Challenges in Directional Antennas

1
Challenges in Directional Antennas

• - Vinay Kolar
• Committee
• Dr. Nael Abu-Ghazaleh
• Dr. Kyoung-Don Kang

2
Synopsis

• Directional Antennas- A new technology in Ad hoc
  networking.
• Scope of the thesis
• Analysis of behavior in chain topology.
• Identify Head of Line (HoL) blocking and propose
  new queuing policy.
• Passive discovery of directional neighbors.
• Use them for a better routing.

3
Presentation contents

• Background
• Analysis
• The HoL problem and solution
• Directional routing
• Overall Summary
• References

4
Omni vs. Directional antenna

• Omni
• Directional

5
MAC protocol - 802.11

• Above physical layer
• Proper transmission of packet over channel
• Carrier Sensing
• Carrier Sense Multiple Access with Collision
  Avoidance (CSMA/CA)
• Back-off exponentially
• Virtual Carrier Sensing
• Be silent if you listen to any other
  communication.
• NAV Table

6
Handshake in 802.11
Channel busy for 6ms
Y
Whole transaction 6 ms
A
B
X
Whole transaction 7 ms
Channel busy for 7ms
7
MAC layer reliability

• Try to retransmit the data for RETRANSMIT_LIMIT
• Failure Give up and inform Routing layer

8
Hidden and Exposed terminals
D can interrupt B-C
A
B
C
D
E
Exposed terminal problem
Hidden terminal problem
9
Types of Directional Antennas
Sector of the antenna

• Switched beam
• Cheaper and less complex
• Each sector can point beam in one fixed direction

Antenna with 8 sectors
10
Types of Directional Antennas

• Steerable Antennas
• More intelligent
• Precision focus
• Null steering
• Greater complexity

Y
X
Sender
Interferer
11
Gain of Directional Antenna

• Ability to focus beam in particular plane/angle.
• Gain(directional) gt Gain(Omni)
• Gain at both ends
• Transmitter Gain
• Receiver Gain
• Range(directional) gt Range(Omni)

X
12
DMAC

• Assumptions
• Capable of operating in omni and directional mode
• AoA for a signal can be captured from antenna
• RTS-CTS handshake similar to 802.11
• Omni RTS
• Directional RTS

13
Angle of Arrival (AoA) cache

• Table of ltnode, anglegt tuples.
• Add/Update
• If X hears from Y at angle z, then X adds/updates
  ltY,zgt in its AoA cache.
• Delete
• If X fails to reach Y in direction z for
  DIRECTIONAL_TRANSMIT_LIMIT.
• Timer expires

14
Omni vs. directional mode
Packet to X
X in AoA Cache?
NO
Transmit in omni
YES
Z Get AoA for X From AoA cache
Transmit packet directionally at Z degrees
15
Directional Virtual Carrier Sensing (DVCS)

• Directional NAV (DNAV) table
• If RTS-CTS is overheard in direction z
• Mark sector as busy
• For a constant ?,
• (z- ?),(z ?),duration
• Before transmitting
• Check if channel is busy

16
Pros of Directional Antennas

• Spatial reuse
• Extended range Energy savings

A
17
Cons of Directional Antennas

• Deafness
• Head of Line (HoL) blocking
• Omni does not reach directional neighbors
• Routing is unable to find directional neighbors

18
Presentation contents

• Background
• Analysis
• The HoL problem and solution
• Directional routing
• Overall Summary
• References

19
Analysis of chain topology

• Linear chain topology
• Qualnet simulator
• 200 m apart
• Omni distance 250m
• Directional send Omni receive 340m
• Directional send Directional receive 450 m

20
Throughput vs. Hop count
Directional performs better than omni But is
everything fine in directional?
21
RTS drops

• Standard deafness
• 1 RTS drop
• Simultaneous RTS
• 2 RTS drops
• Rare
• Back-to-back RTS
• 2 RTS drops
• Deafness a major factor!

22
Consequences of Deafness

• RTS drops
• Hidden terminal
• Solved in 802.11.
• Reoccurs in DMAC
• Inconsistent Virtual Carrier sensing

23
Consequences of Deafness

• Back-offs, NRTE, Route Error
• What is the probability of A succeeding?
• After RETRANSMIT_LIMIT, an NRTE is caused

A
B
C
Drop
Pkt1 Pkt2 Pkt3
Pkt2 Pkt3
A backs off
24
Effect of Queue length

• Head of Line blocking
• Wait time 30 transmit time!
• 13 RTS drops in between

25
Source Dictation

• Window 32 vs. Window 2
• Similar to 802.11 6

26
Source dictation

• Window 32 vs. Window 2

Description Win 32 Win 2
Number of NRTEs 14 0
Number of RTS drops 5493 924
Number of DATA drops 26 4
Number of ACK drops 6 2
Throughput 312354 496133
27
TCP Delays

• Due to NRTE, TCP DATA packet gets dropped gt Hole
  in TCP window
• Duplicate TCP ACKs for every packet received
• More ACK Traffic
• More Head of line blocking

28
Geometry of topology

• Zig-zag pattern( 90 degrees between edges)
• More channel re-use than linear chain because of
  parallel transmissions?
• Throughput is 25 lesser than chain
• Number of DATA packet losses 6 times higher

29
DATA packet collusion
30
Azimuth analysis

• Let YZp be the Azimuth Gain
• for node Y in direction of Z.
• Power difference at Y
• mod (YZp YXp)
• Power difference between the legs of connection

31
Conclusion of Analysis

• Deafness results in
• Packet losses
• Longer back-offs
• NRTE
• Route error
• Hidden terminal
• Inconsistent DNAV updates
• Effectiveness of directional antennas not being
  used
• Longer range - discovery of routes not done
• Channel reuse - HoL

32
Conclusion of Analysis

• Well behaved source helps.
• Similar to 802.11 6
• Intricate function of topology.
• Angle between the paths of communication matters
• In spite of the weakness, DMAC performs better
  than 802.11.

33
Presentation contents

• Background
• Analysis
• The HoL problem and solution
• Directional routing
• Overall Summary
• References

34
Avoiding HoL blocking

• Existing Queueing mechanism
• Strict priority FIFO queuing
• Ineffective for DMAC
• What is needed to avoid HoL?
• Mechanism to find out the time interval for which
  the channel might be busy in a particular
  direction
• Sensing the channel in direction of each packet?

35
Avoiding HoL blocking

• If such mechanism is present
• Schedule the packet with least wait time.
• Use DNAV!!
• For given directions, check DNAV and record wait
  times for each packet.
• Choose packet with minimum wait time.

36
Avoiding HoL blocking

• Is DNAV accurate?
• What if the node was deaf and DNAV was not
  updated?
• Live with it !!
• Chances of marking wrong angle in DNAV?

37
Avoiding HoL blocking

• Marking right information in DNAV
• When X gets a packet from Z when it is locked
• Update only the wait time
• Do not update the angle
• Update angle and wait time when X is in omni mode

38
Avoiding HoL blocking

• Terminologies
• Interlinking queue
• Routing layer inserts the packet into this queue
• MAC picks up the packet from this queue
• MAC Queue
• New queue for the proposed protocol from which
  the DMAC will pick the packets for transmitting
• A MAC Queue can accommodate a maximum of
  MAC-QUEUE-SIZE packets.

39
Proposed queuing policy

• If MAC Queue is not full
• Buffer packets from Interlinking queue to MAC
  Queue
• Check MAC Queue for the packet of least wait time
  (respecting priority)
• Transmit that packet

40
Omni-directional packets

• Have the maximum wait time
• If an omni packet is head of Interlinking queue
• Transmit all packets from MAC Queue
• Schedule omni packet
• Disadvantage
• Packets which are behind the omni packet will not
  be scanned till the omni packet is sent.
• Starving of omni packet

41
Results
Throttling connection

• 1-2 obstructs 4-3 flow
• If 1-2 is very high, then chances of a packet 4-3
  being transmitted is low
• 4-5 packet gets blocked

42
Results

• Demonstrate throughput improvement of 4-5

43
Results

• Queue size is varied

44
Results Queue 30

• Que 30
• Orig
• vinkolar_at_topaz diffquesizes grep "Through"
  ch2-6.omni.stat.orig.20sec.que30.1536pkt.bal
  grep Server
• 2, ,1024, Application, CBR
  Server,Throughput (bits/s) 1238076
• 3, ,1024, Application, CBR
  Server,Throughput (bits/s) 680668
• 5, ,1025, Application, CBR
  Server,Throughput (bits/s) 155997
• Persec
• vinkolar_at_topaz diffquesizes grep "Through"
  ch2-6.omni.stat.persec.20sec.que30.1536pkt.bal
  grep Server
• 2, ,1024, Application, CBR
  Server,Throughput (bits/s) 1291659
• 3, ,1024, Application, CBR
  Server,Throughput (bits/s) 626318
• 5, ,1025, Application, CBR
  Server,Throughput (bits/s) 230436
• Total from 4
• 680668155997836665
• 626318230436856754
• Around 2 improvement!

45
Results - Grid

• Improvements
• Throughput
• End to end delay
• Refer thesis

46
Results Grid

• Throughput
• End to end delay
• Upto 30 improvement
• Refer thesis
• Grid with random connections
• Refer thesis

47
Results - Grid

• Normalized throughput for grid
• Why do we get lesser gain when MAC Queue is
  increased from 20 to 30?

48
Results - Grid

• Improvement in End to end delay

49
Conclusions

• Channel utilization is low in normal DMAC
• Identified and proposed a solution to solve the
  HoL
• Proposed a scheme to solve incorrect AoA updates
• Good results with incorrect DNAV
• Greater improvement if deafness is solved

50
Future Work

• Study HoL with DMACs which reduce deafness
• Reduce the omni-directional packet block
• Without letting omni-packets to starve
• Study the effects when the number of sectors are
  varied

51
Presentation contents

• Background
• Analysis
• The HoL problem and solution
• Directional routing
• Overall Summary
• References

52
Routing layer DSR

• Responsible for
• Finding and maintaining routes
• DSR
• On demand
• Source routing
• Each packet has the set of nodes through which it
  has to be routed

53
Directional Routing

• Terminologies
• Omni-directional neighbor
• Neighbor nodes that can be reached by omni
  transmission
• Directional neighbor
• Neighbor nodes that can ONLY be reached by
  directional transmission

54
Directional Routing

• Routing layers cannot find directional neighbors
• Broadcasts are omni
• Existing Solution
• Sweep the directional beam through all sectors
  instead of sending one omni transmission 2
• Main Disadvantage
• Overhead involved !!

55
Proposed solution

• Is there a better way?
• DMAC Layer stores all known neighbors in AoA
  cache
• Use it to find directional neighbors
• How to propagate it to routing layer?
• Interface between DMAC and routing.
• How can the routing layer use it?
• Tailored to the needs.

56
Proposed solution

• The Interface
• Semantics
• Invocation
• Directional DSR(DDSR)
• Implements the interface
• Shortens routes in source route

57
Design

• Upcall interface
• Interface from MAC layer to routing layer
• Calls a function whenever a node is
• Added into AoA
• During new neighbor discovery
• Purged from AoA
• NRTE
• Timeout

58
DDSR

• One-hop table
• Maintains a list of one hop neighbors
• Updated during the upcall
• Adds or deletes the neighbor

59
DDSR Route Compaction

• Each packet will have source route
• Scan source route from destination to source
• If node is present in one-hop table, then this is
  the next hop

A,B,C,D,E
60
Till date

• Premature stage
• Route repair and Route maintenance needs to be
  built
• Routing by choosing next hops which are out of
  the source route
• Detection of neighbors by AoA
• Current AoA updated only when it listens to RTS
  or DATA packet

61
Preliminary Results

• Nodes are 150m apart
• CBR runs from
• Node 1 to Node 7
• Node 7 to Node 1
• One of the best cases for route compaction

62
Preliminary Results

• End to end delay is always lower
• Reduced number of hops

63
Preliminary results

• Throughput
• Improvement most of the times

64
Preliminary results

• Number of packet drops are VERY high in DDSR
• Route repair mechanisms?

65
Grid topology

• Results in Grid topology
• Around 20 improvement in End-to-End delay
• Lot of Packet drops
• Throughput loss at high rate.
• Results in Sparse Grid
• End to end delay improvement
• Packet losses high again!
• Throughput remains the same!

66
Results - Grid

• We dont always win in end to end delay

67
Results - Grid

• Throughput loss packet losses!!
• Need better route repair mechanism!!

68
Results Sparse Grid

• Good improvement in end to end delay
• No improvement in throughput
• Packet losses is high again!

69
Conclusions

• Proposed mechanism to use higher range in routing
• Without overhead
• A scheme to shorten paths in DSR
• Preliminary results are encouraging

70
Future work

• AoA Updates
• Update for all packets
• Route error and route maintenance
• Routing through nodes not in source route
• Low overhead neighbor discovery
• Takes advantage of inactive nodes

71
Overall Summary

• More effective handshake is needed in DMAC
• Deafness creates a significant degradation
• Spatial reuse can be made more effective HoL
• Routing can be optimized with little overhead

72
References

• 1 Choudhury, R. R., and Vaidya, N. H.
  Deafness A Problem in Ad Hoc Networks when
  using Directional Antennas
• 2 Choudhury, R. R., and Vaidya, N. H. Impact
  of Directional Antennas on Ad Hoc Networks
  Routing.
• 3 Korakis, T., Jakllari, G., and Tassiulas, L.
  A MAC protocol for full exploitation of
  directional antennas in ad-hoc wireless networks
• 4 Takai, M., Martin, J., Bagrodia, R., and Ren,
  A. Directional virtual carrier sensing for
  directional antennas in mobile ad hoc networks.
• 5 Choudhury, R. R., Yang, X., Vaidya, N. H.,
  and Ramanathan, R. Using directional antennas
  for medium access control in ad hoc networks.
• 6 Xu, S., and Saadawi, T. Revealing the
  problems with 802.11 medium access control
  protocol in multi-hop wireless ad hoc networks.