MAC Layer Resource Allocation for Wireless Body Area Networks

Qinghua Shen1,Xuemin(Sherman)Shen1,Tom H.Luan2,and Jing Liu3

(1.Department of Electrical and Computer Engineering，University of Waterloo，ON N2L 3G1，Canada; 2.School of Information Technology，Deakin University，Melbourne，Australia; 3.Department of Electronic Engineering，Shanghai Jiao Tong University，China)

Wireless body area networks(WBANs)can provide low⁃cost，timely healthcare services and are expected to be widely used for e⁃healthcare in hospitals.In a hospital，space is often limited and multiple WBANs have to coexist in an area and share the same channel in order to provide healthcare services to different patients.This causes severe interference between WBANs that could significantly reduce the network throughput and increase the amount of power consumed by sensors placed on the body.There⁃fore，an efficient channel⁃resource allocation scheme in the medium access control(MAC)layer is crucial.In this paper，we devel⁃op a centralized MAC layer resource allocation scheme for a WBAN.We focus on mitigating the interference between WBANs and reducing the power consumed by sensors.Channel and buffer state are reported by smartphones deployed in each WBAN，and channel access allocation is performed by a central controller to maximize network throughput.Sensors have strict limitations in terms of energy consumption and computing capability and cannot provide all the necessary information for channel allocation in a timely manner.This deteriorates network performance.We exploit the temporal correlation of the body area channel in order to minimize the number of channel state reports necessary.We view the network design as a partly observable optimization prob⁃lem and develop a myopic policy，which we then simulate in Matlab.

medium access control(MAC);wireless body area networks(WBANs);resource allocation;interference mitigation;partially observ⁃able optimization

1 Introduction

I t is widely recognized that current hospital⁃centric healthcare services do not efficiently meet the needs of an ageing population.A promising solution is e⁃health⁃care，which involves using ICT to support healthcare practices[1].The key part of an e⁃healthcare system is a wire⁃less body⁃area network(WBAN)comprising multiple sensors that monitor the physical condition of patients[2].A WBAN enables continuous，remote monitoring of patients，and this in⁃creases the efficiency of medical staff and reduces the cost of hospital healthcare[3].With an e⁃healthcare system，medical staff can limit the number of unnecessary visits to a patient. Data collected by the sensors enables medical staff to keep track of the condition of individual patients and react to emer⁃gencies in advance.For example，signs of cardiac arrest could be detected hours in advance，and a life⁃threatening situation averted[4].

However，there are still fundamental challenges to the wide⁃spread use of sensor⁃based WBAN in hospitals and the provi⁃sion of guaranteed communication services for critical medical traffic.If medical staff are to respond rapidly，patient vital signs such as heart rate，blood pressure，respiratory rate，tem⁃perature，pulse oximetry，and level of consciousness all need to be monitored in real time.This means that the data must be transmitted accurately and without unacceptable delay[5]，[6]. In addition，sensors on the body typically have limited power. This means that underlying communication protocols must be efficient，i.e.，transmission errors and retransmission must be kept to a minimum.Because sensors have limited computing capability and data buffer，real⁃time data that cannot be trans⁃mitted within a given period is dropped，and this leads to a high report⁃dropping ratio.An e⁃healthcare network demands more efficient communication because of the critical nature of medical traffic and the limited resources of sensors.Also，be⁃cause floor space is usually quite limited in a hospital，multi⁃ple WBANs for different groups of patients are often located inthe same area，and interference between WBANs can be se⁃vere.Therefore，MAC layer resource allocation has to be effi⁃cient and provide the high QoS needed for e⁃healthcare sys⁃tems.

In this paper，we exploit the temporal channel correlations around human bodies to develop a centralized MAC protocol for e⁃healthcare systems.Specifically，we consider a scenario where multiple WBANs exist in an area，such as a ward，and contend the channel for transmission.Each WBAN includes sensors on the body that continuously transmit collected data to the Internet(via a data sink)and to the smartphone of the pa⁃tient.Because of variations in the body⁃area channel，the trans⁃mission link between a sensor and smartphone may not be al⁃ways available.To maximize network throughput and reduce the packet⁃drop ratio，it might seem reasonable to permit the WBAN with good channel quality and cached data in sensors to transmit.However，this means that the real⁃time channel and buffer state information of all WBANs is required.Be⁃cause sensors on the body have limited computing and energy resources，this information cannot always be accurately mea⁃sured and provided by the sensors.This leads to inefficient use of channels.

To address this issue，we use the temporal channel correla⁃tions to guide MAC resource allocation.Specifically，we repre⁃sent the channel state with a belief state and use this metric to allocate access to the channel.The belief state，which is not chosen for transmission，is updated according to the statistical information.Only one sensor from a WBAN needs to report its current state so that channel⁃state reports are minimized.Giv⁃en the incomplete nature of network state information，we treat the throughput⁃maximization problem as a partly observable optimization problem.We first analyze the dynamics of the be⁃lief states and buffer states.Then，we create a myopic policy and investigate its drawback in terms of incurred packet drop⁃ping.Then，we propose a modified myopic policy in which the future impact of a current decision is approximated.Finally，we compare our proposed policy with Round Robin(RR)to demonstrate its effectiveness.

The remainder of this paper is organized as follows.In sec⁃tion 2，we discuss related works.In section 3，we discuss the system model to be studied.In section 4，we formulate the problem.In section 5，we discuss the policy design.In section 6，we give simulation results.In section 7，we conclude and discuss future research directions.

2 Related Works

In this section，we review works on resource allocation in the MAC layer of a WBAN as well as works that cover the prob⁃lem of partly observable optimal control.

Resource allocation in the WBAN MAC layer has long been an important research topic.In[7]，a fuzzy logic algorithm is used in a hospital environment to adjust the MAC layer control parameters according to real⁃time network information.To sup⁃port the transmission of medical traffic within coexisting WBANs，IEEE 802.15.6 has been proposed.The standard de⁃scribes collaborative and non⁃collaborative methods，such as beacon shifting and channel hopping，for eliminating interfer⁃ence between WBANs[8]，[9].However，it does not specify how to use these methods with different network settings.Inter⁃ference⁃mitigation schemes for WLANs，including the busy tone scheme[10]-[13]，are not suitable for WBANs because energy⁃consuming control signals quickly drain the power of sensors.To reduce the amount of energy consumed by sensors，the scheduling problem can be formulated using game theory，and heuristic cooperation can be used in the scheduling policy [14].In[15]，a network that can tolerate concurrent transmis⁃sion of multiple WBANs is described.The authors propose an uncomplicated scheduling scheme inspired by the random in⁃complete coloring scheme.However，in all of these mentioned works，variations in the body area channel are not considered.

In this paper，we describe the throughput maximization prob⁃lem as a partly observable optimization problem because of the incomplete nature of network state information.The partly ob⁃servable optimization problem has been studied extensively. Research on this topic started with scheduling over a single random process.In[16]and[17]，a random process governed by a Markov chain is considered.The authors show that the conditional probability distribution of the current state(given previous control decisions and observations)is sufficient statis⁃tical information for an optimal scheduling policy.Moreover，the convex property of a corresponding value function is proved[16].This is the key to obtaining an optimal policy.In [18]，the authors studied optimal policies for scheduling over multiple random processes.

3 System Model

In this section，we describe the system model in terms of net⁃work，channel，and traffic and then describe the channel⁃ac⁃cess scheme.

3.1 Network Model

We consider a network comprising NwWBANs and one cen⁃tral controller(Fig.1).The central controller allocates channel resources to WBANs and forwards the medical information through a wired network.Each WBAN corresponds to a single patient.Within each WBAN，there is a smart phone and one sensor.In this paper，only one sensor is considered because a single sensor can monitor most vital signals nowadays[19]. The smart phone collects information from the sensor in the same WBAN and transmits this information to the central con⁃troller.A sensor on the body has limited power and computing capacity whereas a smartphone has greater power supply and computing capacity.Sensors are set to turn a radio on at a pre⁃defined time but are in sleep mode most of time.In contrast，the smartphone is always turned on.The timing of the system is partitioned into slots，and the duration of each of these slots is denoted T.

▲Figure 1.System model.

3.2 Channel Model

Medical traffic is transmitted in two hops from the sensor to the central controller(Fig.1).As in IEEE 802.15.6[20]，[21]，we denote the first⁃hop body⁃surface⁃to⁃body⁃surface channel CM3 and the second⁃hop body⁃surface⁃to⁃external channel CM4.The CM4 channel is modeled as free space wireless channel，and the transmissions of the second hop are error⁃free because of the ample transmission power and clear channel conditions.CM3 has severe and varying path loss due to the ab⁃sorption of human body.There are also strong temporal corre⁃lations of channel between neighboring time slots.

To obtain the features of CM3 without any loss of generality，we use the Gilbert Elliot(GE)model(Fig.2)[22]，[23].In this model there are two channel states:on(error⁃free transmission) and off(unsuccessful transmission).

Let Ci(n)denote CM3 channel state for the ith WBAN over the nth time slot.If the channel is on，Ci(n)=1;otherwise Ci(n)=0.Let RicandΠcidenote the probability transition ma⁃trix and stationary distribution of CM3 for the ith WBAN，re⁃ spectively.According to the GE model，Riccan be given as

▲Figure 2.On/off wireless channel model(Gilbert Elliot model).

where giis the conditional probability of the channel changing from off to on，andgi≜Pr{Ci(n)=1|Ci(n-1)=0}for n∈{1，2，...}.The conditional probability of the channel changing from on to off is given by bi,andbi≜Pr{Ci(n)=0|Ci(n-1)=1}for n∈{1，2，...}.The corresponding stationary distribution is giv⁃en byΠci=[bi/(bi+gi)，gi/(bi+gi)].In the GE model，if a cha⁃nnel tends to stay in its current state，the channel is positively correlated，i.e.，1>bi+gi.

3.3 Traffic Model

The medical data is collected and summarized as a report by the sensor and transmitted to the smartphone at the end of each time slot.A report contains health information that is re⁃quired for rapid response[4].We denote the number of pack⁃ets in a report Np.The transmission of reports from sensor to smartphone follows the sum of the Bernoulli process，and the arrival rate for the ith WBAN is given by λi.Because condition of a patient changes much slower than the channel variations，we only consider a scenario where λi<1.The computing capa⁃bility of sensors is limited;therefore，we assume that the sen⁃sor buffer can only store a limited number of packets.With a loss of generality，we consider that a sensor can only cache one report within each time slot.If a new report arrives but the pre⁃vious report is still cached in the buffer，the previous report is evicted from the buffer and replaced by the new report.The buffer state of the ith WBAN at the beginning of time slot n is given by qi(n)∈{0，1}，where qi(n)=0 means that the sensor buf⁃fer of the ith WBAN is empty at the beginning of timeslot n; otherwise，qi(n)=1.

3.4 Channel-Access Scheme

The goal of MAC is to maximize network throughput without allowing the sensors to consume too much energy.The channel⁃access scheme is described as follows.At the beginning of each time slot，the central controller sends out a beacon to choose a WBAN for transmission during this time slot.Be⁃cause only one WBAN is scheduled，interference between WBANs is avoided.Let s(n)denote the index of the WBAN chosen during time slot n，and s(n)=i.The smartphone of the ith WBAN sends a beacon to the sensor in the same WBAN.If CM3 is on and there is one report to be transmitted，i.e.，Ci(n)=1 and qi(n)=1，the transmission from the sensor to smart⁃phone will be successful and the sensor’s buffer will be emp⁃tied.Then the smartphone forwards the report to the central controller;otherwise，only the channel state is reported to the central controller for future scheduling.

There are two pieces of information about network stateavailable to the central controller.One piece of information is the statistics about the random processes of CM3 channels and the medical report event arrival.In practice，the central con⁃troller can obtain these statistics by learning over a period of time.Specifically，the smartphone in each WBAN can learn the channel statistics of that WBAN first and forward them to the central controller.Adaptive learning algorithms[24]，[3] could be used to increase learning accuracy in real time.We assume that the central controller can obtain accurate informa⁃tion about the body area channels.In the future，we will consid⁃er the impact of imperfect and delayed information on MAC de⁃sign.The other piece of information is the partial real⁃time in⁃formation about the network state.As described in the channel⁃access scheme，the central controller has the information about the WBAN it chooses at the end of each time slot.Thus，real⁃time information about the network state is partly available.

To make a proper decision at each time slot，the central con⁃troller maintains the belief states of the channel and buffer states of all WBANs based on both statistical information and partial real⁃time information.Let Ω(n)≜[ω1(n)，…，ωNw(n)]de⁃note the belief states of the channel states of all WBANs at the beginning of time slot n，where ωi(n)is the belief state of the channel state of the ith WBAN over time slot n.The belief states evolve as follows.If real⁃time information of the ith WBAN is available，the central controller updates its belief state of the ith WBAN according to real information;otherwise，the central controller updates this belief according to statisti⁃cal information.The belief state evolution can be written as

whereTic(γ)is an evolution operator of the belief channel state of the ith WBAN.For the ON⁃OFF channel model，the op⁃erator is

As described in[16]，the above belief state is a sufficient sta⁃tistic that depicts current channel state given the channel state is a Markov process.

4 Problem Formulation

In this section，we formulate the problem as a partly observ⁃able optimization problem.Then，we investigate the value func⁃tion of the proposed problem for policy design.

4.1 Reward and Objectives

We first design a reward to facilitate the decision⁃making of the central controller.The reward should favor higher through⁃put and not favor packet drop.If the ith WBAN is chosen and transmission is successful，Bi(n)units of reward are received by the network.Let Ri(n)denote the reward obtained in time slot n when the ith WBAN is chosen.Ri(n)is given by

If the channel of the chosen WBAN is off or the buffer of the chosen WBAN is empty，the reward is zero;otherwise，Bi(n) amount of reward accumulated.Let Bi(n)equal the probability of one medical report arriving since the last successful trans⁃mission.If a WBAN is not given channel access for a long time，the reward that can be received by the WBAN is small because many packets may have been lost.

The issue for the central controller is which WBAN should be given channel access at each time slot.A control policy for this problem is π:Ω(n)→s(n)，a function that maps the belief state Ω(n)to the action s(n).The goal of the central controller is to maximize the average reward of the network over infinite horizon，which is a common measure in communication system [18].Thus，the control problem can be written as

Let π*denote the optimal solution to P1，then

Because the real⁃time state of the full network is not observ⁃able，P1 is a partly observable optimization problem.If only the channel state is considered，P1 becomes a partly observ⁃able Markov decision process problem(POMDP)[16].A POM⁃DP has larger state space compared with an observable optimi⁃zation problem and is more difficult to solve.Our problem is more difficult than a POMDP because we consider random traf⁃fic arrival.P1 is a dynamic programming problem;thus，we study the value function of P1 for policy design.

4.2 Value Function

Value function analysis involves breaking an optimization problem over multiple periods into sub⁃problems at different points in time.The value function at time slot n is given by Vn(Ω(n)).This is the maximum expected reward that the net⁃work can have at time slot n.We consider a case where the central controller chooses the ith WBAN at the beginning of time slot n and updates the network state information at the end of time slot n.The reward that can be obtained from time slot n comprises the expected immediate rewardE[Ri(n)]and the maximum expected reward from timeslot n+1，namely Vn+1(Ω(n+1)|s(n)=i，Ci(n)).Thus，the value function of P1 at time slot n can be written as

For a POMDP，the value function is piecewise linear andconvex[17].However，for a general partly observable problem，the value function may not have this property.Generally，(7) can be solved backwards to obtain the value ofV1() Ω(1)and the optimal policyπ*.However，because computation com⁃plexity increases exponentially，the value function and optimal policy cannot be obtained in real time by the central controller.

5 Policy Design

Because obtaining the optimal solution to P1 is difficult，we first investigate the properties of the channel and buffer dynam⁃ics and from this investigation we propose a policy.

5.1 System Dynamics

The belief state of the ith WBAN is ωiat any time slot，and we study what the belief state will be after k consecutive time slots during which the ith WBAN is not chosen by the central controller.Let(Tic(ω(n)))k≜Pr{C(n+k)=1|ω(n)}(k=0，1，2，...，n)denote the belief state evolution for k consecutive unob⁃served time slots.Then we have[18]

Fig.3 shows howTci(ω(n))kchanges over time given the positive correlation of CM3，i.e.，1-bi-gi>0.The belief state eventually converges toΠci(2)，which is the stationary probability that the channel is on.This suggests that a policy should work in the following way.If the ith WBAN is chosen and the channel of the ith WBAN is on，the central controller should give preference to this WBAN again in the near future in order to utilize the ON state.In contrast，if the ith WBAN is chosen and the channel of the ith WBAN is off，the central con⁃troller should not give preference to this WBAN in near future in order to avoid wasting channel resource.

▲Figure 3.Evolution of belief state of channel.

Second，given the initial state qi(n)=0，we study the proba⁃bility that exactly one report arrives during the duration k con⁃secutive time slots without transmission.This is the reward we set for a successful transmission.Let N(kT)denote the number of arrived reports during a period kT.For a sum of Bernoulli process，the probability of m events arriving during a period kT is e-λkT(λkT)mm!Thus，given the initial state qi(n)=0，the probability that exactly one report arrives during a duration of k time slots is:

From(9)，the probability has a maximum value at 1/λ.Be⁃cause the system is slotted，the corresponding number of time slot k can be either the minimum integer larger than 1/(λT)or the maximum integer smaller than 1/(λT).

Fig.4 shows howPr{} N(kT)=1changes over time.The probability of exactly one report arrival increases to its peak as time goes and then decreases.This result suggests that a con⁃trol policy should have the following property.If the central controller determines that the buffer of the ith WBAN is emp⁃ty，it needs to wait for a period of time to revisit the ith WBAN for a new report arrival.However，if the duration is larger than 1/λ，the probability of report loss increases，and this leads to a smaller reward.Thus，the central controller should not wait too long for a revisit.

5.2 A Modified Myopic Policy

We construct a myopic policy and，through analysis，point out that this policy results in high report dropping.Drawing on our previous analysis of system dynamics，we propose a modi⁃fied myopic policy that addresses this issue by approximating the expected future reward.

If only the dynamics of the channel is considered，as in[18]，the optimal policy is to stick to the WBAN that is on.Specifi⁃cally，if a WBAN is found to be on，the central controller should keep choosing this WBAN until the channel turns off. With the random report arrivals in WBANs，the above policy is no longer optimal.After a successful transmission，the proba⁃bility of a report arrival is low(Fig.4).Thus，even though the channel in the previous slot is on，the central controller does not give preference to this WBAN.

Obtaining an optimal policy is complex;therefore，we devel⁃op a myopic policy.The central controller's objective is simpli⁃fied to maximize the expected reward for a current time slot based on the belief states and ignores the impact of the current decision on the future reward.Let πmdenote the myopic poli⁃cy.It can be written as

The myopic policy πmhas two issues.First，the belief states of the channel converge in a homogeneous network setting (where all WBANs have the same statistics).Therefore，the central controller needs a scheme to choose from multiple WBANs with the same expected rewards.This is not addressed in the myopic policy.Second，with a myopic policy，if a WBAN has not be chosen for more than 1/(λT)consecutive time slots，the chance that the central controller will choose this WBAN decreases because the expected reward is smaller.This will cause reports to be dropped in that WBAN.This problem is rooted in the myopic philosophy.An optimal control policy that takes into account future reward does not have such an is⁃sue because if a WBAN is not chosen for more than 1/(λT)con⁃secutive time slots，the central controller tends to choose this WBAN.Otherwise，the expected future reward is smaller，and this leads to a smaller total reward.In other words，after 1/(λT)，the myopic policy significantly deviates from the opti⁃mal policy.

We propose a modified myopic policy to address the above issues.First，when multiple WBANs have the same maximum expected reward，the central controller chooses a WBAN using a random picker.This random picker makes a choice accord⁃ing to a random number that it generates from a probability density function.In this work，we use a uniform distribution. Second，the impact of future rewards is considered.Because the complexity of obtaining the accurate future reward is high，the future reward is approximated heuristically.Specifically，we increase the expected reward of the current time slot for the WBANs that have been waiting more than 1/(λT)time slots:

whereτis the number of slot that the ith WBAN has been wait⁃ing more than 1/(λT)，and w is scaling factor.Over time，(wτλT+1)/λTincreases.Thus，the WBANs that have been wai⁃ting more than 1/(λT)time slots have an increased chance of being chosen.This helps solve the second issue introduced by the myopic policy.

6 Simulation Results

In this section，we evaluate the performance of the proposed MAC layer resource allocation through Matlab simulations.

6.1 Simulation Setup

We simulate a scenario similar to that shown in Fig.1.A central controller is placed at the center of the network，and there are total Nwpatients，each with a WBAN for health moni⁃toring.The CM3 channels are simulated using the GE model，and the CM4 channels are simulated as being error⁃free for packet transmission.The initial channel states of the patients are generated randomly according to the stationary distribution of the channel states.In practice，the arrival rate of vital signs，such as blood pressure and heart rate，is usually less than 100 Kbps whereas the arrival rate of ECG and EMG signals is near⁃er 1 Mbps[5].Thus，the network is either congested or uncon⁃gested.This leads us to evaluate the effectiveness of the pro⁃posed resource allocation policy when the network is congest⁃ed and uncongested.Given the normalized service capacity，the congested network can be given by

and the uncongested network can be given by

In the simulation，we vary the report arrival rate to change the network condition.Letλcandλudenote the report arrival rate for a congested network and uncongested network，respec⁃tively.Let Nwcand Nwudenote the number of patients in the con⁃gested network and uncongested network，respectively.

For simplicity，we consider a homogeneous network where the report arrival rate and the channel statistics of all WBANs are the same.We omit the index of parameters，and the set⁃tings are shown in Table 1.

In each experiment，we compare our proposal with the round⁃robin(RR)scheme[25]and myopic(Myo)policy[26].The RR scheme is chosen because it is simple and starvation⁃free. In the RR scheme，the central controller assigns channel ac⁃cess opportunity to WBANs in a circular way.The WBAN cho⁃sen in time slot n is given by

▼Table 1.System parameters for simulation

where mod is the modulo operator.Let kN mod N=N，for k∈{0，1，2，..，n}because the network index starts from 1 in this work.The myopic algorithm is chosen because it is optimal when only the channel dynamics are considered[26].The myo⁃pic algorithm is to choose the WBAN with the best belief state of channel.

From here，we call our proposal MyoMo.In each simulation，we report the number of successful transmissions，number of reports dropped，and number of wasted transmission opportuni⁃ties using our proposal and existing proposals.A transmission opportunity is wasted if a WBAN is chosen but transmission is unsuccessful，either because the channel is OFF or the buffer is empty.The improvements of MyoMo and RR on Myo are ex⁃pressed as percentage ratio of performance difference to the performance of Myo.Each simulation was 1000 s in duration.

6.2 Performance Evaluation

Here，we report simulation results in terms of 1)network throughput(measured by the number of successful transmis⁃sions，2)the number of reports dropped，and 3)channel utiliza⁃tion(measured by the number of wasted transmission opportu⁃nities).

6.2.1 Uncongested Scenario

Fig.5 shows the perfor⁃mances of RR，Myo and MyoMo algorithms in an un⁃congested network.Fig.5a，Myoperformstheworst，with about 450 less suc⁃cessful transmissions than RR and MyoMo.MyoMo performsslightlybetter than RR，with about 50 more successful transmis⁃sionsonaverage.Myo causesabout400more dropped reports than RR andMyoMo，andmost WBANs drop fewer reports under MyoMo than under RR(Fig.5b).The lesser performance of Myo com⁃pared to RR and MyoMo can also be seen in Fig.5c. The superiority of RR and MyoMo over Myo in an uncongested network is shown in Fig. 5f.Compared to Myo，MyoMo and RR have 82%and 77% more successful transmissions，respectively;approximately 15%fewer dropped reports;and approximately 10%fewer wasted transmission opportunities.The reason that Myo per⁃forms the worst is that Myo only takes into consideration the channel state and always chooses the WBAN with the best channel.However，the chosen WBAN may have an empty buf⁃fer，and this leads to a high number of wasted transmission op⁃portunities.With low channel utilization，the number of reports dropped is high，and the number of successful transmissions is low.The reason that RR performs almost as well as MyoMo is because in an uncongested network，a wasted transmission op⁃portunity is more likely caused by an empty buffer than a chan⁃nel that is off.The number of WBANs in the network is denot⁃ed Nwu.In the RR algorithm，each WBAN needs to wait for Nwutime slots for a transmission opportunity.When Nwuis suffi⁃ciently large，the probability of an empty buffer is small.In oth⁃er words，the RR algorithm aims to avoid an empty buffer.It re⁃duces the number of wasted transmission opportunities and im⁃proves the number of successful transmissions.In an uncon⁃gested network，the MyoMo algorithm and related reward(11) is dominant because the buffer is taken into account.MyoMo aims to avoid an empty buffer，and this is similar to RR.As a result，the performance of RR is similar to MyoMo when the network is uncongested.

▲Figure 5.Performance for unsaturated scenario.

Fig.5d and Fig.5e show the difference between WBANs in terms of the number of channels that are on and the number of report arrivals，respectively.In Fig.5d，the 0 on the y⁃axis is the average number of channels that are on.The difference inthe number of channels that are on in different WBANs can be more than 1000.The reason for this is twofold:1)the initial channel states are generated randomly according to the station⁃ary distribution of the channel states，and 2)the channel states evolve according to a probability transition matrix(1).The number of report arrivals is different for different WBANs(Fig. 5e).These differences are the result of the random Poisson number generation method we used in Matlab.The difference between the number of channels that are on and the number of report arrivals causes performance to vary between WBANs us⁃ing the same algorithm.

6.2.2 Congested Scenario

Fig.6 shows the performance of the RR，Myo and MyoMo algorithms in a congested network.MyoMo performs the best，with about 300 more successful transmissions than either RR nels that are off increases.Unlike RR，MyoMo exploits the tem⁃poral channel correlation and uses the belief state of the chan⁃nel to make a decision.Thus，MyoMo is less likely to choose a WBAN with a channel that is off，and this leads to better per⁃formance.

From Fig.6 and Fig.5，the number of successful transmis⁃sions in a congested network is greater than in an uncongested network whereas the number of wasted transmission opportuni⁃ties in a congested network is smaller than that in an uncon⁃gested network.In a congested network，wasted transmission opportunities due to empty buffer are greatly reduced，and this leads to a higher number of successful transmissions.

7 Conclusion and Future Work

◀Figure 6. Performance in a congested network.

We have proposed a MAC layer resource⁃allocation scheme or Myo，which performs the worst.However，the difference in performance between Myo and RR is smaller.These trends can also be seen in report dropping Fig.6b and channel utiliza⁃tion Fig.6c.The improvements on Myo brought about by RR and MyoMo are shown in Fig.6f.Compared to Myo，MyoMo completes 20%more successful transmissions，has 2%fewer report drops，and has 6%fewer wasted transmission opportuni⁃ties.Compared to Myo，RR completes 8%more successful transmissions，has 1%fewer dropped reports，and has 3%few⁃er wasted transmission opportunities.MyoMo outperforms RR in congested networks because the number of wasted transmis⁃sion opportunities caused by empty buffers reduces，and the number of wasted transmission opportunities caused by chan⁃ for a WBAN.Through theory and simulation，we have demon⁃strated the effectiveness of our proposal in terms of increasing network throughput and channel utilization in both congested and uncongested networks.In future work，we will consider heterogeneous medical data traffic with differential services provisioned using MAC layer resource allocation.

[1]Q.Shen，X.Liang，X.Shen，X.Lin，and H.Y.Luo，“Exploiting geo⁃distributed clouds for a e⁃health monitoring system with minimum service delay and privacy preservation，”IEEE Journal of Biomedical and Health Informatics，vol.18，no. 2，pp.430-439，2014.doi:10.1109/JBHI.2013.2292829.

[2]Q.Shen and W.Zhuang，“Energy efficient scheduling for delay constrained com⁃munication in wireless body area networks，”in Proc IEEE GLOBECOM 2012，Anaheim，USA，pp.262-267.

[3]X.Liang，X.Li，Q.Shen，R.Lu，X.Lin，X.Shen，and W.Zhuang，“Exploiting prediction to enable secure and reliable routing in wireless body area networks，”in Proc.IEEE INFOCOM 2012，Orlando，USA，pp.388-396.doi:10.1109/IN⁃FCOM.2012.6195777.

[4]J.Welch，J.Moon，and S.McCombie，“Early detection of the deteriorating pa⁃tient:the case for a multi⁃parameter patient⁃worn monitor，”Biomedical Instru⁃mentation and Technology，vol.46，no.2，pp.57-64，2012.

[5]H.Cao，V.Leung，C.Chow，and H.Chan，“Enabling technologies for wireless body area networks:a survey and outlook，”IEEE Communication Magazine，vol. 47，no.12，pp.84-93，Dec.2009.doi:10.1109/MCOM.2009.5350373.

[6]H.Su and X.Zhang，“Battery⁃dynamics driven TDMA MAC protocols for wire⁃less body⁃area monitoring networks in healthcare applications，”IEEE Journal on Selected Areas in Communications，vol.27，no.4，pp.424-434，May 2009. doi:10.1109/JSAC.2009.090507.

[7]B.Otal，L.Alonso，and C.Verikoukis，“Highly reliable energy⁃saving MAC for wireless body sensor networks in healthcare systems，”IEEE Journal on Selected Areas in Communications，vol.27，no.4，pp.553-565，May 2009.doi:10.1109/ JSAC.2009.090516.

[8]IEEE Standard for Local and Metropolitan Area Networks Part 15.6:Wireless Body Area Networks，IEEE Standard 802.15.6，2012.

[9]K.S.Kwak，S.Ullah，and N.Ullah，“An overview of IEEE 802.15.6 standard，”in Proc.3rd International Symposium on Applied Sciences in Biomedical and Communication Technologies，Rome，Italy，2010，pp.1-6.doi:10.1109/ISA⁃BEL.2010.5702867.

[10]P.Wang and W.Zhuang，“A token⁃based scheduling scheme for WLANS sup⁃porting voice/data traffic and its performance analysis，”IEEE Transactions on Wireless Communications，vol.7，no.5，pp.1708-1718，May 2008.doi: 10.1109/TWC.2007.060889.

[11]P.Wang，H.Jiang，and W.Zhuang，“A new MAC scheme supporting voice/data traffic in wireless ad hoc networks，”IEEE Transactions on Mobile Computing，vol.7，no.12，pp.1491-1503，Dec.2008.doi:10.1109/TMC.2008.73.

[12]H.Jiang，P.Wang，and W.Zhuang，“A distributed channel access scheme with guaranteed priority and enhanced fairness，”IEEE Transactions on Wireless Communications，vol.6，no.6，pp.2114-2125，Jun.2007.doi:10.1109/ TWC.2007.05746.

[13]Information Technology-Telecommunications and Information Eχchange Be⁃tween Systems Local and Metropolitan Area Networks-Specific Requirements Part 11:Wireless LAN Medium Access Control and Physical Layer Specifica⁃tions，IEEE Standard 802.11，2012.

[14]L.Wang，C.Goursaud，N.Nikaein，L.Cottatellucci，and J.Gorce，“Cooperative scheduling for coexisting body area networks，”IEEE Transactions on Wireless Communications，vol.12，no.1，pp.123-133，Jan.2013.doi:10.1109/ TWC.2012.120412.112073.

[15]S.Cheng and C.Huang，“Coloring⁃based inter⁃WBAN scheduling for mobile wireless body area networks，”IEEE Transactions on Parallel and Distributed Systems，vol.24，no.2，pp.250-259，Feb.2013.doi:10.1109/TPDS.2012.133. [16]R.D.Smallwood and E.J.Sondik，“The optimal control of partially observable Markov processes over a finite horizon，”INFORMS Operations Research，vol. 21，no.5，pp.1071-1088，1973.

[17]R.D.Smallwood and E.J.Sondik，“The optimal control of partially observable Markov processes over the infinite horizon:discounted costs，”INFORMS Oper⁃ations Research，vol.26，no.2，pp.282-304，1978.

[18]K.Liu and Q.Zhao，“Indexability of restless bandit problems and optimality of whittle index for dynamic multichannel access，”IEEE Transactions on Infor⁃mation Theory，vol.56，no.11，pp.5547-5567，Nov.2010.doi:10.1109/ TIT.2010.2068950.

[19]H.Luo，“Wearable mini⁃size intelligent healthcare system，”US Patent App.11/ 675，684，2007.

[20]K.Y.Yazdandoost and K.Sayrafian⁃Pour，“Channel model for body area net⁃work(ban)，”IEEE P802.15⁃08⁃0780⁃09⁃0006，Apr.2009.

[21]E.Reusens，W.Joseph，B.Latré，B.Braem，G.Vermeeren，E.Tanghe，L.Mar⁃tens，I.Moerman，and C.Blondia，“Characterization of on⁃body communication channel and energy efficient topology design for wireless body area networks，”IEEE Transactions on Information Technology in Biomedicine，vol.13，no.6，pp.933-945，Nov.2009.doi:10.1109/TITB.2009.2033054.

[22]S.Rashwand and J.Mišic＇，“Channel and error modeling for wireless body area networks，”Springer Mobile Networks and Applications，vol.19，no.3，pp.276-286，Jun.2013.doi:10.1007/s11036⁃013⁃0449⁃8.

[23]P.Sadeghi，R.Kennedy，P.Rapajic，and R.Shams，“Finite⁃state Markov model⁃ ing of fading channels—a survey of principles and applications，“IEEE Sig⁃nal Processing Magazine，vol.25，no.5，pp.57-80，Sept.2008.doi:10.1109/ MSP.2008.926683.

[24]J.Gauvain and C.Lee，“Maximum a posteriori estimation for multivariate Gaussian mixture observations of Markov chains，”IEEE Transactions on Speech and Audio Processing，vol.2，no.2，pp.291-298，Apr.1994.doi: 10.1109/89.279278.

[25]X.Shen，H.Yu，J.Buford，and M.Akon，Handbook of Peer⁃to⁃Peer Network⁃ing.New York:Springer，2010.

[26]Q.Zhao，B.Krishnamachari，and K.Liu，“On myopic sensing for multi⁃channel opportunistic access:structure，optimality，and performance，”IEEE Transac⁃tions on Wireless Communications，vol.7，no.12，pp.5431-5440，Dec.2008. doi:10.1109/T⁃WC.2008.071349.

Biographiesphies

Qinghua Shen(q2shen@uwaterloo.ca)received his BS degree and MS degree in electrical engineering from Harbin Institute of Technology，China，in 2008 and 2010.He is currently working toward his PhD degree in the Department of Electri⁃cal and Computer Engineering，University of Waterloo，Canada.His research inter⁃ests include resource allocation for e⁃healthcare system and cloud computing.

Xuemin(Sherman)Shen(sshen@uwaterloo.ca)received his BSc(1982)degree from Dalian Maritime University，China，in 1982.He received his MSc degree(1987) and PhD degree(1990)in electrical engineering from Rutgers University，USA，in 1987 and 1990.He is professor and university research chair in the Department of Electrical and Computer Engineering，University of Waterloo，Canada.Dr.Shen’s research focuses on resource management in interconnected wireless/wired net⁃works，wireless network security，social networks，smart grid，and vehicular ad hoc and sensor networks.He has chaired or co⁃chaired many committees of internation⁃al IEEE conferences and is the editor⁃in⁃chief of IEEE Network,Peer⁃to⁃Peer Net⁃working and Applications,and IET Communications.He has also been the founding editor，associate editor，or guest editor for many other peer⁃reviewed publications. Dr.Shen is a registered professional engineer in Ontario，Canada;fellow of the IEEE;fellow of the Engineering Institute of Canada;fellow of the Canadian Acade⁃my of Engineering;and distinguished lecturer at the IEEE Vehicular Technology So⁃ciety and IEEE Communications Society.

Tom H.Luan(tom.luan@deakin.edu.au)received his BE degree from Xi'an Jiao⁃tong University，China，in 2004.He received his MPhil degree from Hong Kong Uni⁃versity of Science and Technology，in 2007.He received his PhD degree from the University of Waterloo，Canada，in 2012.He currently lectures on mobile and appli⁃cations in the School of Information Technology，Deakin University，Australia.His research interests include mobile cloud computing，mobile application and service development，vehicular networking，and wireless content distribution.

Jing Liu(Jingliu_lj@stju.edu.cn)received her BS，MS and PhD degrees from Xidi⁃an University，China，in 1998，2001 and 2005.Since July 2005，she has worked in the Department of Electronic Engineering，Shanghai Jiao Tong University，China. She is currently an associate professor at Shaighai Jiao Tong University.From 2012 to 2013，she was a visiting professor in the Department of Electrical and Computer Engineering，University of Waterloo，Canada.Her current research interests include wireless body area networking，wireless sensor networking，cooperative communica⁃tion，cognitive communication，and LTE networking.

10.3969/j.issn.1673-5188.2014.03.002

t

2014⁃03⁃31

This research has been supported by a research grant from the Natural Science and Engineering Research Council(NSERC)under grant No. CRDPJ 419147⁃11,and Care In Motion Inc.,Canada.

http://www.cnki.net/kcms/detail/34.1294.TN.20140821.0900.001.html,published online 21 August,2014