Proceedings of SPIE ITCOM Conference, Sept 2003, Orlando, FL

QoS Considerations in Wireless Sensor Networks for Telemedicine Fei Hu a * Sunil Kumar b ** Computer Engineering, RIT, 83 Lomb Memorial Drive, Rochester, NY 14623-5603; b Electrical & Computer Engineering, Clarkson University, Potsdam, NY 13699-5720

a

ABSTRACT The integration of telemedicine with medical micro sensor technology (Mobile Sensor Networks for Telemedicine applications -- MSNT) provides a promising approach to improve the quality of people’s lives. This type of network can truly implement the goal of providing health-care services anytime and anywhere. Our research in this field generates the following outcomes that are reported in this paper: (1) We propose a mobile sensor network infrastructure to support the third-generation telemedicine applications; (2) An energy-efficient query resolution mechanism in large-scale mobile sensor networks is used for critical medical data collections; (3) To provide the guaranteed mobile QoS for arriving multimedia calls, a new multi-class call admission control mechanism is proposed which is based on dynamically forming a reservation pool for handoff requests. We used discrete-event-based simulation model using OPNET to verify our scheme. The simulation results show that our system can satisfy the adaptive QoS requirements in large-scale telemedicine sensor networks. Keywords: Sensor networks, Mobile Networks, Telemedicine, CAC, QoS, Data Query 1. INTRODUCTION Advances in micro-electro-mechanical systems (MEMS) technology, medical information processing, wireless communications, and digital electronics have enabled the development of low-cost, low-power, multifunctional medical micro sensor nodes. These sensors consist of sensing, data processing, and communicating components, and leverage the idea of sensor networks based on the collaborative effort of a large number of nodes [1-3]. Mobile Telemedicine is one of the most exciting technologies today for improving the public health condition [4]. This paper proposes the sensor-based mobile telemedicine architecture to resolve the large-scale medical query resolution and multimedia transmission issue. As a matter of fact, the integration of telemedicine with medical micro sensor technology provides a promising approach to improving the quality of people’s lives. This type of network can truly implement the goal of providing health-care services anytime and anywhere. Mobile sensor network is a new and emerging field with many challenging issues [2]. Unfortunately there is very little research conducted in Mobile Sensor Networks for Telemedicine systems (MSNT). Most work separately focuses on either mobile networks or telemedicine system and cannot seamlessly integrate the two fields together. Some of the most challenging issues and related works done on MSNT can be summrized as follows: • A promising MSNT architecture that can utilize the features of advanced 3G (Third Generation) wireless networks, such as high bandwidth and multimedia transmission. Future telemedicine system cannot use ADSL or PSTN phone to transmit data since the patients can move around at any speed. A simple mobile-phone-based telemedicine system is proposed in [4]. But it does not consider multimedia QoS issue and network scalability problem.

*

[email protected]; phone: (585) 475-5139. [email protected]; phone: (315) 268-6602

**

•

•

An efficient CAC (Call Access Control) mechanism to support QoS requirements of different types of medical services in 3G mobile networks. For example, call from ambulence should get the high priority call access services while average patients can transmit normal data such as blood pressure in a low-priority access level. Recently very limited work has been reported in the literature regarding CAC schemes in multi-class wireless networks. S. K. Das et al. [5] developed an integrated framework for QoS provisioning at a lower layer such as the radio link layer combining a novel CAC strategy. However, it does not provide implementation details of channel reservation and handoff prioritization, which is very important in mobile networks. Another CAC scheme based on adaptive bandwidth reservation has been proposed by Oliveira et al. in 1998 [6]. However, it does not minimize the new-call blocking rate effectively compared to handoff-call dropping rate. Low-energy query resolution in large-scale sensor networks. With a small amount of medical sensors it is possible to collect all measurements from each patient and send them to a central hospital, and perform data processing centrally. However, in large-scale sensor networks with limited energy it is not a scalable approach. It is best to view such sensor networks as distributed databases [7]. The query should adapt to mobile environments since patients can move around.

This paper reports our research results related to MSNT protocols and algorithms. (1) (2)

We use an efficient query protocol to minimize the energy consumption during medical query resolution in large-scale wireless sensor networks. Our query strategy considers the mobility of patients in 3G cellular networks. For guaranteeing the M-QoS (Mobile QoS1) of each class of medical calls, we propose a new notion of Reservation Ordering (RO) of handoff requests. RO is about the assigning of admission priorities for multiclass medical calls. Our admission priority determination is made according to patients’ time-varying movement behaviors and the desired M-QoS requirements of the multi-class calls themselves. We use an efficient dynamic CAC scheme to meet the target handoff dropping probability of real-time medical services such as remote surgery command system.

The rest of this paper is organized as follows. The infrastructure of mobile sensor networks for telemedicine applications is described in Section 2. Section 3 provides our low-energy query resolution scheme in large-scale medical sensor networks. In Section 4, we state the detailed procedure for forming priority-based handoff request reservation pool that is based on accurate next-cell prediction. This is followed by the presentation of our call admission mechanism. Our simulation results and corresponding analysis are provided in Section 5. Finally, we conclude the paper with a discussion of further work in Section 6.

2. SENSOR-BASED TELEMEDICINE NETWORKS The main features of the proposed “3G Mobile Telemedicine based on Sensor networks” are as follows: 1) We utilize the infrastructure of CDMA2000-based 3G wideband wireless cellular networks. 2) To implement ‘communication anywhere’, we assume a multi-layer hierarchical structure that includes the following cell-sizes: pico-cell, micro-cell, macro-cell and satellite-cell (Fig. 1). A mobile user can use soft handoff to contact the base station with the strongest communication signal strength. 3) Sensor Telemedicine Networks are seamlessly integrated with the aforementioned 3G mobile networks. In the future telemedicine systems, patients will carry medical sensors that sense the health parameters, such as body temperature, blood pressure, pulse oxymetry, ECG, breathing activity, and so on. In addition, serious patients can also carry other sensors that help the medical center carry out remote monitoring. Typical examples are location sensor, motion or activity sensor, microphone sensor, and camera sensor. Typically a patient will carry a wrist-device (called as supersensor in this paper) with a stronger battery and higher memory compared to the other medical sensors, to perform 1

Mobile Quality-of-Service (M-QoS) is a new concept, which is defined particularly for WATM environments and has not been specified by the ATM Forum WATM Group [8]. M-QoS can be viewed to be comprised of the traditional QoS found in wired ATM augmented with wireless QoS. Wired QoS relates to link delay, cell delay jitter, throughput and other parameters supported by typical ATM. Wireless QoS is a set of performance parameters associated with wireless link such as channel error rate and with mobile units such as Handoff-call Dropping Probability (HDP) and New-call Blocking Probability (NBP).

multiple-hop ad hoc communication. Super-sensors will be used to collect sensing data from body sensors and communicate with other super-sensors. However, these tiny wrist-devices will not have as much power as today’s cell phone, to perform two-way communication with the base stations. They therefore transmit data through multi-hop routing algorithm. The functional block diagram of a typical medical sensor is shown below in Fig .2.

Fig. 1: Hierarchical mobile network structure [13]

Sensor Unit Sensor

Control Unit

Signal Processing

Transceiver

Battery

Fig. 2: Medical Sensor Architecture

Handoff-prioritized calls from serious patients or elderly people

Cell 1

Real-time Handoff-guaranteed calls from Ambulance

Cell 2 +

911

Fig.3 Three types of medical services There are three types of medical services (calls) in our telemedicine system:

Non-real-time calls containing medical data

(1) Real-time calls from Ambulance Patients using the video (camera) sensors, GPS system, and other advanced medical sensors to establish a rate-guaranteed connection with the medical center. We treat them as “handoffguaranteed” calls since we should reserve wireless bandwidth in the next cell to guarantee their on-going connections when they move through different cells. These calls are given the highest priority. (2) Handoff-prioritized calls from serious patients or elder people. The serious patients can use a high-power backpack computer or medical cell phone (see Fig.3) to directly perform two-way communication with the medical center. We give the second highest priority to these calls when they need to hand over to a new cell. (3) Non-real-time calls from Cluster-heads who collect medical data from wrist-worn super sensors of average patients or normal people. These super sensors can aggregate all kinds of medical data and locally perform some preprocessing such as filtering and compression. Please note that they cannot communicate with the medical center directly since they do not have enough power. Actually they would “wake up” periodically or when urgent medical conditions are detected by the body sensors. They could transmit data, using multi-hop communication, to the cluster-head who could be a cell-phone of a nearby person/patient or a specialized facility driving around and collecting medical data from these sensors, and then communicate with the medical center. We provide their communication the best-effort services when the command center sends out a query to a certain cell to collect medical data. These super-sensors utilize “ad hoc multi-hop transmissions” to relay query results.

2. DATA QUERY IN SENSOR NETWORKS Little research work has been done on the data query problem in sensor networks. The ACQUIRE mechanism described in [12] provides a superior query optimization for responding to a particular kind of queries: complex, one-shot queries for duplicated data. However, it does not consider scalability. We propose a concentric-circle-based network self-organization protocol to efficiently forward the query messages between the medical center and the super-sensors (see Fig. 4). The medical center periodically sends the cluster-forming information to each super-sensor based on their location, maximum transmission range and maximum sensing range. These super-sensors will perform “neighbor discovery” and “cluster forming” operations. Each cluster has a clusterhead that is responsible for collecting query results from its cluster members and forwarding it to the next cluster-head along the shortest path – a straight line. This network architecture can efficiently save query energy since the query response can be sent back to the medical center in a shortest distance. As we know, the wireless communication consumes the most percentage of the total sensor battery consumption compared to local data processing. Thus a shorter path can significantly save the transmission energy. Our network architecture considers different scenarios. For instance, there may be very few or no patients in some areas due to different node densities. Thus in a path we may have an empty cluster (assume each cluster has the same diameter). In such a situation, a cluster-head should find another closest cluster-head to continue the forwarding of the query results. Besides patients can move out of a cluster or join a cluster. Also, the patient’s super-sensors may fail due to battery exhaustion. Thus the self-organization algorithm should adapt to the dynamic network topology changes. Our scheme integrates query processing with routing protocols so that we can efficiently use in-networking data aggregation instead of end-to-end data transmission to reduce the amount of data of query results. The idea is to combine the data coming from different sources en route, eliminate redundancy, minimize the number of transmissions and thus save energy. This paradigm shifts the focus from the traditional address-centric approaches (finding short routes between pairs of addressable end-nodes) to a more data-centric approach (finding routes from multiple sources to a single destination that allows in-network consolidation of redundant data). In our simulation experiment, we assume the patients are moving around at a certain velocity. We form clusters based on the scheme shown in Fig.3. In each cluster, we adopt Zone Routing Protocol to manager the message communication between each cluster head and its members [13]. Each query will forward from cluster head to cluster head. Because the members in one cluster do not need to communicate with the members in another cluster, we can save much communication traffic in sensor networks.

Our next step will take advantage of the mobility of patients to improve the query efficiency. Currently we periodically form clusters to adapt the mobility nature of our sensor networks.

Fig. 4: Proposed sensor query architecture 4. PRIORITY CONTROL FOR MULTIMEDIA MOBILE CALLS The challenging task of bandwidth assignment for multimedia (i.e. multi-class) mobile medical calls, which include patients’ sensing data, voice, pictures and video data, should take into consideration largely different QoS profiles of each class such as HDP (Handoff Dropping Rate), latency tolerance and desired amount of wireless effective bandwidth. This requires that we assign each class of calls different priorities during resource allocation, unlike in single class case where all calls are assumed to have the same priority. We propose the Reservation Ordering (RO) to make sure that the priority order for each submitted handoff request reservation is maintained. To determine the value of RO for each handoff call, we define a term Class Urgency (CU), which represents the desired priority. CU of the incoming multimedia calls is determined by their M-QoS parameters, such as delay tolerance and HDP. We give the Ambulance Communication the highest CU and normal patients’ calls the lowest CU. However, CU cannot be used as the only factor for determining the value of RO. For example, when an MH (mobile host of the patient) is moving close to the new base station such that the signal from the old base station is too week to maintain the normal signal-to-noise ratio, we should probably serve the handoff request of this MH immediately even though its CU is low, otherwise the call will be dropped. In other words, the RSS (received signal strength) value can become another factor for determining the RO priority. In addition, varying speeds of the MH can be an important parameter in 3G mobile network where very rapid fading is common due to its small cell size and low used power. A faster MH will generally require an earlier handoff than a slower one. However, the MH in reservation area may get stuck in traffic jams or at traffic lights. For these cases, it may not be proper to assign higher priority to this MH just because its RSS is low. We define the RO priority as weighted scheme:

RO = [W1 × (∆RSS ∆ t ) + W2 × (RSS ) + W3 × (Class Urgency )] where

(W1 + W2 + W3 = 1 )

here, ∆RSS / ∆t reflects the MH speed, and RSS becomes weaker when the distance of MH from its BS becomes larger. In multi-class network, we should assign W1, W2 and W3 based on the influence that above-mentioned three factors may have on RO. A reasonable weight assignment for W1, W2 and W3 is 0.1, 0.4 and 0.5, in that order. Here CU is assigned highest weight (W3 = 0.5) because it plays an important role in multimedia network. Furthermore, weight (W2 = 0.4) given to position information is much higher than that given to speed. Note that we normalize the value of ∆RSS / ∆t and RSS between 0 and 1.There are already many good ways for measuring MH velocity such as in [10]. Thus it is not difficult to obtain the value of ∆RSS / ∆t .

The following pseudo-code describes the necessary system operations each time a MH handoff request message is sent to the next-cell’s BS. Compute RO for that MH IF this message is a Reservation Canceling {Re-mark the channels for that MH from ‘Reserved’ to ‘Free’ in the pool; Delete the buffer unit for that MH in the Reservation Queue if it exists; } Else IF this message is a Reservation Confirming IF there is already a ‘Reserved’ Channel Block (CB) for that MH in the pool { Modify its RO to the new value; Reorder all the CB based on their new RO value in the pool;} ♣ Else /* This is a new reservation */ { Delete the buffer unit for that MH in the Reservation Queue if it exists; IF available free bandwidth ≥ Desired bandwidth { Insert a new CB in the reservation pool based on RO priority;} Else /* available free bandwidth < Desired bandwidth */ { Buffer it into the Reservation Queue}

Fig.5 Flow chat for our proposed multi-class CAC algorithm

Our multi-class CAC algorithm is shown in Figure 5. Two crucial details are worth mentioning: (1) Handoff Call Dropping: When an arriving handoff call is served according to its RO priority, it may not find the corresponding CB in the reservation pool (this can happen when a MH fails to reserve a CB due to the nonperfect next-cell prediction algorithm)2. Therefore, the handoff call has to compete with new calls for the free channels. However, network congestion can occur due to too many arriving calls. This can result in the handoff

2

For example, assuming there are three neighboring cells: Cell 1, Cell 2 and Cell 3. And a handoff user is in Cell 1. It cooperates with all neighboring Base Stations and runs ‘Next-cell predicting algorithm’. Assuming the result of the algorithm tells the system that the ‘Reservation’ should happen in Cell 2. Then the system will ‘reserve' channels in Cell 2. Unfortunately, the algorithm could make a mistake at a small probability, and the user actually handoffs to Cell 3 instead of Cell 2. Once the user enters Cell 3, it could not find its corresponding ‘Reserved’ channels because it actually reserved channels in Cell 2.

call still being not served. Next, it will try to use GC3. If all the GC are used by other handoff calls, this handoff call will be dropped. (2) New Call Blocking: If a new call cannot find free channels, it will be immediately blocked. It should be pointed out that the new calls should be served based on their priorities. However, for determining their priorities, only class urgency is used instead of RO as in the case of handoff calls.

5. SIMULATION RESULTS To evaluate the performance of the proposed 3G sensor network QoS management system, a network model of macro cells with channel capacity C was constructed using a discrete-event driven simulator – OPNET & C (Fig. 6). The cell radius is assumed to be 500m, which is a typical size for future mobile telemedicine system. Three different velocities are assumed: 2m/s (walking), 10m/s (normal-speed car), and 20m/s (high-speed vehicle). Furthermore we assume that the three classes of calls have the same percentages of three velocities in order to emphasize the influence of class urgency on the computation of RO. A cluster of seven cells is assumed and each cell keeps contact with its six neighboring cells. Based on the modeling analysis in Section F, a number of call generators generate Poisson arrivals of new and handoff call requests from different service classes. Each medical sensor has the physical architecture of Fig. 7 and operates based on the finite state machine of Fig. 8. The simulation of wireless bandwidth reservation and allocation algorithm is shown in Fig. 9.

Fig. 6 Simulation topology: 3G Mobile Sensor Networks

Fig. 7 Medical Super-Sensor Node

3

The concept of GC can also be use in CDMA-based 3G wireless networks. In CDMA networks, a channel means a specific pseudonoise code for a user.

Fig. 8 Finite State Machine of each sensor node

Fig 9 Wireless Resource Allocation Simulation As shown in Fig. 10, 11, our call admission control algorithm shows a good stability. The handoff dropping rate and new call blocking rate are converged into the stable values after some simulation time. If we assume that the position and velocity of MH do not have much influence on the RO of each handoff call, except for the CU of each class 4, we can see the effect of RO on improving HDP of each class of handoff calls.

4

This can be achieved by assuming that each class of calls has the same percentage of all types of moving users such as pedestrians and cars.

We consider two classes of calls: Class 1 corresponding to handoff-guaranteed traffic (Ambulance) and Class 2 corresponding to normal patients’ traffic (non-real-time). Two important cases are considered: light handoff load (HCD = 25%) and heavy handoff load (HCD = 75%). The reason for choosing these two cases is that we may see the effect of RO on HDP more clearly.

Fig. 10 Handoff dropping rate (Adaptive vs. non-adaptive)

Fig. 11 New call blocking rate

Fig. 12 Handoff Dropping Rate becomes lower due to priority control Figure 12 (a) ~ (d) are our simulation results. The X-axis represents the percentage of a given class of calls among all handoff calls. It varies from 20% to 100%. The Y-axis is the value of HDP multiplied by 10,000. It can be seen that HDP of Class 1 calls decreases when RO is employed. Although in light handoff load case, the reduction is not very obvious (Figure 12 (a)), in heavy handoff load case the effect of RO is very dominant (Figure 12 (b)). This is not a surprising result since RO can assign class 1 calls the highest priority when only CU is considered. Unfortunately, HDP increases for class 2 calls (Figure 12 (c) and (d)), especially in heavy handoff load case (Figure 12 (d)). This is because class 2 calls get the lowest priority when their RO is compared to other classes. When the network is under congestion, the class 2 calls have the highest probability for being dropped among the three classes of calls.

6. CONCLUSIONS AND FUTURE WORK This paper reported our mobile telemedicine work that is based on 3G mobile sensor networks. This paper addressed the problem of providing M-QoS guarantee for multi-class medical services in the 3G mobile networks. A multi-weighted algorithm for computing priorities of handoff requests was proposed in order to serve arriving multi-class calls with highly diverse QoS parameters. A dominant feature of our approach is combining patient handoff behaviors with the call admission procedure.

Some of future work will include the integration of routing algorithms with sensor query forwarding and dynamic lowenergy security-key distribution problem in the transmission of crucial medical data.

REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9.

T. D. Moore, “Telemedicine: The future?” Generics Group, Version 2.0, available from http://www.genericsgroup.com. I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless sensor networks: a survey”, Computer Networks (38), Elsevier, 2002, pp 1-20. H. Joseph, B. Swafford and S. Terry, “MEMS in the medical world,” Sensors Magazine, April 1997, available from: http://www.sensorsmag.com. B. Woodward, R.S. H. Istepanian, “Design of a Telemedicine system using a mobile telephone,” IEEE Transactions on Information Technology in Biomedicine, Vol. 5, No.1, March 2001. S. K. Das, et al., “A Call Admission and Control Scheme for Quality-of-Service (QoS) Provisioning in Next Generation Wireless Networks,” Wireless Networks 6, pp. 17-30, 2000. Carlos Oliveira, et al., “An Adaptive Bandwidth Reservation Scheme for High-Speed Multimedia Wireless Networks,” IEEE JSAC, vol. 16, no. 6, pp. 858-873, August 1998. R. Govindan, J. Hellerstein, W. Hong, S. Madden, M. Franklin, S. Shenker, The Sensor Network as a Database, Technical Report 02-771, Computer Science Department, University of Southern California, September 2002. C-K Toh, “Wireless ATM and Ad-Hoc Networks : Protocols and Architectures,” pp. 88-98, Kluwer Academic Publishers, 1997 (ISBN: 079239822X). L. Huang, Sunil Kumar and C.-C. Jay Kuo, Adaptive Resource Allocation for Multimedia QoS Management in Wireless Networks, IEEE Trans. Vehicular Technology (submitted: under revision).

10. T. Liu, P. Bahl and I. Chlamtac, "Mobility Modeling, Location Tracking, and Trajectory Prediction in Wireless ATM Networks," IEEE Journal on Selected Areas in Communications (JSAC), Vol. 16, No. 6, pp. 922-936, August 1998. 11. Fei Hu and Neeraj K. Sharma, “Multimedia Call Admission Control with QoS-guaranteeing in Next-generation

Wireless Networks: A Reservation-pool Approach,” Computer Networks (to appear), 2003. 12. N. Sadagopan, B. Krishnamachari, and A. Helmy, "The ACQUIRE mechanism for efficient querying in sensor networks", First IEEE International Workshop on Sensor Network Protocols and Applications (SNPA), in conjunction with IEEE ICC 2003, May 2003, Anchorage, AK, USA. 13. Michael Feramez, “Introduction to 3G Mobile”, (Presentation), Radcom Consultants Austalia Inc. 2002. 14. Z. Haas, M. Pearlman, “The performance of Query Control Schemes for the Zone Routing Protocol,” ACM SIGCOMM’98.