THROUGHPUT ANALYSIS OF IEEE
802.11 DCF BASIC IN PRESENCE
OF HIDDEN STATIONS
Shahriar Rahman
Stanford Electrical Engineering
http://ee.stanford.edu
Outline of Talk
802.11 DCF Protocol Overview
Problem with DCF Basic Access
Modeling Hidden Stations
DCF Throughput Models
Simulation Results
Discussions & Conclusion
Future Work
Q&A
IEEE 802.11 DCF
802.11 operates on DSSS,
FHSS or IR PHY
MAC provides CSMA/CA
through NAV (~’CS’)
Basic & RTS/CTS accesses
Congestion, timing and
backoff mechanisms
On modeling DCF ->
Bianchi; Wu, et. al.
A Problem with DCF Basic
 2-way handshaking
 Assumes that there is no
other transmission during
this slot!!!
 What if there is a hidden
station???
A
B
C
D
Saturation Throughput Model
Bianchi provides a saturation throughput model
based on a Markov model of backoff mechanismPsuccess E[P]
S=
Pidle s + Psuccess Ts + Pcollision Tc
Pidle = 1- Ptr and Psuccess = Ptr Ps
Pcollision = Ptr (1 - Ps)
Ptr = 1 – (1 – t) n and Ps = nt (1 – t) n-1 /Ptr
Ts and Tc measures time durations of a successful
transmission and collided transmission
Hidden Station Model - Static
Kleinrock and Tobagi’s hearing graph-
1
2
3
4
5
11001
11001
00111
00110
01001
1, 2
(a)
3, 4
(b)
5
(c)
Each station can hear some and not others =>
Pr(reachable) with assumption static => no transition
Generalize this to an n-station WLAN and decompose
into a k-group reachability graphPr(n) = S (Nr(j) /Nt(j) ) / k
Take average stations per group => expected number
of hidden stations in the network
Hidden Station Model - Dynamic
2
1
2
k
Adjacency
graph
4
1
k-state Markov chain
3
Extend static model and allow transitions between k
states, over n stations? => adjacency graph
Pr(reachable->reachable) => use control parameter, m
Pr(hidden->*) = 1/l, Pr(reachable->hidden) = (1-m)/(l-1)
Balance equations:
Pr(j) + (1 – l) Ph(j) = 1
(1 - m)/(1 - l) Pr(j) = (1/l) Ph(j)
Solve to get:
Pr(j) = 1 / (1 + l(1 – m))
Our Throughput Model - Saturation
1.0
Normalized throughput
Worst case throughput
loss => hidden stations
always transmit
Ptr = 1; Ps = Nret (1 – t) Nre-1
This changes throughput
to- PsE[P]/(PsTs + PcollTc)
I also changed Tc to
include ACK_TimeoutDIFS+E[P]+SIFS+ACK_..
Huge degradation of
throughput for either
static or dynamic WLANs
Will see simulations agree
.80
n = 50
n = 20
n = 10
n=5
.60
.40
.20
.00
.10
.30
.50
.70
.90
Probability of hidden stations
Our Throughput Model – Finite Load(1)
Similar grouping into k groups, but now with
identical loads, li individually and S li = l per group
Packet from a group must be successful both from its
group and all other groups-
Further, transmission probabilities from k contending
groups consisting some stations each
Plug Ps and Ptr into throughput equation
Can be used for both basic and RTS/CTS
Our Throughput Model – Finite Load(2)
Now have hidden groups, but assume same rate per
group persists (i.e. allow only same rate within group)
Extend the previous Ps and Ptr to separate out
reachable and hidden stations, in adjacency graph, i.e.,
Assumption that reachable >= hidden. Is it valid?
It is not obvious how to calculate t. One idea may be
from scheduler’s history at stations
Certainly justifies RTS/CTS, MACAW, DCF+, etc.
Simulation Topology & Traffic
2
<=250m
4
1
3
5
 Simulations in ns-2
 914MHz Lucent
WaveLAN DSSS PHY
 Omni-antenna with
250m range
 Modified CMU scene
generator to create
 RTS threshold => 3000 bytes
hidden stations, static
 1028 bytes (8224 bits) packets
topology, random
pause time
 Inter-packet gap = 0 (saturation)
and 1/rate (finite load)
 Modified CMU traffic
 CBR traffic over UDP links
generator for variable
packet size, intervals
 Script to calculate various
throughputs from trace
>250m
Saturation Simulation Results
1.0
n=50
Normalized throughput
Simulated with certain
percentage hidden stations
for 5, 10, 20, 50 stations
Results agree with model to
some extent
Differences can be attributed
to hidden stations may not
always have packets (as
assumed in the model)
.80
n=20
n=10
.60
n=5
.40
.20
.00
Still need to experiment with
m and simulate finite load
throughput
.10
.30
.50
.70
.90
Probability of hidden stations
Discussions & Conclusion
Hidden station models are sophisticated and can be
used in many applications involving “carrier sense”
Saturation throughput model is valid and should be
considered as an extension to Bianchi’s DCF model
Proposed finite load model is computationally
expensive and needs further simplification. Finite
load throughput model is an important step towards a
general model of DCF and its derivatives
Though simulations are limited, it provides some
degree of validation to the throughput models
It was a worthwhile investigation indeed helping me
taking EE384* skills to different areas in networking
Summary & Future Work
Summarized prior art in
DCF throughput and
hidden station modeling
Developed static and
dynamic hidden station
models for 802.11 DCF
Developed a finite load
throughput model for
DCF
Integrated hidden station
models for different
types of loads
Showed limited
simulation and …
Fixed relationships among
reachable/hidden stations
Finite load validation with
CBR traffic (per group)
 Finite load validation with
VBR traffic, e.g. Bernoulli
IID, exponential, bursty, ..
Scheduling packets in fixed
src-dst pairs in multichannel medium, e.g. iSLIP
wireless networks 
Q&A
Simulation scripts, code, topologies, traffic pattern
files can be found athttp://www.stanford.edu/~sirahman/80211dcf/
THANK YOU