| 176 | 176 | 2. Since Tb values of all samples from a single RFI scenario (band, fc, rb, gain) are uniformly distributed, we ordered the Tb values obtained from the reference testbed L1B data from lowest to highest, and matched them with the 10 spectrograms generated with different marginalized signal gain values (uniform from -0.276 to +0.317) that simulate variation in physical transmission |
| 177 | 177 | 3. With this method, we are able to obtain more data points to train and test the CNN, instead of using the exact gain values (-20, -10, 0, 10, 20) to form only one sample per scenario, leaving SN2 to SN10 from the L1B data unused |
