One of the paramount concerns in Wireless Sensor Networks (WSNs) is enabling sensors to discern the direction of data transmissiona concept referred to as global direction guidance for data transmission. Before initiating any data transmission, the entire network must establish a transmission guidance framework through a routing scheme. This framework imparts a global direction implicitly to each sensor node, directing it towards the sink node. Generally, three primary routing schemes are employed to establish this global transmission guidance: geographic location-based schemes [33–35], transmission cost-based schemes [36], and hop count-based schemes.

In large-scale WSNs, data transmission, and recovery protocols frequently use methods like retransmission and error correction to overcome abnormalities. Such solutions can, however, result in energy inefficiencies and latency. Furthermore, they might not take into account the large-scale dynamic nature of the WSNs in a comprehensive way, which could limit their ability to grow, adapt, and distribute data efficiently, especially in real-world medical circumstances. Geographic location-based schemes typically require GPS equipment, while transmission cost-based schemes necessitate additional hardware or modules to assess costs, such as energy consumption.

Data aggregation stands out as one of the most efficient methods within WSNs for alleviating network communication overhead. In our scenario, as soon as sensor data is generated by cluster members, it must be transmitted to the cluster head. However, this form of multiple-to-one communication often leads to significant collisions, causing increased packet loss and undermining the efficacy of data collection. Addressing this challenge is a crucial issue, yet practical research in this area remains relatively limited, highlighting the importance of collision control within a cluster. Early Message Ahead Join Adaptive Data Aggregation (E-ADA) [37] scheme for efficient data collection in IoT employs an early notification to enhance data aggregation probability and optimize energy consumption and introduces a delay-optimized convergence Routing. It is largely helpful in reducing transmission latency, demonstrating superior performance in network delay and lifetime compared to existing schemes.

In contrast to traditional wireless networks, sensor nodes within WSNs possess severely constrained power, bandwidth, and storage resources while contending with an error-prone wireless channel. These characteristics render the development of reliable data transmission protocols for WSNs exceptionally challenging. Furthermore, conventional reliable data transmission protocols commonly employed in the internet and other wireless networks, such as TCP and UDP, are ill-suited for the unique constraints of WSNs [38] Despite the emergence of numerous reliable transmission protocols, the majority of them cater to highly specific applications [39], rather than addressing the needs of general WSNs. With reliable data delivery in facial expression recognition, being a broad field of study, it faces certain challenges due to intra-class diversity. The study in [40] introduces a framework using HOG descriptors and Random Forest thereby proving remarkable accuracy on challenging facial datasets.

Energy conservation is a paramount concern in WSNs, manifesting as a primary focus across numerous WSN applications. Typically, two avenues are explored for achieving energy efficiency:

Minimizing energy consumption at the individual node level: This entails reducing energy consumption by decreasing transceiver usage time (e.g., implementing a sleep mechanism) and optimizing transceiver power settings since transceivers and data processing units are the primary energy consumers in sensors.

This paper exclusively delves into the design of low-energy communication strategies and investigates various techniques for energy conservation, including sleep schedule adjustments, energy-efficient MAC protocols, energy-efficient routing protocols, and low-energy architectural designs within IEEE 802.15.4. While these strategies individually achieve energy efficiency, our focus is on optimizing performance across multiple strategies to attain superior results.

In existing hop-count-based routing protocols, WSN sensors are typically divided into sub-zones based on their hop counts from the sink node, as illustrated in Fig. 2 model I. In this model, sensor nodes A and B share the same hop counts despite their differing locations. The work in [41] explores wireless mobile ad-hoc networks, analyzing DSDV, AODV, and DSR routing protocols. This study compares their performance, revealing that while it transfers more data, AODV experiences fewer losses due to path changes. Here, the significant goal is to identify suitable routing protocols for WSNs, addressing technology limitations and proposing an enhanced protocol. The conventional hop-count-based routing approach provides only approximate location information to each sensor, resulting in uneven distribution of next forwarding neighbors. Denoting expected transmission reliability as R and channel error as e (0 < e < 1), and representing the number of redundant transmissions or next forward neighbors as M, the desired forwarding reliability R can be obtained [4,5] as follows:

(1)R=1−eM

As per Eq. (1), the forwarding reliability of nodes A and B in Fig. 3 is determined by the count of the next forward neighbors located at hop h−1. Assuming that the counts of next forward neighbors for nodes A and B at hop h−1 are denoted as MA,f and MB,f, respectively, and considering an average channel error of e, the forwarding reliability of nodes A and B can be expressed as follows using Eq. (1):

(2)RA=1−e(MA,f) (3)RB=1−e(MB,f)

Model II in Fig. 2 shows that MA,f > MB,f, under the same distribution of sensor nodes (Because Node A covers more region area in hop h-1). Thus, node A has higher forwarding reliability than B, namely: (4)RA>RB

It shows that the general hop-count-based routing scheme makes the node at the same hop have different forwarding reliability. We designed an improved hop-count routing scheme to address this problem in the following part.

Different numbers of next forward neighbors cause unbalanced forwarding reliability due to the rough location information (only based on hop count). Thus, one of the possible ways to resolve the current problem is to make the location information of each node more accurate. As is shown in Fig. 2 model I, the communication ranges of nodes A and B cover three hop regions and have different coverage at the forwarding hop h and the backward hop h+1. This kind of information could improve the accuracy of the location in the general hop-count routing scheme. We define a new type of hop count H’ for a given node γ as: (5)H ′=α(h−1)+βh+γ(h+1);α+β+γ=1

To find the values of α1¨¼ β, and γ, we assume the numbers of neighbor nodes of the given node γ at hop h+1, h, and h−1 are Nh+1, Nh and Nh−1, respectively, then we have: (6)α=N(h−1)N(h−1)+Nh+N(h+1) (7)β=NhN(h−1)+Nh+N(h+1) (8)γ=N(h+1)N(h−1)+Nh+N(h+1)

Finally, we can express the new optimized hop count H’ as: (9)H=α(h−1)+βh+γ(h+1)+N(h−1)(h−1)NhhN(h+1)(h+1)N(h−1)+Nh+N(h+1)

Finally, H’ means the average hop counts of neighbor nodes within the communication range. To better understand how the improved hop-count routing scheme works, we show an example of new hop-count recalculating in example I of Fig. 3. Node A has 4, 8, and 4 neighbor nodes at hop 5, 6, and 7, respectively, and node B has 2, 8, and 4 neighbor nodes at hop 5, 6, and 7 separately. Then, we recalculate the new hop count H’ for nodes A (H’A) and B (H’B) as follows:

For node A: (10)α=M5(M5+M6+M7)=4(4+8+2)=27 (11)β=M6(M5+M6+M7)=8(4+8+2)=47 (12)γ=M7(M5+M6+M7)=4(4+8+2)=27 (13)HA ′=α5+β6+γ7=275+476+177≈5.86

For node B: (14)α=M5(M5+M6+M7)=2(4+8+2)=17 (15)β=M6(M5+M6+M7)=8(4+8+2)=47 (16)γ=M7(M5+M6+M7)=2(4+8+2)=17 (17)HB ′=α5+β6+γ7=175+476+277≈6.14

In our enhanced hop-count-based routing scheme, the traditional integer hop counts of nodes A and B are refined into more precise non-integer hop counts. This refinement enables each node to identify an increased number of equally distributed next-forwarding neighbors, thereby alleviating issues associated with the standard hop-count routing approach. As illustrated in Fig. 4, node A and node B serve as an example. In the conventional hop-count routing scheme, node A has two forwarding neighbors, I and J, both at hop 5, whereas node B has only one forwarding neighbor. However, after recalculating the new hop counts, the integer hop counts are further divided into smaller decimal hop counts, as indicated in the figure. Consequently, node B can transmit its packet to nodes H and K, which possess smaller hop counts such as 5.45, as depicted in Fig. 4.

To build a more efficient hop-count-based routing scheme, we have to collect general hop-count information for each node in the network:

The sink node initializes its hop count as ‘0’ and broadcasts a hop-count ‘1’ message to its neighbors. The neighbors set this received hop count as their latest hop count, then the neighbors rebroadcast the message to their neighbors by increasing their hop count by 1. In this course, we implement the adaptive broadcasting algorithm [9] to mitigate collisions of broadcasting.

In WSNs, when an event occurs, it triggers the neighboring nodes to simultaneously transmit event data to the sink node for aggregation, as depicted in Fig. 5. Despite the widespread adoption of the cluster structure for data aggregation, two primary challenges persist:

Cluster head election: In typical event-driven WSNs, selecting the cluster head from the active sensor nodes is a critical step in cluster formation. In our study, we assume that cluster head selection can be performed using existing algorithms, such as those discussed in [10,11,33].

Upon the occurrence of an event, nodes are activated to report the event to their respective cluster head. However, the frequent transmissions to the same cluster head lead to a high collision rate, even when using the CSMA/CA algorithm. Although CSMA/CA was originally designed to coordinate channel access among multiple wireless devices, its collision mitigation efficiency in large-scale WSNs falls short of expectations. Additionally, the CSMA/CA algorithm relies on nodes introducing random delays to avoid collisions without considering the current channel utilization.

Furthermore, in the context of cluster-based WSNs operating under IEEE 802.15.4, most research has focused on minimizing interference between neighboring clusters [36,42], rather than addressing collisions within a single cluster. Therefore, it becomes necessary to enhance the original CSMA/CA algorithm to better suit the requirements of WSNs.

It is important to note that we define the “Enhanced channel utilization” as the state achieved when the network operates at maximum utilization. For a cluster comprising n nodes, we establish the transmission probability (τ) within a random time slot. Subsequently, we calculate the probabilities for three distinct cases during the time interval [t,t+1]: 1.

A channel stays idle:

Based on (18)–(20), we can compute the network utilization as: (21)S=(pSuccTPacket)(pSucccTSucc+pIdleTIdle+pCollTColl)

Here, TSucc presents the average time duration of channel utilization in case of successful transmissions, and TColl is the average time duration if the channel experiences a collision. Here, TIdle is the average time duration in case of the channel stays idle.

To maximize network utilization S, we can get the enhanced transmission probability according to the deduction in [43]: (22)τOpt=sqrt([n+2(n−1)(T∗−1)/n])−1(n−1)(T∗−1)≈1nsqrt(T∗/2)

Here, T^* = TColl/TIdle in (22), the parameter T^* could be a known value by observing the channel utilization. Thus, the Enhanced transmission probability only depends on the number of sensors (the parameter n) in the cluster. However, it is impossible to predict the number of activated sensors in the event, and we have to find another way to maximize network utilization.

As the only centralized coordinator in a cluster, the cluster head can count the time duration of TColl and TIdle while receiving the data from member nodes. Based on the TColl and TIdle acquired from the cluster head. We define the channel utilization L as: (23)L=E[TIdle]E[TColl]=pIdleTIdlepCollTColl=(1−τ)n[1−(1−τ)n−nτ(1−τ)(n−1)].T∗

Here, L is known. Usually, due to a large number of nodes and low transmission probability, we make n→∞ and τ→0, and according to the Taylor formula, we obtain: (24)limn→∞,τ→0(1−τ)n≈1−nτ+n(n−1)2τ2

Combined with (24), we can further simplify (23) as: (25)L=(1−τ)n[1−(1−τ)n−nτ(1−τ)(n−1)].T∗=1(1−τ)(−n)−1−nτT∗=2n(n+1)τ2T∗

Replace τ with τOpt in Eq. (22), we obtain the enhanced L: (26)LOpt=2n(n+1)τ2TColl∗=2n(n+1)(n(T∗/2)(−2))TColl∗≈1

LOpt shows the Enhanced rate of TColl and TIdle while achieving the enhanced network utilization. In other words, when the average collision time duration is equal to the average idle time duration, the channel achieves the best utilization. However, we still do not have clear guidance from L to adjust the transmission behaviour of each node. In this case, we further transform (25) to be: (27)T∗L(1−τ)n=[1−(1−τ)n−nτ(1−τ)(n−1)]

According to Taylor formula, we can derive: (28)1−(1−τ)n−nτ(1−τ)(n−1)τ≈n(n−1)2τ2

Assume that n→∞, n(n−1) ≈ n2, thus we can transfer Eq. (27) to be: (29)n2τ2T∗L=2−2nπ+n2τ2

Replacing nπ by Y, we have: (30)(T∗L−1)γ2+2γ−2=0

From Eq. (30), we have: (31)Y=((1+2(T∗L−1))−1)/(T∗L−1)

Replacing L with LOpt: (32)YOpt=(1+2(T∗L−1))−1T∗L−1

Here, Y is the number of sensor nodes that may transmit in a random time slot. Thus Y−YOpt indicates the difference in the number of active sensor nodes before and after the enhancement.

According to the previous analysis, we propose an Enhanced CSMA/CA algorithm as follow: Channel state estimation. When the cluster head receives the data from its member nodes, it exactly knows the channel utilizing condition. It counts the time duration of TSucc, TColl and TIdle at a certain period i (usually, it is available to exploit this kind of information at the MAC layer), as shown in Fig. 6. Then we can average them to reflect the channel-utilizing condition.

(33)E[TColl]=E[∑(i=1)nTColli] (34)E[TIdle]=E[∑(i=1)nTIdiei]

The channel-using information must be updated frequently to track channel-utilizing conditions more accurately. Thus, we only count the average collision and idle time duration in the last x periods.

Then, we can evaluate the channel utilizing conditions as: (35)L=E[TIdle]E[TColl]

The value of L will be updated every x period and recorded as (L1, L2, L3, …, Ln). Transmission adjustment. After obtaining the latest channel utilizing condition parameter L, the cluster head will enclose this parameter in the Beacon Payload field of the periodically generated beacon frames, as shown in Fig. 7.

After receiving the beacon, each sensor node knows the current channel utilizing condition, adjusting their behaviors to mitigate the possible collisions. To achieve this goal, each transmission attempt will be enabled by a certain probability Pt and rescheduled by probability 1−Pt.

Here, Pt can be defined as: (36)Pt=1−[min(1,YYOpt)]NB+1

Here, NB indicates the back-off times of CSMA/CA or the times the sensor node tries to transmit. The fairness of each transmission among the member nodes will be improved. From (36), we can find that:

A. In the case of Y/Yopt < 1. Here, Y/Yopt < 1 indicates channel utilization is not sufficient. And it should enable more member nodes to transmit for higher channel utilization.

As for NB, the larger NB achieves a higher transmission probability. Thus, the nodes with more extended waiting have a higher priority or more privileges to transmit.

B. In case of Y/Yopt > 1.

Here, Y/Yopt > 1 indicates channel utilization is excessive. And upcoming transmission attempts has to be suspended for lower collisions. The flow of the ECC algorithm is shown in Fig. 8.

As is known to all, redundant transmissions are recognized as one of the most effective methods to deliver desired reliable data. However, there are still some problems in existing research:

Forwarding collisions. In most end-to-end reliable transmission protocols, the source node transmits a packet to next-hop neighbors. The neighbors who receive the source packet correctly continuously forward the packet to next-hop neighbors separately until the sink node receives it. The same source packet can be transmitted via multiple paths to avoid transmission loss due to wireless link failures, as is shown in various transmitters of Fig. 9. However, the reliability is seriously degraded due to the collisions caused by multiple transmissions at each hop [44,45]. These researchers assume an ideal TDMA scheme is adopted to cope with this situation, which is impractical.

Excessive transmissions. Another problem of most end-to-end reliable protocols is that all suitable transmitters have to compete for their communications at each hop without considering whether they have already succeeded in forwarding the packet to the next-hop. As a result, it wastes much more energy.

Explicit acknowledgment (ACK). In the hop-by-hop recovery, an ACK is used to confirm each successful transmission at each hop. However, the ACK consumes network resources under the same wireless link.

To resolve the problems mentioned earlier, our design should meet the following requirements:

Rule 1. Only one node can be selected as the forwarding node at each hop. As is shown in the single transmitter of Fig. 9. In this case, the redundant transmissions are reduced, and the collisions caused by the multiple forwarding nodes are avoided.

The overview of our designed protocol is that the cluster head periodically transmits the collected event information to the sink node via multi-hops. Only one node sends the source packet to its next bop neighbors at each hop. Then, only one of the neighbors who received the source packet correctly will be selected to forward this packet continuously until the sink node receives the source packet, as shown in Fig. 10. Table 2 shows the list of notations used in the reliable transmission framework.

The detailed design of reliable data delivery over multi-hops is shown as follows:

Sleeping forwarding neighbors wake up. When a packet is ready to be forwarded to the next hop, the next forwarding neighbors at the next hop have to be activated for packet reception. To save energy in the inactive neighbor’s nodes waking up duration, we adopt the successive small preambles [1] to wake up the next forwarding neighbors, as shown in Fig. 11.

Election of unique forwarding node. We proposed an efficient algorithm to select a ‘best node’ as the unique forward node to enable a single node for data transmission and disable the other nodes to be “silent”.

Step 1: When a transmitting node (including cluster head) is ready to transmit a packet, it monitors the channel and ensures the channel is not occupied by any other transmission, the same as the CSMA/CA. Then, the source node transmits the packet to neighbors at next-hop. Otherwise, if the channel is occupied, the source node delays the transmission to wait until the channel becomes available.

(37)Tslot(i)=[Tn]HF_max ′−HI ′HF_max ′−HF ′n+[Tnmrand()]

Here, HI shows the enhanced hop counts of node I, H’F, denotes the enhanced hop counts of node F. And H ′Fmax, is the max enhanced hop counts of the neighbor nodes of F. Finally, H ′Fmax−H ′F implies the max communication range of node F.

According to (37), each candidate first selects a time slot based on their different improved hop counts. As shown in (37), the nodes with the more minor hop counts would be allocated the front time slots. In other words, the nodes closer to the sink node have more privileges to become the next forwarding transmitter. As shown in Fig. 12, node I and node J stay closer to the sink node than node K. Thus, node I and node J choose a shorter backoff time slot t2. Furthermore, to prevent two or more candidates from selecting the same time slot, like nodes I and J, each candidate has to randomly choose a time slice to scatter each transmission attempt further. Finally, node I will first start its transmission attempt due to the shortest backoff time delay.

Reduction of excessive transmissions. A forwarding node has to multicast the received packets to next-hop neighbors at each hop to guarantee the expected reliability. However, the necessary number of neighbor nodes could differ depending on the different situations. When a wireless link experiences a high transmission failure, it needs more receiving neighbors to improve the forwarding reliability, as shown in Fig. 13a. On the contrary, when the wireless link condition stays good, the receiving neighbor nodes should be reduced to save energy, as shown in Fig. 13b.

To reduce the excessive transmissions, each forwarding node calculates the thresholds △h according to the average wireless link failure and adds it into the packet header for the next multicast. Then only the neighbors less than (H−△h) are allowed to receive this packet. According to different thresholds △h, a packet from node A can multicast to the different sets of neighbors in different shadow areas SA.

To validate the performance of our protocol, our simulation platform was built in MATLAB. A sensor node is created in our program with the necessary functions of the MAC layer, network layer, transmission layer, and application layer. Then, we deploy a number of sensor nodes in the network to form our target practical application scenario. Most of the parameters are derived from our common experiment hardware and others from our previous experimental experience to guarantee the performance evaluation close to the practical applications.

In our simulation, different numbers of stationary sensor nodes are randomly deployed over 200 ^* 200 m² field, and we randomly select a circular area as the event area. The sensors in the event area will generate the event information, and each event is assumed to be on for the 60 s. The other simulation settings are defined, and the parameters of the simulation we adopted are listed in Table 3.

For better presentation, we use the abbreviations of GHRS, IHRS, and GLRS to stand for general hop-count routing schemes, improved hop-count routing schemes, and geographic location-based routing schemes, respectively.

We abstract the main principles from two kinds of reliable data delivery protocols to prove the better performance of RDDM.

End-to-end error recovery (E2E). Assume that acknowledgment is not applied. The cumulative transmission loss at each hop has to be considered for the expected transmission reliability from the source to the destination. Thus, the expected number of next forwarding neighbors N can be expressed by Eq. (40): (40)N=(log(1−r))/(log−(1−(1−e)h))

Here, r presents desired reliability and indicates channel error, and h is hop count. Hop-by-hop error recovery (H2H). Assume that acknowledgment is applied at each hop. the cumulative packet loss at each hop has not to be considered. Thus, the expected number of next forwarding neighbors N can be derived from the following Eq. (41): (41)N=log(1−r)/loge

An example is given in Table 5 to address the different expected numbers of next forwarding neighbors at the source node in the two models. In this example, we set the expected reliability to 0.9 and calculate the expected transmission paths with different channel error rates and hop counts for two models. Table 5. The expected number of next forwarding neighbors.

In this part, we set different link failure rates to show the different performance on packet delivery rate, time delay, and energy consumption by repeating 100 rounds. Finally, these performance details are addressed as follows.

Fig. 18 compared the average packet delivery ratio in terms of different transmission failure rates. The RDDM consistently outperforms the protocols E2E and H2H in different transmission failure rates. RDDM consistently achieves the lowest packet delivery rate. At the same time, E2E becomes the worst because RDDM can provide the highest packet transmission and forwarding reliability than the other two protocols at each hop.

Fig. 19 compared the average transmission time delay in different failure rates. The figure shows that RDDM outperforms the other two protocols by achieving the lowest time delay. Because RDDM has the highest forwarding reliability at each hop to avoid forwarding failure. Meanwhile, a single node is selected to forward the packet without redundant transmissions in RDDM. Finally, RDDM confirms the successful forwarding by monitoring the channel to reduce the time delay of ACK failures.

Fig. 20 compared the average energy consumption in different transmission failure rates. The figure shows that RDDM achieves the best performance with the lowest energy consumption. We can analyze why (1) RDDM balanced the forwarding reliability and reduced the forwarding failure at each hop. (2) In RDDM, the packet can be forwarded by only one node at each hop, and the lowest energy consumption is achieved by avoiding energy wasting on redundant transmissions. Meanwhile, in E2E and H2H, multiple paths are established, and a packet could be forwarded by multiple forwarding nodes at each hop, resulting in energy wasting for the excessive redundant transmissions.

After the data aggregation, the aggregated data needs to be reliably transmitted from the cluster head to the sink node over multi-hops. Due to wireless link failure, data transmissions may fail to be delivered. To guarantee transmission reliability over multi-hops, we design a reliable data delivery mechanism that ensures desired reliability and achieves better performance on other metrics. Finally, the mathematical analysis and simulation show the better performance we achieved with our design [45]. However, although the better performance of our design is validated, there are some challenges [46–50].