1. Introduction
Gathering sensory data in a minimal amount of time has been a critical concern for years. Examples of time-sensitive applications are smart monitoring and object tracking, in which sensor devices can be deployed over a large area and the applications need to acquire data in a short time [
1]. In reality, while sensors are typically battery-powered [
2], they must serve for years. One of the most prominent methods used to prolong the sensor lifetime is duty-cycling [
3]. Nevertheless, it lengthens the data-gathering time and degrades the application service quality. In this regard, UAVs potentially help, and research studies on UAVs are receiving increasing interest from both academia and industry. A UAV can serve as a mobile station, a relay node, or an edge [
4]. Particularly, for data collection applications, UAV is helpful in various ways. For example, it prolongs the sensor network lifetime and increases channel quality by providing direct short-range communication with sensor devices [
5]. It can also fly over an unattended/disconnected area to gather data, which lowers the deployment costs.
This paper investigates the minimum-latency data aggregation scheduling problem in duty-cycled wireless sensor networks with UAVs. Herein, a UAV is responsible for collecting data from the ground sensors. Because the sensor data are highly correlated, data aggregation techniques are often employed to reduce time, energy consumption, and redundancy. The network splits into several clusters, each of which is managed by a cluster head. Sensor nodes in a cluster send their data to the cluster head (CH), and the UAV only visits these CHs to acquire the aggregated data. Several prior works presented different optimization perspectives for the UAV-enabled data collection problem. authors of [
6,
7] arranged multiple UAVs to minimize the energy consumption of sensor devices. In [
8], the authors emphasized that the age of information should be minimal, but the scheme only facilitated a small number of sensors. Aiming at a similar goal, in [
9], the authors assumed that data become obsolete after a certain deadline. A UAV is intended to gather data from as many devices as possible. Quite a few other researches aimed to minimize the UAV-related metrics, e.g., UAV flight time and UAV energy consumption. These works either require UAV to approach every device [
6], or divide the network into small-size clusters with direct communication between CH and cluster members (CMs) [
10]. This would result in an excessive number of clusters if the network size increases, making the UAV path planning and scheduling more challenging. Furthermore, only optimizing the UAV itinerary might not be efficient, because the data collected at each cluster head can become obsolete after a certain waiting time. To the best of our knowledge, there has been no research on delay-efficient transmission scheduling for sensor networks in conjunction with UAV itinerary scheduling.
Herein, we propose an adaptive clustering and scheduling (ACS) approach that computes the device transmission time tightly based on the UAV itinerary. Each CH is given a deadline to aggregate data from its surroundings. Because the UAV visits a sequence of CHs, the -th CH has more time to gather data than the k-th one. As such, later-visited clusters have larger deadlines and should contain more CMs. Inside each cluster, device transmissions are pipelined based on their duty cycle to maximize the number of devices it can accommodate within the deadline. Our contributions are briefly stated as follows. Note that, the terms ‘sensor’ and ‘device’ are used interchangeably in this paper.
We formulate a joint UAV-sensor device scheduling problem and prove its NP-hardness. Then a simpler problem is derived in which the UAV can fly continuously without hovering at each CH.
The proposed scheduling approach precisely computes the transmission schedule among devices and to the UAV. By calculating the times at which the UAV should approach the CHs, we infer the deadline for each CH to acquire data from the CMs.
From the computed deadlines, we design a collision-free reverse cluster formation and scheduling scheme. Herein, the scheme simultaneously constructs a data aggregation tree in each cluster and assigns transmission times to devices. Note that the last CH with the largest deadline will start to construct the aggregation tree and acquire CMs first, followed by the CH with the second largest deadline, and so forth.
The remainder of this paper is organized as follows.
Section 2 reviews recent work related to the problem we are solving. In
Section 3, we describe network models and assumptions, followed by the problem formulation in
Section 4. The proposed ACS scheme is presented in
Section 5. We evaluate the proposed design in
Section 6, summarize the contributions of this work, and discuss future directions in
Section 7.
2. Related Work
For decades, a mobile sink (or mule) has been used as a solution to collect data from ground sensors [
11]. A mobile sink is useful in situations where there are no dedicated and powerful nodes to collect data from the surroundings, or when the network is disconnected. However, the ground mobile sink is affected by complex terrains and obstacles; therefore, they lack flexibility and speed [
11]. Consequently, it might take a lot of time to collect data from a network. This is problematic since, nowadays, data freshness is also a critical concern. UAV seems to be a viable replacement for the ground mobile sink. Existing research aimed to shorten the UAV itineraries, since the UAV resources are usually limited. For instance, in [
12,
13], the authors assumed that the UAV collects individual sensor data following a fixed route. Herein, not all sensors send data to the UAV. In [
14], the authors took into account the altitude, trajectory, velocity, and wireless link when minimizing data collection time. The UAV shall hover at a maximum altitude so as to maximize the transmission range, thereby shortening the trajectory. Concerning the age of information, the authors of [
8] also targeted a minimal UAV trajectory and proposed solutions using dynamic programming and genetic algorithms. With the same objective, the authors of [
15] modeled the data collection problem as a Markov decision process, in which they used deep reinforcement learning to find an optimal UAV trajectory. In [
16], the authors solved the trade-off between the energy consumption of the UAVs and the number of UAVs required to cover an area. Herein, fewer UAVs mean that each UAV has to cover a larger area. As a consequence, the UAVs have to hover at a higher altitude for a longer period, which drains more energy. In general, prior works, that focused on UAV performance involved determining the traveling order of CHs, differing in how the constraints were considered. Several other studies in this area, such as [
17,
18,
19], have also contributed to this direction.
UAVs can help improve the energy efficiency of WSNs (wireless sensor networks), because traditional data collection solutions for WSNs are used to create energy holes around the sink due to the high traffic going through these nodes [
11]. UAVs, instead, can freely move to any location and alleviate the energy hole problem. For example, the authors of [
20] let the UAV move close to the target sensors, so sensor devices did not need to transmit at high power. In [
21], the authors optimized the transmitting power and transmission reliability of Internet of Things (IoT) devices by clustering and using multiple UAVs, one per cluster. In [
10,
22], the network is split into clusters of equal sizes to balance the energy consumption of sensors and prolong the network lifetime. In [
13], the scheme allows the prioritization of the region of interest and ignores sensors from low-priority regions. In [
6], the authors presented a successive convex optimization technique to achieve a sub-optimal solution. Herein, not only is reliable communication with sensors guaranteed, but the energy consumption of all sensors is also minimized. Lee et al. proposed a LEEF algorithm, which utilizes
k mobile relays to cover
k clustered regions. LEEF strives to equalize the energy consumption of mobile relays [
23].
To the best of our knowledge, although there were many efforts conducted on data collection with UAV, all the existing schemes only focus on the performance metrics of the UAV without considering the aggregation schedule inside each cluster. Moreover, the clusters usually have equal size and are organized in a star topology, which requires a long time to collect data as the CMs have to take turns sending to the CH. Finally, the geographical coverage of a CH is limited by its communication range; therefore, to maintain the direct transmissions between CMs and CH, the number of clusters increases when the deployment area scales up. As a result, the UAV itinerary becomes longer and more inefficient. Our proposed scheme, ACS, arranges to aggregate data at the CHs right before the arrival of the UAV. In addition, the UAV speed and CH visiting order will determine the cluster size. We utilize the sleep–awake pattern of sensors to pipeline transmissions through multi-hop routes such that each CH can accommodate a larger number of CMs in a short aggregation time.
3. Network Model
We model the network as a graph
, in which
V is the set of homogeneous IoT devices (nodes) and
E is a set of communication links between the devices. Herein, there exists a link between two nodes
if and only if the Euclidean distance between them is smaller than the communication range
R. The whole network shares a single communication channel. Two neighbors can communicate with each other, and if a node transmits, all of its neighbors can hear the signal. This leads to collisions if two or more signals simultaneously arrive at the same receiver. The node then cannot receive any of the signals. We use the protocol interference model, which consists of primary and secondary collisions [
3].
Sensor devices are temporally synchronized and use TDMA (time division multiple access) to access the channel. Herein, time is divided into slots and each sensor device sends the data at a specific slot. There will be no collision avoidance mechanism required, such as CSMA (carrier sensing multiple access). Instead, it requires a transmission schedule for the whole network, so that each device can send data without collision. The communication transceivers operate in half-duplex mode, i.e., at a time, they can only either send or receive data. We assume that all devices run in duty cycle mode, in which a device only wakes up for a short time to receive or transmit data, and for the rest of the time, it stays in a sleeping state to conserve energy. Herein, time consists of equal working periods, and each working period is further split into constant
T slots with indices in
. Each device randomly chooses an active slot
. The device periodically wakes up at its active slot to wait for possible incoming transmission and sleeps otherwise. If a device wants to send data to its neighbor, it must wake up and transmit data at the active slot of the receiver.
Figure 1 illustrates a network containing three nodes
, in which node
c is active at slot 0 and
. Nodes
a and
b must send data to node
c in two different working periods, at the active slot 0 of node
c. A time slot is sufficient to transmit a packet.
Sensor data sent by the devices can be aggregated perfectly [
3] so that each node only sends data once. This means that, regardless of the number of received data packets, the intermediate node will aggregate the received data with its data, to form a single packet toward the upstream device. We also assume that the UAV is navigated and positioned so that its wireless links to CHs are sufficiently reliable to guarantee transmission success.
4. Problem Formulation
Table 1 lists the acronyms and notations we use in this paper.
The UAV is assumed to rest at a docking station and remains charged if it is not on duty. On a trip to gather data, it visits a set of CHs, one after another. In each cluster, CH is responsible for aggregating data from other CMs, through multi-hop paths. An intermediate node combines all received data and its own data, then outputs one outgoing transmission. Examples of this aggregation function are MIN, MAX, and AVERAGE. For devices, an aggregation schedule must specify the time slot and the device that sends data to which receiver. In addition, the schedule must incorporate all devices. A valid schedule should induce no collision between transmissions occurring at the same time. For UAVs, the schedule should indicate the departure time, given the known trajectory and speed. Within each cluster, the routing structure forms a tree topology rooted at the CH. Hereafter, we use the terms sender/receiver and child/parent interchangeably.
The aggregation time, or aggregation delay, is the time elapsed from when the UAV departs from the docking station, until when it leaves the last CH in its itinerary after successfully retrieving data. Let
k be the number of clusters and
is the set of CHs, which are known in advance. For example, several nodes with more resources can be deployed in the field as CHs. The CH locations may depend on the terrain. Finding an optimal number
k as well as the optimal positions of the CHs are postponed to future research. Because the whole network is partitioned into
k clusters, we have:
Remark 1. Given the locations of the CHs, we can compute the shortest flying path for the UAV. For simplicity, we assume that the set is ordered and represents the shortest trajectory. Nevertheless, the size and members of each cluster depend on the scheduling scheme.
Let
be the traveling time from CH
to CH
and
,
be the waiting time at the
(
Figure 2). Both
and
are measured in the number of time slots. Let
be the speed of the UAV (m/ time slot), we have the following:
The hovering time
if and only if CH
has yet to finish data collection upon the UAV arrival. We can compute the UAV arrival time
at CH
as follows:
The departure time
of the UAV at CH
, i.e., the data aggregation time at the
i-th cluster, is as follows:
Note that are constant values, and is the aggregation time of the whole network. To minimize , one needs to minimize the total waiting time . Herein, the total waiting time comprises the hovering time and the waiting time at the base station. The selection of must ensure that all devices participate in the data aggregation.
Remark 2. is the aggregation time of the whole network.
Let
be an arbitrary valid transmission schedule, in which
is the set of devices transmitting data at time slot
i:
. The following conditions must hold:
and
Herein, (
5) means that all the devices are assigned transmission time slots, and (
6) ensures that each node only sends data once to its receiver. In addition, all the transmissions occurring at the same time must be collision-free,
:
Within a cluster, the transmission time of each node must be smaller than the aggregation time:
Furthermore, since a parent node may receive data from several children, the transmission time slot of a child must be smaller than that of its parent.
Remark 3. Due to the aggregation constraint, the transmission time slot of a parent must be later than that of its children, i.e., An aggregation schedule can be identified by the routing structure, i.e.,
, which points out a receiver for each device, and the transmission time, i.e.,
, which enforces when the devices should send data. We can formally formulate the optimization problem as follows.
Theorem 1. The problem (10) is NP-hard. Proof. Let us consider a reduced problem of (
10) when
. This problem becomes the conventional minimum-time aggregation scheduling problem in wireless sensor networks, and is NP-hard [
24]. Therefore, problem (
10) is NP-hard. □
5. Adaptive Clustering and Scheduling Scheme
This section presents a heuristic scheme to solve problem (
10). Recall that there are two types of scheduling for WSNs. We summarize the main features as follows.
Bottom-up scheduling. This type requires to construction of an aggregation tree rooted at the CH in advance. Next, the scheduling algorithm identifies the transmission time slot for each tree edge, from the leaves up to the root. The scheduling algorithm iteratively finds transmissions for time slots .
Top-down scheduling. This type does not require having a tree beforehand. The tree is grown during the scheduling process. Unlike the bottom-up approach, we can provide a deadline to collect data at the root node. Depending on the given deadline, a part or the whole network can send data to the root [
25]. The scheduling algorithm iteratively finds transmissions for time slots
, in the reverse order of the bottom-up scheduling approach.
Regarding the bottom-up approach, the process begins with (step 1) splitting the network into clusters, followed by (step 2) constructing the aggregation tree in each cluster, and then (step 3) performing a scheduling algorithm. However, how to optimally cluster the network is hard, as it is impossible to know the aggregation time in each cluster without running step 2 and step 3. This may require numerous trials and errors to achieve a sufficient solution.
On the other hand, top-down scheduling allows for the simultaneous construction of the aggregation tree and scheduling, and the tree size (and, hence, the cluster size) can be adjusted together with the given aggregation deadline.
Figure 3a illustrates a case where three CHs are present, and the UAV leaves the CHs at
, and
. In
Figure 3b, the clusters may have different sizes because the aggregation deadline for each is different. While all the clusters consist of members being scheduled until
, only
and
continue to aggregate data during
, and then only
contains transmissions scheduled during
. Generally, when the UAV flies to the
, the sensor devices can transmit data toward any CH. After the
delivers data to the UAV, the remaining CHs, i.e.,
, continue to gather data. Since the UAV may arrive at the
i-th CH earlier than the time
finishes collecting data from its cluster, it may have to wait for the duration
before departure. Based on
, the data aggregation deadline at the
i-th cluster can be derived from (
4) as follows:
The following theorem proves that we can always simplify problem (
10) by adding a constraint on the waiting time at each CH (or hovering time) as follows:
Theorem 2. For any valid aggregation schedule with , there always exists a schedule with such that and .
Proof. We can identify
as follows:
From (
12) and (
4), we have
because
. Since each CH
can aggregate data from its cluster within
time slots, it can also aggregate data within
time slots. The theorem is proved. □
Using Theorem 2, we can simplify the problem in (
10) to the following problem: Find min
and a scheduling algorithm
such that when giving the deadlines
to the CHs
, all devices are scheduled to send data to the CHs. The scheduling algorithm
must satisfy Equations (5)–(9). Herein,
is calculated by the following:
or, equivalently:
which is simpler than (
4), as the variables
are eliminated.
In reality, the UAV might not be able to fly continuously, while ensuring reliable communication with the selected CHs. However, we can consider as the average speed of the UAV, taking into account the hovering time and acceleration/deceleration time.
5.1. Overall Approach
Equations (
13) and (
14) suggest that the aggregation deadlines at all the clusters can be derived from the aggregation deadline at the last CH
, or from the waiting time
at the docking station
as the hovering times at all the CHs are zero (
). In other words, once either
or
is selected, then
is deterministic. Given a scheduling algorithm
, Algorithm 1 shows a naive method to determine the minimum value of
. The first
is returned if the derived deadlines
given to the CHs are sufficient to cover the entire network. However, this approach requires repeating the same scheduling procedure (lines 3–4) multiple times, making it costly.
Algorithm 1 Naive method to find . |
Input: set of CHs
Output: Minimum - 1:
while True do - 2:
Compute as in ( 14) - 3:
if already covered all nodes within then - 4:
break - 5:
else - 6:
- 7:
end if - 8:
end while
|
Instead, this paper proposes a novel adaptive top-down data aggregation scheduling approach, namely ACS, which grows the aggregation trees from the roots (i.e., the CHs), and at the same time computes the transmission schedule. Herein, ACS starts at the last transmission time slot, i.e.,
, and goes backward to find the senders in each time slot. Assuming that
denotes the reverse scheduling algorithm, Algorithm 2 illustrates an abstract procedure of ACS.
Algorithm 2 One-shot reverse method to find . |
Input: set of CHs Output: The aggregation time.
- 1:
Assign some big value to - 2:
Compute as in ( 13) - 3:
Run the scheduling algorithm - 4:
- 5:
for
do - 6:
- 7:
end for - 8:
return
|
At first, ACS initializes a sufficiently large deadline
for the last cluster (line 1), and computes other deadlines for the remaining CHs (line 2). This is to ensure that after running the scheduling algorithm
(line 3), all the transmission time slots of devices are positive numbers. Next, we take
as the smallest time slot used to schedule (line 4). Because the transmissions are scheduled in a reverse direction, there will be no transmission scheduled for time slots
(
Figure 4). As a result, we can shift all devices’ assigned transmission time slots backward by
time slots, i.e., by
working periods to the left (lines 5–6). The aggregation delay is, thus,
.
Figure 4 shows why we can shift the schedule be a maximum of
working periods to the left. Because the algorithm computes the schedule in reverse order, all the transmissions are assigned time slots in the range [
tmin,
]. However, it is not possible to shift the schedule by
tmin time slots because the new schedule may break the duty cycle settings of the receiver. For example, if
tmin, and a transmission from
u to
v is scheduled at
tmin. It means that
=
tmin. If the algorithm shifts
tmin time slots, then
u will be scheduled to send data to
v at time slot 0, which is not at the active slot of
v. The maximum number of slots that we can shift is, therefore,
.
Remark 4. After initializing with some large deadline and finishing with the earliest transmission at tmin > 0, the transmission schedule is shifted backward by working periods. The aggregation delay is then time slots.
In the next subsection, we will provide a detailed design of the scheduling algorithm .
5.2. Adaptive Clustering and Scheduling
The proposed ACS scheme is presented in Algorithm 3. For the ease of scheduling implementation, we assign transmission time slot to CH for all (lines 1–2). This can be seen as the time transmits data to the UAV. Typically, the algorithm iterates the backward time slot by the time slot, starting from the deadline at the last CH . Herein, the set of candidate receivers P is determined first, followed by finding the senders for them. The set of scheduled devices is denoted by and is initially empty (line 4). ACS stops when all the devices are scheduled, i.e., . Lines 5–15 present the process in which ACS searches for eligible transmitting nodes at time slot t. Herein, P is the set of parent candidates at the current time slot t, which is initially empty. A parent candidate (i.e., u can be a receiver) at time slot t is any node that satisfies the conditions below:
(line 8): candidate receiver u is already scheduled. By checking this condition, ACS makes sure that a CH only starts acquiring CMs to grow its aggregation tree at a proper time. For example, when Algorithm 3 starts at , only grows its aggregation tree. When the time t decreases down to , then begins constructing its own aggregation tree, and so on.
(line 9): candidate receiver u is active at the current time slot t; otherwise, it cannot receive data.
(line 9): candidate receiver u has at least one unscheduled neighbor. Otherwise, no node can be scheduled to send data to u.
Algorithm 3 ACS scheduling scheme. |
Input: , set of CHs , aggregation delay det at each cluster Output: The transmission schedule - 1:
for each do - 2:
- 3:
end for - 4:
- 5:
- 6:
while
do - 7:
- 8:
for do - 9:
if and then - 10:
- 11:
end if - 12:
end for - 13:
- 14:
Find a collision-free matching between P and C - 15:
set of senders in the matched pairs - 16:
- 17:
- 18:
- 19:
end while
|
These nodes in P are the receiver candidates. The set of candidate senders C consists of nodes that are neighbors of at least one node in P and are not scheduled (line 13). In order to identify the sender–receiver pairs that can transmit data simultaneously without collision, ACS performs matching between P and C. The detailed matching algorithm is presented in Algorithm 4. After matching, selected senders will be assigned the transmission time t (line 16), and are added to the set. Then, the algorithm proceeds to the next time step: (line 18).
Finding a matching between the sender set
C and the receiver set
P can be performed in various ways, depending on how we prioritize the nodes. For example, in our previous work [
26], we used the node degree to sort the set
P. This work nevertheless targets the low duty-cycled WSNs, for which we utilize two well-known metrics, link delay and number of neighbors, as follows.
Definition 1. Link delay from a sender u to a receiver v is as follows: Definition 2. The neighbors in a set H of nodes u are as follows: Herein, node
u, after finishing data gathering from
, must wait for a certain time until its intended receiver
v becomes active. The link delay basically means the minimum of such a waiting time.
Algorithm 4 Sender–receiver matching. |
Input: Candidate receivers P, candidate senders C
Output: The set of scheduled nodes - 1:
Sort P in a non-decreasing number of the number of its neighbors in C - 2:
- 3:
while
and
do - 4:
first node in P - 5:
- 6:
- 7:
{Due to collision with the transmission from u} - 8:
- 9:
- 10:
end while - 11:
return
|
Algorithm 4 seeks a collision-free set of transmissions that can happen at the current time slot t. The receivers are in P and senders are in C. Items in P are sorted such that any preceding item has no more neighbors in C than its succeeding ones, i.e., a node with fewer neighbors in C has a higher priority (line 1). After that, the algorithm picks up a node in C with the smallest link delay to v as the sender (lines 4–5). Next, the algorithm removes all the receiver candidates that overhear the transmission from the selected sender u (line 6) from set P and removes all sender candidates whose transmissions can cause a collision at the selected receiver v (line 7). The while loop continues until there is no node left in either P or C.
5.3. Complexity Analysis
In this section, we cover the time complexity analysis of the ACS scheme. We omit the analysis of Algorithm 1, as this algorithm is not used. Below, we will examine the algorithm complexity of Algorithms 2–4.
Algorithm 4 first sorts the candidate receiver set P, which can be conducted in time. Line 3 loops times. In the loop, line 5, line 7, and line 8 each take no more than . Therefore, the time complexity of Algorithm 4 is .
In Algorithm 3, lines 1–5 require no more than to complete. Line 6 loops, at most, times. In each loop, the algorithm checks each node in the network vertex set (line 8); for each node u, it will examine the active slot and the aggregation status of the neighbors of node u (line 9). Checking the aggregation status of node u depends on the network density, i.e., the number of neighbors. In the worst case, this operation could require at most . Then, lines require . Line 14 is Algorithm 4, which requires, at most, . Hence, the time complexity of Algorithm 3 is . It is trivial to see that the main computation part of Algorithm 2 is line 3, which is Algorithm 3. Therefore, the time complexity of the ACS scheme is .
5.4. Step by Step Example
Figure 5 illustrates a complete run of the ACS on a network topology shown in
Figure 5a. Nodes
and
are two CHs, and the algorithm starts with
and
(time slots). A working period contains
slots.
Although the run starts at
, no node is active at slot 0; hence, the algorithm proceeds to time slot
. The set of candidate receivers is
, and the set of candidate senders is
. Two candidate senders,
C and
H, have the smallest link delay to
A, i.e.,
. The algorithm randomly chooses one among the two. Here, we assume that node
C is selected; hence,
(
Figure 5b).
Next, from time slot
, down to time slot
, there is no active candidate receiver. Note that, although node
C is active at time slot
, it does not have any neighbor to be a candidate sender. Then, at
, node
A becomes active again, and the set of candidate senders is
. Now
H is selected because it has the smallest link delay to
A. The transmission time of
H to
A is
(
Figure 5c). Similarly, when
, ACS schedules node
G to transmit data to
H, so
(
Figure 5d).
In the next time step,
, because
and node
B is also active at
,
B is eligible to acquire a sender. The only unscheduled neighbor of
B, i.e., node
F, is scheduled to send to
B at
(
Figure 5e). The algorithm continues with time slot
and
(
Figure 5f,g). The final aggregation schedule is obtained in
Figure 5h, by subtracting the transmission times of all devices’
working periods, i.e., 10 time slots.
7. Conclusions
This paper proposes ACS, an adaptive clustering and scheduling scheme that efficiently arranges the device transmissions and UAV flying schedule. In particular, along the UAV flying trajectory, later-visited CHs should have larger allocated data aggregation deadlines. Upon calculating all deadlines at the CHs, ACS performs a reverse scheduling algorithm to iteratively grow multiple aggregation trees rooted at the CHs. The transmission schedule and sender–receiver selections are pipelined according to the device’s active time slots.
Simulation results confirm the effectiveness of ACS, as it reduces aggregation time by up to compared to other benchmarking schemes. ACS shows a significant improvement over ICS when the working period length is average, i.e., in this case, . The performances of ACS and ICS tend to converge when T is too small or too large. In the future, we intend to optimize the number of CHs and their placement to achieve shorter aggregation times. We also plan to evaluate the impact of various factors, such as fading and the signal-to-interference and noise ratio, on performance. Finally, more scenarios will be investigated, including disconnected regions, larger and denser networks, and deadline-constrained aggregation tasks.