Adaptive Bandwidth Reservation and Scheduling for Efficient Wireless ...

632

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

Adaptive Bandwidth Reservation and Scheduling for Efficient Wireless Telemedicine Traffic Transmission Lu Qiao and Polychronis Koutsakis, Member, IEEE

Abstract—Telemedicine traffic carries critical information with regard to a patients’ condition; hence, it requires the highest transmission priority compared with all other types of traffic in the cellular network. The need for expedited errorless transmission of multimedia telemedicine traffic calls for a guaranteed bandwidth to telemedicine users. This condition, however, creates a tradeoff between the satisfaction of the very strict quality-of-service (QoS) requirements of telemedicine traffic and the loss of this guaranteed bandwidth in the numerous cases when it is left unused due to the infrequent nature of telemedicine traffic. To resolve this complex problem, in this paper, we propose and combine the following two techniques: 1) an adaptive bandwidth reservation scheme based on road map information and on user mobility and 2) a fair scheduling scheme for telemedicine traffic transmission over wireless cellular networks. Index Terms—Bandwidth reservation, mobility, scheduling, telemedicine traffic.

I. I NTRODUCTION

C

URRENTLY, in the cases of motor vehicle accidents, which are the number one cause of violent death in the United States, prehospital teams provide on-scene initial assessment and resuscitation and transmit this information to a physician, mainly through voice communication; therefore, the physician can only make an assessment based on what is described and cannot continuously monitor an injured victim through visual communication (e.g., video) while, at the same time, receiving data of major importance with regard to the victim’s vital signs [7], [10]. According to the national health care provider of the U.K., i.e., the National Health Service (NHS), 1.3 million emergency calls were received in 2007, more than 25% of which were classified as most urgent. Unfortunately, in the cases of 13% of the calls, the patients’ arrival to the hospital was delayed for a variety of reasons, mainly having to do with traffic congestion. It is clear, therefore, that ambulances need to provide the patient with health services that would only have been possible, until recently, when the patient would be admitted to the hospital [33]. Manuscript received January 24, 2010; revised October 20, 2010 and November 19, 2010; accepted November 19, 2010. Date of publication November 29, 2010; date of current version February 18, 2011. The review of this paper was coordinated by Dr. N. Ansari. L. Qiao was with the Department of Electrical and Computer Engineering, McMaster University, Hamilton, ON L8S 4L8, Canada. She is now with the Bank of Montreal, Toronto, ON M5X 1A1, Canada (e-mail: qiaoluu@ gmail.com). P. Koutsakis is with the Department of Electronic and Computer Engineering, Technical University of Crete, 73100 Chania, Greece (e-mail: polk@ telecom.tuc.gr). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TVT.2010.2095472

Telemedicine is already employed in many areas of health care, such as intensive neonatology, critical surgery, pharmacy, public health, and patient education. In addition to ambulance vehicles, it is also of critical importance for the provision of health care services at understaffed areas such as rural health centers, ships, and trains, airplanes, as well as home monitoring (e.g., Americans of age 85 and above represent the fastest growing population in their country; therefore, in the future, they will have to be monitored at home, because it will be impossible for everybody to stay at the hospital) [2], [6], [28], [29]. Mobile health care (M-health; “mobile computing, medical sensor, and communication technologies for health care” [4]) is a new paradigm that brings together the evolution of emerging wireless communications and network technologies with the concept of “connected health care” anytime and anywhere. Various M-health studies have been conducted within the last few years on very significant aspects of public health [4]–[10], [13], [28], [30]–[35]. In many of these studies, the efficient use of the cellular network resources was of paramount importance for the correct and rapid transmission of all types of telemedicine traffic (video, audio, and data), particularly because only a limited bandwidth could be devoted to the transmission of this very urgent and crucial, for the patients’ life, type of traffic (e.g., 40–60 Kb/s in [30], 50–300 Kb/s in [33], 360 Kb/s in [34], and less than 300 Kb/s (downlink) in [35] due to the limitations in the cellular system used in each country and depending on the type of telemedicine traffic transmitted in each study). In [31], the authors suggest that the problem of limited bandwidth in an emergency situation, where the number of patients being transferred in ambulances to the hospital is overwhelming, can only be handled by transmitting vital signs’ data to the hospital in a reduced and packed format over a Third-Generation (3G) cellular network. Hence, one of the main tasks of current research in the field is the design of system hardware and software to implement a mobile telemedicine system that can transmit, with high quality of service (QoS), significant loads of multiplexed information in next-generation wireless cellular networks [7]. II. C ONTRIBUTION OF T HIS PAPER The uplink bandwidth available to mobile users will significantly be larger in next-generation cellular networks. However, one common characteristic of almost all of the aforementioned referenced studies is that they focus solely on the transmission of telemedicine traffic over the cellular network, without taking into account the fact that regular traffic, which represents the bulk of traffic in the network, also has strict QoS requirements.

0018-9545/$26.00 © 2011 IEEE

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

The integration of the sparse but critical telemedicine traffic with regular traffic is a complex problem and not a mere scheduling one, given that even the most sophisticated Multiple Access Control (MAC) scheme cannot properly service an urgent application if there is no bandwidth available in the cell at a given time. The only way to achieve this condition would be at the severe dissatisfaction of regular users who are already active in the specific cell. It should also be mentioned that, in several of the referenced studies, despite the critical importance of the telemedicine systems used, the accuracy in information transmission was low. For example, in [9], a high percentage (27%) of electrocardiogram (ECG) data transmission interruptions took place due to Global System for Mobile Communications (GSM) channel congestion. Therefore, along with the implementation of an efficient scheduling mechanism, a bandwidth reservation scheme is also necessary. This scheme will reserve the proper bandwidth among cells for handoff traffic and, particularly, for handoff telemedicine traffic. This case is the problem that we try to tackle in this paper, for the first time in the relevant literature, to the best of our knowledge (with the exception of a short section in [33], which, however, considers just telemedicine video). In [1] and [27], we have introduced Multimedia Integration Multiple Access Control (MI-MAC), a flexible time-division multiple access (TDMA)-based protocol that was shown to achieve superior performance compared with a number of other TDMA- and wideband code-division multiple access (WCDMA)-based protocols of the literature when integrating various types of multimedia traffic over one cell of a cellular network. Using MI-MAC as a basis, we focus in this paper on the problem of transmitting urgent telemedicine traffic and reserving adequate bandwidth throughout the emergency vehicle’s trajectory. More specifically, we propose a new adaptive bandwidth reservation scheme based on road map information and on user mobility and introduce several new scheduling ideas for the efficient and fair transmission of all types of mobile telemedicine traffic and its integration with regular traffic. This paper is organized as follows. Section III presents our system model. Section IV discusses our adaptive bandwidth reservation scheme. Section V presents our proposed scheduling scheme, and Section VI presents our results and a discussion about these results. Two earlier versions of one part of our contribution, i.e., our proposed scheduling scheme, were presented in [25] and [26]. III. S YSTEM M ODEL The following four types of telemedicine traffic are considered in this paper: 1) electrocardiograph (ECG); 2) X-ray; 3) video; 4) medical still images. Similar to [7] and [13], which use data from the Massachusetts Institute of Technology-Beth Israel Hospital Arrhythmia Database, we consider that ECG data are sampled at 360 Hz with 11 b/sample precision. We set a strict upper bound of just one channel frame (75 ms; this choice is explained in Section V)

633

for the transmission delay of an ECG packet. We consider that a typical X-ray file size is 200 kB [6], [35] and that the aggregate X-ray file arrivals are Poisson distributed with mean λX files/frame. Medical image files have sizes that range between 15 and 20 kB/image [7] and are Poisson distributed with mean λI files/frame. The upper bound for the transmission delay of an X-ray file is set to 1 min (the average download time needed for it in [35]), and the upper bound for the transmission delay of an image is set to 5 s. A discussion on the strictness of these upper bounds will be made in Section V. Because H.263 is still the most widely used video-encoding scheme for telemedicine video, we use, in our simulations, real H.263 videoconference traces from [3] and [21] with a mean bit rate of 91 Kb/s, a peak rate of 500 Kb/s, and a standard deviation of 32.7 Kb/s. Due to the need for very high-quality telemedicine video, the maximum allowed video-packet-dropping probability is set to 0.01% [38]. The respective upper bound for the videopacket-dropping probability of regular MPEG-4 video users is assumed to be equal to 1% [16]. Packets of a video frame, both in the case of telemedicine and in regular video users, are dropped if they are not transmitted before the arrival of the next video frame (i.e., within 40 ms for the MPEG-4 traffic and within an integer multiple of 40 ms for the H.263 traffic). This strict requirement is necessary, because streaming video requires bounded end-to-end delay so that packets arrive at the receiver in a timely fashion to correctly be decoded and displayed. If a video packet does not arrive on time, the play out process will pause, which is annoying to the human eye [34]. For example, in the medical robotic system OTELO [34], the maximum acceptable end-to-end delay for video packets is set to 350 ms.1 The four types of regular traffic that were considered in [1] (voice, e-mail, web traffic, and MPEG-4 videoconference traffic) are also considered in this paper to study the practical scenario of many different types of users who simultaneously attempt to access the network and hence aggravate the access for telemedicine users. We use two real video traces to generate the MPEG-4 videoconference traffic: one video trace with a mean rate of 400 Kb/s, a peak rate of 2 Mb/s, and a standard deviation of 434 Kb/s and another video trace with a mean rate of 210 Kb/s, a peak rate of 1.5 Mb/s, and a standard deviation of 182 Kb/s. The two video traces are chosen with equal probability. Video sources have exponentially distributed sessions with a mean duration of 5 min (this duration has been denoted by global trials as an expected one for users of another wireless cellular video application [12]). In [34], the authors have experimented with various packet sizes, in the range of 128–512 B s, to achieve the best reliable connectivity for medical video streaming, and they found that the best size for this purpose was 300 B. We use this size as the information payload, plus a header of 5 B (i.e., a packet size of 305 B) for the transmission of all types of traffic in this paper. 1 To the best of our knowledge, no telemedicine video models have been studied in the recent literature, and we have not found actual telemedicine video sequences. Therefore, we assumed that a telemedicine video, where the camera focuses on a specific part of the patient’s body, would be similar to a videoconference trace, where the camera focuses on a user’s face or hand. Hence, we used real videoconference traces for this paper. We believe that it is the most reasonable assumption that can be made to get some insight into the problem, in the absence of actual telemedicine video.

634

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

Fourth-generation (4G) mobile data transmission rates of up to 20 Mb/s are planned [22]; therefore, we consider this value to be the uplink rate. IV. A DAPTIVE BANDWIDTH R ESERVATION BASED ON M OBILITY AND ROAD I NFORMATION Within a picocell, spatially dispersed source terminals share a radio channel that connects them to a fixed base station (BS). The BS allocates channel resources, delivers feedback information, and serves as an interface to the mobile switching center (MSC). The MSC provides access to the fixed network infrastructure. The dissatisfaction of a wireless cellular subscriber who experiences forced call termination while moving between picocells is higher than that of a subscriber who attempts to access the network for the first time and experiences call blocking [20]. For this reason, it is important that the system can, at any point in time, accommodate newly arriving handoff calls in any cell of the network. This need becomes imperative for highly urgent traffic such as telemedicine traffic, calling for seamless handoff. On the other hand, the policy of reserving a significant amount of bandwidth for possible handoff calls may lead to a portion of the bandwidth being left unused due to small volumes of handoff traffic, whereas at the same time, the remaining available resources for newly generated traffic from within the cell may not suffice. The discussion in the following section focuses on this problem. A. Network and Mobility Models We consider a hexagonal cell architecture, as shown in Fig. 1. We consider in all cells the uplink (wireless terminals to BS) wireless channel. The following mobility assumptions are adopted from [17] and [18]. Each cell has six neighbors. The cell diameter is 300 m. Roads are modeled by straight lines. Each road is assigned a weight (ωj for road j), which represents the traffic volume. Each new call is generated with a probability of 50% to be moving on the road and 50% to be stationary. Moving users are assumed to be traveling only on the roads and are placed on each road i with the following probability: ωi N j=1 ωj where N is the total number of roads. The initial location of a moving user on a particular road is a uniform random variable between zero and the length of that road. During their call, stationary callers remain stationary, and mobile users travel at a constant speed. Mobile users can travel in either direction of a road with an equal probability and with a speed randomly chosen in the range of [36, 90] km/h. At the intersection of two roads, a mobile user might continue to go straight or turn left, right, or around with probabilities 0.55, 0.2, 0.2, and 0.05, respectively. If a mobile user chooses to go straight or turn right at the intersection, it needs to stop there, with probability 0.5 for a random time between 0 and 30 s due to a red traffic light. If the user chooses to turn left or around, it needs to stop there for a random time between 0 and 60 s due to the traffic signal. Each BS is loaded with the road map

Fig. 1.

Road map and cellular network model.

of its coverage area and its neighboring cells. Mobile stations report their position to the BS of their cell through a control channel. The position information includes the mobile user’s exact location (cell and road), moving direction, and speed and can be provided with an accuracy of 1 m through the Global Positioning System (GPS) [17]–[19]. B. Our Proposed Adaptive Bandwidth Reservation Scheme The following two main categories of methods exist to perform mobile stations’ movement predictions: 1) the locationbased method, which measures the current location and movement direction of the ongoing calls and predicts the next cell to be reached by the mobile stations, and 2) the historybased method, which considers the movements of previous subscribers and assumes that the ongoing subscribers move in similar patterns or tries to deduct each user’s movement pattern [24]. Our proposed method belongs to the first category. The bandwidth reservation schemes proposed in [17]–[19] require mobile stations to report their location information to the BS every T s (60 s in [17], 1 s in [19], and an adjustable interval in [18]). For this period of time, if there is a probability based on the mobile’s trajectory that it will move to a new cell, then a certain amount of bandwidth is reserved in all possible future cells to which the mobile may move. In [17], a prediction is made by the system based on each updated position information and the road map. In [18], the mobile’s recorded moving history is also used for the prediction. In [20], the authors do not use a standard road map for their study but generate a random map for each simulation. The common disadvantage of these approaches is that the length of the report period yields a tradeoff between the prediction accuracy and the computational load imposed on the system. If mobile stations frequently report their position, the computational load of processing this information and using it for bandwidth reservation will be very high; on the other hand, if mobile users’ position reports are very infrequent, the prediction on the mobile’s trajectory can be untrustworthy and lead to an unnecessary waste of the reserved bandwidth in neighboring cells. Because of this tradeoff, the authors in [18] and [19] use additional dynamic mechanisms in their schemes to adjust the amount of reserved bandwidth for users in neighboring

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

cells based on the quality of the system performance, and we will explain in the following discussion why this “corrective” approach, which imposes further computational load on the system, is not needed in our scheme. To eliminate the problems introduced by the time-based location information reports of the mobile station to the BS, we propose the use of distance-based information reports. More specifically, we set the road intersections and cell boundaries in the map to be the check points of our system, as shown in Fig. 1. Each cell boundary represents a unique check point, whereas around each road intersection, four check points are set, one in each possible direction that the user may choose after reaching the intersection. Each of the check points is assumed to be placed at a distance of 10 m from the intersection, which is a distance that even the slowest moving vehicles considered in this paper (with a speed of 36 km/h) will cover within 1 s after passing the intersection (naturally, if there are less than four possible directions for a mobile to follow after reaching an intersection, the number of check points needed around that intersection is smaller than four). Mobile stations only need to update their position information to the BS of their cell when they arrive at a check point. We assume that check points are configured by the wireless network provider into the mobile terminals’ GPS maps; even if the cell boundaries change or the user is handed off to another network (interoperability problem), the user can be notified by the network to download the new map to his/her mobile phone. If the BS of the current cell of a mobile station predicts, based on the station’s location (at a check point) and speed that the station is going to move to another cell (i.e., that the next check point for the station will be a cell boundary), it sends a notification to the BS of that cell, including the current bandwidth used by the station and the estimated arrival time at the next check point. Hence, the proper amount of bandwidth is reserved for the station. For telemedicine video and regular video terminals, the bandwidth that is reserved in the next cell is equal to the remaining bandwidth that the terminal will need to complete its transmission. This bandwidth is declared in each video user’s initial request to the BS, as will be explained in Section V, where we discuss our proposed scheduling scheme. For all other types of users, the bandwidth that is reserved in the next cell is equal to their current bandwidth so that they will seamlessly continue its transmission. Our proposed approach not only guarantees the existence of adequate resources in the new cell for handoff users but also reduces the information update frequency. The prediction and adaptive reservation process executed by the BS can be summarized as follows: For each new user in the system If the mobile station is not stationary When the mobile station arrives at a check point Update the position information Estimate the next arrival check point and calculate the arrival time at that check point If the next check point is a cell boundary Reserve the proper amount of bandwidth for the station in the next cell.

635

Our proposed adaptive bandwidth reservation scheme based on distance-based location updates has the following two major advantages. 1) The position update duration is unique for every mobile user based on their different initial position and speed. This case is impossible to “capture” with the time-based location updates proposed in [17]–[19]. In addition, it creates less computational load than the time-based updates, which are, in most proposals in the literature, synchronized and therefore require simultaneous information transmission from the terminals to the BS. 2) One important parameter used to evaluate a bandwidth reservation scheme is bandwidth efficiency [17], which is calculated by f = Nr /Nq , where Nr is the reserved bandwidth, and Nq is the actual bandwidth utilized by handoff users. The closer f is to 1, the higher the efficiency achieved by the bandwidth reservation scheme becomes, because this case would mean that there is neither lack nor waste of bandwidth in the reservation procedure. The scheme in [17] outperformed other schemes with which it was compared, achieving, at best, f = 1.047379 (the schemes with which [17] was compared achieved a bandwidth efficiency in the range of 1.25). We argue that, with our distance-based approach, the bandwidth efficiency of our scheme can asymptotically reach 1. The reasons for this are given as follows. 1) Our scheme guarantees that there is no waste of bandwidth. For all users ready to hand off, the exact amount of bandwidth that they need is reserved in the new cell, and this cell is known with precision, with the use of the four check points around each intersection. The only case when bandwidth can be wasted is when a mobile station that is predicted to move to another cell makes a stop before entering this cell (this case is not included in our adopted mobility model). Given that the distance between a check point set close to an intersection and a cell boundary is, in all cases, much smaller than a cell diameter of 300 m, this case can be considered a rare exception. 2) Our scheme also guarantees that there will be no lack of bandwidth for handoff users. As will be explained in Section V, the bandwidth needed for all types of telemedicine and regular mobile stations that do not transmit video is 32 Kb/s (only one slot per channel frame). Therefore, this bandwidth is very small compared to the total channel capacity of 20 Mb/s and can generally be reserved in the next cell, with the rare exception of cases when the channel is overloaded with traffic. In the case of handoff telemedicine users, if the needed bandwidth is not available, then e-mail and web users are preempted in the new cell for the system to grant this bandwidth to the high-priority telemedicine traffic. When e-mail and web reservations are canceled, the BS notifies the affected data terminals and places them at the front of the respective (e-mail or web) queue of terminals that await bandwidth allocation. The priorities among telemedicine traffic types in bandwidth reservation are set in the same way as the scheduling priorities, which

636

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

TABLE I T YPES OF T RAFFIC T RANSMITTED IN O UR S YSTEM IN D ESCENDING P RIORITY O RDER

Fig. 2.

Frame structure for the 20 Mb/s channel, with a frame duration 75 ms.

will be explained in Section V. With this guarantee of the existence of bandwidth for telemedicine users that need to be handed off from one cell to another, our system can offer zero blocking probability to telemedicine traffic, which is imperative for ambulances that try to establish connections to BSs as they move toward the hospital [32]. V. P ROPOSED S CHEDULING S CHEME We extend our previous work on the MI-MAC protocol [1], [27] to focus on the efficient and fair scheduling for transmissions from mobile telemedicine users. The uplink channel time is divided into time frames of fixed length. Each frame is divided in time slots, and each slot accommodates exactly one fixed length packet that contains voice, video or data information, and a header. Because we use a different packet size (305 B, instead of 53 B, which was used in our previous work), the frame duration and structure in this paper are different from [1] and [25]–[27]. The frame duration is selected such that an active voice terminal generates exactly one packet per frame, and given that the speech codec is 32 Kb/s, the frame duration is 75 ms, and a frame accommodates 614 slots (slot duration of 0.122 ms). As shown in Fig. 2, which presents the channel frame structure, each frame consists of the following two types of intervals: 1) the request interval and 2) the information interval. The request intervals consist of slots, which are subdivided into minislots, and each minislot accommodates exactly one fixed length request packet. By using more than one minislot per request slot, a more efficient usage of the available request bandwidth (in which users contend for channel access) is possible. We chose the number of minislots per request slot to be equal to 10 to allow for guard time and synchronization overheads for the transmission of a generic request packet and for the propagation delay within the picocell. Handoff telemedicine traffic is treated with full priority, with the use of the adaptive bandwidth reservation scheme discussed in Section IV. The scheduling priority order used by the BS for all traffic types originating from handoffs or from within the cell is shown in Table I. The choice of priorities has been made based on the importance that each of these traffic types currently has for medical care [2], [4]–[10], [13]. The same priority order is used in our adaptive bandwidth reservation scheme, as mentioned in Section IV. For regular traffic, prioritization has

been defined based on the strictness of the QoS requirements for each traffic type. Video and voice have the same QoS requirement of a maximum of 1% packet dropping, but video traffic is much burstier; therefore, it is granted priority over voice. If we did not make this choice, voice users would first enter the channel and keep their reserved slots for the whole duration of their talkspurt, hence preventing video users from acquiring their needed slots and transmitting their packets before the maximum packet delay is reached and the packet is dropped. With regard to the transmission of request packets by the mobile stations to acquire access to the channel, we can ensure the priority of the traffic types with the strictest QoS by using the two-cell stack protocol [15] for contention resolution. The two-cell stack has the advantage that it clearly defines the end of contention among users of the same priority class; hence, users of lower priority classes cannot affect the QoS of users of higher priority classes. The use of the two-cell stack protocol also enables users who move from cell X to cell Y without having been able to access the channel in cell X to transmit their request packets to the BS of cell Y with higher priority than new users who originate in within cell Y. Similar to previous works in the literature (e.g., [14]), we have found that a small percentage of the bandwidth suffices to be used for requests. This percentage is only 0.98% in this paper (six slots used for requests out of the 614 slots of the channel frame), and this value, which corresponds to 60 minislots, has been found through extensive simulations to provide a good tradeoff between allowing sufficient bandwidth for all types of terminals to transmit their requests and allowing a large-enough number of slots for terminals with a reservation to transmit their information packets. We assume the existence of a call admission control (CAC) module at the entrance of the system, through which the BS is notified of the types of telemedicine users who are active in the cell. Therefore, given that the twocell stack protocol needs a minimum of two minislots to resolve contention or to denote the absence of contention for each type of user in a specific channel frame, the BS can decide how many request slots should be dedicated to telemedicine users. For example, in the extreme case when all four types of telemedicine traffic are present, both from handoffs and from

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

traffic within the cell, a minimum of 16 minislots will be needed (two minislots per type of handoff telemedicine traffic and two minislots per type of telemedicine traffic from within the cell) to be dedicated to telemedicine users. If more than 16 minislots are needed to resolve the contention among telemedicine users, then contention will continue until all collisions have been resolved, and only then will users of regular traffic (both handoff and from within the cell) get the opportunity to transmit their request packets. Still, the case of 16 or more minislots being needed to resolve telemedicine traffic contention is quite infrequent because of the sparse nature of telemedicine traffic. Our results show that our scheme can satisfy the strict QoS requirements and the urgency of telemedicine traffic by devoting most of the time less than 16/6140 = 0.26% of the total bandwidth for telemedicine request packets. When less than six request slots are needed to resolve the contention among all types of users (telemedicine and regular), the remaining slots until the end of the request interval are used, only for the current frame, as information slots so that the respective bandwidth may not be wasted. In reality, however small the picocell radius is, the channel fading experienced by each user is different, because users move independent of each other; therefore, in this paper, fading per user channel is considered. We adopt the robust threestate [i.e., good state, short bad state, and long bad (LB) state] Markov chain error model for wireless channels presented in [11], and we use the idea (analyzed in [27]) that the system should take advantage of the “problem” created when a regular video user experiences the LB channel state (error burst) and cannot transmit in its allocated uplink slots. When the BS determines, through probabilistic estimation and monitoring of the slots where a regular video terminal is scheduled to transmit, that the terminal is in the LB state, if that terminal has more reserved slots in the current channel frame, the BS deallocates these slots. Full priority for these slots is given to handoff telemedicine video terminals, followed by telemedicine video users that originate within the cell, then by handoff regular video users, and, finally, by regular video users that originate within the cell. The allocation of the abandoned slots within each priority type is first come, first served (FCFS). With regard to the impact of higher layers, e.g., transmission control protocol (TCP), on the performance of our scheme, recent work (e.g., [40] and [41]) has provided solutions to the problem of differentiating packet losses caused by different layers. One major issue not tackled in [1] and [27] was that of fairness in the scheduling scheme. If users of the same type of traffic are served in a FCFS order, certain users may have their QoS severely violated, whereas other users get exceptional QoS, which could result in a seemingly acceptable average QoS over the total number of users. This approach is, however, unfair to users who arrive later in the network and hence are placed at the bottom of the BS service queue; the problem is particularly significant in the case of telemedicine video and regular video users, where early arriving users may dominate the channel. For this reason, we introduce the following fair scheduling scheme for telemedicine video and regular video users (the scheme is separately enforced among users of each of the two

637

types of traffic, because telemedicine video users have higher priority). The BS allocates bandwidth by currently comparing the channel resources to the total requested bandwidth from all active video users. If the available bandwidth is larger than the total requested bandwidth, all users will be assigned as many slots as they have requested. If, however, the available bandwidth is smaller than the total requested bandwidth, then the available bandwidth will proportionally be shared among video users. More specifically, let M be the number of currently idle information slots in the frame and B(i) be the amount of bandwidth that will be assigned to video terminal i in every channel frame. B(i) is given by  D(i) (1) B(i) = M ∗ D(i)/ i

 where D(i) is the ith user requested bandwidth, and i D(i) is the total bandwidth requested by all of the video terminals at that moment. The aforementioned scheme does not need to be implemented on any of the other types of traffic considered in this paper, because they are allocated only one slot per frame. The simple proportional fair scheduling algorithm, which is previously described and used in this paper, is obviously only one of the many available solutions in the literature for fair scheduling. Given the complex problem of scheduling a large number of traffic types, including urgent telemedicine traffic, the simplicity of the algorithm is the reason that we choose to apply it to our scheme. An additional scheduling policy is needed for X-ray and medical image traffic, because the upper bounds for their transmission delays are equally strict with the upper bounds for telemedicine video and ECG traffic. This strictness can be explained by the fact that X-ray and medical image terminals are allocated only one slot per frame (i.e., 32 Kb/s, which is similar to regular voice, e-mail, and web traffic) to allow for the significantly larger numbers of slots needed by telemedicine and regular video users. Therefore, a typical X-ray file of 200 kB needs 50 s to be transmitted (whereas the upper bound for its transmission delay is set to 1 min), and an average-sized medical image file of 17.5 kB needs around 4.38 s to be transmitted (whereas the upper bound for its transmission delay is set to 5 s). The allocation of only one slot per frame to these types of traffic, although defensive enough to prevent cases where newly arriving telemedicine video traffic cannot find enough resources to transmit, is not the most efficient in terms of bandwidth utilization. On the other hand, if these types of traffic are constantly granted more than one slot per frame, this case could lead to the existence of very few idle slots for the bursty telemedicine video users. Therefore, to maximize system bandwidth utilization, we use the following scheduling policy. After the end of the request interval at the beginning of each frame, the BS is aware of whether there are information slots during the current frame, which will be left idle. These slots are allocated, only for the current frame, to X-ray and medical image users who have already entered the network (with priority to X-ray users) as additional slots to their guaranteed single slot per frame. Hence, the telemedicine traffic transmission is expedited, and channel throughput is increased.

638

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

Fig. 3. Effect of regular video traffic on X-ray traffic.

Finally, there is no need for the preemption of regular video traffic in our scheme. In case more bandwidth is needed than the bandwidth that can be gained from the preemption of e-mail and web users, then in the next scheduling period, for each regular video user (i.e., upon the arrival of the next video frame that the user needs to transmit), the user will not be assigned any slots until the required bandwidth for handoff telemedicine traffic is allocated to the respective telemedicine users. The only case in which the preemption of regular video traffic would be necessary would be the case when telemedicine video traffic became dominant in the network. However, such a case is highly unlikely to practically occur.

Fig. 4. Percentage of regular video users who experience packet loss larger than 1% versus the number of regular video users.

VI. R ESULTS AND D ISCUSSION A. Evaluation of the Performance of the Proposed Scheme We use computer simulations to study the performance of our scheme. The simulator is written in the C programming language. Each simulation point is the result of an average of 100 independent runs (Monte Carlo simulation), each simulating 1 h of network operation. All our results have been derived for 95% t-confidence intervals (constructed in the usual way [23]). We use the road map shown in Fig. 1 for our simulations. The following weight values ωj were assigned, without loss of generality, to the eight roads in the road map: {1, 1, 5, 1, 6, 1, 1, 1}. The two larger weights correspond to the two main roads (shown in Fig. 1 in bold black). We have also experimented with other weight values, and this change had no effect on the nature of our results. Our results are presented in Figs. 3–13 and in Table II. In these results, we vary the percentage of telemedicine traffic over the total channel capacity between 5% and 10%, in increments of 1%. We have used 50 different “traffic combinations” of the four types of telemedicine traffic to create this load; this way, our results can be representative of different real-world cases, where one type of telemedicine traffic might be more dominant than other types in any given moment. At the beginning of each simulation run, one of the 50 “traffic combinations” is chosen at random. We do not provide a detailed results comparison between our proposed adaptive bandwidth reservation/scheduling scheme and other schemes in the literature for the following two reasons: 1) To the best of our knowledge, this paper is the first to handle the integrated transmission of various types of urgent

Fig. 5. Regular video packet dropping versus the number of regular video users.

Fig. 6.

Effect of regular voice traffic on telemedicine video traffic.

telemedicine traffic and regular traffic, and 2) such a comparison, for the transmission of regular traffic, was made in [1] and [27]; therefore, this paper, which is focused on the transmission of telemedicine traffic, will achieve even better results compared with schemes that are not optimized for the transmission of urgent/critical traffic. However, this comparison would be rather unfair. Still, we will provide a few indicative results of such a comparison in this section, and we conceptually compare our scheme with three other recent works in the literature. Fig. 3 shows the effect that the increase in the number of regular video traffic users has on the QoS of X-ray traffic. The maximum number of video users (57) corresponds to 95% of the total traffic being generated by regular video traffic. As

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

Fig. 7.

Effect of regular voice traffic on telemedicine image traffic.

Fig. 8.

Effect of telemedicine video traffic on regular video traffic.

639

Fig. 11. Effect of telemedicine video traffic on X-ray traffic.

Fig. 12. Effect of telemedicine video traffic on telemedicine image traffic.

Fig. 9. Percentage of telemedicine video users who experience packet loss larger than 0.01% versus the number of telemedicine video users.

Fig. 13. Improvement on handoff-voice-packet-dropping probability with the use of adaptive bandwidth reservation. TABLE II I MPACT OF THE B OTTLENECK C ELL ON THE R ESULTS FOR A LL T YPES OF T RAFFIC

Fig. 10. Telemedicine video packet dropping versus the number of telemedicine video users.

640

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

shown in the figure, the average delay for the start time of transmission of X-ray very slowly files increases and does not become larger than 710 ms, even for very large numbers of regular video users. This condition means that the upper bound of 1 min for the transmission delay of an X-ray file is met in all cases examined in this paper (< 710 ms until the transmission starts, i.e., until the specific file acquires a slot in which it will be transmitted, plus 50 s for the actual transmission of the file). The Poisson-distributed X-ray file arrivals in our simulations have a rate λX (files/frame) that varies between 0.001 and 0.005, therefore between 21 and 107 Kb/s. These rates are quite high, because for example, a rate of 0.005 X-ray file transmission per 75 ms frame corresponds to an average transmission of four X-ray files per minute in the seven picocells shown in Fig. 1, which, in real life, would correspond to a massive crisis situation. As also shown in Fig. 3, even for zero video users, there is a delay of slightly more than three channel frames (232 ms) in the start time of transmission of an X-ray file due to the existence of handoff traffic and other types of telemedicine and regular traffic in the cells. For the same range of regular video users (0–57): 1) Telemedicine image files are transmitted, on the average, within less than 170 ms [Poisson-distributed telemedicine image file arrivals in our simulations have a rate λI (files/frame) that varies between 0.03 and 0.24, therefore between 56 and 448 Kb/s]. 2) ECG packets are transmitted, on the average, within half a frame (37.5 ms). 3) The average packet-dropping probability of telemedicine video ranges between 0.0044% and 0.0057%, because it is affected only by the wireless channel errors. All the aforementioned values are far below the acceptable upper bounds; hence, our adaptive bandwidth reservation and scheduling schemes are shown to provide the required QoS for all types of telemedicine traffic. We also simulated two additional cases, where the packet size was smaller or larger than the 305-B packet size used for the aforementioned results. The comparison of the respective results, for a packet size equal to 53 B and 517 B,2 respectively, can reveal the impact of the choice of packet size on the performance of our proposed schemes. As explained in Section V, the change in packet size leads to changes in the frame duration and structure. Therefore, for packets of 53 B, the frame duration is 12 ms, and a frame accommodates 566 slots. For packets of 517 B, the frame duration is 128 ms, and the frame accommodates 619 slots. The respective results for cases 2 and 3 were identical with those for a 305-B packet size. With regard to case 1, our results have shown that telemedicine image files are transmitted, on the average, within less than 177 ms (packet = 53 B) and 168 ms (packet = 517 B). The aforementioned results show that the impact of packet size on our results is negligible. Figs. 4 and 5 show that the increase in the number of regular video traffic clearly affects that type of traffic, because 2 The 53-B packet size was used in our previous work in [1] and [25]–[27]. The 517-B packet size (512-B information payload plus a header of 5 B) was chosen, because the maximum packet size used in [34] was 512 B.

the video-packet-dropping probability of regular video users significantly rises above the 1% acceptable upper bound. This conclusion, combined with the results presented in Fig. 3, shows that only the regular video traffic suffers from an increase in the number of regular video users; therefore, our combined schemes succeed in offering absolute priority to telemedicine traffic. Figs. 4 and 5 also show that the use of our fair scheduling scheme substantially helps improve both the average videopacket-dropping probability over all regular video users and the individual QoS of each regular video user. The improvement in the individual QoS, as shown in Fig. 4, is due to the allocation of available resources in each channel frame proportional to each user’s requested bandwidth. The result shown in Fig. 5 with regard to the improvement in the average video packet dropping is, again, due to the proportionate allocation; our scheme prevents the case where a user whose transmission deadline is not imminent may dominate the channel, hence not allowing users with imminent transmission deadlines to transmit. Figs. 6 and 7 show the almost-negligible effect on the QoS of telemedicine users that stems from the increase in the number of voice users. We present results that show telemedicine video packet dropping and telemedicine images transmission delay, and it is clear in both figures that only in the case of very high voice loads can there be deterioration in telemedicine traffic QoS. The reason, again, is that our combined scheduling and adaptive bandwidth reservation schemes guarantee full priority to all types of telemedicine traffic. The average telemedicine image traffic transmission delay does not exceed 5 s, as shown in Fig. 7 (the average start of transmission delay +4.38 s, as explained in Section V), even when the number of voice users exceeds 1180, which corresponds to 81.7% of the total channel capacity being utilized by voice users. Similar to the results presented in Fig. 3, even for 0 voice users, there is a small delay of more than two channel frames for the transmission of a telemedicine image file due to the existence of handoff traffic and other types of telemedicine and regular traffic in the cells. In addition, the telemedicine-video-dropping probability is shown in Fig. 6 to remain below the strict upper bound of 0.01% until the number of voice users exceeds 1040, which corresponds to 72% of the total channel capacity. On the other hand, Fig. 8 shows the effect that the increase in telemedicine load has on regular video traffic, which is the most bursty of all regular traffic types. This increase results in a very significant increase in regular video packet dropping due to the absolute priority of telemedicine traffic. Even in this case, however, there should be more than 15 telemedicine video users present in the seven picocells for the QoS requirement of a maximum of 1% regular video packet dropping to be violated. Figs. 9 and 10 are similar in nature to Figs. 4 and 5, but in this case, we study the effect of the increase of the number of telemedicine video users to the telemedicine-videopacket-dropping probability. It is clear in the figures, again, that our fair scheduling scheme significantly improves both the average telemedicine-video-packet-dropping probability over all telemedicine video users and the individual QoS of each telemedicine video user. An interesting comparison can be made between Figs. 4 and 5 and Figs. 9 and 10. In Figs. 4 and 5, the QoS requirement for regular video packet dropping

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

(< 1%) is violated when a larger percentage of users experiences a packet loss > 1%, compared with the percentage of telemedicine video users who experience a packet loss > 0.01% when the QoS requirement for telemedicine video packet dropping (< 0.01%) is violated, which is the case both for the implementation without a fairness mechanism and for the implementation with the fairness mechanism that we propose. The reason for this result is that our schemes are designed to offer maximum priority to telemedicine traffic; therefore, when telemedicine video packet dropping increases, the increase is almost “uniform” for all telemedicine video users. On the contrary, regular video users may exhibit differences in their video packet loss (even with the use of the fairness mechanism), because they are of lower priority, and hence, some of these users may not transmit in a traffic overload situation. Figs. 11 and 12 show that only under a medium-to-high number of telemedicine video users can there be an impact on the QoS of telemedicine image traffic and X-ray traffic. The reason is that both types of traffic have a higher priority in our schemes than telemedicine video, as shown in Table I. It should be clarified that, in the results presented in Figs. 11 and 12, all types of regular traffic are present in the system, with the exception of regular video traffic. Our results for all other types and combinations of telemedicine and regular traffic confirm that telemedicine traffic is negligibly affected by an increase in regular traffic, whereas regular traffic is severely affected by increased loads of telemedicine traffic. By increasing the average telemedicine traffic load up to 20% of the total bandwidth (i.e., 4 Mb/s), we have found that the maximum achievable throughput was, in all the scenarios tested, larger than 45% when regular video traffic was present (43% when using 53-B packets and 45.5% when using 517-B packets), and in most cases, it is larger than 55%. The reason that higher throughput cannot be achieved is that the QoS of regular video is first violated in all cases of traffic loads. In the absence of telemedicine traffic, however, the channel throughput achieved by our scheme can surpass 77% in the presence of regular video traffic (74% when using 53-B packets and 78% when using 517-B packets) and 88% when only voice, e-mail, and web traffic is present (83% when using 53-B packets and 88% when using 517-B packets; again, the impact of packet size ranges from very small to negligible). Therefore, our combined adaptive bandwidth reservation and fair scheduling schemes guarantee the absolute priority of telemedicine traffic and can also preserve the required QoS of regular traffic under all the traffic scenarios used in our extensive simulation study. In [27], we compared our scheme with two other TDMA MAC protocols in the literature, and our protocol was shown to outperform both protocols in all the results with regard to the QoS offered to different types of traffic and the channel throughput achieved. Indicatively, we mention here that the use of either [38] or [39] in the simulations conducted in this paper leads to a 9% and 13.5% lower throughput, respectively, on the average, over all the studied scenarios in which the two schemes can provide the required QoS for telemedicine traffic. These scenarios are the ones in which the telemedicine traffic load and the regular video traffic load are both low. In the cases of medium or high telemedicine traffic loads and/or high regular

641

video traffic loads, both schemes cannot provide the required QoS for telemedicine traffic and, in some cases, for regular traffic as well. For the reasons explained at the beginning of this section, we do not proceed with a more detailed results comparison. Fig. 13 shows the significant improvement achieved by the use of our adaptive bandwidth reservation scheme on the QoS of the most widely used cellular application, i.e., voice. The voice-packet-dropping probability of handoff voice users is smaller by 16.73%, on the average, compared with the case when the bandwidth reservation scheme is deactivated. Finally, in the results presented in Table II, we consider the case study of a bottleneck cell. Without loss of generality, we assume that the hospital is placed in cell E, in the map presented in Fig. 1, in a random position along the major road that cuts through cell E. To study this scenario, we made the following change in the mobility model presented in Section IV-A. All mobile users travel toward the hospital, i.e., they do not decide their next movement when they arrive at an intersection based on the probabilities of the mobility model but based on the shorter distance toward the hospital. Upon entering cell E, their speed falls below the minimum of 36 km/h considered in the rest of our simulations and reaches 0 by the time the user arrives at the hospital. When a mobile user reaches the hospital, the user becomes inactive, and another user with the same traffic characteristics is randomly generated in one of the seven cells of the map. The results in Table II show the expected deterioration in all the QoS metrics, in the bottleneck cell, compared to the average respective results over the other six cells. The deterioration increases for traffic types of lower priority. However, in all of the traffic scenarios simulated in which the average traffic load in each cell did not surpass 70% (14 Mb/s), the QoS requirements for all types of traffic were not violated in the bottleneck cell. B. Conceptual Comparison With Relevant Work We also conceptually compare this paper with one practical implementation of telemedicine video traffic [33] and two efficient MAC protocols recently proposed in the literature: one protocol based on WCDMA [36] and one protocol that integrates TDMA, code-division multiple access (CDMA), and carrier sense multiple access (CSMA) [37]. In [33], the authors studied the possibility of using the highspeed packet access (HSPA) suite, which is an improvement over the Universal Mobile Telecommunications System (UMTS), to transmit telemedicine video traffic. Several videoconference sessions were initiated, and best case, worst case, and average scenarios were recorded as a moving ambulance was linked with a consultation using HSPA. In the case where the ambulance was at the city center, it was forced to use links that were already utilized by several other users (this case, as noted by the authors in [33], is the most realistic scenario of the cases examined), and the transmission rates achieved in the uplink were equal to 66% of the rates achieved in noncongested links but had a much higher fluctuation. This condition resulted in large delays, varying between 200 ms and 1300 ms, which, in turn, resulted to videoconference sessions with varying quality.

642

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 2, FEBRUARY 2011

In addition, the speed of the ambulance was shown to play an important role: The higher the speed (which, in critical cases, is absolutely necessary), the larger the delays due to multiple handoffs, which were difficult for the network to seamlessly execute. When the ambulance increased its speed to about 60– 100 km/h the delay significantly increased. In the case of a congested network, the summary of doctors’ feedback on the audio and video quality of the system was as low as 34% positive for the video quality and 56% positive for the sound quality. These results show the need for schemes that will ensure the immediate and high-quality transmission of this type of traffic and other telemedicine traffic types (such as the schemes proposed in this paper), regardless of the traffic load in the cellular network. In [36], six types of traffic [i.e., voice, audio, constant bit rate (CBR) video, variable bit rate (VBR) video, computer data messages, and e-mail messages] are integrated. The protocol’s throughput, as presented in [36] in packets per frame, is lower than the throughput achieved by our scheme in all our simulations for the strict QoS requirements considered in this paper; in addition, the protocol does not consider handoff traffic. In [37], the authors propose an adaptive MAC protocol to address the heterogeneities posed by next-generation wireless networks. The protocol is very efficient in using bandwidth from different network to satisfy, as best as possible, the bandwidth requirements of various types of applications; however, the TDMA network is shown in the performance evaluation to be unable to support even relatively small numbers of voice users and can only well serve best effort traffic. As soon as the traffic load increases, the available bandwidth from CDMA and wireless local area network (WLAN) networks is used. However, even with the use of this additional bandwidth, the protocol can only guarantee, for VBR video traffic, an average bit rate of 239 Kb/s (the minimum bit rate for VBR video is considered 120 Kb/s, and the maximum is 420 Kb/s). This condition is a much lower bandwidth requirement than the bandwidth requirements of regular VBR video traffic used and satisfied in this paper. In addition, although the average bit rate in [37] is satisfied, there is no mention in of the average video packet dropping achieved with A-MAC. In addition, no urgent traffic, e.g., telemedicine, is included in [36] and [37]. This condition would mean that the throughput would further drop in [36] and the incapability of the TDMA component to handle the additional VBR traffic would increase in [37] should such an urgent type of traffic be included in the traffic mixture. In our view, the basic idea in [37] (MAC protocols that adapt to multiple network structures with QoS-aware procedures) is the direction toward which MAC protocol designs should head in the near future. It is, however, of major importance that each individual access method (e.g., TDMA, CDMA, and WLAN protocol) is designed toward maximizing bandwidth utilization while guaranteeing the QoS of even the most urgent types of traffic. This condition was the goal of our design in this paper with regard to the TDMA component, and we believe that the only way of achieving this goal is to combine the scheduling scheme with an equally effective adaptive bandwidth reservation scheme such as the one proposed in Section IV. Because we view this scheduling mechanism as part of a larger system that will integrate different access

methods, we did not aim at exactly matching the specifications of current 3G networks; the way to incorporate both the scheduling and adaptive bandwidth reservation mechanisms into the standards of next-generation (4G) networks will be part of our future work. The choice of parameters that we have made from the relevant literature leads us to believe that the expected results from this future work will closely match, qualitatively, those of the simulations presented in this paper. VII. C ONCLUSION The importance of errorless and expedited telemedicine traffic transmission over wireless cellular networks quickly becomes a very important issue due to the current systems’ inability to provide high QoS to transmissions from mobile telemedicine terminals. Recent studies have solely focused on the transmission of telemedicine traffic over the cellular network, without taking into account the equally important fact that regular traffic also has strict QoS requirements. Therefore, a practical solution should focus on the efficient integration of the two traffic categories. In this paper, we have proposed, for the first time in the relevant literature to the best of our knowledge, the combination of a fair and efficient scheduling scheme and an adaptive bandwidth reservation scheme that enable the integration of highest priority telemedicine traffic transmission with regular wireless traffic over cellular networks. We have included, in our extensive simulation study, all major types of current telemedicine applications (i.e., ECG, X-ray, video, and medical still images), as well as the most popular “regular” applications (i.e., voice, video, e-mail, and web). With the use of distancebased information reports to achieve seamless handoff, as well as new scheduling ideas, our proposal, which is evaluated over a hexagonal cellular structure, provides full priority and satisfies the very strict QoS requirements of telemedicine traffic without violating the QoS of regular traffic, even in the case of high traffic loads. R EFERENCES [1] P. Koutsakis, S. Psychis, and M. Paterakis, “Integrated wireless access for videoconference from MPEG-4 and H.263 video coders with voice, email and web traffic,” IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1863–1874, Sep. 2005. [2] A. Bhargava, M. Farrukh Khan, and A. Ghafoor, “QoS management in multimedia networking for telemedicine applications,” in Proc. IEEE Workshop Softw. Technol. Future Embedded Syst., Hakodate, Japan, 2003, pp. 39–42. [3] P. Seeling, M. Reisslein, and B. Kulapala, “Network performance evaluation using frame sizes and quality traces of single-layer and two-layer video: A tutorial,” IEEE Commun. Surveys Tuts., vol. 6, no. 3, pp. 58–78, Third Quart., 2004. [4] S. Garawi, R. S. H. Istepanian, and M. A. Abu-Rgheff, “3G wireless communications for mobile robotic teleultrasonography systems,” IEEE Commun. Mag., vol. 44, no. 4, pp. 91–96, Apr. 2006. [5] E. A. V. Navarro, J. R. Mas, J. F. Navajas, and C. P. Alcega, “Performance of a 3G-based mobile telemedicine system,” in Proc. IEEE CCNC, Las Vegas, NV, 2006, pp. 1023–1027. [6] S. C. Voskarides, C. S. Pattichis, R. Istepanian, C. Michaelides, and C. N. Schizas, “Practical evaluation of GPRS use in a telemedicine system in Cyprus,” in Proc. 4th IEEE Int. EMBS Special Topic Conf. Inf. Technol. Appl. Biomed., Birmingham, U.K., 2003, pp. 39–42. [7] Y. Chu and A. Ganz, “A mobile teletrauma system using 3G networks,” IEEE Trans. Inf. Technol. Biomed., vol. 8, no. 4, pp. 456–462, Dec. 2004. [8] C. H. Salvador, M. P. Carrasco, M. A. G. de Mingo, A. M. Carrero, J. M. Montes, L. S. Martin, M. A. Cavero, I. F. Lozano, and

QIAO AND KOUTSAKIS: ADAPTIVE BANDWIDTH RESERVATION AND SCHEDULING FOR TELEMEDICINE TRAFFIC

[9]

[10]

[11] [12] [13]

[14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28]

[29] [30] [31] [32]

J. L. Monteagudo, “Airmed-cardio: A GSM- and Internet-services-based system for out-of-hospital follow-up of cardiac patients,” IEEE Trans. Inf. Technol. Biomed., vol. 9, no. 1, pp. 73–85, Mar. 2005. S. Pavlopoulos, E. Kyriacou, A. Berler, S. Dembeyiotis, and D. Koutsouris, “A novel emergency telemedicine system based on wireless communication technology—AMBULANCE,” IEEE Trans. Inf. Technol. Biomed., vol. 2, no. 4, pp. 261–267, Dec. 1998. J. R. Gallego, A. Hernandez-Solana, M. Canales, J. Lafuente, A. Valdovinos, and J. Fernandez-Navajas, “Performance analysis of multiplexed medical data transmission for mobile emergency care over the UMTS channel,” IEEE Trans. Inf. Technol. Biomed., vol. 9, no. 1, pp. 13–22, Mar. 2005. M. Bottigliengo, C. Casetti, C.-F. Chiasserini, and M. Meo, “Short-term fairness for TCP flows in 802.11b WLANs,” in Proc. IEEE INFOCOM, Hong Kong, 2004, pp. 1383–1392. E. C. Baig, Will Consumers Tune In to a Tiny TV in Their Hand? Jul. 2006. [Online]. Available: http://www.usatoday.com/tech/wireless/ 2006-08-17-mobile-tv_x.htm C.-Y. Huang and S.-G. Miaou, “Transmitting SPIHT compressed ECG data over a next-generation mobile telecardiology testbed,” in Proc. 23rd IEEE Int. Conf. Eng. Med. Biol. Soc., Istanbul, Turkey, 2001, pp. 3525–3528. N. M. Mitrou, G. L. Lyberopoulos, and A. D. Panagopoulou, “Voice and data integration in the air interface of a microcellular mobile communication system,” IEEE Trans. Veh. Technol., vol. 42, no. 1, pp. 1–13, Feb. 1993. A. C. Cleary and M. Paterakis, “An investigation of stack-based algorithms for voice packet transmission in microcellular wireless environments,” in Proc. IEEE ICC, Seattle, WA, 1995, pp. 1375–1381. C.-F. Tsai, C.-J. Tsang, F.-C. Ren, and C.-M. Yen, “Adaptive radio resource allocation for downlink OFDMA/SDMA systems,” in Proc. IEEE ICC, Glasgow, U.K., 2007, pp. 5683–5688. D.-S. Lee and Y.-H. Hsueh, “Bandwidth-reservation scheme based on road information for next-generation cellular networks,” IEEE Trans. Veh. Technol., vol. 53, no. 1, pp. 243–252, Jan. 2004. S. Choi and K. G. Shin, “Adaptive bandwidth reservation and admission control in QoS-sensitive cellular networks,” IEEE Trans. Parallel Distrib. Syst., vol. 13, no. 9, pp. 882–897, Sep. 2002. W.-S. Soh and H. S. Kim, “A predictive bandwidth reservation scheme using mobile positioning and road topology information,” IEEE/ACM Trans. Netw., vol. 14, no. 5, pp. 1078–1091, Oct. 2006. F. R. Yu, V. W. S. Wong, and V. C. M. Leung, “A new QoS provisioning method for adaptive multimedia in wireless networks,” IEEE Trans. Veh. Technol., vol. 57, no. 3, pp. 1899–1909, May 2008. [Online]. Available: http://www-tkn.ee.tu-berlin.de/research/trace/ trace.html [Online]. Available: http://www.nttdocomo.com/pr/2002/001039.html A. M. Law and W. D. Kelton, Simulation Modeling and Analysis, 2nd ed. New York: McGraw-Hill, 1991. R. Zander and J. M. Karlsson, “A rate-based bandwidth borrowing and reservation scheme for cellular networks,” in Proc. IEEE Veh. Technol. Conf.—Fall, Los Angeles, CA, 2004, pp. 1123–1128. L. Qiao and P. Koutsakis, “Guaranteed bandwidth allocation and QoS support for mobile telemedicine traffic,” in Proc. IEEE Sarnoff Symp., Princeton, NJ, 2008, pp. 1–5. L. Qiao and P. Koutsakis, “Fair and efficient scheduling for telemedicine traffic transmission over wireless cellular networks,” in Proc. IEEE Veh. Technol. Conf.—Spring, Barcelona, Spain, 2009, pp. 1–5. P. Koutsakis, M. Vafiadis, and A. Lazaris, “A new bandwidth allocation mechanism for next generation wireless cellular networks,” Wireless Netw., vol. 16, no. 2, pp. 331–353, Feb. 2010. G. Schreier, D. Hayn, P. Kastner, S. Koller, W. Salmhofer, and R. Hofmann-Wellenhof, “A mobile-phone-based teledermatology system to support self-management of patients suffering from psoriasis,” in Proc. 30th IEEE Int. EMBS Conf., Vancouver, BC, Canada, 2008, pp. 5338–5341. H. F. Rashvand, V. T. Salcedo, E. M. Sanchez, and D. Iliescu, “Ubiquitous wireless telemedicine,” IET Commun., vol. 2, no. 2, pp. 237–254, Feb. 2008. C. Hu, W. Liu, Y. Pan, Z. Liu, J. Liao, and M. Q.-H. Meng, “Wireless acquisition system for the real ambulance field,” in Proc. IEEE Int. Conf. ROBIO, Bangkok, Thailand, 2008, pp. 1081–1086. J. A. Zubairi and S. Misbahuddin, “Ambulatory data aggregation and reduction for transmission over limited bandwidth network,” in Proc. Int. Symp. CTS, Baltimore, MD, 2009, pp. 356–360. A. Zvikhachevskaya, G. Markarian, and L. Mihaylova, “Quality-ofservice consideration for the wireless telemedicine and e-health services,” in Proc. IEEE WCNC, Budapest, Hungary, 2009, pp. 1–6.

643

[33] K. Banitsas, S. Tachakra, E. Stefanidis, and K. Boletis, “Using HSPA to improve the telemedical links on a moving ambulance,” in Proc. 30th IEEE Int. EMBS Conf., Vancouver, BC, Canada, 2008, pp. 739–742. [34] R. S. H. Istepanian, N. Y. Philip, and M. G. Martini, “Medical QoS provision based on reinforcement learning in ultrasound streaming over 3.5G wireless systems,” IEEE J. Sel. Areas Commun., vol. 27, no. 4, pp. 566–574, May 2009. [35] M. G. Hadjinicolaou, R. Nilavalan, T. Itagaki, S. C. Voskarides, C. S. Pattichis, and A. N. Schizas, “Emergency TeleOrthoPaedics m-health system for wireless communication links,” IET Commun., vol. 3, no. 8, pp. 1284–1296, Aug. 2009. [36] X. Wang, “An FDD wideband CDMA MAC protocol with minimumpower allocation and GPS-scheduling for wireless wide area multimedia networks,” IEEE Trans. Mobile Comput., vol. 4, no. 1, pp. 16–28, Jan./Feb. 2005. [37] M. C. Vuran and I. F. Akyildiz, “A-MAC: Adaptive medium access control for next-generation wireless terminals,” IEEE/ACM Trans. Netw., vol. 15, no. 3, pp. 574–587, Jun. 2007. [38] D. A. Dyson and Z. J. Haas, “A dynamic packet reservation multiple access scheme for wireless ATM,” Mobile Netw. Appl., vol. 4, no. 2, pp. 87–99, May 1999. [39] S. Elnoubi and A. M. Alsayh, “A packet reservation multiple access (PRMA)-based algorithm for multimedia wireless system,” IEEE Trans. Veh. Technol., vol. 53, no. 1, pp. 215–222, Jan. 2004. [40] K. Xu, Y. Tian, and N. Ansari, “TCP-jersey for wireless IP communications,” IEEE J. Sel. Areas Commun., vol. 22, no. 4, pp. 747–756, May 2004. [41] Y. Tian, K. Xu, and N. Ansari, “TCP in wireless environments: Problems and solutions,” IEEE Commun. Mag., vol. 43, no. 3, pp. 27–32, Mar. 2005.

Lu Qiao received the B.Sc. and M.Sc. degrees in electronics and information engineering from Beijing Jiaotong University, Beijing, China, in 2006 and 2008, respectively, and the M.A.Sc. degree in electrical and computer engineering from McMaster University, Hamilton, ON, Canada, in 2008. Since 2008, she has been with the Bank of Montreal, Toronto, ON, as a Senior Application Software Developer for capital market trading products. Her research interests include multimedia integration over wireless networks.

Polychronis Koutsakis (M’09) was born in Hania, Greece, in 1974. He received the five-year Diploma in electrical engineering from the University of Patras, Patras, Greece, in 1997 and the M.Sc. and Ph.D. degrees in electronic and computer engineering from the Technical University of Crete, Chania, Greece, in 1999 and 2002, respectively. For three years (2003–2006), he was a Visiting Lecturer with the Department of Electronic and Computer Engineering, Technical University of Crete. From July 2006 to December 2008, he was an Assistant Professor (tenure-track) with the Department of Electrical and Computer Engineering, McMaster University, Hamilton, ON, Canada, with his research being funded by the National Sciences and Engineering Research Council of Canada (NSERC). Since January 2009, he has been with the Department of Electronic and Computer Engineering, Technical University of Crete, as an Assistant Professor. His research interests are focused on the design, modeling, and performance evaluation of computer communication networks, particularly on the design and evaluation of multiple access schemes for multimedia integration over wireless networks; on-call admission control and traffic policing schemes for both wireless and wired networks; multiple access control protocols for mobile satellite networks, wireless sensor networks and powerline networks; and traffic modeling. He is the author of more than 85 peer-reviewed papers in the aforementioned areas, has served as a Guest Editor for an issue of the ACM Mobile Computing and Communications Review and as a Project Proposal Reviewer for NSERC and the Dutch Technology Foundation, and serves as an Editor for a number of journals. Dr. Koutsakis is a Voting Member of the IEEE Multimedia Communications Technical Committee. He was a Technical Program Committee (TPC) Cochair for the Fourth ACM International Workshop on Wireless Multimedia Networking and Performance Modeling in 2008 and annually serves as a TPC Member for most of the major conferences in his research field.