Chapter 2 Survey on ETC and Communication Technologies
2.4. Summary
The capacity of existing DSRC based ETC system using single channel for transaction process is not enough for instantaneously huge traffic flow with high traveling speed. And the trend of ETC systems is integration of ITS systems and MLFF ETC systems so that traditional DSRC technologies such as RFID and infrared are not appropriate. WAVE/DSRC is a communication technology which is designed for high speed, high dynamic network topology. The features of multi-channel operation and low transmission latency delay make it satisfy the requirements. Thus, development of WAVE/DSRC based MLFF ETC system with high system capacity which takes advantages of multi-channel operation and compatibility of ITS is important.
14
Chapter 3
System Architecture
3.1. Introduce to the Multi-Channel ETC
Figure 3-1 Multi-channel contention free ETC architecture
Figure 3-1 shows the architecture of developed contention-free polling based ETC system. The contention based communications between R-RSU and vehicles are used for
15
SCH reservation. When vehicles enter reservation zone and receive channel reservation beacon, called R-Beacon, broadcasted from R-RSU on CCH, vehicles attempt to access medium and send request for channel reservation to R-RSU on CCH. Each T-RSU uses independent SCH to poll vehicles for transaction process in transaction zone. According to the polling based communication mode and several independent SCH, T-RSUs play roles of coordinators for vehicles’ transaction processes. One of the T-RSUs uses control channel (CCH) to service vehicles which are failed to channel reservation before entering transaction zone. The other T-RSUs use independent SCHs to poll vehicles for transaction process.
R-RSU periodically broadcasts R-Beacon packets on CCH covering both transaction zone and reservation zone. The necessary information such as identify authentication or timestamp is including in R-Beacon. Vehicles in reservation zone without channel reservation receiving R-Beacons will send a channel reservation request, called R-REQ, to R-RSU on CCH. There may be several vehicles that attempt to send R-REQ packets to R-RSU at the same time. These vehicles will contend to each other and try to access medium based on MAC protocol. After R-RSU receives the R-REQ packet from a vehicle, the channel allocator, which is designed for coordination of SCH usage, chooses an appropriate SCH number for the requested vehicle and then sends an R-RES packet back to the vehicles.
As the requested vehicle received the R-Response packet, vehicle sets its own SCH number to the reserved SCH number in R-RES packet and sends an ACK packet back to R-RSU.
R-RSU will add vehicles to the corresponding polling queue of T-RSU based on the reserved channel number after receiving the ACK packet. The packet flow of channel reservation is shown in Figure 3-2.
.
16
Figure 3-2 Packet flow of channel reservation process
In SCH interval (SCHI), T-RSUs poll vehicles in corresponding polling queues. The polling scheduler decides polling queue sequence. It means the order of vehicles which wait to be polled is decided by polling scheduler. The polling scheduler may happen in the end of CCHI or at the beginning of each polling cycle in SCHI. In first case the polling sequence is the same in one SCHI. In second case the order of polling sequence is decided in run time.
Before T-RSU sending a polling packet, the polling scheduler chooses an appropriate vehicle in polling queue to be polled.
The polling based transaction process is a three handshake process. After vehicles
17
receiving the polling packet, they will send back an ACK packet to the corresponding T-RSU based on their own SCH number on SCH. The transaction process is complete when T-RSU receives the ACK packet from the polled vehicle.
If vehicles cannot reserve channel successfully before entering transaction zone, the one of T-RSUs will use contention base access scheme for backup transaction process to these vehicles on CCH. The backup transaction process is a three handshake process similar to other contention based ETC systems. The flow chart of the ETC system including channel reservation process and transaction process is shown in Figure 3-3.
18
Figure 3-3 Flow chart of channel reservation and transaction process
3.2. Challenges of the system
About contention
The goal of contention access scheme is to let vehicles reserve channel successfully
19
before they entering transaction zone. The length of reservation zone is one of factors that influences channel reservation successful probability. The longer reservation zone means more vehicles stay in reservation zone for longer time. After received R-Beacon, vehicles in reservation zone without channel reservation will attend to access medium to send the request to R-RSU at the same time. The contention of channel reservation between vehicles will be more serious due to the longer length of reservation zone.
We conduct a simple analysis of channel reservation successful ratio to illustrate the above problem [18]. Assume there is x-lane highway ETC system and length of the reservation zone is 𝑙𝑅. The R-RSU broadcasts R-Beacon at fixed interval time 𝐼𝐵.The vehicle arrival rate of each lane is λ. The average vehicle speed is v. Thus, the number of vehicles in reservation zone 𝑁𝑅 can be expressed as following:
𝑁𝑅 = 𝑥 ∙ 𝜆 ∙ 𝑙𝑅∙1𝑣 (1)
Each vehicle in reservation zone that attempt to send packets has the same probability, means the probability is reciprocal of the number of vehicle. If there are more vehicles in reservation, lower probability is. Actually, there is no sense about less than 1 vehicle. The number of vehicles in reservation zone 𝑁𝑅 is an average value which means it is a probability distribution. Here we use Poison probability distribution to describe the number of vehicles in reservation zone. The probability that exact x vehicles in reservation zone can be express as following:
Pr (x = k) =𝑁𝑅𝑘 ∙ 𝑒𝑘!−𝑁𝑅 (2)
By using the probability distribution Pr (x = k) and the transmission probability which is reciprocal of the number of vehicles, we obtain one transmission successful ratio 𝑃𝑆:
𝑃𝑆 = ∑𝑚𝑘=1 1𝑘Pr(𝑥 = 𝑘) (3)
20
In 802.11p, if medium is busy, vehicles require to wait until medium is free and a random back-off time, called contention window. The contention window initializes at an initial value 𝐶𝑊𝑚𝑖𝑛. If next time that vehicles try to access but medium is still busy, the contention window value will double until reach a maximal value 𝐶𝑊𝑚𝑎𝑥. The initial value and maximal value of contention windows are defined in IEEE 802.11p based on the priority level. Each packet transmission time 𝑡𝑇𝑥 equals to 𝑅𝐵
𝑇𝑥 where symbol B is packet size and 𝑅𝑇𝑥 is transmission rate. When medium is busy, means a vehicle is transmitting, the other vehicles require to wait until the transmission over. The average waiting time of each busy medium access is half of packet transmission duration 12𝑡𝑇𝑥. The expected time 𝑡𝑠 of one
time successful transmission can be expressed as following:
𝑡𝑠 = 𝑡𝑇𝑥 + ∑ [ 𝑃𝑠(1 − 𝑃𝑠)𝑥∙
(𝑚𝑖𝑛{2𝑥−2∗ 𝐶𝑊𝑚𝑖𝑛, 𝐶𝑊𝑚𝑎𝑥} +𝑥2𝑡𝑇𝑥)]
𝑚𝑥=1 (4)
In CCHI, vehicles should complete a three-handshake process for channel reservation.
It requires at least three times successful packets transmission. If successful packet transmissions between vehicles and R-RSU are less than three in a CCHI, vehicles fail to
Vehicles need to wait a little time for first R-Beacon reception. To simplify the question,
21
we assume the average waiting time 12𝐼𝐵 for vehicles to receive first R-Beacon when
entering reservation zone. The expected number of R-Beacon reception will rely on the remaining length of reservation zone and vehicle speed. R-RSU sends R-Beacon in fixed interval. Each packet reception has probability, called packet error rate 𝑃𝑃𝐸𝑅, to be failed due to surrounding environment effect such as raining or fogs. Thus, the channel reservation successful ratio P𝑅 can be expressed as following:
P𝑅 = 1 − (𝑃𝑃𝐸𝑅+ (1 − 𝑃𝑃𝐸𝑅) ∑2𝑦=0[( 𝑦 ) 𝑃𝑠𝑦(1 − 𝑃𝑠)𝑇𝑐𝑐ℎ−𝑦])max (
𝑙𝑅−𝑣
𝑣2𝐼𝐵 ∙ 𝐼𝐵1, 0)
(6) Table 3-1 Parameters for contention mathematical Analysis
λ 1 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠 𝑙𝑎𝑛𝑒𝑠 ∙ 𝑠⁄
V 25 𝑚⁄ 𝑠
𝐶𝑊𝑚𝑖𝑛 3 slots
𝐶𝑊𝑚𝑎𝑥 7 slots
Slot time 0.000013 s
B 1000 bytes
𝑅𝑇𝑥 3 Mb⁄ 𝑠
𝑡 0.05 s
22
Figure 3-4 Mathematical analysis of channel reservation successful probability
About channel allocation
In our ETC system, there are multi T-RSUs set in transaction zone for polling-based transaction process. T-RSUs have each own polling queue with variable polling queue length.
0
Length of Reservation Zone (m)
v = 60 km/hr.
Length of Reservation Zone (m)
v = 60 km/hr.
v = 90 km/hr.
v = 120 km/hr.
23
When vehicles entering reservation zone and requesting for channel reservation, R-RSU should decide a T-RSU to service them based on vehicle location and vehicle speed and each corresponding polling queue state. In Figure 3-5, vehicle A and vehicle B request for channel reservation. The SCH#2 is an intuitive solution for these requests due to the shorter polling queue length. In fact, we should consider not only polling queue length but also vehicle speed and vehicle location when designing a channel allocator. In a situation, vehicle C and vehicle D are also in polling queue #2 and almost leaving transaction zone without operating transaction process. Vehicles in polling queue #1 are far away from transaction zone with lower speed, means they will stay in transaction for a longer time. If we add vehicle A and vehicle B into polling queue #2, the T-RSU should spend time polling vehicle A and vehicle B even they are still far away from transaction zone. For vehicle C and vehicle D, they have fewer times to be polled due to longer polling queue length by adding vehicle A and vehicle B into it. It may decrease the transaction successful probability.
Figure 3-5 Channel allocation scheme
24
About polling scheduling
The goal of polling scheduling is increasing transaction successful probability. In polling based transaction process, transaction successful probability and the expected polling times are positive correlation, means the transaction successful probability is increasing due to more polling chance. A vehicle expected polling times is impacted by many factors such as polling queue length, vehicle speed and the number of T-RSUs. We conduct an analysis about transaction successful probability. We assume that there are 𝑁𝑠 T-RSUs and each T-RSU has the same polling queue length 𝐿𝑄 which can be obtained from equation (1) and equation (6):
𝐿𝑄 =𝑁𝑁𝑅 ∙ 𝑃
𝑠𝑐ℎ (7)
The polling based transaction process is three handshake processes, so time duration of one polling process 𝑡𝑝 equals to 3𝑡𝑇𝑥. The transaction process starts when vehicle entering transaction zone and receiving the first polling packet from T-RSU. This waiting time can is related to polling queue length 𝐿𝑄 and current channel interval. If vehicles enter transaction zone in SCHI, the time 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_ they only need to wait is relying on polling queue length:
𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_𝑠 = 12∙ 𝐿𝑄∙ 𝑡𝑝 (8)
If vehicles enter transaction zone in CCHI, they should wait until channel changing to SCHI.
𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_ = 12𝑡 + 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_𝑠 (9) The total time vehicles stay in transaction zone is equal to 𝑙𝑣𝑇 , the symbol 𝑙𝑇 is length
25
of transaction zone. If transaction zone is not long enough before vehicle receiving first polling packet, we say that the time duration 𝑡𝑠𝑡 vehicles spend for transaction process is 0.
We assume vehicle entering transaction zone in SCHI or in CCHI has the same probability.
Thus the time duration 𝑡𝑠𝑡 can be expressed as following:
𝑡𝑠𝑡 = 12(𝑚𝑎𝑥 {𝑙𝑣𝑇− 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_ , 0} + 𝑚𝑎𝑥 {𝑙𝑣𝑇− 𝑡𝑤𝑎𝑖𝑡𝑖𝑛𝑔_𝑠 , 0}) (10)
For compatibility of other existing commercial applications which also share SCH bandwidth, our ETC system should not use all of SCHI for contention free period. The ratio 𝑟 𝐹𝑃 is contention free period of SCHI that is reserved for our ETC system. Thus, we can obtain the expected polling times of one vehicle in transaction zone:
𝑝𝑜𝑙𝑙𝑖𝑛𝑔= 12𝑡𝑠𝑡∙ 𝑟 𝐹𝑃∙𝐿 1
𝑄 ∙ 𝑡𝑃 (11)
If there is one packet failed in three handshake process, the transaction is failed for this time. There are expected 𝑝𝑜𝑙𝑙𝑖𝑛𝑔 times three handshake process for transaction process.
Thus, the total transaction successful probability 𝑃𝑇 can be expressed as:
𝑃𝑇 = 1 − (1 − (1 − 𝑃𝑃𝐸𝑅)3)𝑇𝑝𝑜𝑙𝑙𝑖𝑛𝑔 (12) Table 3-2 Parameters for polling scheduling mathematical Analysis
𝑙𝑅 100 m
𝑙𝑇 5 m
𝑟 𝐹𝑃 0.4
26
Figure 3-6 Mathematical analysis of transaction successful probability
How to determine the lengths of the two regions
As shown in Figure 3-4, the longer reservation zone may not increase channel reservation successful ratio. The contention is the main problem. In longer reservation zone, the more vehicles will attempt to access medium and cause collision. Even vehicles have more time to reserve channel in long reservation zone, the effect of lower successful packet transmission probability is more serious. The other problem is polling queue length also increasing with length of reservation zone as shown in equation (7). In equation (11), the long polling queue length 𝐿𝑄 causes less expected polling times 𝑝𝑜𝑙𝑙𝑖𝑛𝑔 and decreases the transaction successful probability. However, channel reservation successful probability is low in short reservation zone shown as Figure 3-4. The collision is relieved due to fewer vehicles in shorter reservation zone, but the transmission times of channel reservation request are also decreasing.
27
In equation (10) and equation (11), the expected polling times and the transaction successful probability are increasing with long transaction zone. If ETC system using fewer T-RSUSs, the transaction zone should increase due to longer polling queue length in equation (7). However, the limitation of transaction zone is ALPR technology as described in above chapter.
28
Chapter 4
Design and Development
4.1. How to contention
The MAC layer of WAVE/DSRC follows 802.11e.In 802.11e, there are two medium access schemes: Enhance Distributed Channel Access (EDCA) and HCF Controlled Channel Access (HCFF). EDCA extends from DCF and supports Quality of Service (QoS). To support QoS, packets from upper layer send into one of four access categories with distinct transmission priority. The packets in access categories with higher transmission priority have shorter contention window. AC_BK is the access category with lowest priority, it usually accounts for transmission of network background flow. The second is AC_BE which means best effort for general network applications with no packet scheduling. AC_VI and AC_VO is the highest priority level for transmission of video or audio date.
In WAVE/DSRC, RSU broadcasts WAVE Service Advertisement (WSA) messages periodically with highest priority level AC_VO as service provider. If vehicles receiving WSA message are interest of that service, they send a request to RSU for service register.
The similar procedure so does our ETC system. In reservation zone, R-RSU broadcasts R-Beacon and response, and vehicles send back request and ack message. All of these messages have highest priority same as WSA message. There is no consideration of vehicles’
29
speed and locations in our ETC system because we desire our ETC system is compatible in WAVE/DSRC standard without modification.
4.2. Channel Allocation
WAVE/DSRC standard does not define how to decide an appropriate channel for a multi-channel access scheme. The major goal of our ETC system is reliability of transaction process. The SCHs that channel allocator assigns to vehicles should let vehicle have high probability to be polled for more times. The expected polling time is related to polling queue state, vehicle speed and vehicle location as described in above chapter. Here we apply load balance policy which only considers polling queue state in our ETC system. When vehicles requesting for channel reservation, channel allocator selects SCH number with the shortest corresponding polling queue length in T-RSU. If there are many T-RSUs have the shortest polling queue length, channel allocator will select by T-RSU ID in sequence. For example, in Figure 4-1, vehicle A, B, C and D reserved channel to R-RSU in sequence. The channel allocator first assigns SCH#3 which has lowest polling queue length to vehicle A. After added vehicle A into polling queue #3, polling queue #2 and polling queue #3 has the same queue length. When vehicle B requests for channel reservation, channel allocator assigns SCH#2 to vehicle B. Then it will assign SCH#3 to vehicle C after adding vehicle B into polling queue #2. Finally, the length of all there polling queue are the same. Thus channel allocator will assign polling queue #1 to vehicle D when it starts channel reservation.
30
Figure 4-1 Channel allocation method in our ETC system
4.3. Polling Schedule
As describing in above, WAVE DSRC follows 802.11e MAC layer. In 802.11e, there is no description about polling list management. In our ETC system, the polling scheduler of each T-RSU should decide the sequence of polling queue in run time or at end of CCHI. For guaranty of reliability, all vehicles should be polled at least one time before them leveling the transaction zone. In our ETC system, the polling scheduler runs at the end of CCHI and sorts queues based on expected deadline time of vehicles. If vehicles are going to leave transaction zone, then we will put them to the front of queue. For example, in Figure 4-2, there are four vehicles A, B, C, and D waiting to be transacted. Vehicle A has higher speed than vehicle C so that it will leave transaction zone faster. Thus the polling scheduler will put vehicle A in front of vehicle C in polling queue. Although vehicle B and vehicle D is not in transaction zone, we still put them into polling queue and try to poll them. The reason is that we cannot make sure when them will enter transaction zone due to GPS error and acceleration caused by drivers.
31
Figure 4-2 Polling schedule method in our ETC system
4.4. How to poll
Each time T-RSU sends polling packet to the vehicle which is on the top on polling queue and then move it to the end of queue. For example, in Figure 4-2, after T-RSU send polling packet to vehicle A and then move A to the end of queue. The sequence of polling queue becomes C, B, D, A. The T-RSU will delete vehicles in polling queue in two cases:
transaction success or times out. If there is no packet error in three handshake process between T-RSU and vehicles, the transaction process is complete and then T-RSU deletes vehicles from polling queue. Otherwise, T-RSU still keeps vehicle in polling queue even it has been passing transaction zone without being transacted. Thus, we should have some mechanisms to determine whether vehicles in polling queue are still in transaction zone or reservation. Here we use expected leaving time as evaluation criteria. If the expected leaving time of vehicles in polling queue plus a threshold ∆ is earlier than current time, we say that these vehicles have been time out without be transacted and delete them from polling queue.
The threshold ∆ is calculated by lower bound of speed limitation 𝑣𝑚𝑖𝑛 and length of total ETC system which equals to length of reservation zone 𝑙𝑅 plus length of transaction zone
32
𝑙𝑇 as following, where ɛis a system parameter to ensure correctness of system in GPS error situation.
∆= ɛ ∙𝑙𝑅𝑣 + 𝑙𝑇
𝑚𝑖𝑛 (13)
33
Chapter 5
Experimental Design
In this thesis, we focus on the system based on WAVE DSRC in normal environment even in harsh environment whether it is still reliable. Our ETC system will compare to the ETC system which using single channel contention based transaction scheme for most existing ETC systems. The compared ETC system uses only CCH to process transaction operation which just similar to our ETC system transaction operation on SCH. According to the experiment results, we will show that our ETC system is more appropriate for MLFF architecture and more reliable than single channel contention based ETC system. Then, the results also show the effect of varied length of reservation zone and the reliability in distinct numbers of T-RSU situations.
Sometimes, there is huge number of vehicles passing a toll gate at the same time. ETC systems should be able to service these overwhelming vehicles simultaneously and transact them as many as possible. In simulations, the vehicle arrival process is commonly modeled as Poisson process [19]. For Poisson process of vehicle arrival rate, the vehicle inter-arrival time distribution is an exponential variable. The vehicles arrival rate is defined as an average number of vehicles that appearing at each lane per second in this thesis. Base on exponential distribution, the cumulative probability function is shown as Figure 5-1. 1. The probability
34
that less than one vehicle arrival at a lane in one second is 64%. It still has 0.68% probability
that less than one vehicle arrival at a lane in one second is 64%. It still has 0.68% probability