A Feedback-Based Power Control Algorithm Design for VANET
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
A Feedback-Based Power Control Algorithm Design for VANET Xu Guan*, Raja Sengupta*, Hariharan Krishnan**, Fan Bai** *Systems Program, CEE, University of California, Berkeley, **General Motors R & D Center
Abstract— We consider the problem of adjusting transmission power for vehicle-to-vehicle broadcast safety communication in vehicular ad hoc networks. Given a target communication range designated by a vehicle safety application, the power control algorithm is designed to select a transmission power no greater than necessary for the targeted range. The power control algorithm results in higher communication reliability since collisions are minimized for safety communications. Our main idea is to add a power tuning feedback beacon during each safety message exchange. Our simulation results show that the power control algorithm can reduce the transmission power significantly to achieve higher packet reception rate than the case without any power control. This algorithm is mainly applied to communications-based vehicle safety applications but could be useful for other applications as well. Keywords- Power control, VANET, V2V
I. INTRODUCTION In this paper, we propose an adaptive transmission power control algorithm for vehicle-to-vehicle (V2V) communications-based safety applications. There is a well established effort in government and industry to develop vehicle safety communication (VSC) system, which uses broadcasts of safety related information from nearby vehicles to enhance vehicle safety[1,2,3]. These broadcast safety messages are periodic in nature, and contain information such as position, speed, heading, braking profile, etc., enabling the receiving vehicle to detect hazardous motions by another vehicle, thereby alerting the driver of that situation. Given the critical safety nature, these messages need to be delivered reliably and timely to a desired range. For example, if a vehicle is stopped in the middle of a highway, it needs to send safety messages to vehicles that might be approaching it from behind. Since the comfortable stopping distance of a vehicle at freeway speeds could be about 250 meters, the stopped vehicle seeks to broadcast its safety messages to a desired range of 300 meters. Similar performance requirements for other VSC applications are listed in the literature [1,2,4]. We assume that these safety broadcast messages are to be delivered in single hop, using IEEE 802.11p (or Dedicated Short Range Communications, DSRC) technology pursued by the industry and government [1,5]. The power limits prescribed by the Federal Communications Commission (FCC) for DSRC spectrum are as high as 33 dBm for vehicle on board units, so that a desired communication range of 300 meters for these safety messages can be easily reached in one hop.
It was originally hoped that all safety messages could be sent at some maximum power suited to a maximum range chosen to accommodate all types of safety messages. The literature suggests such a maximum range might be about 300 meters [1,2]. However, the literature also shows that if all messages were broadcast with enough power to cover 300 meters in the DSRC channels [1], message loss rates caused by MAC collision is between 20% and 40% [6]. These findings have motivated us to study algorithms for VSC applications that intelligently choose a target range no greater than necessary, and to analyze power control algorithms that choose a transmission power no greater than necessary for the targeted range. Hence, our problem formulation is as follows. We assume each message is presented to our power control layer with a target communication range (distance). The aim of our power control algorithm is to match the transmission power to the target communication range so that the channel bears no more power than required to deliver the message. We only try to adapt to slowly changing channel characteristics such as path loss and shadowing effect caused by buildings or vehicles intervening between vehicles. These kinds of conditions change at the time-scales of vehicle motion, i.e., in the seconds or hundreds of milliseconds. The safety messages themselves are sent at these time-scales. Our power control algorithm, like most such algorithms, is based on receiver feedback. By focusing on the slowly changing channel, we aim to piggyback the feedback information on the safety messages themselves. Furthermore, the safety messages being standardized always contain the GPS position of the sender. Assuming the receiving vehicle knows its own GPS position, it can compute the range to the sender. Because target range information is included in the message, the receiver knows whether the message has exceeded its target range. Then, it uses this piece of knowledge to provide feedback. The above is the basic idea and information structure of our design. Our idea utilizes the concept of cross-layer design, which couples power control with application design requirements. Similarly, GPS information from application is also used in our power control algorithm. The emphasis on range distinguishes our problem from power control problems attacked in the mobile ad hoc networks (MANET) literature. Power control scheme in MANET focuses on creating multi-hop network connectivity graphs that maximize multi-hop throughput, or deliver sufficient throughput while minimizing energy [7-13]. However, the objective of transmission power control in
978-1-4244-2026-1/08/$25.00 © 2008 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
vehicular ah hoc networks (VANET) for VSC applications is different as stated earlier. Several papers on power control for VANET’s [14,15,16] are directly related to our work. They use power control as a means of congestion control. The work in [14] proposes to monitor the speed of neighboring vehicles and then estimate vehicular traffic density. When the traffic is too dense, the transmission power is reduced. The estimation of vehicular traffic density from vehicular speed is a difficult one since the relationship is not of a deterministic nature. The paper [15] on the other hand, proposes to estimate data traffic density by using information available from the radio itself. Each node tracks the number of nodes from which it detects carrier sensing signal. Implicitly, this information is translated into an estimate of the data load in its neighborhood and is used to control the transmission power. Different from ours, neither paper considers the problem of adapting the transmission power to the target range specification of messages, which is important for VSC applications. The simulation results in section IV show that our power control algorithm outperforms the basic mechanism which just uses enough power to cover the worst-case, i.e, maximum transmission power for all safety messages. On the other hand, the results also show that there is scope for considerable improvement, especially when the data traffic loads are high. Congestion control is not the objective of our power control algorithm. Within protocol stack, our protocol is intended to locate on top of carrier sense multiple access (CSMA) but under transport protocols. CSMA achieves part of congestion control through its back-off mechanism. Transport protocols also can achieve congestion control functions, e.g., transmission control protocol (TCP). Unlike the established MAC and transport-layer protocols in wired network which respond to congestions by throttling the data rate, congestion control mechanisms in wireless network can also reduce congestion by adjusting transmission power. Clearly a future design needs to designate a transmission range that is both desired by the safety applications and appropriate for congestion control purpose. The paper is organized into the following sections: the next section will formulate the problem in detail and compare the power control algorithms in traditional ad hoc network and vehicle ad hoc network. Section III is the algorithm design and discussion. Section IV gives the simulation results to verify the algorithm. And finally, Section V provides the conclusions and future work planned. II.
discussion. The goal of our power control algorithm is to tune the transmission power so that safety messages can reach the target range without much excess. Excess range of communication beyond the required target range may cause interference and reduce the reliability of the VSC system. From a layering perspective, our power control algorithm utilizes the information from the application layer, to match the transmission power level to the specified target range. The target range, designated by the VSC application, serves as an input to the power control algorithm. Our power control algorithm is a best-effort approach to determine the transmission power level, while it does not replace the functionality of congestion control. We believe that congestion control, by changing target range or data rate requirements, should work independently with the power control algorithm. Here, we are only examining the control of transmission power. Given a target communication range, the required transmission power level cannot be computed accurately in an open-loop manner by using the propagation model, partly due to the nondeterministic and time varying nature of the channel. We assume that the channel changes slowly compared to the periodic broadcast message interval, so that our algorithm can rely on the historical observation to estimate the current channel situation and adjust the transmission power accordingly. III.
ALGORITHM DESIGN
A. Scheme description We propose to adjust the transmission power level of safety messages based on feedback from other vehicles. Our scheme is based on the fact that vehicles are equipped with GPS, and each one broadcasts its location periodically as part of the safety messages. The information required for power control is piggybacked as a header on the safety messages. Here we assume the safety messages are sent periodically. As shown later, we simulate a variety of rates in section IV. The receiver calculates the difference of its own position from that of the sender (sender positions included in the safety message received) to derive the relative distance between the sender and itself. ID
Position Data
Target Range
Feedback Beacon
Payload
PROBLEM DEFINITION
In a VSC system design, it is assumed vehicles are equipped with GPS and can obtain vehicle data from the vehicle bus. Therefore each one is aware of its safety-related information as speed, position, heading, brake information, etc. The information exchange between vehicles is conducted by the DSRC radio with an adjustable transmission power module. For the Cooperative Collision Warning (CCW) application, the message is broadcasted periodically [1]. We assume the communication distance is designated in the V2V safety message, and is called the target range in the following
Figure 1. Packet format
The power control protocol header is shown in figure 1. Among other things, our broadcast message header includes the Target Range and a Feedback Beacon. The specified Target Range denotes the valid range up to which the application would like the message to be transmitted. Each node (i.e. vehicle) maintains a list of nodes, called the speaker list. This list is used to keep track of the sender IDs of certain packets
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
that was received by each node. When a message is received, the receiver node computes the relative distance between itself and the sender. If the relative distance is larger than the Target Range specified in the received message, the receiver node includes the sender ID in its speaker list. When a node broadcasts, it chooses some of the node IDs from the speaker list, and assembles them to constitute the Feedback Beacon as shown in figure 1. The reason for including only a subset of node IDs from its speaker list is to reduce the feedback traffic in the communication channel. We use the MAC address to uniquely represent the node ID. In our current implementation, we choose the last ten IDs added to the speaker list within the last power update interval in our Feedback Beacon. However, the choice of IDs that would be included in the Feedback Beacon can also be based on other information such as relative distances between nodes, etc. After each broadcast, the speaker list is cleared by the node. Note that each node includes a subset of the speaker list in the Feedback Beacon with every broadcast packet. When a node receives messages, it searches for its own ID in the Feedback Beacon of the received messages. If its own ID is found, it concludes that in the last message sending interval its broadcast packet has been received by this sender and that this sender is located beyond the Target Range that was specified. In order to improve the reliability of the scheme, each node maintains a counter to record the number of nodes out of the specified Target Range that have reported the successful reception of its packet. In each message sending interval, if the counter value is larger than a chosen integer N, the transmission power will be decreased by a fixed step; else the power will be increased by a fixed step. The following pseudo code in figure 2 demonstrates how the different modules collaborate.
communication range with the current transmission power. This figure illustrates a freeway with 3 lanes, and each dot shown is a vehicle. When Tx range is larger than the Target Range, the sending node will become aware based on the received feedback of other nodes that are outside the target range. Therefore, the transmission power ought to be reduced. Else if the node can not find its ID in the Feedback Beacons received, it should increase the transmission power. Figure 3 provides an illustration of the two cases. In general, the wireless channel is non-deterministic due to multi-path and other fading effects. The effects of the non deterministic property of the channel are addressed in our design by threshold N used for the counter in the pseudo code. The node will decrease transmission power only if there are at least N nodes out of the Target Range that have replied with the sender’s ID in their Feedback Beacon. This conservative policy is chosen in order to maintain the maximum reliability of the packet reception within the Target Range and to prevent chattering of the transmission power selected. A good selection of N will result in selection of the required transmission power that yields and effective transmission range slightly larger than the required Target Range, without much chattering. To reduce the size of the Feedback Beacon, we include only a subset of nodes IDs from the speaker list. In general, this speaker list would consist of a large number of node IDs. If all the Node IDs from the speaker lists were included in the Feedback Beacon (which could be large), the channel cannot tolerate the amount of the feedback provided, which in the end, will congest the channel. However, if only a few IDs from the speaker list are included in the Feedback Beacon, then the scheme would work effectively without congesting the channel. Thus there is a tradeoff in the selection of proper feedback.
During the beacon time interval: update the speaker list search in the received beacons for ones own ID if ID is found counter++ At the end of the beacon time interval: if(counter>N) power level=power level-delta else power level=power level+delta if(power level>MAX) power level=MAX_POWER if(power level<MIN) power level=MIN_POWER counter=0 clear the speaker list
Figure 2. Pesudo code of the power control algorithm
In the pseudo code, MAX_POWER and MIN_POWER are the maximum and minimum power level supported by the radio. B. Scheme analysis The algorithm can be explained further using figure 3. Let r be the Target Range, and Tx range be the effective
Figure 3. Demonstration of the algorithm
The fixed step increase and decrease of transmission power proposed in the scheme is based on the assumption that the channel changes relatively slowly compared to the power update in each message exchange interval. The node can learn
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
the channel condition by the feedback it receives to determine the transmission power level required. The power control scheme can only be effective to slow-fading channel characteristics, and packet-by-packet power tuning scheme can be challenged by fast fading channel characteristics. The power control problem and proposed algorithm for VANET should be exercised meaningfully within the context of the applications considered. IV.
power to be 700 mW. It gives a sufficiently high message reception probability at the maximum target range of 300 meters in the absence of any multiple access interference. The rest of the simulations are run with vehicular traffic flowing as per the parameters in table I. This is a dense traffic condition. A power control algorithm should be required in this condition. The mobility trajectory simulates a freeway without entrances, exits or intersections.
SIMULATION
This section describes our evaluation of the design. The performance measure is the probability of message loss at a randomly chosen receiver within the target range. To be more precise, let packet i be sent with target range Ri, and at this time instant, there be ni nodes within the range Ri, out of which mi nodes receive the packet successfully. The packet loss rate for packet i, pi=1- mi/ni. The overall packet loss rate is the average over all packets loss rate. For a safety message this measure should be as small as possible. We evaluate the design by comparing it to two baselines. The first baseline represents the case without power control. In the absence of a power control algorithm, we assume that all messages have to be sent at a power matched to the maximum range. As discussed in the introduction, we assume this maximum range is 300 meters and use our simulator to find the corresponding maximum power level as follows. In the “no power control” case, all messages are then sent at this maximum power level. We compute the corresponding message loss probability and compare it to the message loss probability using power control. The algorithm is simulated using the OPNET simulator [17]. We have modified the 802.11a module to model DSRC. The propagation model is, P ( d ) = P ( d 0 ) − 10 γ log X
σ
10
(
d )+ X d0
σ
,
~ N ( 0 , 5 . 5 dB ), γ = 2 . 75 ,
in which P(d) and P(d0) are the receiving power at distance d and reference distance d0, and γ is the path loss exponent. X σ is used to model channel fading. To derive the “no power control” baseline we send messages from a single sender to a single receiver, as shown in figure 4. At each distance we use multiple transmission power levels. At each power level and distance we compute the message loss probability at the receiver to obtain the curves in figure 5. The receiver sensitivity is -87dbm. We assume a slow-fading channel, so that the channel characteristics do not vary drastically during the packet transmissions.
Figure 4. Single node to single node transmission
For example, when transmission power is 50mW, about 80% of the messages can be received 200 meters away. Based on figure 5, we choose the “no power control” transmission
Figure 5. Packet receiving prob. vs. Tx power TABLE I.
SIMULATION PARAMETERS
Lanes lane length average spacings average speed traffic flow receiver sensitivity data rate payload size TABLE II.
4 1600m 17.0m 11.6m/s 1920vph/lane -87dBm 6Mbps 300 bytes
POWER CONTROL ALGORITHM PARAMETERS
maximum transmission power minimum transmission power N delta power Feedback beacon size
1W 10mW 5 10mW 60 Bytes*
* The beacon at most contains ten nodes ID. Each node ID is its MAC address of 6 bytes.
We next present results on a series of simulations in which the messages are sent at an average rate of 1 every 50 msec, 100 msec, 200 msec, 300 msec, and 500 msec. The results appear in figures 6,7,8,9,10. In each figure, the dash line denotes the message loss rate using our power control scheme. The sending process is exponentially distributed. Other parameters are as in table I and the power control parameters are as in table II. In each of these simulations we derive another baseline. The second baseline is derived by running a series of simulations using the mobile traffic specified by table I. In each simulation all nodes use a constant power level. For a given density of vehicles and data traffic, one would expect there to be an optimal power level. If the power level is too low, too many messages will be lost due to channel attenuation. If the
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
power level is too high, too many messages will be lost due to multiple access interference. For example, for a sending rate of 100 msec, the optimal power message loss (baseline 2) is about 6%, the message loss with our power control scheme is 20.5%, and that without power control (baseline 1) is 45.1%. Table III gives these three set of data points for each sending rate. We observe, as expected, that the power control always does better than the “no power control” case. Also, in most cases, the performance of our power control protocol is a reasonable approximation of optimal solution (indicated as second baseline). These observations seem to indicate that our power control algorithm help to improve the reliability of vehicle safety applications. Thus, we believe that power control is a promising technical approach to enhance the performance of safety applications. TABLE III.
Figure 7. Packet loss rate vs. power level (packet sending interval=200ms)
PACKET LOSS RATE IN DIFFERENT SCHEMES
message sending interval(ms)
packet loss rate using power control (%)
packet loss rate using optimal operating power (%)
500 300 200 100 50
1.33 2.71 3.92 20.5 61.3
1.11 1.99 2.59 6.70 13.0
packet loss rate using maximum power of 700mW (%) 1.38 4.10 9.61 45.1 78.0
On the other hand, a careful comparison between the performance of our scheme and the performance of optimal solution is also interesting. As the channel gets more congested (i.e., at higher sending rates), we find that our power control algorithm does not converge to the optimal performance. For example, at 50 msec, the loss rate using optimal power setup can be as low as 13% while our power control algorithm causes a 61.3% loss rate. Our conjecture is that the algorithm is exhibiting bi-stable behavior. If feedback is not being correctly received due to collisions, higher power levels will be used as a nature result, leading to even more severe collisions and less correctly received feedback. This way, the algorithm is converging to an unstable equilibrium. We are currently investigating this phenomenon so as to further improve the robustness of our protocol design.
Figure 8. Packet loss rate vs. power level (packet sending interval=300ms)
Figure 9. Packet loss rate vs. power level (packet sending interval=500ms)
In the case of message sending interval 200ms, the power level statistic is shown by the CDF curve in figure 11. This CDF curve indicates that the algorithm can result in a stationary state in terms of the transmission power.
Figure 6. Packet loss rate vs. power level (packet sending interval=100ms)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
[2]
[3]
[4] [5]
[6] Figure 10. Packet loss rate vs. power level (message sending interval=50ms) [7] [8]
[9]
[10]
[11]
Figure 11. CDF of transmission power (message sending interval=200ms)
[12]
V. CONCLUSIONS & FUTURE WORK We have developed a power control algorithm and protocol to determine the transmission power for reliable vehicle safety communication. Our simulation results show that the algorithm performs better than the alternative solutions. The alternative to an adaptive scheme (such as ours) would be to send all messages at some maximum power chosen to match a maximum range covering the worst case requirement for safety messages. We have also compared our algorithm to an optimal performance using a constant power level that is exactly matched to the vehicular traffic density and data traffic load. This theoretical-oriented optimal solution always does better than our power control algorithm and indicates the potential for further improvement in our design. We find that the more the data traffic loads on the channel, the greater the potential for improvement to our current design.
[13]
Our future work is focused on understanding the conjecture about instability discussed in section IV and modifying the control design accordingly. We also seek to combine the ideas in this paper with those on power control for congestion control in the literature. This would properly complement the flow control delivered by conventional medium access control and transport protocols. REFERENCES [1]
“Vehicle Safety Communications Project Task 3 Final Report: Identify Intelligent Vehicle Safety Applications Enabled by DSRC”, DOT HS 809 859, Mar. 2005.
[14]
[15] [16]
[17]
Xu Q. et.al.,”Medium Access Control Protocol Design for VehicleVehicle Safety Messages”. Accepted by the IEEE Transactions on Vehicular Technology, April 2006. (To appear 2007) C.L.Robinson et.al.,”Efficient Coordination and Transmission of Data for Cooperative Vehicular Safety Applications”, Proc. of the second ACM VANET, 2005. Xu, Q., et.al., “Vehicle-Vehicle Safety Messaging in DSRC”. Proc. of the first ACM VANET, 2004. “Wireless Access in Vehicular Environment (WAVE) in Standard 802.11 Information Technology Telecommunications and Information Exchange Between Systems, Local and Metropolitan Area Networks, Specific Requirements, Part 11: Wireless LAN Medium Access Control(MAC) and Phisical Layer(PHY) Specifications, IEEE 802.11p/D1.0”, Feb. 2006. T. K. Mak et.al., “A Multi-Channel VANET Providing Concurrent Safety and Commercial Services.”, Proc. of the second ACM VANET, 2005. V. Rodoplu et.al., “Minimum Energy Mobile Wireless Networks,” IEEE JSAC, vol. 17, no. 8, Aug. 1999, pp. 1333-44. J. Monks, et.al., “A Power Controlled Multiple Access Protocol for Wireless Packet Networks,” Proc. IEEE INFOCOM Conf., 2001, pp. 219-28. S. -L. Wu, et.al., “Intelligent Medium Access for Mobile Ad Hoc Networks with Busy Tones and Power Control,” IEEE JSAC, vol. 18, no. 9, 2000, pp. 1647-57. A. Muqattash et.al., “Power Controlled Dual Channel (PCDC) Medium Access Protocol for Wireless Ad Hoc networks,” Proc. IEEE INFOCOM Conf., 2003, pp. 470-80. R. Wattenhofer et al., “Distributed Topology Control for Power Efficient Operation in Multihop Wireless Ad Hoc Networks,” Proc. IEEE INFOCOM Conf., 2001, pp. 1388-97. L. Li et.al., “Minimum Energy Mobile Wireless Networks Revisited”, Proc. of ICC , June 2001. S. Narayanaswarmy et.al., “Power Control in Ad-hoc Networks: Theory, Architecture, Algorithm and Implementation of COMPOW protocol,” Proc. of European Wireless Conference – Next Generation Wireless Networks: Technologies, Protocols, Services and Applications, 2002. H. Xu, M.Barth, “A Transmission-Interval and Power-Level Modulation Methodology for Optimizing Inter-Vehicle Communications”, Proc. of the first ACM VANET, 2004. M. Torrent-Moreno et.al., “Fair Sharing of Bandwidth in VANETs”, Proc. of the second ACM VANET, 2005. M. Torrent-Moreno et.al., “Distributed Fair Transmit Power Adjustment for Vehicular Ad Hoc Networks”, Proc. of the Third Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON), 2006 http://www.opnet.com/
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
A Feedback-Based Power Control Algorithm Design for VANET Xu Guan*, Raja Sengupta*, Hariharan Krishnan**, Fan Bai** *Systems Program, CEE, University of California, Berkeley, **General Motors R & D Center
Abstract— We consider the problem of adjusting transmission power for vehicle-to-vehicle broadcast safety communication in vehicular ad hoc networks. Given a target communication range designated by a vehicle safety application, the power control algorithm is designed to select a transmission power no greater than necessary for the targeted range. The power control algorithm results in higher communication reliability since collisions are minimized for safety communications. Our main idea is to add a power tuning feedback beacon during each safety message exchange. Our simulation results show that the power control algorithm can reduce the transmission power significantly to achieve higher packet reception rate than the case without any power control. This algorithm is mainly applied to communications-based vehicle safety applications but could be useful for other applications as well. Keywords- Power control, VANET, V2V
I. INTRODUCTION In this paper, we propose an adaptive transmission power control algorithm for vehicle-to-vehicle (V2V) communications-based safety applications. There is a well established effort in government and industry to develop vehicle safety communication (VSC) system, which uses broadcasts of safety related information from nearby vehicles to enhance vehicle safety[1,2,3]. These broadcast safety messages are periodic in nature, and contain information such as position, speed, heading, braking profile, etc., enabling the receiving vehicle to detect hazardous motions by another vehicle, thereby alerting the driver of that situation. Given the critical safety nature, these messages need to be delivered reliably and timely to a desired range. For example, if a vehicle is stopped in the middle of a highway, it needs to send safety messages to vehicles that might be approaching it from behind. Since the comfortable stopping distance of a vehicle at freeway speeds could be about 250 meters, the stopped vehicle seeks to broadcast its safety messages to a desired range of 300 meters. Similar performance requirements for other VSC applications are listed in the literature [1,2,4]. We assume that these safety broadcast messages are to be delivered in single hop, using IEEE 802.11p (or Dedicated Short Range Communications, DSRC) technology pursued by the industry and government [1,5]. The power limits prescribed by the Federal Communications Commission (FCC) for DSRC spectrum are as high as 33 dBm for vehicle on board units, so that a desired communication range of 300 meters for these safety messages can be easily reached in one hop.
It was originally hoped that all safety messages could be sent at some maximum power suited to a maximum range chosen to accommodate all types of safety messages. The literature suggests such a maximum range might be about 300 meters [1,2]. However, the literature also shows that if all messages were broadcast with enough power to cover 300 meters in the DSRC channels [1], message loss rates caused by MAC collision is between 20% and 40% [6]. These findings have motivated us to study algorithms for VSC applications that intelligently choose a target range no greater than necessary, and to analyze power control algorithms that choose a transmission power no greater than necessary for the targeted range. Hence, our problem formulation is as follows. We assume each message is presented to our power control layer with a target communication range (distance). The aim of our power control algorithm is to match the transmission power to the target communication range so that the channel bears no more power than required to deliver the message. We only try to adapt to slowly changing channel characteristics such as path loss and shadowing effect caused by buildings or vehicles intervening between vehicles. These kinds of conditions change at the time-scales of vehicle motion, i.e., in the seconds or hundreds of milliseconds. The safety messages themselves are sent at these time-scales. Our power control algorithm, like most such algorithms, is based on receiver feedback. By focusing on the slowly changing channel, we aim to piggyback the feedback information on the safety messages themselves. Furthermore, the safety messages being standardized always contain the GPS position of the sender. Assuming the receiving vehicle knows its own GPS position, it can compute the range to the sender. Because target range information is included in the message, the receiver knows whether the message has exceeded its target range. Then, it uses this piece of knowledge to provide feedback. The above is the basic idea and information structure of our design. Our idea utilizes the concept of cross-layer design, which couples power control with application design requirements. Similarly, GPS information from application is also used in our power control algorithm. The emphasis on range distinguishes our problem from power control problems attacked in the mobile ad hoc networks (MANET) literature. Power control scheme in MANET focuses on creating multi-hop network connectivity graphs that maximize multi-hop throughput, or deliver sufficient throughput while minimizing energy [7-13]. However, the objective of transmission power control in
978-1-4244-2026-1/08/$25.00 © 2008 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
vehicular ah hoc networks (VANET) for VSC applications is different as stated earlier. Several papers on power control for VANET’s [14,15,16] are directly related to our work. They use power control as a means of congestion control. The work in [14] proposes to monitor the speed of neighboring vehicles and then estimate vehicular traffic density. When the traffic is too dense, the transmission power is reduced. The estimation of vehicular traffic density from vehicular speed is a difficult one since the relationship is not of a deterministic nature. The paper [15] on the other hand, proposes to estimate data traffic density by using information available from the radio itself. Each node tracks the number of nodes from which it detects carrier sensing signal. Implicitly, this information is translated into an estimate of the data load in its neighborhood and is used to control the transmission power. Different from ours, neither paper considers the problem of adapting the transmission power to the target range specification of messages, which is important for VSC applications. The simulation results in section IV show that our power control algorithm outperforms the basic mechanism which just uses enough power to cover the worst-case, i.e, maximum transmission power for all safety messages. On the other hand, the results also show that there is scope for considerable improvement, especially when the data traffic loads are high. Congestion control is not the objective of our power control algorithm. Within protocol stack, our protocol is intended to locate on top of carrier sense multiple access (CSMA) but under transport protocols. CSMA achieves part of congestion control through its back-off mechanism. Transport protocols also can achieve congestion control functions, e.g., transmission control protocol (TCP). Unlike the established MAC and transport-layer protocols in wired network which respond to congestions by throttling the data rate, congestion control mechanisms in wireless network can also reduce congestion by adjusting transmission power. Clearly a future design needs to designate a transmission range that is both desired by the safety applications and appropriate for congestion control purpose. The paper is organized into the following sections: the next section will formulate the problem in detail and compare the power control algorithms in traditional ad hoc network and vehicle ad hoc network. Section III is the algorithm design and discussion. Section IV gives the simulation results to verify the algorithm. And finally, Section V provides the conclusions and future work planned. II.
discussion. The goal of our power control algorithm is to tune the transmission power so that safety messages can reach the target range without much excess. Excess range of communication beyond the required target range may cause interference and reduce the reliability of the VSC system. From a layering perspective, our power control algorithm utilizes the information from the application layer, to match the transmission power level to the specified target range. The target range, designated by the VSC application, serves as an input to the power control algorithm. Our power control algorithm is a best-effort approach to determine the transmission power level, while it does not replace the functionality of congestion control. We believe that congestion control, by changing target range or data rate requirements, should work independently with the power control algorithm. Here, we are only examining the control of transmission power. Given a target communication range, the required transmission power level cannot be computed accurately in an open-loop manner by using the propagation model, partly due to the nondeterministic and time varying nature of the channel. We assume that the channel changes slowly compared to the periodic broadcast message interval, so that our algorithm can rely on the historical observation to estimate the current channel situation and adjust the transmission power accordingly. III.
ALGORITHM DESIGN
A. Scheme description We propose to adjust the transmission power level of safety messages based on feedback from other vehicles. Our scheme is based on the fact that vehicles are equipped with GPS, and each one broadcasts its location periodically as part of the safety messages. The information required for power control is piggybacked as a header on the safety messages. Here we assume the safety messages are sent periodically. As shown later, we simulate a variety of rates in section IV. The receiver calculates the difference of its own position from that of the sender (sender positions included in the safety message received) to derive the relative distance between the sender and itself. ID
Position Data
Target Range
Feedback Beacon
Payload
PROBLEM DEFINITION
In a VSC system design, it is assumed vehicles are equipped with GPS and can obtain vehicle data from the vehicle bus. Therefore each one is aware of its safety-related information as speed, position, heading, brake information, etc. The information exchange between vehicles is conducted by the DSRC radio with an adjustable transmission power module. For the Cooperative Collision Warning (CCW) application, the message is broadcasted periodically [1]. We assume the communication distance is designated in the V2V safety message, and is called the target range in the following
Figure 1. Packet format
The power control protocol header is shown in figure 1. Among other things, our broadcast message header includes the Target Range and a Feedback Beacon. The specified Target Range denotes the valid range up to which the application would like the message to be transmitted. Each node (i.e. vehicle) maintains a list of nodes, called the speaker list. This list is used to keep track of the sender IDs of certain packets
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
that was received by each node. When a message is received, the receiver node computes the relative distance between itself and the sender. If the relative distance is larger than the Target Range specified in the received message, the receiver node includes the sender ID in its speaker list. When a node broadcasts, it chooses some of the node IDs from the speaker list, and assembles them to constitute the Feedback Beacon as shown in figure 1. The reason for including only a subset of node IDs from its speaker list is to reduce the feedback traffic in the communication channel. We use the MAC address to uniquely represent the node ID. In our current implementation, we choose the last ten IDs added to the speaker list within the last power update interval in our Feedback Beacon. However, the choice of IDs that would be included in the Feedback Beacon can also be based on other information such as relative distances between nodes, etc. After each broadcast, the speaker list is cleared by the node. Note that each node includes a subset of the speaker list in the Feedback Beacon with every broadcast packet. When a node receives messages, it searches for its own ID in the Feedback Beacon of the received messages. If its own ID is found, it concludes that in the last message sending interval its broadcast packet has been received by this sender and that this sender is located beyond the Target Range that was specified. In order to improve the reliability of the scheme, each node maintains a counter to record the number of nodes out of the specified Target Range that have reported the successful reception of its packet. In each message sending interval, if the counter value is larger than a chosen integer N, the transmission power will be decreased by a fixed step; else the power will be increased by a fixed step. The following pseudo code in figure 2 demonstrates how the different modules collaborate.
communication range with the current transmission power. This figure illustrates a freeway with 3 lanes, and each dot shown is a vehicle. When Tx range is larger than the Target Range, the sending node will become aware based on the received feedback of other nodes that are outside the target range. Therefore, the transmission power ought to be reduced. Else if the node can not find its ID in the Feedback Beacons received, it should increase the transmission power. Figure 3 provides an illustration of the two cases. In general, the wireless channel is non-deterministic due to multi-path and other fading effects. The effects of the non deterministic property of the channel are addressed in our design by threshold N used for the counter in the pseudo code. The node will decrease transmission power only if there are at least N nodes out of the Target Range that have replied with the sender’s ID in their Feedback Beacon. This conservative policy is chosen in order to maintain the maximum reliability of the packet reception within the Target Range and to prevent chattering of the transmission power selected. A good selection of N will result in selection of the required transmission power that yields and effective transmission range slightly larger than the required Target Range, without much chattering. To reduce the size of the Feedback Beacon, we include only a subset of nodes IDs from the speaker list. In general, this speaker list would consist of a large number of node IDs. If all the Node IDs from the speaker lists were included in the Feedback Beacon (which could be large), the channel cannot tolerate the amount of the feedback provided, which in the end, will congest the channel. However, if only a few IDs from the speaker list are included in the Feedback Beacon, then the scheme would work effectively without congesting the channel. Thus there is a tradeoff in the selection of proper feedback.
During the beacon time interval: update the speaker list search in the received beacons for ones own ID if ID is found counter++ At the end of the beacon time interval: if(counter>N) power level=power level-delta else power level=power level+delta if(power level>MAX) power level=MAX_POWER if(power level<MIN) power level=MIN_POWER counter=0 clear the speaker list
Figure 2. Pesudo code of the power control algorithm
In the pseudo code, MAX_POWER and MIN_POWER are the maximum and minimum power level supported by the radio. B. Scheme analysis The algorithm can be explained further using figure 3. Let r be the Target Range, and Tx range be the effective
Figure 3. Demonstration of the algorithm
The fixed step increase and decrease of transmission power proposed in the scheme is based on the assumption that the channel changes relatively slowly compared to the power update in each message exchange interval. The node can learn
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
the channel condition by the feedback it receives to determine the transmission power level required. The power control scheme can only be effective to slow-fading channel characteristics, and packet-by-packet power tuning scheme can be challenged by fast fading channel characteristics. The power control problem and proposed algorithm for VANET should be exercised meaningfully within the context of the applications considered. IV.
power to be 700 mW. It gives a sufficiently high message reception probability at the maximum target range of 300 meters in the absence of any multiple access interference. The rest of the simulations are run with vehicular traffic flowing as per the parameters in table I. This is a dense traffic condition. A power control algorithm should be required in this condition. The mobility trajectory simulates a freeway without entrances, exits or intersections.
SIMULATION
This section describes our evaluation of the design. The performance measure is the probability of message loss at a randomly chosen receiver within the target range. To be more precise, let packet i be sent with target range Ri, and at this time instant, there be ni nodes within the range Ri, out of which mi nodes receive the packet successfully. The packet loss rate for packet i, pi=1- mi/ni. The overall packet loss rate is the average over all packets loss rate. For a safety message this measure should be as small as possible. We evaluate the design by comparing it to two baselines. The first baseline represents the case without power control. In the absence of a power control algorithm, we assume that all messages have to be sent at a power matched to the maximum range. As discussed in the introduction, we assume this maximum range is 300 meters and use our simulator to find the corresponding maximum power level as follows. In the “no power control” case, all messages are then sent at this maximum power level. We compute the corresponding message loss probability and compare it to the message loss probability using power control. The algorithm is simulated using the OPNET simulator [17]. We have modified the 802.11a module to model DSRC. The propagation model is, P ( d ) = P ( d 0 ) − 10 γ log X
σ
10
(
d )+ X d0
σ
,
~ N ( 0 , 5 . 5 dB ), γ = 2 . 75 ,
in which P(d) and P(d0) are the receiving power at distance d and reference distance d0, and γ is the path loss exponent. X σ is used to model channel fading. To derive the “no power control” baseline we send messages from a single sender to a single receiver, as shown in figure 4. At each distance we use multiple transmission power levels. At each power level and distance we compute the message loss probability at the receiver to obtain the curves in figure 5. The receiver sensitivity is -87dbm. We assume a slow-fading channel, so that the channel characteristics do not vary drastically during the packet transmissions.
Figure 4. Single node to single node transmission
For example, when transmission power is 50mW, about 80% of the messages can be received 200 meters away. Based on figure 5, we choose the “no power control” transmission
Figure 5. Packet receiving prob. vs. Tx power TABLE I.
SIMULATION PARAMETERS
Lanes lane length average spacings average speed traffic flow receiver sensitivity data rate payload size TABLE II.
4 1600m 17.0m 11.6m/s 1920vph/lane -87dBm 6Mbps 300 bytes
POWER CONTROL ALGORITHM PARAMETERS
maximum transmission power minimum transmission power N delta power Feedback beacon size
1W 10mW 5 10mW 60 Bytes*
* The beacon at most contains ten nodes ID. Each node ID is its MAC address of 6 bytes.
We next present results on a series of simulations in which the messages are sent at an average rate of 1 every 50 msec, 100 msec, 200 msec, 300 msec, and 500 msec. The results appear in figures 6,7,8,9,10. In each figure, the dash line denotes the message loss rate using our power control scheme. The sending process is exponentially distributed. Other parameters are as in table I and the power control parameters are as in table II. In each of these simulations we derive another baseline. The second baseline is derived by running a series of simulations using the mobile traffic specified by table I. In each simulation all nodes use a constant power level. For a given density of vehicles and data traffic, one would expect there to be an optimal power level. If the power level is too low, too many messages will be lost due to channel attenuation. If the
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
power level is too high, too many messages will be lost due to multiple access interference. For example, for a sending rate of 100 msec, the optimal power message loss (baseline 2) is about 6%, the message loss with our power control scheme is 20.5%, and that without power control (baseline 1) is 45.1%. Table III gives these three set of data points for each sending rate. We observe, as expected, that the power control always does better than the “no power control” case. Also, in most cases, the performance of our power control protocol is a reasonable approximation of optimal solution (indicated as second baseline). These observations seem to indicate that our power control algorithm help to improve the reliability of vehicle safety applications. Thus, we believe that power control is a promising technical approach to enhance the performance of safety applications. TABLE III.
Figure 7. Packet loss rate vs. power level (packet sending interval=200ms)
PACKET LOSS RATE IN DIFFERENT SCHEMES
message sending interval(ms)
packet loss rate using power control (%)
packet loss rate using optimal operating power (%)
500 300 200 100 50
1.33 2.71 3.92 20.5 61.3
1.11 1.99 2.59 6.70 13.0
packet loss rate using maximum power of 700mW (%) 1.38 4.10 9.61 45.1 78.0
On the other hand, a careful comparison between the performance of our scheme and the performance of optimal solution is also interesting. As the channel gets more congested (i.e., at higher sending rates), we find that our power control algorithm does not converge to the optimal performance. For example, at 50 msec, the loss rate using optimal power setup can be as low as 13% while our power control algorithm causes a 61.3% loss rate. Our conjecture is that the algorithm is exhibiting bi-stable behavior. If feedback is not being correctly received due to collisions, higher power levels will be used as a nature result, leading to even more severe collisions and less correctly received feedback. This way, the algorithm is converging to an unstable equilibrium. We are currently investigating this phenomenon so as to further improve the robustness of our protocol design.
Figure 8. Packet loss rate vs. power level (packet sending interval=300ms)
Figure 9. Packet loss rate vs. power level (packet sending interval=500ms)
In the case of message sending interval 200ms, the power level statistic is shown by the CDF curve in figure 11. This CDF curve indicates that the algorithm can result in a stationary state in terms of the transmission power.
Figure 6. Packet loss rate vs. power level (packet sending interval=100ms)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
[2]
[3]
[4] [5]
[6] Figure 10. Packet loss rate vs. power level (message sending interval=50ms) [7] [8]
[9]
[10]
[11]
Figure 11. CDF of transmission power (message sending interval=200ms)
[12]
V. CONCLUSIONS & FUTURE WORK We have developed a power control algorithm and protocol to determine the transmission power for reliable vehicle safety communication. Our simulation results show that the algorithm performs better than the alternative solutions. The alternative to an adaptive scheme (such as ours) would be to send all messages at some maximum power chosen to match a maximum range covering the worst case requirement for safety messages. We have also compared our algorithm to an optimal performance using a constant power level that is exactly matched to the vehicular traffic density and data traffic load. This theoretical-oriented optimal solution always does better than our power control algorithm and indicates the potential for further improvement in our design. We find that the more the data traffic loads on the channel, the greater the potential for improvement to our current design.
[13]
Our future work is focused on understanding the conjecture about instability discussed in section IV and modifying the control design accordingly. We also seek to combine the ideas in this paper with those on power control for congestion control in the literature. This would properly complement the flow control delivered by conventional medium access control and transport protocols. REFERENCES [1]
“Vehicle Safety Communications Project Task 3 Final Report: Identify Intelligent Vehicle Safety Applications Enabled by DSRC”, DOT HS 809 859, Mar. 2005.
[14]
[15] [16]
[17]
Xu Q. et.al.,”Medium Access Control Protocol Design for VehicleVehicle Safety Messages”. Accepted by the IEEE Transactions on Vehicular Technology, April 2006. (To appear 2007) C.L.Robinson et.al.,”Efficient Coordination and Transmission of Data for Cooperative Vehicular Safety Applications”, Proc. of the second ACM VANET, 2005. Xu, Q., et.al., “Vehicle-Vehicle Safety Messaging in DSRC”. Proc. of the first ACM VANET, 2004. “Wireless Access in Vehicular Environment (WAVE) in Standard 802.11 Information Technology Telecommunications and Information Exchange Between Systems, Local and Metropolitan Area Networks, Specific Requirements, Part 11: Wireless LAN Medium Access Control(MAC) and Phisical Layer(PHY) Specifications, IEEE 802.11p/D1.0”, Feb. 2006. T. K. Mak et.al., “A Multi-Channel VANET Providing Concurrent Safety and Commercial Services.”, Proc. of the second ACM VANET, 2005. V. Rodoplu et.al., “Minimum Energy Mobile Wireless Networks,” IEEE JSAC, vol. 17, no. 8, Aug. 1999, pp. 1333-44. J. Monks, et.al., “A Power Controlled Multiple Access Protocol for Wireless Packet Networks,” Proc. IEEE INFOCOM Conf., 2001, pp. 219-28. S. -L. Wu, et.al., “Intelligent Medium Access for Mobile Ad Hoc Networks with Busy Tones and Power Control,” IEEE JSAC, vol. 18, no. 9, 2000, pp. 1647-57. A. Muqattash et.al., “Power Controlled Dual Channel (PCDC) Medium Access Protocol for Wireless Ad Hoc networks,” Proc. IEEE INFOCOM Conf., 2003, pp. 470-80. R. Wattenhofer et al., “Distributed Topology Control for Power Efficient Operation in Multihop Wireless Ad Hoc Networks,” Proc. IEEE INFOCOM Conf., 2001, pp. 1388-97. L. Li et.al., “Minimum Energy Mobile Wireless Networks Revisited”, Proc. of ICC , June 2001. S. Narayanaswarmy et.al., “Power Control in Ad-hoc Networks: Theory, Architecture, Algorithm and Implementation of COMPOW protocol,” Proc. of European Wireless Conference – Next Generation Wireless Networks: Technologies, Protocols, Services and Applications, 2002. H. Xu, M.Barth, “A Transmission-Interval and Power-Level Modulation Methodology for Optimizing Inter-Vehicle Communications”, Proc. of the first ACM VANET, 2004. M. Torrent-Moreno et.al., “Fair Sharing of Bandwidth in VANETs”, Proc. of the second ACM VANET, 2005. M. Torrent-Moreno et.al., “Distributed Fair Transmit Power Adjustment for Vehicular Ad Hoc Networks”, Proc. of the Third Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON), 2006 http://www.opnet.com/
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings.
