Table 3 The separation results of different optimizers in the form of \(\zeta\) and \(\textrm{PI}\) with SNR=10dB
From: A MLE-based blind signal separation method for time–frequency overlapped signal using neural network
Optimizer | SGD | SGDM | Adagrad | Adam | Adadelta | RMSprop | RMSpropM |
|---|---|---|---|---|---|---|---|
Parameter setting | \(\eta = 2.0\) | \(\begin{matrix}\eta = 0.8\\ \rho =0.89\end{matrix}\) | \(\begin{matrix} \eta = 5.0\\ \epsilon =10^{-8}\end{matrix}\) | \(\begin{matrix} \eta = 1.9\\ \beta _{1} =0.845\\ \beta _{2} =0.89\\ \epsilon =10^{-8}\end{matrix}\) | \(\begin{matrix}\eta = 152\\ \rho = 0.1\\ \epsilon =10^{-6} \end{matrix}\) | \(\begin{matrix}\eta = 0.24\\ \rho =0.24\\ \epsilon =10^{-8} \end{matrix}\) | \(\begin{matrix}\eta = 0.2\\ \rho =0.2\\ \beta =0.3\\ \epsilon =10^{-8} \end{matrix}\) |
Correlation coefficient (\(\zeta\)) | \(\begin{matrix}0.7112\\ 0.7119\\ 0.7264\end{matrix}\) | \(\begin{matrix}0.7522\\ 0.7270\\ 0.7696\end{matrix}\) | \(\begin{matrix}0.7398\\ 0.7335\\ 0.7564\end{matrix}\) | \(\begin{matrix}0.8865\\ 0.8868\\ 0.8852\end{matrix}\) | \(\begin{matrix}0.8865\\ 0.8868\\ 0.8853\end{matrix}\) | \(\begin{matrix}0.8850\\ 0.8854\\ 0.8853\end{matrix}\) | \(\begin{matrix}0.8864\\ 0.8868\\ 0.8852\end{matrix}\) |
Performance index (\(\textrm{PI}\)) | 0.9680 | 0.8029 | 0.7667 | 0.0239 | 0.0239 | 0.0302 | 0.0249 |