2. Outline
1. V2X Use cases, trends,
requirements
2. V2X strategies &
enablers
3. Three Case
studies & open
problems
4. Conclusions
3. 5G imposes new use cases w/ a
myriad of requirements
33
720Mbps+ for multi-viewing angles
<50ms e2e latency
120Mbps for compressed 8K video at 120fps.
Multi-Gbps for uncompressed 4K/8K videos
1ms e2e latency from action to feedback
Ultra high reliability
Latency,
throughput,
reliability
Our focus
4. V2X Societal Impact
“One in four trucks in Europe runs empty”
What a waste! CTO, SCANIA
2/3 of the oil used around the world
goes to power transportation vehicles,
of which 50% goes to cars and light
trucks
+ traffic fatalities +
Time wasting,
limited mobility,
etc
5. V2X: a mix of requirements
Generally strict requirements
• High reliability
• High throughput (in the long run)
• Low and strictly bounded latency
• Latency and reliability of communication are
scenario specific
Normal Requirements
- Max packet error rate: 10−3
- Payload size: 400 bytes
- Max latency: 100ms
Application-specific Throughput requirements
• High throughput applications (occupancy grid,
full sensor LIDAR, Radar, camera)
• Moderate throughput applications (sharing of
planned trajectories, sharing of course traveling
decisions)
• Low throughput applications (short emergency
messages, perdiodic DSRC-like broadcast
messages) TODAY!
FP7-METIS Requirements
- Max packet error rate: 10−5
- Payload size = 1600 bytes
- max latency: 5ms
Safety
Dynamic
Maps
Full Autonomy
Very strict
strict
Relaxed requirements
To say the least, requirements
are not well understood.
6. V2V Communication Use Cases
Informed safety/hazard
decisions
Expand the sensing
range of the vehicle
Allows interactions between
vehicles with different
automation levels
Direct V2V (3GPP Scenario 1)
PC5/ProSe
7. V2I Communication Use Cases
Use for precise navigation
(location tracking)
Altering drivers of nearby
vehicles and red lights
Effective with non-connected
cars, bicycle, and pedestrians
Multi-connectivity for reliable
transmissions
8. Cooperative Intelligent Transport
Systems (C-ITS)
Cooperative adaptive cruise control
(CACC)
• Cars share information on speed
changes in real time
√ Allows for a more efficient
adaptive cruise control system, as if all
vehicles drive as one single unit
√ Reduced braking and accelerating,
ease traffic and cut carbone emissions
Traffic Safety
• Electronic emergency brake light
• Traffic jam ahead
• Stationary vehicle warning, etc.
Proactive Mobility (not studied much)
Uberization of mobility
CooperativeVehicle-to-anything communication
9. • Japan: 10MHz of spectrum in 760 MHz
– Additionally 5775-5845 MHz for DSRC
• USA: 75 MHz in 5850-5925 MHz
• Europe: auto industries expect market
introduction in 2017
– Spectrum: 70 MHz in 5855-5925 MHz
– 63 GHz to 64 GHz (V2V/V2R
communications)
• Likely to be used for truck platooning
V2X spectrum likely to be above 6GHz
V2X Spectrum
10. Non safety apps also
possible
IEEE 802.11p, IEEE 1609.x
Supports very
low data rates
(6 Mbps max)
DSRC not designed
for uRLLC
!
Current legacy solutions for V2X comm. are based
on IEEE 802.11p
• Main problem: mainly optimized for a WLAN-
type of environment with no-/very low mobility.
1. Nevertheless, using blindly D2D underlay for
V2V communications may cause significant
degradation to system performance
2. Traditional D2D cannot work for V2V communications
(in particular safety applications) due to strict
requirements on latency and reliability with small
message payload
DSRC (in US) = ITS-G5 (Europe)
- Rapidly changing CQI/CSI vs
slowly-changing
- High mobility vs. Low mobility
- Highly dynamic network
topologies vs. Slow dynamic
V2V vs. D2D
Analogy with
LTE vs WiFi
11. DSRC versus LTE-A for V2X
Features DSRC LTE-A
Channel width 10 MHz Up to 100 MHz
Frequency Band 5.86 – 5.92 GHz 450 MHz – 4.99 GHz
Bit Rate 3 -27 Mb/s 100’s of Mb/s to 1 Gb/s
Range Up to 1 km Up to 30 Km
Capacity Medium Very high
Coverage Intermittent Ubiquitous
Mobility support Medium High
Market
penetration
Low Potentially high
DSRC employs CSMA/CA and its variants
• Exponential Backoff based mechanisms (Excessive delays)
• Even after obtaining channel access, delays can be caused due to insufficient SINR
• Inability to transmit and received simultaneously
• Short range (100-300meters) compared to long-range of LTE-V
Shall we use
DSRC or LTE-V?
It depends
12. LTE-V
V2V
V2V
V2I
V2V through
D2D mode in
LTE-A
Direct communication or
through infrastructure
Higher data rates than
DSRC (up to 1Gbps)
+ Long-Range
RSU helps vehicles discover
other nearby vehicles
• DENMS (decentralized environment notification message) & CAM (cooperative awareness message) In Europe
(USA), CAM 800 (300) bytes packet with repetition rate of 2 (10) Hz + maximum latency of 100 ms
• METIS: 1600 bytes delivered in 5ms (10Hz) sent over a 10MHz safety channel
• High-resolution map information can be exchanged among vehicles
• CSAM cooperative situational awareness messages
Types of Messages
Periodic vs. event driven
13. • Arrival of vehicles follow poisson process
with 3 seconds average interval
- Each lane has different average speed
- Vehicle transmit power is 20dbm per 10
MHz (max is 33 dBm)
- Minimum of SINR=6dB is required
(application-specific)
Traffic Model
- Packet loss rate
- Latency of Packet delivery
- Network utilization
- Channel Busy Ratio (CBR)
- Information dissemination rate (IDR)
no. Of copies of a packet delivered per
unit time from a single vehicle to its
neighbors up to a given distance d_max
- Information Age (IA)
V2X Performance Indicators
- Freeway/highway (multilane) vs. Manhattan-like
- Bidirectional vs. Unidirectional traffic
- Vehicle in one lane can relay traffic of vehicles on same or different lane
- Connected vs. Disconnected RSUs
- Connected RSUs as relays for information dissemination
- Parameters of interest are:
- Vehicle type and position
- Speed and Inter vehicle spacing
- Isolated vs. Clustered vehicles Deployment Scenarios
14. Knobs: Rate vs. Power/range vs. Message (size/content) congestion control based on PER, CBR, IDR, IA
- V2X is all about increased reliability (PLR) leveraging:
Redundant transmissions over time slots (coded slotted Aloha) either at RSU or vehicles acting
as RSUs.
Redundant transmissions Over RSUs
Redundant transmissions Over vehicles acting as RSUs
Network coding as an enabler
Epidemics exploiting important vehicles
- V2I vs. V2V
- Unicast vs. Multi-cast (eMBMS)
- Exploiting multiconnectivity and data replication at collection of RSUs
- Prioritizing packet scheduling (information map + objects)
- Full duplexing: same RB can be used by more vehicles in the cell if far enough to each other
- Fairness: resource allocation on a differentiated weighted fair basis; for eg. priority messages
• Location aware resource allocation vs. CSI-based
• Static vs. Dynamic zones
• Cell partitioning
• How much spectrum is needed to ensure reliable V2X? 80MHz needed to ensure 99% reliability
whereas 802.11p dedicates only 10MHz insufficient spectrum
• Vehicle coordination: Centralized via RSU/cloud vs. Decentralized solutions
Strategies & Enablers
15. V2X Requirements & Enablers
Efficient radio transmission of
small/ medium size data packets
Support connectivity for up to
3000 of devices per sqkm
Scalable data rates 1kbps to 10
Mbps
High link reliability (E2E), >
99.999%
Low latency (E2E), 10 ms
Redundancy/fall back to other
technologies
Full availability (area/cell,
including out of coverage ~ 100%)
Requirements Solutions/Enablers
Mobility up to 500kmph
Flexible frame structure and
resource singaling (TDD/FDD)
Network Coding
Diversity, multiconnectivity
Enhanced (random) access
resources
Context based latency and
reliability aware RRM including
link adaption
Enhanced mobility for continuous
connectivity
RRC mechanisms
Traffic safety
Dynamic Maps
16. Edge/Fog/Cloud enabled V2X
Control Plane
Data Plane
Cloud
Edge/fog
Latency
functionality split
(control & data
plane separation
RSUs with MEC reduce latency
Storage/computing
18. Setting and State-of-the art
Use case → automotive safety services
• Cell is divided into spatially disjoint zones.
• The zone layout and RB set are fixed and do not change over time
• Single V-UE within a cell sector is allowed to reuse specific RB
• Centralized RB allocation by the eNB/RSU is considered
M. Botsov, M. Klugel, W. Kellerer, and P. Fertl, “Location dependent resource allocation for mobile device-to-device
communications,” IEEE Wireless Communications and Networking Conference, Istanbul, Turkey, Apr. 2014.
19. Dynamic Proximity-aware Resource
Allocation in V2V Communications
Due to the localized and proximity nature of the service,
the solution involves:
Spatial and
temporal
resource reuse
Eliminate need of full
Channel State Information
(CSI) knowledge at RSU
M. I. Ashraf, M. Bennis, C. Perfecto, and W. Saad, “Dynamic Proximity-aware Resource Allocation in Vehicle-to-Vehicle
(V2V) Communications,” IEEE Globecom, Washington DC, Dec. 2016.
Spectral Clustering
(Location/Load-aware)
20. System Model
• Single cell scenario with Manhattan deployment of V2V-
pairs, LOS, NLOS, speed of vehicles, small packet size
• Dynamic zone formation (geographical information and
load based dissimilarities)
• Multiple V2V-pairs coexist in the network
• V2V pairs reuse the specific RB (Intra-zone matching).
• Let 𝒦 = 1, … , 𝐾 be the set of V-UE pairs, 𝒩 = {1, … , 𝑁}
be the set of RBs, 𝒵 = {1, . . . , 𝑍} be the set of zones.
21. Problem Formulation
Let 𝑆𝑧 be the total number of V-UE pairs
Satisfying the target SINR inside each zone
𝑧 ∈ 𝓏. The cost per-zone for a given
matching 𝜂 𝑧 by:
Hare-Niemeyer method is used for calculating the set of 𝒩𝑧
for each zone.
Coefficients 𝛼 and 𝛽 > 𝛼 are weight parameters that indicate
the impact of the load and number of satisfied V-UE pairs.
1) Dynamic Zone Formation (by RSU)
- Neighborhood based Gaussian similarity [𝑫]
• 𝜎 𝑑 controls the impact of the neighbourhood size.
• 𝜖 𝑑 is the range of the Gaussian distance similarity.
– Load based similarity [𝑪]
– Combining the Similarities
• 𝜃 controls the impact of load similarity and distance.
– Spectral Clustering for zone formation 𝔃.
1) Intra-zone resource allocation as matching game
with externalities.
– many-to-one matching game per zone with
externalities between RBs and V-UEs
– Distributed algorithm that allows RBs and V-UEs to
self-organize and to maximize their own utilities
within their respective zones.
RB1
RB2
26. • Cluster is created with Spectral
clustering algorithm
– Categorizes similar vehicles
(distance between vehicles) in
same clusters while dissimilar
vehicles in different clusters
– Gaussian similarity, distance
between vehicle b and b’
where σd controls the impact of
neighborhood size
• Vehicle with high SINR act as the
cluster head (also close to the cluster)
which coordinates the scheduling
within the cluster.
• Allocates orthogonal resource
blocks among cluster members.
Vehicle to Infrastructure
Sbb =
SINR Heatmap
28. Mmwave-enabled V2V
• Research around mmW
communications for the automation
industry has gained significant
momentum in the last 12 months
• DSRC supports very low data rates
(up to 27Mbps).
• mmW comms in 60Ghz unlicensed
band seem a viable approach for high
bandwidth connected vehicles.
Need to tackle mmW specific
transmission challenges:
• Directionality Steering needed to
avoid deafness
• Beamwidth Selection
• Effect of Blockage
33. Autonomous Driving requires
Communication
Sensors including camera
gather and process data in
the order of Gbps
How much of this data needs to be
shared with peer vehicles?
Coming of
flood of
data
4x 8K HD
cameras generate
215 Gbps!!
Fully autonomous driving V2X Augmented Reality
Self-driving cars will
generate over 4000
GB per day and each..
34. Autonomous Driving and
Communication Needs
Challenge: Safely Detecting Hidden Objects
• Sensors on a car see only line of sight
objects
• Hidden objects affect autonomous cars
– “Google’s car requires occasional human
intervention to prevent accident”
– “Future of autonomous driving depends on
detecting hidden objects & blind spots” -
DARPA Challenge
Expand field of view
beyond line of sight
Network-assisted Autonomous
35. A Request-Response Approach
Vehicle requests only data of interest
◦ E.g. at blind spots, intersections, etc.
Request
for R
Response for R
Cube R
Vehicles which sense the data, may respond
(otherwise use “cloud”)
Fundamental
tradeoffs:
Who should respond?
What if many
respond?
Latency constraints
Diversity vs.
redundancy
37. Modern Cars are Moving Wireless
Systems
A vehicle is a
Computer
Storage
Wireless battery charging
Satellite, LTE
GPS
Computing
VLC
LiDAR, RADAR
WiFi, DSRC
Bluetooth
Storage ComputingProcessing Communication
BatterySensors
Use
case
Use
case
Tremendeous
value $$$$
38. (Few) Open Problems
From Classical Platooning (Consensus in
velocity)
• Maximize fuel efficiency, reduce
collisions, etc.
• Distributed vs. Cloud-enabled.
• Merge and split platoons
Proactive platooning
• Controller suggests to a vehicle if to
join or meet someone in the future,
where are they going?
• Trucks are closed by they could platoon
make them go faster
– This is a local decision control
Predictive control decision at road
intersection on whether it is beneficial for a
vehicle to catch up another vehicle at next
intersection.
• Distributed control problem as you
increase no. of trucks, more fuel is
saved
• trucks close in time and space adjust
speed and platoon and save fuel during
platooning
𝑣2 𝑣0𝑣1𝑣3
𝑑3 𝑑1
sensing,/reasoning/
perceiving/decision-
making
URLLC
INSIDE
39. Conclusions
• V2X is one of the most fundamental use case in
5G with technological and societal impact
• Multidisciplinary approach (wireless, control,
computing, etc.)
• Cuts across ITS, smart cities, IoT, etc.
• An ongoing research work
Question: LTE or WiFi enabled V2X BOTH!
40. Communication
Time frequency
resources, spectrum
Transport (ITS)
road, lane, etc.
Big Data
Maps, context-
awareness, etc
Control
Platooning, vehicle
coordination
Driver in the loop
User driving routing
preference
Societal
Improved fuel
consumption
Accident prevention
Smarter ticies
Editor's Notes
A car requests a cube that it is interested in, for e.g. its blind spot, the next intersection, etc.
Vehicles which possess this information can respond with sensor data in that cube.
In this example, [animate], the red car wants to know about the area around the intersection in its blind spot –that is cube R.
So, [animate] the red car sends a request for R.
[animate] And the green Car can send a response for R.
So let’s see how our system achieves this.
First, the car sends its proposed path to the destination to the cloud.
The cloud checks if it has enough sensor data along this path.
If yes, it moves on to analyze the data and check if the path is safe.
If not, It requests for any missing data from other cars.
The other cars respond to the cloud with the missing data.
The car now has sufficient information to check if the path is safe, and if so reports it to the car.
If it is not safe, the cloud runs a path planning algorithm to find and alternate path which is sent to the car.