One of the challenges faced in optical communication system is the signal deterioration due to dispersion. Various conventional method such as dispersion compensation fiber (DCF) and fiber bragg grating (FBG) are available to compensate dispersion. In order to avoid the maintenance and performance degradation due to aging, it is essential to incorporate the signal processing based compensation methods. Recently, due to the emergence of artificial intelligence, various deep learning techniques are used to compensate dispersion. The main objective of the dispersion compensation techniques is to improve the quality of the signal over long-haul transmission. In order to achieve this, we propose three variants of modified Automatic Dispersion Compensation Network (ADC-Net) such as LeADC-Net, AlexADC-Net and ResADC-Net to compensate dispersion in optical fiber for Intensity Modulation/ Direct Detection (IM/ DD) system. In this approach trained weights and biases are applied to the output optical signal to recover the original input optical signal in order to improve the bit error rate (BER) and quality factor. The lower mean square error (MSE) of 0.004 and 9 . 24 × 1 0 − 12 , mean absolute error (MAE) of 0.031 and 1 . 63 × 1 0 − 6 , and root mean square error (RMSE) of 0.065 and 3 . 04 × 1 0 − 6 was achieved for magnitude prediction in ResADC-Net and phase angle prediction in LeADC-Net compared to other proposed architectures during the training process. The higher quality factor (Q factor) of 45.513 dB is achieved in ResADC-Net compared to other proposed architecture. Performance improvement is achieved in ResADC-Net compared to other proposed architectures and state-of-the-art methods. • Improves BER and Q factor with ResADC-Net architecture. • No need of DCF, FBG or regenerative repeater up to 90 000 km. • Low MSE, MAE, and RMSE achieved in ResADC-Net compared to ADC-Net. • High degree of similarity achieved in ResADC-Net compared to ADC-Net.
Rajendran et al. (Fri,) studied this question.