Fast mode decomposition for few-mode fiber based on lightweight neural network

Jiajia Zhao; Guohui Chen; Xuan Bi; Wangyang Cai; Lei Yue; Ming Tang

doi:10.3788/COL202422.020604

Chinese Optics Letters, Volume. 22, Issue 2, 020604(2024)

Fast mode decomposition for few-mode fiber based on lightweight neural network

Jiajia Zhao¹, Guohui Chen¹, Xuan Bi¹, Wangyang Cai¹, Lei Yue¹, and Ming Tang^2、*

¹School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha 410114, China

²Wuhan National Laboratory for Optoelectronics (WNLO) and National Engineering Laboratory for Next Generation Internet Access System, School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China

show less

Figures & Tables(14)

Fig. 1. Pattern decomposition based on MobileNetV3_Light neural network.

Download full size

View in Article

Fig. 2. Traditional convolution and depth-separable convolution.

Download full size

View in Article

Fig. 3. MobileNetV3_Light network structure.

Download full size

View in Article

Fig. 4. MobileNetV3 block network structure diagram.

Download full size

View in Article

Fig. 5. Test flow chart.

Download full size

View in Article

Fig. 6. Average correlations across training periods for the six model cases.

Download full size

View in Article

Fig. 7. Simulated near-field light-field map, reconstructed near-field light-field map, residual images, and their correlation.

Download full size

View in Article

Fig. 8. Relation between the mode number and correlation.

Download full size

View in Article

Fig. 9. Simulated and reconstructed images under the influence of different intensities of noise and their correlation.

Download full size

View in Article

Table 1. Detailed Parameter Settings of MobileNetV3_Light Network Model Structure

View table

View in Article

Table 1. Detailed Parameter Settings of MobileNetV3_Light Network Model Structure

Input	Operator	Exp size	#out	SE	NL	s
224² × 3	Conv2d	–	8	–	HS	2
112² × 8	Bneck, 3 × 3	16	16	√	RE	2
56² × 16	Bneck, 3 × 3	72	24	–	RE	2
28² × 24	Bneck, 3 × 3	88	24	–	RE	1
28² × 24	Bneck, 5 × 5	96	40	√	HS	2
14² × 40	Bneck, 5 × 5	240	40	√	HS	1
14² × 40	Bneck, 5 × 5	120	48	√	HS	1
14² × 48	Bneck, 5 × 5	144	48	√	HS	1
14² × 112	Bneck, 5 × 5	288	96	√	HS	2
7² × 96	Bneck, 5 × 5	576	96	√	HS	1
7² × 96	Conv2d, 1 × 1	–	576	√	HS	1
7² × 576	AvgPool, 7 × 7	–	–	–	–	1
1² × 576	Conv2d, 1 × 1, NBN	–	1024	–	HS	1
1² × 1024	Conv2d, 1 × 1, NBN	–	K	–	–	1

Table 2. Average Error of the Six Model Weights
View table
View in Article
Table 2. Average Error of the Six Model Weights
$\bar{Δ p_{1}}$ $\bar{Δ p_{2}}$ $\bar{Δ p_{3}}$ $\bar{Δ p_{4}}$ $\bar{Δ p_{5}}$ $\bar{Δ p_{6}}$
Average weights error 0.47% 0.48% 0.42% 0.48% 0.53% 0.55%

Table 3. Average Error of the Relative Phase of the Six Modes
View table
View in Article
Table 3. Average Error of the Relative Phase of the Six Modes
$\bar{Δ θ_{1}}$ $\bar{Δ θ_{2}}$ $\bar{Δ θ_{3}}$ $\bar{Δ θ_{4}}$ $\bar{Δ θ_{5}}$
Average weights error 0.47% 0.48% 0.42% 0.48% 0.53%

Table 4. Time Spent in Different Phases of Testing
View table
View in Article
Table 4. Time Spent in Different Phases of Testing
T₁ T₂ T₃ T₄
Predicting model weight and phase 267.5 min 2.41 s 3.86 s 36.24 s

Table 5. Parameter Size of Different Neural Network Models
View table
View in Article
Table 5. Parameter Size of Different Neural Network Models
MobileNetV3_Light MoblieNetV2 Xception Resnet50 VGG-16
Parameters 2.5 × 10⁶ 3.4 × 10⁶ 22.85 × 10⁶ 25.56 × 10⁶ 138.36 × 10⁶
Mode size 6.5 MB 14.2 MB 88 MB 98 MB 528 MB

Tools

Get Citation

Copy Citation Text

Jiajia Zhao, Guohui Chen, Xuan Bi, Wangyang Cai, Lei Yue, Ming Tang, "Fast mode decomposition for few-mode fiber based on lightweight neural network," Chin. Opt. Lett. 22, 020604 (2024)

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites