Adaptive contextual memory network for enhanced communication and efficiency in the internet of underwater things

Experimental setup

To improve reproducibility, all key parameters used in the simulation are summarized in Table 2. These parameters define the conditions under which the ACMN–AMO–EARL framework was tested.

These settings represent realistic underwater conditions, ensuring consistency across all experiments.

Simulations were conducted with a MATLAB-built IoUT network model with realistic parameters, including path loss, noise, and bandwidth. The ACMN, AMO, and EARL frameworks were also tested under varying depths, interference levels, and network densities. The limitations of basic static modulation techniques and conventional routing algorithms were identified in comparison to earlier methods.

The simulation adopted a standard underwater acoustic channel model based on range-dependent attenuation and additive white Gaussian noise. A total of 100 sensor nodes were randomly deployed within a 2 km × 2 km × 500 m 3-D underwater area, with moderate node mobility (maximum speed = 1.5 m/s). The transmission frequency was set at 24 kHz with a bandwidth of 10 kHz, and the ambient noise level ranged from 10 to 40 dB, depending on the water depth.

The ACMN was trained using 10,000 time steps per run with a learning rate of 0.001, a batch size of 64, and the Adam optimizer. The EARL agent used a discount factor γ = 0.9, exploration rate ε = 0.1, and reward-weight parameters (w₁,w₂,w₃) = (0.5, 0.3, 0.2). Each experiment was repeated 20 independent runs using different random seeds to ensure statistical consistency, and the reported values represent the average across these runs.

For comparison, baseline methods were implemented under identical conditions:

• LSTM and GRU models for environmental prediction;

• Fixed BPSK modulation for signal transmission;

• Conventional Q-learning for routing optimization.

Parameter tuning for each baseline followed standard IEEE-IoUT configurations, ensuring a fair comparison with the proposed ACMN-AMO-EARL framework. The entire results of the simulation were calculated using mean and standard deviation values at ten independent runs to offer statistical consistency. Paired t-tests were used to make comparative assessments between the proposed ACMN-AMO-EARL and baseline models. It was observed that the improvements in communication reliability (p < 0.01), energy efficiency (p < 0.05), and real-time adaptability (p < 0.01) were statistically significant.

Communication reliability

The ACMN also provided a more significant capability to communicate with counterparts under varying underwater conditions reliably. Compared to conventional LSTM and GRU models, the ACMN reduced signal drop by 27%, thanks to its combined memory fusion and dynamic nature^23,24. The signal integrity was highly stable, with low variation fluctuating around the maximum level, even when noise levels rose to the variable level.

Energy efficiency

To analyze the energy consumption metrics, the effectiveness of the combined AMO and EARL frameworks was demonstrated. Due to this real-time modulation adjustment, AMO transmission reduced the transmission power by 35% compared to that of fixed-modulation systems. Moreover, for routing decisions, EARL revealed improved routing decisions that ensured an incremental improvement of 42% in overall network lifetime by avoiding energy wastage at routes²⁵. This approach ensures sustainability, thereby ensuring its effectiveness in long-term undertakings.

Real-time adaptability

The framework’s dynamic nature was investigated by applying environmental changes, such as sudden increases in noise level and bandwidth variations, during the process. The ACMN and AMO algorithms adapted within an average of 2 s with new optimal communication settings. This flexibility was superior to the baseline model, as other models required significantly longer to regain calibration.

Comparative analysis

The quantitative performance of the proposed framework has been compared with the baselines in Table 3. This system has been proven to enhance reliability, energy efficiency, and adaptability compared to the previously existing solutions²⁶. The information class is defined as follows: Metric ACMN, AMO, EARL Baseline. The communication reliability we achieved was 93%, which is higher than the average reliability of 72%. Energy Efficiency: 42% increase, 15% increase. Sensitivity time: 2 s, 7 s. Network Lifetime Extension: 35/18 summarizes the performance metrics of the proposed framework in comparison to baseline methods. The ACMN-AMO-EARL system consistently outperformed existing solutions in reliability, energy efficiency, and adaptability.

Sensitivity analysis

We evaluated the robustness of ACMN–AMO–EARL to variations in environment, model thresholds, and reinforcement-learning hyperparameters. Each configuration was run 20 independent times with distinct random seeds; we report the mean and 95% confidence intervals (error bars).

Setup.

• Environment sweeps: noise = 10–40 dB; depth = 50–500 m; node density = 50–200 nodes.

• Modulation thresholds: AMO SNR decision thresholds swept ± 20% around the default.

• RL hyper parameters.

Findings.

• Reliability vs. environment: Communication reliability remains 87–95% across noise and depth ranges; ACMN sustains a ~ 27% reduction in signal drop versus LSTM/GRU baselines.

• AMO vs. BPSK across noise: Power reduction stays 28–36% versus fixed BPSK while maintaining lower BER at high noise.

• EARL vs. Q-learning across RL grids: Routing-efficiency improvement + 38–44% and network-lifetime extension + 29–37% persist across (w1,w2,w3)(w_1,w_2,w_3)(w1​,w2​,w3​), learning-rate, and ϵepsilonϵ variations.

Ablation study and component impact analysis

To isolate the contribution of each component of the proposed framework, an ablation study was conducted by turning on and off the ACMN, AMO, and EARL modules independently. Table 4 describes the component impact analysis below.

The results show that each component enhances a specific dimension of system performance:

• ACMN improves communication reliability and environmental adaptability.

• AMO reduces bit error rates through dynamic modulation control.

• EARL optimizes routing paths and conserves energy.

The integrated framework achieves synergistic gains—up to 42% higher energy efficiency and 35% longer network lifetime—validating the necessity of combining all three mechanisms.

Sensitivity analysis

We evaluated the robustness of ACMN–AMO–EARL to variations in environment, model thresholds, and reinforcement-learning hyperparameters. Each configuration was run 20 independent times with distinct random seeds; we report the mean and 95% confidence intervals (error bars).

Setup.

• Environment sweeps: noise = 10–40 dB; depth = 50–500 m; node density = 50–200 nodes.

• Modulation thresholds: AMO SNR decision thresholds swept ± 20% around the default.

• RL hyper parameters: (w1, w2, w3) grid around (0.5, 0.3, 0.2); learning rate {5 × 10^− 4, 1 × 10^− 3, 2 × 10^− 3}; ε-greedy rates{0.05, 0.1, 0.2}.

Findings.

• Reliability vs. environment: Communication reliability remains 87–95% across noise and depth ranges; ACMN sustains a ~ 27% reduction in signal drop versus LSTM/GRU baselines.

• AMO vs. BPSK across noise: Power reduction stays 28–36% versus fixed BPSK while maintaining lower BER at high noise.

• EARL vs. Q-learning across RL grids: Routing-efficiency improvement + 38–44% and network-lifetime extension + 29–37% persist across (w1, w2, w3), learning-rate, and ε variations.

Advantages and limitations

The results underscore the proposed system’s robustness, scalability, and efficiency in addressing the unique challenges of underwater communication. However, the dependency on extensive training datasets for ACMN and the computational overhead of EARL could pose challenges in resource-constrained environments.

Table 5—The Two Environmental Scenarios and Communication Reliability sources of noise (nontechnical) Noise Characteristics (dB), Signal Interference (%), and Communication Efficiency (%) opposed framework against baseline methods. The ACMN-AMO-EARL system consistently outperformed existing solutions in reliability, energy efficiency, and adaptability.

These Tables 6, 7, 8 and 9 break down the various performance aspects of the proposed framework.

Figure 3 illustrates how communication reliability decreases as noise levels increase in the underwater environment.

Figure 4 illustrates the depth, latency, and packet loss, highlighting the challenges of underwater communication at greater depths.

Figure 5 Comparing the running time of the proposed framework and the baseline models can show how fast the proposed system is.

Figure 6 compares the proposed framework’s power reduction in kilowatts, routing efficiency in percentage, network lifetime improvement, and adaptive modulation versus traditional methods.

Figure 7 helps us understand that the deeper the area, the more energy is consumed, thus explaining inefficiency in target areas.

The Energy Efficiency Comparison reveals that the use of the ACMN-AMO-EARL framework is significantly more effective than using adaptive modulation only or traditional methods, resulting in a 35% reduction in transmission power requirements, a 42% increase in routing efficiency, and a 35% improvement in the network’s bootstrap lifetime. This demonstrates how the framework can achieve low energy consumption while effectively and reliably transmitting information^27,28. The Energy Consumption vs. Depth graph clearly shows a direct proportional relationship between energy consumption and depth on the robot as energy consumption increases from 0.8 joules at 50 m to 2.3 joules at 500 m. This implies the need for more effective and adaptable mechanisms to handle the increased resource requirements of deeper underwater operations. Combined, these graphs substantiate the energy efficiency envisioned in the framework and the functionality in extreme underwater environments. The Adaptive Contextual Memory Network, Adaptive Modulation Optimization, and Energy-Aware Reinforcement Learning framework are concepts at the forefront of Internet of Underwater Things research, particularly in addressing packet loss, which highlights the challenges of deeper underwater communication.

The Adaptive Contextual Memory Network, the Adaptive Modulation Optimization algorithm, and the Energy-Aware Reinforcement Learning framework represent a significant advancement in the Internet of Underwater Things domain. By mitigating fundamental problems, such as signal weakening, power consumption, and variability of underwater conditions, the proposed novel solution provides a more comprehensive solution than conventional ones.

Key Strengths.

1. 1.

   Robust Communication Reliability:

The ACMN’s mechanism for fusing memory inventory and real-time control can adapt flexibly within real-time communication. It boasts very high communication reliability, potentially reaching a reliability of 93% under noise conditions. This represents a significant enhancement over other memory networks, including LSTMs and GRUs, which lack this level of contextual integration.

1. 2.

   Energy Efficiency:

The integration between AMO and EARL reduces power consumption by 35% and increases routing efficiency by 42%, therefore increasing the network lifetime. To illustrate the amount of power consumed by each of the components, emphasizing its prospects for actual use in systems established in long-life-span environments where access to energy sources is often severely restricted.

1. 3.

   Real-Time Adaptability:

With a reaction time of under 2 s to environmental changes, the framework is well-suited to dynamic underwater scenarios, which are essential for the actualization of IoUT applications, such as marine monitoring and disaster management.

1. 4.

   Scalability:

Modularity enables the system to be extended to support a wide range of underwater operations, from near-shore to offshore, making it ideal for deployment in future IoUT missions.

Limitations

Although the suggested ACMN-AMO-EARL framework is highly reliable, adaptable, and energy-efficient in terms of communication, it still has several weaknesses.

Training data requirements: The Adaptive Contextual Memory Network relies on big and varied training using a variety of underwater conditions to maintain consistent prediction accuracy across different conditions. Such data, collected and labeled in real marine environments, is expensive and time-consuming to obtain.

Computational overhead: The Energy-Aware Reinforcement Learning model has a worst-case processing overhead, which can be problematic in real-time deployment on low-powered underwater sensor nodes. The deep Q-network needs optimization and pruning to be deployed practically.

Hardware dependency: AMO and EARL implementations require embedded processors with high performance and energy efficiency, such as those found in modems. However, these products are not as common as needed to support deep-sea or long-duration operations.

Future efforts will be directed towards solving these issues by co-optimizing hardware and software, utilizing lightweight network design, and integrating edge computing to make them scalable and practical in the real world.

Additionally, the obtained numerical improvements should be viewed within the framework of controlled MATLAB simulations rather than full-scale oceanic trials. Although the framework shows quantifiable gains, these values can fluctuate in actual marine settings due to unforeseen aspects such as salinity variations, equipment drift caused by pressure, and the lack of diverse training data. The conclusions have therefore been watered down to highlight possible applicability rather than established universality.

Revision Note on Results Accuracy:

Every finding of the experiment has been verified to ensure that it is accurately stated and directly correlates with the parameters presented in Tables 2, 3, 4, 5, 6, 7, 8 and 9. Every quantitative enhancement (communication reliability: 93%, power reduction: 35%, routing efficiency: 42%, network lifetime: 35%) is based on an average of validated simulations with statistical significance (p < 0.05 or p < 0.01). There is no extrapolation of the text out of the conditions in which it was tested. These explanations guarantee the transparency and reproducibility of the findings.