Suppression of LED Nonlinear Distortion by Improvement CNN‒BiLSTM‒Self-Attention Cascade Equalization Algorithm Based on Neighborhood Rough Set

Kejun Jia; Jiaqi Che; Jiaxin Liu; Yuqin Xian

doi:10.3788/AOS250539

Acta Optica Sinica, Volume. 45, Issue 10, 1006003(2025)

Suppression of LED Nonlinear Distortion by Improvement CNN‒BiLSTM‒Self-Attention Cascade Equalization Algorithm Based on Neighborhood Rough Set

Kejun Jia^*, Jiaqi Che, Jiaxin Liu, and Yuqin Xian

Author Affiliations

School of Computer and Communication, Lanzhou University of Technology, Lanzhou 730050, Gansu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(26)

Fig. 1. O-OFDM cascade equalizer system

Download full size

Fig. 2. CNN structure diagram

Download full size

Fig. 3. BiLSTM network structure

Download full size

Fig. 4. Self-Attention structure

Download full size

Fig. 5. UAR and LAR in C₁ and C₂ classes

Download full size

Fig. 6. Diagram when LAR is larger than UAR

Download full size

Fig. 7. Classification strategy process optimized by NRS

Download full size

Fig. 8. LMS cascade improved CNN‒BiLSTM‒Self-Attention equalizer

Download full size

Fig. 9. ACO-OFDM system attribute dependence. (a) 4QAM; (b) 16QAM

Download full size

Fig. 10. DCO-OFDM system attribute dependence. (a) 4QAM; (b) 16QAM

Download full size

Fig. 11. Accuracy of classification strategies for different O-OFDM systems. (a) 4QAM; (b) 16QAM

Download full size

Fig. 12. Distribution of indoor VLC geometric scenes

Download full size

Fig. 13. Multipath channel gain between LED and PD. (a) PD is centrally located [3 m,3 m,0.85 m]; (b) PD is at the edge [0.5 m,0.5 m,0.85 m]

Download full size

Fig. 14. Performance of 4QAM ACO-OFDM system under different combination parameters. (a) MSE; (b) BER

Download full size

Fig. 15. Performance of 16QAM DCO-OFDM system under different combination parameters. (a) MSE; (b) BER

Download full size

Fig. 16. Training process curves of ACO-OFDM and DCO-OFDM systems. (a) ACO-OFDM; (b) DCO-OFDM

Download full size

Fig. 17. BER of ACO-OFDM and DCO-OFDM systems under different training set sizes. (a) ACO-OFDM; (b) DCO-OFDM

Download full size

Fig. 18. BER of cascade equalization algorithm for ACO-OFDM system with PD at position [3 m,3 m,0.85 m]. (a) 4QAM; (b) 16QAM

Download full size

Fig. 19. BER of cascade equalization algorithm for DCO-OFDM system with PD at position [3 m,3 m,0.85 m]. (a) 4QAM; (b) 16QAM

Download full size

Fig. 20. BER of cascaded equalization algorithm for different O-OFDM systems with PD at [0.5 m,0.5 m,0.85 m] under 16QAM. (a) ACO-OFDM; (b) DCO-OFDM

Download full size

Fig. 21. Cascade equalization algorithm complexity of ACO-OFDM and DCO-OFDM systems with PD at position [3 m,3 m,0.85 m]. (a) 4QAM; (b) 16QAM

Download full size

Table 1. Simulation parameters

View table

Table 1. Simulation parameters

Parameter	Value
LED position coordinate [x, y, z] /m	[3, 3, 3.5]
PD position coordinate [x, y, z] /m	[3, 3, 0.85], [0.5, 0.5, 0.85]
PD field angle $ψ_{F O V}$ /(°)	60
LED half-power angle $τ_{1 / 2}$ /(°)	70
PD area A_R /cm²	1
Reflectivity of the wall $ρ_{i}$	0.8
Reflective cell area $Δ A$ /m²	0.01

Table 2. Different parameter combinations of CNN‒BiLSTM‒Self-Attention model
View table
Table 2. Different parameter combinations of CNN‒BiLSTM‒Self-Attention model
Number L₁ L₂ c Number L₁ L₂ c
1 1 1 1×1 4 2 2 1×1
2 1 1 1×2 5 2 2 1×2
3 1 1 1×3 6 2 2 1×3

Table 3. Preset parameters of CNN‒BiLSTM‒Self-Attention model

View table

Table 3. Preset parameters of CNN‒BiLSTM‒Self-Attention model

Hyper-parameter	Value
Number of filters	16
Number of hidden nodes	300
Learning rate	0.001
Batch size	120
Training sequence length	10% of total data
Conv layer activation function	ReLU
Self-Attention layer activation function	Softmax
Optimizer	Adam
Epochs	400

Table 4. Parameters of LED‒Wiener model
View table
Table 4. Parameters of LED‒Wiener model
Parameter b₀ b₁ b₂ I_max k V_tov
Value 1 0.15 0.1 0.5 2 0.2

Table 5. Complexity of CNN‒BiLSTM‒Self-Attention algorithm
View table
Table 5. Complexity of CNN‒BiLSTM‒Self-Attention algorithm
Algorithm Complexity
Conventional CNN‒BiLSTM‒Self-Attention $O (K^{2} \cdot Q^{2} \cdot T + K^{2} \cdot Q^{2} \cdot T + L_{2} \cdot d^{2} \cdot T + T^{2 d})$
Improvement CNN‒BiLSTM‒Self-Attention $O (K^{2} \cdot Q^{2} \cdot T + L_{2} \cdot d^{2} \cdot T + T^{2 d})$

Tools

Get Citation

Copy Citation Text

Kejun Jia, Jiaqi Che, Jiaxin Liu, Yuqin Xian. Suppression of LED Nonlinear Distortion by Improvement CNN‒BiLSTM‒Self-Attention Cascade Equalization Algorithm Based on Neighborhood Rough Set[J]. Acta Optica Sinica, 2025, 45(10): 1006003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites