The main goal of this paper is to determine the optimal DC bias value for DC-biased optical orthogonal frequency division multiplexing (DCO-OFDM)- based Light fidelity (Li-Fi) systems using machine learning (ML) algorithms. The reason for this is that in DCO-OFDM, applying either a significant or a slight DC bias can lead to inefficiencies, such as increased clipping noise or reduced optical power. Therefore, it is crucial to determine the most appropriate DC bias level. So, to achieve this, ML algorithms were implemented. This paper compares ML techniques, including hybrid linear regression with K-Nearest Neighbors (KNN) and hybrid polynomial regression with KNN, for predicting the ideal DC bias value in DCO-OFDM-based Li-Fi systems. The models were evaluated based on their Root Mean Square Error (RMSE), the Coefficient of Determination (R2), and the Mean Absolute Percentage Error (MAPE). The results show that hybrid polynomial regression with KNN outperforms the other models, achieving a lower RMSE of 0.18847, a higher R2 of 96.908%, and a lower MAPE of 9.271%. In addition, to validate the model's performance using real-world measurements, a hardware implementation of the Li-Fi system was developed using an Arduino-based receiver to collect actual signal data, which was then used to further evaluate prediction accuracy. The hardware results further confirmed the superiority of the hybrid polynomial regression with KNN model, achieving an RMSE of 0.296 and an R2 score of 81.37%, consistent with the simulation outcomes and demonstrating its robustness in real-time scenarios. These findings suggest that hybrid polynomial regression with KNN can be an effective technique for predicting the optimal DC bias value in DCO-OFDM-based Li-Fi systems.

Free Space Optical Orthogonal Frequency Division Multiplexing (FSO- OFDM) has emerged as a powerful distinct transmission framework for next-generation wireless links; here the performance of FSO system using OFDM in Low Earth Orbit/Medium Earth Orbit (LEO/MEO) inter-satellite communication is investigated under the combined effects of thermal, background noise and solar storm disturbances over a Gamma-Gamma (GG) channel model. An impact of channel impairments are analysed for BPSK, QPSK and M-ary (M = 64,128,256) QAM transmission. A bit error rate in the received signal is analysed for different refractive index with weak and strong turbulence conditions. To increase the transmission link distance and reduce the bit error rate, we employed zero forcing equalization techniques at the receiver. We examine the different transmission channel length from 1000 to 11000m. Lower order PSK modulation schemes such as BPSK and QPSK indorse reliable performance for link ranges up to 7000m due to their robustness to turbulence and noise. The simulation results substantiate that in weak turbulence OFDM-FSO system, 256-QAM achieves high spectral efficiency with a BERvalue of 10-12 at 23dB SNR. Under severe turbulence, BPSK is optimal up to 7000m with a BERof 10-12 at 22dB SNRin weak, and 27dB in strong turbulence, whereas for long distance up to 11000m, 256-QAM becomes more efficient, reaching a BERof 10-6 at 27.5dB SNR. The analysis further demonstrate that for an identical symbol rate, the BER achieved at 11000m in weak turbulence closely bouts the BER obtained at 9000m when the channel experiences strong turbulence like solar storm. Furthermore, the developed equalization framework exhibits superior performance compared to traditional attitudes, demonstrating its capability to sustain stable long-range OFDM-FSO communication for LEO/MEO inter-satellite links.

The global demand for ultra-high-speed data transmission intensifies, Intensity-Modulation Direct-Detection (IM/DD) systems have recently turned as the pillar of short-reach optical networks like 5G/6G front haul, hyper-scale data-centres and Optical Wireless Communication (OWC/Li-Fi). Out of all the modulation candidates, Direct-Current Biased Optical Orthogonal Frequency Division Multiplexing (DCO-OFDM) is one of the most promising modulation formats due to the benefits of high spectral efficiency and immunity to the multipath fading and ISI due to the inherent nature of this format. Nonetheless, high DC bias currents must be used to prevent signal clipping in optical emitters (e.g., Light Emitting Diodes (LEDs) and Laser Diodes (LDs)) due to the large power fluctuations, which results in inefficient energy utilization and an accelerated deterioration of the component. In addition, if these peaks become larger than the small linear dynamic range of inexpensive hardware then they cause strong non-linear distortion, greatly increasing the Bit Error Rate (BER) while also limiting the modulation order possible (e.g. 64-QAM to 1024-QAM). This review totally examines the Direct Clip framework a further developed type of digital flag handling (DSP) that is utilized to limit these impediments. While typical clipping regards signal peaks as random errors, Direct Clip takes advantage of Iterative Clipping and Filtering (ICF) and Noise Shaping to actively push the signal envelope into the ideal domain of operation for hardware. This paper illustrates the trade-offs between computational complexity, spectral leakage, and fidelity through individual blocks to show how Direct-Clip architectures facilitate the utilization of low-resolution Digital-to-Analog Converters (DACs) and non-linear optical components for high-throughput applications. These findings confirm that Direct-Clip Optical-OFDM is much more than an incremental advance, but rather one of the core building blocks for sustainable, low-cost, high-capacity Optical connectivity at the next level of software-defined networks.

Abstract This paper proposes a channel-uncertainty-aware hybrid learning spectrum sensing (CUA-HLSS) framework for optical orthogonal frequency division multiplexing (OFDM) networks operating under imperfect channel estimation. In practical optical OFDM systems, channel estimation errors occur due to noise, limited pilot resources, and hardware impairments, which degrade the performance of conventional spectrum sensing techniques relying on fixed thresholds and ideal channel assumptions. To address this issue, the proposed framework explicitly models channel estimation uncertainty and incorporates it into the sensing process through uncertainty-aware feature extraction and hybrid learning-based adaptive decision-making. The CUA-HLSS approach employs multiple sensing features, including normalized subcarrier energy, subcarrier energy variance, residual channel error power, and a channel uncertainty factor, to enhance detection reliability, particularly in low signal-to-noise ratio (SNR) environments. Simulation results demonstrate that the proposed algorithm achieves a detection probability greater than 0.85 at −10 dB SNR, whereas classical energy detection, matched filter, and cyclostationary sensing methods achieve only 0.45–0.65 under the same conditions. Compared with CNN- and RNN-based sensing methods, CUA-HLSS provides a 15–20 % improvement in detection probability while maintaining a false alarm probability below 0.05. Furthermore, the framework requires only 4.2 ms runtime and 5.0 mJ energy per sensing decision.

WMLP-CRF: Weighted Multi Layer Perceptrons with Conditional Random Field for Fiber-Induced Nonlinearity Mitigation in Long-Haul Coherent Optical OFDM

The necessity for reliable healthcare monitoring following the COVID-19 pandemic has highlighted the limitations of RF-based devices in medical settings. Visible light communication (VLC), which provides inherent security and is resistant to RF interference, is a good alternative. This work proposes a VLC system using DC-biased optical (DCO) orthogonal frequency division multiplexing (OFDM) for resilient, high-speed biomedical data transmission through indoor optical fading channel. In this system, data is modulated using quadrature amplitude modulation (QAM) with order 4 and 16. Four equalization methods; block-type, comb-type, superimposed training (ST), and blind channel estimation (CE); are implemented across three patient positioning scenarios. We integrate blind channel estimation and SLM-based PAPR reduction for a dynamic healthcare LiFi scenario. A comprehensive comparative analysis of CE techniques is conducted under realistic patient positioning (LOS/NLOS) conditions. An analysis of the trade-off between spectral efficiency and energy-to-noise ratio is examined in this context. Simulation results reveal a high Peak-to-Average Power Ratio (PAPR), reaching 15dB with block-type CE. To mitigate this, Selected Mapping (SLM) is applied with three complex phase sequences, and three real sequences, achieving up to 4 dB PAPR reduction with no Bit Error Rate (BER) degradation. At 28dB SNR, BER values were [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] for blind, block-type, comb-type, and ST CE, respectively. Spectral efficiency declines with increased multipath, yet blind CE maintains the highest performance, reaching 0.9bits/s/Hz with 20 multipath components. Additionally, complex phase vectors in SLM provide an extra 1dB PAPR gain over real-valued versions.

Abstract The growing demand for batteryless indoor Internet of Things (IoT) devices has stimulated interest in visible light communication (VLC) systems capable of simultaneously delivering information and harvesting energy from ambient illumination. This paper presents a photovoltaic (PV)-based VLC system employing asymmetrically clipped optical orthogonal frequency-division multiplexing (ACO-OFDM) to enable simultaneous lightwave information and power transfer (SLIPT) using a single PV receiver. A bias-tee front-end separates the harvested direct-current (DC) component from the data-bearing alternating-current (AC) signal, while an adaptive normalized least mean squares (NLMS) equalizer is applied to mitigate illumination-dependent gain drift, PV nonlinearity, and maximum power point tracking (MPPT)-induced ripple. Comprehensive simulations evaluate bit error rate (BER), spectral efficiency (SE), and harvested power under practical indoor conditions. Results show that reliable communication with BER below 10 −3 is achieved at approximately 18 dB SNR for 16-QAM and 25 dB for 64-QAM. The system attains spectral efficiency approaching 1 bit/s/Hz and 1.5 bit/s/Hz for 64-QAM at moderate-to-high SNRs while maintaining energy-neutral operation at illumination levels above approximately 800 lux. These findings confirm the feasibility of PV-based VLC as an effective solution for low-power, batteryless indoor IoT applications.

Abstract This paper investigates peak-to-average power ratio (PAPR) reduction and its impact on bit error rate (BER) and spectral characteristics in optical orthogonal frequency division multiplexing (Optical-OFDM) systems using different numbers of sub-carriers. Conventional optical-OFDM suffers from high PAPR, leading to severe nonlinear distortion, BER degradation, and increased out-of-band radiation. To address these issues, classical PAPR reduction techniques such as selective mapping (SLM), partial transmit sequence (PTS), and C-PTS-PSO are analyzed and compared with a proposed PTS-PSO scheme. Simulation results demonstrate that the proposed method consistently achieves the lowest PAPR across 64, 256, and 512 sub-carrier configurations, with PAPR reductions of up to 8–9 dB at a CCDF of 10 −3 compared to conventional optical-OFDM. The reduced PAPR significantly improves BER performance, where the proposed scheme achieves SNR gains of approximately 6–7 dB at a BER of 10 −3 for 64 and 256 sub-carriers relative to conventional methods. Furthermore, power spectral density (PSD) analysis confirms substantial suppression of out-of-band emissions, achieving more than 50 dB reduction compared to conventional Optical-OFDM. These results highlight that the proposed PTS-PSO method effectively mitigates nonlinear effects, enhances spectral efficiency, and improves overall system reliability.

• VLC systems face high PAPR challenges limiting efficiency and reliability. • Proposed hybrid techniques mitigate PAPR while improving BER performance. • Approach combines nonlinear companding with advanced precoding methods. • Tailored for ACO-OFDM’s unipolar structure and distortion resilience. • Introduced noise cancellation model enhances BER by suppressing noise. Visible light communication (VLC) systems that employ optical orthogonal frequency division multiplexing (O-OFDM) face a critical challenge in the form of a high peak-to-average power ratio (PAPR), which limits transmission efficiency and system reliability. This study proposes and evaluates a set of hybrid techniques aimed at mitigating PAPR while simultaneously enhancing bit error rate (BER) performance. The approach integrates nonlinear companding transformations with various precoding methods, specifically tailored for asymmetrically clipped optical OFDM (ACO-OFDM), a promising scheme for VLC due to its unipolar signal structure and resilience to nonlinear distortions. In addition, a dedicated noise cancellation (N.C.) model is introduced to further improve BER performance by suppressing channel and clipping noise. Simulation results demonstrate significant improvements, with PAPR reductions reaching up to 6.329 dB, 7.123 dB, and 7.549 dB, and corresponding E b / N 0 differences of − 0.727 dB, 1.435 dB, and 1.799 dB with respect to conventional ACO-OFDM. These outcomes highlight the efficacy of the proposed hybrid methods and their practical suitability for real-world VLC systems, offering a balanced trade-off between power efficiency and signal fidelity.

In high-power semiconductor lasers, efficient thermal management is crucial, particularly in compact transistor-outline (TO) can packages where heat dissipation is challenging. The conventional glass window used in these packages, though transparent in the visible regime, exhibits low thermal conductivity and infrared transparency, leading to heat accumulation, elevated junction temperatures, and degraded device performance. To overcome this limitation, we fabricated and optimized a thermally and optically transparent polyethylene (PE) film as a replacement for the commercial glass window. By tailoring its thickness to enhance mid-infrared transparency while maintaining optical transparency at 660 nm and mechanical strength, the 100-μm PE film enables efficient radiative cooling, resulting in a window temperature reduction of up to 9.4 °C compared to the glass configuration. This improvement yields, for instance, a reduced threshold current (54 mA reduction), increased output power (by ∼ 50 mW), and enhanced slope efficiency (0.11 W/A enhancement) at 35 °C and higher. At an ambient temperature of 80 °C, thermal fitting reveals an effective chip temperature of 72.8 °C, corresponding to a 7.2 °C cooling advantage over the glass-covered device. Our approach sustains optical modulation performance of ∼1.8-1.9 GHz modulation bandwidth and improves direct current optical orthogonal frequency-division multiplexing (DCO-OFDM) net data rate to 2.11 Gb/s. These results establish the PE film as a robust, low-cost, scalable, and effective thermal management solution for enhancing the performance and reliability of high-power optoelectronic devices in thermally constrained environments.

Abstract In Intensity-Modulated Direct Detection Optical Orthogonal Frequency Division Multiplexing (IM/DD-OOFDM) communication systems, a lower Peak-to-Average Power Ratio (PAPR) is essential for improving system performance. In order to reduce the high PAPR, a Migrating Birds Optimization (MBO) is integrated with the classical Selective Mapping (SLM) method in IM/DD-OFDM system. This approach successfully optimizes the values of phase factors, minimizes the search numbers, and reduces computational complexity. To evaluate the impact of the MBO-SLM method on PAPR reduction performance, the parameters of the MBO-SLM method play a crucial role in decreasing the PAPR in an IM/DD optical OFDM system. The MBO-SLM method demonstrates improved Bit Error Rate (BER) performance, Power Spectral Density (PSD), and power saving performance compared to other PAPR reduction methods. Numerical results indicate that the proposed PAPR reduction method outperforms other existing methods for optical OFDM signals. Specifically, by implementing this technique in the IM/DD optical OFDM communication system, we achieved a decrease in PAPR from 10.5 dB to 4.95 dB at a Complementary Cumulative Distribution Function (CCDF) of 10 –3 , resulting in a reduction of 5.55 dB. Additionally, the computational complexity of the MBO-SLM method shows a 91% improvement over the Discrete Elephant Herding Optimization (DEHO) based Partial Transmit Sequence (PTS) method when the search number is set to 512.

Abstract Optical orthogonal frequency division multiplexing (Optical OFDM) is a key modulation technique for high-speed optical communication systems; however, its performance is significantly degraded by chromatic dispersion and phase noise, which impair subcarrier orthogonality and introduce inter-symbol and inter-carrier interference. To overcome these limitations, this work proposes an enhanced signal detection scheme for optical OFDM based on a hybrid equalization framework integrated with quasi-reduced maximum likelihood detection (QRM-MLD). The proposed approach is evaluated against conventional detection methods, including zero-forcing equalization, minimum mean square error, maximum likelihood, and successive interference cancellation. MATLAB-based simulation results demonstrate that the hybrid QRM-MLD scheme achieves substantial performance gains under various channel impairment conditions. At a target bit error rate of 10 −3 , the proposed method requires up to 10 dB lower signal-to-noise ratio compared to uncompensated optical OFDM and offers approximately 2–7 dB improvement over conventional detection schemes. Capacity analysis further confirms its superiority, achieving a value of 290 at an SNR of 50 dB. Overall, the proposed hybrid detection strategy significantly improves detection accuracy, spectral efficiency, and robustness, making it a strong candidate for next-generation high-capacity optical OFDM communication systems.

Layered asymmetrically clipped optical orthogonal frequency division multiplexing (LACO-OFDM), which can offer high efficiency in both spectral and energy under the unipolar signal constraint, is a state-of-the-art multi-carrier modulation format for visible light communication (VLC) systems. However, entropy loading (EL), which can squeeze out the last few bits of the multi-carrier system, has not been demonstrated in LACO-OFDM systems yet. Unlike the single-layer multi-carrier system, where orthogonal subcarriers allow straightforward EL application, the multi-layer based LACO-OFDM complicates EL application due to non-orthogonal subcarriers caused by inter-layer interference (ILI) from residual clipping noise (RCN) in demodulation. After theoretically analyzing the impact of probabilistic constellation shaping (PCS) on RCN in LACO-OFDM, a probability-aware RCN power estimation method based on constellation trimming (CT) is proposed to accurately estimate the effective signal-to-noise-ratio (SNR) of the RCN-contaminated subcarriers. Numerical studies have been conducted to evaluate the performance of the proposed method under various channel conditions, which show that a "win-win" outcome, i.e., higher estimation accuracy with lower computational complexity, can be achieved with a proper threshold in CT. Based on this, a total power constrained EL algorithm is proposed for LACO-OFDM VLC systems, to the best of our knowledge, for the first time. The performance investigation of the proposed EL algorithm in the LACO-OFDM VLC system has been carried out via both simulation and experiment. The experimental results show that: (a) compared with the existing bit loading (BL) method which also considers the impact of RCN in LACO-OFDM, our EL method offers higher generalized mutual information (GMI) across the whole range of received optical power (ROP) under test, with an average GMI improvement of ~10% and an maximum GMI gain up to 0.2 bits/symbol; (b) setting the CT threshold to 0.5 yields GMI comparable to the case without CT while substantially reducing the computing time (e.g., by ~80% at the ROP of 28.44 µW), with time savings becoming more pronounced at higher ROP levels.

Beamforming and waveform design are key essentials for integrated sensing and communication (ISAC) in future wireless networks, which also inspires the emerging research of optical wireless ISAC (OW-ISAC). In this paper, we investigate the joint design of optical phased array (OPA) beamforming and orthogonal frequency division multiplexing (OFDM) waveform, thus extending the radio-frequency (RF)-ISAC paradigm to the optical band. First, the system model for the OPA-based OW-ISAC framework is introduced to facilitate simultaneous communication and sensing. Next, signal processing techniques are tailored for an optical OFDM waveform with precisely separated communication and sensing subcarriers. Then, the joint optimization problem of OPA beamforming and OFDM waveform design is formulated, decomposed, and resolved to reach a flexible compromise between communication and sensing performances. Moreover, numerical simulations validate the effectiveness of the proposed OW-ISAC scheme and reveal the trade-off between communication and sensing functionalities. Consequently, the capabilities of reliable communication and precise sensing will establish OW-ISAC as a powerful complement to RF-ISAC in the near future.

The line-of-sight (LOS) requirement of free-space optical (FSO) systems can be relaxed by employing optical intelligent reflecting surfaces (IRSs). In this paper, we show that an IRS-assisted FSO system employing a square-law photo detector (PD) receiver can be modeled as a linear end-to-end system if the receiver lens area is sufficiently large. Based on this linear model, we characterize the impact of IRS-induced delay dispersion and derive an analytical expression for the corresponding channel impulse response (CIR), which reveals the dependence of the end-to-end channel on the characteristics of the incident and reflected beams’ wavefronts, the position of transmitter and receiver, the size and phase shift profile of the IRS, and the incident beamwidth on the IRS. For transmission, we consider an on-off keying (OOK) and a DC-clipped optical orthogonal frequency-division multiplexing (DCO-OFDM) system. Our simulation results reveal that a maximum effective delay spread of 0.7 ns is expected for in-plane reflection from a square IRS with an area of 1 m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, which induces ISI for bit rates larger than 10 Gbps. We show that while the maximum delay spread is approximately independent of the IRS phase shift profile, the received power for focusing and quadratic phase shift profiles is larger than that for linear phase shift profiles. We also show that the IRS-induced delay dispersion can be mitigated by equalization at the receiver. Our results reveal that DCO-OFDM performs better than OOK modulation with zero forcing linear equalization (ZF-LE), whereas OOK modulation with decision feedback equalization (DFE) always outperforms DCO-OFDM.

To allow a good signal-to-noise ratio (SNR) over a wide bandwidth for signals arriving from an unknown angle, many optical wireless communication (OWC) systems need a wide-aperture high-bandwidth detector. However, the physical properties of a photodiode (PD) challenge the design for high bandwidth and a large detection area, simultaneously. This problem is also known as the bandwidth versus area trade-off of an amplified PD receiver. In this paper, we study the limitations of a single-PD OWC receiver on throughput, particularly on bandwidth, photon capturing area, and electrical noise, to formulate a model that evaluates and optimizes the throughput of a communication link for different signal modulation schemes. We optimize the signal power spectral density (PSD) of pulse-amplitude modulation (PAM) and of DC-biased optical orthogonal frequency-division multiplexing (DCO-OFDM) modulation for throughput, considering that large modulation bandwidths suffer disproportionally from noise enhancement in typical receiver amplifiers. This worsens if large PDs are used, but we show that to a large extent, the corresponding stronger signal can compensate for the increased noise. We study the optimum detector size, in combination with adaptations of the spectrum and, in particular, the maximum frequency of the modulation. While the limitations of photonic emitters have been extensively studied before, this is one of the first papers that models how the receiver limits the throughput of PAM and OFDM.

Optical orthogonal frequency-division multiplexing (OFDM) and probabilistic constellation shaping (PCS) have emerged as powerful techniques to enhance the performance of optical wireless communications (OWC) systems. While PCS improves spectral efficiency and adaptability, we show that its integration with optical OFDM can inadvertently increase the peak-to-average power ratio (PAPR) of the signal, exacerbating clipping distortion due to signal clipping. This letter investigates the impact of PCS on the PAPR of direct current-biased optical OFDM (DCO-OFDM) waveforms and proposes an optimization of PCS that maximizes channel capacity, considering clipping distortion. The optimization problem is shown to be complex and non-convex. We thus present a suboptimal yet efficient solving approach based on projected gradient descent to solve the problem. Simulation results demonstrate the superiority of the proposed approach over the conventional uniform signaling, particularly under severe clipping distortion conditions.

Visible Light Communication (VLC) systems commonly employ optical orthogonal frequency division multiplexing (O-OFDM) to achieve high data rates, benefiting from its robustness against multipath effects and intersymbol interference (ISI). However, a key limitation of asymmetrically clipped direct current biased optical–OFDM (ACO-OFDM) systems lies in their inherently high peak-to-average power ratio (PAPR), which significantly affects signal quality and system performance. This paper proposes a joint chaotic encryption and modified μ-non-linear logarithmic companding (μ-MLCT) scheme for ACO-OFDM–based VLC systems to simultaneously enhance security and reduce PAPR. First, image data is encrypted at the upper layer using a hybrid chaotic system (HCS) combined with Arnold’s cat map (ACM), mapped to quadrature amplitude modulation (QAM) symbols and further encrypted through chaos-based symbol scrambling to strengthen security. A μ-MLCT transformation is then applied to mitigate PAPR and enhance both peak signal-to-noise ratio (PSNR) and bit-error-ratio (BER) performance. A mathematical model of the proposed secured ACO-OFDM system is developed, and the corresponding BER expression is derived and validated through simulation. Simulation results and security analyses confirm the effectiveness of the proposed solution, showing gains of approximately 13 dB improvement in PSNR, 2 dB in BER performance, and a PAPR reduction of about 9.2 dB. The secured μ-MLCT-ACO-OFDM not only enhances transmission security but also effectively reduces PAPR without degrading PSNR and BER. As a result, it offers a robust and efficient solution for secure image transmission with low PAPR, making it well-suitable for emerging wireless networks such as cognitive and 5G/6G systems.

The direct current (DC) in optical orthogonal frequency division multiplexing (DCO-OFDM) scheme is commonly adopted in light fidelity (Li-Fi) technology as it offers a spectrally efficient solution. A prior study adopted a machine learning (ML)-based solution to predict the optimum DC bias using key parameters, including the statistical properties of the OFDM transmitted signal and a polynomial regression model. However, the model's robustness decreased when the data structure was shuffled, indicating limited generalization to unseen data. This study builds upon that work by utilizing the same dataset and improving the prediction model with advanced ML tools, such as the LazyPredict algorithm (LPA), to systematically evaluate and select a regression model. A robust ML regressor selection process is proposed to ensure the reliability of predictions. Additionally, a comprehensive data analysis is conducted to assess the importance of features affecting the optimum DC bias. The results demonstrate that the ensemble learning algorithm, Random Forest (RF), outperforms other models with an R-squared of 0.953 and an RMSE of 0.233. A Friedman statistical test was applied to validate the results over five iterations of model training. Furthermore, hyperparameter tuning and bootstrap sampling were employed to conduct a deeper investigation into the model's performance and stability. The proposed model significantly enhances the accuracy and robustness of DC bias prediction compared to previous approaches, ensuring consistent performance across different data distributions.

The direct current (DC) in optical orthogonal frequency division multiplexing (DCO-OFDM) scheme is commonly adopted in light fidelity (Li-Fi) technology as it offers a spectrally efficient solution. A prior study adopted a machine learning (ML)-based solution to predict the optimum DC bias using key parameters, including the statistical properties of the OFDM transmitted signal and a polynomial regression model. However, the model’s robustness decreased when the data structure was shuffled, indicating limited generalization to unseen data. This study builds upon that work by utilizing the same dataset and improving the prediction model with advanced ML tools, such as the LazyPredict algorithm (LPA), to systematically evaluate and select a regression model. A robust ML regressor selection process is proposed to ensure the reliability of predictions. Additionally, a comprehensive data analysis is conducted to assess the importance of features affecting the optimum DC bias. The results demonstrate that the ensemble learning algorithm, Random Forest (RF), outperforms other models with an R-squared of 0.953 and an RMSE of 0.233. A Friedman statistical test was applied to validate the results over five iterations of model training. Furthermore, hyperparameter tuning and bootstrap sampling were employed to conduct a deeper investigation into the model’s performance and stability. The proposed model significantly enhances the accuracy and robustness of DC bias prediction compared to previous approaches, ensuring consistent performance across different data distributions.