Semi-supervised Learning for Automatic Modulation Recognition Using Haar Time–Frequency Mask and Positional–Spatial Attention

Figure 6

Sample distribution of (a) instantaneous statistical features[47], (b) entropy features [48], and (c) high-dimensional features (HTF-PSA-SSL) after using t-SNE to visualize features under 12 dB signals. The high-dimensional features of HTF-PSA-SSL are well-aggregated, while the instantaneous statistical features and entropy features are heavily scattered.

(a)

(b)

(c)