1. R2D2: Embracing Device-to-Device Communication in Next Generation Cellular Networks
Tarun Bansal*, Karthik Sundaresan+,
Sampath Rangarajan+ and Prasun Sinha*
Speaker: Zhixue Lu*
*Ohio State University and +NEC Labs America

2. Device-to-Device (D2D) Communication
Normally smartphones communicate with each other through cellular base station
Without D2D
With D2D

3. D2D Traffic Applications
(b) Peer-to-Peer:
Public Safety when traditional infrastructure is not available
Benefit: No interference
(a) Base Station Assisted:
Localized communication and Machine-to-Machine (M2M) communication
Benefit: Service provider helps with security, neighbor discovery and ensuring Quality-of-Service
Our focus is on integration of Base Station Assisted D2D Traffic with
existing cellular communication

4. Our analysis for sectored deployments:
D2D Benefits
Spatial Reuse of Resources
Multiple D2D transmissions per cell [Janis 2009, Doppler 2009, Lee 2013]
Our analysis for sectored deployments:
Very little benefit from additional spatial reuse D4 D1 D2 D1 D2 D4
Cell divided into 3 sectors with each sector covering 120o
Not possible to schedule multiple D2D transmissions in the same sector. D3 D3

5. D2D Benefits (contd.) Offloading
Fewer time slots taken [Janis 2009, Doppler 2009, Lee 2013]
Time slot 1
Time slot 2
Time slot 1
Without D2D
With D2D

6. Benefit of D2D in sectored deployments (Static Channel Allocation)
classified
as cellular
Spatial Reuse
Benefit
Offload
Benefit
Tries to schedule multiple D2D transmissions per sector
Offload benefit is significant (1 time slot instead of 2)
Additional Spatial Reuse benefit (due to D2D) is not significant in sectored deployments due to small sector size

7. Identifying third D2D Benefit: Flexible Load
Use as flexible load
Resources need to be fixed in both Uplink and Downlink directions
In practice, UL-DL traffic distribution varies both in space and time e.g. residential vs. commercial, morning vs. evening
D2D can go on either on the Uplink or Downlink resources
Use resources that would have been otherwise wasted
Uplink Flows
Downlink Flows
D2D flows
time
time
frequency
frequency

8. How do we maximize these two benefits?
D2D Benefits
Offloading (1 time slot instead of 2)
Use as flexible traffic (Place on either DL or UL resource)
How do we maximize these two benefits?

9. Static vs. Dynamic Channel Allocation
Interior Channels
Too much traffic: Borrow channels from neighboring sectors
Exterior Channels
Two co-located sectors
can have the same channel
Static Channel
Allocation
Dynamic Channel
Allocation leverages spatial
traffic variations

10. Challenges in Incorporating D2D with Dynamic Channel Allocation
UL or DL communications (directional) can still go simultaneously
D2D (omnidirectional) may cause interference to coexisting transmissions in sectored deployments (not considered before)
Determining interference from D2D transmission requires knowledge of path loss between all users (costly). D1 D2 D3
Similarly, D2D transmission cannot coexist
with UL or D2D
D2D presents a new challenge: Transmissions may cause interference if co-located sectors are using the same resource.

11. D2D Interference (Dynamic Channel Allocation)
classified
as cellular
Lower throughput
with D2D due to
collisions among
neighboring
sectors
Omnidirectional nature of D2D makes it non-trivial to schedule transmissions

12. Objective
How to do dynamic resource allocation in multi-cell deployments with D2D traffic?
How to schedule UL, DL and D2D transmissions while avoiding interference?
Previous work only looks at resource allocation without D2D traffic, OR
Scheduling with D2D in single cell deployments with no sectorization.

13. R2D2 Contributions
A light-weight scalable solution (R2D2) that works at two different time-scales
Phase 1: Allocate resources to each base station at coarser time scale (cross-sectors)
Phase 2: Allocate resources to each flow independently at each base station at finer time scale (co-located sectors) while avoiding interference

14. R2D2 Contributions
Practical: Works in sectored deployments with multiple cells
Proposed multiple scheduling algorithms with provable guarantees
Showed using simulation results that proposed algorithms perform close to optimal in practice

15. Phase 1: Cross-Base Station Resource AllocationJ M L
Cell\ Historical Demand DL UL
D2D
Cell J X
Cell L
Cell M
Resources Available
Cell\ Resources Allocated DL UL
Cell J X
Cell L
Cell M
R2D2
Phase 1
Algorithm
Input
Output

16. Phase 1: Cross-Sectors Resource Allocation
Objective: Allocate UL and DL resources across 3 sectors in each of the directions
Proposed algorithm satisfies following properties:
Flexibly places the D2D traffic on UL and DL resources
Localized and Scalable
Ensure no interference across sectors belonging to different base stations
See Paper for more constraints and detailed solution

17. Phase 2: Per-frame Scheduling
Each sector was a part of different interfering set in Phase 1:
Same resource
may get assigned to
co-located sectors D2 D3 D1

18. Phase 2: Per-frame Scheduling (contd.)
Schedule UL, DL and D2D transmissions such that total throughput is maximized while avoiding interference
Assign a time-frequency resource block to each flow in the three sectors
Challenge: Path loss information between devices is unknown

19. Phase 2 Contributions See Paper for detailed solutions
Scheduling algorithms at each Base Station:
A greedy polynomial algorithm with approximation ratio of ½
A faster greedy polynomial algorithm with approximation ratio of ¼
More challenging since D2D can go on either DL resources or UL, but not both
A greedy polynomial algorithm with approximation ratio of 1/3
Time-Divisioned
System
Frequency-Divisioned
System
See Paper for detailed solutions

20. Simulation Setup Simulation with 19 base stations and 57 sectors
Modeled practical parameters including path loss, shadowing
Other algorithms simulated
R2D2 Low Complexity
D2D Dynamic Genie (Optimal, knows path loss information between all users)
No D2D Dynamic (Uses dynamic resource allocation and classifies all D2D traffic as Cellular)
Existing Algorithm Static (Uses static resource allocation and does not use D2D as flexible traffic, Janis et al., IJCNS 2009)

21. Results with variation in D2 traffic
R2D2 Low Complexity performs within 5% of optimal
R2D2 Low Complexity gives a throughput of 2.5x and 4.9x compared to existing solution and No D2D, respectively
Benefits coming from
Intelligent resource allocation during Phase 1 while placing D2D traffic flexibly
Interference-free scheduling during Phase 2
within 5%
2.5x
4.9x

22. Conclusions
Investigated the problem of incorporating D2D communication in next generation cellular deployments with sectorization
Identified a new benefit of D2D communications: Flexibility
Towards that, proposed a solution that works at two different time granularities that ensures scalability
Synergy between the two phases makes R2D2 light weight while avoiding all interference
Proposed multiple algorithms with provable approximation ratios for resource allocation and scheduling
Thank You