Distributed Sequential Access MAC Protocol for Single-Hop Wireless Networks

Wireless Pers Commun DOI 10.1007/s11277-013-1142-8 Distributed Sequential Access MAC Protocol for Single-Hop Wireless Networks Ki-seok Lee Cheeha Kim Springer Science+Business Media New York 2013 Abstract Medium access control overhead is the primary reason for low throughput in wireless networks. Performing blind contentions, contentions without any information of other contenders, and exchanging control message are time-consuming control operations. In this study, we propose a new MAC protocol called distributed sequential access MAC (DSA-MAC) which provides the transmission order without any explicit control operations. It may induce very light control overhead; therefore, compared to existing wireless MAC protocols, DSA-MAC can remarkably enhance network throughput. Keywords Unique ID-based MAC Sequential access Wireless networks 1 Introduction In recent years, short-range single-hop wireless networks such as mobile hotspots [1], Bluetooth WPAN [2], and resource-sharing home networks have been widely used. These networks adopt traditional wireless MAC protocols including contention based MAC, e.g., 802.11 DCF [3] and polling based MAC, e.g., 802.15.1 [4]. Although both types of MAC protocols are easy to implement, they consume some portion of the MAC link capacity. Contention-based MAC protocols perform blind contention whereas polling-based MAC protocols exchange control messages for polling and polling-list management. Because of heavy control overheads, traditional MAC protocols rarely achieve high network throughput, which is essential for supporting multimedia content communications and real-time applications [5]. In addition, wireless MAC should be evolved to over the scalability problem in terms of parameters such as physical data rate or the number of users because the scale of networks largely grows every year [6]. A single-hop wireless network may be characterized such that every node can monitor the network. Based on the characteristics of being able to monitor the medium, we propose a K. Lee C. Kim (B) Networking and Distributed Systems Lab, Room 341, PIRL, POSTECH, Hyoja-Dong, Pohang 790-784, Korea e-mail: mic0@postech.ac.kr

K. Lee, C. Kim new MAC protocol, called distributed sequential access MAC (DSA-MAC), which excludes any operations for blind contention or control message exchange because it can control the medium access sequence using the monitoring result. DSA-MAC enables all active nodes to sequentially access the medium according to their assigned order of unique IDs. All active nodes transmit data in turn from the smallest ID node to the largest ID node. Each active node can find its turn simply by counting the number of previous data transmissions. DSA- MAC enables a node to acquire a unique ID through a distributed ID reporting process when it becomes active. It also enables all active nodes to detect unused IDs and dynamically rearrange their ID accordingly. Most operations involved in DSA-MAC require very light overhead so that DSA-MAC can remarkably enhance network throughput and support good scalability; this result is proved by comparing medium utilization. 2 Distributed Sequential Access MAC (DSA-MAC) 2.1 Operation Overview In DSA-MAC, time is divided into a sequence of periods, each consisting of three subperiods: synchronization, ID reporting, and data transmission (Fig. 1). At the end of each ID reporting subperiod, all active nodes can find their largest ID and determine the medium access sequence according to the order of IDs assigned to them. To proceed to the ID reporting process, all nodes need to be synchronized using beacon signals, which need not be decoded. Initially, any node can send a beacon signal. Subsequently, the last data transmission node in a period sends a beacon signal in the next period. All active nodes start reporting their IDs after the beacon signal. When a node becomes active, it should acquire a new ID. For the moment, we assume that all active nodes have their own unique IDs. The acquisition of a new ID will be described in the subsequent section. All active nodes perform the ID reporting process in the same way as the contention process of CSMA/IC [7], which is outlined here. Note that ID reporting helps a newly active node to independently acquire a new ID. For ID reporting, all active nodes keep their IDs in binary form, having the same number of bits, called the length of the ID field. The number of ID reporting slots in the ID reporting subperiod is defined to be equal to the length of the ID field. All ID reporting slots, which are arranged in consecutive order, match each bit of the ID starting from the most significant bit. Each node executes the following operations from the first ID reporting slot. At the ID reporting slot that represents the bit 1 in its ID, the node sends a bit-on-signal. At the ID reporting slot that represents the 0 bit in its ID, the node senses the medium. When it senses that the medium is busy, it stops the ID reporting process; otherwise the node continues. According to this rule, all nodes, except the node that has the largest ID (D in Fig. 1), stop the ID reporting process. Therefore, all nodes can find the largest ID (0110) among all active nodes. In CSMA/IC, only the node, which has the largest ID, gets the medium access opportunity. On the other hand, in the proposed scheme, all active nodes get medium access opportunities using the largest ID information, which is used to indicate the number of transmissions in the subsequent data transmission subperiod and currently unused IDs. In the data transmission subperiod, each active node transmits data in the order of IDs from 0 ID node to the largest ID node. To prevent conflict between two consecutive transmissions, the minimum idle time, called the inter transmission space (ITS), is defined between them. Each node can find its turn by counting the number of data transmissions. In Fig. 1, node A, which has ID 0001, can transmit after a transmission because it has the second ID. However,

Distributed Sequential Access MAC Protocol Node states 1 2 3 A B C D E ID 0001 ID 0000 ID 0011 ID 0110 NewI D 0111 ~ 1111 ID 0010 ID 0101 ID 0100 ID 0011 Check points Sense medium Bit-on signal ITS SIFS ITS SIFS 1 2 3 ITS ITS SIFS ITS ITS ITS SIFS ITS Beacon TX node B A C K TX Node A A C K ID 0010 is vacant. à rearrange ID TX Node C A C K ID 0011 is vacant. à rearrange ID TX NodeD A C K Time Synchronization subperiod ID Reporting subperiod Data transmission subperiod period (t p) *TX: Transmission *I TS :InterTransmission Space Fig. 1 Three subperiods of DSA-MAC some IDs between 0 and the largest ID may not be used (e.g. 0010 in Fig. 1) andsome transmission turns should be skipped (the third transmission turn corresponding to 0010). The unused IDs are called vacant IDs. A vacant ID can be detected if no transmission is detected at the expected time. At the first check point in Fig. 1, it is expected that a node whose ID is 0010 transmits data but there is not such a node. So, 0010 is a vacant ID. Because a vacant ID induces unnecessary idle time, our proposed scheme introduces the vacant ID removing process; whenever nodes detect a vacant ID, every node which has larger ID than the vacant ID decreases its ID by 1. This process works even when multiple vacant IDs exist and they are successive or distributed. At the first check point in Fig. 1, all nodes can detect vacant ID 0010. Nodes which have larger IDs than 0010, node C and D decrease their IDs by 1. After this process, node C whose updated ID is 0010 transmits data. At the second check point in Fig. 1, all nodes can detect vacant ID 0011. Node D whose ID is larger than 0011 decreases its ID by 1. At the check point 3 after this process, 0011 is still vacant ID. Therefore, node D performs another vacant ID removing process. After that, node D whose ID becomes 0011 will transmit data. In summary, a node whose ID is larger than the vacant ID decreases its ID until no consecutive idle ITSs are detected. Note that it takes only an additional ITS time to detect. So, the time cost to for removing a vacant ID is very light compared to what we obtain from contention-free transmissions. When a node fails to count the number of transmissions and misses its transmission turn, it may cause a collision. For example, if node with ID 0100 misses counting a transmission, it stays idle during its transmission turn and waiting node having larger IDs than it by 1, node with ID 0101, moves forward one transmission turn. Therefore, two nodes collide with each other. When two nodes collide, they must find new IDs in the next period. 2.2 ID Management Operations In this subsection, we present our proposed self ID acquisition scheme which is based on medium monitoring, and a dynamic ID length control scheme for reducing the number of ID collisions and vacant IDs.

K. Lee, C. Kim After ID reporting period, a newly active node randomly selects an ID that is larger than the current largest ID and participates in ID reporting process in the next period. Therefore, there is no conflict with existing IDs, but ID collision, which causes data transmissions to collide, is possible when multiple newly active nodes choose the same ID at the same time. Note that ID collision is inevitable in distributed ID acquisition. When collision occurs, all collided nodes repeat this ID acquisition until they acquire unique IDs. Unfortunately, the probability of ID collision remains the same in successive ID acquisition. Thus, to reduce the probability, the length of ID field increases in our proposed scheme when the probability is larger than a threshold specified in Eqs. (1) (6). After ID acquisition, one of the newly selected IDs naturally becomes the largest ID in the next period because all newly selected IDs are larger than any IDs in this period. Therefore, the data transmission subperiod will be extended until the largest ID node among the newly active nodes transmits data. There could be some vacant IDs after random selection. The number of ID collisions and vacant IDs depends on the length of the ID field. As the length of the ID field increases, the range of selectable IDs for newly active nodes also increases and the number of ID collisions is likely to drop. On the other hand, the interval between newly selected nodes increases. Therefore, the number of unused IDs between the largest ID in the previous period and the largest ID after ID acquisition also increases. An ID length control scheme is proposed to manage this trade-off relation between the number of ID collisions and the number of vacant ID. The metric used for ID length control is time wasted in a period, t w, which is caused by vacant IDs and ID collisions within a period. t w can be calculated using Eqs. (1) (6) where N unused is the number of unused IDs, N new is the number of newly active nodes during aperiodandt ITS is the length of ITS. It is assumed that all active nodes have the same transmission time (t x ). t w = 2 t ITS E[N vacant ]+t x E[N IDcollision ] (1) m E[N vacant ]= (P[N new = k] E k,m [N vacant ]) (2) E k,m [N vacant ]= k=1 m ( x k 1 ) m k (x k + E k,m [N IDcollision ]) x=k E[N IDcollision ]=m (3) (P[N new = k] E k,m [N IDcollision ]) (4) k=2 P[N new = k] =e λt (λt p p) k k! ki=2 (i k C i ( m 1 )i (1 m 1 )k i ) (k 2) E k,m [N IDcollision ]= 0 (k = 1) E[N vacant ] is the average number of vacant IDs in a period and E k,m [N vacant ] is the average number of vacant IDs when N new = k and N unused = m. E[N IDcollision ] is the average number of nodes which experience ID collision in a period and E k,m [N IDcollision ] is the average number of nodes which experience ID collision in a period when N new = k and N unused = m. We assume that N new follows a Poisson distribution with rate λ (Eq. 3) and we approximate E k,m [N IDcollision ] adopting a binomial distribution (p = m 1 ) (Eq. 4). Using these assumptions, E[N IDcollision ], E[N vacant ] and t w can be obtained. (5) (6)

Distributed Sequential Access MAC Protocol Periodically solving Eqs. (1) (6) to find the ID field length that minimizes t w in dynamically changing network conditions, especially λ, requires high computing power. To avoid complex computations, we propose a simple guide derived from rigorous computing over reasonable parameters. A node measures the number of new nodes for a certain number of periods to acquire N new, and monitors the ID reporting process to acquire N unused. When measured N new < 0.6, the ID field length should be increased by 1 if average N unused < 2. When 0.6 < measured N new < 2, the ID field length should be increased by 1 if N unused < 4. When average measured N new > 2, the ID field length should be increased by 1 if N unused < 8. The above three rules can also be applied to the ID field length decrement. The ID field length should be decreased by 1 if the above conditions are satisfied after the ID field length decrement. All active nodes insert a 0 bit in the most significant bit for ID field length increment and eliminate the most significant bit for ID field length decrement. 2.3 PHY Layer Issues ID reporting signal stability and synchronization accuracy are important engineering issues in the DSA-MAC protocol. The ID reporting signal should be differentiated from background noise. If the magnitude of the background noise is smaller than the carrier sensing threshold, nodes can easily differentiate the ID reporting signal, even in a one-dimensional signal space (magnitude axis) [8]. The synchronization errorinduced by the synchronization beacon signal arrival time difference, which untunes the starting point of the ID reporting period, is equal to the maximum propagation time in the network. An ID reporting slot is long enough to compensate for the synchronization error. In addition, the effect of clock drift is insignificant because synchronization is performed at the beginning of every period. 3 Numerical Results In this section, we compare the control overhead of our proposed MAC with CSMA/IC [7] and CP-Multipoll [9] using the PHY layer parameters of 802.11 standard [3], indicated by italic type in the following paragraphs. Because DSA-MAC rearranges the access sequence at the end of every period, we calculate the total overhead for N active, the number of active nodes, times transmissions. The DSA-MAC overhead is SynchronizationTime + IDReportingTime + (N active + 1) IT S (Fig. 1). SynchronizationTime (23µs), which equals aslottime +3 aairpropagationtime, includes abundant propagation time for synchronization. IDReportingTime is IDlength aslottime(20 µs). ITS is defined to take 15 µs(2 (arxrfdelay + arx- PLCPDelay + amacprocessingdelay) +arxtxturnaroundtime), which is larger than asifstime(10 µs) to detect vacant ID in a stable way. Therefore, the total overhead of our distributed sequential access MAC is 23 µs + IDlength 20 µs + (N active + 1) 15 µs. CSMA/IC [7] is a collision-free contention-based MAC protocol which is known to require the smallest overhead among all contention-based MAC protocols. CSMA/IC also performs contention by comparing the unique IDs of all contenders after a short synchronization process. The CSMA/IC overhead per contention becomes anidlesensingtime + SynchronizationTime + IDComparingTime. The total CSMA/IC overhead is N active (15 µs + 23 µs + IDlength 20 µs) because the contention process completes only in IDlength times contention slots. The contention-period multi-polling protocol (CP-Multipoll) [9] is a polling-based MAC protocol that always succeeds in polling. A dedicated node assigns unique backoff values to

K. Lee, C. Kim (a) (b) Fig. 2 Medium utilization comparisons. a Data rate: 54 Mbps, b data rate: 54 4 Mbps all nodes (STAs) in the polling list. All STAs perform the CSMA/CA contention process using these uniquely assigned back-off values. The best case overheads of CP-Multipoll is N active BackoffValueAssigningTime +N active (adifstime(50 µs)+aslottime(20 µs)). BackoffValueAssigningTime is a data transmission time that includes at least asifstime(10 µs) + apreamblelength(144µs) + aplcpheaderlength(48µs). BackoffValueAssigningTime is not a repetitive control overhead, so it is excluded in overhead calculation.

Distributed Sequential Access MAC Protocol Figure 2 clearly shows that our DSA-MAC includes a much smaller control overhead than the aforementioned MAC schemes. Though the contention overhead of CSMA/IC is compatible with DSA-MAC, a single data transmission is made after each contention process. CP-Multipoll also includes a polling overhead, which should be performed repetitively. Due to the light control overhead, DSA-MAC is more scalable than others. As the number of nodes increase, only the utilization of DSA-MAC increases. And, when the data rate speed becomes high, the utilization loss of DSA-MAC is less than that of others. 4Conclusion As the demand for real time communications and multimedia communications of portable electronic devices grows, the throughput requirement for communications between devices comprising a single-hop network becomes high. In this study, we propose a DSA-MAC protocol that controls the medium access sequence without any control operations. DSA- MAC has little control overhead; therefore, we believe that our proposed MAC is and suitable for high-network-throughput applications and support high medium utilization even when the scale of the network becomes large. Acknowledgments This research was supported by the MKE (The Ministry of Knowledge Economy), Korea, under the ITRC (Information Technology Research Center) support program supervised by the NIPA(National IT Industry Promotion Agency) (NIPA-2012-H0301-12-3002). References 1. Pack, S. H., Rutagemwa, H., Shen, X., Mark, J. W., & Cai, L. (2007). Performance analysis of mobile hotspots with heterogeneous wireless links. IEEE Transactions on Wireless Communications, 6(10),3717 3727. 2. Ali, K. A., & Mouftah, H. T. (2011). Wireless personal area networks architecture and protocols for multimedia applications. Ad Hoc Networks, 9(4), 675 686. 3. IEEE 802.11 WG: IEEE Standard 802.11-2007. Part11: Wireless LAN medium access control (MAC) and physical layer (PHY) specifications. 4. IEEE 802.15.1 WG: IEEE Standard 802.15.1-2005. Part15.1: Wireless medium access control (MAC) and physical layer (PHY) specifications for wireless personal area networks (WPANs). 5. IEEE 802.15 WG: IEEE Standard 802.15-04/312r0. Applications and usage scenarios for mesh-wpan. 6. Teymoori, P. (2012). DT-MAC: An efficient and scalable medium access control protocol for wireless networks. IEEE Transaction on Wireless Communications, 99, 1 11. 7. Lee, K., Oh, S. & Kim, C. (2012), A dynamic ID management protocol for CSMA/IC in ad hoc networks. Ad Hoc Networks. http://dx.doi.org/10.1016/j.adhoc.2012.11.008. 8. Goldsmith, A. (2005). Wireless communications. Cambridge: Cambridge University Press. 9. Lo, S.-C., Lee, G., & Chen, W.-T. (2003). An efficient multipolling mechanism for IEEE 802.11 wireless LANs. IEEE Transactions on Computer, 52(6), 764 778.

K. Lee, C. Kim Author Biographies Ki-seok Lee got his B.S. degree in Computer Engineering from Kyungpook National University in 2006, and in 2013 he got his Ph.D. in Computer Science and Engineering from Pohang University of Science and Technology. Cheeha Kim is a Professor (1989-present) at CSE, POSTECH, Korea. He got his B.S. degree in Electronics Engineering from Seoul National University in 1974 and M.S. in Computer Science from University of Maryland, College Park, in 1984. In 1986, he got his Ph.D. from University of Maryland, College Park, in Computer Science. He is a Member of the IEEE Computer society, IEEE Communication society, and ACM. He has chaired a number of international conferences on computer communications, and served as an Editor for LNCS. His research interests include computer communications, mobile computing, cognitive radio, sensor networks, distributed systems and performance evaluation.