Академический Документы
Профессиональный Документы
Культура Документы
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 1
Abstract—This paper studies the problem of cHarging tAsk Scheduling for direcTional wireless chargEr networks (HASTE), i.e., given
a set of rotatable directional wireless chargers on a 2D area and a series of offline (online) charging tasks, scheduling the orientations
of all the chargers with time in a centralized offline (distributed online) fashion to maximize the overall charging utility for all the tasks.
We prove that HASTE is NP-hard. Then, we prove that a relaxed version of HASTE falls within the realm of maximizing a submodular
function subject to a partition matroid constraint, and propose a centralized offline algorithm that achieves (1 − ρ)(1 − 1e )
approximation ratio to address HASTE where ρ is the switching delay of chargers. Further, we propose a distributed online algorithm
and prove it achieves 12 (1 − ρ)(1 − 1e ) competitive ratio. We conduct simulations and field experiments on a testbed consisting of 8
off-the-shelf power transmitters and 8 rechargeable sensor nodes. The results show that our distributed online algorithm achieves
92.97% of the optimal charging utility, and outperforms the comparison algorithms by up to 15.28% in terms of charging utility.
ters and receivers equipped with directional antennas. In wireless charging task scheduling issue for omnidirectional
this model, the power charging area for a charger and the wireless chargers in offline scenarios, which are fundamen-
power receiving area for a device are modeled as sectors. tally different from our paper. In the conference version of
A rechargeable device can be charged via wireless by a this paper [51], we initiated the first study on scheduling
charger with non-zero power if and only if they are located wireless charging tasks for directional wireless chargers and
in each other’s covered sector. All wireless chargers can designing online algorithms.
freely adjust its orientation in [0, 2π) while rechargeable
devices cannot. Moreover, a charging task initiated by a 1.3 Key Technical Challenges
rechargeable device consists of five elements: the position
We are faced with three major challenges to address HASTE.
and orientation of its associated device, the release time and
The first challenge is that HASTE is non-linear and is NP-
end time of the task, and its required charging energy. To
hard. HASTE is nonlinear because that the orientation of
evaluate the effectiveness of wireless charging for a task, we
chargers can be freely scheduled; a task can be either cov-
define the task’s charging utility as a linear and bounded
ered by a charger and have a certain constant power incre-
function with its harvested energy from its release time to
ment or not with no power increment, which has the flavor
its end time.
of 0-1 integer programming; the charging utility function
With these models, we consider two scenarios for charg-
is linear but bounded, let alone that we extend our results
ing task scheduling, i.e., offline and online. In the offline sce-
to the case where the utility function is a general concave
nario, information for all charging tasks is known a priori,
function. In addition, by reducing from the classical NP-
and thereby the scheduling policies for all chargers at any
hard separate assignment problem, we prove that HASTE is
moment can be determined beforehand. To accommodate
NP-hard.
practical concerns, we assume that each charger needs an
The second challenge is how to design an efficient cen-
amount of time for switching its orientation, which we
tralized offline algorithm for HASTE in the offline scenario
call switching delay. In the online scenario, charging tasks
while considering the switching delay of chargers. The
stochastically arrive, and chargers reschedule their orien-
switching delay happens if and only if a charger’s next
tations in realtime. Nevertheless, in addition to switching
intended orientation is different from its current orientation,
delay, each charger needs an additional amount of time for
which implies that the switching delay as well as its caused
recomputing the scheduling policies with negotiating with
performance loss is history-dependent. Moreover, the per-
neighboring chargers, which we call rescheduling delay. To
formance loss is difficult to evaluate as there are potentially
avoid global management effort and reduce update cost,
multiple tasks are affected by a charger’s switching delay,
we desire a distributed and local algorithm which is scal-
and the charging utility function for tasks is non-linear.
able with network size. For both scenarios, we want to
The third challenge is how to design an efficient dis-
dynamically schedule the orientations of chargers as time
tributed online algorithm for HASTE in the online scenario
goes on such that the overall weighted charging utility for
where all chargers are asynchronous and the rescheduling
all charging tasks is maximized. Moreover, we stress that
delay needs to be considered. To the best of our knowledge,
chargers can be either in the working mode for the offline
there are neither existing distributed online algorithms di-
scenario or in that for the online scenario, but cannot switch
rectly applicable to our problem even when the reschedul-
between these two different statuses. To sum up, we state
ing delay is omitted, nor existing online algorithms that deal
our problem HASTE as follows. Given a set of rotatable
with the case in our considered scenario with rescheduling
directional wireless chargers on a 2D area and a series of
delay being concerned for which the response is delayed
offline (online) charging tasks, scheduling the orientations of
and the algorithm is not truly “online”.
all the chargers with time in a centralized offline (distributed
online) fashion to maximize the overall charging utility for
all the tasks. 1.4 Proposed Approach
To address the first challenge, we propose that rather than
considering all possible orientations in [0, 2π) for chargers,
1.2 Prior Art we can safely consider a limited number of orientations for
On one hand, there exist numerous literatures [22]–[37] them without causing performance loss, and therefore, ex-
studying on the mobile charging case where one single tract the so-called “dominant task sets” as the corresponding
or multiple chargers travel in a field to charge wireless sets of covered tasks. Then, we neglect the switching delay
rechargeable devices to guarantee their normal working. for wireless chargers, and thus reformulate the original con-
They are fundamentally different from ours as we consider tinuous optimization problem into a discrete optimization
static chargers. problem HASTE-R. Further, we prove that the reformulated
On the other hand, the other works consider wireless problem is exactly a problem of maximizing a submodular
charger networks consisted of static wireless chargers, but function subject to a partition matroid constraint, which
nearly none of them investigate charging task scheduling. greatly facilitates approximation algorithm design.
In particular, most of them focus on scheduling issues in To address the second challenge, based on the theoret-
coarse granularity rather than task levels, such as those ical results obtained by addressing the first challenge, we
overlooking the harmful effect of high electromagnetic ra- can either use a simple greedy algorithm that achieves 12
diation (EMR) [16], [17], [38]–[41] and those taking the approximation ratio [52] or a randomized algorithm with
EMR safety into consideration [42]–[48]. To the best of our the optimal approximation guarantee, namely, 1− 1e approx-
knowledge, there is only one work [49], [50] investigate imation ratio [53]. Nevertheless, as the former is not good
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 3
enough and the latter is too computationally demanding, we working time in a charging time period [30], [31]. We refer
tailor the TABULARGREEDY algorithm proposed in [54], readers to survey [56] for more related works.
[55] to address HASTE-R as it can achieve an approximation Second, the other works are dedicated to wireless
ratio between 12 and 1 − 1e (1 − 1e as default in our setting) charger networks consisted of static wireless chargers, but
depending on the value of a control parameter and resulting nearly none of them consider charging task scheduling.
in different time complexity. Further, to bound the perfor- First, most of them study scheduling issues in coarse gran-
mance loss of switching delay, we exploit the concavity of ularity rather than task levels. On one hand, some works
the utility function and consider all the caused performance (e.g., [16], [17], [38]–[41]) overlook the detrimental effect of
loss for all impacted tasks in the worst case, and prove that the electromagnetic radiation (EMR) to human health. For
the switching delay introduces a constant factor of 1 − ρ in instance, He et al. considered the triangular deployment
the ultimate achieved approximation ratio, i.e., (1−ρ)(1− 1e ), problem of wireless chargers [38]. They attempted to min-
of the proposed algorithm, where ρ is the switching delay. imize the number of chargers while rechargeable tags can
To address the third challenge, we propose a distributed receive sufficient power. In addition, we first proposed the
online algorithm based on the proposed centralized offline directional charging problem based on empirical experimen-
algorithm to HASTE. We prove that if the rescheduling tal results, and investigated the ominidirectional charging
delay is neglected, as long as the local executions of a problem using directional chargers in [16], the wireless
charger and its neighbors are in order and repeat regularly charger placement problem for directional charging in [17],
with time, the achieved global charging utility is the same [39]–[41]. On the other hand, other literatures [42]–[48] take
as that of the centralized offline algorithm. Further, by the EMR safety into consideration, and guarantee that the
leveraging the concavity of the utility function and the sub- EMR intensity at any point in the area does not exceed a
modularity of the objective function in HASTE, we bound predefined EMR threshold. For instance, we presented and
the performance loss of scheduling delay, and prove that studied how to schedule non-adjustable chargers [42], [43]
our distributed online algorithm achieves 21 (1 − ρ)(1 − 1e ) and adjustable chargers [44], [45] to maximize the charging
competitive ratio. utility for chargers under the EMR safety constraint. Niko-
letseas et al. [46] considered more practical constraints such
as the energy limitations of chargers and devices, the non-
1.5 Evaluation Results
linear constraints in the time domain, and their goal is to
We conducted simulations and field experiments to evalu- optimize the amount of energy transferred from chargers to
ate our proposed algorithms. Our simulation results show devices and truly utilized. Moreover, we reported a wireless
that our proposed distributed online algorithm can achieve charger placement scheme that ensures EMR safety in [47].
92.97% of the optimal charging utility which corroborates Second, to the best of our knowledge, there is only one work
our theoretical findings, outperform the other two compari- [49], [50] that study the wireless charging task scheduling.
son algorithms by 10.96%. We implemented our algorithms Nevertheless, [49], [50] consider omnidirectional wireless
on a testbed consisting of 8 off-the-shelf TX91501 power chargers whose charging power is adjustable and focus
transmitters produced by [19] and 8 rechargeable sensor on offline scenarios, which are fundamentally different
nodes associated with 8 charging tasks. Our experimental from our paper. Moreover, we launched the first study on
results show that our distributed online algorithm out- scheduling wireless charging tasks for directional wireless
performs the comparison algorithms by up to 15.28% on chargers and designing online algorithms in the conference
average, and 29.63% at most. version of this paper [51].
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 4
We adopt the general and practical directional charging 3.2 Charging Utility Model
model proposed in [16]–[18]. As Figure 1 shows, a charger
We adopt a linear and bounded charging utility model for
si with working orientation denoted by vector −r→
θi can only
harvested energy for a task, which is similar to the charging
charge devices in a charging area in the shape of a sector
utility model for received power proposed in [17]. That is,
with charging angle As and radius D. A rechargeable device
the charging utility for a task is first proportional to the
oj with orientation denoted by vector −r→
φj can only receive
harvested energy of its associated device, and then reaches
non-zero power in a receiving area in the shape of a sector
a constant if the harvested energy exceeds a predetermined
with receiving angle Ao and radius D. The charging power
threshold, i.e.,
from si to oj is given by
( 1
α
, 0 ≤ ||si oj || ≤ D, Ej
· x, x ≤ Ej
(||si oj ||+β)2 U(x) = (1)
−−→ −→ 1, x > Ej
si oj · rθ − ksi oj kcos(As /2) ≥ 0,
i
Pr (si , θi , oj , φj ) =
and −
o−→ −→
j si · rφj − koj si kcos(Ao /2) ≥ 0.
where Ej is the required charging energy of charging task
0, otherwise Tj .
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 5
3.3 Problem Formulation and Hardness Analysis Algorithm 1: Dominant Task Sets Extraction
Let θi (t) (θi : R≥0 7→ {[0 2π) ∪ Ø}) be the func- Input: The wireless charger si , all charging tasks
tion of orientation for charger si with time t. Suppose {Tj }mj=1
the value of θi (t) at the kth time slot is θi,k if charger Output: All dominant task sets
m
si is not switching; otherwise, θi (t) is set to Ø and 1 Find the subset of charging tasks in {Tj }j=1 that cover
defined in θi (t) for all the chargers so that the overall T i is included in the covered set. During the rotating
charging utility is maximized. With all above, we define process, if the rotated angle is larger than 2π , then
the problem of cHarging tAsk Scheduling for direcTional terminate. If not, goto Line 3.
wireless chargEr networks (HASTE) as follows.
m
X Z tje n
X
(P1) max U= wj · U ( Pr (si , θi (t), oj , φj )dt)
θi,k
j=1
j
tr i=1
4 P ROBLEM R EFORMULATION
In this section, considering the complexity of the formu-
(
Ø, kTs < t ≤ (k + ρ)Ts
, θi,k 6= θi,k−1
s.t. θi (t) =
lation P1 of HASTE, we reformulate HASTE to make it
θi,k , (k + ρ)Ts < t ≤ (k + 1)Ts
tractable. In particular, we first propose a dominant task sets
θi,k , kTs < t(k + 1)Ts , otherwise
extraction algorithm for chargers to reduce the continuous
where k ∈ Z+
0 , and θi (0) = Ø solution space for orientations of chargers to a discrete one
0 ≤ θi,k < 2π. with limited choices. Then, we consider a relaxed version
The following theorem shows the complexity of HASTE. of HASTE, i.e., HASTE-R, and prove it falls into the realm
of maximizing a submodular function subject to a partition
Theorem 3.1. HASTE is NP-hard. matroid constraint, which assists the further algorithm de-
sign.
Proof: Due to space limit, we only sketch the proof
here. Suppose ρ → 0, tjr = 0 and tje = Ts for all charging
tasks, which means each task simply occupies the first time 4.1 Extraction of Dominant Task Sets
slot and we only need to consider one round scheduling Though each charger can continuously rotate within [0 2π),
in this time slot. Moreover, suppose the required charging we do NOT need to consider all possible orientations.
energy for each task Ej is so small that as long as a task Instead, we only need to care about the possible sets of
is covered by a charger, it certainly obtains an amount of covered tasks, whose number is obviously limited for any
energy greater then Ej and therefore achieves a charging given charger. Further, among these sets we only need to
utility of wj in the overall charging utility. Besides, though consider the following specific ones.
the orientation of chargers can be freely chosen in [0 2π), its Definition 4.1. (dominant task set) Given a set of tasks T 1i
covered sets of charging tasks can be enumerated in a fixed covered by a charger si with some orientation, if there doesn’t
number of steps and are limited, as we will see in Algo- exist another set of tasks T 2i covered by si with some other
rithm 1. Consequently, with the above settings, our problem orientation such that T 1i ⊂ T 2i , then T 1i is a dominant
changes to choosing the orientation for each charger among task set.
its candidate choices such that the overall charging utility of
We describe our algorithm for extracting dominant task
all tasks is maximized. We can prove this simplified problem
sets in Algorithm 1. Basically, the considered charger rotates
is NP-hard by reducing from the classical NP-hard separate
for 2π and extracts the dominant task sets one by one. We
assignment problem [61], which is defined as follows: given
use a toy example for illustration in our conference version
a set of bins and a set of items to pack in each bin, a
of this paper [51].
value for assigning item j to bin i, and a separate packing
constraint for each bin, i.e., for bin i, a family Ii of subsets of
items that fit in bin i, packing items into bins to maximize 4.2 Problem Relaxation and Reformulation
the aggregate value. Here we can regard each charger as a As the switching delay is hard to be analyzed for optimiza-
bin, each task as an item, each set of covered tasks for a tion, we first consider a relaxed version of HASTE, HASTE-
candidate orientation of charger si as a subset in the family R for short, by neglecting the switching delay of all chargers,
Ii for bin i, the achieved utility of a task as the value for and then analyze HASTE. We will bound the performance
assigning this item to a bin, and therefore, we can reduce loss for the relaxation in our proposed algorithms.
any instance of the separate assignment problem to the Suppose the obtained set of dominant task sets for
considered simplified problem. As the separate assignment p p
charger si is Γi , the pth dominant task set in Γi is Γi . Let xi,k
problem is NP-hard [61], HASTE is also NP-hard. be a binary indicator denoting whether the pth dominant
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 6
task set in Γi in the kth time slot is selected or not. For Besides, it can be easily proved that M = {S, I} is a matroid
convenience of expression, we define by verifying the three properties proposed in Definition 4.3.
Thus we have the following lemma.
( α
(||si oj ||+β)2
, 0 ≤ ||si oj || ≤ D,
Pr (si , oj ) =
0, otherwise. Lemma 4.1. The constraint in the scheduling problem RP1 can
p
be written as a partition matroid on the ground set S .
Moreover, we abuse the notation slightly by defining Γi 3
oj as ∃Tj 0 ∈ Γip |Tj 0 .oj 0 = oj . That is, there exists a charing Accordingly, problem RP1 can be rewritten as
p (RP2)
task Tj 0 in Γi and its associated position of rechargeable
device is oj . Then, the problem HASTE-R can be formulated m tje /Ts
X X X
as max f (X) = wj · U ( Pr (si , oj )Ts )
X
j=1 j p
(RP1) k=tr /Ts +1 Γi,k 3oj , i∈[n],
p
p∈{p|Θi,k =X∩Θi,k }
m tje /Ts
X∈I
X X X
max UR = wj · U ( xpi,k Pr (si , oj )Ts ) s.t.
p
xi,k p
j=1 j
k=tr /Ts +1 Γi 3oj ,
i∈[n], p∈[|Γi |]
where X is the decision variable, and f (X) (f : 2S → R≥0 )
|Γi |
is the objective function. Note that we abuse the notation
p p
slightly, and here Γi,k 3 oj means ∃Tj 0 ∈ Γi,k |Tj 0 .oj 0 = oj .
X
s.t. xpi,k = 1, (xpi,k ∈ {0, 1})
p=1 For RP2, we have the following critical lemma.
where xpi,k s
are the decision variables, Γi is the pth domi-
p Lemma 4.2. The objective function f (X) in RP2 is a monotone
nant task set in Γi . submodular set function.
Clearly, RP1 is a combinatorial optimization problem. Proof: By Definition 4.2, we need to check whether
To facilitate further analysis, we first give the following f (X) satisfies the three listed conditions.
definitions. First, when there are no active scheduling policies, i.e.,
Definition 4.2. [62] (submodular set function) Let S be a X = ∅, the received energy for any task is zero, then we
finite ground set. A real-valued set function f : 2S → R have f (X) = 0.
is normalized, monotonic and submodular if and only if it Second, let A be a set of scheduling strategies in S and
satisfies the following conditions, respectively: e ∈ S\A. For simplicity, define
tje /Ts
1) f (∅) = 0; X X
2) f (A ∪ {e}) − f (A) ≥ 0 for any A ⊆ S and e ∈ S\A; g(X, j) = U( Pr (si , oj )Ts )
j p
(3)
Γi,k 3oj , i∈[n],
3) f (A ∪ {e}) − f (A) ≥ f (B ∪ {e}) − f (B) for any k=tr /Ts +1
p
p∈{p|Θi,k =X∩Θi,k }
A ⊆ B ⊆ S and e ∈ S\B .
as the achieved utility for task Tj . It is easy to see that
Definition 4.3. [62] (matroid) A matroid M is a strategy g(A ∪ {e}, j) − g(A, j) ≥ 0 because there are possibly more
M = (S, L) where S is a finite ground set, L ⊆ 2S is a chargers cover task Tj as all possible dominant task sets that
collection of independent sets, such that: p p
cover Tj , i.e., Γi,k (i ∈ [n], p ∈ {p|Θi,k = A ∩ Θi,k }) would
1)∅ ∈ L; be enlarged as A becomes A ∪ {e}, and the utility function
2)if X ⊆ Y ∈ L, then X ∈ L; U(.) is non-decreasing. Hence we have
3)if X, Y ∈ L, and |X| < |Y |, then ∃y ∈ Y \X , X ∪ m
X
{y} ∈ L. f (A ∪ {e}) − f (A) = wj · [g(A ∪ {e}, j) − g(A, j)] ≥ 0. (4)
j=1
Sk
Definition 4.4. [62] (partition matroid) Given S = i=1 Si0 is Third, let A and B be two sets such that A ⊆ B ⊆ S and
the disjoint union of k sets, l1 , l2 , . . . , lk are positive integers, element e ∈ S\B . On one hand, it is easy to see that
a partition matroid M = (S, I) is a matroid where I = tj tj
e e
{X ⊂ S : |X ∩ Si0 | ≤ li for i ∈ [k]}. Ts
X X Ts
X X
Pr (si , oj )Ts − Pr (si , oj )Ts
We will show that the problem RP1 fits perfectly in t
j p
Γi,k 3oj , t
j p
Γi,k 3oj ,
k= Tr +1 k= Tr +1
the realm of maximizing a monotone submodular function s i∈[n], p∈P1 s i∈[n], p∈P2
subject to a partition matroid. First, we define Γi,k = Γi tj
e tj
e
Ts Ts
(k ∈ [K]) as the set of dominant task sets for charger si at the X X X X
= Pr (si , oj )Ts − Pr (si , oj )Ts
kth time slot, where K is the total number of time slots and j p
Γi,k 3oj , j p
Γi,k 3oj ,
p t
k= Tr
t
k= Tr
the notation [n] = {1, 2, . . . , n}. Then, we define Θi,k as the s
+1
i∈[n], p∈P3 s
+1
i∈[n], p∈P4
p (5)
corresponding scheduling policy for Γi,k , i.e., the orientation
p p p p
that covers Γi,k = Γi , for charger si at the kth time where P1 = {p|Θi,k = {A ∪ e} ∩ Θi,k }, P2 = {p|Θi,k =
p p
slot, define Θi,k = {Θi,k }p∈[|Γi,k |] as the set of scheduling A ∩ Θi,k }, P3 = {p|Θi,k = {B ∪ e} ∩ Θi,k }, and P4 =
p
policies for si at the kth time slot, and define a ground set {p|Θi,k = B ∩ Θi,k }. On the other hand, it is clear that
of all scheduling policies S = {Θi,k }i∈[n],k∈[K] . Further, we (U(x1 + ∆x) − U (x1 )) − (U(x2 + ∆x) − U (x2 )) ≥ 0, (6)
define the scheduling policies for all chargers at all K time
slots as X , which is subject to |X ∩ Θi,k | ≤ 1. Therefore, as for any x2 ≥ x1 ≥ 0 and ∆x ≥ 0 due to the concavity of the
Θi,k s are disjoint sets, we write the independent sets as charging utility function U(.).
Consequently, we have [g(A∪{e}, j)−g(A, i, q)]−[g(B∪
I = {X ⊆ S : |X ∩ Θi,k | ≤ 1 for i ∈ [n], k ∈ [K]}. (2) {e}, j) − g(B, j)] ≥ 0, and therefore,
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 7
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 8
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 9
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 10
Charging Utility
Charging Utility
Charging Utility
0.72 0.8 0.69 0.75
Fig. 2. As vs. charging utility (cen- Fig. 3. Ao vs. charging utility (cen- Fig. 4. ρ vs. charging utility (central- Fig. 5. C vs. charging utility (cen-
tralized offline algorithm) tralized offline algorithm) ized offline algorithm) tralized offline algorithm)
Charging Utility
Charging Utility
1
7.3.1 Impact of Charging Angle As 0.8
0.65 1
0.8
0.65
0.6
0.6
0.6
Our simulation results show that on average HASTE outperforms 0.4
0.55
0.6
0.4
0.55
0.5
0.45
0.2 0.45
0 0
4.34% and 6.03%), respectively, in terms of As . Figure 2 shows 70
60 10
0.4 70
60 10
0.4
0.35
20 20
that the charging utilities of HASTE, GreedyUtility, and Δt (min)
50
40 40
30 0.35
Δt (min)
50
40 40
30 0.3
30 50 Ej (kJ) 30 50 Ej (kJ)
GreedyCover steadily increase with the charging angle of
chargers As , and achieve the same maximum overall charg- Fig. 6. Required charging energy Fig. 7. Required charging energy
ing utility when As = 360◦ . Note that for simplicity, we & task duration vs. charging utility & task duration vs. charging utility
(centralized offline algorithm) (distributed online algorithm)
still use HASTE to denote our proposed centralized offline
algorithm or distributed online algorithm to HASTE in all 7.3.4 Impact of Color Number C
simulation figures if no confusion arises. This observation
Our simulation results show that on average the achieved charg-
is consistent with our intuition as the larger the charging
ing utility of HASTE steadily increases with color number C .
angle, the larger the chance that a charger can cover more
Figure 5 shows the box plot of the charging utilities of
charging tasks even with the same orientation, and all the
HASTE. It can be seen that the average charging utility
chargers cover the same set of tasks regardless of their
of HASTE increases by 3.29% when the color number C
orientations when As = 360◦ and thus make no difference
increases from 1 to 8. The maximum and minimum charging
in the performance for the three algorithms. Moreover, the
utilities of HASTE also smoothly increase with C . The
solution for HASTE with the color number C = 4 always
variance of charging utility for the eight colors is at most
outperforms that with C = 1, and has a performance gain
of 0.39% on average (at most 2.59%).
8.56 × 10−3 .
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 11
0.9
0.76 0.7 0.8
Charging Utility
Charging Utility
Charging Utility
Charging Utility
0.8
0.72 0.67 0.75
0.7
0.68 0.64 0.7
0.6
HASTE-DO(C=4) HASTE-DO(C=4) HASTE-DO(C=4)
0.64 HASTE-DO(C=1) HASTE-DO(C=1) 0.61 HASTE-DO(C=1) 0.65
0.5
GreedyUtility-DO GreedyUtility-DO GreedyUtility-DO
GreedyCover-DO GreedyCover-DO GreedyCover-DO
0.6 0.4 0.58 0.6
30 60 90 120 150 180 210 240 270 300 330 360 30 60 90 120 150 180 210 240 270 300 330 360 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8
As (°) Ao (°) ρ C
Fig. 8. As vs. charging utility (dis- Fig. 9. Ao vs. charging utility (dis- Fig. 10. ρ vs. charging utility (dis- Fig. 11. C vs. charging utility (dis-
tributed online algorithm) tributed online algorithm) tributed online algorithm) tributed online algorithm)
Besides, we can see that the charging utility for each of the 5
Our simulation results show that on average HASTE outperforms 7.4.5 Impact of Required Charging Energy and Task Dura-
GreedyUtility and GreedyCover by 6.83% and 8.95% (at most tion
8.68% and 10.96%), respectively, in terms of Ao . Figure 9
Our simulation results show that the achieved charging utility
illustrates that the charging utilities of the three algorithms
of HASTE steadily increases with a decreasing charging energy
monotonically increase with the receiving angle of devices
or an increasing task duration. Similar to the setting for the
Ao . The reason is that tasks with larger receiving angles can
centralized offline algorithm, we set the required charging
potentially be charged by more chargers, and thus receive
energy being randomly selected from [0.5Ej 1.5Ej ], and
more energy on average. Moreover, it is clear that the
task duration from [0.5∆t 1.5∆t]. Figure 7 shows that when
increasing trends of charging utilities for these algorithms
Ej downgrades from 50 kJ to 10 kJ and ∆t rises from
are first fast and then become slow as Ao increases from 30◦
30 min to 70 min, the achieved charging utility increases by
to 360◦ . Besides, HASTE with C = 4 outperforms HASTE
45.47%. The increasing speed for charging utility decreases
with C = 1 by 1.42% on average (at most 2.23%). Again,
when Ej increases or ∆t decreases, which demonstrates a
the charging utilities for the distributed online algorithms
marginal diminishing gain property.
are less than their corresponding centralized offline version.
7.4.6 Communication Cost
7.4.3 Impact of Switching Delay ρ
Our simulation results show that the number of messages and
Our simulation results show that on average HASTE outperforms
the number of rounds for a time slot increase quadratically and
GreedyUtility and GreedyCover by 5.20% and 7.3% (at most
linearly, respectively, with the number of chargers. We set the
5.20% and 7.31%), respectively, in terms of ρ. Figure 10 shows
number of color C to 1, and plot the average numbers of
that the charging utilities for all the algorithms steadily
messages and rounds in Algorithm 3 in Figure 12. We can
decrease with switching delay ρ. Especially, HASTE with
see that when the number of chargers increases from 10
C = 4 outperforms HASTE with C = 1 by 1.98%. When
to 100, the numbers of messages and rounds increase by
the switching delay is even up to one time slot, i.e., ρ = 1,
223.77% and 952.29%, respectively. The number of rounds
the achieved charging utilities for all the algorithms only
linearly increases because the number of neighboring charg-
slightly degrade compared with ρ = 0. This is because most
ers linearly increases. Further, as the number of messages
chargers keep still most of the time, and thus the caused
in each round also grows proportionally to the number of
performance loss is little.
neighboring chargers, it thus grows quadratically with the
7.4.4 Impact of Color Number C number of neighboring chargers, or the number of chargers.
Our simulation results show that on average the achieved charg- This finding supports Theorem 6.1.
ing utility of HASTE steadily increases with color number C .
Figure 11 demonstrates the box plot of the charging utilities 7.5 Insights
of HASTE when the color number C increases from 1 to 8. First, we investigate the impact of distribution of positions
We can see that both of the maximum and minimum charg- of charging tasks on the overall charging utility. Suppose
ing utilities of HASTE steadily increase with C . Moreover, there are 50 tasks distributed in a 50 m × 50 m area, and
on average the average charging utility of HASTE increases Ao = As = π/3. The required charging energy and charging
by 3.08% when the color number C increases by 1. Besides, duration for all tasks are randomly chosen from [5 kJ 20 kJ]
the variance of charging utility for all the eight colors is at and [10 min 120 min], respectively. The positions of tasks
most 8.42 × 10−3 , which indicates the stable performance of are randomly generated following a 2D Gaussian distribu-
our algorithm. tion with both x- and y - coordinates obeying a Gaussian
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 12
1
Charging Utility 0.35
Charging Utility
0.4 0.8
0.3
0.3
0.6
1
0.2
0.25
U∝ Ej o
0.2
0.4 87 [1 3]
o
0.1
4 114 [2 6]
348
o
[1 2]
7
0.2
0 0.15 12 8
25 o
39 [4 8]
20 o
25 342 [0 3]
15 20 0.1 0 o 20
σy 10 15 0 1 2 3 4 5 6 7 8 9 10 54 [2 5] o
5 5
10
σx Ej (J) ×104 2 298 [1 5]
0 0 3 9
o o 6
228 [1 5] 296 [3 5]
16
Fig. 13. Overall charging utility vs. Fig. 14. Individual charging utility o
5 20
o
[3 4] 4
o
[3 4]
o
0 [2 5] 25 [0 3]
Guassian distribution variance vs. required charging energy 15 18 138
o
[3 4]
1
151
o
[3 6]
2 10
14
17
o
305 [2 3]
o
112
o 130
o
[0 4]
11 245 [0 3]
[3 5]
3
o
13 4 20 [3 5]
1 19
0 Task
Charger
0 1 2 3 4 5
Charging Utility
0.6
some tasks are over-charged while the others are starved out Fig. 17. Charging utility of 20 tasks for the centralized offline algorithms
can be effectively avoided, and according to the concavity
of the charging utility function, the overall charging utility
will be enhanced. Second, we study the impact of Ej on
9 C ONCLUSION
the individual charging utility of each charger. Compared The key novelty of this paper is on proposing the first
with the above setting, we uniformly distribute 50 chargers scheduling algorithm for charging tasks in directional wire-
and 200 tasks. The required charging energy is a random less charging networks. The key contributions of this paper
number in [5 kJ 100 kJ]. Figure 14 shows that generally are proposing a centralized offline algorithm that achieves
the charging utility first can achieve 1 for a small Ej , and (1 − ρ)(1 − 1e ) approximation ratio where ρ denotes the
then rapidly decreases when Ej continues growing. The switching delay, and a distributed online algorithm that
maximum individual charging utility is approximately in- achieves 12 (1 − ρ)(1 − 1e ) competitive ratio, and conducting
versely proportional to Ej , as shown by the curve in Figure both simulations and field experiments for evaluation. The
14. The reason is that to achieve the same charging utility, key technical depth of this paper is in transforming the
a task with a higher required Ej needs a higher average problem into maximizing a submodular function subject to
charging power from its surrounding chargers, which is not a partition matroid constraint, bounding the performance
cost efficient. Thus, higher Ej leads to lower charging utility. loss caused by the switching delay and proving the approx-
imation ratio for the centralized offline algorithm, making
the centralized offline algorithm distributed and bounding
8 F IELD E XPERIMENTS the performance loss caused by the rescheduling delay and
We have conducted field experiments to evaluate our proving the competitive ratio for the distributed online algo-
scheme. First of all, we implemented our proposed schemes rithm. Our simulation and field experimental results show
on a small textbed. Due to space limit, we refer readers to that our proposed distributed online algorithm can achieve
the conference version of this paper [51] for details. 92.97% of the optimal charging utility, outperform the other
Next, we implemented our proposed schemes on a large two comparison algorithms, and its communication cost
testbed which consists of 16 TX91501 power transmitters moderately increases as the charger number scales up.
and 20 rechargeable sensor nodes. Figure 16 shows the
topology of this large testbed, which is much more irreg-
ular than the small testbed as it is randomly generated. ACKNOWLEDGEMENT
Similarly, Figures 17 and 18 show that HASTE respectively This work was supported in part by the National Natural
outperforms GreedyUtility and GreedyCover by 4.38% and Science Foundation of China under Grant 61872178, in part
10.12% on average, and by 13.27% and 23.60% at most in by the Natural Science Foundation of Jiangsu Province
the centralized offline scenario; and by 6.04% and 15.28% under Grant No. BK20181251, in part by Supported by the
on average, and by 22.58% and 29.63% at most for in the Fundamental Research Funds for the Central Universities
distributed online scenario. under Grant 14380062, in part by the open research fund
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13
1
[18] X. Xu et al., “Optimizing charging efficiency and maintaining sen-
0.8 sor network perpetually in mobile directional charging,” Sensors,
Charging Utility
vol. 19, no. 12, p. 2657, 2019.
0.6
[19] “www.powercastco.com,” 2018.
0.4 [20] “https://blog.ossia.com/news/ossia-and-t-mobile-pilot-wireless-
0.2 HASTE-DO(C=4) charging-asset-trackers-in-walmart-warehouses,” 2019.
GreedyUtility-DO [21] “https://wi-charge.com/applications,” 2019.
GreedyCover-DO
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 [22] C. Wang et al., “Multi-vehicle coordination for wireless energy
Task replenishment in sensor networks,” in IEEE IPDPS, 2013, pp.
1101–1111.
Fig. 18. Charging utility of 20 tasks for the distributed online algorithms [23] ——, “Recharging schedules for wireless sensor networks with ve-
hicle movement costs and capacity constraints,” in IEEE SECON,
2014, pp. 468–476.
of Key Lab of Broadband Wireless Communication and [24] W. Xu et al., “Towards perpetual sensor networks via deploying
multiple mobile wireless chargers,” in IEEE ICPP, 2014, pp. 80–89.
Sensor Network Technology (Nanjing University of Posts
[25] L. Chen et al., “Charge me if you can: Charging path optimization
and Telecommunications), Ministry of Education, in part by and scheduling in mobile networks,” in ACM MobiHoc, 2016, pp.
the National Natural Science Foundation of China under 101–110.
Grant 61802172, Grant 61672276, Grant 61832005, Grant [26] S. Nikoletseas et al., “Interactive wireless charging for energy
61472184, Grant 61872173, and Grant 61872082, in part balance,” in IEEE ICDCS, 2016, pp. 262–270.
[27] L. Fu et al., “Minimizing charging delay in wireless rechargeable
by the Key Research and Development Project of Jiangsu sensor networks,” in IEEE INFOCOM, 2013, pp. 2922–2930.
Province under Grant BE2015154 and Grant BE2016120, [28] L. He et al., “Esync: An energy synchronized charging protocol for
and the Collaborative Innovation Center of Novel Software rechargeable wireless sensor networks,” in ACM MobiHoc, 2014,
Technology and Industrialization, Nanjing University, and pp. 247–256.
[29] ——, “Mobile-to-mobile energy replenishment in mission-critical
in part by the Jiangsu Innovation and Entrepreneurship robotic sensor networks,” in IEEE INFOCOM, 2014, pp. 1195–1203.
(Shuangchuang) Program, the Guangdong Leading Talent [30] Y. Shi et al., “On renewable sensor networks with wireless energy
Program No. 2016LJ06D658. transfer,” in IEEE INFOCOM, 2011, pp. 1350–1358.
[31] L. Xie et al., “On renewable sensor networks with wireless energy
transfer: the multi-node case,” in IEEE SECON, 2012, pp. 10–18.
[32] S. Guo et al., “Mobile data gathering with wireless energy replen-
R EFERENCES ishment in rechargeable sensor networks,” in IEEE INFOCOM,
2013, pp. 1932–1940.
[1] A. P. Sample et al., “Design of an rfid-based battery-free pro- [33] L. Xie et al., “Bundling mobile base station and wireless energy
grammable sensing platform,” IEEE Transactions on Instrumentation transfer: Modeling and optimization,” in IEEE INFOCOM, 2013,
and Measurement, vol. 57, no. 11, pp. 2608–2615, 2008. pp. 1636–1644.
[2] L. Xie et al., “Wireless power transfer and applications to sensor [34] ——, “On traveling path and related problems for a mobile station
networks,” IEEE Wireless Communications, vol. 20, no. 4, pp. 140– in a rechargeable sensor network,” in ACM MobiHoc, 2013, pp.
145, 2013. 109–118.
[3] S. Li et al., “Wireless power transfer for electric vehicle appli- [35] C. Wang et al., “Improve charging capability for wireless recharge-
cations,” IEEE Journal of Emerging and Selected Topics in Power able sensor networks using resonant repeaters,” in IEEE ICDCS,
Electronics, vol. 3, no. 1, pp. 4–17, 2015. 2015, pp. 133–142.
[4] P. E. Glaser, “An overview of the solar power satellite option,” [36] ——, “A hybrid framework combining solar energy harvesting
IEEE Transactions on Microwave Theory and Techniques, vol. 40, no. 6, and wireless charging for wireless sensor networks,” in IEEE
pp. 1230–1238, Jun 1992. INFOCOM, 2016, pp. 1–9.
[5] S. F. Bush, Smart grid: Communication-enabled intelligence for the [37] Z. Dong et al., “Energy synchronized task assignment in recharge-
electric power grid. John wiley & sons, 2014. able sensor networks,” in IEEE SECON, 2016, pp. 1–9.
[6] “https://www.wirelesspowerconsortium.com/,” 2018. [38] S. He et al., “Energy provisioning in wireless rechargeable sensor
[7] “https://www.wirelesspowerconsortium.com/blog/273/wireless- networks,” IEEE Transactions on Mobile Computing, vol. 12, no. 10,
power-market-surges-as-usage-leaps-forward,” 2018. pp. 1931–1942, 2013.
[8] Z. Wang et al., “Adaptively directional wireless power transfer [39] H. Dai et al., “Wireless charger placement for directional charg-
for large-scale sensor networks,” IEEE Journal on Selected Areas in ing,” IEEE/ACM Transactions on Networking, vol. 26, no. 4, pp.
Communications, vol. 34, no. 5, pp. 1785–1800, 2016. 1865–1878, 2018.
[9] L. Wang et al., “Millimeter wave power transfer and information
[40] N. Yu et al., “Placement of connected wireless chargers,” in IEEE
transmission,” IEEE Globecom, pp. 1–6, 2015.
INFOCOM, 2018. To appear.
[10] N. Deng and M. Haenggi, “A novel approximate antenna pattern
[41] X. Wang et al., “Heterogeneous wireless charger placement with
for directional antenna arrays,” IEEE Wireless Communications Let-
obstacles,” in ACM ICPP, 2018, p. 16.
ters, vol. 7, no. 5, pp. 832–835, 2018.
[11] T. A. Khan and R. W. Heath, “Wireless power transfer in millimeter [42] H. Dai et al., “Safe charging for wireless power transfer,” in IEEE
wave tactical networks,” IEEE Signal Processing Letters, vol. 24, INFOCOM, 2014, pp. 1105–1113.
no. 9, pp. 1284–1287, 2017. [43] ——, “Safe charging for wireless power transfer,” IEEE/ACM
[12] T. A. Khan and R. W. Heath Jr, “Wireless power transfer in Transactions on Networking, vol. 25, no. 6, pp. 3531–3544, 2017.
millimeter wave,” Wireless Information and Power Transfer: Theory [44] ——, “SCAPE: Safe charging with adjustable power,” in IEEE
and Practice, pp. 139–156, 2018. ICDCS, 2014, pp. 439–448.
[13] “http://www.powercastco.com/pdf/p2110-evb-reva.pdf,” 2018. [45] ——, “SCAPE: Safe Charging With Adjustable Power,” IEEE/ACM
[14] O. Georgiou, “Simultaneous wireless information and power Transactions on Networking, vol. 26, no. 1, pp. 520–533, Feb 2018.
transfer in cellular networks with directional antennas,” IEEE [46] S. Nikoletseas et al., “Low radiation efficient wireless energy
Communications Letters, vol. 21, no. 4, pp. 885–888, 2017. transfer in wireless distributed systems,” in IEEE ICDCS, 2015,
[15] L. Fan et al., “Secure wireless information and power transfer pp. 196–204.
based on tilt adaptation in 3-d massive mimo systems,” IEEE [47] H. Dai et al., “Radiation constrained wireless charger placement,”
Access, vol. 7, pp. 5531–5540, 2019. in IEEE INFOCOM, 2016, pp. 1–9.
[16] H. Dai et al., “Omnidirectional chargability with directional anten- [48] ——, “Robustly safe charging for wireless power transfer,” in IEEE
nas,” in IEEE ICNP, 2016, pp. 1–10. INFOCOM, 2018. To appear.
[17] ——, “Optimizing wireless charger placement for directional [49] ——, “Radiation constrained scheduling of wireless charging
charging,” in IEEE INFOCOM, 2017, pp. 2922–2930. tasks,” in ACM MobiHoc, 2017, pp. 17–26.
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TMC.2020.2997602, IEEE
Transactions on Mobile Computing
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14
[50] ——, “Radiation constrained scheduling of wireless charging Ke Sun received his B.S. degree in Computer
tasks,” IEEE/ACM Transactions on Networking, vol. 26, no. 1, pp. Science from Nanjing University of Aeronautics
314–327, Feb 2018. and Astronautics, Jiangsu, China, in 2016. He is
[51] ——, “Charging task scheduling for directional wireless charger currently a Master student in Nanjing University.
networks,” in ICPP, 2018. To appear. His research interests are in the area of of mo-
[52] G. L. Nemhauser et al., “An analysis of approximations for max- bile computing.
imizing submodular set functions-I,” Mathematical Programming,
vol. 14, no. 1, pp. 265–294, 1978.
[53] G. Calinescu et al., “Maximizing a monotone submodular function
subject to a matroid constraint,” SIAM J. Comput., vol. 40, no. 6,
pp. 1740–1766, 2011.
[54] M. Streeter et al., “Online learning of assignments,” in Advances in
Neural Information Processing Systems 22, Y. Bengio, D. Schuurmans,
J. D. Lafferty, C. K. I. Williams, and A. Culotta, Eds. Curran Alex X. Liu received his Ph.D. degree in Com-
Associates, Inc., 2009, pp. 1794–1802. [Online]. Available: http:// puter Science from The University of Texas at
papers.nips.cc/paper/3719-online-learning-of-assignments.pdf Austin in 2006, and is a professor at the De-
[55] D. Golovin et al., “Online submodular maximization under a ma- partment of Computer Science and Engineer-
troid constraint with application to learning assignments,” arXiv ing, Michigan State University. He received the
preprint arXiv:1407.1082, 2014. IEEE & IFIP William C. Carter Award in 2004, a
[56] X. Lu et al., “Wireless charging technologies: Fundamentals, stan- National Science Foundation CAREER award in
dards, and network applications,” IEEE Communications Surveys & 2009, and the Michigan State University Withrow
Tutorials, vol. 18, no. 2, pp. 1413–1452, 2016. Distinguished Scholar Award in 2011. He has
[57] C. Lin et al., “Minimizing charging delay for directional charging served as an Editor for IEEE/ACM Transactions
in wireless rechargeable sensor networks,” in IEEE INFOCOM, on Networking, and he is currently an Associate
2019, pp. 1819–1827. Editor for IEEE Transactions on Dependable and Secure Computing
[58] “http://www.sevenoak.biz/pantiltheads/sk-ebh01.html,” 2018. and IEEE Transactions on Mobile Computing, and an Area Editor for
[59] R. Zhang et al., “MIMO broadcasting for simultaneous wireless Computer Communications. He received Best Paper Awards from ICNP-
information and power transfer,” IEEE Transactions on Wireless 2012, SRDS-2012, and LISA-2010. His research interests focus on
Communications, vol. 12, no. 5, pp. 1989–2001, 2013. networking and security.
[60] Z. Ding et al., “Application of smart antenna technologies in
simultaneous wireless information and power transfer,” IEEE
Communications Magazine, vol. 53, no. 4, pp. 86–93, 2015.
[61] L. Fleischer et al., “Tight approximation algorithms for maximum Lijun Zhang received the B.S. and Ph.D. de-
general assignment problems,” in ACM SODA, 2006, pp. 611–620. grees in Software Engineering and Computer
[62] S. Fujishige, Submodular functions and optimization. Elsevier, 2005, Science from Zhejiang University, China, in 2007
vol. 58. and 2012, respectively. He is currently an as-
[63] H. Dai et al., “Practical scheduling for stochastic event capture in sociate professor of the Department of Com-
energy harvesting sensor networks,” International Journal of Sensor puter Science and Technology, Nanjing Univer-
Networks, vol. 18, no. 1-2, pp. 85–100, 2015. sity, China. Prior to joining Nanjing University,
[64] “https://en.wikipedia.org/wiki/online algorithm,” 2019. he was a postdoctoral researcher at the De-
partment of Computer Science and Engineering,
Michigan State University, USA. His research
interests include machine learning, optimization,
information retrieval and data mining.
1536-1233 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 18,2020 at 15:33:20 UTC from IEEE Xplore. Restrictions apply.