Post on 28-Aug-2020
transcript
Robust Topology Control for Indoor Wireless Sensor Networks
Chenyang Lu Computer Science and Engineering
Why Do We Need Topology Control? Reducing transmission power can reduce power
consump8on and reduce channel conten8on
But it’s challenging: Links have irregular and probabilis8c proper8es Link quality can vary significantly over 8me Human ac8vity and mul8-‐path effects in
indoor networks
3
Outline Empirical study ART algorithm Implementa8on and evalua8on Conclusion
4
Existing Empirical Studies Many studies explore link performance at a fixed
transmission power [Srinivasan 2006], [Woo 2003], [Reijers 2004], [Zhou 2004], [Lai 2003]
[Son 2004] evaluates older Chipcon CC1000 radios [Lin 2006] uses a simplified indoor environment (all nodes
have line-‐of-‐sight)
Our study considers modern, 802.15.4-‐compliant CC2420 radios in a complex office environment
5
Is Topology Control Bene;icial? Testbed Topology
0 dBm -‐15 dBm -‐25 dBm
6
Is Topology Control Bene;icial? Packet Recep8on Ra8o (PRR) Distribu8on Across Links
368 links (70.2%) receive NO packets at -‐25 dBm
Compared to 82 links (15.6%) @ -‐5 dBm
105 links (20.2%) receive ≥ 95% of packets at -‐25 dBm
7
Is Topology Control Bene;icial?
!!" !!# !$" !$# !" ##
#%$
#%!
#%&
#%'
#%"
#%(
#%)
#%*
#%+
$
,TX -./01,23456
788
,
,
$&(!9$!(
$#(!9$!+
$#"!9$!)
$#'!9$#"
Impact of TX power on PRR
3 of 4 links fail @ -‐10 dBm ...
... but have modest performance @ -‐5 dBm Insight 1: Transmission power should be set on a per-‐
link basis to improve link quality and save energy.
8
What is the Impact of Transmission Power on Contention?
High contention
Low signal strength
Insight 2: Robust topology control algorithms must avoid increasing conten8on under heavy network load.
9
Is Dynamic Power Adaptation Necessary?
Link 110 -‐> 139
10
Is Dynamic Power Adaptation Necessary?
Long-‐Term Link Stability
Insight 3: Robust topology control algorithms must adapt their transmission power in order to maintain
good link quality and save energy.
11
Are Link Indicators Robust Indoors? Two instantaneous metrics are ohen proposed as
indicators of link reliability: Received Signal Strength Indicator (RSSI) Link Quality Indicator (LQI)
Disagreement over which is a bejer indicator of PRR [Srinivasan 2006]: “RSSI is under-‐appreciated” [Lin 2006]: LQI and RSSI are both good proxies for PRR (ATPC alg.) TinyOS 2.1: LQI used to es8mate channel quality
Can you pick an RSSI or LQI threshold that predicts whether a link has high PRR or not?
12
Are Link Indicators Robust Indoors? Links 106 -‐> 129 and 104 -‐> 105
RSSI threshold = -‐86 dBm, PRR threshold = 0.9
6% false posi8ve rate 8% false nega8ve rate
RSSI threshold = -‐85 dBm, PRR threshold = 0.9
4% false posi8ve rate 62% false nega8ve rate
RSSI threshold = -‐84 dBm, PRR threshold = 0.9
66% false posi8ve rate 6% false nega8ve rate Insight 4: Instantaneous LQI and RSSI are not robust
es8mators of link quality in all environments.
13
Adaptive and Robust Topology control (ART)
w=10
Power Level = 7 6?
Target PRR = 80%
Ini8alizing Steady Trial
6 7
14
Avoiding Contention Naïve policy: When # of transmission failures goes above
threshold, then increase power level But what if this makes things worse?
Remember, higher power → more conten8on
Ini8ally increase power when # of failures > threshold, but remember # of failures in last window
If # of failures is worse than last 8me, then flip direc8on and decrease power instead
Cheaply tracks “gradient” of power-‐to-‐PRR curve
15
Experimental Setup ART implemented using TinyOS 2.1 CVS
Adds 392 bytes of RAM and 1582 bytes of ROM
Window size = 50, PRR threshold = 95%
Three experiments: Link-‐level Data collec8on High conten8on
16
Link-‐Level Performance Selected 29 links at random from 524 detected in empirical
study Transmijed packets round-‐robin over each link in batches
of 100, cycled for 24 hours (15000 packets/link)
PRR Avg. Current
Max Power 56.7% (σ = 2.5%) 17.4 mA (σ = 0) ART 58.3% (σ = 2.1%) 14.9 mA (σ = 0.32)
17
Link-‐Level Performance Link 129 -‐> 106
18
Data Collection Deployed Collec8on Tree Protocol [Gnawali 2008] rou8ng
algorithm and selected one testbed node as sink All 27 other nodes take turns sending batches of 200
packets 1800 total packets/node over 4 hours
Compare against maximum power and PCBL [Son 2004] Collects large amount of bootstrapping data (2 hrs. on testbed) Uses lowest power sewng with PRR ≥ 98% “Blacklists” links with PRR < 90%
19
Data Collection
0.7
0.75
0.8
0.85
0.9
0.95
1
Max-‐Power PCBL ART
Packet Delivery Ra
te
Packet Delivery Rate
20
Data Collection
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Max-‐Power PCBL ART
Rela=v
e En
ergy Con
sump=
on
CTP data Protocol overhead
Energy Consump8on
21
Data Collection Hop-‐Count vs. PRR
Max-‐power starves nodes with most expensive paths
22
Handling High Contention Select 10 links at random from testbed Send packets over all 10 links simultaneously as possible
(batches of 200 packets for 30 min.)
Compare again against PCBL and max-‐power Also run ART without “gradient” op8miza8on to isolate its
effect on PRR
23
Handling High Contention
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Max-‐Power PCBL ART ART (w/o gradient)
Packet Recep
=on Ra
te
Packet Recep8on Rate
24
Handling High Contention
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Max-‐Power PCBL ART ART (w/o gradient)
Rela=v
e En
ergy Con
sump=
on
Energy Consump8on
25
Handling High Contention
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Max-‐Power PCBL ART ART (w/o gradient)
Rela=v
e En
ergy/PRR
Energy Efficiency
50.9% more energy efficient than max-power
40.0% more energy efficient than max-power
26
Handling High Contention Distribu8on of PRR
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95 1
CDF(Pa
cket Recep
=on Ra
te)
Packet Recep=on Rate
Max-‐Power PCBL ART ART (w/o gradient)
27
Conclusions Our empirical study shows important new nega8ve results:
RSSI and LQI are not always robust indicators of link quality indoors
Profiling links even for several hours is insufficient for iden8fying good links
Inherent assump8ons of exis8ng protocols!
ART is a new topology control algorithm which is robust in complex indoor environments
ART achieves bejer energy efficiency than max-‐power without bootstrapping or link starva8on
28
References G. Hackmann, O. Chipara and C. Lu, Robust Topology Control for
Indoor Wireless Sensor Networks, ACM Conference on Embedded Networked Sensor Systems (SenSys), November 2008.
Y. Fu, M. Sha, G. Hackmann and C. Lu. Prac8cal Control of Transmission Power for Wireless Sensor Networks, IEEE Interna8onal Conference on Network Protocols (ICNP'12), October 2012,
29