 9 May 2014
Biomedical wireless sensor networks are a key technology to support the development of new applications and services targeting patient monitoring, in particular, regarding data collection for medical diagnosis and continuous health assessment. However, due to the critical nature of medical applications, such networks have to satisfy demanding quality of service requirements, while guaranteeing high levels of confidence and reliability. Such goals are influenced by several factors, where the network topology, the limited throughput, and the characteristics and dynamics of the surrounding environment are of major importance. Harsh environments, as hospital facilities, can compromise the radio frequency communications and, consequently, the network’s ability to provide the quality of service required by medical applications. Furthermore, the impact of such environments on the network’s performance is hard to manage due to its random and unpredictable nature. Consequently, network planning and management, in general or step-down hospital units, is a very hard task. In such context, this work presents a quality of service based management tool to help healthcare professionals supervising the network’s performance and to assist them managing the admission of new sensor nodes (i.e., patients to be monitored) to the biomedical wireless sensor network. The proposed solution proves to be a valuable tool both, to detect and classify potential harmful variations in the quality of service provided by the network, avoiding its degradation to levels where the biomedical signs would be useless; and to manage the admission of new patients to the network.
KeywordsPatient monitoring Quality of service Admission control Biomedical wireless sensor networks
Patients’ monitoring in general or step-down hospital units is crucial to avoid clinical worsening of inpatients ([Alemdar and Ersoy 2010]; [Nouira and Trabelsi 2012]). However, in most cases, despite the necessity of continuous monitoring, nursing professionals measure vital or physiological signs (e.g., temperature, pulse and respiratory rates, blood pressure or oximetry) manually a few times a day. Such episodic measurements can be complemented by efficient and reliable real-time monitoring systems, bringing out an enhancement of the quality of care, while freeing the nursing staff to provide extra attention to the inpatients ([Chipara et al. 2010]; [Ko et al. 2010b]).
Depending on its application and purpose, real-time patient monitoring systems have distinct requirements ([Baig and Gholamhosseini 2013]). In one hand, continuous monitoring of cardiac or cerebral signs (i.e., electrocardiogram or electroencephalogram, respectively) requires high throughput networks (e.g., IEEE 802.11-based networks). On the other hand, gathering of vital or physiological signs can be done through Biomedical Wireless Sensor Networks (BWSNs) based on the IEEE 802.15.4 standard ([Ko et al. 2010a]). BWSNs are distinct from IEEE 802.11-based wireless networks in several aspects. Unlike IEEE 802.11 networks, they form a distributed, self-organised and energy-efficient low data rate network without the need of a physical infrastructure. Due to its flexibility and low energy consumption, BWSNs have been successfully used in several patient monitoring scenarios ([Alemdar and Ersoy 2010]; [Honeine et al. 2011]).
The QoS requirements of BWSNs are, typically, defined at the earliest stages of the project and addressed across the different layers of the communication stack. However, due to the dynamic nature of hospital environments BWSNs can be exposed to hostile situations regarding the radio communications. Consequently, the QoS provided by the network can change very often in an unpredictable way ([Ko et al. 2010b]). In such harsh environments the QoS degradation can be related with either random or deterministic factors. Random factors, such as the dynamics of the network or hospital environment, the radio interferences, or the patients’ mobility need to be monitored and classified in order to detect situations in which they can cause a severe degradation of the QoS required by the applications using the BWSN. Deterministic factors, such as the network congestion due to the overpopulated network can be avoided by using QoS assessment and admission control tools to deploy the nodes in appropriate areas. In this context, this paper contributes with a QoS-based management system that can be used both, to assess the QoS provided by the BWSN and to assist healthcare professionals managing the admission of new sensor nodes (i.e., new patients to be monitored) to BWSNs while knowing its impact on the QoS provided by the network.
In what follows, some related works are outlined in Section “Related work”, and then the proposed QoS-based network management system is presented in Section “Proposed QoS-based network management system”. Section “Time domain analysis of the QoS metrics” presents the methodology used to detect and classify QoS degradation events. In Section “Experiments and results”, the proposed QoS-based network management system is assessed and the results analysed. Finally some conclusions are drawn.
The emergence of WSNs as an important communication infrastructure regarding the development of real-time monitoring applications, such as patient monitoring, increases the demand for both QoS monitoring systems and admission control systems. As discussed in ([Pereira et al. 2012]) and ([Lindh and Orhan 2009]), both QoS monitoring systems and admission control systems are keystones to achieve high levels of reliability and performance in WSNs; in fact, they complement each other.
In the last few years, a few approaches have been proposed for both QoS monitoring and admission control for WSNs. The authors of ([Sun et al. 2005]) propose SenProbe, an end-to-end capacity estimation tool that can be used to performance analysis and network deployment planning. SenProbe uses one-way active measurements to estimate the end-to-end path capacity by injecting packet trains in the network to measure the packet dispersion. As an alternative to real-time active measurements, the authors of ([Yang and Kravets 2005]) propose a contention-aware admission control protocol designed for IEEE 802.11 ad-hoc networks, which estimates the available channel capacity by allowing each node to measure the time that the communication channel is busy. The work presented in ([Orhan and Lindh 2011]) offers a measurement-based performance and admission control system, where the admission decision is made by the network coordinator, based on real-time measurements of the packet loss ratio provided by a performance meter. The performance meter runs on each sensor node and it is continually tracking the number of sent and received data packets and bytes. By running on each sensor node, the performance meter contributes to its energy depletion and consequently reduces the network lifetime. Since WSNs are composed, typically, of energy-constrained sensor nodes, the increase of energy consumption on their nodes must be avoided. In fact, depending on the network application it could be a strong weakness. Although proposing different solutions, all these studies support the importance and need of using both QoS monitoring and admission control systems, to manage and improve the performance of WSNs.
This paper proposes a new QoS-based network management system for BWSNs comprised by two modules, namely the QoS monitoring module and the admission control module. The QoS monitoring module collects relevant information about the network’s performance and uses it to detect and classify QoS degradation events. Based on this information, reports are generated to inform the healthcare professionals, or who is in-charge, about the network’s performance. On its turn, the admission control module uses a probe-based admission control procedure to decide where a new sensor node can join the network. The decision where to admit the new sensor node on the network is based on time-domain analysis of the performance metrics used to assess the QoS provided by the network, namely the packet reception ratio. Within the proposed method, the probe-based procedure uses a “virtual sensor node” called QoS Probe to mimic the presence of a new real sensor node on its neighbourhood. The use of a virtual sensor node” enables to assess the network performance from a remote location, in order to decide about the best location to admit the new sensor node within the network. To avoid the problem of energy depletion in the energy-constrained sensor nodes, the proposed QoS-based network management system collects all the data required for decision making at the sink that, generally, is connected to the power line. Regarding the energy consumption in the sensor nodes, the QoS Probe is required to run only on request and for short time periods, being the extra energy consumption insignificant.
Proposed QoS-based network management system
Regarding the BWSNs used within the suggested QoS-based network management system, this section presents the software used by its nodes. Namely, its architecture and components are described.
The suggested QoS-based network management consist of two modules, namely the QoS monitoring module and the admission control module. The QoS monitoring module embraces the following objectives: to monitor the real-time QoS supported by the network, and to detect and classify QoS degradation events. On its turn, the admission control module is in charge of managing the admission location of new sensor nodes to the network to prevent its performance degradation due to network congestion.
The QoS monitoring module analyses the incoming traffic at the sink to extract relevant metrics (e.g., the packet reception rate) in order to assess the QoS provided by the network. Then, such metrics are evaluated using time-domain techniques to detect QoS degradation events. Regarding the QoS degradation events, they can be classified into two categories: hard degradation events and soft degradation events. Hard degradation events are related to the failure of static thresholds previously defined for each metric; in such cases, the normal operation of the applications using the network may be affected. On the other hand, soft degradation events are related to variations in the metrics within the imposed thresholds. Although soft degradation events may not cause the failure of the applications using the network, they must be detected and classified. For instance, a small QoS degradation may lead to a hard degradation event if persisting the necessary time. To prevent such situations, the time-domain analysis is used not only to compute the actual value of each metric but also to investigate the metrics dynamic.
Time domain analysis of the QoS metrics
The QoS metrics can be considered as static or dynamic. Static metrics do not depend on the network operation and are constant along the time. On the other hand, dynamic metrics vary according to network operation and are essential to know the real network status and performance ([Xiao et al. 2004]). Regarding the dynamic metrics, they can be assessed considering only its current value, or taking into consideration its past behaviour. By taking into consideration its behaviour in the near past, it is possible to assess the metrics in view of its tendency to the near future.
where m(t) represents the measured value of the metric, m∗(t) represents the real value of the metric, and the noise function w g n(t) represents the metric’s variation due to random interferences and the natural fluctuations of the network.
Computing the metrics value
The number of observations used to estimate the value of each metric depends both, on the network operation and on the deployment environment. As a matter of example, consider a network deployed inside a hostile environment in which the quality of the wireless links can often vary considerably. In such case it is reasonable to use a large number of observations to estimate the metrics’s value, in other words, it is necessary to analyses the network during the last T seconds, being T the time necessary to compute the metrics’s observations being considered. Therefore, the value of N must be determined taking in consideration the network typical operation and the target deployment environment.
Detecting relevant variations in the metrics
The network operation can be affected both by external events (e.g., radio interferences) and by internal events (e.g., node dead due to energy depletion). In both circumstances, these events may reflect itself as perturbations in the metrics being used to assess the network’s QoS. Detecting such perturbations is of major importance to prevent the degradation of the network’s performance.
The MDI gives quantitative information about the change on the metric energy.
After calibration, the MDI can be used to detect potential degradation events, the higher the MDI, the greater is the potential degradation of the network performance. To calibrate the MDI, several approaches can be used, depending on the QoS monitoring strategy. One possibility consists of using the maximum MDI achieved during the normal network operation as a threshold to detect degradation events and fire alert messages. Other possibility is using the MDI’s mean value achieved during the network normal operation.
Detecting small variations in the metrics
Although useful to detect significant perturbations on the metrics being analysed, the previous analysis is insensitive to small and monotonic variations on the metrics. In other words, the previous analysis is insensitive to variations resulting in . To detect such small variations three figures-of-merit are used, namely, the Metric Tendency (MT), the Zero Crossing Rate (ZCR) and the classification of the most recent observation of the metric as a single global minimum (MIN) or a single global maximum (MAX) within the sample being processed.
Consider the functions h i ,i=1,...,p. The product function of these functions h i is represented by . In the case of h1≡h2≡...≡h p ≡h, we obtain . By id we denote the identity function.
Within the sample being processed, if the is to decrease (↓), if the is to increase (↑) and, finally, if the is to be constant (→).
where sgn is the sign function.
The order of the filter (i.e., the length of the sample being processed) used to compute the previous figures-of-merit depend on the QoS monitoring policy in use. Considering a small value of N, the QoS monitoring system becomes very sensitive and reactive. In opposition, a QoS monitoring system less reactive must use a higher N value. Nevertheless, it is necessary to find equilibrium between the order of the filter and the desired reactiveness of the QoS monitoring system.
By using these three figures-of-merit it is possible to detect small and monotonic variations, over the sample being processed, on the metric being analysed. By detecting such situations, corrective measures can be taken to prevent the further degradation of the QoS provided by the network.
QoS assessment and network performance classification
Considering the classification presented in ([Truong et al. 2006]), the QoS metrics used to assess the network’s performance can be sorted into two sets: those that need to be maximised, denoted as m m a x and, the remaining that need to be minimised m m i n . Based on this classification, and using the features previously presented, it is possible to analyse the performance of the network and classify suspicious events.
Network performance assessment and classification if QoS metric belongs to m max
≤E t h r e s h o l d
≤E t h r e s h o l d
Not degrading or recovering
≤E t h r e s h o l d
≤E t h r e s h o l d
→ or ↑
>E t h r e s h o l d
Network performance assessment and classification if QoS metric belongs to m min
≤E t h r e s h o l d
≤E t h r e s h o l d
Not degrading or recovering
≤E t h r e s h o l d
≤E t h r e s h o l d
→ or ↓
>E t h r e s h o l d
Experiments and results
Network simulators are widely used to assess and compare the performance of WSNs ([Liang 2009]). Their use has advantages and disadvantages. They allow easy and fast network deployments, controllable and flexible scenarios and the repeatability of the obtained results. However, due to the oversimplified channel models and protocols, simulations may not represent the reality. On the contrary, real deployments avoid problems relating to simplifications of models and protocols, but such deployments are considerably harder to implement and deploy. For these reasons, a hybrid approach based on the cross level emulation and simulation tool known as COOJA ([Osterlind et al. 2006]) has been used in this work. COOJA is a flexible WSNs simulator designed for simulating networks running Contiki OS ([Dunkels et al. 2004]). The BWSN used in this work was developed with the Contiki OS and simulated on COOJA using the framework presented in ([Abreu et al. 2012]).
80 m × 80 m
5 × 5 grid (see Figure 4)
Number of Nodes
1 sink and 25 sensor nodes
Unit Disk Graph Medium Distance Loss
IPv6 with 6LowPAN
User Datagram Protocol (UDP) ([Postel 1980])
RPL with the MRHOF ([Gnawali and Levis 2012])
< 70 bytes (one packet)
Reporting Interval (s)
Evaluation and results
To evaluate the proposed QoS-based network management system, this section provides several simulation and the corresponding results. The evaluation consists of two parts. First, the QoS monitoring module is used to detect and classify potential harmful events relating to QoS degradation. Then, the admission control module is used to verify if a new sensor node can be admitted by the BWSN in the requested area. In both scenarios the PRR is taken as a figure-of-merit to evaluate the QoS provided by the network. The PRR was defined as: , where M is the number of sensor nodes on the network, R i is the number of data packets received by the sink from the node i, and S i is the number of data packets sent by the node i.
Evaluating the QoS monitoring module
Regarding the QoS monitoring module, it was tested in order to evaluate its capacity to detect not only small and monotonic variations in the PRR (i.e., the metric being used to assess the QoS provided by the network) able to cause QoS degradation in the long term, but also sudden variations in the PRR potentially dangerous for QoS.
A detailed observation of Figure 6 permits to verify (around 440 s and 490 s) that using both, information about the metric’s dynamic and the previous defined figures-of-merit, the QoS monitoring module was able to detect slight and monotonic variations in the PRR. Concerning the detection and classification of potentially harmful QoS degradation events using the energy feature, it is important to notice that small values of the MDI correspond to small variations on the metric being analysed.
In order to introduce selectivity on the detection of significant degradation events, and at the same time avoid alarm fatigue due to over-alarming, a threshold was introduced in the MDI. In this way, the maximum MDI value obtained during the normal network operation (i.e., M D I=4) was used as a minimum limit to detect a degradation event and fire the corresponding alarm. In other words, degradation events are detected and fired if the MDI value exceeds the one used to calibrate the QoS monitoring system. In addition, a real application of the proposed method can introduce other mechanisms to improve selectivity.
In addition to the present analysis, the proposed method can be used to assess the performance of each individual node within the network. This possibility, while requiring additional computational power and memory, allows obtaining additional information about the network performance. In particular, monitoring individual nodes is crucial to identify the source of the performance degradation events, which is impossible when analysing the overall performance of the network.
Evaluating the admission control module
To test the admission control module, the aforementioned probing procedure was used to evaluate if a new sensor node can be admitted to the network in the place under test. The probing procedure starts on request (e.g., by a healthcare professional) and stops when one of the following conditions is observed: the metric value crosses a pre-established limit or the metric tendency (to increase or decrease) is interrupted, in other words for the metrics belonging to m m a x , and for the metrics belonging to m m i n .
Regarding the evaluation of the admission control module, two scenarios were considered. For the first, each sensor node within the network was set to generate constant bit rate traffic of about 0.5 packets per second, where the network is considered uncongested and able to admit a new sensor node while maintaining a PRR of, at least, 98%. In the second scenario, the sensor nodes were set to send 1 data packet per second to the network, and it has to fulfil a PRR of, at least, 91%. Herein, the network is congested and unable to admit a new sensor node while maintaining the requested QoS.
Results obtained from the admission control procedure after a measurement time of about 90 s, see Figure 9
End of measurement
PRR at the beginning
PRR at the end
of the measurement
of the measurement
Results obtained from the admission control procedure, after a measurement time of about 200 s, see Figure 11
End of measurement
PRR at the beginning
PRR at the end
of the measurement
of the measurement
Biomedical wireless sensor networks have to fulfil high levels of confidence and reliability in order to be used in patient monitoring scenarios, particularly those where data collection is vital for diagnosis or to improve the quality of care provided in hospital units and nursing houses. This paper presented a QoS-based network management system that comprises two modules namely, the QoS monitoring module and the admission control module. The QoS monitoring module is able to detect and classify suspicious events regarding the degradation of the QoS provided by the network. More, by using not only the metric value but also its dynamic, the QoS monitoring module is able to detect and classify QoS degradation events even before the metric reaches critical values. By using this ability, the QoS monitoring system can be used to fire advertising messages to the network manager informing about the potential QoS degradation event. On its turn, the network manager can take measures to mitigate the effects of such events, preventing the performance of the network from degrading. Regarding the admission control system, it can be used to assess if the network can admit a new sensor node, i.e., a new patient being monitored. By using a “virtual sensor node to mimic the presence of the new real sensor node on its neighbourhood while assessing the network, the proposed admission control system is able to test the network even from a remote location. In this way, the proposed QoS-based network management system allows to monitor and manage the network from a centralised location, which is an important feature regarding patient monitoring scenarios in hospital facilities or nursing homes.
This work has been supported by the PhD grant of Portuguese Foundation for Science and Technology (FCT - Fundação para a Ciência e a Tecnologia), SFRH/BD/61278/2009. Miranda was also supported by Portuguese funds through the CIDMA - Center for Research and Development in Mathematics and Applications, and the Portuguese Foundation for Science and Technology.
- Abreu C, Ricardo M, Mendes P: Framework for QoS performance assessment on biomedical wireless sensor networks. In Biodevices 2012 - international conference on biomediacal electronics and devices. Portugal: Vilamoura; 2012. . http://dblp.uni-trier.de/db/conf/biostec/biodevices2012.html#AbreuRM12Google Scholar
- Alemdar H, Ersoy C: Wireless sensor networks for healthcare: a survey. Computer Networks 2010, 54(15):2688-2710. . http://www.sciencedirect.com/science/article/pii/S1389128610001398 10.1016/j.comnet.2010.05.003View ArticleGoogle Scholar
- Baig M, Gholamhosseini H: Smart health monitoring systems: an overview of design and modeling. J Med Syst 2013, 37: 1-14.View ArticleGoogle Scholar
- Baker SD, Hoglund DH: Medical-grade, mission-critical wireless networks [designing an enterprise mobility solution in the healthcare environment]. IEEE Eng Med Biol Mag 2008, 27(2):86-95.View ArticleGoogle Scholar
- Chipara O, Lu C, Bailey TC, Roman G-C: Reliable clinical monitoring using wireless sensor networks: experiences in a step-down hospital unit. In Proceedings of the 8th ACM conference on embedded networked sensor systems, SenSys ‘10. New York: ACM; 2010:155-168.View ArticleGoogle Scholar
- Dunkels A, Gronvall B, Voigt T: Contiki - a lightweight and flexible operating system for tiny networked sensors. Local computer networks, 2004 2004, 38-38. 29th annual IEEE international conference on. pp 455–462 doi:10.1109/LCN.2004.38Google Scholar
- Gnawali O, Levis P: The minimum rank objective function with hysteresis. 2012.View ArticleGoogle Scholar
- Honeine P, Mourad F, Kallas M, Snoussi H, Amoud H, Francis C: Wireless sensor networks in biomedical: Body area networks. Systems, signal processing and their applications (WOSSPA) 2011. 2011 7th international workshop on. pp 388–391 doi:10.1109/WOSSPA.2011.5931518Google Scholar
- Ko J, Gao T, Rothman R, Terzis A: Wireless sensing systems in clinical environments: improving the efficiency of the patient monitoring process. IEEE Eng Med Biol Mag 2010a, 29(2):103-109.View ArticleGoogle Scholar
- Ko J, Lu C, Srivastava MB, Stankovic JA, Terzis A, Welsh M: Wireless sensor networks for healthcare. Proc IEEE 2010b, 98(11):1947-1960.View ArticleGoogle Scholar
- Liang X: QoS provisioning for wireless sensor networks: algorithms, protocols and modeling. 2009.Google Scholar
- Lindh T, Orhan I: Performance monitoring and control in contention-based wireless sensor networks. Wireless communication systems, 2009. ISWCS 2009 2009. 6th international symposium on. pp 507–511 doi:10.1109/ISWCS.2009.5285347Google Scholar
- Más I, Karlsson G: Probe-based admission control for a differentiated-services internet. Comput Netw 2007, 51(13):3902-3918. 10.1016/j.comnet.2007.04.009View ArticleGoogle Scholar
- Karlsson G, Más, I: A model for endpoint admission control based on packet loss. In NETWORKING 2008 Ad Hoc and sensor networks, wireless networks, next generation internet, volume 4982 of Lecture notes in computer science. Edited by: Das A, Pung H, Lee F, Wong L. Springer Berlin/Heidelberg; 2008:506-517. http://dx.doi.org/10.1007/978-3-540-79549-0_44Google Scholar
- Nouira K, Trabelsi A: Intelligent monitoring system for intensive care units. J Med Syst 2012, 36: 2309-2318. 10.1007/s10916-011-9698-xView ArticleGoogle Scholar
- Orhan I, Lindh T: Measurement-based performance and admission control in wireless sensor networks. Int J Adv Syst Meas 2011, 4(1 & 2):32-45.Google Scholar
- Osterlind F, Dunkels A, Eriksson J, Finne N, Voigt T: Cross-level sensor network simulation with cooja. Local computer networks, proceedings 2006. 31st IEEE conference on. pp 641–648 doi:10.1109/LCN.2006.322172Google Scholar
- Pereira V, Silva JS, Monteiro E: A framework for wireless sensor networks performance monitoring. World of Wireless, Mobile and Multimedia Networks (WoWMoM) 2012. IEEE international symposium on a. pp 1–7 doi:10.1109/WoWMoM.2012.6263784Google Scholar
- Postel J: User datagram protocol. RFC 1980., 768(Standard): . http://www.ietf.org/rfc/rfc768.txtGoogle Scholar
- Sun T, Chen L-J, Yang G, Sanadidi MY, Gerla M: Senprobe: path capacity estimation in wireless sensor networks. Proceedings of the 3rd international workshop on measurement, modelling, and performance analysis of wireless sensor networks 2005. . http://nrlweb.cs.ucla.edu/publication/show/452Google Scholar
- Truong Hl, Samborski R, Fahringer T: Towards a framework for monitoring and analyzing qos metrics of grid services. Proceedings of the second IEEE international conference on e-science and grid computing 2006. Society Press . http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.89.7823 Society Press .Google Scholar
- Xiao L, Wang J, Lui K-S, Nahrstedt K: Advertising interdomain qos routing information. IEEE J Sel Area Comm 2004, 22(10):1949-1964. 10.1109/JSAC.2004.836004View ArticleGoogle Scholar
- Yang Y, Kravets R: Contention-aware admission control for ad hoc networks. IEEE Trans Mobile Comput 2005, 4(4):363-377.View ArticleGoogle Scholar
- Younis M, Akkaya K: Strategies and techniques for node placement in wireless sensor networks: a survey. Ad Hoc Netw 2008, 6(4):621-655. 10.1016/j.adhoc.2007.05.003View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.