design principles for energy harvesting wireless
TRANSCRIPT
Design Principles for Energy Harvesting Wireless Communication Networks
Aylin Yener
Acknowledgment: NSF 0964364
Introduction
Ubiquitous
Mobile / Remote
Energy-limited
6/16/20142
Many sources
Abundant energy
Green
Energy Harvesting
Wireless Communications
Energy Harvesting Wireless Networks
Energy Harvesting Networks
Wireless networking with rechargeable (energy
harvesting) nodes:
Green, self-sufficient nodes,
Extended network lifetime,
Smaller nodes with smaller batteries.
A relatively new field with increasing interest.
6/16/20143
Some Applications
Wireless sensor networks
Greencommunications
6/16/20144
Harvesting Energy
Fujitsu’s hybrid device
utilizing heat or light.
Image Credits: (top) http://www.fujitsu.com/global/news/pr/archives/month/2010/20101209-01.html(bottom) http://www.zeitnews.org/nanotechnology/squeeze-power-first-practical-nanogenerator-developed.html
Nanogenerators built at
Georgia Tech, utilizing strain
6/16/20145
New Network Design Challenge
A set of energy feasibility constraints based on energy harvests govern the communication resources.
Main design question:
When and at what rate/power should a rechargeable (energy harvesting) node transmit?
Optimality? Throughput; Delivery Delay
Outcome: Optimal Transmission Schedules6/16/2014
6
One Energy harvesting transmitter.
Find optimal power allocation/transmission policy that departs maximum number of bits in a given duration T.
Energy available intermittently.
Up to a certain amount of energy can be stored by the transmitter BATTERY CAPACITY.
Throughput Maximization
[Tutuncuoglu-Y.’12]
6/16/20147
Energy harvesting transmitter:
Energy arrives intermittently from harvester Transmitter has backlogged data to send within a
deadline T. Stored in a finite battery of capacity
System Model
Ei
transmitter receiver
Energy queueData queue
Emax
Emax
6/16/20148
Energy arrivals of energy at times
Arrivals known non-causally by transmitter, Design parameter: power rate .
iE is
)( pr
E0
t
TE1 E2 E3
s1 s2 s3s0
System Model
6/16/20149
Notations and Assumptions
Power allocation function:
Energy consumed:
Transmission with power p yields a rate of r(p)
Short-term throughput: T
dttpr0
))((
)(tp
T
dttp0
)(
6/16/201410
Power-Rate Function
Transmission with power p yields a rate of r(p)
Assumptions on r(p):
i. r(0)=0, r(p) → ∞ as p → ∞ ii. increases monotonically in piii. strictly concaveiv. r(p) continuously differentiable
Example: AWGN Channel,
)( pr
Rate
Power
Nppr 1log
21)(
6/16/201411
Power-Rate Function
r(p) strictly concave, increasing, r(0)=0 implies
)( pr
Rate
Power
pppr in decreasing llymonotonica is )()tan(
Given a fixed energy, a longertransmission with lower power departs more bits.
Also, r -1(p) exists and is strictly convex
6/16/201412
Lazy Scheduling, El Gamal 2001
Battery Capacity:
Energy Constraints
(Energy arrivals of Ei at times si)
Energy Causality: nn
n
i
t
i stsdttpE
'0)( 1
1
0
'
0
nn
n
i
t
i stsEdttpE
')( 1
1
0
'
0 max
nn
n
i
t
i stsnEdttpEtp '0)(0)( 1
1
0
'
0 max , ,
Set of energy-feasible power allocations
6/16/201413
Energy “Tunnel”
cE
t1s 2s
0E1E
2EmaxE
Energy Causality
Battery Capacity
6/16/201414
Optimization Problem
Maximize total number of transmitted bits by deadline T
Convex constraint set, concave maximization problem
T
tptptsdttpr
0)()(..,))((max
nn
n
i
t
i stsnEdttpEtp '0)(0)( 1
1
0
'
0 max , ,
6/16/201415
Necessary conditions for optimality of a transmission policy
Property 1: Transmission power remains constant
between energy arrivals.
Proof: By contradiction
r(p)dtr(p(t)) dt(t)) r(p
elsetptttptttp
tp
,tttptp
* of concavity strict to due Then
Define
energy total given with ][0, some for Let
00
222
111*
2121
)(],[)(],[)(
)(
)()(
6/16/201416
Let the total consumed energy in epoch be
which is available in energy queue at
Then a constant power transmission
is feasible and strictly better than a non-constant
transmission.
Necessary conditions for optimality
],[ 1ii ss totalE
ist
],[, 11
iiii
total sstss
Ep
Transmission power can change only at is
6/16/201417
Property 2: Battery never overflows.
Proof:
pr(p)dtr(p(t))dt(t))pr(
elsetp
tptp
TT
in increasing is since Then
Define
time at overflows of energy an Assume
00
)(
],[)()(
Necessary conditions for optimality
6/16/201418
Necessary conditions for optimality of a transmission policy
Property 3: Power level increases at an energy arrival instant
only if battery is depleted. Conversely, power level decreases
at an energy arrival instant only if battery is full.
Policy can be improved Policy cannot be improved
p(t)
p’(t)p*(t)
r(p(t))dt(t))dtpr(
6/16/201419
Necessary conditions for optimality of a transmission policy
Property 3: Power level increases at an energy arrival instant
only if battery is depleted. Conversely, power level decreases
at an energy arrival instant only if battery is full.
Policy can be improved Policy cannot be improved
p(t)p’(t)
r(p(t))dt(t))dtpr(
p*(t)
6/16/201420
Necessary conditions for optimality of a transmission policy
Property 4: Battery is depleted at the end of transmission.
Proof:
increasing is since Then
Define
p(t) after remains of energy an Assume
r(p)dtr(p(t))dt(t))pr(
elsetp
TTtptp
TT
00
)(
],[)()(
6/16/201421
Necessary Conditions for Optimality
Implications of Properties 1-4:
Structure of optimal policy: (Property 1)
For power to increase or decrease, policy must meet the upper or
lower boundary of the tunnel respectively (Property 3)
At termination step, battery is depleted (Property 4).
An algorithmic solution can be found recursively, see
[Tutuncuoglu-Y.12]
constant ,}{ ,0
)( 1nnn
nnn psiTt
itiptp
6/16/201422
Energy “Tunnel”
cE
t1s 2s
0E1E
2EmaxE
Energy Causality
Battery Capacity
6/16/201423
Shortest Path Interpretation
Optimal policy is identical for any concave power-rate function!
Let , then the problem solved becomes:
The throughput maximizing policy yields the shortest path through the energy tunnel for any concave power-rate function.
1)( 2 ppr
)(..1)(min0
2
)(tptsdttp
T
tp
length of policy path in energy tunnel
)(..1)(max0
2
)(tptsdttp
T
tp
6/16/201424
Shortest Path Interpretation
Property 1: Constant power is better than any other alternative
Shortest path between two points is a line (constant slope)
E
t06/16/2014
25
Alternative Solution(Using Property 1)
Transmission power is constant within each epoch:
KKT conditions optimum power policy.
NnEpLEts
prL
n
iiii
N
iiipi
,...,10..
)(.max
max1
1
}1,{)( ,N,iiepocht,ptp i
(Li: length of epoch i)
(N: Number of arrivals within [0,T])
6/16/201426
0
0
1max
1
n
iiiin
n
iiiin
EpLE
EpL
full is battery when only positive only 's
empty is battery when only positive are 's
Solution
Complementary Slackness
Conditions: nEpLE
nEpL
n
iiiin
n
iiiin
0
0
1max
1
n
nN
nj jjnp
positive a at decreases positive a at increases
1)(
1*
6/16/201427
(Water Filling-Goldsmith 1994)
Directional Water-Filling
[Ozel, Tutuncuoglu, Ulukus, Y., 2011]
Harvested energies filled into epochs individually
0 tO O O
0E 1E 2E
Water levels (vi)
6/16/201428
Directional Water-Filling
Harvested energies filled into epochs individually
Constraints:
Energy Causality: water-flow only forward in time
0 tO O O
0E 1E 2E
Water levels (vi)
6/16/201429
Directional Water-Filling
Harvested energies filled into epochs individually
Constraints:
Energy Causality: water-flow only forward in time
Battery Capacity: water-flow limited to Emax by taps
0 tO O O
0E 1E 2E
Water levels (vi)
6/16/201430
Example
31
0
20 E51 E
1s
p
12 E 93 E 74 E
2s 3s 4s
10max E
6/16/2014
Directional Water-Filling
Energy tunnel
and directional
water-filling
approaches
yield the same
policy
E
t0
0 tO O O
0E
OO O
0E
1E 2E 3E 4E 5E
1E 2E
3E4E
5E
6/16/201432
Directional Water-Filling
Energy tunnel
and directional
water-filling
approaches
yield the same
policy
E
t0
0 tO O O OO O
6/16/201433
Simulation Results
Improvement of optimal algorithm over an on-off transmitter in a simulation with truncated Gaussian arrivals.
6/16/201434
Extension to Fading Channels
[Ozel-Tutuncuoglu-Ulukus-Y.‘11]
Find the short-term throughput maximizing and transmission completion time minimizing power allocations in a fading channel with known channel states.
Finite battery capacity
6/16/201435
System Model
AWGN Channel with fading h :
Each “epoch” defined as the interval between two “events”.
)1log(21),( hphpr
0E 2E 3E 6E 7E
0 tx x x x
Fading levels
L1 L4 L7
h1h2=h3=h4
h5h6=h7
h8
0,..5,4,1 E
O O O O O
6/16/201436
TM Problem with Fading Transmission power constant within each epoch:
Maximize total number of transmitted bits by deadline T
Solution once again is directional waterfilling.,...,MnEpLEts
phL
n
iiii
M
iii
i
pi
10..
)1log(2
max
max1
1
}1,{)( ,M,iiepocht,ptp i
6/16/201437
n
M
ni iin h
p 11*
Directional Water-Filling for fading channels
0 tO O Ox
0E 2E 4E
Fading levels (1/hi)
Water levels (vi)
x
Same directional water filling model with added
fading levels.
Directional water flow (Energy causality)
Limited water flow (Battery capacity)
6/16/201438
Multiple EH Transmitters
How to allocate power when there are more than one energy harvesting transmitters sharing the same medium?
How do the network parameters affect the optimal policy?
Many recent multi-node models, e.g., MAC (and BC) [Ozel,Yang,Ulukus’11,’12], Relay [Cui, Zhang,’12], [Oner, Erkip’13], [Varan, Y.’13], …, Two-way Relay [Tutuncuoglu, Varan, Y.’13],…
6/16/201439
Interference and EH: Gaussian IC with EH Transmitters
E1[i]
Transmitter 1 Receiver 1
Energy queue
E1,max
E2[i]
Transmitter 2 Receiver 2
E2,max
1
1
a
b
2212
1211
ZXXbY
ZXaXY
[Tutuncuoglu-Y. ’12]
6/16/201440
,N, n,j
EipiEts
ipipr
j
n
ijj
N
i
121
][][0..
])[],[(max
max,1
12100
21 p,p
Sum-Throughput Maximization Problem:Find optimal transmission power/rate policies that maximize the total amount of data transmitted to both receivers by a deadline T=Nτ.
Problem Definition
6/16/201441
Claim:
Concavity of sum-rate
2121 ),( ppppr and in concave jointlyis
Given any transmission scheme achieving a sum-rate r(p1,p2), one can utilize time-sharing to construct concave sum-rate:
0,,10,)1(..
),()1(),(),,(
max),(
2
2121
21
21*
ppppptspprppr
ppr
ppr
jjjj
6/16/201442
Convex problem allows the solution to be found
using coordinate descent between and
Optimal Policy
Iterative Generalized Directional Water-filling(IGDWF):
constrained water-filling with generalized water levels:
nN
ni iip
n
n
prp
v
)()(
6/16/201443
][1 ip ][2 ip
Two-user IGDWF
44
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
GDWF for
User 1
6/16/2014
Two-user IGDWF
45
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
GDWF for
User 2
6/16/2014
Two-user IGDWF
46
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
GDWF for
User 1
6/16/2014
Two-user IGDWF
47
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
GDWF for
User 1
GDWF for
User 2
6/16/2014
Take-away
Multiple energy harvesting transmitters sharing the same medium: transmit policy of one depends on the others.
Care need to be exercised in iteratively finding the water-filling solutions.
Policies do depend heavily on the channels, some instances converge in one iteration and/or result in simplified algorithms, e.g., strong interference.
6/16/201448
Multiple EH Transmitters:The concept of Energy Cooperation[Gurakan-Ozel-Yang-Ulukus ’12]
Intermittent energy nodes may be energy deprived!
Relay can “receive”
the energy to forward
the data.
Energy cooperation
between nodes can be
very useful!
6/16/201449
Wireless Energy Transfer
Already present in RFID systems
New technologies like strongly coupled
magnetic resonance reported to achieve
high efficiency in
mid-range
Image Credits: (top) http://www.siongboon.com/projects/2012-03-03_rfid/image/inlay.jpg(middle) http://www.witricity.com(bottom) http://electronics.howstuffworks.com/everyday-tech/wireless-power2.htm
Transfer energy to a 60-watt bulb with 50 percent efficiencyfrom 6-feet & 90 percent efficiency from 3-feet (MIT).75 percent efficiency from two to three feet away (Intel).
6/16/201450
Wireless Energy TransferIn-body (in-vein) wireless devices
6/16/201451
Image Credits: (top) http://www.extremetech.com/extreme/119477-stanford-creates-wireless-implantable-innerspace-medical-device (bottom) http://www.imedicalapps.com/2012/03/robotic-medical-devices-controlled-wireless-technology-nanotechnology/
Poon, 2012
Energy Harvesting and Cooperating Models (EH-EC)
52
Time slotted model, N slots
with length T, indexed by i
K transmitters receive energy
packets of size Ej[i] at the
beginning of the ith time slot
][1 iE
][2 iE
1
2
][1 ip
][2 ip
Received energy stored in an infinite size battery
In slot i, node k transmits with power pk[i]
6/16/2014
EC-EH
53
In time slot ,transmitter sends transmitter an energy of with efficiency
Uni-directional energy transfer is a special case with
][1 ip
][2 ip
n
ik
K
jjkkjkjk
batk TipiiiEnE
1 1,,, ][][][][][
1
2
][1 iE
][2 iE
Harvested energy Received and sent energy
Energy used for transmission
1221
Kkjαα j,kk,j ,...,1,,0,0
i kj
][, ijk k,jα
Energy in node k’s battery at the end of the ith time slot:
6/16/2014
EC-EH
54
Energy Constraints:
Non-negativity of transmit power and transferred energy:
0][][][][][1 1
,,,
n
ik
K
jjkkjkjk
batk TipiiiEnE
NnKkjnnp kjk ,...,1,,...,1,,0][,0][ ,
Energy Causality: Energy required by transmission or transfer is available, i.e., harvested:
What is the sum-capacity of EC-EH-MAC and EC-EH-T(wo)-W(ay)-C?
6/16/2014
Sum-Capacity:
EC-EH-TWC
55
),,0(~ 2
21122
12211
kkN
NXhXY
NXhXY
N
21 1log211log
21 ppCTWC
S
p1
p2
][1 iE ][2 iE
1 21h
2h
1,2
2,1
6/16/2014
EC-EH-MAC
56
),,0(~
...2
11
NN
NXhXhY KK
:by and Normalize
k
k
hXXYY
h
','
2
K
kk
MACS pC
11log
21 Sum-Capacity:
R
p1
p2
...
pK
1h
2h
Kh
1
2
K
][1 iE
][2 iE
][iEK
1,2
2,1K,1
1,K
K,22,K
6/16/2014
Problem Statement
57
Nn,...,Kj,k
TipiiiE
npnts
ipipipCN
n
ik
K
jjkkjkjk
kjk
N
iKS
nnp jkk
,...,1,1
0][][][][
,0][,0][..
])[],...,[],[(1max
1 1,,,
,
121
][],[ ,
Find maximum achievable sum-rate by optimizing the energy transfer and energy expended for tx.
6/16/2014
[Tutuncuoglu-Y. ’13]
Simplifying the Problem
58
1) Optimal Routing of Energy Transfers:Use equivalent transfer efficiency values that reflect the optimal routing of energy transfers.
2) Procrastinating Policies:Restrict to a subset of policies that delay energy transfer unless transferred energy is used immediately
3) Decomposition:Solve energy transfer and power allocation separately
6/16/2014
Redefine effective efficiency values as
where is any feasible energy transfer path
Routing Energy Transfers
59
2,1 3,2
1 2 3
3,1 Energy can be transferred through multiple paths.
Optimal policy chooses the highest efficiency path.
Transferring and receiving energy simultaneously is suboptimal.
jeeeekeejk mm
,,,),...,(, ...max211
1
),,...,,,( 21 jeeek m
6/16/2014
Procrastinating Policies
60
Definition: A procrastinating policy satisfies
i.e., the energy received by a node is not greater than the energy required for transmission within that time slot.
In a procrastinating policy, a node does not transfer energy unless the receiving node intends to use it immediately.
Lemma: There exists at least one procrastinating policy that solves the sum-capacity problem.
Restrict search space to such policies
][][1
,, iTipK
kjkjkj
6/16/2014
Sum-Capacity Problem
61
Nn,...,Kk
TipiE
npts
ipipCN
n
ikk
k
N
iKSnpk
,...,1,1
0][][
,0][..
])[],...,[(1max
1
11
*
][
Define consumed powers
Sum-Capacity problem can be decomposed as
][][1][][ ,1
,, iiT
ipip kj
K
jkjjkkk
,...,Kkj
iTipits
iiT
ipCC
K
jjkkjk
kj
K
jkjjkkS
iS
jk
1,
][][,0][..
])[][(1][max
1.,
,1
,,][
*
,
Power Allocation Energy TransferSolved directly (single slot)Solved via IGDWF
6/16/2014
EC-EH-TWC
62
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
2]1[1 E 5]2[1 E
4]2[2 E 7]4[2 E
5.01,22,1
6/16/2014
EC-EH-TWC
63
T 2T 3T 4 T
4
0
T 2T 3T 4 T
3
9
12
0
8
6
2
6
v1
v2
2]1[1 E 5]2[1 E
4]2[2 E 7]4[2 E1]1[2,1
2]4[1,2
2]1[1 p 2]2[1 p 2]3[1 p 1]4[1 p
0]1[2 p
2]2[2 p 2]3[2 p
7]4[2 p
5.01,22,1
6/16/2014
EC-EH-MAC
64
Energy transfer direction is determined by
Power allocation problem is solved as if a single transmitter with energy arrivals
jkjka ,max
][][1
iEaiEK
kkk
T0 t2T 3 T 4T
i
nnE
1][
]1[E]2[E
]3[E
]4[E
6/16/2014
Simulations
65
Energy transfer from 3 to 1 is optimal after this point, allowing sum-capacity improvement
6/16/2014
104.011.02.03.01
][
,]10,0[~][],[,]1,0[~][
/10
,110,105,100
sec1,100
1,3
,
32
1
190
3
2
1
jk
mJUiEiEmJUiEHzWN
dBhdBhdBh
,TN
MAC
TWC-Simulations
66
Uni-directional energy transfer may be suboptimal in both
directions
5.0,5.0,]10,0[~][,],0[~][
/10
,100,100
sec1,100
1,22,1
2
1
190
2
1
aamJUiEmJEhUiE
HzWN
dBhdBh
,TN
6/16/2014
676/16/2014
Information Theory of EH Transmitters
So far, we have assumed
sufficiently long time slots and
utilized the known rate
expressions.
What if energy harvesting is
at the channel use level, i.e.,
each input symbol is individually
limited by EH constraints?
Tx
iE
Rx)( pRRate
iXENC
iE
DEC1 1 0 0 0 1
iY
686/16/2014
Energy Harvesting (EH) Channel:
[Tutuncuoglu-Ozel-Ulukus-Y.‘13]
The channel input is restricted by an
external energy harvesting process.
State: available energy
Has memory (due to energy storage)
Depends on channel input
Causally known to Tx (causal CSIT)
ENCODERiXW
Energy Harvesting
Energy Storage
Information Theory of EH Transmitters
ii YX
696/16/2014
EH Channel
ENCODERW
iE
DECODERW
CHANNELiX iY
iS
XYP |
ii SX
),min( max1 EEXSS iiii
(Ch. input constrained by state)
(State has memory)(State evolves based on ch. input)1max E
BATTER
Y
1,0iX
706/16/2014
Capacity with Zero Storage
Let , and encoder can use arriving energy, i.e.,
Memoryless channel with causal state, [Shannon 1958]
0max E
)()(max qpHpqHCpZS
ENCODER DECODERii YX
WW
iE
. 1,0,1,0,0
ii
ii
XthenEifXthenEif
716/16/2014
Capacity with Infinite Storage
2121
,1),(
qqqH
CIS
As , a save-and-transmit scheme proposed for the
AWGN ch [Ozel, Ulukus, 12] is optimal.
Any codeword with can be conveyed without error
maxE
pX ][
ENCODER DECODERii YX
WW
iE
iSmaxE
726/16/2014
The Binary EH Channel
[Tutuncuoglu-Ozel-Y.-Ulukus’13]
Unit battery, Binary noiseless channel, Timing channel equivalent: encoding strategy; upper bound by
providing state info at the decoder.
ii YX 1max E
ENCODER DECODERii YX
WW
maxEiE
iS
][
)()(max 1 )1(1
))1((
)( TE
tpTHC t Tq
qH
tpBEHC
t
t
T
736/16/2014
A tighter upper bound than [Tutuncuoglu-Ozel-Y.-Ulukus’13]:
A lower bound on i.e., the information leaked to the receiver about the harvesting process.
)|;( UTZI
New Results on BEHC[Tutuncuoglu-Ozel-Y.-Ulukus’14](will be presented at ISIT 2014)
Improved encoding scheme as well:
ZUNZUZUZU
V1)mod(
1
746/16/2014
New Results on BEHC
Asymptotically optimal achievable
rate
Improved upper bound on capacity
Capacity within 0.03 bits
756/16/2014
EH DMC with CSIT&CSIR[Ozel-Tutuncuoglu-Ulukus-Y.’14](will be presented at ISIT 2014)
The battery state is available causally at both Tx and Rx Input symbol , with consuming units
of energy.Xi {0,1,..., K}
Si
ENCODER CHANNEL DECODERiX iY
WW
maxEiE
iS
Information flows both through the physical channel and the battery state. (e.g., communication is possible without channel)
iS iS
Xi k k
766/16/2014
EH DMC with CSIT&CSIR
ENCODER CHANNEL DECODERiX iiYY 21
WW
),|,( )1(121 iiii yxyyp
Δ
)1(2)1(1 ii YY
ENCODER CHANNEL DECODERiX iY
WW
maxEiE
iS iS iS
)|( ii xyp
Original EH channel with
CSIT and CSIR:
Equivalent channel with
feedback:[Chen-Berger ‘05]
Feedback does not increase capacity Weissman, Goldsmith 2009
776/16/2014
EH DMC with CSIT&CSIR
Conclusion• New networking paradigm: energy harvesting nodes
New design insights arise from new energy constraints!
Realistic concerns, e.g. storage capacity, storage
efficiency impact transmission policies.
Multi-terminal scenarios need to be handled with care.
Cooperation with an energy harvesting relay or energy
cooperation brings additional insights and possibilities.
Information theoretic formulations are challenging but
promising to yield new insights.6/16/2014
78
Current Studies and Open Problems
Information theoretic limits, optimal coding schemes for energy harvesters
Operational principles of energy harvesting receivers
Impact of EH on signal processing PHY algorithms
Impact on network protocols
Efficient online schedules, simple practical implementations
Papers: 6/16/2014
79
http://wcan.ee.psu.edu/