Table 3 DL model architecture
From: Modified state activation functions of deep learning-based SC-FDMA channel equalization system
Parameter | Value |
|---|---|
Sequence input size | 128 |
LSTM layer size (No. of Hidden Units) | 128 |
Fully connected layer size (No. of Classes) | 256 |
Loss function | Default (cross-entropy) and Sum Squared Errors (SSE) |
Mini batch size | 1000 |
Numbers of Epochs | 6 |
Optimization approaches | Adam, RMSProp, and SGdm |
Gate Activation Function (GAF) | Sigmoid |
State Activation Function (SAF) | From Table 1 |
Training Options | Â |
Initial Learning Rate | 0.05 |
Learning Rate Drop Factor | 0.8 |