Noise Floor Estimation Based on Deep CNNs
Hao Huang, Jianqing Li, Jiao Wang, Hong Wang · 2020
This paper proposed a new method for noise estimation based on deep learning. We treat the wideband power spectrum as a one-dimensional (1-D) gray image and regard the noise floor estimation problem as a curve regression task. We design an end-to-end deep learning model based on convolutional neural networks (CNNs) to accomplish the task. By using sufficient numbers of simulation noise floor labeled spectra samples to train the model, experimental results show that our model can effectively regress the noise floor of the wideband power spectra. Comparing to the nonlinear recursive smoothing filter method, our method not only can be suitable for the single narrow carrier signal noise floor estimation but also gain good results when there exist multiple carriers in the wideband power spectrum.