4. Summary

In this chapter, two cost-effective techniques for intensity and phase-modulated systems have been proposed and demonstrated. Both methods employ ANN for pattern recognition in conjunction with features extraction approaches and digital signal processing.

In the first method, asynchronous amplitude sampling is the features extraction method. For high-speed optical communications, new approach using ANN trained by the features of linear optical sampling is implemented. For the demonstration of the proposed method, 10 Gbps NRZ-OOK, 40 Gbps NRZ-DQPSK, 100 Gbps NRZ DP-QPSK, 160 Gbps DP-16QAM and 1 Tbps WDM-Nyquist NRZ-DP-QPSK modulation formats are considered. The efficiency of this technique is demonstrated in the presence of different transmission link parameters, such as CD, DGD and OSNR. Simulation results demonstrate successful recognition from a known bit rates with higher estimation accuracy, which exceeds 99.8%. For this method, asynchronous sampling with a rate greater than symbol rates is successfully utilized to have the maximum features for each received signal. Thereby, due to the simplicity of ANN implementation and the use of only amplitude samples, the proposed techniques enable the identification of various modulation formats at different bit rates with high accuracies.

The second technique presents a new achievement using the continuous wavelet transform (CWT) for features extraction. It offers the best time and frequency localization. In that case, Haar wavelet and SVD followed by ANN pattern-recognition are used to achieve the classification process. This method is advantageous because its cost effectiveness and its flexibility. To demonstrate the validity of this technique, we consider the classification of 40 Gbps NRZ-OOK, and three multi-carriers modulation schemes such as 160 Gbps OFDM DP-16QAM, 400 Gbps DC-PDM-QPSK and 1 Tbps WDM-Nyquist NRZ-DP-QPSK. The effect of each channel parameter to the probability of recognition has been also observed. In particular, it has been found that the correct identification was observed at higher OSNR values. While, an increase of CD and DGD affects the accuracy of recognition and limits the measurement ranges. Despite the presence of link impairments, and because the CWT is resistant to the noise in the signal, the classification of these formats remains possible with good precision.
