An Adaptive Energy Efficient MAC Protocol for the Medical Body Area Network

An Adaptive Energy Efficient MAC Protocol for the Medical Body Area Network N. F. Timmons* and W. G. Scanlon** *WiSAR Lab, Letterkenny Institute of Technology Letterkenny, Donegal Email: nick.timmons@lyit.ie **School of Electronics, Electrical Engineering and Computer Science Queen’s University, Belfast Email: w.scanlon@qub.ac.uk Abstract—Medical body area networks will employ both implantable and bodyworn devices to support a diverse range of applications with throughputs ranging from several bits per hour up to 10 Mbps. The challenge is to accommodate this range of applications within a single wireless network based on a suitably flexible and power efficient medium access control protocol. To this end, we present a Medical Medium Access Control (MedMAC) protocol for energy efficient and adaptable channel access in body area networks. The MedMAC incorporates a novel synchronisation mechanism and initial power efficiency simulations show that the MedMAC protocol outperforms the IEEE 802.15.4 protocol for two classes of medical applications. Keywords—BAN, wireless, sensor, network, protocol, MAC. I. INTRODUCTION The rapid expansion of wireless technology has inevitably led to the possibility of widespread un-tethered medical and health monitoring. Health monitoring systems which use cable as a medium can now be replaced with wireless connections. Point to point wireless links such as with the Medical Implant Communication Service (MICS) [1], MedRadio, and single sensor biotelemetry have been deployed over the last few years. However, thinking has moved on to the benefits of a body-centric communication networks. In the medical domain there are potentially a multitude of ultra low power wireless sensor networks (WSN) and body area network (BAN) applications with data rates ranging from 0.01 bps to 10’s Mbps. While medical BAN applications could support both in- patient and out-patient care, it is convenient to distinguish between implantable and wearable sensing activities (Table 1). TABLE I. SENSING ACTIVITIES IN MEDICAL BANS. Wearable BAN devices Implanted BAN devices EEG ECG SpO2 pulse oximeter, Glucose Fall detection Emergency call Performance assessment Glucose sensor; Cardiac arrhythmia: pacemaker, cardiovertor, defibrillator Intracranial pressure sensing Wireless capsule drug delivery Deep brain stimulation: retinal sensors, Parkinson’s, epilepsy Insulin pump A. Medical BAN Technical Requirements The main challenge in medical BANs is to balance the demands of the hard energy constraint associated with low power wireless sensor devices, with the quality of service (QoS) demands of the wide range of sensing and control applications. For example, a battery powered implanted medical device ideally must have a lifetime of up to 10 years. Unlike other wireless networks, it is generally impractical to charge or replace exhausted batteries, and therefore battery lifetime defines node lifetime. Since the transceiver communication operations consume much more energy than the processing operations, it is a primary objective to minimise transmit and receive operations to maximise node lifetime. Therefore, the medium access control (MAC) protocol in a BAN must be highly energy efficient. The main energy saving features that must be exhibited by a well designed MAC protocol are: collision avoidance, overhearing, control packet overhead, receiver idle listening, and transmitter over- emitting. Important attributes such as latency, throughput, and bandwidth utilisation, may be secondary in priority in generalised WSNs. However, in medical BANs life critical applications will place a priority on latency, security and guaranteed throughput. Unlike a WSN, a BAN will have a limited number of nodes, typically 10–15. The demands of scalability, which are a feature of many WSN applications, are not an issue, and multi-hopping will be limited to 2 or 3 hops at most. Multi- hopping is used to overcome harsh propagation conditions in and around the body [2] and is required to link ultra-low power devices. This affects the topology, which for a BAN, is most suited to a hybrid STAR network. B. Existing Asynchronous and Synchronous MACs The main approaches for accessing the media in an energy efficient manner in WSNs are asynchronous: Low Power Listening (LPL), or synchronous: Scheduled Contention or TDMA slot allocation. The shortcomings of these protocols when applied to a medical BAN application can be 978-1-4244-4067-2/09/$25.00 © 2009 IEEE Wireless VITAE’09587 summarised below: • TDMA schemes are contention free but are not flexible, adaptive and scalable. Synchronisation mechanisms are an overhead cost in terms of energy used. • LPL schemes are scalable, flexible and adaptive but are susceptible to high energy cost in transmitter and receiver due to extended preamble mechanisms. • Scheduled contention MACs are scalable, flexible, adaptive, however require maintenance of a schedule to reduce collision cost overhead control. Scheduling overhead has a cost in terms of energy consumption. The IEEE 802.15.4 [3] standard has been examined as a platform for BANs however there are some limitations in meeting the requirements of IEEE 802.15.6 (BAN group [4]), in terms of power consumption [5] and also, as this paper will show, in flexibility and adaptability. The IEEE 802.15.4 is not adaptive to channel quality variation, not optimised for heterogeneous devices, or adaptive to multiple applications, and there is no guarantee for life-critical transmissions. Its beacon mode has an overhead cost in energy whereas the non-beacon mode has better energy consumption at the expense of reduced flexibility [5]. Its rigid superframe size and beacon interval leads to over-provisioning (higher energy) or under-provisioning (poorer QoS delivery). A medical BAN will have to accommodate two types of data which are characteristic of medical services: periodic and non-periodic. Periodic data is traditionally best suited to a TDMA type protocol, i.e., monitoring temperature, glucose levels etc., whereas non-periodic data is best suited to a contention style MAC, i.e., medical emergency. The key driver for this new work is that no wireless standard has yet been adopted which governs medical BANs incorporating implantable and wearable devices. The proposed MedMAC solution attempts to provide flexibility, scalability, and adaptability, combined with ultra low power consumption. Section II will set out the proposed architecture for the MedMAC, while Section III will compare its power efficiency with the IEEE 802.15.4. II. MEDICAL MEDIUM ACCESS CONTROL PROTOCOL (MEDMAC) A. Scope The key features of this protocol include: contention free channel access over a variable number of TDMA channels; energy efficient and dynamically adjustable time slots; a novel adaptive and low-overhead TDMA synchronisation mechanism; optimised energy efficiency by dynamically adjusting the QoS requirements using ongoing traffic analysis; and optional contention period used for low grade data, emergency operation, and network initialization procedures. All devices will sleep or run idle to save power when not transmitting or receiving. Within each superframe (period between beacon transmissions), slots will be allocated to devices by the coordinator and will be given up by the device when not in use. The most appropriate topology for the medical BAN is a STAR topology with the central coordinator worn outside the body or fixed in a bedside position. The main features of MedMAC are summarised below: • device synchronisation maintained without waking for regular beacon; • maximises energy/bandwidth efficiency by dynamically adjusting QoS provision based on traffic analysis; • a single adaptable superframe structure to facilitate three classes of QoS, combining contention and dynamic slot reservation e.g.: o Class 0: Low grade data, asymmetric, < 1000 bps e.g. temp monitoring, respiratory, pulse sensor. o Class 1: Medium grade data, asymmetric/symmetric - ≤ 250 kbps, e.g.: ECG, EEG, blood pressure, SPO2. o Class 2: High grade data up to 10 Mbps. asymmetric/symmetric; medical imaging, video, EMG, Capsule Endoscope; • dynamic and adaptive bandwidth allocation - number of devices from 2–256 with a dynamic slot duration; • flexible delivery of data while maintaining device sleep time of > 99%; • adaptive packet size depending on channel quality, and traffic and QoS analyses; B. Proposed MedMAC Architecture TDMA is an attractive solution for medical BAN applications as it is suited to periodic data. Guaranteed timeslots for each device removes the possibility of collisions from other nodes in the network and the resultant waste in energy. The novel aspect of this proposal is that each device will have the exclusive use of the channel for a fixed timeslot, without the synchronisation overhead normally associated with the TDMA mechanism. Another novel aspect of the MedMAC is the capability to accommodate variable slot sizes simultaneously for heterogeneous applications. To accommodate emergency or life-critical scenarios, low grade data applications and network set up procedures, the MAC will also have an optional adaptable contention period. In a life-critical scenario TDMA slots can be overridden with contention access giving priority to emergency messaging. The novel aspect of this will be the flexibility in the ratio of the slotted period to the contention period, which can vary depending on applications and associated traffic demand. 1) Beacon Period and Multi-Superframe The MedMAC will incorporate a multi-superframe structure for all three classes of MAC and is shown in Fig. 1. The basic unit of the structure is a dynamic and programmable period (superframe) bounded by a beacon frame sent at regular intervals by the coordinator. The beacon period consists of an optional contention period and a contention free period made up of timeslots. The contention free period and the contention period make up a total number of timeslots ranging from 2 to 256; the proportion of time slots in each period is dynamically 588 adjustable and dependent on the number of devices and associated applications. The durations of the superframe and the timeslots are also programmable and dependent on the application requirements including sleep/power saving demands. Figure 1. Multi-Superframe structure for the MedMAC protocol. The MedMAC protocol relies on synchronisation between the coordinator and other nodes to maintain timeslot synchronisation. In low power wireless networks waking up the receiver every superframe to listen for a beacon is a significant drain on energy. The MedMAC protocol incorporates a novel synchronisation mechanism where a node can sleep through a number of beacon periods without losing synchronisation. The duration of the multi-superframe is defined by this synchronisation scheme and is equal to the number of beacon periods through which the node can sleep. The node receiver is only activated to listen for the beacons bounding the multi-superframe and can ignore all other beacons (Fig. 1). 2) Adaptive Guard Band Algorithm Maintaining synchronisation of devices while sleeping through beacons can be achieved by using a combination of timestamp scavenging and an Adaptive Guard Band Algorithm (AGBA). In timestamp scavenging, synchronisation can be maintained using an updated timestamp field as long as there are regular packets from the coordinator to a node (e.g., Data, Ack, and Beacon). However, if there is an interruption in the regular delivery of packets from the coordinator, or simply that an application requires a very low throughput then the node cannot rely on scavenging packets from the coordinator to maintain its timing. In this case, the AGBA allows the node to sleep through many beacon broadcasts without losing synchronisation with the coordinator. It does so by introducing a guard band for each of the timeslots which can be dynamically adjusted to track the actual drift of local time bases. The Drift Adjustment Factor (DAF) minimises the waste of bandwidth using guard bands. The nodes can sleep through many beacon periods and need only be refreshed with new timing corrections by capturing the beacon at the beginning of the next multi-superframe. Upon start-up the AGBA introduces a guard band to each timeslot which allows for the maximum possible drift of the combined coordinator and node timing crystals. All nodes will be informed by the beacon when the AGBA is initially invoked at the start of a multi-superframe. The beacon packet will also contain information which defines slot allocation, beacon period, and multi-superframe duration. At the start of the multi-superframe all nodes are brought into synchronisation by the timestamp from the coordinator. After this point the algorithm will be invoked to calculate the guard band for each node. This calculation is based on the maximum drift of the crystals and the time elapsed from start of multi- superframe to the end of the timeslot (see Eqn. 1). The time elapsed depends on the timeslot number allocated to the node, the timeslot duration, and the beacon period. These results are known as the default guard band (GB) values. )__()__()_( RXtolcrysTXtolcryselapsedtimeGB ××= (1) The default GB values can be calculated and stored in the coordinator during the device registration period. From the default GBs the coordinator calculates the values for the associated timeslot start times for each node with reference to the beacon start time for the current multi-superframe. These values will be transmitted to each node during the initialization phase of the network where they will be stored as default values. For each subsequent beacon period the node will recalculate the guard band required based on the increasing value of time elapsed. A revised timeslot start time based on the new guard band will be calculated for each new beacon period. This can continue until the end of the multi-superframe when all nodes will receive a beacon which corrects the timing of each node to the coordinator. The size of the multi-superframe is dictated by the permitted maximum size of GBs. The larger the GB allowed, the larger the multi-superframe. For the experiments reported in this paper an arbitrary 50% rule is used which limits the maximum GB size to 50% of the timeslot. In subsequent multi-superframes the AGBA and DAF mechanisms will be deployed unless one of the following conditions occurs: timestamp scavenging; initial multi- superframe following device registration; or default GB conditions. The individual default GB duration for each slot is calculated using the following equations. The GB for the first slot in the first beacon period will only have a single guard band given by: )_).((, tolXSDmnGB = (2) For the remaining slots in the first beacon period the following equation will determine the GB for each slot. This equation is iterative and incorporates the GBs of previous slots into the time elapsed: ( ) tolX nGBGBGBSDntolX mnGB _1 12.....221._ , − − ++++ = (3) Beacon period with n timesslots Multi-superframe with m beacon per Contention free period Optional contention period Beacon 589 Now to calculate the GBs in the subsequent beacons of the multi-superframe the following generic equation can be used: ( ) tolX n GBGBGBBPmSDntolX mnGB _1 1 2... 2 2 1 ).1(._ , − − ++++−+ = (4) Alternatively, GB values can also be calculated by using: )_.(1,, tolXBPmnGBmnGB +−= (5) From the GB values the new slot start times (SST) can be determined for each node: mnGBmGBmGBSDnmSSTmnSST ,12...,22,12)1(,1, −++++−+= (6) and mGBBPmSSTmSST ,1)1(1,1,1 −−+= (7) Where n indicates slot number in a beacon period and its range is max1 SNn ≤≤ where maxSN is the maximum number of slots; m is beacon period number in a multi-superframe and its range is max1 BPNm ≤≤ where maxBPN is the maximum number of beacon periods in the multi-superframe; SD is the slot duration measured in seconds; tolX _ is the combined tolerances (ppm), of the coordinator and the device time base crystals; BP is the beacon period measured in seconds and SST1,1 is the beacon start time. Figure 2. Multi-Superframe with beacon frame, timeslots and guard bands. 3) AGBA with Drift Adjustment Factor For increased energy saving the AGBA incorporates a novel optional feature called the drift adjustment factor (DAF). Using AGBA to generate the default GBs means that they are the worst case values and will increase with time until the next time correction at the beginning of the next multi-superframe. However, in practical cases the actual crystal drift may be a lot less than the default GB. The Drift Adjustment Factor (DAF) will assess the relative drift between the default GB and the actual drift (AD) and make an adjustment to the default GB accordingly. If the drift does not grow at the same rate as predicted by the AGBA then the actual GB can be reduced. This adjustment is made at the end of the multi-superframe by the coordinator based on information received from the nodes during the multi-superframe. The new GBs are calculated and the new timeslot start times are delivered to the nodes by the beacon at the beginning of the next superframe. Each time a packet arrives at the coordinator from a node it delivers the actual timeslot start time or end time. From this value the coordinator calculates the AD and records the worst- case AD for each node in the multi-superframe. The node with the worst case AD will be the reference for the DAF (ADref). This avoids the possibility of over-correction. The percentage difference between the predicted GB and the AD is calculated for the most recent worst case node. If this is greater than 5 % of the timeslot then adjustment is made to the GB value. If %5,, > − SD pnADpnGB of time slot (8) then invoke Drift Adjustment Factor (DAF). If ADref is larger than the corresponding GB in the first beacon period of that multi-superframe, i.e.: If prefpmref GBAD ,1,,, > then invoke default AGBA algorithm. If prefpmref GBAD ,1,,, = then let prefpref GBGB ,1,1,1, =+ If prefpmref GBAD ,1,,, < then apply the following equation to reset the GB for the corresponding timeslot in the first BP of the next multi-superframe: ⎟⎟⎠ ⎞ ⎜⎜⎝ ⎛ − −=+ 2 ,,,1, ,,1,1, pmrefADprefGB pmrefGBprefGB (9) Where p is current multi-superframe number, p+1 is new multi-superframe number, m is beacon period number, and ref is timeslot number of the reference slot. This effectively allows the GB excess to be divided by 2 in consecutive super-multiframes until the percentage difference between previous GB and current AD is equal to or less than 5. If it is less than 5% then DAF will increase the GB using equation (9). ⎟⎟⎠ ⎞ ⎜⎜⎝ ⎛ − −=+ 2 ,,,1, ,,1,1, pmrefADprefGB pmrefGBprefGB (10) The new GB can be determined for the corresponding time slot in the first beacon period (m=1) of the next multi- superframe. This is the reference GB and it is used to extrapolate the remaining GBs for all the timeslots. The difference between the new guard band (GBref) in the first beacon period of the next multi-superframe and the corresponding GB of the first BP in the previous multi- superframe is calculated in order to generate the GB adjustment factor R. pmref GB pref GB pmref GB R ,, 1,1,,, + − = (11) The guard bands for all the nodes in the first beacon period