|
|
|
Duncan Sharp |
|
2002 08 12 |
|
Communication Networks Laboratory |
|
http://www.ensc.sfu.ca/research/cnl |
|
School of Engineering Science |
|
Simon Fraser University |
|
|
|
|
|
Motivation & prior work |
|
Background – mobile radio systems |
|
Traffic source – public safety system |
|
Traffic analysis |
|
overall (long term summary) traffic patterns |
|
detailed analysis (arrival & departure
processes) |
|
Conclusions & further work |
|
|
|
|
|
Ad hoc networks – self organizing networks where
user terminals form the infrastructure |
|
In theory, a good fit for mobile radio systems,
especially for |
|
public safety, |
|
emergency response & |
|
disaster recovery |
|
|
|
|
|
|
Ad hoc is relatively “hot” research topic |
|
Open issues for mobile ad hoc networks |
|
routing protocols |
|
multicast & QoS support (for real time
voice) |
|
scaling the above in variable density graphs |
|
especially when topology is changing (dynamic) |
|
|
|
|
... there is very little suitable mobile radio
type multicast voice traffic data in the public domain |
|
|
|
|
Simple (simplex) mobile radio |
|
Extend the range / coverage (repeaters) |
|
Extend the capacity (add channels) |
|
Improve efficiency (trunking) |
|
Extend the range again (multi/simulcast) |
|
|
|
|
|
A group of compatible radios communicate amongst
themselves on a single channel / frequency (simplex) |
|
Terms |
|
push-to-talk (PTT) – a call |
|
talk group (voice multicast) |
|
single hop ad hoc network |
|
|
|
|
More complex |
|
|
|
Terms |
|
repeater |
|
fixed infra-structure |
|
ad hoc stub network |
|
|
|
|
“Busy” talk group (channel) utilization ranges
20% à
50% |
|
Add more talk groups by adding more radio
channels at repeater |
|
User radios are frequency agile |
|
Each repeater channel (transceiver) is a
separate talk group |
|
User selects talk group by changing channels |
|
|
|
|
|
|
Share channels as a common traffic pool to
improve overall utilization (queuing system) |
|
Set aside a control channel to take requests and
assign a channel for each PTT |
|
|
|
|
|
Add cells for wide area coverage |
|
|
|
Terms |
|
“multicast” adjacent cells use different
frequencies |
|
“simulcast” adjacent cells use same frequency
(synchronize phase & timing in overlap areas) |
|
|
|
|
|
E-Comm system architecture |
|
multi agency – police, fire, ambulance, etc |
|
digital trunked system |
|
Vancouver is simulcast cell |
|
System characteristics |
|
Traffic data files |
|
call summary reports |
|
call detail activity reports |
|
|
|
|
|
|
|
|
|
|
Each user agency (e.g., police, fire, ambulance)
is assigned one or more “talk groups” (multi-cast or “party-lines” for work
teams to converse), for example: |
|
Dispatch talk groups |
|
Chat talk groups |
|
Information talk groups |
|
Tactical / Incident talk groups |
|
Etc... |
|
|
|
|
Most data taken from Vancouver cell as it is the
busiest |
|
Taken from System Control Console on discs (CD,
magnetic) |
|
Various file formats (usually “csv” text) |
|
Manipulated in Excel |
|
|
|
|
from a stable 51 day period beginning 2001 09 24 |
|
|
|
|
Two days of records used (2001 11 01-02) |
|
Caller & callee identification removed |
|
Data voids identified & considered |
|
|
|
|
|
Notation |
|
Summary data - exploration of daily and weekly
patterns |
|
Detailed data - exploration of call arrival
& departure processes |
|
informal model fitting |
|
primary tools:
Excel, S-Plus |
|
|
|
|
|
Traffic is an arrival process |
|
calls arrive at some time Ta, measured in sec
from 0 |
|
inter-arrival time Ti (time elapsed between call
arrivals); thus arrival rate per sec is 1/Ti |
|
and a departure process |
|
calls have a duration or holding time Tc; thus
departure rate is 1/Tc |
|
Traffic is measured by averaging over an
integration period, Tp. In Erlangs
(utilization): A = Tc / Ti |
|
|
|
|
weekly patterns are discernable |
|
|
|
|
Average, maximum & minimum of all
Wednesdays during 51 day study period |
|
|
|
|
Average, maximum & minimum of all
Fridays during 51 day study period |
|
|
|
|
Average traffic per day |
|
Average peak hour per day |
|
Maximum peak recorded |
|
|
|
|
Analyzing the arrival and departure processes |
|
Trace overviews |
|
Holding times (Tc) |
|
Inter-arrival times (Ti) |
|
|
|
|
Plotting arrival time (Ta) against call holding
time (Tc) |
|
|
|
Talk group (traffic “sources”) |
|
System (traffic from multiple talk groups) |
|
|
|
|
|
|
|
|
|
Analysis of departure process in terms of call
holding time (Tc) |
|
lag plots, |
|
distribution (frequency & cdf) plots |
|
autocorrelation plots |
|
Looking at |
|
one talk group (chat) |
|
system (multiple talk groups, 10) |
|
|
|
|
|
|
|
|
|
|
10 genuine talk group traffic traces
combined into system |
|
|
|
1 day |
|
|
|
|
|
|
looks exponential after for Tc > ~3 s |
|
|
|
|
|
|
|
Analysis of arrival process in terms of
inter-arrival times (Ti) |
|
lag plots (look at call sessions) |
|
distribution (frequency & cdf) plots |
|
autocorrelation plots |
|
variance-time plots |
|
Usually looking at one hour segments for reasons
of stationarity |
|
|
|
|
Does not have the single “shot-gun” blast
look of a single random process |
|
|
|
|
a series of sessions is evident |
|
|
|
|
Arrival time between “sessions” |
|
|
|
Threshold for session decision set at 25 sec |
|
|
|
|
Arrival times between calls within each
“session” |
|
|
|
Threshold for session decision set at 25 sec |
|
|
|
|
|
|
Session threshold time may contribute to “ill”
behavior |
|
|
|
|
Distribution “appears” more exponential |
|
|
|
|
|
|
does not show evidence of long range dependency |
|
|
|
|
10 genuine talk group traces combined into
system |
|
|
|
Note artifacts from 0.01 sec resolution |
|
|
|
|
10 genuine talk group traces combined into
system |
|
|
|
|
10 genuine talk group traces combined into
system |
|
|
|
|
Evidence of long range dependency (LRD) |
|
|
|
|
Call holding time is approximately exponential
for individual talk group & system traffic (Markovian departure
process) |
|
For individual talk groups, call arrivals are
approximately Poisson but with different mean arrival rate for inter /
intra session (Markovian arrival process) |
|
For combined talk groups (system), call arrivals
show evidence of LRD |
|
|
|
|
|
|
|
|
|
LRD in the arrival process may affect how system
performance is estimated, so further work: |
|
investigate cause of the LRD |
|
investigate design point (busy period)
definition / determination |
|
investigate performance estimating methods |
|
Also of interest is to analyze the spatial
distribution of talk group traffic (mobility) |
|
|
|
|
Shahir Popatia, E-Comm, for access and use of
the data |
|
Keith Bandy, Planetworks, for data extraction
and initial manipulation |
|
Ljiljana Trajkovic, SFU, for insight into the
data and processes |
|
|
|
|
P. Cohen et al, “Traffic Analysis for Different
Classes of Users of Land Mobile Comm Systems”, IEEE Vehicular Technology
Conference, pp. 283-285, 1983 |
|
G. Stone, “Appendix D of SRSC Final Report,
Public Safety Wireless Communications User Traffic Profiles & Grade of
Service Recommendations”, US Dept of Justice, 13 March 1996 |
|
D. Tang & M. Baker, “Analysis of
Metropolitan-Area Wireless Network”, Proceedings of 5th Mobicom, pp .13-23,
1999 |
|
G. Hess, “Land-Mobile Radio System Engineering”,
Artech House, 1993 |
|
D. Duffy et al, “Statistical Analysis of
CCSN/SS7 Traffic Data from Working CCS Subnetworks”, IEEE JSAC, Vol. 12,
No. 3, pp. 544-551, April 1994 |
|
|
|
Notes