1. Routing with Directional Antennas

2. Questions on what is of interest
Transmission of Routing Updates or Control Messages – directional ? omni-directional ?
What are the effects of the extension of range ? – Recap : Directional Range > Omni-directional range.
Is the additional spatial re-use providing any advantages ? More stable routes ?
What is the impact of mobility ?
Is there any interaction between the MAC and the routing layers ? If so what is it ?
What might be other challenges ?

3. [8] R. Roy Choudhury and N. H
[8] R.Roy Choudhury and N.H.Vaidya, “Impact of Directional Antennas on Ad Hoc Routing”, Proceedings of IFIP Personal and Wireless Communications, 2003.
Link:

4. Objective and Strategies
Evaluate the impact of the use of directional antennas on ad hoc routing – in other words to answer the questions that we brought up.
Steps that they take:
Perform simulations to understand how the presence of directional antennas affects routing.
Use the insights gained to propose changes to existing routing strategies (or design new strategies as might be appropriate).
Analyze the effects of the new schemes.

5. Reported Related Work
This is a relatively new area – only three papers so far.
One of the papers simply looks at performance enhancements due to the use of directional antennas on on-demand routing.
A second paper looks at computing maximally disjoint routes (useful due to spatial re-use – no interference between routes.)
Ram’s paper simply uses a link state routing protocol.

6. Antenna Model Used
Cone-Sphere Model.
N beams – so the angular width of each beam is 2P/N.
Two modes of operation – Omni & Directional – node can toggle between modes.
Directional gain Gd > Omnidirectional Gain Go.
Thus, a node has two neighborhoods :
Directional Neighborhood.
Omni-directional Neighborhood.

7. Directional-Omni (DO) and Omni-Omni (OO) Neighbors 
B is a DO neighbor of A if it can receive A’s directional transmissions even if B is in the omni mode.
B is a OO neighbor of A if it can receive A’s omni transmissions.
Note that the OO neighbors are a sub-set of the DO neighbors.
Is this a tongue twister or what ?

8. The MAC
Of course not the MAC shown 
We need a MAC that can use directional antennas.
The authors use DiMAC – their own MAC protocol.
RTS/CTS handshake is directional.
At each node, DiMAC maintains what is called a directional NAV (network allocation vector) table – used in other similar MAC protocols.
This helps tabulate the direction from which a RTS or a CTS is received from each neighbor.

9. MAC protocol continued
Each node uses the lookup table and determines the direction of the table – transmits RTS in that direction.
The recipient node listens omni-directionally.
It figures out the beam on which the RTS was received and sends out the CTS in that direction.
Nodes that overhear RTS or CTS or both defer transmissions for the proposed duration of the data transfer.
Protocol suffers from the deafness problem

10. Deafness
A node say Node C, attempts to initiate a dialogue with another node (say Node A).
However, Node A is talking to someone else (say Node B) and is beamformed in the direction of B.
Node C’s signals are not received by Node A – it is deaf to Node C’s signals.
Node C interprets the the absence of a reply to be a collision.
Repeats this transmission multiple times. Finally drops packet.
Can be potentially treated as a route failure.
             

11. Routing: DSR over DMAC
RECAP: DSR is an on-demand routing protocol for ad hoc networks; so far assumes the presence of omni-directional antennas
Sequence: RREQ, RREP when route is not available in cache. Use the route until it fails. Failure communicated using an RERR message.
Source routing.

12. Route Discovery -- Sweeping
Omni-directional broadcast is emulated by sequentially transmit the packet in each direction.
This can increase delay – if N directions, each sweep takes N times the time taken by an omni-directional transmission.
But, sweeping can reach the DO neighbors.

13. Mobility: Scanning
In mobile scenarios, the direction of a particular neighbor (as indicated in the table) could become stale.
Scanning is used to address this.
Scan  HELLO packets are transmitted sequentially on each antenna beam.
When a node receives such a HELLO message it responds using the same (appropriate) antenna beam.
Note: Neighbor discovery is more complex now.

14. Partial Scanning
Scanning can be expensive.
Typically if communication is regular the node might not have moved far.
Partial scan is where a node searches for a lost neighbor using only “K” beams adjacent to the beam that was previously in use for that neighbor.
Reduces overhead how does one choose K ?
However, for bursty communications this might deteriorate to a “scan”.

15. Simulations and Performance
Use of Qualnet for simulations.
CBR Traffic
1500 x 1500 square region.
Directional DSR (including sweeping and directional routes) and DiMAC.
Several scenarios considered.
Metrics of interest are Route Discovery latency (RDL) and throughput.

16. Intuitive Thoughts
RREQ messages get delayed due to Sweeping as one might expect.
Shorter routes might be discovered due to extension of range.
Deafness might be a problem – how critical ?

17. Studying Route Discovery Latency (RDL)
Smaller beamwidth  longer range.
If the distance of separation between the source and destination is small not much to be gained – the directional hop count and omni hop count are almost same.
Smaller hop count offset by increase due to sweeping delay.
At larger distances the advantage of the higher transmission range dominates.

18. Dependent on node density as well.
Intuitively as density becomes higher one might expect to find more DO neighbors – so one might expect hop count to decrease thereby improving performance.
Not the case ! Interference due to side lobes increases the possibility of collisions.
Performance enhancements not significant.
Unclear from the paper : Why does omni transmissions not have the same problem which is, why is the degradation more significant in the case of directional schemes ? Is this an artifact of the protocols themselves ?

19. Throughput
Even though hop-count is expected to decrease throughput does not go up significantly.
This is because due to sweeping delays, the optimal path may not be found.
The nearest neighbor may not be the first one to be found – so sub-optimal paths may be discovered earlier.

20. Delayed Route Reply Optimization
Replying to all RREQs is not a good thing.
First – increase in overhead.
Second, when destination is responding to the first RREQ, it might miss others (directional schemes) – remember RREQ is broadcast.
Unclear from paper: Is RREQ missed in between ?
So wait for a pre-specified time T before sending an RREP – T = r * Tsweep; r is a system parameter.
Tsweep is the time taken to finish a full sweep.

21. The authors also find that deafness can create significant problems when linear topologies are used.
This causes the directional schemes to in fact fall somewhat below the omni schemes in linear topologies in the presence of multiple flows.
Refer to the paper for the details.
In random topologies however, significant benefits are seen – shorter routes, higher spatial re-use. Partitions are prevented.

22. Effect of transmission range.
As we proceed from omni, performance increases.
However, as we increase the transmission range (reduce beamwidth), sweeping delays increase and the problems due to deafness exacerbate. Thus, shortest RREQ routes are never received in spite of the delayed optimization.
Artifact of the protocols (especially sweeping).
At extremely small beamwidths, performance degrades.

23. S Number of Control packets X Area blocked by each packet
Routing Overhead
New metric:
S Number of Control packets X Area blocked by each packet
S Number of data packets
a =
Intuitively, network capacity consumed by each control packet is proportional to the interference region caused by the packet.
Initially it is seen that the sweeping overhead much higher than omni transmissions as one might expect.

24. Selective Forwarding Optimization
Figure from [8]
Do not forward RREQ in the direction received.
Forward control packets on those beams that are diagonally opposite to the beam with which the control packet was received.
Reduces overhead significantly but still higher than DSR.

25. Impact of Mobility Partial Scanning used. Seems to work fairly well.
Ultimately, the use of the techniques seem to demonstrate promise in terms of using directional antennas.
However, many of the schemes can be energy intensive – especially the sweeping part.
Open questions on how to make these schemes improve energy efficiency to a greater extent – already benefits in terms of reduced collision rate, increased throughput, reduced transmissions due to shorter routes.

26. Next Time Onwards : We share the Air time 

27. In terms of receiving abstracts for your talks:
Just Kidding !!! Thanks!