1. Introduction
Various 5G wireless technologies were standardized and developed by academia and industries related to the field of telecommunications, and are now ready to be deployed with several key features such as enhanced mobile broadband (eMBB), ultra reliable low latency communications (URLLC), etc. A new radio for the 5G wireless networks is expected to use various radio frequencies, above or below 6 GHz. As compared to the 28, 60, and 73 GHz spectrums, the frequencies near the 3 GHz spectrum receive considerable attention from service providers because the millimeter Wave (mmWave) suffers from significant pathloss and link blockage caused by obstacles (e.g., buildings, vehicles and human beings) owing to severe penetration loss and reflection. However, it is advantageous to still consider the mmWave spectrum to obtain several hundred mega-hertz (MHz) of bandwidth because of the scarcity of spectrum below 6 GHz. Thus, measurement campaigns and demonstrations of mmWave communications have been conducted in real testbeds [
1,
2,
3]. Subsequently, mmWave channel models are established based on these measurement results [
4,
5,
6,
7].
The 5G era has reserved the development of various applications and services for the near future as they require gigabit data rate, fast response (less than 1 ms for end-to-end delay), and massive connections. One such interesting application is vehicular communication for autonomous driving and in-vehicle infotainment. Autonomous driving systems primarily rely on sensor devices equipped on vehicles, such as LiDAR, near-medium range radar, long-range radar (LRR) and cameras. However, communications between vehicles or a vehicle and an infrastructure such as a road-side unit (RSU) are essential to control a fleet of vehicles to avoid chain-reaction crashes instead of controlling individual cars. Moreover, an autonomous vehicle has rich data, such as high-resolution video (e.g., UHD 4K) from dash cameras, which can be used to improve vehicle control precisely. For instance, the curvature recognition of the road in front of the vehicle can be improved from 30 m to 50 m in distance ahead if using 2 MP instead of 0.3 MP data [
8].
In the last decade, dedicated short-range communications (DSRC) and IEEE 802.11p/wireless access in vehicular environment (WAVE) standards have guided vehicle-to-vehicle (V2V) communications, which enables vehicles to exchange safety messages to alarm or notify vehicle status and infotainment data using a 10 MHz bandwidth channel that supports up to 27 Mbps data rate. Further, the third generation partnership program (3GPP) introduced long-term evolution (LTE) based vehicle to infrastructure (V2I) communication and device-to-device (D2D) communication for the V2V communication. However, neither technologies is sufficient to transmit the increasing sensor data of future smart cars. Accordingly, 3GPP [
9] is developing a new standard applying the 5G radio (e.g., mmWave) to vehicular communications in Release 19. Recently, various academic studies on mmWave-based vehicle-to-everything (V2X) communications were conducted. Practical mmWave vehicular testbeds have been built to evaluate the performance of V2I or V2V [
10,
11]. Moreover, the channel models of the V2X communications are exploited in [
12,
13,
14,
15] and several ideas to reduce beam alignment and training overhead are proposed in [
16,
17,
18,
19]. The problems of mmWave beam alignment and width selection to reduce interference among vehicles are solved by a distributed algorithm in [
20,
21,
22]. Kim [
8] proposed a channel assignment algorithm for mmWave V2V beams to avoid reciprocal beam interference.
The IEEE 802.11ad standard, also known as the Wireless Gigabit Alliance (WiGig), which specifies the physical and link layers for mmWave communications, was released approximately a decade ago to provide a gigabit connection between proximate devices. The WiGig uses a 60 GHz mmWave spectrum and supports more than 4 Gbps data rate with 2 GHz channel bandwidth. Furthermore, indirect communication using a relay node is possible when a link blockage occurs; a sending station chooses a relay node that has a line-of-sight (LOS) link to a receiver. In this study, we investigated the applicability of the IEEE 802.11ad standard to a gigabit V2V communication (GiV2V) system in a vehicular testbed, for which we utilized off-the-shelf device modules based on the IEEE 802.11ad standard and installed them on the roof of the vehicles. A LOS direct connection satisfying the IEEE 802.11ad requirements between two vehicles was configured and the throughput and connectivity were measured while driving on the roads. This measurement was performed on low-speed campus roads and high-speed city roads, respectively, to evaluate the effect of vehicle mobility. To the best of our knowledge, this study was the first attempt at the measurement of V2V communications using commercial IEEE 802.11ad modules.
The highlights of our contributions in this study are as follows:
We conducted mmWave V2V communications using commercial IEEE 802.11ad modules.
We analyzed inter-vehicle connectivity by mmWave of short-range radio.
We compared the mmWave V2V communications in different driving environments.
According to our experimental results, the IEEE 802.11ad modules perform only in a short radio range of approximately 10–20 m at boresight due to significant pathloss, which causes frequent disconnections between vehicles, especially during high mobility. However, the average disconnection time is approximately 1 s, irrespective of the vehicle speed, even though the deviation of disconnection differs based on the speed; thus, high-speed vehicles demonstrate a larger deviation of disconnection time than low-speed vehicles. Regardless of the frequent disconnections, the experimental results demonstrate that the mmWave connectivity can transmit a large amount of data within a few seconds in the future smart cars.
The remainder of the paper is organized as follows. In
Section 2, we introduce the background and related works on mmWave and vehicular communication technology. We describe the GiV2V system in
Section 3.
Section 4 describes our experimental configuration, followed by the results in
Section 5. Finally, we discuss and conclude our study in
Section 6.
2. Related Works
In the last decade, the mmWave spectrum was popularly explored to enlarge the bandwidth and increase the throughput in mobile communications for a last decade. Rappaport et al. introduced seminal results of the experiments using a wideband sliding correlator channel sounder with steerable directional horn antennas at both the transmitter and receiver in New York city in [
1] from 2011 to 2013, and presented more results on the frequency bands of 28, 38, and 73 GHz bands in [
2,
3].
Based on those studies, mmWave channel models are demonstrated in [
4,
5], where empirically-based propagation channel models are proposed for the 28, 38, 60, and 73 GHz mmWave bands. Samimi et al. presented a 3-D statistical channel impulse response model for urban LOS and non-LOS (NLOS) channels developed from the 28 and 73 GHz ultra-wideband propagation measurements presented [
6,
7]. They later added small-scale fading measurements for the 28 GHz outdoor mmWave ultrawideband channels using directional horn antennas [
23]. Additionally, 28 GHz wideband propagation channel characteristics for mmWave urban cellular communication systems are presented in [
24].
Sun et al. characterized a mmWave indoor propagation channel based on a wideband measurement campaign at 73 GHz in an office-type environment [
25]. For outdoor access channel modeling at 60 GHz mmWave spectrum, Weiler et al. [
26] established a quasi-deterministic channel model and a link level-focused channel model and, subsequently, the authors of [
27] introduced a new quasi-deterministic (Q-D) approach for modeling mmWave channels, which allows natural description of scenario-specific geometric properties, reflection attenuation and scattering, ray blockage, and mobility effects.
Andrews et al. provided a comprehensive overview of mathematical models and analytical techniques for mmWave cellular systems based on stochastic geometry [
28]. In addition, a system-level analysis of the success probability for cell selection and random access delay in the mmWave cellular systems is conducted in [
29].
Currently, the mmWave communication is now considered for V2V or V2I communications, which are required for autonomous or proxy driving in future smart cars [
9]. We survey existing works related to the mmWave V2X communications as in
Table 1.
Several studies on measurement campaigns and channel modeling for the mmWave V2V or V2I communications are reported. Ben-Dor et al. conducted multipath and angle-of-arrival (AOA) measurements at 60 GHz of outdoor peer-to-peer channels in an urban campus courtyard for transmission and into a vehicle [
30] in 2011, demonstrating that varying the transmitter and receiver separation provides different root mean square (RMS) delay spreads from 2.73 ns to a maximum value of 12.3 ns for the LOS and NLOS antenna pointing scenarios. Loch et al. developed a practical mmWave vehicular testbed to evaluate performance, where fixed beam-steering approach enables the RSUs to transmit large amounts of data in a considerably short amount of time for a wide range of speeds [
10]. Park et al. investigated mmWave blockage characteristics based on measurements collected in a typical V2V environment at 28 GHz [
11].
Based on these measurements and simulations, the V2X channel models are exploited. Va et al. reviewed the state-of-the-art in measurements related to mmWave vehicular channels [
12]. In contrast to the previous studies conducted with the two-ray model on flat road surfaces, more realistic settings with road undulation, road surface curvature, and blockage by other vehicles are considered to improve the accuracy of pathloss prediction. The propagation mechanisms reflection and diffraction of mmWave on realistic road surfaces and geometries at 60–77 GHz are demonstrated in [
31,
32]. Further, Antoescu et al. proposed channel propagation models for mmWave V2X communications using ray-tracing simulations [
13], which include the effects of link blockage, scattering and multipath fading. Wang et al. provided research results on propagation characteristics of V2V channels, particularly for shadowing effects induced by obstructing vehicles between a transmitter and receiver [
14]. In [
15,
33], a geometric multiple-input multiple-output (MIMO) channel model for mmWave mobile-to-mobile (M2M) applications based on the two-ring reference model is proposed.
Several studies on the V2X network based on a stochastic geometry model are reported that investigate and optimize the connectivity and throughput with varying blockage, beam direction and vehicle density. Tassi et al. proposed a stochastic model of mmWave-based RSUs infra-structure to vehicle communications [
34], where they investigated the blockage probability and throughput with varying vehicle densities and speeds at a multi-lane highway. Lorca et al. presented a theoretical analysis of the Doppler power spectrum in the presence of beamforming at the transmitter and/or the receiver in V2I systems [
35]. Wang et al. analyzed the coverage of urban mmWave micro-cellular networks based on stochastic geometry with a LOS probability function of randomly oriented buildings for a V2I scenario [
36]. Perfecto et al. analyzed the interplay between the beamwidth assignment and the scheduling period in V2V communications [
20] and proposed an optimization algorithm to establish a V2V link having optimal beam width, using swarm intelligence based on the channel and queue state information [
21]. Va et al. proposed a swarm intelligence to efficiently pair vehicles of V2V links and optimize beam widths considering the channel state information and queue state information [
22].
Some studies propose approaches to utilize sensors (such as radars, GPS, camera, etc.) to reduce the beam alignment overhead among vehicles based on vehicle position, posture, or other sensed information. Gonzalez et al. [
19] and Kumari et al. [
18] proposed a set of algorithms to perform the beam alignment in a V2I scenario, by extracting information from the IEEE 802.11ad module or the LRR radar signal to configure the beams or create a joint waveform for automotive radars and mmWave V2V communications using the same hardware. In [
16], Choi et al. proposed a high-level solution for a key challenge of the mmWave beam training overhead where the information derived from the sensors or DSRC are leveraged as lateral information for the mmWave communication link configuration. Mavromatics et al. leveraged vehicle sensory data of position and the motion for beam forming by DSRC beacons [
17].
Similarly, location-based beam alignment and training are considered in the following studies. Va et al. proposed a mmWave-beam switching approach based on the position information (for example, the information available via GPS) from the train control system for efficient beam alignment [
37] and presented an optimization of beam design to maximize the data rate for non-overlap beams in the LOS to the RSU [
38]. Garcia et al. also proposed a location-aided beam forming strategy and analyzed the resulting performance considering the antenna gain and latency [
39]. Maschietti et al. formulated the optimum beam alignment solution of a Bayesian team decision problem with novel and less complicated algorithms for optimality [
40]. Va et al. leveraged the position of a vehicle along the past beam measurements to rank desirable pointing directions that can reduce the required beam training based on a popular machine learning method used in recommender systems [
41]; moreover, they proposed the utilization of the position of the vehicle to query a multipath fingerprint database that provides prior knowledge of potential pointing directions for reliable beam alignment [
42]. In [
43], Wang et al. also introduced machine learning with the past beam training records for optimal beam pairing by exploiting the locations and sizes of the receiver and its neighboring vehicles.
Eltayeb et al. proposed a blockage detection technique for mmWave vehicular antenna arrays that jointly estimate the locations of the blocked antennas along with the attenuation and phase-shifts that result from the suspended particles [
44]. In such a blockage, a joint optimization problem to select a relay and link to circumvent obstacles and to reduce delivery latency in the 60 GHz mmWave networks was modeled by He et al. [
45] together with a less complex algorithm decomposing the problem into tractable sub-problems. Furthermore, Taya et al. proposed a multi-hop relaying through dynamic vehicle deployment to increase the coverage of V2X, formulate the deployment problem as an optimization problem, and obtain its lower and upper bounds of performances [
46].
In [
47], Petrov et al. showed that the interference from the adjacent lanes can be reasonably approximated using two-dimensional stochastic models without any significant loss of accuracy. This interference may significantly affect the performance of the communication systems as highly directional antennas are used by spatial configurations. Kim et al. proposed a channel assignment algorithm for mmWave beams to avoid the inter-beam interference from the uncoordinated beams in mmWave V2V communications [
8].
Multiple connections to legacy network (such as 3G or LTE) and mmWave base stations can provide seamless connectivity to moving vehicles. Giordani et al. introduced a method with multi-connectivity to a mmWave cell and a conventional microwave cell for robust connectivity and handover based on a sounding signal sweeping and instantaneous measurement of the received signal strength measurement [
48]; further, they proposed a novel uplink multi-connectivity system for the efficient control plane applications, such as handover, beam tracking, and initial access [
49].
For security in vehicular mmWave communication systems, Eltayeb et al. proposed physical layer security techniques by injecting artificial noise in controlled directions using multiple antennas [
50].
These previous studies on the vehicular communications attempt to establish a vehicular channel model of mmWave V2V or V2I communications through simulation or mathematical modeling and propose ideas to reduce beam alignment overhead and training based on location or sensor data. In this study, we demonstrated the feasibility of V2V communication using the IEEE 802.11ad with the 60 GHz mmWave spectrum, specifically in a LOS environment. For the IEEE 802.11ad standard, Jacob et al. explored a channel model for system level simulations with medium access control (MAC) protocols to investigate the influence of moving humans in the framework of IEEE 802.11ad standard [
51]. Coll et al. evaluated the IEEE 802.11ad standard for V2V communication through simulation [
52], demonstrating that the MAC operation and beamforming processes result in a high overhead, and the uncoordinated transmitting stations can significantly degrade network throughput. However, to the best of our knowledge, there has been no experimental study with off-the-shelf IEEE 802.11ad devices in a driving testbed.
6. Discussion
According to the measurement results, the GiV2V connectivity differs from the driving environment and the degree of mobility affects the disconnection duration. Wireless communication parameters, pathloss and slow/fast fading of the GiV2V can be affected typically by varying vehicle speed in the different driving environment, but other parameters except the pathloss can be negligible considering the short radio range of the WiGig. However, the probabilities of disconnectivity in two different driving environments are compared, as shown in
Figure 12. This figure illustrates the probability density function (PDF) of the disconnection interval in the campus and city measurements. As the probability of disconnectivity followed an exponential distribution, we inputed the measured values to the exponential distribution using the least-square error (LSE) minimization. The
of the exponential distribution (i.e., 1/
) was 0.65 and 0.86 for the campus and city roads, respectively; the expected intervals (
) for the campus and city were 1.16 and 1.54 s, respectively. Although certain disconnectivity intervals were exceptionally longer in the city case, we can still argue that the average disconnection was comparable, at 1.1 and 1.5 s for campus and city measurements, respectively.
Based on the results on connectivity duration shown in
Figure 13, the maximum duration of maintaining a connection was 19.4 s in the campus test, while it was 53.5 s in the city scenario because vehicles stayed close at intersections for a long time owing to the traffic signal. Thus, the arithmetic mean of the connectivity time was higher in the city case (2.38 s) than the campus case (1.45 s). However, the value of geometric mean of the campus and city experiments were 0.7 and 0.6 s, respectively, and the median value of each was 0.5 and 0.4 s, respectively; the average connectivity duration in the campus scenario was higher than the city case, except during several long connectivity durations.
Observation 7. The IEEE 802.11ad-based V2V links provided short connection of 1–2 s irrespective of the vehicle speed.
We list the summary of the experimental results of V2V communication using IEEE 802.11ad in
Table 6. Based on the results, we concluded that the short radio range of IEEE 802.11ad caused too frequent disconnections even at slow vehicle speed. Moreover, the connectivity occurred differently corresponding to the driving environment; in campus, the vehicles demonstrated intermittent connections continuously, while vehicles demonstrated long connections only at intersection. However, the durations of the connections on the two roads were comparable and they allowed vehicles to deliver more data to each other. For instance, a vehicle could transmit approximately 300 MB during the short average connection time, 2 s.
However, those intermittent connections require more efficient and intelligent transport layer protocols and buffer management for retransmission and packet reordering. The conventional TCP congestion and flow control can suffer from retransmission of lost packets and connection management. Thus, light-weight protocols such as UDP with control signaling are probably appropriate and redundancy by fountain or network coding is advantageous for a reliable transmission.
Compared to our convoy experiments on the same lane, the disconnectivity interval becomes longer and the vehicle could not receive data continuously from the same transmitter if the vehicle moved individually without platooning. In this scenario, delay tolerant communication and data duplication on vehicular caches are necessary, where a receiver vehicle losing a transmitter due to long disconnection can request the data from a neighboring vehicle having duplicated data. As a future work, we will investigate distributing large data in a vehicular cloud using the GiV2V mmWave links and vehicular storages in the delay tolerant networking.