Time-Synchronization Method for CAN–Ethernet Networks with Gateways
Round 1
Reviewer 1 Report
The paper is interesting, well-written and clearly presented. The topic of time synchronization for in-vehicle applications is very relevant, especially with the advent of autonomous driving systems and applications.
The structure of the paper is good and the different sections provide enough information to be well understood. However, the main section that could be improved in the article is the evaluation and results part, since only a limited number of nodes is used to test the proposed architecture/protocol and a few number of measurements is conducted. If possible, this evaluation should be extended, to verify the variation of the time synchronization metrics with the increasing number of nodes in the hybrid network.
Another point is that it should be made clear is how the proposed solution for applying IEEE 1588 in the CAN network compares with previous literature, namely the work mentioned in reference [15]. In addition, the same clarification should be provided for the hybrid architecture comparison with the paper referrenced in [16], despite the different technologies used.
Author Response
Response to Reviewer 1 Comments
Point 1: The paper is interesting, well-written and clearly presented. The topic of time synchronization for in-vehicle applications is very relevant, especially with the advent of autonomous driving systems and applications.
The structure of the paper is good and the different sections provide enough information to be well understood. However, the main section that could be improved in the article is the evaluation and results part, since only a limited number of nodes is used to test the proposed architecture/protocol and a few number of measurements is conducted. If possible, this evaluation should be extended, to verify the variation of the time synchronization metrics with the increasing number of nodes in the hybrid network.
Response 1: I agree that evaluation with a realistic network would present more convincing argument for the efficacy of the proposed method. It was mainly due to limited resources of the lab, i.e., not enough CAN slave nodes and an oscilloscope with only four channels.
Nonetheless, more CAN nodes will adversely affect the accuracy because, as shown in Figure 5, t2*-t2 < t2**-t2. Because CAN network length rarely exceeds 20 m, we repeated the experiment with longer cable (15m) for CAN slave 3. Following your remark on the number of measurements, we repeated ten times. The new results showed that the average error increased from 6.704 to 8.184 us while the worst-case error increased from 8.68 us to 9.68 us. Comparison shows that the synchronization error increased, as you expected, but I think the increase is not big enough to nullify the efficacy of the proposed method.
To incorporate your comment, I have replaced the experimental results with those with longer cable and more repeated measurements in Section 4.
Point 2: Another point is that it should be made clear is how the proposed solution for applying IEEE 1588 in the CAN network compares with previous literature, namely the work mentioned in reference [15]. In addition, the same clarification should be provided for the hybrid architecture comparison with the paper referrenced in [16], despite the different technologies used.
Response 2: I think that the results cannot be directly comparable because reference [15] deals with only CAN nodes while our synchronization is done over two networks connected by a gateway. Anyway, the synchronization error of [15] is 160 us while that of [16] is 6 us.
Thank you very much for your valuable comments.
Reviewer 2 Report
This paper presents the adaptation of the IEEE 1588 precision time protocol (PTP), initially developed for Ethernet networks to a heterogeneous network composed of an Ethernet and a CAN network linked by a gateway.
The protocol has been modified to take into account the gateway transmission delay.
An experimental platform composed of a master Ethernet node and three CAN slave nodes has been evaluated. Measures show that the modified protocol has a synchronization precision of 10 microseconds. Results obtained are the same as the AutoSAR synchronization protocol for CAN networks.
This paper is built and focused on a single case study studied in depth. As a consequence, it creates a feeling of lack of generality.
As your study belongs to the domain of autonomous driving technologies, you do not justify the choice of these two networks among those used in vehicles. Indeed, the classical communication architectural schema of vehicles is mainly hybrid. It is often composed of CAN, LIN, FlexRay, MOST and Ethernet networks linked by a gateway.
How could you adapt your work to such communication vehicle architectures?
How could you extend your work to several subnetworks?
There is no related work of time synchronization protocols. Please, add a related work section.
Page 2. The PTP protocol is shortly compared with only AutoSAR synchronization protocol.
The choice of the PTP protocol is justified due to it nanoseconds scale precision, whereas AutoSAR is limited to 10 microseconds. However, the experimental results of you work have too a 10 microsecond time precision. The results obtained are not better than AutoSAR results. Please, justify differently your choice.
The Ethernet node is the synchronization master of your protocol. Please, justify this choice. Could it be possible that the synchronization master will be a CAN node?
Is it possible that the synchronization master will be located on the gateway? In that case, the gateway will be the central synchronization module of all other subnetworks.
There is no synchronization node on the Ethernet network, except the master node? Could you explain why? Is it possible to add such nodes in your experiment?
Some misspellings:
Line 20: “(CAN) and Ethernet. .”
Line 22: “to be implementED”
Line 42: “provides g high bandwidth”
Author Response
Response to Reviewer 2 Comments
Point 1: This paper presents the adaptation of the IEEE 1588 precision time protocol (PTP), initially developed for Ethernet networks to a heterogeneous network composed of an Ethernet and a CAN network linked by a gateway.
The protocol has been modified to take into account the gateway transmission delay.
An experimental platform composed of a master Ethernet node and three CAN slave nodes has been evaluated. Measures show that the modified protocol has a synchronization precision of 10 microseconds. Results obtained are the same as the AutoSAR synchronization protocol for CAN networks.
This paper is built and focused on a single case study studied in depth. As a consequence, it creates a feeling of lack of generality.
As your study belongs to the domain of autonomous driving technologies, you do not justify the choice of these two networks among those used in vehicles. Indeed, the classical communication architectural schema of vehicles is mainly hybrid. It is often composed of CAN, LIN, FlexRay, MOST and Ethernet networks linked by a gateway.
Response 1: The initial idea of this paper came from our participation in an autonomous bus project. Our role was to design a network for various sensors and controllers. Many devices developed by traditional automotive industry were quipped with CAN interface while those of relatively new technology like LIDAR were with Ethernet interface. This reflects the fact that CAN is almost a de facto standard for automotive applications. And because some of relatively new type of sensors need to send raw sensor data without processing, high data rate is required, which CAN cannot provide and are available with Ethernet. With data via these two protocols, perception software wants time of measurement for more accurate sensor fusion. This can be easily done on the Ethernet side because synchronization is a part of TSN. However, there is no notion of common time among CAN nodes.
To incorporate your comment, I have revised several paragraphs in the introduction to convey the following description.
CAN is widely used like a de facto standard and has no provision for shared notion of time among nodes. FlexRay and MOST can also be used for autonomous vehicle, but they are not as common as CAN and already have their own synchronization mechanisms. LIN is usually for slow devices such as electric seat motors, thus is not considered in this research.
Point 2: How could you adapt your work to such communication vehicle architectures?
Response 2: I don’t think that our work can be directly applicable to such architectures. Instead, we need some type of synchronization procedure among time masters, i.e., grand master of Ethernet, Timing Master of MOST, Time Guardian of FlexRay.
Point 3: How could you extend your work to several subnetworks?
Response 2: If several CAN networks are connected with an Ethernet network, the developed procedure can be periodically repeated for each CAN-Ethernet pair. However, if several CAN networks are connected in tandem and one CAN network is connected with an Ethernet, we need to do more research for this case.
Point 4: There is no related work of time synchronization protocols. Please, add a related work section.
Response 4: We have extended the 4th paragraph of the introduction to incorporate your comment.
Point 5: The PTP protocol is shortly compared with only AutoSAR synchronization protocol. The choice of the PTP protocol is justified due to it nanoseconds scale precision, whereas AutoSAR is limited to 10 microseconds. However, the experimental results of you work have too a 10 microsecond time precision. The results obtained are not better than AutoSAR results. Please, justify differently your choice.
Response 5: In fact, we had no choice but PTP because it is bult in with TSN Ethernet. We tried to extend the usage of PTP toward CAN area. Even though PTP guarantees 1 us precision for TSN, it is expected that the precision will deteriorate for heterogeneous network with a gateway. The precision of AutoSAR is about the same as our precision, but I’d like to point out that AutoSAR synchronization is only on a CAN network while the proposed one is over a CAN-Ethernet network. In addition, AutoSAR may not be implemented on all CAN nodes due to its cost and computing requirements.
Point 6: The Ethernet node is the synchronization master of your protocol. Please, justify this choice. Could it be possible that the synchronization master will be a CAN node?
Response 6: PTP provides a procedure to select a node with the highest clock precision (grand master). Therefore, there is always a synchronization master on the Ethernet part while there is no such mechanism on CAN. It is not impossible to have a CAN time master, but it requires further research to accommodate such a case.
Point 7: Is it possible that the synchronization master will be located on the gateway? In that case, the gateway will be the central synchronization module of all other subnetworks.
Response 7: I think that is quite possible. As I mentioned that there is a procedure for Ethernet nodes to compete to be a grand master, the gateway as a node on the Ethernet can be the grand master and it is expected to reduce synchronization error on the CAN network.
Point 8: There is no synchronization node on the Ethernet network, except the master node? Could you explain why? Is it possible to add such nodes in your experiment?
Response 8: Yes, there can be other nodes on Ethernet. They synchronize following the master node on Ethernet. But, we ignored them because they didn’t affect the synchronization performance for CAN-Ethernet network.
To reflect your comments on the direction of future research, I have added a sentence in the last paragraph of the conclusion section.
Point 9: Some misspellings:
Line 20: “(CAN) and Ethernet. .”
Line 22: “to be implementED”
Line 42: “provides g high bandwidth”
Response 9: This part has been modified in the paper.
Thank you very much for your valuable comments.
Please see the attachment.
Author Response File: 
Author Response.docx
Round 2
Reviewer 2 Report
Remarks and comments have been correctly taken into account
