Author details

The robustness of this technique is evaluated and the simulated results show that the correct modulation identification scheme is possible even at all channel parameters ranges: OSNR between 10 and 35 dB, CD from 85 to 510 ps/nm and DGD in the range of 0–20 ps for all modulation formats mentioned previously. We reached higher identification accuracies at high bit rates which facilitates the performance monitoring process (OPM) for the network

From the results obtained, it is found that the method employing the amplitude histograms is simple and more robust to impairments link, than the other one using CWT, for features extraction. Moreover, the asynchronous amplitude histograms (AAH) generation followed by ANN is rapid on time response compared to the method using CWT for features extraction. Besides the

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

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 vari-

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.

modulation formats, identification ambiguities is more apparent for the second method.

conjunction with features extraction approaches and digital signal processing.

ous modulation formats at different bit rates with high accuracies.

management.

26 Advanced Applications for Artificial Neural Networks

4. Summary

Latifa Guesmi<sup>1</sup> \*, Habib Fathallah2,3 and Mourad Menif<sup>1</sup>

\*Address all correspondence to: latifa.guesmi@supcom.tn

1 GresCom Laboratory, University of Carthage, High School of Communication of Tunis (Sup'Com), Ghazala Technopark, Ariana, Tunisia

2 Computer Department, College of Science of Bizerte, University of Carthage, Tunis, Tunisia

3 KACST-TIC in Radio Frequency and Photonics for e-Society, King Saud University, Riyadh, Saudi Arabia
