Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Na Yang*, Ilker Demirkol^, Wendi Heinzelman*
* Dept. of Electrical and Computer Eng., University of Rochester, NY, USA
^ Dept. of Telematics Eng., Universitat Politecnica de Catalunya, Barcelona, Spain
IWCMC’11 - Multimedia over Wireless Symposium
1/20
2/20
Optimize wireless image transmission
◦ in a user-centric way
More formally:
Objective
Minimizing
communication
energy for
battery-operated
cameras
Constraint
User-defined
image quality
Cross-layer Approach
•Image quality Application
layer
•Packet length MAC layer
•Transmit power Physical
layer
3/20
4/20 [1] Sabir et al., IEEE Tr. on Image Processing, 2006.
(internal parameter): packet-level error tolerance rate, used to relate image quality constraint to packet-level optimizations.
Image compression, transmission, reception
5/20
Minimize average energy per image ◦ to successfully transmit and receive one image for
a given distance and image quality threshold
{ , , }
max
min ( , , , ),
. . ( , , , ) ,
,
0 ,
0,
0,
t Limage t L
P L
t L g
g
L
t
E P L d
s t PSNR P L d PSNR
d d
L L
P
(Packet length) (Transmit power) (Packet-level error tolerance rate)
Optimization parameters
(Use-defined image quality)
6/20
Energy consumption model ◦ Packet structure
◦ Average energy per packet
Only optimize payload length
··image pkt
L
V RE E
L
V pixels/image R bits/pixel
N : average number of transmissions
Time on the working state
1
·
on L UH H P
L UH H P
b
LLT T T T T
L L L LR
Uncoded BPSK
ct cr
on t c
pk tx
t c
t
r
E N E E E
N T P P P
3.5
0
·2 41
,2
ltGP
BN d
b tP P d e
AWGN channel
7/20
Energy consumption model ◦ Average number of transmissions N
Maximum number of bit errors in an accepted packet
0
1L UH
L UH
L UH
L LL L ii i
acc L L b b
i
Pr C P P
Summation of all the possibilities of having less than or equal to errors. i : number of bit errors in a packet
1
1
1[ ] 1
n
acc acc
n acc
N E n n Pr PrPr
L UHL L
n : number of transmissions
8/20
How to set δ?
BER , R vs. received image quality [1]
Source distortion Channel distortion
9/20
BERacc = Pb ?
0
· 1
, , ,·
L UH
L UH
L UH
L LL L ii i
L L b b
iacc t L
L UH acc
i C P P
BER P L dL L Pr
average number of errors in accepted packets
PSNR constraint
BERacc constraint
Fix source coding rate R
10/20
Differentiating , , and bP accBER
bP
11/20
Find the optimal solutions by using numerical optimizations in MATLAB
Optimal value searching range
User-defined conditions
max
7 1
7 1
10 bits
10 10
10 10
L
b
L L
P
10 m 150 m
12.8 dB 28.4 dB
g
g
d
PSNR
* * *, ,L bL P
12/20
Parameter Notation ZigBee mote WiFi Mote
Details MICAz mote [9]
Microchip ZG2100M transceiver module [10]
Carrier frequency fc 2.4 GHz 2.4 GHz
Data rate Rb 250 kbps 2 Mbps
Upper layer overhead length
LUH 160 bits 160 bits
PHY/MAC layer header length
LH 32 bits 32 bits
Preamble length LP 32 bits 32 bits
Payload length LL 128 bytes 2048 bytes
Transmit circuit power Pct 33 mW 379.5 mW
Receiver circuit power Pcr 59.1 mW 280.5 mW
Transmit power Pt 12.9 mW 128.7 mW
Image size V=512 x 512 pixels Maximum payload length Lmax= 105 bits=12.5 KB
13/20
The optimal cases
14/20
Almost no retransmissions (retransmissions are a very energy-inefficient way to enhance the received signal quality.
*
max
* decreases decreasesg b
L L
PSNR P
Minimum energy per image
Sensitivity analysis of the optimization parameters
PSNR=19.5 dB, d=30 m
Larger packet, relative shorter overhead.
Small Pt , more retransmissions; Large Pt , waste of energy.
Small , threshold too stringent; Large is OK, since Pb cannot be too large
15/20
Performance gain over error-free transmissions
Proposed
•Source
distortion
•Channel
distortion
Reference [6]
•Source
distortion
•Error-free
transmission
Saves around 10-20% of the energy for middle-to-large distances.
Better energy by distributing the distortion to both source coding and channel error distortion.
16/20
Performance gain over ZigBee and WiFi systems with default parameters
Under the same PSNR constraint: 28.4 dB
At short distances, save 35% over ZigBee mote and 18% over WiFi mote.
At large distances, reference energy increases sharply due to large number of retransmissions.
Proposed
•Optimized Pt
•Optimized LL
Reference
•Fixed Pt
•Fixed LL
17/20
Proposed optimization approach significantly reduces the total energy compared with a ZigBee and a WiFi commercial mote for middle-to-large distances.
For the same image quality, distributing distortion in both source coding and transmission processes achieves lower energy than only allowing distortion in source coding.
18/20
[1] M. F. Sabir, H. R. Sheikh, R. W. Heath, and A. C. Bovik, “A joint source-channel distortion model for JPEG compressed images,” IEEE Transactions on Image Processing, vol. 15, pp. 1349–1364, 2006.
[6] T. Wang, W. Heinzelman, and A. Seyedi, “Minimization of transceiver energy consumption in wireless sensor networks with AWGN channels,” in 46th Annual Allerton Conference on Communication, Control, and Computing, pp. 62 –66, 2008.
[9] MICAz mote, http://www.openautomation.net, “Data sheet for the MICAz motes, Crossbow Technology Inc..”
[10] Microchip ZG2100M/ZG2101M WiFi transceiver module, http://ww1.microchip.com/downloads/en/DeviceDoc/70624A.pdf.
19/20