Algorithm 3: Find optimal ${\alpha }_1$ and ${\alpha }_2$.

/* Load data on Fading channel  ;                          */ 1 snr_k ← post-processing SINRs, bler ← BLER  /* Calculate ${\gamma }_{\mathrm{eff}}$ with ${\alpha }_1$ and ${\alpha }_2$                            */ 2 snr_eff = -1 * alpha1 * tf.log(tf.reduce_mean (tf.exp(-1*snr_k/alpha2), axis = 1))  /* Decide target SNR by regression                           */ 3 target_snr ← predicted snr corresponding to BLER  /* Calculate loss function                               */ 4 loss = tf.reduce_sum(tf.abs(tf.subtract (target_snr,snr_eff)))  /* Select a training algorithm between Adagrad and RMSprop; Adagrad is  selected in this case.                                 */ 5 train = tf.train.AdagradOptimizer(0.1).minimize(loss)  /* train = tf.train.RMSPropOptimizer(0.1).minimize(loss) ← when RMSProp is  selected.                                        */  /* Training data 4000 times                              */ 6 with tf.Session() as sess: 7 sess.run(init) 8 for i in range(4000): 9  sess.run(train)  /* Calculate MSE in test data set                           */ 10 regressed_bler ← estimated BLER, y_data ← BLER 11 mse = np.mean(np.square(np.subtract(np.asarray(y_data), np.asarray(regressed_bler))))