Deep Learning Aided Channel Estimation and Signal Detection for Underwater Optical Communication
Shi Jia,Huang Aiping,Tao Linwei
DOI: https://doi.org/10.3788/CJL202249.1706004
2022-01-01
Chinese Journal of Lasers
Abstract:ObjectiveAt present, underwater wireless optical communication (UWOC) is widely concerned in underwater communication because of its high transmission efficiency and excellent transmission capacity. For underwater acoustic communication, the transmission delay is large because of the limited bandwidth of sound wave in kilohertz frequency region. UWOC technology can achieve the data transmission rate of Gbit/s, while maintaining low transmission delay. In addition, UWOC can carry more data because of the shorter wavelength of light. However, light wave propagation in UWOC channel is affected by absorption, scattering and other factors. The absorption effect is irreversible, and the light energy is converted into other forms of energy, causing the signal to decay. In scattering, the direction in which each photon is emitted varies randomly, so that the energy captured by the receiver is reduced. In order to accurately evaluate the complex UWOC channel information, many scholars have studied the absorption, scattering and turbulence effects in different water areas and characterized these effects. This greatly improves the accuracy of UWOC channel modeling and channel estimation. However, due to the complexity of underwater environment, the channel state information estimated by traditional methods is usually not accurate enough and the recovered signals have high bit error rate (BER). Based on the above research, this paper designs a scheme of underwater optical communication channel estimation (CE) and signal detection (SD) aided by deep learning method.MethodsThis paper presents an end-to-end solution to the challenging CE and SD problems in UWOC systems using a deep neural network (DNN). Firstly, the scheme uses optical orthogonal frequency division multiplexing (OOFDM) system as the system model, and classical UWOC channel as the channel model. Then the DNN model is built according to the channel characteristics of UWOC and the DNN is trained off-line using the simulated data under different UWOC channels. The scheme combines CE and SD in UWOC system and uses DNN, and it can directly estimate and detect the transmitted signal. Finally, the performance of DNN scheme is compared with those of traditional methods including least squares (LS) and minimum mean square error (MMSE) in different water areas, and the influence of pilot number, cyclic prefix (CP) and transmission distance on the detection performance of DNN scheme in OOFDM system is discussed.Results and DiscussionsIn this paper, the following simulations are carried out to prove the performance of the proposed DNN method CE and SD scheme. The DNN model proposed is trained and tested using two types of water body data, and the BER performance of DNN under different signal-to-noise ratios (SNRs) is compared with those of traditional LS and MMSE estimation methods. When the number of pilots is 64, the BER performance of the proposed DNN method is better than that of the traditional LS method and comparable to that of the MMSE method. However, when the pilot number is reduced to 8, in order to improve the spectral efficiency, the performance of the MMSE method is significantly reduced, and the performance of DNN method is better than those of the two traditional methods (Fig. 5). In OOFDM system, CP is usually added to eliminate the inter-symbol interference, but it also increases the bandwidth and energy loss. In the following simulation, we remove CP and compare the performance of DNN method with traditional methods in different water areas. The simulation results show that the traditional LS and MMSE methods cannot evaluate the channel effectively any more, while the DNN method is still effective (Fig. 6). In the underwater environment, with the increase of the transmission distance, the energy loss of the optical signal is caused by the absorption and scattering effects. In order to characterize the performance of the DNN method at different transmission distances, the transmission distance is set to 12 m and 15 m respectively in simulation. When the transmission distance is 12 m, the performance of DNN method is much better than LS method but slightly worse than MMSE method. When the transmission distance is 15 m, the DNN method outperforms the MMSE method (Fig. 7). It is shown that DNN-based CE and SD scheme has high research value in long-distance UWOC.ConclusionsIn this paper, a CE and SD scheme of OOFDM system based on DNN is designed for UWOC system. After the DNN model is trained off-line under several classic water types, it can be deployed in OOFDM system to estimate and restore the received signal. Simulation results show that the DNN method presented in this paper performs well in CE and SD under different UWOC channels. In addition, DNN method has advantages in detection accuracy, and it is also more robust in comparison with traditional LS and MMSE methods. The DNN model is proven powerful in dealing with CE and SD in UOWC system. The proposed method is a new approach to studying CE and SD in UWOC, and its feasibility is verified by simulation.