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

Table 4

The recognition accuracy of HTF-PSA-SSL on public datasets RML2016.10a, RML2016.10b and RML2016.04c with 1% (88) labeled samples.

SNR (dB)	RML2016.10a (%)	RML2016.10b (%)	RML2016.04c (%)
−20	9.18	10.27	8.91
−18	9.09	11.05	8.68
−16	11.18	14.43	8.87
−14	14.45	18.78	9.09
−12	20.00	23.35	12.10
−10	27.09	30.23	17.41
−8	34.91	40.62	21.11
−6	53.09	51.27	36.54
−4	71.27	71.98	67.41
−2	81.18	83.28	79.14
0	87.64	87.90	86.05
2	89.45	89.58	89.88
4	90.45	91.13	91.23
6	92.09	90.02	90.74
8	92.45	90.62	89.63
10	93.18	91.87	91.23
12	92.09	91.75	91.73
14	92.45	91.72	88.27
16	93.18	92.72	89.38
18	92.36	93.25	91.48