Adaptive Data Rate Enhancement for Long Range Wide Area Network with Dynamic Margin

Long Range Wide Area Network (LoRaWAN) is widely recognized as a leading solution within the Low Power Wide Area Network (LPWAN) landscape, offering a scalable and energy-efficient framework for long-range communication [1, 2]. However, in dense network scenarios, its performance is often compromised by high contention and stochastic channel variations, which lead to reduced packet delivery ratios (PDRs) and increased energy consumption [3, 4].

The performance and longevity of LoRaWAN devices are fundamentally governed by transmission parameters. Key among these is the Spreading Factor (SF), which defines the chips-per-symbol ratio, effectively balancing data throughput against signal robustness. Transmission Power (TP) dictates signal intensity, directly influencing both coverage range and energy consumption. Furthermore, Bandwidth (BW) determines channel capacity and data rate, while the Coding Rate (CR) enhances error correction capabilities to mitigate interference. Therefore, the optimal configuration of these parameters is critical to achieving a balanced trade-off between communication range, reliability, and energy efficiency [5-7].

The dynamic optimization of transmission parameters based on Channel State Information (CSI) has become a primary focus in various modern wireless communication systems [8]. However, in the context of LoRaWAN, the standard Adaptive Data Rate (ADR) mechanism often relies on static parameters that fail to adapt to rapid channel fluctuations. The LoRaWAN specification defines the ADR mechanism to optimize transmission parameters. While standard ADR implementations demonstrate robustness in stable environments, they suffer from a critical rigidity due to the use of a fixed 10 dB safety margin, which often leads to suboptimal resource allocation in dynamic conditions [9, 10]. This reliance on a static margin creates a fundamental trade-off that standard ADR fails to resolve.

In stable channel conditions, the fixed 10 dB margin is often excessively conservative, compelling devices to utilize higher SF or TP than required, which results in unnecessary energy consumption and airtime wastage [11]. Conversely, in volatile channels prone to bursty interference or shadowing, this fixed margin may prove inadequate, leading to link instability and significant packet loss [12]. Consequently, the lack of an adaptive safety margin mechanism in standard ADR constitutes a critical deficiency in optimizing dense, stationary LoRaWAN deployments.

In this context, this paper proposes the Dynamic Margin Adaptive Data Rate (DM-ADR). This mechanism replaces the fixed safety margin with a bounded dynamic margin derived directly from the standard deviation of recent uplink Signal-to-Noise Ratio (SNR) samples. This approach allows the network server to explicitly capture fluctuations caused by the environment by expanding the margin during volatile periods to prevent packet loss and reducing it during stable periods to conserve energy. We implemented DM-ADR in the Network Simulator 3 (NS-3) network simulator and evaluated it against standard ADR and representative schemes. The results demonstrate that alignment of the safety margin with observed SNR variability significantly improves the trade-off between reliability and energy efficiency in large-scale deployments.

The remainder of this paper is organized as follows: Section 2 reviews the related work, while Section 3 provides an overview of the standard ADR in LoRaWAN. Section 4 details the proposed DM-ADR method, followed by the description of the simulation configuration in Section 5. Section 6 presents the results and discussion, and Section 7 concludes the paper and outlines future work.

LoRaWAN is a standard LPWAN protocol designed to support long-range communication with low power consumption for IoT applications. The network architecture operates on a star-of-stars topology composed of three main elements: end-devices, gateways, and back-end servers [19]. Unlike simple point-to-point configurations, LoRaWAN utilizes a star topology where the gateway centrally receives chirp signals, relieving sensor nodes from the burden of routing processes. This architecture not only ensures superior energy efficiency but also facilitates end-to-end data security mechanisms (such as Advanced Encryption Standard (AES)) and centralized network management using platforms like Chirpstack, making it an ideal standard for disaster mitigation in remote regions [20].

The physical layer of LoRaWAN utilizes LoRa modulation, a proprietary spread spectrum technology developed by Semtech [21]. The performance of a LoRa link is governed by critical transmission parameters, primarily the SF, TP, and BW. The SF ranges from 7 to 12, while TP typically varies between 2 dBm and 14 dBm. There is an inherent trade-off in configuring these parameters: utilizing a high SF and high TP extends the transmission range up to 15 km in suburban areas but results in a decreased data rate and significantly higher energy consumption. Conversely, lower SF and TP settings yield higher data rates and reduced power consumption but shorten the effective communication range [22].

ADR is a critical mechanism within the LoRaWAN protocol designed to optimize network capacity and energy efficiency by dynamically adjusting transmission parameters, primarily the SF and TP. In the standard LoRaWAN ADR mechanism, the process of adjusting the SF and TP is based on the maximum SNR value of the uplink packets received by the gateway. The network server then compares this maximum SNR value with the minimum required SNR for a specific SF, according to the sensitivity table for the LoRa module. The standard ADR algorithm can be seen in the following Algorithm 1 [11]:

$SNR_{max}$ is the maximum value of the last 20 SNR values received by the gateway from a LoRaWAN device. This value represents the best signal quality achieved within a specific timeframe, typically during the last 20 uplink transmissions. On the other hand, the SNR requirement ($SNR_{req}$) is the minimum SNR value needed for a LoRa signal to be received and demodulated correctly at a specific SF. This value indicates the receiver's sensitivity threshold. The larger the SF, the lower the $SNR_{req}$, indicating the system is more resistant to noise but has a slower data rate. The required SNR values are usually already defined based on the results of physical LoRaWAN testing, as follows (for 125 kHz BW) in Table 1. The resulting margin is translated into an integer number of adaptation steps, $N_{step}=int\left[S N R_{margin}/3\right]$, following the ADR convention that approximately 3 dB corresponds to one control step. When $N_{step}$ is positive, the link has excess margin, and the algorithm first increases the data rate by decrementing the SF step by step until the minimum SF is reached or the steps are exhausted. Any remaining positive steps are then applied to reduce TP in 3 dB decrements to improve energy efficiency. When $N_{step}$ is negative, the link margin is insufficient, and the algorithm increases TP in 3 dB increments, up to $T P_{max }$, to restore reliability. The algorithm outputs the updated SF and TP values for subsequent transmissions.

Table 1. Required Signal-to-Noise Ratio (SNR) of each Spreading Factor (SF) [23]

In the standard ADR mechanism in LoRaWAN, the margin value is fixed at 10 dB as part of the link budget calculation to maintain communication reliability. This margin serves as a buffer or safety factor against various conditions that can degrade signal quality, such as fading, inter-device interference, environmental variations, or differences in receiver sensitivity in the real world compared to theoretical values. In other words, even though the device's SNR is already high enough to support a certain SF, the system still reduces it by 10 dB to allow for tolerance if channel conditions suddenly worsen.

A comparative evaluation of the proposed DM-ADR against the baseline Standard ADR was conducted within a discrete-event simulation environment using NS-3. The implementation was realized through custom C++ simulation scripts built on top of the widely used LoRaWAN module for NS-3, which provides an end-to-end LoRaWAN stack (PHY, MAC, and network-side control) suitable for reproducible ADR studies. These simulations employed a standard LoRaWAN network module [24, 25].

The simulation parameters used can be seen in Table 2, which includes physical configuration aspects and MAC, as well as simulation environment parameters such as the number of devices, data transmission period, and simulation duration. All experiments were conducted in the EU868 band over a 5000 × 5000 m deployment with a single gateway and 100 to 700 end devices. Uplink traffic is generated periodically every 10 minutes using a 20-byte payload, and each run lasts 2 days to capture steady-state ADR behaviour under sustained channel contention.

Table 2. Simulation setup

To validate the proposed method's adaptiveness to realistic channel variations, the simulation environment utilizes the Log-Normal Shadowing model, with parameters rigorously aligned with the benchmark study by Slabicki et al. [11]. By adopting the propagation configuration and stochastic parameters defined in this established framework, we ensure that the simulated SNR fluctuations are statistically consistent with actual field environments. This approach confirms the method's robustness against realistic channel impairments without the need for a separate physical measurement campaign. This composite model accounts for both the deterministic signal attenuation due to distance and the stochastic variations caused by obstacles in the propagation path. The path loss $P L(d)$ at a distance d from the gateway is expressed as:

$P L(d)=P L\left(d_0\right)+10 n \log _{10}\left(\frac{d}{d_0}\right)+X_\sigma$                  (4)

where, $P L\left(d_0\right)$ represents the path loss at a reference distance $d_0$ and $n$ denotes the path loss exponent, which characterizes the signal decay rate specific to the environment (e.g., urban or rural). To simulate the irregularity of the wireless channel and the presence of obstacles (shadowing), a zero-mean Gaussian random variable $X_\sigma$ with standard deviation $\sigma$ (in dB) is introduced. The parameters are set to $n=2.32$ and $P L\left(d_0\right)=128.95$ [11], consistent with the log-distance formulation available in NS-3 propagation-loss modelling. Finally, to reduce randomness and improve statistical confidence, each scenario is repeated for 5 independent iterations, and the reported results are averaged across runs.

The energy consumption of LoRaWAN end devices in this study is modelled by considering the three main operational states of the radio transceiver: transmission (TX), reception (RX), and standby or sleep. The energy consumed in each state is defined as the product of the supply voltage, the current drawn, and the duration of the state, expressed as:

$E=V \times I \times t$                   (5)

where, V is the supply voltage, I is the current in amperes, and t is the duration in seconds. The total energy consumed by a node over the course of the simulation is therefore given by:

$E_{{total }}=E_{T X}+E_{R X}+E_{ {standby }}$                (6)

Transmission energy is primarily determined by the configured TP and the time-on-air (ToA) of each packet. The ToA is computed according to the LoRa physical layer parameters, specifically the SF, BW, CR, and payload size (PL). The symbol duration is defined as

$T_{s y m}=\frac{2^{S F}}{B W}$                         (7)

$\begin{aligned} { ToA } & =\left(n_{ {preamble }}+4.25\right) \times T_{ {sym }} +N_{ {payload }} \times T_{ {sym }}\end{aligned}$             (8)

where, $n_{{preamble }}=8$ (default in LoRaWAN), and $N_{ {payload }}$ is the number of payload symbols derived from the LoRa specification. Thus, the energy consumed during transmission can be calculated as:

$E_{T X}=V \times I_{T X} \times T o A$                  (9)

This formulation ensures that higher SFs (longer ToA) and higher transmit powers both result in higher energy costs, reflecting the reliability–lifetime trade-off in LoRaWAN. After each uplink, a Class A LoRaWAN end device opens two receive windows (RX1 and RX2). Even if no downlink packet is received, the device remains in reception mode, consuming a constant current for the duration of these windows. The reception energy is therefore modeled as:

$E_{R X}=V \times I_{R X} \times\left(t_{R X 1}+t_{R X 2}\right)$                  (10)

where, $I_{R X}$ is the current consumption in reception mode, and $t_{R X 1}$, $t_{R X 2}$ are the durations of the two windows. When not transmitting or receiving, the device remains in standby or sleep mode, drawing a minimal current $I_{ {standby}}$. Although this contribution is small in high-traffic scenarios, it becomes dominant when devices send packets infrequently. The standby energy is given by:

$E_{ {standby }}=V \times I_{ {standby }} \times t_{{standby }}$                  (11)

At the beginning of the simulation, each device is initialized with a fixed energy budget of 10,000 J. During runtime, the simulator continuously subtracts the consumed energy from this budget according to the device state. From this, we derive the average energy consumption per device, the total energy consumed by the network, and the percentage of energy depletion. This framework allows us to evaluate the trade-offs between link reliability and energy efficiency across different ADR algorithms.

This study presented DM-ADR, a dynamic margin adaptation scheme designed to overcome the rigidity of the fixed safety margin inherent in the Standard LoRaWAN ADR. By leveraging short-term SNR variability to adjust the link budget proactively, the proposed method balances reliability and energy efficiency in high-density scenarios. Our simulation-based evaluation using the NS-3 framework demonstrates that DM-ADR delivers significant performance gains under the conditions tested. In a dense network comprising 700 nodes, the proposed algorithm improved the PDR by up to 23.10 percentage points and reduced the average energy consumption per node by approximately 52% compared to the Standard ADR baseline. These improvements are particularly evident in environments characterized by high shadowing variability, where (σ = 6). In such scenarios, DM-ADR maintained a PDR above 90%, whereas the performance of Standard ADR degraded significantly.

Future work will focus on extending the DM-ADR framework to multi-gateway architectures and mobile LoRaWAN nodes. Additionally, we intend to validate the algorithm's performance through real-world testbed experiments to confirm its robustness under heterogeneous traffic patterns and unpredictable physical-layer impairments.