Research Progress in Orbital Angular Momentum Recognition for Laser Beams Based on Artificial Intelligence (Invited)

Shiyun Zhou; Yishu Wang; Jinyu Yang; Chunqing Gao; Shiyao Fu

doi:10.3788/AOS231987

Acta Optica Sinica, Volume. 44, Issue 14, 1400002(2024)

Research Progress in Orbital Angular Momentum Recognition for Laser Beams Based on Artificial Intelligence (Invited)

Shiyun Zhou^1,2,3, Yishu Wang^1,2,3, Jinyu Yang^1,2,3, Chunqing Gao^1,2,3, and Shiyao Fu^1,2,3、*

Author Affiliations

¹School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China

²Key Laboratory of Information Photonics Technology, Ministry of Industry and Information Technology, Beijing 100081, China

³Key Laboratory of Optoelectronic Imaging Technology and System, Ministry of Education, Beijing 100081, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. OAM recognition schemes based on machine learning model. (a) SOM model^[65]; (b) 143 km communication experiment based on ANN^[66]; (c) structure of shallow CNN model^[67]; (d) ANN model for multiple OAM recognition^[68]; (e) structure of FNN model^[69]; (f) structure of 8-CNN model^[70]; (g) structure of 10-CNN model^[71]; (h) two stage 10-CNN model for high-precision decoding^[72]; (i) structure of adaptive deep ELM model^[73]

Download full size

Fig. 2. OAM recognition schemes based on deep learning model. (a) Visualization of VGGNet model^[76]; (b) structure of deep CNN model^[77]; (c) structure of MetaNet model^[78]; (d) structure of Adjusted ENN model^[80]; (e) flow chart of fast OAM spectrum analysis based on DRN^[81]; (f) structure of opto-electronic neural network^[82]; (g) structure of SNN model^[83]; (h) flow chart of identifying classic unseparable base phase differences in OAM based on DRN^[84]

Download full size

$OAM recognition schemes based on hybrid learning model. (a) Structure of CNN and R-CDT hybrid model[85]; (b) structure of 3-layer D2NN model[86]; (c) structure of SVM and PCA hybrid model[87]; (d) structure of astigmatic transformation (AT) and CNN hybrid model[88]; (e) structure of angular spectrum (AS) and CNN hybrid model[89]; (f) structure of diffraction and CNN hybrid model (DCNN)[90]$

Fig. 3. OAM recognition schemes based on hybrid learning model. (a) Structure of CNN and R-CDT hybrid model^[85]; (b) structure of 3-layer D²NN model^[86]; (c) structure of SVM and PCA hybrid model^[87]; (d) structure of astigmatic transformation (AT) and CNN hybrid model^[88]; (e) structure of angular spectrum (AS) and CNN hybrid model^[89]; (f) structure of diffraction and CNN hybrid model (DCNN)^[90]

Download full size

Fig. 4. AI-based OAM recognition schemes under atmospheric turbulence. (a) Structure of 6-CNN model^[92]; (b) structure of deep CNN model^[93]; (c) structure of TACCNN model^[94]; (d) structure of CNN and ConvLSTM hybrid model^[95]; (e) structure of turbulence compensation phase screen network^[96]

Download full size

Fig. 5. AI-based OAM recognition schemes under ocean turbulence. (a) Structure of 6-CNN model used in Ref. [97]; (b) structure of 6-CNN model used in Ref. [98]; (c) structure of D²NN model^[99]; (d) structure of HOEDNN model^[100]

Download full size

Fig. 6. AI-based OAM recognition schemes under other disturbances. (a) CNN model under noise disturbance^[101]; (b) DenseNet model under misalignment disturbance^[102]; (c) DenseNet model under smoke disturbance^[103]; (d) CNN model under scattering field turbulence^[104]; (e) CNN model under white noise and atmospheric turbulences^[105]

Download full size

Table 1. Numerical methods, machine learning, and deep learning models involved in this paper

View table

Table 1. Numerical methods, machine learning, and deep learning models involved in this paper

Type of methodology	Method	Abbreviation
Numerical method	Radon-cumulative-distribution transform^［48］	R-CDT
Numerical method	Principal component analysis^［49］	PCA
Machine learning	Self-organizing map^［50］	SOM
	Artificial neural network^［51］	ANN
	Support vector machine^［52］	SVM
	Convolutional neural network^［53］	CNN
	Feedforward neural network^［54］	FNN
	Extreme learning machine^［55］	ELM
Deep learning	Efficient neural network^［56］	ENN
	Spiking neural network^［57］	SNN
	Deep residual network^［41］	DRN
	Densely connected convolutional network^［58］	DenseNet
	Diffractive deep neural network^［59］	D²NN
	Long short-term memory^［60］	LSTM

Table 2. OAM recognition schemes based on AI technology

View table

Table 2. OAM recognition schemes based on AI technology

Type of AI technology	Reference	Year	Model	Range	Count	Evaluation
Machine learning	[65]	2014	SOM	16	1	E-AER=1.7%
	[66]	2016	ANN	4	1	E-A>80%
	[67]	2018	3-CNN	[ $-$ 10, 10]	2	S-A>88%
	[68]	2018	ANN	[ $-$ 10, 10]	2	S-MSE=0.1
	[69]	2019	FNN	[ $-$ 25, 25]	1	S-A=99.55%
	[70]	2020	8-CNN	[ $-$ 50, 50]	1	E-A=98.54%
	[71]	2021	10-CNN	[1, 8]	2	E-A>98%
	[72]	2023	10-CNN	[ $-$ 4, 8]	1	E-A=99.84%
	[73]	2023	ELM	[1, 10]	1	S-RMSE=0.0632
Deep learning	[74]	2016	VGGNet	[1, 110]	1	S-A>74%
	[77]	2018	CNN	[1, 100]	1	E-A>99%
	[78]	2020	MetaNet	255	1	E-A>90.2%
	[80]	2022	Adjusted ENN	[ $-$ 10, 10]	7	S-MSE=10^-6
	[83]	2023	SNN	[1, 9]	9	E-A>99.1%
	[82]	2023	Opto-electronic neural network	[ $-$ 10, 10]	21	S-MSE=10^-3-10^-5
	[84]	2023	DRN	—	1	E-MSE=0.00632
	[81]	2023	DRN	[ $-$ 150, 150]	50	S-RMSE=0.002
Hybrid learning	[85]	2018	CNN+R-CDT	3	1	E-A=98.50%
	[86]	2019	3-D²NN	10	1	S-A>80%
	[87]	2020	SVM+PCA	15	1	E-A=98%
	[88]	2021	AT+CNN	[ $-$ 3, 3]	1	E-A=99%
	[89]	2021	AS+CNN	[ $-$ 6, 6]	1	S-A=97.9%
	[90]	2022	DCNN	[ $-$ 7, 7]	1	E-AF=97.82%

Table 3. AI-based OAM recognition schemes under disturbances

View table

Table 3. AI-based OAM recognition schemes under disturbances

Disturbance		Reference	Year	Model	Setting		Range	Evaluation
Disturbance		Distance	Scale				Range	Evaluation
Nonhomogeneous medium	Atmosphere	[92]	2019	6-CNN	50 m	$10^{- 12}$	[ $-$ 15, 15]	E-A=89.48%
		[106]	2019	6-CNN	3 km	$10^{- 15}$	[ $-$ 8, 8]	E-A>84%
		[93]	2019	14-CNN	50 m	$10^{- 13}$	0, 1, 3	E-MP=98.34%
		[107]	2020	6-CNN	5 km	$10^{- 14}$	[1, 16]	S-A>70%
		[94]	2020	TACCNN	—	D∕r₀=5.28	20	E-MP>70%
		[95]	2022	ConvLSTM	1 km	$10^{- 15}$	[ $-$ 3, 4]	E-MSE=0.025
		[108]	2022	DenseNet	—	D∕r₀=2.67	$-$ 20, 20	E-MP=90.51%
		[96]	2023	ResNet	—	D∕r₀=5	$-$ 20, 20	E-MP>90%
	Ocean	[97]	2021	6-CNN	100 m	$5 \times 10^{- 14} K^{2}$	[1, 10]	E-A=99%
		[98]	2022	6-CNN	10 m	$10^{- 14} K^{2}$	[ $-$ 8, 8]	E-A=88.5%
		[99]	2022	D²NN	>50 m	$10^{- 12} K^{2}$	12	S-A=82.5%
		[100]	2023	HOEDNN	—	$10^{- 12} K^{2}$	16	E-A≈100%
Other		[101]	2020	CNN	Noise		16	E-A>90%
		[102]	2021	DenseNet	Misalignment		[ $-$ 5, 5]	E-A=98.35%
		[103]	2022	DenseNet	Smoke		[1, 8]	E-A=97%
		[104]	2022	CNN	Scattering		[1, 8]	E-A=99%
		[105]	2023	CNN	Noise plus turbulence		16	S-A=97%

Tools

Get Citation

Copy Citation Text

Shiyun Zhou, Yishu Wang, Jinyu Yang, Chunqing Gao, Shiyao Fu. Research Progress in Orbital Angular Momentum Recognition for Laser Beams Based on Artificial Intelligence (Invited)[J]. Acta Optica Sinica, 2024, 44(14): 1400002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Reviews

Received: Dec. 26, 2023

Accepted: Feb. 27, 2024

Published Online: Jul. 4, 2024

The Author Email: Fu Shiyao (fushiyao@bit.edu.cn)

DOI:10.3788/AOS231987

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology