Performance Comparison of Different Rotated QAM Based DMT Schemes in Underwater Optical Wireless Communication

In the paper, different rotated QAM-based DMT schemes are proposed and experimentally demonstrated. To overcome the severe high-frequency fading in the underwater optical wireless communication (UOWC) channel, a deviation correction-based maximum likelihood detection (DC-MLD) is introduced into the rotated-QAM scheme. It can improve the precision of the de-rotated process and achieve a better performance than the MLD. Besides, other rotated-QAM schemes using machine learning, including k-means clustering, support vector machine (SVM), and logistic regression (LR), are investigated. These schemes do not need channel state information to achieve the de-rotated process. Meanwhile, the training overhead and execution time of SVM and LR are compared. The experimental results show that, after 5-m air and 4.2-m water channel, compared with the 16QAM DMT, the rotated-16QAM DMT can improve more than 2.5dB received optical power (ROP) sensitivity at the bit error rate (BER) of 1e-3 under the data rate of 3 Gb/s. Meanwhile, the proposed rotated-16QAM scheme with DC-MLD can save more 0.8dB ROP than that with MLD at BER of 2e-4. Besides, it has better performance than the rotated-16QAM DMT with k-means, LR, and SVM.

Automated summary

This paper proposes and experimentally demonstrates different rotated QAM-based DMT schemes. To overcome high-frequency fading in the underwater optical wireless communication (UOWC) channel, a deviation correction-based maximum likelihood detection (DC-MLD) is introduced. Other rotated-QAM schemes using machine learning, such as k-means clustering, support vector machine (SVM), and logistic regression (LR), are investigated. The experimental results show that the rotated-QAM schemes outperform the traditional 16QAM DMT in terms of received optical power (ROP) sensitivity and bit error rate (BER).