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

Table 3

Ablation study on public dataset RML2016.10a.

Description	−20 dB (%)	−16 dB (%)	−12 dB (%)	−8 dB (%)	−4 dB (%)	0 dB (%)	4 dB (%)	8 dB (%)	12 dB (%)	16 dB (%)
	8.82	10.09	13.45	22.09	45.64	66.91	72.36	75.00	75.18	77.27
+	8.82	10.45	17.82	32.91	70.91	87.36	89.27	91.96	92.00	92.64
CNN5	7.91	9.55	14.09	29.00	48.64	73.96	80.37	82.87	83.27	82.96
CNN5 + spatial	8.09	10.36	15.00	29.60	60.64	81.36	83.73	84.32	85.96	85.27
CNN5 + positional	9.09	11.00	18.45	33.05	69.36	83.91	88.82	89.73	89.55	89.03
Eps13 [44]	8.73	9.18	15.27	25.73	48.96	68.18	72.46	77.68	81.55	81.36
Eps13 + frequency mask	9.00	10.64	15.91	29.77	55.09	75.00	81.82	82.33	84.82	83.45
Eps13 + time mask	8.27	10.46	17.64	32.82	60.64	76.32	84.48	84.82	85.03	83.36
HTF-PSA-SSL	9.18	11.18	20.00	34.91	71.27	87.64	90.45	92.45	92.09	93.18

Note. Bold values indicate the result of HTF-PSA-SSL.