* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download RT-WiFi: Real-Time High Speed Communication Protocol for
Wireless USB wikipedia , lookup
Policies promoting wireless broadband in the United States wikipedia , lookup
Asynchronous Transfer Mode wikipedia , lookup
Wake-on-LAN wikipedia , lookup
Computer network wikipedia , lookup
Network tap wikipedia , lookup
Deep packet inspection wikipedia , lookup
Internet protocol suite wikipedia , lookup
Airborne Networking wikipedia , lookup
List of wireless community networks by region wikipedia , lookup
Wireless security wikipedia , lookup
Recursive InterNetwork Architecture (RINA) wikipedia , lookup
Cracking of wireless networks wikipedia , lookup
IEEE 802.11 wikipedia , lookup
RT-WiFi: Real-Time High Speed Communication Protocol for Wireless Control Systems Yi-Hung Wei∗ 1 , Quan Leng∗ , Song Han∗ , Aloysius K. Mok∗ , Wenlong Zhang† , Masayoshi Tomizuka† ∗ Department † Department of Computer Science, The University of Texas at Austin {yhwei, qleng, shan, mok}@cs.utexas.edu of Mechanical Engineering, University of California, Berkeley {wlzhang, tomizuka}@berkeley.edu Abstract— Applying wireless technologies in control systems can significantly enhance the system mobility and reduce the deployment and maintenance cost. Existing wireless technologies, however either cannot provide real-time guarantee on packet delivery or are not fast enough to support high speed control systems, which typically require 1KHz or higher sampling rate. Nondeterministic packet transmission and insufficiently high sampling rate will severely hurt the control performance. To address this problem, in this paper, we present our design and implementation of a real-time high speed wireless communication protocol called RT-WiFi. RT-WiFi is a TDMA data link layer protocol based on 802.11 physical layer to provide deterministic timing guarantee on packet delivery and high sampling rate up to 6KHz. It incorporates configurable components for adjusting design trade-off including sampling rate, latency variance, reliability, and compatibility to existing Wi-Fi network, thus can serve as an ideal communication platform for supporting a wide range of high speed wireless control systems. We implemented RTWiFi on commercial off-the-shelf hardware and integrated it into a mobile gait rehabilitation system. Our extensive experiments demonstrate the effectiveness of RT-WiFi in providing deterministic packet delivery in both data link layer and application layer, which further eases the controller design and improve the control performance. I. Introduction Wireless control systems (WCSs) have received significant attention these years because of their great advantages of enhanced mobility, easier deployment as well as reduced configuration and maintenance cost. WCSs have been widely applied in many application domains including industrial process control and automation [1], healthcare and biomedical systems [2],smart structures [3], and intelligent robotic systems [4], to name a few. A key step in building a wireless control system is to choose the most appropriate wireless communication protocol according to its desired control behavior. Control designers generally look at three most critical design factors when making the decision: sampling rate, reliability and real-time data delivery. These factors directly influence the quality of control, and the performance of the whole system. Many existing wireless technologies have been adopted in wireless control systems. However each of these protocols can only support a small number of specific control applications. Some of them are based on low data rate physical layers, like 802.15.4, and focus on the real-time data delivery 1 The first two authors have equal contribution to this work. Fig. 1: Vision of RT-WiFi protocol and reliability performance, thus are only suitable for lowspeed control applications; Others including Wi-Fi focus more on improving network throughput but pay less attention to the deterministic behavior of data delivery, so that cannot be adopted in control systems which have stringent timing requirements. The lack of a flexible and high speed real-time wireless communication platform makes it difficult to find an ideal wireless communication protocol for many wireless control applications and complexify the control designers’ job. Taking our recent work [4] of designing a semi-autonomous remote robotic system as an example, because no available wireless technology can support both high data rate sensing flow and real-time control flow at the same time, we have to use a combination of two wireless technologies, Wi-Fi and WirelessHART, which complicates the system design, prolongs the developing time and increases the maintenance cost. To address this problem, in this paper we introduce RTWiFi, a flexible real-time high speed communication protocol designed for supporting a wide range of wireless control systems. We keep the following goals in mind when we design the protocol: Real-time Data Delivery and High Sampling Rate: Control applications rely on bounded latency for estimating and controlling the state of target system. Our previous research [5] shows that a control system usually prefers a sampling rate higher than 1kHz. Thus, RT-WiFi is based on 802.11 physical layer. To achieve deterministic timing behavior, we adopt TDMA mechanism in the RT-WiFi data link layer design for channel access and set stringent timing requirement on the transactions in each time slot. Flexible Data Link Layer Configuration: Different control applications may have different communication requirements on data delivery, it is hard to apply a rigid design to a wide range of applications. We envision to design RT-WiFi as a configurable platform, such that control designers have great flexibility to choose their own design parameters according to their target applications. These design tradeoff include sampling rate, energy efficiency, communication reliability, real-time data delivery, and co-existence to existing Wi-Fi networks. Transparent System Design: In order to make RT-WiFi easily available and simple to integrate with the existing applications, it should reuse current hardware and make minimum changes on softwares. Users should be able to use RT-WiFi on the commercial off-the-shelf (COTS) IEEE 802.11 network interface, and existing application should be able to run on top of RT-WiFi with no or only minimum changes. Additionally, the transparent system design allows us to use existing 802.11 security techniques and build mesh networks easily. Our contributions in this paper are three-fold: • • • We designed the RT-WiFi protocol to provide a flexible real-time high speed wireless communication platform for a wide range of wireless control systems. According to their target applications, control designers can customize their design tradeoff among sampling rate, jitter level, reliability, and compatibility with regular Wi-Fi network. RT-WiFi supports up to 6kHz sampling rate in our selected COTS hardware, and to the best of our knowledge, this is among the highest sampling rates that can be supported by any existing wireless real-time communication protocols. We built a prototype for the RT-WiFi protocol on COTS hardware. We shared our first-hand experience in addressing the challenges we met during the RTWiFi implementation and discussed how to choose design parameters that helps users to customize RTWiFi depending on their target applications. We presented a comprehensive case study to utilize the RT-WiFi protocol to integrate heterogenous sensors and assistive robotic devices together to build a mobile gait rehabilitation system and evaluate the performance through extensive experiments. The remainder of this paper is organized as follows: In Section II we give a review on existing wireless protocols applied in wireless control systems. We present the system design of RT-WiFi in Section III, and describe the implementation details in Section IV. We describe the use case of a mobile gait rehabilitation system in Section VI. We conducted extensive performance evaluation on RT-WiFi and summarize our experimental results in Section V. Section VII concludes the paper and discusses the future work. II. Related Works There are some well-developed wireless protocols which can be used for wireless control. They include Bluetooth [6] and Zigbee [7] for home automation, WirelessHART [1] and ISA100.11a [8] for industrial process control, MBStar [9] for body area networks and so on. However these protocols are not perfect for high speed wireless control. Bluetooth [6] cannot guarantee real-time data delivery, except for voice packets. Zigbee [7] can provide real-time data delivery in beacon-enabled mode, but the data rate is limited to 250kbps. WirelessHART [1] is a Time division multiple access (TDMA)-based wireless protocol, which is specially designed for industrial process control. It can provide realtime communication, but the maximum sampling frequency is only 100Hz. ISA100.11a [8] is a another wireless protocol which is developped by International Society of Automation (ISA) to support industrial wireless control neeeds. Similar to the problem of WirelessHART, the data rate of ISA100.11a is low. MBStar is designed to provide higher sampling rate than WirelessHART for process control, but the highest rate it can reach is 400Hz, which is still far away from the preferred sampling rate for some applications. Zigbee, WirelessHART ,MBStar and [10], [11] cannot provide high enough sampling rate because their physical layer (IEEE 802.15.4 [12]) is designed for low rate wireless personal network. Comparatively Wi-Fi, based on IEEE 802.11 [13], is a wireless protocol designed for high speed wireless local area newtork and can have data rate up to 150Mbps (IEEE 802.11n). However, Wi-Fi does not provide any realtime guarantee on data delivery. Wi-Fi normally operates in distributed coordination function (DCF) mode, in which each station use distributed randomized algorithm to access the channel to avoid collision. Thus the delivery time of data is not deterministic. In Wi-Fi, there is another optional access method: Point coordination function (PCF) [13] In PCF, the access point (AP) operates as a point coordinator (PC) and determines which station has the right to transmit the data. Unlike DCF, PCF is a centralized channel access method. PC divides the interval between two beacon frames into two periods: contention free period and contention period. In contention free period, PC coordinates the data transmission and in contention period, stations use DCF to access the channel. Although PCF can be used to schedule packet transmission, it cannot rely on any timing information, and thus does not provide real-time data delivery guarantee. Also, once a station get the access to the channel, it may occupy the channel for a non-deterministic time. Thus, it is not sure when the subsequent packets will be sent. TDMA is a channel access method which relies on accurate timing information and may potentially provide real-time data delivery. Some TDMA protocol have been proposed [14], [15], [16], [17]based on IEEE 802.11. However, both of [14], [15] are focused on the time synchronization issue of multi-hop network. They are not targeted at providing flexible platform for control application, and they does not consider the coexistence with regular Wi-Fi network. On the other hand, the purpose of [16], [17] is to provide efficient long range wireless network in rural areas. It is because the original CSMA/CA EĞƚǁŽƌŬ DĂŶĂŐĞƌ ŽŶƚƌŽů ƉƉůŝĐĂƚŝŽŶ ZdͲtŝ&ŝ ĐĐĞƐƐWŽŝŶƚ ƉƉůŝĐĂƚŝŽŶ ŽŶƚƌŽůƉƉůŝĐĂƚŝŽŶƐ dƌĂŶƐƉŽƌƚ hWͬdW EĞƚǁŽƌŬ /W ZdͲtŝ&ŝ YƵĞƵĞƐ ZdͲtŝ&ŝ ^ƚĂƚŝŽŶϭ ZdͲtŝ&ŝ ^ƚĂƚŝŽŶϮ ZdͲtŝ&ŝ ^ƚĂƚŝŽŶϯ dŝŵĞƌ &ůĞdžŝďůĞŚĂŶŶĞů ĐĐĞƐƐŽŶƚƌŽůůĞƌ DĞƐƐĂŐĞ ,ĂŶĚůŝŶŐ DŽĚƵůĞ D ĐƚƵĂƚŽƌϭ ^ĞŶƐŽƌϭ ĐƚƵĂƚŽƌϮ ^ĞŶƐŽƌϮ ĐƚƵĂƚŽƌϯ ^ĞŶƐŽƌϯ DĞƐƐĂŐĞ ,ĂŶĚůŝŶŐ DŽĚƵůĞ Fig. 2: A example of RT-WiFi-based wireless control system >ŝŶŬ^ĐŚĞĚƵůĚĞƌ ^ƵƉĞƌĨƌĂŵĞ dĂďůĞ >ŝŶŬ dĂďůĞ ĞǀŝĐĞ WƌŽĨŝůĞ YƵĞƵĞ mechanism in Wi-Fi network is inefficient when Wi-Fi is used for outdoor long-distance communication. The main objective of them is to increase the throughput of the network, but not to provide high sampling rate real-time data delivery. III. RT-WiFi Protocol Design In this section, we describe the design details of RT-WiFi protocol. RT-WiFi is a data link layer protocol based on 802.11 PHY and aims to provide real-time data delivery for a wide range of wireless control systems from low-speed industrial process control to high speed mechanical device control. The design goals of RT-WiFi are to provide both realtime high sampling rate and application-dependent flexible data link layer configuration. Real-time data delivery is crucial to control systems because control applications rely on precise timing for monitoring and controlling the state of a system. High sampling rate is required to achieve good quality of control and to support ever-increasing number of sensors and actuators in a control systems. For instance, a communication protocol is required to support an aggregate 2KHz sampling rate for a control system with one sensor and one actuator, if both of them demands 1KHz sampling rate. For supporting a wide range of control applications, the design of RT-WiFi shall consider the various requirements of different types of control applications. The design philosophy of RT-WiFi is to provide enough freedom for control designer to choose their preferred communication behavior, which best fits to their existing controller design. At the same time, the design of RT-WiFi minimizes the modification on the original Wi-Fi protocol, so that it can be transparent to both the upper layer software stack and underlying hardware, and provide the most compatibility and usability. Figure 2 shows the typical architecture of a control system that adopts RT-WiFi network. In this case, a system architect plans to deploy three sensors and three actuators in this control system. Depending on the physical proximity of each sensor/actuator, they are attached to different RT-WiFi stations. In a RT-WiFi network, a network manager and control programs are running on top of RT-WiFi access point (AP). The network manager is software in the application layer that is responsible for configuring the RT-WiFi network based on the communication requirements specified by a control designer. The network manager dynamically allocates the communication links and generates network configuration to configure the data link layer of RT-WiFi AP and RT-WiFi stations. After the configuration stage, RT-WiFi network periodically transmit the sender data to and receive control data from the control application. In the WŚLJƐŝĐĂů /ϴϬϮ͘ϭϭŽŵƉĂƚŝďůĞ,ĂƌĚǁĂƌĞ Fig. 3: System architecture of RT-WiFi protocol following, we are focusing on the design and implementation of the RT-WiFi data link layer. Fig. 3 presents the overall architecture of RT-WiFi protocol. At the very bottom, RT-WiFi adopts the physical layer of IEEE 802.11 [13]. IEEE 802.11 is one of the most pervasively adopted wireless technologies and supports up to 150Mbps bandwidth which is sufficient for most wireless control systems. Control application users can easily deploy the RTWiFi data link layer on a commercial off-the-shelf IEEE 802.11 compatible hardware for supporting high speed and real-time data delivery. Sitting on top of the 802.11 PHY is the RT-WiFi data link layer which will be elaborated in the remaining of this section. The RT-WiFi data link layer provides a nice abstraction for the upper layers, thus standard UDP or TCP-based applications will be supported seamlessly. Various control protocols can be well supported by wrapping the their control payload in TCP/UDP packets. The core of the RT-WiFi protocol is the TDMA-based data link layer. Since there is no centralized channel access controller of a regular Wi-Fi network, it adopts carrier sense multiple access with collision avoidance (CSMA/CA) to avoid collision. This mechanism helps improve the network throughput but not feasible for supporting high speed real-time traffic with stringent timing requirements. As Fig. 4a shows, the transmission of the regular Wi-Fi packet with a hard deadline will be blocked by a nondeterministic of time because of carrier sense, and further delayed by random backoff mechanism. To address this problem, we adopts TDMA mechanism and a centralized channel and time management to access the channels according to strict timing. Fig. 4b presents the channel access pattern of a RT-WiFi network. Since only one node can access a certain channel in a given time slot, RT-WiFi provides collision free and deterministic communications. As shown in Fig. 3, there are three key components in the RT-WiFi data link layer: a timer for maintaining global synchronization among all RT-WiFi nodes and triggering timing events; a link scheduler for coordinating the access to shared media, and executing the dedicated event at scheduled time points; and a flexible channel access controller which dynamically configures the hardware parameters for executing the timing event according to the target application behavior. We shall explain the design of each of these components in ƵƐLJ DĞĚŝƵŵ /&^ ĞĨĞƌĐĐĞƐƐ ͙͘͘͘ ĂĐŬŽĨĨƐůŽƚƐ Đϭ /&^ ĂƚĂ ĞĨĞƌĐĐĞƐƐ ƚϭ ĐϮ ^ůŽƚϬ ĂƚĂ W ƌŽĂĚĐĂƐƚ ĂĐŬŽĨĨ ƐůŽƚƐ ƚϮ Đϭ͕ĐϮ͗ĚĂƚĂƌĞĂĚLJƉŽŝŶƚ ƚϭ͕ƚϮ͗ĚĂƚĂƚƌĂŶƐŵŝƚƉŽŝŶƚ (a) Regular Wi-Fi dŝŵĞ^ůŽƚ ĂƚĂ dŝŵĞ^ůŽƚ ĂƚĂ ^ŚĂƌĞĚ ^ůŽƚϮ ^ůŽƚϯ ^ůŽƚϰ ^ůŽƚϱ ^ůŽƚϲ ^ůŽƚϳ ^dϭ W ^dϮ W ^dϯ W W ^dϭ W ^dϮ W ^dϯ Fig. 5: An example of a superframe with 8 time slots. There are three key data structures maintained in the link scheduler: dŝŵĞ^ůŽƚ ĂƚĂ (b) RT-WiFi Fig. 4: Comparison of media access methods between regular Wi-Fi and RT-WiFi the following sections. A. Timer Design To achieve high sampling rate, RT-WiFi has very stringent timing requirements on each device. For instance, if the required aggregate sampling rate is 5kHz, the timer is required to deliver an accurate timer interrupt for every 200µs, and all tasks related to that sample (including data transmission, acknowledgement and possible in-slot retransmission) shall be completed within that short interval (called time slot). A time slot is a rigid channel access unit in the RT-WiFi network. In order to provide better timing predictability for the target applications, each RT-WiFi node can only access the channel in their pre-assigned time slots. Depending on the current sampling rate, the size of a time slot is required to be larger than the inverse of the sampling rate. Fig. 6a shows a generic RT-WiFi slot timing, and the slot size is determined by Equation 1. The guard interval is used to ensure the consecutive transmission does not interfere with each other due to synchronization inaccuracy. Data transmission is followed by the guard interval, and the transmission time depends on the transmission rate and packet size. When a station successfully receives the data, it should response an acknowledgement (ACK) message after short inter-frame space (SIFS). If sender does not receive ACK after the ACK wait period, then it assumes the packet is lost. T T imeS lot ≥ TGuardInterval + T Data + T S IFS + T ACK ^ůŽƚϭ (1) Our timer design is based on the Timing Synchronization Function (TSF) in IEEE 802.11. TSF is achieved by having all nodes keep a local 1MHz TSF timer, and synchronize to the master clock in RT-WiFi Access Point periodically through the beacon frames. The timer then triggers accurate time interrupts to ensure the correct operation of the data link layer. We shall present the implementation details of our timer module on WiFi chips in Section IV-B. B. Link Scheduler If accurate time synchronization with the master clock in the Access Point is achieved, every station in the RT-WiFi network forms a consensus to a global time. Based on this globally synchronized timing information, a link scheduler is designed to coordinate the channel access among the stations. Link: A link describes the communication behavior within a time slot. It is specified by its source, destination and the associated link type. We define four link types in RT-WiFi: transmit, receive, broadcast and shared. A broadcast link is used to transmit a data frame to all the neighbor nodes in the network; a transmit link is used for dedicated data frame transmission to the given destination; a receive link is used for receiving a data frame from the given source; and a shared link is for multiple source nodes to compete for transmitting a data frame to the same destination. Notice that, RT-WiFi is designed to be compatible to existing Wi-Fi systems. A RT-WiFi node can support both realtime and non-real-time traffic simultaneously. Link scheduler shall differentiate the priority of different traffics and assign them to either dedicated or shared links. Superframe: Superframe is a sequence of consecutive time slots. It defines the device’s communication pattern with neighbors and the superframe will repeat itself infinitely. Fig. 5 illustrates an example of a superframe configured in a RTWiFi AP with 8 time slots (the time slot length is 250µs). In this RT-WiFi network 3 stations are present, and each station is configured with a pair of dedicated transmit and receive link to the AP in the superframe. Device Profile: RT-WiFi Access Point keeps the device profile for the centralized network manager to generate superframe and links for each RT-WiFi station. The device profile is gathered when a RT-WiFi station joins a RT-WiFi network, and the device information is forwarded to network manger via RT-WiFi AP. The network manger installed in the RT-WiFi AP performs access control, generates the schedule information by looking up the sampling rate requirement of the new joined device that is specified by control designer, and reply the superframe and link configuration for that station. The network manager design and schedule construction is out of the scope of this paper and will be briefly discussed in the future work. To execute the TDMA schedule specified by the superframe, link scheduler is invoked by every timer interrupt. The link type of current time slot is determined by looking up the superframe using the timing information. Depending on the link type, link scheduler either requests a frame for the destination node from the message handling module or redirect the receiving frame to the upper layer. Before put the transmit frame into the physical layer, link scheduler setup the transmission parameters by using the flexible using channel access controller, which shall be described in the following section. C. Flexible Data Link Layer Design To support a wide range of control applications, we apply a flexible design in RT-WiFi data link layer. RT-WiFi allows the available computation resources and the specific control applications. dŝŵĞ^ůŽƚ ĂƚĂ 'ƵĂƌĚ /ŶƚĞƌǀĂů ^/&^ < In-slot retry: The in-slot retransmission mechanism is shown in Fig. 6b. the retransmission is invoked immediately if the sender does not receive an ACK message from the receiver. The retransmission time of a in-slot retry should not exceed the length of a time slot. <tĂŝƚ (a) RT-WiFi timeslot with successful transmission dŝŵĞ^ůŽƚ ĂƚĂ 'ƵĂƌĚ /ŶƚĞƌǀĂů ĂƚĂ ^/&^ <tĂŝƚ Out-of-slot retry: If a packet is failed to transmit in one time slot, RT-WiFi node can retransmit it in the next available link. Notice that the retransmission heavily depends on the desired application behavior. For example, in the next available link, if new control/sensor data is available, then it does not make sense to retransmit the old data. < <tĂŝƚ (b) RT-WiFi timeslot with in-slot retransmission dŝŵĞ^ůŽƚ ŽƵŶĚĞĚ>ĞŶŐƚŚ EŽƌŵĂůtŝ&ŝdƌĂĨĨŝĐ /&^ ZdͲtŝ&ŝĂƚĂ ^/&^ < 'ƵĂƌĚ /ŶƚĞƌǀĂů (c) RT-WiFi slot timing when considering Wi-Fi traffic Fig. 6: RT-WiFi time slot design users to choose the following design parameters to achieve the expected application behavior. 1) Sampling rate: Higher sampling rate helps control applications perform more accurate monitoring on the target systems, which further improves the quality of control. Increasing the sampling rate, however is challenging because higher sampling rate requires smaller size of the time slot, which is heavily dependent on the following factors. Packet size and Data rate: Apparently, RT-WiFi can not afford to transmit packet if the transmission time of a packet if larger than the time slot size according to Eq. 1. IEEE 802.11 specifications allow the use of multiple transmission rates at the physical layer that utilize different modulation and coding schemes. For example, the 802.11g supports twelve data rate from 1Mbps to 54Mbps. Higher transmission rate provides higher potential throughput, but it is also less resilient to noise and easily prone to error [18]. A flexible data link layer design should be able to select the adequate rate adaptively which best fit the current channel condition and desired time slot size. Computational resource: Increasing the sampling rate of the TDMA data link layer will reduce the time slot size. For executing tasks in a shorter time, more computational resource are required. The maximum sampling rate supported by a RTWiFi node is limited by its computation capability. Synchronization accuracy: The guard interval in Fig. 6 is reserved for the drift between two RT-WiFi devices. The minimum slot size is related to the synchronization accuracy because the size of time slot depends on the size of the guard interval. We can reduce the guard interval size by utilizing more accurate clock or decrease the synchronization interval. 2) Reliability: RT-WiFi applies two retransmission mechanisms to improve the reliability of a communication link. These two mechanisms can be used either independently or combined together. To use which retransmission mechanism depends on 3) Co-existence with regular Wi-Fi network: Due to the pervasively existing Wi-Fi signal, the RT-WiFi data link layer design needs to take its co-existence performance with regular Wi-Fi network into consideration. With the presence of Wi-Fi networks in the operational environment, the RT-WiFi data link layer should keep real-time characteristics while minimizing its influence to surrounding regular Wi-Fi networks as well. We assume that a regular Wi-Fi node obeys the CSMA/CA channel access mechanism and the maximum consecutive channel access time of a regular Wi-Fi node can be bounded by a constant time. This can be achieved by restricting the maximum transmit unit (MTU) size in the MAC layer of Wi-Fi network, and limiting the lowest available transmission speed of that Wi-Fi network. This assumption must be hold because if any Wi-Fi node (i.e. a jammer) can occupy a channel and continuously transmit data for arbitrarily long time then there is no way to achieve real-time data delivery. If this assumption is valid, then RT-WiFi can apply the following two mechanisms to improve its co-existence performance with regular Wi-Fi network. Performing Carrier Sense: Without the presence of WiFi traffic in the operation environment, a RT-WiFi node does not perform carrier sense in its assigned links before the transmission. This is because the network manager will make sure that there is no temporal or spatial reuse in the operating environment. However, if there are regular Wi-Fi networks operating in the surrounding environment, the co-existence performance between RT-WiFi and regular Wi-Fi is poor because the probability for the two types of traffic to collide with each other is high. In this case, RT-WiFi nodes should perform clear channel assessment and only start the transmission when the channel is clear. Shortening Inter-frame Spacing (IFS): IFS is an interval defined between the transmission of two consecutive Wi-Fi packets. Wi-Fi nodes are required to wait for a pre-defined IFS before they can start to transmit the next frame. To allow the RT-WiFi node has higher priority to access a channel, a shorter IFS can be set for RT-WiFi node so that they can start to transmit before regular Wi-Fi node. Fig. 6c illustrates the slot timing when we consider the co-existence of Wi-Fi and RT-WiFi traffic. The real-time data delivery still can be achieved because RT-WiFi only wait for a bounded time, and RT-WiFi node has a higher priority to occupy a channel. The price of improving the co-existence performance is to delay the data transmission of RT-WiFi traffic. We shall expect higher jitter in the sensor and control data if we enable this functionality, and it also limits the highest supported sampling rate because we need to reserve blocking time from Wi-Fi traffic. D. Association Process Before joins to a RT-WiFi network, a RT-WiFi node has no information about the TDMA schedule. It behaves like a regular Wi-Fi node, which uses CSMA/CA to access the channel, and it follows the same authentication and association process as regular Wi-Fi node to join a RT-WiFi network. After the association process is completed, a RT-WiFi node wait for the next beacon message, which consists of the TDMA scheduling information. The scheduling information consists of time slot size, superframe size, TSF value of the next superframe, and the TDMA schedule. After processing this message, RT-WiFi station synchronizes its local clock with the TSF value provided by RT-WiFi AP, configures timer interrupt, and starts to execute the TDMA schedule. Notice that the TDMA scheduling information is appended in the beacon frame by using vendor specific field [13], which retain the original format of a beacon frame. Thus, a regular Wi-Fi station can easily associate with a RT-WiFi AP without any modification. To support regular Wi-Fi station with RTWiFi network, the co-existence technique that mentioned in the preivous seciton should be enabled. Currently, a station can only directly associate with a RT-WiFi AP, which forms a star network. We will mention how to extend the network topology of RT-WiFi to a multi-hop mesh network in the future work. IV. System Implementation In this section, we present the implementation details of the RT-WiFi protocol on Atheros AR9285 [19] Wi-Fi chip. We first summarize our hardware and software platform, and then describe the challenges and our solutions when implement the key components in the RT-WiFi protocol. A. Hardware and Software Platform RT-WiFi does not rely on specific IEEE 802.11 hardware implementation, and our design should be able to port to any IEEE 802.11 compatible hardware. We choose Atheros AR9285 Wi-Fi chip for our proof of concept implementation because it is one of the most commonly used Wi-Fi chip and it has open source driver module in Linux operating system. AR9285 Wi-Fi chip supports IEEE 80211 b/g/n specification. AR9285’s maximum transmission rate is 150Mbps and it operates in the 2.4GHz frequency band. We use Ubuntu 12.04 as the operating system, which runs Linux kernel 3.2.0. On top of that, We adopt open source Linux compat-wireless driver [20] as the foundation of RTWiFi. Two software modules originated from compat-wireless driver, mac80211 and hardware-dependent driver module, are incorporated with the TDMA design to build the MAC layer of RT-WiFi. mac80211 is a framework for developing software based MAC driver, it is used for supporting the IEEE 802.11 frame management and serving as the interface between upper network layer and the message handling module of RT-WiFi MAC layer. We use hardware driver module ath9k for supporting AR9285, and a software interface is implemented for message handling and flexible channel access configuration. Since we only made very limited modification to create an interface between RT-WiFi module and the hardware-dependent modules for building RT-WiFi design, it will be easy to port the design of RT-WiFi to other IEEE 802.11 compatible hardware. B. Timer Module The basic time unit in RT-WiFi network is a time slot. Depending on the adopted version of IEEE 802.11 standard, the size of the time slot could be estimated by Equation 1. In the following, we give a concrete example by using IEEE 802.11g with ERP-OFDM modulation scheme. TGuardInterval depends on the accuracy of timing synchronization, which relies on the Timing Synchronization Function (TSF) in IEEE 802.11. According to the IEEE 802.11 standard [13], the drift rate of TSF should be within ±10ppm. Users can control the maximum amount of drift between AP and stations by controlling the broadcasting interval of TSF. For example if the beacon is broadcasting every 100ms, then the maximum drift is within 20µs. To determine the value of T Data , we assume the application payload is 200 bytes, UDP header is 8 bytes, and IPv4 header is 20 bytes. The MAC header is 32 bytes with 4 bytes frame check sequence. According to [21], the physical layer convergence procedure (PLCP) preamble and header delay are 20µs. If the data is transmitted by the maximum data rate (54Mbps), then (200 + 8 + 20 + 32 + 4) ∗ 8 = 59.11µs. (2) 54 ∗ 106 T S IFS is 10µs for 802.11g. The total size of ACK frame is 14 bytes. Similary, we can derive T ACK = 22.07µs. Theoretically the minimum slot size for the previous setting is T Data = 20 + T T imeS lot ≥ 20 + 59.11 + 10 + 22.07 = 111.18µs. (3) The size of the time slot for in-slot retry and co-existence with regular Wi-Fi can be calculated by similar manner. Based on TSF, timer sets its local hardware timer in AR9285 and triggers the timer interrupt for each time slot. Originally, the handler of this timer interrupt is deferred and handled by a kernel tasklet later. However, in our implementation, we observed that this tasklet could be blocked by other tasklet that has the same priority, which delayed the execution of time sensitive RT-WiFi task. In order to avoid this delay for meeting the stringent timing requirement of RT-WiFi, instead of deferring execution of RT-WiFi task to tasklet we directly executed it in interrupt service routine. Notice that this design reduces the delay for executing RT-WiFi task with the cost of increasing kernel response time. However, the average execution time of RT-WiFi task for Intel Atom N450 CPU is less 20µs, which only has limited influence to the kernel response time. C. Link Scheduler and Message Handling Module To implement TDMA-based channel access, we build two message handling modules in RT-WiFi for sending and receiving packets. To ensure efficient message handling, the transmission message module consists of multiple queues, one for each dedicated real-time communication link and one for W /ŶƚĞůŝϯϮϯϮϬD Ϯ͘Ϯ',nj ^dϭ /ŶƚĞůŝϳϮϲϮϬD Ϯ͘ϳ',nj ^dϮ /ŶƚĞůƚŽŵEϰϱϬ ϭ͘ϲϲ',nj ^dϯ /ŶƚĞůĞůĞƌŽŶD ϵϬϬD,nj (a) Setting 1 Wϱ /ŶƚĞůŝϳϮϲϮϬD Ϯ͘ϳ',nj Wϲ /ŶƚĞůŝϯϮϯϮϬD Ϯ͘Ϯ',nj ^dϱ /ŶƚĞůĞůĞƌŽŶD ϵϬϬD,nj ^dϲ /ŶƚĞůƚŽŵEϰϱϬ ϭ͘ϲϲ',nj EĞƚǁŽƌŬͲ EĞƚǁŽƌŬͲ (b) Setting 2 Fig. 7: An overview of the testbed setting (a) Interference free environment non-real-time communication link. The transmission message module accepts packet from mac80211 module, and will be controlled by Link Scheduler to transmit a scheduled packet to the lower layer at each time slot. The receiving message module enqueues messages to upper layer whenever it gets a message. Both real-time and non-real-time traffic could be present in a RT-WiFi node. In order to identify the traffic type in the data link layer, we utilize the type of service (TOS) field in the IP header. Application developer can create a socket with specified TOS, by using user level API i.e. setsockopt(). Then mac80211 module will map TOS to an access class. Currently, RT-WiFi maps the highest priority class to RT-WiFi queues, messages in that class will be sent by dedicated transmit links. For non-real-time packets, the link scheduler will transmit them by shared links. After the RT-WiFi associates with the AP, link scheduler configures the timer interrupt, and will start to transmit and receive messages according to the configured communication schedule. D. Flexible Channel Access Controller To achieve flexible channel access, we implement an adaptive channel access controller. The controller configures the hardware setting according to the user specified parameters. Changing sampling rate is done with modifying the hardware timer. AR9285 has a feature called retry series, which can be used to set a series of in-slot retry with a corresponding transmission rate. If out-of-slot retry is enabled, RT-WiFi will confirm the delivery of a frame by checking the ACK message. If the ACK message is missing, RT-WiFi will re-queue that frame into RT-WiFi queue. To enable co-existence mode with regular Wi-Fi network, we can manipulate the hardware register of AR9285 to control its carrier sense. AR9285 has eight hardware queues, and we can map the RT-WiFi traffic to one of the hardware queue. For configuring the IFS, we can set the IFS of the hardware queue for RT-WiFi traffic, which lets RT-WiFi traffic has higher priority to access a channel. V. Performance Evaluation To validate our design of the RT-WiFi protocol and evaluate its performance in providing high speed real-time wireless communication, we setup a testbed for the performance comparison between RT-WiFi and regular Wi-Fi. (b) Office environment Fig. 8: Experiment environment As shown in Fig. 8a, the testbed consists of one Access Point (AP) and three stations (STA1, STA2 and STA3) which form into a star network topology. All the four devices are equipped with the same Atheros AR9285 IEEE 802.11 compatible wireless interface but different CPUs with varied computation power. This is to demonstrate that the implementation of RT-WiFi only introduces limited computation overhead, and can even run on resource-constrained embedded devices. To emulate the sensing and control flows in a wireless control system, we install a UDP socket program in each device. Sensor data with a fixed size payload are transmitted from each station to the Access Point according to the configured sampling rate. On the other direction, the Access Point periodically transmits control data with the same packet size back to each station. Notice that sensors, actuators, and controllers are all running in the application layer. The overall delay from a sensor to a controller or from a controller to an actuator includes the application layer to MAC layer delay in each device and the MAC layer to MAC layer delay between the devices. In this section we focus on evaluating the MAC layer performance. We will evaluate the application layer performance in Section VI through a case study. We compare the MAC layer to MAC layer performance between RT-WiFi and regular Wi-Fi in two test scenarios, an interference-free environment and an office environment with intensive Wi-Fi traffic. The main performance metrics we used in the experiments include the data link layer transmission latency, and the packet loss ratio. The data link layer transmission latency is calculated as the difference of a frame’s TSF timestamps between the receiver side and the sender side; The packet loss ratio measures the percentage of packets lost by tracking the sequence number of each packet. A. Interference-free Environment To create an interference-free environment for performance comparison, we deploy the testbed in a parking lot of a hiking trail at Austin city suburban area around Texas State Highway Loop 360 and Loop 1. Table I summarizes the comparison results between RT-WiFi and regular Wi-Fi in this test scenario. The results include the max, mean and standard deviation of the MAC layer to MAC layer transmission delay, and the packet loss ratio for each of the six links. The empirical cumulative density function (ECDF) of the transmission latency for uplinks (STAs to AP) and downlinks (AP to STAs) are presented in Figure 9. We observed from the experiment results that RT-WiFi protocol can provide accurate real-time packet delivery that more than 99.98% of packets are delivered within the assigned 500µs time slot and the standard deviation of the latency in RT-WiFi network is less than 5.3µs. The occasional packet loss is caused by the interference from other visitors’ mobile devices. This is verified by observing uncontrolled probe request frames from unknown devices to RT-WiFi AP when visitors are nearby. On the contrast, Wi-Fi network has higher average delay and larger transmission variation. This is because four nodes in the Wi-Fi network compete for media access by utilizing CSMA-CA and random backoff mechanisms. The mean delay of downlinks are higher than that of uplinks is because the packets sent from AP not only have to wait for channel access but also are delayed due to the queuing effect. RTWiFi network is seldom affected by the queuing effect. This is because to delay a packet transmission in the AP, one packet needs to be deferred for more than two time slots (1000µs) according to our configured link schedule. As shown in Fig. 9, only less than 0.01% of packets in RT-WiFi network has latency larger than 1000µs. The latency variation in Wi-Fi network is up to 90 times to that of RT-WiFi network, and the maximum delay is more than 30 times to RT-WiFi, which poses great challenges for the controller design. Another observation from Fig. 9a is that the minimum latency of RT-WiFi on the uplinks is higher than that of regular Wi-Fi. This is due to the delay of the TDMA guard interval and the implementation overhead in the MAC layer software module. However, as Fig. 9b shows, the difference Empirical CDF 1 Wi−Fi RT−WiFi 0.8 0.6 0.4 0.2 0 0 500 1000 1500 2000 2500 Latency (µs) 3000 3500 4000 (a) Latency ECDF for uplinks (STAs→AP) 1 Empirical CDF The following configuration is used in our experiments. We run each experiment for 600 seconds. Both the Wi-Fi and RT-WiFi are configured to use channel 6 of IEEE 802.11 b/g/n. There are in total six pairs of UDP sockets in the experiment setting between the stations and the Access Point (STA1→AP, STA2→AP, STA3→AP, AP→STA1, AP→STA2 and AP→STA3). For each UDP socket, we publish the data every 4ms. The application layer payload is fixed as 460 bytes. For RT-WiFi network, we set the time slot size as 500µs, and we use the same superframe and link schedule as shown in Fig. 5. Due to the limit of space, in this paper we only compare the regular Wi-Fi with the baseline of RTWiFi protocol by fixing the data rate as 54Mbps, and disabling both the retransmission and co-existence mechanisms. For the more comprehensive comparison between regular Wi-Fi and RT-WiFi, please refer to our technical report. Wi−Fi RT−WiFi 0.8 0.6 0.4 0.2 0 0 500 1000 1500 2000 2500 Latency (µs) 3000 3500 4000 (b) Latency ECDF for downlinks (AP→STAs) Fig. 9: MAC layer transmission latency comparison between Wi-Fi and RT-WiFi in an interference-free environment of the minimum latency between regular Wi-Fi and RT-WiFi is smaller for the downlinks. This is because the minimum latency of regular Wi-Fi is increased by queuing delay. B. Office Environment To evaluate the performance of RT-WiFi network with presence of interference, we also deploy the testbed on the 5th floor in UT GDC building. This floor is covered with more than 10 Wi-Fi APs, which spread to all the orthogonal Wi-Fi channels in ISM 2.4GHz band. Under this setting, we repeat the same experiments as in the interference-free environment. Fig. 10 compares the MAC layer latency between RT-WiFi and Wi-Fi network in the office environment. Compared with the interference-free environment, the latency of regular Wi-Fi network is increased because the office environment has more inference from existing Wi-Fi network. Thus, the CSMA/CA and random backoff mechanism has higher chance to defer the transmission of regular Wi-Fi packets. The latency distribution of RT-WiFi network in Fig. 10 is similar to Fig. 9, because we adopts the same TDMA design, which targets to provide small latency variation. Table II shows the latency and packet loss ratio of RT-WiFi network and Wi-Fi network. The maximum latency of RT-WiFi network is increased up to 4.2ms because there are more uncontrolled mobile devices around in the office environment. The loss ratio of RT-WiFi network is raised up to 10% due to the collision to background traffic. However, because of the TDMA design of RT-WiFi network, the mean delay in the office environment remains similar to the interference-free environment, which will be convenient for control designer to reuse their controller design. Moreover, the delay variation of RT-WiFi is still small (less than 30µs), which makes control application easy to compensate the packet loss. Link STA1→AP STA2→AP STA3→AP AP→STA1 AP→STA2 AP→STA3 Max Delay(µs) RT-WiFi Wi-Fi 535 16448 529 13387 525 13589 827 16472 544 17465 1055 17049 Mean Delay (µs) RT-WiFi Wi-Fi 173 191 172 181 174 202 184 250 187 298 188 248 Delay Standard Deviation (µs) RT-WiFi Wi-Fi 4.88 231.60 2.52 214.15 3.01 236.75 5.29 342.24 3.98 360.66 4.87 325.50 Loss Ratio RT-WiFi Wi-Fi 0.19% 0% 0.16% 0% 0.21% 0% 0.06% 0% 0.04% 0% 0.01% 0% TABLE I: Latency and loss ratio comparison in an interference-free environment Empirical CDF 1 Wi−Fi RT−WiFi 0.8 0.6 0.4 0.2 0 0 500 1000 1500 2000 2500 Latency (µs) 3000 3500 4000 (a) Latency ECDF for uplinks (STAs→AP) Empirical CDF 1 Wi−Fi RT−WiFi 0.8 0.6 0.4 0.2 0 0 500 1000 1500 2000 2500 Latency (µs) 3000 3500 4000 (b) Latency ECDF for downlinks (AP→STAs) Fig. 10: MAC layer transmission latency comparison between Wi-Fi and RT-WiFi in an office environment Regular Wi-Fi RT-WiFi baseline RT-WiFi co-ex RT-WiFi co-ex + retry Max Delay (µs) 62629 953 2507 2831 Mean Delay (µs) 580 183 220 254 Delay Standard Deviation (µs) 1679.04 27.75 149.27 166.37 Loss Ratio 0% 50.21% 10.92% 4.96% TABLE III: Latency and loss ratio comparison with present of large network traffic On the other hand, Wi-Fi network has zero packet loss because it keeps retransmitting until the data is delivered. However, retransmitting old sensor data does not give effective system information if new sensor data is available. According to Table II the maximum delay of regular Wi-Fi network is up to 100ms, and the standard deviation is increased to 2800µs. The large delay of sensor data will make controller out track of the system state, and large latency for delivering control data could also makes target system out of control for a long period. Also the high latency variance of Wi-Fi network could be very challenging for controller design. C. Flexible Channel Access Controller To evaluate the performance of flexible channel access controller, we setup a testbed in the same office environment with two networks as shown in Fig. 8b. Network-A is a regular Wi-Fi network, which utilizes network traffic generator, iperf [?], to generate a 10Mbps UDP traffic from STA5 to AP5. The MAC layer maximum transmission unit of Wi-Fi network is configured as 300 bytes, for bounding the packet transmission time. For Network-B, we run the same UDP socket program, which publishes data every 4ms from AP6 to STA6. We configured Network-B by using four settings, regular Wi-Fi, RT-WiFi baseline, RT-WiFi with co-existence, RT-WiFi with co-existence and one in-slot-retry, and observe the latency and packet loss ratio for link (STA6→AP6). Table III shows the experiment results. Because of the interference from Network-A, the mean delay of regular Wi-Fi is increased to 580µs, which is 47% higher than the average mean delay of the upload links in the previous experiment. The packet loss ratio of baseline RT-WiFi network is also increased dramatically to 50.21%, due to the large interference from Network-A. For RT-WiFi network with co-existence mode, the loss ratio is decreased to 10.92%. It is because we enable the carrier sense mechanism, but the carrier sense can not detect potential hidden terminals. The variance also increases to 149.27 since the transmission may be blocked by traffic from Network-A. The loss ratio is further decreased to 4.96% when we enable in-slot-retry, which also increases the mean delay and delay variance. VI. Case Study - A Mobile Gait Rehabilitation System To demonstrate the benefit of applying RT-WiFi in wireless control systems and applications, we built a networkedbased rehabilitation system with RT-WiFi integrated. In the proposed system, RT-WiFi network serves as the wireless communication links between control applications and the rehabilitation device for improved system mobility and telerehabilitation. Control algorithms run in the application layer of a host computer, which also serves as RT-WiFi AP, and the rehabilitation device is connected to the RT-WiFi station. Fig. 11 gives an overview of the network-based rehabilitation system, which consists of two smart shoes [22] with embedded air press sensors, multiple IMU motion sensors, a robotic assistive device [23] and a host computer running control applications. There are two types of wireless communication links in the system, all based on RT-WiFi. One is used to transmit sensing signals from sensing devices (air pressure sensors and IMU sensors) to the control applications in the host computer [24]. Information on this wireless link is used for health monitoring and rehabilitation strategy design, so the sampling rate can be chosen from 100Hz to 1kHz depending on the high-level decision making algorithms (100Hz is chosen in the current design). The other type of wireless link is for controlling the robotic assistive device, which requires at least 1kHz sampling rate for precise motion control. More details of the system design can be found in [2]. Link STA1→AP STA2→AP STA3→AP AP→STA1 AP→STA2 AP→STA3 Max Delay(µs) RT-WiFi Wi-Fi 3865 10078 4193 81499 3861 75298 1197 78089 1342 78923 2186 77860 Mean Delay (µs) RT-WiFi Wi-Fi 176 401 171 348 174 429 184 788 189 790 189 799 Delay Standard RT-WiFi 25.86 27.62 25.16 16.86 15.19 19.03 Deviation (µs) Wi-Fi 1491.69 1000.60 1221.72 2861.42 2806.56 2855.89 Loss Ratio RT-WiFi Wi-Fi 8.64% 0% 9.97% 0% 7.55% 0% 9.52% 0% 8.09% 0% 9.31% 0% TABLE II: Latency and loss ratio comparison in an office environment achieve this synchronization, we utilize IEEE 1588 Precision Time Protocol (PTP) [26] by using its software implementation PTP daemon [27], and using a dedicate PCI Express Ethernet link for PTP. The synchronization error between the two system clocks was measured to be under 40µs. This was precise enough to measure the end-to-end application layer latency and compare the control performance between RT-WiFi and regular Wi-Fi. B. Emulation of a Wireless Control System Fig. 11: An overview of the network-based rehabilitation system Fig. 12: Integration of smart shoes with a RT-WiFi station. A. System Integration We take the smart shoe as an example to show how we integrate the sensing and control devices into the rehabilitation system through RT-WiFi. Smart shoe is a human gait detecting device, which has been developed for health monitoring from our previous research. Our network-based gait rehabilitation system periodically requests sensing signals from air pressure sensors for abnormal gait detection. There are four air pressure sensors embedded in each shoe, and sensing signal from each sensor is encapsulated as 8-byte packet. Including a 8-byte time index, we have 72 bytes of data in total for two shoes to be transmitted in each time step. Fig. 12 shows how we integrated smart shoes with RT-WiFi network. The real-time sensor data were first collected from an analog-input module NI 9221 [25] on a NI 9116 compactRIO chassis, and then sent through a UDP socket from an Ethernet port of the NI 9022 real-time controller on the compactRIO chassis to a RT-WiFi station. The RT-WiFi station then forwarded the sensor data through the RT-WiFi wireless communication link to the RT-WiFi AP on which the controller was running. To evaluate the performance of the network-based rehabilitation system, we need to measure the application layer latency between the smart shoes and the RT-WiFi AP, which relies on a precise system clock synchronization between them. To To better demonstrate the performance of the RT-WiFi wireless communication protocol, we ran some simulations based on the data traces we collected through smart shoes. Note that in the real control system, there should be two links for transmitting both sensing and control signals as shown in Fig. 11. As the first step of wireless control system integration, we integrated smart shoes to acquire the network dynamics (latency and packet loss ratio), and then ran some simulations to evaluate the performance of the wireless control system before we integrated the robotic hardware. The model of our robotic device in this networked rehabilitation system was used for simulation. We ran both regular Wi-Fi and RT-WiFi network at 1kHz for 60 seconds in an in-door office environment. At each time step, we recorded whether the sensing packet was successfully transmitted to the wireless AP. If so, we recorded the latency between the application layer of wireless station and wireless AP. In this simulation, the control signal was implemented every 1ms, and a sensing packet would be ignored if its latency was larger than 1ms. To compare the performance of wireless network for control system applications, we define the effective packet loss ratio (EPLR) as follows: EPLR = Nl + Nd × 100%, N (4) where Nl is the number of lost packet, Nd is the number of packets with latency longer than the sampling interval, and N is the total number of packets transmitted over a certain period. A proportional-derivative (PD) controller was implemented as the main controller, and the wireless control system was asked to track a sinusoidal signal with unit magnitude. Simulations were conducted with different controller gains and reference frequencies to test the effectiveness of network dynamics. The root-mean-square (RMS) error are given in Table IV. Empirical cumulative distribution functions (ECDFs) of tracking errors are given in Figs.13 to 15. Based on our simulation, the EPLR for regular Wi-Fi is 26.07%, and the EPLR for RT-WiFi is 9.5%, which is 63.56% RMS Error (deg) f = 1Hz RMS Error (deg) f = 2Hz RT-WiFi 1.633 0.796 0.160 7.236 3.323 0.636 Wi-Fi Improvement 2.097 1.037 0.257 9.872 4.526 1.140 1 Emperical CDF Controller Gain Kp Kd 0.5 0.005 1 0.01 5 0.05 0.5 0.005 1 0.01 5 0.05 22.13% 23.24% 37.74% 26.70% 26.58% 44.21% 0.6 0.4 0.2 0 TABLE IV: RMS tracking error in simulations WiFi RT-WiFi 0.8 0 1 2 3 4 5 Tracking Error (deg) 6 7 8 (a) ECDF of tracking error with 1Hz reference 1 WiFi RT-WiFi 0.8 Emperical CDF Emperical CDF 1 0.6 0.4 0.2 0 0 2 4 6 8 10 Tracking Error (deg) 12 14 16 WiFi RT-WiFi 0.8 0.6 0.4 0.2 0 0 2 4 6 8 10 Tracking Error (deg) (a) ECDF of tracking error with 1Hz reference 12 14 16 (b) ECDF of tracking error with 2Hz reference Fig. 14: ECDF of tracking error with K p = 1, Kd = 0.01 WiFi RT-WiFi 0.8 1 0.6 0.4 0.2 0 0 2 4 6 8 10 12 14 16 Tracking Error (deg) 18 20 22 24 26 Emperical CDF Emperical CDF 1 (b) ECDF of tracking error with 2Hz reference WiFi RT-WiFi 0.8 0.6 0.4 0.2 0 0 0.5 1 Fig. 13: ECDF of tracking error with K p = 0.5, Kd = 0.005 1.5 2 Tracking Error (deg) 2.5 3 (a) ECDF of tracking error with 1Hz reference VII. Conclusion and Future work In this paper, we present the design principles and implementation details of a real-time wireless communication protocol called RT-WiFi to support high speed wireless control systems. RT-WiFi is a TDMA data link layer protocol based on 802.11 physical layer to provide high sampling rate and deterministic timing guarantee on packet delivery in wireless control systems. It incorporates configurable components for adjusting design tradeoff including sampling rate, real-time performance, communication reliability, and compatibility to existing Wi-Fi networks. RT-WiFi can be implemented on commercial offthe-shelve hardware and supports up to 6kHz sampling rate. 1 Emperical CDF lower than regular Wi-Fi. As we can see from Table IV that compared with regular Wi-Fi, RT-WiFi consistently yields much smaller RMS tracking errors under different settings of the controller gains. This demonstrates its significant improvement on control performance compared with regular Wi-Fi. From Figs.13 to 15, we can see that the probability to achieve small tracking errors is always higher for RT-WiFi than regular Wi-Fi. Moreover, the system is more sensitive to packet loss when the controller gain is smaller, and larger tracking errors occur when we ask the system to tracking faster reference signals. In clinic test, to protect patients from injury, we need to make sure that the position tracking error is no larger than 10 degrees at all times. Due to the limitation of actuator power, we also prefer small control gains (e.g.,K p = 0.5, Kd = 0.005). In this setting, only RT-WiFi can be employed. WiFi RT-WiFi 0.8 0.6 0.4 0.2 0 0 0.5 1 1.5 2 2.5 3 Tracking Error (deg) 3.5 4 4.5 5 (b) ECDF of tracking error with 2Hz reference Fig. 15: ECDF of tracking error with K p = 5, Kd = 0.05 We integrate RT-WiFi into a mobile gait rehabilitation system. Our results of extensive experiments validate our design and demonstrate the significant improvement from RT-WiFi on both data link layer and application layer latency in wireless control systems compared with regular Wi-Fi, which further lead to easier controller design and better control performance. As the future work, we are working towards reducing the form factor of the RT-WiFi communication module, addressing the relatively high power consumption issue, developing a resource management and scheduling framework for RT-WiFi system to extend the network topology to mesh structure and support multi-hop real-time sensing and control flows to meet end-to-end delay constraints. With all these pieces in place, we envision that RT-WiFi will serve as an ideal communication platform for a wide range of real-time wireless control systems. [21] Acknowledgment This work is supported by National Science Foundation under Grant CMMI-1013657. [22] [23] References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] J. Song, S. Han, A. K. Mok, D. Chen, M. Lucas, M. Nixon, and W. Pratt, “WirelessHART: Applying wireless technology in real-time industrial process control,” in Real-Time and Embedded Technology and Applications Symposium (RTAS). IEEE, 2008, pp. 377–386. W. Zhang, X. Zhu, S. Han, N. Byl, A. K. Mok, and M. Tomizuka, “Design of a network-based mobile gait rehabilitation system,” in IEEE International Conference on Robotics and Biomimetics (ROBIO). IEEE, 2012, pp. 1773–1778. B. Li, Z. Sun, K. Mechitov, G. Hackmann, C. Lu, S. J. Dyke, G. Agha, and B. F. Spencer, “Realistic case studies of wireless structural control,” in International Conference on Cyber-Physical Systems (ICCPS), 2013. S. Han, A. K. Mok, J. Meng, Y.-H. Wei, P.-C. Huang, Q. Leng, X. Zhu, L. Sentis, K. S. Kim, and R. Miikkulainen, “Architecture of a cyberphysical avatar,” in International Conference on Cyber-Physical Systems (ICCPS), 2013. Y.-H. Wei, Q. Leng, S. Han, A. K. Mok, W. Zhang, M. Tomizuka, T. Li, D. Leith, and D. Malone, “RT-WiFi: Real-time high speed communication protocol for wireless control systems,” in Real-Time Systems Symposium Work-in-Progress Session (RTSS WiP), 2012. “Bluetooth,” http://www.bluetooth.com. “Zigbee Alliance,” http://www.zigbee.org. “ISA100,” http://www.isa.org/isa100. X. Zhu, S. Han, P.-C. Huang, A. K. Mok, and D. Chen, “Mbstar: A real-time communication protocol for wireless body area networks,” in Euromicro Conference on Real-Time Systems (ECRTS). IEEE, 2011, pp. 57–66. V. Gabale, B. Raman, K. Chebrolu, and P. Kulkarni, “LiT MAC: addressing the challenges of effective voice communication in a low cost, low power wireless mesh network,” in Proceedings of the First ACM Symposium on Computing for Development. ACM, 2010, p. 5. B. Raman, K. Chebrolu, S. Bijwe, and V. Gabale, “PIP: a connection-oriented, multi-hop, multi-channel TDMA-based MAC for high throughput bulk transfer,” in Proceedings of the 8th ACM Conference on Embedded Networked Sensor Systems. ACM, 2010, pp. 15–28. “IEEE 802.15 WPAN Task Group 4,” http://www.ieee802.org/15/pub/ TG4.html. “IEEE 802.11 working group,” http://www.ieee802.org/11/. D. Koutsonikolas, T. Salonidis, H. Lundgren, P. LeGuyadec, Y. C. Hu, and I. Sheriff, “TDM MAC protocol design and implementation for wireless mesh networks,” in Proceedings of the 2008 ACM CoNEXT Conference. ACM, 2008, p. 28. P. Djukic and P. Mohapatra, “Soft-TDMAC: A software TDMAbased MAC over commodity 802.11 hardware,” in IEEE International Conference on Computer Communications (INFOCOM). IEEE, 2009, pp. 1836–1844. A. Dhekne, N. Uchat, and B. Raman, “Implementation and evaluation of a TDMA MAC for wifi-based rural mesh networks,” in Proceedings of the 4th ACM workshop on networked systems for developing regions. ACM, 2009. Y. Ben-David, M. Vallentin, S. Fowler, and E. Brewer, “JaldiMAC – taking the distance further,” in Proceedings of the 4th ACM workshop on networked systems for developing regions. ACM, 2010, p. 2. M. Lacage, M. H. Manshaei, and T. Turletti, “IEEE 802.11 rate adaptation: a practical approach,” in Proceedings of the 7th ACM international symposium on Modeling, analysis and simulation of wireless and mobile systems. ACM, 2004, pp. 126–134. “Atheros AR9285,” http://www.qca.qualcomm.com/media/product/ product 79 file1.pdf. “Linux Compat-wireless Driver,” http://wireless.kernel.org/. [24] [25] [26] [27] D. Vassis, G. Kormentzas, A. Rouskas, and I. Maglogiannis, “The IEEE 802.11 g standard for high data rate WLANs,” Network, IEEE, vol. 19, no. 3, pp. 21–26, 2005. K. Kong and M. Tomizuka, “A gait monitoring system based on air pressure sensors embedded in a shoe,” IEEE/ASME Transactions on Mechatronics, vol. 14, no. 3, pp. 358–370, 2009. K. Kong, J. Bae, and M. Tomizuka, “A compact rotary series elastic actuator for human assistive systems,” IEEE/ASME Transactions on Mechatronics, vol. 17, no. 2, pp. 288–297, 2012. J. Bae, K. Haninger, X. Wai, D.and Garcia, and M. Tomizuka, “A network-based monitoring system for rehabilitation,” in Advanced Intelligent Mechatronics (AIM), 2012 IEEE/ASME International Conference on. IEEE, 2012, pp. 232–237. “National Instruments,” http://www.ni.com/compactrio/. J.C. Eidson, Measurement, control, and communication using IEEE 1588, Springer Publishing Company, Incorporated, 2010. “Precision Time Protocol Daemon,” http://sourceforge.net/projects/ ptpd/.