Optimizing Routing Protocol Design for Long-Range Distributed Multi-Hop Networks

Abstract

The advancement of communication technologies has facilitated the deployment of numerous sensors, terminal human–machine interfaces, and smart devices in various complex environments for data collection and analysis, providing automated and intelligent services. The increasing urgency of monitoring demands in complex environments necessitates low-cost and efficient network deployment solutions to support various monitoring tasks. Distributed networks offer high stability, reliability, and economic feasibility. Among various Low-Power Wide-Area Network (LPWAN) technologies, Long Range (LoRa) has emerged as the preferred choice due to its openness and flexibility. However, traditional LoRa networks face challenges such as limited coverage range and poor scalability, emphasizing the need for research into distributed routing strategies tailored for LoRa networks. This paper proposes the Optimizing Link-State Routing Based on Load Balancing (LB-OLSR) protocol as an ideal approach for constructing LoRa distributed multi-hop networks. The protocol considers the selection of Multipoint Relay (MPR) nodes to reduce unnecessary network overhead. In addition, route planning integrates factors such as business communication latency, link reliability, node occupancy rate, and node load rate to construct an optimization model and optimize the route establishment decision criteria through a load-balancing approach. The simulation results demonstrate that the improved routing protocol exhibits superior performance in node load balancing, average node load duration, and average business latency.

1. Introduction

The LoRa network can meet the range and power requirements of various applications. Traditional LoRa networks adopt a star topology, where terminal nodes transmit data to the gateway using a single-hop transmission mode. This approach is flexible and has low latency, as terminal nodes only need to focus on their data without additional data forwarding, effectively reducing terminal energy consumption and extending the terminal’s lifespan [1,2,3]. However, when the number of terminal nodes is extensive, their deployment locations are scattered, and the number of gateways is insufficient, issues such as insufficient communication coverage, signal interference and congestion, and increased packet loss rates may arise. Addressing these issues by increasing the number of gateways is costly, difficult to maintain, and poorly scalable. Therefore, distributed multi-hop networks become a feasible solution [4]. Reference [5] analyzes the challenges that need to be addressed in LoRa multi-hop and mesh network solutions, emphasizing scalability and complexity. Reference [6] examines the use of multi-hop LoRa in smart city applications to extend coverage and reduce energy consumption at terminal nodes. The authors compare LoRa single-hop and multi-hop networks, concluding that multi-hop networks provide better performance, although there are concerns regarding the scalability of their implementation. In reference [7], the authors compare packet transmission in LoRa single-hop and multi-hop networks, indicating that mesh topologies can effectively monitor large areas; however, there are significant limitations related to security and latency. This paper proposes an improved protocol based on the OLSR protocol, designed for LoRa distributed multi-hop networks.

In distributed multi-hop networks, the OLSR protocol provides immediate routing information, meeting the transmission needs of various types of business data. During route establishment and maintenance, nodes exchange periodic broadcast Hello messages for link sensing and neighbor detection and broadcast Topology Control (TC) messages to transmit routing information and the overall network topology status. This consumes part of the channel’s resources, causing network congestion. The OLSR protocol reduces the number of retransmissions of the same message within the same area through the Multipoint Relay (MPR) selection mechanism, thereby reducing the network overhead [8]. However, the traditional MPR selection algorithm results in redundant MPR sets, increasing the network overhead. It only uses the hop count as the routing criterion, making it difficult to adapt to actual network link conditions.

To address the redundancy in the MPR set selected by the OLSR protocol’s MPR selection mechanism, reference [9] proposes a Necessity-first Algorithm (NFA) to select the MPR set, reducing the number of topology control messages and MPRs. However, the authors did not extend their consideration to parameters such as PDR, throughput, delay, and energy consumption. Reference [10] proposes an enhanced MPR selection algorithm that retains the advantages of the traditional MPR selection algorithm and uses an extended local database of three-hop neighboring nodes for MPR selection, further reducing the number of topology control messages and the network overhead. Reference [11] introduces the min–max algorithm, considering Quality of Service (QoS) and node density, and selects MPR nodes based on the maximum signal range. By introducing the min–max algorithm to enhance OLSR performance, this algorithm considers QoS and node density. The min–max algorithm is used to select MPR nodes based on the maximum signal range, analyzing the QoS parameters using different numbers of nodes.

Addressing the single routing criterion in the OLSR protocol, reference [12] proposes the ML-OLSR protocol, which calculates link reliability using distance information, derives node reliability (SDN), and introduces mobility-aware and load-aware algorithms. Reference [13] applies metrics such as Expected Transmission Count (ETX), minimum loss, and minimum delay to optimize the OLSR protocol’s performance. Reference [14] proposes an optimal routing strategy based on the Dijkstra algorithm to establish predictive communication, though incorrect predictions may result in erroneous route selections, increasing the network overhead.

After an in-depth study of the existing literature, this paper, addressing the shortcomings of the MPR selection algorithm in the OLSR protocol, proposes an improved MPR selection algorithm based on connection necessity, resolving redundancy by deleting neighbor nodes with minimal coverage. Addressing the single routing criterion issue in the OLSR protocol, this paper proposes allocating SF to different services in LoRa multi-hop networks, determining service priority, and considering constraints such as communication latency and link reliability. Multifactor decision criteria and a load cost matrix are formulated to plan transmission paths globally from terminal nodes carrying different services to the gateway.This paper proposes the LB-OLSR protocol, aimed at addressing issues such as insufficient communication coverage, signal interference, congestion, and increased packet loss rates that may arise in traditional LoRa networks under certain complex environments. The proposed method seeks to improve network traffic distribution and enhance network balance, designing a protocol suitable for LoRa multi-hop networks that need to handle a large volume of requests while effectively utilizing resources.

The remainder of this paper is structured as follows: Section 2 outlines the LoRa distributed multi-hop network model and analyzes the related issues. Section 3 presents a connection necessity-based MPR selection algorithm to address the deficiencies of traditional MPR algorithms. Section 4 introduces the load-balancing routing optimization model and the service routing model. Section 5 details the implementation process of load-balancing routing optimization. Section 6 describes the experimental testing scenarios and the results obtained. Finally, Section 7 concludes the paper and offers insights into future work.

2. Network Model and Problem Analysis

2.1. Network Model

The LoRa distributed multi-hop network model, as illustrated in Figure 1, primarily comprises three modules: the control center module, the LoRa distributed network module, and the sensor module [15,16]. The sensor module is responsible for sensing and measuring environmental parameters. Within the distributed network module, LoRa nodes within the gateway’s coverage area directly send the data collected by the sensors to the gateway. If there are obstacles between the LoRa nodes and the gateway, or if the nodes are subject to interference from other wireless devices or signals, preventing direct communication with the gateway, the LoRa nodes will use a routing algorithm to select the optimal relay nodes to transmit data until it reaches the target node or gateway [17]. The LoRa network server receives the data packets, removes duplicates, and forwards the data to the application server.

2.2. Problem Analysis

The MPR selection mechanism is central to the OLSR protocol. Its purpose is to ensure that nodes in the network periodically send topology control information to all two-hop neighbors to quickly acquire the overall network topology while minimizing the retransmission of topology control packets, thereby reducing network control overhead [18,19,20]. However, the traditional MPR selection algorithm’s preference for nodes with high coverage can lead to redundant MPR nodes, which increases the flooding of TC messages and consequently causes exponential growth in the network overhead. Given the limited channel resources, this heightens the risk of network congestion, potentially leading to the accumulation and expiration of control packets, route failures, or even network paralysis. Hence, MPR sets should be chosen based on the necessity of connection. If other MPR sets’ two-hop neighbors also cover a node covered by an MPR set’s two-hop neighbor, it indicates that this node is redundant in the MPR set and should be removed.

The traditional OLSR protocol uses only hop count as the standard for route establishment and employs the Dijkstra algorithm to compute and generate node routing information. This approach needs to pay more attention to the quality of links in real environments, making it challenging to accommodate situations with high traffic volumes or uneven network conditions. This can result in untimely route updates, link breaks, and traffic congestion. Therefore, considering only hop count may not optimize path selection [21,22,23]. LoRa communication parameters should be allocated according to different service requirements to adapt flexibly to various service scenarios. Considering that terminal devices in a distributed network may act as relay devices for multiple services during transmission, route establishment should dynamically select the optimal route for data transmission based on communication delay constraints, link reliability constraints, and node load conditions, aiming to achieve load-balanced routing optimization.

3. MPR Selection Algorithm Based on Connection Necessity

This paper proposes an improved MPR selection algorithm from the necessity perspective, eliminating the redundancy present in traditional algorithms. The algorithm flow is shown in Figure 2. The relevant concepts involved in the algorithm process are as follows:

• Node Depth: The depth of a one-hop neighbor node k (where node k is a one-hop neighbor of node N) represents the number of neighbors with strictly symmetric links to node k, excluding the one-hop neighbors of node N. For example, in Figure 3, the depth of node 1 is 2, including nodes a and b.
• Node Coverage: The coverage of a one-hop neighbor node k indicates the number of neighbors with strictly symmetric links to node k after removing the nodes already included in the MPR set and their covered two-hop neighbors, excluding the one-hop neighbors of node N. For example, in Figure 3, if node 1 is already included in the MPR set, the coverage of node 2 is 2, including nodes c and d, but excluding node b.

Table 1. Description of parameters in MPR selection algorithm based on connection necessity.

Algorithm Parameter	Parameter Description
$\mathrm{MPR}\text{\_}\mathrm{Sel}$	MPR selection nodes
$\mathrm{Sel}\text{\_}\mathrm{MPRS}$	MPR node set
${\mathrm{Sel}\text{\_}\mathrm{MPRS}[:]}_{size}$	Number of nodes in the MPR set
${\mathrm{FN}[:]}_{size}$	Number of one-hop neighbors
FN[i]	The i-th hop neighboring node
${\mathrm{SN}[:]}_{size}$	Number of two-hop neighbors
SN[i]	The i-th two-hop neighboring node
$\mathrm{Avail}\text{\_}\mathrm{FN}$	Remaining available one-hop neighbors
$\mathrm{Avail}\text{\_}\mathrm{SN}$	Remaining available two-hop neighbors
${\mathrm{ACC}}_{uni}$	Unique path between two-hop and one-hop neighbors
Cov[i]	Two-hop neighbors covered by node i
${\mathrm{Cov}\left[\mathrm{i}\right]}_{size}$	Maximum coverage of node i
${\mathrm{Max}\text{\_}\mathrm{Cov}}_{size}$	Depth of node i
${\mathrm{Dep}\left[\mathrm{i}\right]}_{size}$	Maximum depth of node i

describes the parameters of the MPR selection algorithm based on connection necessity. Algorithm 1 presents the MPR selection algorithm based on connection necessity.

In this study, networks were formed by randomly setting 100, 200, 300, and 400 nodes, while controlling the coverage radius of the nodes to achieve an average number of neighboring nodes of 10, 20, 40, and 60. We compared the number of global MPR nodes obtained using the improved MPR selection algorithm with that of the standard MPR selection algorithm under the same network scale. The results indicated that, under varying densities, the improved MPR selection algorithm demonstrates a reduction in the number of calculated MPR nodes ranging from 2% to 12% compared to the standard algorithm. This finding suggests that the improved MPR selection algorithm can significantly reduce the redundancy of MPR nodes, as shown in Figure 4.

The improved MPR selection algorithm significantly reduces redundant nodes within the MPR set from a necessity standpoint. As illustrated in Figure 5, in a small-scale sparse network, where the number of neighbor nodes per node is relatively few, the traditional MPR selection algorithm results in fewer redundant MPR nodes. Hence, the improvement effect is not pronounced. However, in a large-scale dense network, where the number of neighbor nodes per node is relatively high, the traditional MPR selection algorithm leads to many redundant MPR nodes. The improved algorithm can significantly reduce redundancy. This demonstrates that the advantages of the improved MPR selection algorithm are more evident in large-scale, dense networks, effectively reducing the control message overhead, cutting down on network topology maintenance costs, and improving routing computation efficiency.

Algorithm 1 MPR selection algorithm based on connection necessity.

Input:  $FN=FN$[MPR_Sel], $\mathrm{Avail}\text{\_}\mathrm{FN}=FN$[MPR_Sel], $SN=SN$[MPR_Sel], $\mathrm{Avail}\text{\_}\mathrm{SN}=SN$[MPR_Sel] Output:  $\mathrm{Sel}\text{\_}\mathrm{MPRS}$   1: for each $i\in [1,{\mathrm{FN}[:]}_{size}]$ do   2:    for each $j\in [1,{\mathrm{SN}[:]}_{size}]$ do   3:      if ${\mathrm{ACC}}_{uni}=1$ then   4:          $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ = $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ + FN[j]   5:          $\mathrm{Avail}\text{\_}\mathrm{FN}$ = $\mathrm{Avail}\text{\_}\mathrm{FN}$ - $\mathrm{Sel}\text{\_}\mathrm{MPRS}$   6:          $\mathrm{Avail}\text{\_}\mathrm{SN}$ = $\mathrm{Avail}\text{\_}\mathrm{SN}$ - Cov[$\mathrm{MPR}\text{\_}\mathrm{Sel}$]   7:      else if ${\mathrm{Cov}\left[\mathrm{FN}\right[\mathrm{i}\left]\right]}_{size}={\mathrm{Max}\text{\_}\mathrm{Cov}\left[\mathrm{FN}\right]}_{size}$ && ${\mathrm{Dep}\left[\mathrm{FN}\right[\mathrm{i}\left]\right]}_{size}={\mathrm{Max}\text{\_}\mathrm{Dep}\left[\mathrm{FN}\right]}_{size}$ then   8:          $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ = $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ + FN[j]   9:          $\mathrm{Avail}\text{\_}\mathrm{FN}$ = $\mathrm{Avail}\text{\_}\mathrm{FN}$ - $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ 10:          $\mathrm{Avail}\text{\_}\mathrm{SN}$ = $\mathrm{Avail}\text{\_}\mathrm{SN}$ - Cov[$\mathrm{MPR}\text{\_}\mathrm{Sel}$] 11:      end if 12:    end for 13:    if $\mathrm{Sel}\text{\_}\mathrm{MPRS}=SN$ then 14:       break 15:    else 16:       continue 17:    end if 18: end for 19: for each $k\in [1,{\mathrm{Sel}\text{\_}\mathrm{MPRS}[:]}_{size}]$ do 20:    if $Cov$[MPR_Sel[k]]$=Cov$[Sel_MPRS - Sel_MPRS[k]] then 21:        $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ = $\mathrm{Sel}\text{\_}\mathrm{MPRS}$ - $\mathrm{Sel}\text{\_}\mathrm{MPRS}\left[\mathrm{k}\right]$ 22:        if $\mathrm{Cov}\left[\mathrm{Sel}\text{\_}\mathrm{MPRS}\right]=SN$ then 23:           break 24:        else 25:           continue 26:        end if 27:    end if 28: end for 29: return $\mathrm{Sel}\text{\_}\mathrm{MPRS}$

4. Load-Balancing Routing Optimization Scheme Design

The load-balancing routing optimization strategy comprehensively considers communication delay, link reliability, node occupancy rate, and node load rate to plan transmission paths that meet the constraint conditions for different service data types in a LoRa distributed multi-hop network. The load-balancing routing optimization strategy is shown in Figure 6.

The specific optimization strategy is as follows: Initially, the gateway broadcasts a routing request to obtain information such as the location of the end nodes, link information, end node occupancy rate, and end node load status within the network. Next, it obtains the service type of the end nodes and, at the same time, calculates the node weight matrix based on the end node location information to determine the service priority. Then, it determines the SF that meets the service type for end nodes of different service priorities under constraint conditions, and plasn the transmission path for each end node [23,24]. Afterward, from a global optimum perspective, the node load balance degree and the average node load rate are minimized, and the link with the lowest load cost is selected as the path to reach the aggregation node. Finally, the routing optimization ends.

4.1. Load-Balancing Optimization Model

In a LoRa distributed multi-hop network, end nodes not within the gateway coverage range need to relay service data to the gateway through multiple hops via relay nodes. When multiple end nodes in the network carrying different service types report service data simultaneously, some end nodes will assume the task of relaying service data for other nodes. To avoid network congestion caused by excessive load on some end nodes, load-balancing routing optimization is necessary to ensure that end nodes operate within a reasonable load range [25].

The load-balancing optimization model for a LoRa distributed multi-hop network uses communication delay, link reliability, node occupancy rate, and node load rate as constraint conditions to minimize the node load balance degree while maintaining a lower average node load rate. This paper addresses only the allocation of the SF, with all other LoRa packet parameters remaining consistent.

• Communication Delay

Communication $T_{\mathrm{delay}}$ is the time required for business data to be transmitted from LoRa end nodes to the gateway. As indicated in Equation (1), it primarily consists of four components [26]: transmission delay $T_{td}$, propagation delay $T_{pd}$, queuing delay $T_{qd}$, and processing delay $T_{cd}$.

$$T_{\mathrm{delay}}=T_{td}+T_{pd}+T_{qd}+T_{cd}$$

(1)

Given that the propagation delay and the processing delay are relatively small and can be considered negligible, the communication delay is shown as in Equation (2).

$$T_{\mathrm{delay}}=T_{td}+T_{qd}$$

(2)

Transmission delay $T_{td}$ refers to the total air time taken by business data while transferring between nodes. The transmission delay $T_{td}^{M_i}$ for business data from node $M_i$ to the gateway mainly depends on the number of hops $N_l$, the data packet transmission rate $T_{tk}^{SF=k}$, and the number of data packets $B_{M_i}$, as shown in Equation (3).

$$T_{td}^{M_i}=N_l\times T_{tk}^{SF=k}\times B_{M_i}$$

(3)

Queuing delay $T_{qd}$ refers to the time business data must wait for processing at different relay nodes $S_i$ when reaching the aggregation node from a node $M_i$. As shown in Equation (4), $N_l$ indicates the number of relay nodes traversed by business data from the source node to the aggregation node, $B_{S_j}$ represents the number of data packets in the queue of nodes $S_i$ that have higher priority than those of node $M_i$, and ${\alpha }_k^{M_j}=1$ indicates that the data transmission path from node $M_i$ passes through node k; otherwise, ${\alpha }_k^{M_j}=0$.

$$T_{qd}^{M_i}=\sum _{k=1}^{N_l}\left({\alpha }_k^{M_j}\times B_{S_j}\times T_{pk}^{SF=k}\right)$$

(4)

Average communication delay refers to the time it takes for a data packet to be transmitted from the source node to the gateway, as shown in Equation (5). Different business types require different communication delays, with $T_{\mathrm{maxDelay}}$ denoting the maximum allowable average communication delay for a given business type.

$$\overline{T_{\mathrm{delay}}}={\displaystyle \frac{T_{\mathrm{delay}}}{B}}$$

(5)

$$\overline{T_{\mathrm{delay}}}\le T_{\mathrm{maxDelay}}$$

(6)

2.

Link Reliability

Signal propagation can be easily obstructed by obstacles and interference sources, leading to reflection, diffraction, and scattering, resulting in signal attenuation. The logarithmic path loss model is shown in Equation (7), where $d_0$ is the reference distance, usually set to 1; d is the distance between the transmitting and the receiving nodes; $\mu$ is the path loss exponent, and $X_{\sigma }$ is a normal random variable with a mean of 0 and a standard deviation $\sigma$ [27].

$$\begin{array}{c}{P}_l=P_l\left(d_0\right)+10\mu log{\displaystyle \frac{d}{d_0}}+X_{\sigma }\end{array}$$

(7)

$$\begin{array}{c}{X}_{\sigma }={\displaystyle \frac{1}{\sqrt{2\pi }}}{e}^{-{\displaystyle \frac{x^2}{2{\sigma }^2}}}\end{array}$$

(8)

The condition for successful data reception at a node is shown in Equations (9)–(11), where $P_t$ denotes transmission power, $P_l$ represents path loss, $\mathrm{sensitive}$ is the receiving sensitivity of the node, $NF$ represents the noise figure of the receiving node, and $SN{R}_{SF}$ represents the signal-to-noise ratio at $SF$ [28].

$$\begin{array}{c}{P}_r=P_t-P_l\end{array}$$

(9)

$$\begin{array}{c}\mathrm{sensitive}=-174+NF+10{log}_{10}BW-SN{R}_{SF}\end{array}$$

(10)

$$\begin{array}{c}{P}_r>\mathrm{sensitive}\end{array}$$

(11)

Link redundancy is defined in Equation (12), with higher link redundancy indicating greater reliability between nodes.

$$\begin{array}{c}{Q}_{rel}=P_r-\mathrm{sensitive}\end{array}$$

(12)

This study considers using link redundancy between nodes to measure link reliability. Different SFs result in different link redundancies, thus offering varying levels of link reliability. To meet the constraints of different business types, the link reliability $R_l$ is defined, where $Q_{max}$ is the maximum link redundancy between nodes; higher link redundancy between nodes indicates better link reliability.

$$\begin{array}{c}{R}_l=\left\{\begin{array}{c}{\displaystyle \frac{Q_{rel}}{Q_{max}}},Q_{rel}<Q_{max}\hfill \\ 0,\mathrm{others}\hfill \end{array}\right.\end{array}$$

(13)

The link reliability $R_{M_i}$ for the path taken by business data from node $M_i$ to the gateway is shown in Equation (14), where $N_i$ denotes the number of links.

$$\begin{array}{c}{R}_{M_i}=1-{\displaystyle \prod _{j=1}^{N_i}}\left(1-R_j\right)\end{array}$$

(14)

Different business types need to satisfy different link reliability requirements, as shown in Equation (15), where ${Success}_{M_i}$ represents the required success rate for a specific business type.

$$\begin{array}{c}{R}_{M_i}\ge {Success}_{M_i}\end{array}$$

(15)

3.

SF

When planning transmission paths for LoRa end nodes, it is necessary to allocate an appropriate $SF$ based on the business type to meet the requirements for communication delay and link reliability. By calculating the communication delay $T_{\mathrm{delay}}$ and link reliability $R_l$, different business types can obtain the minimum and maximum $SF$ $S{F}_{min}$ and $S{F}_{max}$ that satisfy the constraints. The $SF$ is thus determined as shown in Equation (16).

$$\begin{array}{c}SF=⌊{\displaystyle \frac{S{F}_{min}+S{F}_{max}}{2}}⌋\end{array}$$

(16)

4.

Node Occupancy Rate

The node occupancy rate measures the proportion of time that a node is occupied within a specific timeframe. The node occupancy rate $R_{S_i}$ for node $S_i$ is computed as shown in Equation (17), where $N_s$ represents the number of business nodes, and $M_{S_i}$ represents the maximum number of business data that can be handled by node $S_i$ [29].

$$\begin{array}{c}{R}_{S_i}={\displaystyle \frac{{\sum }_{j=1}^{N_s}{\alpha }_i^{M_j}}{M_{S_i}}}\end{array}$$

(17)

The node occupancy rate $R_{S_i}$ should satisfy Equation (18).

$$\begin{array}{c}{R}_{S_i}\le 1\end{array}$$

(18)

5.

Node Load Rate

The node load rate measures the degree of busyness and resource utilization of a node. The node load rate $C_{s_i}$ for node $S_i$ is calculated as shown in Equation (19), where $N_s$ represents the number of business nodes, $B_{M_j}$ denotes the volume of business data from node $M_j$ whose transmission path passes through node $S_i$, and $B_{S_i}$ is the maximum volume of business data that can be handled by node $S_i$.

$$\begin{array}{c}{C}_{s_i}={\displaystyle \frac{{\sum }_{j=1}^{N_s}\left({\alpha }_i^{M_j}\times B_{M_j}\right)}{B_{S_i}}}\end{array}$$

(19)

The node load rate $C_{S_i}$ should satisfy Equation (20).

$$\begin{array}{c}{C}_{S_i}\le 1\end{array}$$

(20)

The average node load rate reflects the overall busyness of the network, as shown in Equation (21), where $N_t$ denotes the number of nodes in the network.

$$\begin{array}{c}\overline{C_{S_l}}={\displaystyle \frac{{\sum }_{i=1}^{N_t}\left(C_{S_i}\times B_{S_i}\right)}{N_t}}\end{array}$$

(21)

6.

Node Load Balance

Node load balance $C_b^{cv}$ assesses the distribution of load among nodes in the network. The calculation is formulated in Equations (22)–(24), where $C_b^{\sigma }$ represents the standard deviation of the node load balance, and $\overline{C_b}$ denotes the average node load balance.

$$\begin{array}{c}{C}_b^{cv}={\displaystyle \frac{C_b^{\sigma }}{\overline{C_b}}}\end{array}$$

(22)

$$\begin{array}{c}{C}_b^{\sigma }=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^{N_t}{\left(C_{S_i}-\overline{C_b}\right)}^2}{N_t}}}\end{array}$$

(23)

$$\begin{array}{c}\overline{C_b}={\displaystyle \frac{{\sum }_{i=1}^{N_t}{C}_{S_i}}{N_t}}\end{array}$$

(24)

7.

Optimization Objectives

Through the analysis of the load-balancing optimization model, the routing load-balancing problem is transformed into minimizing the node load balance while satisfying constraints on communication delay, link reliability, node occupancy rate, and node load rate. The optimization objectives are shown in Equations (25) and (26), which aim to minimize the node load balance and average node load rate.

$$\begin{array}{c}min\left(C_b^{cv}\right)\\ \mathrm{s}.\mathrm{t}.\left\{\begin{array}{c}{T}_{\mathrm{delay}}\le T_{\mathrm{maxdelay}}\\ R_{M_i}\ge {\mathrm{Success}}_{M_i}\\ R_{S_i}\le 1\\ C_{S_i}\le 1\end{array}\right.\end{array}$$

(25)

If the node load balance is equal, the set of parameters that minimizes the average node load rate is considered.

$$\begin{array}{c}min\left(\overline{C_{S_i}}\right)\\ \mathrm{s}.\mathrm{t}.\left\{\begin{array}{c}{T}_{\mathrm{delay}}\le T_{\mathrm{maxDelay}}\\ R_{M_i}\ge {\mathrm{Success}}_{M_i}\\ R_{S_i}\le 1\\ C_{S_i}\le 1\end{array}\right.\end{array}$$

(26)

4.2. Load-Aware MPR Selection Algorithm Improvement

When a node no longer meets the load rate and occupancy rate constraints, yet is still selected as an MPR node, this can lead to the congestion of topology messages, which is obviously inappropriate. Therefore, in selecting MPR nodes, it is essential to prioritize nodes with low occupancy rates and low load rates to avoid causing topology information congestion. In the OLSR protocol, nodes express their willingness to undertake network traffic through the “willingness” option in the HELLO packet. This willingness ranges from 0 to 7, representing different levels of willingness. If the willingness option is “willing never”, the node will never be selected as an MPR by other nodes; if the willingness option is “willing always”, the node will always serve as an MPR. By default, the value is set to “willing default”.

This paper proposes adjusting the willingness value of nodes by considering both the load rate and occupancy rate comprehensively, as shown in

Table 2. Load-Balanced willingness settings.

	Occupancy Rate	Low	Medium	High
Load Rate
Low		$\mathrm{WILL}\text{\_}\mathrm{HIGH}$	$\mathrm{WILL}\text{\_}\mathrm{HIGH}$	$\mathrm{WILL}\text{\_}\mathrm{DEFAULT}$
Medium		$\mathrm{WILL}\text{\_}\mathrm{DEFAULT}$	$\mathrm{WILL}\text{\_}\mathrm{DEFAULT}$	$\mathrm{WILL}\text{\_}\mathrm{LOW}$
High		$\mathrm{WILL}\text{\_}\mathrm{LOW}$	$\mathrm{WILL}\text{\_}\mathrm{LOW}$	$\mathrm{WILL}\text{\_}\mathrm{LOW}$

. Based on different combinations, the settings are categorized into $\mathrm{WILL}\text{\_}\mathrm{LOW}$, $\mathrm{WILL}\text{\_}\mathrm{HIGH}$, and $\mathrm{WILL}\text{\_}\mathrm{DEFAULT}$ [30].

4.3. Business Routing Model

The traditional OLSR protocol establishes routes based solely on hop count, ignoring the link quality. Thus, it struggles in high traffic volumes or uneven network conditions. This leads to issues like delayed route updates and traffic congestion. Therefore, hop-count-based routing may fail to find the optimal path.

The routing optimization problem is decomposed into transmission path planning for terminal nodes with different business priorities, calculating the node cost matrix. Under constraint satisfaction, the path with the minimum cost is selected as the transmission path for business data. The following analysis covers the establishment of the business routing model.

• Establishing the link weight matrix

Assuming in a distributed multi-hop LoRa network, the number of terminal nodes is n − 1, and the gateway is 1, the network model can be represented by the coordinate matrix $\mathrm{node}\text{\_}\mathrm{Positions}$, distance matrix $\mathrm{node}\text{\_}\mathrm{Distance}$, and adjacency matrix $\mathrm{node}\text{\_}\mathrm{Adjoin}$. Suppose the terminal nodes are numbered (1, 2, 3 … n − 2, n − 1), and the gateway is numbered n.

The coordinates of terminal nodes and gateways are known and fixed, represented by the coordinate matrix $\mathrm{node}\text{\_}\mathrm{Positions}$, as shown in Equation (27). The first n − 1 rows represent the coordinates of terminal nodes from $(x_1,y_1,z_1)$ to $(x_{n-1},y_{n-1},z_{n-1})$, and the n-th row $(x_n,y_n,z_n)$ represents the gateway’s coordinates.

$$\begin{array}{c}\mathrm{node}\text{\_}\mathrm{Positions}={\left(\begin{array}{ccc}{x}_1& y_1& z_1\\ x_2& y_2& z_2\\ ⋮& ⋮& ⋮\\ x_n& y_n& z_n\end{array}\right)}_{n\times 3}\end{array}$$

(27)

The distance matrix between terminal nodes $\mathrm{node}\text{\_}\mathrm{Distance}$ is shown in Equation (28), where $d_{ij}$ represents the distance between terminal node i and terminal node j.

$$\begin{array}{c}\mathrm{node}\text{\_}\mathrm{Distance}={\left(\begin{array}{ccc}{d}_{11}& \cdots & d_{1n}\\ ⋮& \ddots & ⋮\\ d_{n1}& \cdots & d_{nn}\end{array}\right)}_{n\times n}\end{array}$$

(28)

In route planning, to prevent loops or improper routing between terminal nodes and the aggregation node, the distance matrix $\mathrm{node}\text{\_}\mathrm{Distance}$ needs adjustment and optimization according to constraints, as shown in Equation (29), where $d_{ij}=\infty$ indicates that terminal node i is unreachable from terminal node j.

$$\begin{array}{c}{d}_{ij}=\left\{\begin{array}{c}\infty ,i=j\hfill \\ \infty ,d_{in}<d_{jn}\hfill \end{array}\right.\end{array}$$

(29)

From the distance matrix, the link loss matrix $\mathrm{link}\text{\_}\mathrm{Loss}$ between terminal nodes can be derived, as shown in Equation (30), where $l_{ij}$ represents the path loss between terminal node i and terminal node j.

$$\begin{array}{c}\mathrm{link}\text{\_}\mathrm{Loss}={\left(\begin{array}{ccc}{l}_{11}& \cdots & l_{1n}\\ ⋮& \ddots & ⋮\\ l_{n1}& \cdots & l_{nn}\end{array}\right)}_{n\times n}\end{array}$$

(30)

Similarly, adjustments and optimizations are made to the link loss matrix $\mathrm{link}\text{\_}\mathrm{Loss}$ to prevent loops or improper routing between terminal nodes and the aggregation node, adhering to constraints shown in Equation (31), where $l_{ij}=\infty$ denotes that terminal node i is unreachable from terminal node j.

$$\begin{array}{c}{l}_{ij}=\left\{\begin{array}{c}\infty ,i=j\hfill \\ \infty ,d_{in}<d_{jn}\hfill \end{array}\right.\end{array}$$

(31)

The adjacency matrix $\mathrm{node}\text{\_}\mathrm{Adjoin}$ is a three-dimensional matrix, as shown in Equation (32), representing adjacency for different SFs of (7, 8, 9, 10, 11, 12).

$$\begin{array}{c}\mathrm{node}\text{\_}\mathrm{Adjoin}={\left(\begin{array}{ccc}{a}_{11}& \cdots & a_{1n}\\ ⋮& \ddots & ⋮\\ a_{n1}& \cdots & a_{nn}\end{array}\right)}_{n\times n\times 6}\end{array}$$

(32)

To ensure that selected relay nodes fall within the transmission range of terminal nodes under the chosen $SF$, while also preventing overly close distances between the relay and its own nodes, an appropriate distance threshold needs to be set. The constraints are shown in Equation (33), where $R_{SF}$ represents the single-hop transmission distance for different $SF\mathrm{s}$, and $a_{ij}=1$ indicates that node j can serve as a relay for node i; otherwise, $a_{ij}=0$.

$$\begin{array}{c}{a}_{ij}=\left\{\begin{array}{c}0,a_{ij}<(1-\mu )R_{SF}\hfill \\ 1,(1-\mu )R_{SF}\le a_{ij}\le R_{SF}\hfill \\ 0,a_{ij}>R_{SF}\hfill \end{array}\right.\end{array}$$

(33)

After establishing the adjacency matrix $\mathrm{node}\text{\_}\mathrm{Adjoin}$, all possible bidirectional links for terminal nodes under different $SF\mathrm{s}$ can be determined. Using a distributed network model within a large building as an example, the simulation model consists of 140 terminal nodes and one gateway distributed within a space of 100 m × 50 m × 18 m, simulating the characteristics of a multi-layer large building. When SF = 7 is selected, the adjacency matrix under constraint conditions provides a feasible link diagram, as shown in Figure 7.

To construct the link weight matrix $\mathrm{D}\text{\_}\mathrm{W}$, which represents the impact of communication distances between nodes on the selection of relay nodes, the link loss matrix $\mathrm{link}\text{\_}\mathrm{Loss}$ is element-wise multiplied by the adjacency matrix $\mathrm{node}\text{\_}\mathrm{Adjoin}$ under SF = k. This results in the link weight matrix $\mathrm{D}\text{\_}\mathrm{W}$ corresponding to SF = k, as shown in Equation (34).

$$\begin{array}{c}\mathrm{D}\text{\_}\mathrm{W}=\mathrm{link}\text{\_}\mathrm{Loss}\ast {\mathrm{node}\text{\_}\mathrm{Adjoin}}_{SF=k}\end{array}$$

(34)

2.

Construction of the load matrix

The load matrix $\mathrm{C}\text{\_}\mathrm{W}$ is constructed to represent the impact of node load conditions on the selection of relay nodes, as shown in Equation (35). Here, $B_{S_i}$ represents the maximum amount of business data that node $S_i$ can handle, and $C_{S_i}$ represents the load rate of the node. It is assumed that the load capacity of the aggregation node is infinite, i.e., the cost of business data from terminal nodes to the aggregation node is 0.

$$\begin{array}{c}{C}_{-}\mathrm{W}={\left(\begin{array}{ccc}{\Delta }_1\hfill & \cdots \hfill & {\Delta }_n\hfill \end{array}\right)}_{1\times n}\end{array}$$

(35)

$$\begin{array}{c}{\Delta }_k={\displaystyle \frac{1}{B_{S_i}\times \left(1-C_{S_i}\right)}}\end{array}$$

(36)

3.

Construction of the Load Cost Matrix

For routing optimization of terminal nodes of different business types, it is necessary to comprehensively consider the communication distance between nodes, link loss, and load conditions of each node. The Dijkstra algorithm calculates and generates node routing information, thus finding the minimum load cost transmission path for terminal nodes of different business types.

The load cost matrix $\mathrm{T}\text{\_}\mathrm{W}$ is established by first repeating the load matrix $\mathrm{C}\text{\_}\mathrm{W}$ n times along the row direction to generate a new n × n matrix. This matrix is then element-wise multiplied by the link weight matrix $D_{-}{W}_{SF=k}$, as shown in Equation (37).

$$\begin{array}{c}{\mathrm{T}}_{-}\mathrm{W}=repeat\left(\mathrm{C}\text{\_}\mathrm{W},\mathrm{n},1\right)\ast {\mathrm{D}}_{-}{\mathrm{W}}_{SF=k}\end{array}$$

(37)

5. Implementation Method of Load-Balanced Routing Optimization

5.1. Route Establishment Process

In route establishment, nodes need to collect and compute link states and neighbor information, adding this information to the HELLO packets. Nodes periodically broadcast these packets within their one-hop radius. Nodes can establish a neighbor relationship table upon receiving HELLO messages from neighboring nodes. Each node selects a subset of nodes as MPR nodes based on the MPR selection algorithm and maintains an $\mathrm{MPR}\text{\_}\mathrm{Selector}$ table. Nodes only forward broadcast messages from senders that are included in their $\mathrm{MPR}\text{\_}\mathrm{Selector}$ set. Upon receiving measured information from neighbor nodes, MPR nodes add this information to TC packets and periodically broadcast them network-wide, allowing distributed collection and maintenance of the network’s topology information. Nodes in the network acquire complete topology control information and the load condition of each node, thereby calculating each node’s load cost matrix. The Dijkstra algorithm is employed during route updates to compute and generate the minimum cost paths for nodes [31].

At the initial stage, each node initializes its load cost matrix to an all-zero matrix. Nodes extract topology control information from TC packets sent by neighbor nodes, including the destination addresses, the previous-hop address in the packet, communication distances, data transmission rates, and the load of terminal nodes. The node load matrix is calculated according to Equation (36), and the load cost matrix of the node is updated according to Equation (37). Through this process, the load cost to reach nodes in the $\mathrm{MPR}\text{\_}\mathrm{Selector}$ set from the current node, with the previous hop being a neighbor, is updated. Upon completion, if the current node is an MPR node, the previous hop address is set to the node itself, and this TC packet is broadcasted network-wide. After a period, the load cost matrices for reaching other nodes in the network are updated. This approach ensures that the network dynamically adjusts the routing based on load conditions, thereby optimizing load distribution and improving network performance.The routing establishment process is shown in Figure 8.

5.2. Routing Optimization Process

In establishing routes, a node load cost matrix has been developed to address the issue of specific end nodes in LoRa distributed multi-hop networks bearing excessive load. Nodes with an occupancy rate of 1 in the network will be removed during the route optimization process. No new requests or tasks will be assigned to these nodes, and the new traffic will be redirected to nodes with lower occupancy rates, dynamically adjusting resource allocation.The routing optimization process is shown in Figure 9.

Algorithm 2 outlines the steps involved in the routing optimization process.

Algorithm 2 Routing optimization process.

Input: End node business type, end node location information Output: Optimal transmission path Step1: Aggregating nodes’ broadcast route requests to obtain network-wide topology information; Step2: Obtain the business type and location information of end nodes with transmission tasks, initialize node weight matrix, and determine the priority order for end node route planning; Step3: Execute the following process based on the priority order determined in Step2 for each $i\in [1,\mathrm{number}\mathrm{of}\mathrm{business}\mathrm{nodes}]$ do    if $\mathrm{there}\mathrm{are}\mathrm{end}\mathrm{nodes}\mathrm{to}\mathrm{be}\mathrm{planned}$ then      go to Step4    else      routing optimization    end if end for Step4: Obtain the end nodes for which transmission paths need to be planned, calculate SF based on communication delay and link reliability constraints; Step5: With the calculated SF from Step4, determine the transmission range of end nodes, calculate the link loss matrix and adjacency matrix of end nodes, and establish a load cost matrix considering the load situation of end nodes in the network; Step6: Utilize Dijkstra’s shortest path algorithm to plan the transmission path for end nodes to reach the gateway; Step7: Update the load situation of all end nodes in the network if $\mathrm{end}\mathrm{nodes}\mathrm{meet}\mathrm{the}\mathrm{load}\mathrm{rate}\mathrm{constraint}$ then    go to Step8 else    delete the path and search for alternative paths end if Step8: Remove end nodes with an occupancy rate of 1, return to Step3 for end node re-planning of transmission paths; return Optimal transmission path.

6. Simulation Verification

The experiment involved constructing an LB-OLSR protocol model on a simulation platform and comparing it with the traditional OLSR protocol to validate the effectiveness of the improvements.

6.1. Simulation Parameter Settings

A simulation scenario with 200 end nodes and 1 gateway distributed in a 100 m × 50 m × 18 m space was set up in the simulation software, with specific simulation parameters, as shown in

Table 3. Simulation parameter settings.

Parameter Name	Parameter Value
Scenario spatial size	100 × 50 × 18
Number of end nodes	60–200
Number of aggregators	1
Number of simulations	20
BW/kHz	125
SF	7–12
Coding rate CR	4/5
Hello message interval/s	5
TC message interval/s	8

.

The business types of terminal nodes in large buildings are simplified to simulate business characteristics in a stratified large-building scenario. The specific parameters for business types are set as shown in

Table 4. Parameter settings for terminal node business types.

Business Type	Transmission Delay	Reliability	Success Rate	Business Proportion	Payload	Priority
Anomaly alarm	Low	High	98%	5%	5 Byte	1
Control	Low	High	95%	20%	10 Byte	2
Energy monitoring	Medium	Medium	90%	40%	20 Byte	3
Environmental monitoring	Medium	Medium	80%	30%	20 Byte	4
Other business	High	Medium	80%	5%	10 Byte	5

.

In the simulation experiments of fixed-layout scenarios, where the number and locations of all nodes are fixed, business nodes are randomly generated for each simulation group. The experiments involve 20 groups, and after the end of each simulation, the load balance of the nodes is calculated. In the simulation experiments for fixed-node-number scenarios, 20 groups of simulations are conducted, with the locations of nodes randomly changed within each simulation group. Each simulation group involves 20 repetitions based on the fixed-layout scenario, and the average result of these 20 repetitions is considered the outcome. In mixed-scenario simulations, where the number of terminal nodes ranges from 60 to 200, simulations with fixed node numbers are conducted for every 20 additional terminal nodes. This approach ensures changes in both the number and positions of nodes in the network.

6.2. Analysis of Simulation Results

In the fixed-layout scenario simulations, the load balance of terminal nodes within different groups is illustrated in Figure 10. The optimization degree of node load balance varies across different simulation groups due to terminal nodes’ differing numbers and positions carrying monitoring tasks. When the number of business nodes is large, more terminal nodes in the network act as relay nodes to transmit business data, thereby improving the load balance optimization degree of the LB-OLSR algorithm. In fixed-node-number scenario simulations, it is observed that the load balance achieved using the LB-OLSR algorithm is superior to that of the OLSR algorithm across different groups. This indicates that the load-balancing routing optimization method proposed in this study significantly improves load balance.

In the fixed-node-number scenarios, the load balancing of terminal nodes across different groups is shown in Figure 11. Random changes in node positions are made within each simulation group based on the fixed-layout scenario, with 20 simulations conducted to obtain the average result. This approach mitigates the impact of randomly generated business nodes and randomly distributed terminal nodes on the experimental results, thus improving result reliability. In these simulations, the load balancing of nodes utilizing the LB-OLSR algorithm remains stable between 0.2 and 0.4, indicating superior performance compared to the OLSR algorithm, which fluctuates between 0.7 and 0.9. The impact of terminal node position distribution leads to variations in load balancing, but overall, the load balancing tends towards stability.

In the fixed-node-number scenarios, the remaining energy balance of nodes achieved by the LB-OLSR algorithm is significantly lower than that of the OLSR algorithm, indicating that the proposed method in this section results in a more balanced distribution of residual energy among terminal nodes. In various experimental groups, fluctuations in the balance of node residual energy were observed, which is likely attributed to the random distribution of terminal node positions.The experimental results are shown in Figure 12.

In the mixed-scenario simulation experiment, 20 terminal nodes are incrementally added to the fixed-node-count scenario, and 20 simulations are performed for each increment. The average result is taken to mitigate the impact of randomly generated business nodes and the random distribution of terminal nodes on the experimental results. In these simulations, the node load balancing under different terminal node counts is illustrated in Figure 13. The load balancing of nodes using the LB-OLSR algorithm gradually decreases and stabilizes as the number of terminal nodes increases. Conversely, the load balancing in the OLSR algorithm mostly remains between 0.9 and 1, showing no significant fluctuation with an increasing number of terminal nodes. When the number of terminal nodes is low in large-scale buildings, their spatial distribution is sparse, and the distance between terminal nodes is large. Consequently, the choice of relay nodes to transmit business data from the business nodes to the gateway is limited, significantly impacting the load balancing of the LB-OLSR algorithm. However, when the number of terminal nodes is high, the spatial distribution becomes dense, and the distance between nodes is shorter, providing ample relay nodes. This reduces the impact on route optimization.

In the mixed-scenario simulation experiment, the remaining energy balance of the nodes in the LB-OLSR algorithm gradually stabilizes at 0.65. This stability is attributed to the dense spatial distribution of terminal nodes within large buildings when the number of terminal nodes increases beyond a certain threshold, resulting in an ample selection of relay nodes and thereby reducing the impact on routing planning. Conversely, in the OLSR algorithm, as the number of terminal nodes increases from 60 to 120, the failure to consider node load and occupancy rates during the planning of transmission paths for service nodes leads to excessive loads on certain nodes, resulting in a noticeable increase in the remaining energy balance among the nodes.The experimental results are shown in Figure 14.

The average node load duration under different terminal node counts is depicted in Figure 15. Compared to the OLSR algorithm, the average node load duration is longer under the LB-OLSR algorithm. This is because the algorithm focuses on optimizing the nodes’ average load rate and load balancing. As a result, the chosen transmission paths to the gateway may not always have the most minor hops or the shortest distance, which can increase the number of times business data are forwarded. On the other hand, the OLSR algorithm uses the minimum hop count as the routing criterion, resulting in a shorter average node load duration. As the number of terminal nodes increases, more relay node choices become available for business data forwarding, potentially reducing hop count and shortening transmission paths, thereby decreasing the average node load duration.

The average business delay under different terminal node counts is shown in Figure 16. The LB-OLSR algorithm results in a shorter average business delay than the OLSR algorithm. The proposed load-balancing algorithm comprehensively considers communication delay, link reliability, node occupancy rate, and node load rate. It dynamically adjusts traffic distribution strategies based on node load conditions and routing performance within the network, thereby avoiding congestion and node load imbalance and reducing queuing delays for business data. As the number of terminal nodes increases, the load-balancing algorithm can more effectively utilize these nodes to share the load, accelerating business data processing and reducing business transmission delays.

6.3. Discussion

The simulation results presented above serve as conceptual validation, laying a solid foundation for the application of the described LoRa distributed routing protocol in real-world scenarios. This paper focuses on simulation analysis specifically for indoor environments such as large buildings, verifying the performance of the improved algorithm through simulations.

In practical applications, the LB-OLSR protocol can be effectively integrated into existing Internet of Things (IoT) deployments. To achieve this, it is essential to ensure that the hardware devices used can support the functionalities of the protocol. These devices should possess good signal strength and low power consumption characteristics. Additionally, environmental factors may pose challenges to signal transmission in real deployments.

Firstly, physical obstacles can significantly impact the propagation of LoRa signals; the height and density of buildings and trees can cause signal attenuation when penetrating obstacles, potentially leading to data loss or delays. Secondly, climatic conditions, such as rainfall or temperature fluctuations, may affect the speed and quality of signal propagation. Electromagnetic interference is also a major influencing factor; in urban or industrial environments, other wireless devices may generate electromagnetic interference that affects the quality of LoRa signals. Furthermore, in high-density network environments, simultaneous data transmission from multiple nodes may lead to channel congestion, increasing the collision rate of data packets and thereby impacting transmission efficiency. The LB-OLSR protocol enhances the adaptability and robustness of the network through load-balancing strategies, but the influence of environmental factors must still be considered in practical applications.

To effectively integrate the LoRa distributed network load-balancing routing protocol into existing Internet of Things (IoT) deployments, a series of steps must be undertaken. The first step involves a thorough demand analysis, which includes assessing the current LoRa network architecture, determining the number of devices, data traffic, and coverage area, and identifying bottlenecks and load-balancing requirements within the network. Subsequently, the protocol’s scalability, fault tolerance, and support for low-power devices should be considered. Following this, a well-structured LoRa network topology must be designed to ensure the optimal distribution of gateways and nodes, thereby enhancing signal coverage and data transmission. It is also essential to establish data flow directions and routing strategies to achieve effective load balancing. In terms of hardware selection, it is necessary to choose hardware devices that support LoRa technology, ensuring they possess adequate processing capabilities and communication ranges. The use of edge computing devices may also be considered to reduce latency and enhance data processing capabilities. Regarding software configuration, network management software can be employed to monitor traffic and performance, allowing for timely adjustments to load-balancing strategies. In summary, the effective integration of this protocol into existing IoT deployments necessitates a comprehensive evaluation of the current network architecture, the design of a rational network topology, and the selection of high-performance gateways and node devices that support LoRa technology. Additionally, appropriate software must be configured to monitor traffic and performance, followed by small-scale testing and optimization to ensure stability under high-load conditions. Ultimately, through regular maintenance and updates to the devices, the protocol can be effectively integrated into the existing IoT deployment, thereby enhancing the overall efficiency and reliability of the network.

From the user’s perspective, the LB-OLSR protocol offers multiple benefits, optimizing route selection and reducing redundant multiple forwarding nodes to ensure that users can receive the required data in a timely manner. The LB-OLSR protocol intelligently adjusts data transmission strategies based on network load conditions, avoiding network congestion and thereby improving the overall response speed. By reducing redundant control messages and optimizing data transmission paths, the LB-OLSR protocol can lower energy consumption within the network, which is particularly important for battery-powered LoRa nodes, extending the lifespan of the devices. These advantages make the LB-OLSR protocol more attractive for practical applications, meeting users’ demands for efficient, stable, and secure IoT communication.

7. Conclusions

This paper proposes the LB-OLSR protocol as an ideal method for constructing LoRa distributed multi-hop networks. The OLSR protocol faces issues such as redundant MPR nodes, which increase control message flooding and consume limited channel resources. To address this, we propose a connectivity necessity-based MPR selection algorithm to reduce redundant nodes within the MPR set. The traditional OLSR protocol uses the minimum hop count as the routing criterion, which is inadequate for scenarios with heavy traffic or uneven network conditions. Our approach considers various communication requirements by incorporating communication delay, link reliability, node occupancy rate, and node load rate as constraints. Business nodes are prioritized accordingly, and load cost matrices are used to plan transmission paths that meet these constraints. Our simulation results indicate that the improved LB-OLSR protocol achieves a more balanced load among terminal nodes in the network compared to the traditional OLSR protocol, enhancing traffic distribution and resulting in reduced average service delay. In future research efforts, network performance can be optimized through metrics such as network overhead, jitter, and packet loss, particularly in scenarios where terminal nodes experience high loads and have limited energy. Enhancing the longevity of terminal nodes to achieve more balanced residual energy across the network becomes particularly crucial. To accomplish these research objectives, it will be necessary to develop more advanced routing optimization algorithms to address increasingly complex routing optimization challenges.