3.1. Identification of Use Cases
As noted in the introduction, cross-border contexts create challenging situations for CAM, specially regarding service continuity. In order to plan the infrastructure and technology required for operating 5G-enabled CAM in cross-border roads it is necessary to conduct some experiments. It is a common approach in engineering to define a set of use cases that represent the most significant interactions with the system under test. 3GPP identified a set of V2X scenarios grouped into five categories [
32]: advanced driving, vehicle platooning, extended sensors, remote driving and vehicle QoS support.
Vehicle platooning aims at green driving by grouping a set of vehicles to travel together one after the other. The vehicles that are part of the platoon exchange periodic data to move in a cooperative way. Here, autonomous vehicles can automatically join and leave platoons. Autonomous platooning is expected to be adopted first by trucks, which are some of the main users of cross-border corridors. Platooning optimises transport by using roads more effectively, reducing traffic jams and consequently delivering goods faster.
“Extended sensors” is focused on extending the perception obtained by the onboard sensors, with sensor data received from surrounding vehicles or road side units (RSUs). This way, vehicles generate an enhanced perception of the environment beyond what their own sensors can detect. High-resolution data streams produced by cameras and Lidars impose highly demanding communication requirements. Extended Sensors may require querying data from distant RSUs or vehicles. This could imply communicating with devices located in different countries in a cross-border scenario.
Remote driving enables a remote driver or a V2X application to operate a remote vehicle. In this CAM use case, a remote operator takes control of the vehicle when a breakdown or complex environment impedes the autonomous vehicle’s trajectory. The remote operator can be located in a different country from the vehicle position, so the constraints on a cross-border context may present operational limitations.
Advanced driving implies complex manoeuvres such as overtaking or cooperative collision avoidance that require sharing the driving intentions with the vehicles in proximity. Poor performance of the communication pipeline in terms of latency or reliability may lead to a decrease in the level of automation and/or increase likelihood of accidents. Such a safety-critical use case needs to be tested in as many situations as possible, including roaming or supporting MECs when located in different countries.
Vehicle QoS support could be considered a horizontal use case, whereby the vehicle is able to anticipate network QoS fluctuations and adjusts its service requirements accordingly. Moreover, the network is able to allocate connectivity resources to preserve a QoS to satisfy application needs and the application can adapt its traffic demands to meet network performance. The overall goal is to offer a smooth user experience, meaning the coordination of different network infrastructures of both sides in a cross-border context to perform seamless transitions.
3.2. Use Cases’ Requirements
Each of the identified use cases can be implemented with different levels of automation while they all require high performance parameters in terms of connectivity. In this paper we focus on highly automated driving, that is, SAE L4 and L5. Here, the requirements are even more demanding as there is no handover to manual driving.
The spatial- and time-accuracy required by L4 and L5 autonomous vehicles with respect to object localisation are key factors when translating the nominal performance of communication technologies to operational parameters of use cases. The latency in the positioning messages received from another vehicle or entity adds uncertainty to the transmitted localisation value, with a higher impact on the longitudinal direction. This effect is depicted in
Figure 1 for the longitudinal error.
The localisation requirements for autonomous vehicles were studied in [
33], and they concluded that a maximum lateral error of 0.57 m and a maximum longitudinal error of 1.40 m are acceptable for passenger vehicles travelling on highways—the most common road type at cross-border corridors. Considering that state-of-the art ego-vehicle localisation methods have in general sub-metre accuracy in the order of several decimetres [
34], there is little room for positioning error induced by network latency. To study the potential effect of the latency on the localisation error, we have measured the position error induced by different latency values using a public driving dataset. More specifically, we have used three highway scenarios from the CommonRoad dataset [
35]. The scenarios were partly recorded from real traffic and partly hand-crafted to create challenging situations. The selected scenarios include lane merges and curves to make them more challenging. More information about the selected CommonRoad dataset scenarios can be found in
Table 2.
The dataset provides position, velocity, acceleration and heading values of each vehicle recorded at 10 Hz, which is the standard frequency for sending this kind of vehicle data [
36]. In this study and based on SAE J2945/1 [
37], the receiver estimates the current position of the transmitter vehicle based on the latest received message assuming that the transmitter vehicle is moving at a constant acceleration and heading. For each vehicle position data in the dataset, we have estimated the position after a certain time lapse that would correspond to the message latency. The position error added by the latency is calculated as the Euclidean distance between the vehicle position estimation and the actual position at the time that the message reaches the receiver. This ground truth is obtained by using a quadratic interpolation between the closest points that are part of the dataset.
As specified by the European Committee for Standardization (CEN) and European Committee for Electrotechnical Standardization (CENELEC) [
38], positioning accuracy is represented with a set of three statistical values given by the 50th, 75th and 95th percentiles of the cumulative distribution function (CDF) of the position error. The empirical CDF obtained in our study with the datasets of
Table 2 is depicted in
Figure 2 for latencies in the range of 1 to 100 ms. A latency higher than 100 ms would make impossible the real-time processing of the received data, this being the maximum tolerable latency for vehicle-to-vehicle (V2V) communication [
39]. For a latency of 25 ms or below, the error’s 95th percentile is below 5 cm, which is negligible for a localisation problem. With 50 ms latency, the 95th percentile is almost 10 cm, and reaches 18 cm with 100 ms latency. Depending on how close the transmitter vehicle’s positioning measurement errors are to the error bounds, these decimetre level errors added by the network latency can definitely affect to the localisation of the transmitter vehicle in the receiver vehicle’s frame. It is also important to consider that even if the network latency does not increase the localisation error outside of the error bounds, the positioning message needs to be received with enough time in advance to trigger a timely reaction.
3GPP defined some performance requirements in [
32] for use cases involving highly automated driving that are supposed to consider all these aspects. The performance requirements are summarised in
Table 3. Note that the latency requirements are in line with previous calculations; thus, a latency lower than 50 ms is required by all use cases, and it is much lower for advanced driving and remote driving due to their safety-critical nature and for extended sensors because of the required synchronisation and alignment of perception data coming from different sensors (e.g., video stitching or Lidar point cloud fusion). Extended sensors stands out as the most demanding use case in terms of data rate and communication range. Previous 5G-V2X requirements compiled by 3GPP noted modest data rate requirements for remote driving under the assumption that a vehicle only transmits lightweight processed data [
40]. However, more advanced remote driving approaches involve transmission of raw sensor data, so the whole onboard sensor suite can be processed in a remote server. This would of course require some data rate requirements similar to or even more demanding than extended sensors. Advanced driving requires a high transmission rate of 100 messages per second. This is exactly the same recommendation given in [
33], derived as the time required between successive localisation updates. Vehicle platooning does not stand out in any of the performance indicators proposed by 3GPP, but the requirements as a whole are still very demanding.
The latency of V2V communication under a LTE network for multi-operator environments with regional split was studied in [
41] and a latency of 58 ms was estimated in inter-operator communications without inter-operator handover. In the same work, the inter-operator handover, when vehicles have to detach from one operator and then attach to the other one, was estimated to have 300 ms of latency. It is then clear than enhanced 5G features are required to meet the demanding latency requirements.
To sum up, the identified CAM use cases have demanding performance requirements that cannot be met by LTE infrastructures. This is even more clear in a cross-border context that requires roaming. The following section proposes some implementation options to overcome this issue that exploit features present at 3GPP Release 16 and beyond.