Power-efficient Communication Protocol for Social Networking Tags for Visually Impaired Problem Social Networking Tags System for Visually Impaired is an project aims to utilize electronic id technology to enable visually impaired persons to identify surrounding people in a social setting such as a conference. The exchange of information is the main enabler of the application, and the design of an efficient communication protocol among the resource-limited unsynchronized electronic id tags is the key to the entire system. Featuring CR2025 coin battery packed with limited power, adding to the fact that participants and their tags are all constantly moving, this communication protocol has to be able to establish synchronization and exchange information in the short time span when people walk passed each other while accomplishing this in a power efficient manner. This report highlights the design of this application-specific communication protocol, and compares its performance with other protocols considered but not chosen during the design process. Previous Infrastructure Social Networking Tags System for Visually Impaired project, an initiative in Harris Mobile Computing Lab in University of Florida, is a project sponsored by NSF. The scope of the project includes the hardware design of a brand new electronic id tag, the design and implementation of an application-specific communication/synchronization protocol between tags, and the end system (blind unit) that process collected information and present to visually impaired persons. The smart active electronic id tag hardware has a size of only 35mm x 29mm, featuring a low-power 32kHz PIC16F690 microcontroller, a low-power nrf24l01 chip running Wibree [1] short range radio. It operates with 2.2V to 3.0V power supply, and features a physical on/off switch, while the transmission power level can be controlled using software. Offering a short range radio transmission with maximum transmission rate of 1Mbps, the tag is designed to transmit information in bursts within a short range to ensure good power performance. Literature Survey Since the communication protocol of interest is application-specific, it is unlikely that a totally comparable protocol can be found and employed. But a particular data link layer/routing protocol from Motorola designed for sensor network nodes [2] address certain common aspects as our project. With the assumption that sensor nodes goes into sleep mode at majority of time and only wakes up periodically in order to conserve energy, Callaway described Mediation Device Protocol (MD) in which nodes take stochastic turns to be MD, the central synchronization manager, that records time differences between awaken time slots of each node, and synchronizes these nodes so they can be awake at the same time. This protocol results in long message latency and low system throughput, but it is suitable for non-real-time, low
data density applications such as environmental monitoring. 802.11 MAC protocol [3] introduces CSMA/CA basic protocol that utilizes carrier sensing, random backoff value and countdown during channel idleness to allow multiple wireless nodes to hit the airwave. An optional reservation scheme with RTS/CTS is provided to avoid collision, which is suitable for longer message (hence greater collision penalty) and higher node density. Motivation The MD protocol from Motorola is extremely efficient for non-real-time and low data density applications, and works well in ad-hoc fashion with resource-stricken sensor nodes. However, our target application requires a higher data transmission density. Although our application is not a real-time system, it does require information to be delivered to visually impaired persons in a timely fashion. On top of that, the high mobility of tags negates the benefit of establishing MD as the center of neighborhood synchronization, when persons wearing tags have a fair chance of roaming away before the next synchronization point. 802.11 MAC protocol provides some insights about sharing airwave between multiple nodes, but as WiFi is not used as the underlying communication module, it does not make sense to implement the whole 802.11 protocol suite. It is also too costly a protocol to use with this hardware and on this application. A more application specific approach better suited to the hardware and resources available is needed. Technique (Solution) and Implementation The basic communication scheme for this protocol is described below. The protocol can be divided into two parts, the first concerns about the power-efficient scheduling of communication task of a single blind unit/single guest tag unit exchange; and the second concerns about single blind unit communicating with multiple guest tag units. Based on our preliminary study, the effective range of communication of our tags is 5 feet. Under a social setting, a well-visional person would take about 2 minutes to cross that distance, and it would take even longer for visually impaired persons. The periodic guest information solicitation is therefore set to be performed every 3 minutes (180 seconds). Every 180 seconds, the blind unit would broadcast a solicitation message piggybacked with the synchronization information. The guest tags, once received this solicitation, reply back with personal information at the specified synchronized transmission period. For blind unit to get all information about surrounding guests, we need to make sure at least one solicit request will hit the guest unit while it is in Rx mode. The designed protocol is shows as in Figure 1, where optimal x is calculated to minimize the cost function, which is the overall system power consumption, including both blind and guest units. For 1280 Byte solicit packet, it takes 10 ms to transmit, and the optimal listening period would be x = 1250 ms. 2
Solicit x Sleep Solicit Sleep Solicit Sleep y Collecting Period Blind Unit Sleep Listening Reply Guest Unit x 180 sec Figure 1. Communication Protocol y C Listening C Listening Sleep Blind Unit Sleep R Reply Data Random Delay R L Guest Units Figure 2. Communications MAC Protocol The collecting period on the blind unit y is yet to be decided at this point. The primary consideration for this decision relies on what kind of MAC protocol to be used and the associated time parameters when there are multiple guest units present. After evaluating several proposals, RTS/CTS reservation scheme, as shown in Figure 2, is chosen for its efficiency and simplicity. Each guest unit randomly pick a delay factor K less than 256, and waited K * 130 µs (carrier detect time) before sending out the RTS. Selecting K to be 256 gives us a probability of 83% that no collision would happen even if there are 10 tags trying to reply to the same solicitation request. If the guest unit senses there are transmissions in the air, it suspends the countdown of K, and resumes once the on-air signal is terminated. The blind unit listens to the incoming RTS or data transmission; if a RTS is received, it replies with CTS and waits for actual information to come; if actual data packets arrive without a previously cleared RTS, the data is discarded. Based on this RTS/CTS scheme, the length of collecting period y is set to 237.6 ms. Analytical evaluation of this protocol will be shown in the results section. This protocol is currently being implemented on the actual electronic id tags and CCS compiler is chosen for its code efficiency, built-in functions, low cost and availability of convenient IDE. The focus of this report is on the analysis of the power efficiency of the designed protocol, and the full demo and implementation is beyond the scope. 3
Results For evaluation of the design, we compare the presented protocol to two baseline alternatives that were considered during the design and the implementation process as shown in Figure 3. Based on the data sheet of the microprocessor/transceiver used in the hardware, we know the following parameter for a fact. 1. 12mA of current need to be supplied to Transceiver when in Tx mode, 14 ma is needed for Rx mode. 2. In sleep mode (standby), only 32 µa would be supplied, and is considered ignorable because of the 3 order of differences. For the operation intended with the system, the effective transmission range is around 5 ft and the resulting propagation delay of less than 0.01 µs can be ignored. a. Remote Trigger b. Continuous Listening c. Periodic Announcement Listen Transmit Sleep Data Collect (Repeat of Fig 1) Figure 3. Communication Protocol Comparison The three protocols include remote trigger (Figure 3a), which would be the most efficient protocol, since the guest unit only wakes up when a solicitation request from the blind unit arrives. In practice, however, the power of the signal arrived is not strong enough to activate the trigger. Remote trigger protocol would serve as the lower bound of the power consumption evaluation. In Figure 3b, the continuous listening protocol makes guest unit continuously listening for the solicit request, allowing them to respond to the inquiry promptly. The continuous operation makes this an extremely expensive proposition, and it serves as an upper bound during the protocol design process. The power consumption in 3-minute period for these three protocols is reported in Table 1. Table 1. Whole-System Power Consumptions of Three Candidate Communication Protocols Protocol Remote Trigger Continuous Listening Periodic Announcement Power consumption 290 ms-ma 2.16 x 10 6 ms-ma 34919 ms-ma (1x) (~260,000x) (120x) For MAC protocol, we examined four different proposals, the random delay scheme, extended time slot scheme, centralized clock system and the RTS/CTS reservation scheme. Because the difference between current supplied to the transceiver in Rx and TX modes is only marginal in a short period of time and we assume the voltage supplied to the system is regulated at a constant level, the power consumption can be fairly accurately estimated solely by the time it takes to complete the multiple-tag data exchanges. Each of these four MAC protocols require each guest tag to pick a random delay value K, but they are 4
used differently. The random delay scheme is similar to the 802.11 basic MAC protocol which delay transmissions based on a countdowns from K when sensing idle airwave. With 130 µs as the basic unit of delay, this scheme would take 133.15 ms to complete one round, but there is a great chance collision would happen if tags cannot detect all transmissions in the air. This constraint does not satisfy the environment of the targeted application, hence is ruled out early in the design. The extended time slot scheme assign a 10 ms slot for each of the K (delay) value, and each guest tag would reply based on the randomly selected K value after listening for the idle airwave since the communication protocol described above allow different tags to be synchronized. Assuming reasonable clock drift on each tag and the blind unit, this scheme would work reasonably well, even in highly dense populated locality such as dance floor, only at most 17% of reply may collide, as the reliability of the reply rise sharply to more than 96% if there are only 5 tags trying to reply. This scheme, however, would need 2560 ms to complete, which is an order higher than any other schemes. The centralized clocked system make the blind unit send out clocked K number every 130 µs, and it works much like how patrons get a number tag in a busy post office, when send out reply only when their number is called. This scheme would take 228 ms to complete in the worst case, and avoids the problem of clock drifting as well as the packet collision. However, the blind unit needs to switch between Rx and Tx modes 256 times in a high frequency. The last MAC protocol, which is shown in Figure 2, is RTS/CTS reservation protocol. Used as 802.11 optional bandwidth reservation scheme, each tag count down based on the random K value of 130 µs slots, when the countdown reaches 0, the guest unit send out a RTS, which is responded by the blind unit with CTS, and the information exchange ensues. In the worst case, this scheme takes 237 ms to complete the data exchange, and also has low probability for packet collision. Although the time it takes to complete the round is slightly higher than centralized clocked system, it is attractive also for its simplicity. Based on these two considerations, RTS/CTS reservation scheme is chosen as the MAC protocol. The discussion above is summarized in Table 2. Table 2. Time Required to Complete Multiple Data Exchange Four Candidate MAC Protocols MAC Protocol Random Delay Extended Slot Centralized Clock RTS/CTS Time Required (ms) 133.15 2560 228 237 Drawback All nodes need to Collision still possible, High frequency Collision of RTS hear all messages time consuming Rx/Tx modes still possible in the air flipping Conclusion and Future Directions Assistive systems for the visually impaired have been one of the major focuses for health care, wearable computer and intelligent environment communities. This project is different, however, because it focuses on providing assistance in enhancing their social interactions. One of the keys to the success of this project is to design a communication/synchronization protocol that can handle unsynchronized tags passing 5
one another in fleeting moments without wasting lots of energy. This project considers hardware, software and situational constraints and evaluates multiple proposals to come up with the final design, and analytically evaluate the costs and benefits of these proposals. The fully functional prototype system is expected some time in May 2007. A usability study involving visually impaired subjects is scheduled to take place starting from June 2007. We hope to evaluate the real benefit this infrastructure-less system can provide in a realistic setting following the preliminary usability study. Acknowledgement I would like to thank Dr. Mishra for his suggestions regarding the direction and points of consideration on this protocol design project. I would also like to express my gratitude to Dr. Helal for his leadership and continuous support on the overall Social Networking Tags System for Visually Impaired project and the funding agency NSF. Many thanks also go to my collaborators Mr. Ed Koush and Mr. Raja Bose. References 1. Wibree Open Industry Initiative, Wibree Data Sheet, http://www.wibree.com/technology/wibree_2pager.pdf 2. E. Callaway, Wireless sensor networks: architectures and protocols, Chapter 1 7, Auerbach Publications, 2004 3. IEEE 802.11, 1999 Edition (ISO/IEC 8802-11:1999) IEEE Standards for Information Technology Telecommunications and Information Exchange Between Systems Local and Metropolitan Area Network Specific Requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, http://standards.ieee.org/getieee802/download/802.11-1999.pdf 6