LoRa enables low-power, long-range wireless communication due to the simple modulation of the special modulation scheme and pure ALOHA. However, this structure not only has advantages but also disadvantages. There is no carrier sensing for packets sent from other terminal nodes, and the message transmission time due to high SF increases the probability of packet collision as the number of terminals increases.
LoRaWAN has this problem, so it provides adaptive data rate (ADR) to solve the above problems. ADR can control symbol duration, data rate, and energy consumption to minimize side effects as the number of LoRa terminals increases. As shown in Algorithm 1, the ADR collects approximately 20 packets initially and then analyzes the received packet strength to determine the SF increase/transmission power (TP) increase. By adjusting SF and TP properly, near-far problem can be solved by minimizing packet collision and increasing the strength of transmission signal to the node far from gateway [
3].
The ADR uses the signal-to-noise ratio (SNR) of the collected packets to determine SF and TP. The ADR varies greatly depending on the number of packets collected and how to set the representative values. As a result, much research has been conducted to maximize the performance of ADR. The research by Hauser et al. [
4] changed the existing ADR logic to make network resource distribution more efficient. Reynders et al. [
5] significantly reduced the retransmission rate of packets by adding logic to change the channel according to the distance between the gateway and the terminal node to the ADR. There are several characteristics that can lead to an unbalanced distribution of data extraction rate (DER) between nodes. Since SF does not have perfect orthogonality, it is not possible to detect weak packet signals due to strong packet signals. For this reason, packets of nodes with large signal attenuation are not delivered well. In view of these issues, Abdelfadeel et al. [
6] proposed the Fair Adaptive Data Rate Algorithm (FADR). FADR is an ADR in which LoRa achieves an effective data extraction rate with an appropriate spreading factor and transmission power while excluding excessively high transmission power. Kim et al. [
7] suggested congestion estimation through logistic regression to improve the transmission performance degradation of LoraWAN without considering congestion. They proposed a congestion classifier based on logistic regression, which improves performance in terms of transmission delay. When using the same data rate and not considering collision, throughput is greatly reduced. Therefore, Kim and Yoo [
8] proposed a contention-aware adaptive data rate for throughput optimization and applied a gradient projection method to find the optimal value. Through this, the throughput was improved to a value close to the theoretical maximum value. El-Aasser et al. [
9] proposes the algorithms of sensitivitySF and assignmentSF, which are smarter SF setting techniques than the existing ADR, to increase both the throughput between the nodes of the same tier and the overall throughput, and achieve higher throughput than the existing ADR. For fair SF allocation, Cuomo et al. [
10] proposed ordered water-filling-based EXPLoRa-KM (
K-means) and EXPLoRa-TS (time symbol) techniques. KM mitigates the critical region where the collision occurs seriously, and TS plays a role to make the traffic load of each SF constant. Through this, fair allocation was achieved. Zhou et al. [
11] proposed Data Rate and Channel Control (DRCC) to support massive number of LoRa nodes and improve resource utilization. The channel was evaluated based on data extraction rate (DER), SF allocation was performed based on DER, and load balancing of the channel was performed in consideration of packet collision according to node density. Through this, it was proved that the proposed technique is excellent in dense deployment scenarios. Ta et al. [
12] derived optimized SF for fair resource allocation using Exponential Weights for Exploration and Exploitation (EXP3) algorithm to support large-scale LoRa nodes. Through the simulation using real data such as non-uniform node distribution and inter-spreading factor, collision was reflected, and as a result, fair resource allocation was achieved.
As such, we are conducting advanced research to modify the existing ADR or add technologies such as machine learning for efficient and reliable usage of network resources.