Performance Enhancement of In-Vehicle 10BASE-T1S Ethernet Using Node Prioritization and Packet Segmentation

Following the appearance of electrical vehicles and autonomous driving, a new in-vehicle network architecture is required that should be able to process substantial sensor data and communicate with other vehicles or infrastructure. Ethernet is considered a promising technology for replacing existing communication networks due to its stability and large bandwidth. Among various types of Ethernet, 10BASE-T1S can play a significant role in connecting multiple nodes in a bus structure at each zone of the zone-based network architecture. Although its latency is reduced using the physical layer collision avoidance (PLCA) algorithm, it is not small enough to be adopted in safety and powertrain domains, which require a very small delay of less than a few hundred microseconds. Therefore, this study uses node prioritization and packet segmentation to overcome the limitations of the existing PLCA algorithm. The former changes the transmission sequence of nodes while the latter reduces the waiting time for a packet. This paper suggests the algorithms of these schemes and analyzes the performance.

verified in the industry. Among the diverse types of auto-23 motive Ethernet, 100BASE-T1 satisfies the requirements for 24 bandwidth, delay, synchronization and network management 25 of vehicular networks. It is good for implementing a net-26 work in a tree topology, which requires a lot of switches to 27 The associate editor coordinating the review of this manuscript and approving it for publication was Leandros Maglaras . connect many electronic control unit (ECU) nodes as shown 28 in Fig. 1(a) [3]. 29 However, many existing ECUs are connected in a bus 30 topology such as controller area network (CAN) and local 31 interconnect network (LIN). The bus network is suitable for 32 connecting many ECUs in the vehicle because it reduces 33 the cable length compared to the point-to-point structure. 34 The 10BASE-T1S that operates at 10 Mbps in the bus net-35 work can replace existing CAN and LIN [4]. It has other 36 advantages such as light cable weight, simple equalization; 37 it does not need forward error correction, echo cancellation 38 and hybrid circuit [5]. Therefore, it will be used efficiently 39 in a zone-based in-vehicle network as shown in Fig. 1(b), 40 which can reduce the number of ECUs and total cable length. 41 Zonal architecture divides the vehicle into several zones, 42 where each zone gateway communicates with ECU nodes 43 inside its zone. Since many existing ECUs are connected in 44 bus topology, 10BASE-T1S can be used for this purpose. 45 TABLE 1. Properties of in-vehicle networks [8].
Properties of the traditional in-vehicle networks are 55 explained in Table 1 [8] for the comparison with that of 56 10BASE-T1S. In particular, CAN could be used in the 57 transmission of delay-sensitive data by assigning a small ID 58 to a high-priority data [8], [9]. When a collision happens 59 among frames from different nodes, IDs of them are com-60 pared; only the frame with the smallest ID is allowed to con-61 tinue its transmission while other frames stop transmission. 62 In this way, urgent messages are transmitted with low latency 63 without interruption from other nodes in the same bus. 64 If 10BASE-T1S is to replace or cooperate with CAN, 65 its latency should be very small. Latency requirements for 66 in-vehicle networks are dependent on domains. Backbone 67 communication generally requires an end-to-end latency of 68 less than 10 msec [10], [11]. In contrast, the control loop sig-69 nal generated from an existing CAN device requires a latency 70 of less than 100 µsec. The additional delay should be con-71 sidered if the data pass through the gateway. Fig. 2 explains 72 the properties and latency requirements for each domain. 73 According to the figure, safety and engine/powertrain 74 domains have very tight latency requirements, whereas com-75 fort and Human-Machine-Interface (HMI) domains have a 76 relatively generous limit. Concerning the periodic messages 77 usually found in powertrain domains, the maximum allowed 78 delay is about 10 % of the period [12]. For instance, when the 79 period of a control loop signal for controlling a motor or an 80 actuator is 1 msec, the delay limit is 0.1 msec.

81
The 10BASE-T1S, which uses the same bus topology as 82 the existing CAN, utilizes the physical layer collision avoid-83 ance (PLCA) function, reducing the delay by preventing colli-84 sions in the physical layer. However, it is difficult to satisfy all 85 latency requirements because each node must wait for its turn 86 according to the round-robin-based protocol. Even urgent 87 nodes should wait until all preceding nodes complete their 88 transmissions. Assuming maximum Ethernet packet length 89 (i.e., 1,530 bytes) with 8 nodes and 10 Mbps data rate the 90 delay reaches up to 9.8 msec, which exceeds the delay limit of 91 engine/powertrain and safety electronics as shown in Fig. 2. 92 Therefore, we reduce the waiting time by assigning higher 93 priority to these delay-sensitive nodes. Even in this case, 94 the waiting time from the packet transmitted just before the 95 priority packet is inevitable. One way to decrease it is to 96 reduce the packet size. For this purpose, we propose a packet 97 segmentation. There are studies to achieve low latency in other layers 99 of vehicular communications. A network layer approach is 100 VOLUME 10, 2022 essential to secure the performance in inter-vehicle network.  according to the priority of nodes and their messages [16].

124
If a high-priority message occurs in a sequence of low priority 125 messages, the cycle is restarted to transmit the high-priority 126 message first. However, data with low priority, in this case, 127 can experience too much delay due to the extended cycle 128 length. A study has used the concept of priority in CAN-over-129 PLCA Ethernet [17]. In this case, however, only one priority  The high-speed Ethernet backbone connects each zone to 159 the vehicle server [7]. Ethernet backbone can provide deter-160 ministic, high bandwidth and fault-tolerant connectivity [18]. 161 The zone gateway, positioned at the center of each zone, 162 is connected to many ECUs inside the zone through the 163 automotive Ethernet in a star or bus architecture as illustrated 164 in Fig. 3. Each node in the 10BASE-T1S Ethernet can operate 165 as an independent PLCA node or gateway for the CAN or 166 CAN-FD (flexible data rate) bus. If it is available at very low 167 cost, then it will replace CAN or CAN-FD; but at the present 168 time, it is likely to coexist with them via CAN-to-Ethernet 169 conversion at the gateway [19].  whereas the payload includes the data and overhead of the 174 CAN-FD. The CAN protocol guarantees low latency to delay-175 sensitive data through the arbitration process using the CAN 176 ID. However, PLCA does not have this function, and low 177 latency cannot be achieved for time-critical data. This paper 178 introduces node prioritization and packet segmentation in 179 PLCA to satisfy the delay requirement. This concept is not 180 limited to the zone-based structure, but more applications can 181 be found in this architecture.  and (2). In these equations, R b denotes the data rate, L beacon 220 is the BEACON timer length, L to represents the timeout 221 timer length counting the TOs to yield, L commit denotes the 222 COMMIT signal length, and L data,max refers to the maximum 223 Ethernet packet (i.e. 1,530 bytes, including the header and 224 frame check sequence). When the data rate is 10 Mbps, and 225 N is 7, the minimum cycle length corresponds to 27.6 µsec. 226 Under the same conditions, the maximum length becomes 227 9.8 msec, too large for delay-sensitive data. The existing 228 PLCA scheme cannot meet the hard latency requirements 229 in Fig. 2; thus, this study introduces a priority-based PLCA 230 algorithm.
The delay that occurs from the PLCA protocol may exceed 237 the latency requirement of some ECU node. In order to over-238 come this problem, nodes in a 10BASE-T1S bus are classified 239 into regular or priority nodes. Delay-sensitive nodes such as 240 engine/powertrains or safety electronics can be categorized 241 as priority nodes; a unique priority number is stored at each 242 node.

243
The priority assignment in this study utilizes the inter-244 packet gap (IPG) to indicate the priority request information. 245 The IPG is the period of at least 12 bytes after each Ethernet 246 packet. The transmitting node is prohibited from sending data 247 during this period. This IPG is divided into multiple request 248 slots and a control message slot as depicted in Fig. 7. The 249 formers are used to request priority transmissions and the 250 latter to send PLCA control messages. Each priority node 251 can send transmission request messages in its assigned slot 252 when it has data to send. Then, it sends a COMMIT mes-253 sage in the control section and the Ethernet packet. In this 254 way, the priority node can send its payload prior to other 255 nodes. In this process, once the higher-priority section is 256 used, the lower-priority sections should be left empty to avoid 257 transmission conflict. If there is no priority node to send a 258 message, the node in the original round-robin sequence takes 259 the TO. In this paper, section IV analyzes the performance 260 by assuming two priority nodes, which usually transmit short 261 packets from the CAN or CAN-FD bus. This study assumes that the bus comprises eight nodes, 263 where two nodes are assigned priority levels. The node with 264 ID 0 (or Node 0) is the primary node, and Nodes 3 and 5 are 265 assumed to be priority nodes. Node 3 has higher priority and 266 can send a request message in the Priority-1 slot, whereas 267 VOLUME 10, 2022 Node 5 can use the Priority-2 slot in Fig. 7. The operation 268 of the PLCA in the absence of a priority packet is depicted 269 in Fig. 8(a), where each node has a TO in sequence. If it has 270 data, it sends the COMMIT message first, followed by the 271 Ethernet packet. If it has no data, the SILENCE period occu-272 pies the slot. Once the Ethernet packet is sent, the IPG period 273 follows, where the priority request and control messages are 274 included. After the transmission of Node 2 in Fig. 8(a), Node 275 3 surrenders its opportunity during the IPG period, and Node 276 4 takes the TO and sends data.

277
If a priority node has data to transmit, it sends a request 278 message to have the TO during the coming IPG period. The latency of the proposed scheme can be calculated as 287 follows. A higher-priority node is referred to as P#1, whereas 288 a lower-priority node is P#2. It is assumed that the generated 289 data are stored in a buffer and transmitted when a TO is given.  where PLCA node no denotes the number of nodes in the 305 PLCA bus, and L priority,max denotes the maximum-sized 306 packet of the priority nodes. If R b is 10 Mbps, PLCA node no 307 is 8 bits, L to is 32 bits, L beacon is 20 bits, L commit is 5 bits, 308 L data,max is 1,530 bytes, L priority,max is 121 bytes and IPG 309 is 12 bytes, then, the maximum latencies of P#1 and P#2 310 are 1.3 and 1.36 msec, respectively. L priority,max is assumed 311 to be 121 bytes because the maximum size of the CAN-312 FD frames is 91 bytes, and the header and frame check 313 sequence in its Ethernet mapping correspond to 30 bytes, 314 as presented in Fig. 4 [20]. Compared to the maximum cycle 315 length of 9.8 msec in conventional PLCA, the latency of the 316 proposed priority node is reduced to 1.36 msec, almost 1/7 317 of its original value. Minimum latency occurs when data are 318 generated just before their TO, as indicated in Fig. 9. It is 319 calculated as in Eq. (5), the same as in the existing PLCA.

322
The previous section states that node prioritization can reduce 323 the latency of urgent nodes. However, the reduced latency is 324 not small enough to satisfy the requirements of ECUs in all 325 domains shown in Fig. 2. Therefore, the concept of packet 326 segmentation is proposed in this section to further reduce the 327 latency.

329
By introducing priority nodes in 10BASE-T1S Ethernet, the 330 maximum latency can be reduced to 1.36 msec, enabling the 331 network to support more services listed in Fig. 2. However, 332 the reduced latency still does not satisfy the requirements 333 of engine/powertrain-related ECUs. The limit in the latency 334 reduction is attributed to the Ethernet packet size. As the 335 10BASE-T1S Ethernet uses the CSMA/CD protocol, even 336 a high-priority node must wait until the end of the current 337 data transmission. Therefore, the packet length affects the 338 delay, and we propose packet segmentation to further reduce 339 the delay. Unlike priority data, which usually have a short 340 length, a normal Ethernet packet has a long payload of up 341 to 1,530 bytes. According to the proposed segmentation, 342 a packet longer than a predefined size is divided into multiple 343 segments. Then, each segment is transmitted in one cycle. 344 Therefore, the segment size should be determined by consid-345 ering the required latency. Packet segmentation has been used 346 in Gigabit Passive Optical Network (GPON), TTEthernet 347 amounts to 20 bytes while the size of the Ethernet packet is 387 reduced due to the segmentation, it makes the regular nodes 388 experience more latency. Therefore, the size of a segment 389 should be chosen considering latency boundaries of both the 390 priority packets and the regular packets. The effect of the 391 segment sizes is analyzed in section IV.

393
This section explains the implementation algorithm for the 394 proposed packet segmentation in more detail. Fig. 12 presents 395 the operation of the reconciliation sublayer, where func-396 tion blocks of PLCA are placed [4]. In addition to the 397 existing PLCA functions, two more blocks, the SEG-398 MENT_CTRL and REASSEMBLY_CTRL blocks indicated 399 in color blocks, are added to implement the segmentation. 400 The SEGMENT_CTRL block is responsible for the packet 401 segmentation before transmission. If the packet from the PLS 402 is longer than the predefined size, it is segmented and stored 403 in a buffer. Once it takes the TO, it sends a tx_cmd signal, such 404 as COMMIT, node ID and segment information, followed by 405 a segment via the TXD <3:0> bus. This block sends the COL 406 signal to the PLS until the buffer is empty so that the MAC 407 does not transmit a new packet. The REASSEMBLY_CTRL 408 block saves the received segments and reassembles them 409 into a complete packet. For this purpose, it contains multiple 410 buffers to store different segments according to their node 411 IDs. Checking the segment information messages determines 412 whether to reassemble the segments in the buffer and forward 413 them to the MAC layer.

414
FIGURE 12. Operation of the reconciliation sublayer for segmentation. Fig. 13 illustrates the state diagram of the PLCA control, 415 which is executed at the PLCA_CTRL_FSM block in Fig. 12. 416 The CHECK_PRIORITY block is connected to the trans-417 mission and reception blocks, and the priority timer works 418 during the IPG period. If there is a priority packet to transmit 419 at this node, it waits for the IPG interval assigned in the 420 SEND_PRIORITY block. Then, it transmits its request mes-421 sage during the IPG and transmits the packet. However, if a 422 higher priority node transmits request message in the same 423 IPG interval, the lower priority node gives up transmitting and 424 VOLUME 10, 2022  transmits its node ID in the NODE_ID block, and sends the 433 segment information in the SEGMENT block so that the 434 receiving node checks if the received packet is segmented 435 or is the last segment. Finally, the transmitting node starts 436 sending the SYNC signal in the SYNC1 block.

437
The two algorithms described above, priority allocation 438 and segmentation, may result in additional computational 439 complexity. The transmitter needs to send the packets within 440 the segment size, and the receiver should reassemble multiple 441 packet segments. For this purpose, each node should own 442 additional queues, count the packet length, and check the 443 IDs of each received segments. Although this additional com-444 plexity should be considered in the design of the algorithm, 445 it is expected that the advantage of latency reduction is large 446 enough to compensate for the complexity in the bus topology 447 of vehicular network.

449
In this section, we analyze the performance of the proposed 450 PLCA algorithm. The analysis parameters are described in 451 Table 2, and MATLAB was used in the simulation. The 452 number of nodes in the PLCA bus is eight, and the data rate 453 is 10 Mbps. The beacon timer, which waits for a BEACON 454 signal, is set to 20 bits, and the TO timer, which waits for the 455 TO, is set to 32 bits. The lengths of COMMIT and PRIORITY 456 are 5 bits each. The payload length of a priority node is 457 set to 42 to 91 bytes considering the CAN-FD mapping in 458 Fig. 4, where the CAN-FD frame has a maximum of 91 bytes, 459 including bit stuffing. If the length of the priority packet is 460 increased, then it will cause more delay at the lower priority 461 nodes and regular nodes. However, the effect will not be 462 considerable since the number of priority nodes is limited. 463 The payload length generated at other regular nodes is defined 464 as 42 to 1,500 bytes. Each node is configured to have a 465 100-Kbyte queue.  Fig. 15 represents the pseudocode of the simulation, 467 describing the data transmission process by PLCA, employ-468 ing prioritization and segmentation. Each node generates 469 30,000 packets according to the Poisson distribution and 470 updates the queue using those packets by the variable 'cur-471 rent time'. Each node has a TO in a round-robin manner 472 for T seconds. The node with the TO is called the 'current 473 node', and the node with high priority is the 'priority node'. 474  node is 1.3 and 1.36 msec, respectively. For regular nodes 507 in the simulation, the average delay is 0.36 msec and the 508 maximum delay is about 4 msec when the load is 0.5. These 509 results reveal that the proposed PLCA is effective for multiple 510 delay-sensitive ECU nodes demanding delays of less than 511     The throughput of the bus is saturated rapidly with the small 541 segment size because a packet is divided into many seg-542 ments, and each requires overhead parts. The throughput is 543 two priority nodes compared to the single priority case at the 546 same load value since priority data have smaller packet size. 547 However, the throughput is proportional to the load before a 548 load of 0.7; thus, it is not degraded even after prioritization 549 and segmentation. This outcome is attributed to the PLCA 550 that prevents packet collision in the bus.