Addressing Slot Drift in Decentralized Collision
Free Access Schemes for WLANs
Wunan Gong and David Malone
Hamilton Institute, NUI Maynooth, Ireland.
20 November 2012
TDMA vs. CSMA
• TDMA has no collisions.
• Someone needs to schedule TDMA.
• CSMA has collisions.
• No coordination required.
Both are a mess if slots are unclear.
Counting Down
(20us)
Someone else transmits,
stop counting!
Transmissions (~500us)
Data and then ACK
Collision followed by timeout
In 802.11 the slot lengths are a mess, but carrier sense saves us.
Hybrid TDMA/CSMA
Recent proposals for 802.11-like MACs that combine CSMA and
TDMA.
1. L-BEB, CSMA/ECA, (Barcelo, . . . )
2. SRB (Yong, . . . )
3. ZC (Lee, Walrand)
These proposals are like Reservation ALOHA, but adapted for
carrier sense.
802.11’s DCF
• After transmission choose rand(0,CW − 1).
• Wait until medium idle.
• Count down in slots.
• Transmit when get to 0 (if you have a packet).
• If ACK then CW ← CWmin else CW ← 2CW .
L-BEB
• After unsuccessful choose rand(0,CW − 1) otherwise CW /2.
• Wait until medium idle.
• Count down in slots.
• Transmit when get to 0 (if you have a packet).
• If ACK then CW ← CWmin else CW ← 2CW .
Gives an idea of a cycle.
Station 1’s view
1
2
3
4
5
6
7
8
1
2
3
4
TX
7
8
1
2
3
4
TX
Station 2’s view
5
TX
5
6
7
8
1
2
3
4
TX
ZC
• After unsuccessful TX choose random free slot from previous
cycle, otherwise otherwise same slot in next round.
• Wait until medium idle.
• Count down in slots.
• Transmit when get to your slot.
DCFA
Decentralized Collision-Free Access.
• Can be much better than DCF.
• Complications with many stations.
• Collision/noise confusion.
• Some 802.11e aspects a little tricky.
• How important is slot counting?
Slot Drift
8
DCF
ZC
L−MAC
7.5
System Goodput(Mb)
7
6.5
6
5.5
5
4.5
4.0
0
1
2
3
4
5
6
slot drift probability %
7
Cycle len C = 16, number of stations N = 16.
8
9
10
Why would slots drift?
Exposed/hidden nodes could cause problems, but also:
• Sensing Errors: the 802.11 standard requires only a 90% CCA
detection for 9us slots;
• clock errors: different clock rates lead to different slot
boundaries;
• hardware/software miscounting look at primitives available on
Broadcom hardware.
How to improve this?
Help MAC make better decisions.
• Global View Synchronisation: Provide a benchmark against
which slots can be counted.
• Smart Collision Free: Use global view to partition the free
slots.
GVS
1. On TX we announce our notion of slot number.
2. On every announcement, record the offset.
3. If we see two announcements of same offset, renumber.
Two types of renumbering: initialisation (index-shifting) and stable
(index-keeping).
SCF
• Improvement to ZC.
• Needs to identify collisions.
• Partitions free slots.
1
1
Idle slot
2
2
Colliding slot
3
4
Success slot
3
5
GVS Effectiveness
1
0.9
0.8
GVS−ZC 2%
GVS−SCF 2%
GVS−ZC 5%
GVS−SCF 5%
0.7
CDF
0.6
GVS−ZC 10%
GVS−SCF10%
0.5
0.4
0.3
0.2
0.1
0
2
N = C = 16
4
6
8
10
12
14
Number of stations with the most common view in a slot
16
SCF Effectiveness
Average Convergence Speed (Number of Slots)
300
250
200
150
ZC, C=N=16
100
SCF,C=N=16
ZC,C=N=32
SCF,C=N=32
50
1
No slot drift.
4
8
12
16
20
24
28
Number of Stations with the most common view
32
Performance
7.8
ZC
SCF
GVS−ZC
GVS−SCF
DCF
7.5
system goodput (Mb)
7
6.5
6
5.5
5
4.5
0
N = C = 16
1
2
3
4
5
6
slot drift probability %
7
8
9
10
Conclusion
• Nice DCFA algorithms available.
• Depends on well-defined slots.
• GVS: Helps correct slot drift.
• SCF: Helps ZC choose slots.
• Probably not the best you can do.
• How common is slot drift?
• Implement all this?