Organizations: College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China. · College of Advanced Interdisciplinary Studies, National University of Defense Technology, Changsha 410073, China. · College of Science, National University of Defense Technology, Changsha 410073, China. · Hunan Research Center of the Basic Discipline for Physical States, National University of Defense Technology, Changsha 410073, China.
As cognitive radars evolve towards the end-to-end processing of one-dimensional complex baseband sequences, deep networks face severe adversarial threats. However, existing attacks encounter dual limitations in practical electronic warfare: low cross-architecture transferability and severe vulnerability to physical channel distortions, which degrades their tactical reliability. To overcome these bottlenecks, we propose a physical prior-driven Segmented Spatio-Temporal Attack (SSTA). First, a global-local feature decoupling segmentation mechanism interrupts the adversarial gradients' dependence on long-range sequence coherence, bridging the cross-architecture feature gap. Second, moving beyond static norm constraints devoid of physical semantics, we abstract channel distortions into differentiable spatio-temporal operators, integrating them into momentum iterations via an Expectation over Transformation (EoT) strategy. Comprehensive cross-evaluations across 9 heterogeneous models spanning CNNs, RNNs, and Transformers demonstrate that SSTA outperforms the state-of-the-art baseline by 10.36% in Average Black-Box Attack Success Rate. Furthermore, baseband equivalent simulations confirm that the generated waveforms survive extreme physical distortions and, after matched filtering, achieve zero main-lobe attenuation (0 dB) with noise floor lift strictly within Constant False Alarm Rate tolerances. SSTA successfully unifies digital-domain deception and physical-domain fidelity, providing crucial support for CEW security assessment.
Figures & tables
Fig. 1: The overall framework of Segmented Spatio-Temporal Attack.
Jamming Type
Key Parameter Range
Modulation Characteristics
DFTJ
False targets: 3–6, Delay: 1–10 μs
Multi-pulse linear time shifting
ISRJ
Retransmissions: 1–4, Sampling period: 5–20 μs
Periodic spectral notches
RDJ
Lag time: 1–10 μs
Constant frequency shift + time delay extension
SNJ
Duty cycle: 20%–50%, Delay: 1–5 μs
Intermittent sampled noise retransmission
SPJ
Bandwidth: 20–40 MHz
Narrowband spectrum dwelling
SWJ
Sweep period: 40–80 μs , Bandwidth: 20 MHz
Time-varying sloped sawtooth spectrum
TABLE I: Summary of Jamming Signals in the Simulation Dataset
Algorithm
ResNet18
DenseNet121
EfficientNet
SEResNet
DeepLSTM
DeepGRU
CNN-BiLSTM
SwinTrans
Conformer
Average BB-ASR
FGSM [ 40 ]
34.89
18.29
45.71
41.34
51.46
46.20
36.06
38.62
51.81
40.49
MI-FGSM [ 14 ]
47.79
24.50
39.04
55.62
50.19
49.86
32.58
33.24
59.64
43.61
VMI-FGSM [ 16 ]
63.80
62.44
44.69
63.98
61.01
66.39
43.56
32.60
71.40
56.65
DIM [ 17 ]
49.54
22.81
50.02
55.95
43.33
53.81
24.98
33.62
65.25
44.37
TIM [ 18 ]
28.59
13.78
31.52
31.48
30.39
28.48
27.57
38.59
42.58
30.33
SIA [ 19 ]
53.69
48.84
51.78
57.92
53.01
62.60
33.10
35.60
52.55
49.90
TABLE II: Comprehensive Black-Box Transfer Attack Success Rates (%) Across 9 Heterogeneous Models
Fig. 2: Heatmap of transfer attack success rates across the 9×9 cross-evaluation matrix.
Transfer Path
MI-FGSM
SIA
VMI-FGSM
SSTA
CNN → RNN
47.41
61.66
63.34
76.85
CNN → Trans
20.30
25.85
33.55
52.73
RNN → CNN
44.31
51.02
58.55
60.07
RNN → Trans
26.72
30.90
36.95
42.55
Trans → CNN
45.81
42.18
51.38
66.50
Trans → RNN
54.95
54.65
60.37
71.17
TABLE III: Transfer Attack Success Rates (%) Across Different Feature Extraction Mechanisms
Fig. 3: Radar chart of cross-architecture transfer attack success rates on six core paths.
Variant
AWB-ASR (%)
ABB-ASR (%)
Gap (%)
Baseline (MI-FGSM)
93.11
43.89
49.22
V1: Time-Only
86.00
52.07
33.93
V2: Space-Only
91.44
65.19
26.25
V3: Time+Phase
82.56
57.22
25.34
V4: Time+JNR
75.44
58.69
16.75
SSTA (Full)
85.00
67.46
17.54
TABLE IV: Performance Comparison of Different Ablation Variants
Fig. 4: Performance comparison trends of different ablation variants.
Fig. 5: Evolution curves of Average BB-ASR with respect to the iteration number K .
Segments ( N )
1
2
3
5
7
10
AWB-ASR
64.33
81.22
85.00
85.78
87.00
87.56
ABB-ASR
47.26
64.31
67.46
68.68
69.61
69.71
TABLE V: Attack Success Rates (%) Under Different Segmentation Counts N
Fig. 6: Evolution dynamics of transfer success rates across different architectural paths with respect to segmentation count N .
EoT Samplings ( T )
1
5
10
15
20
AWB-ASR (%)
77.67
82.00
85.00
84.33
85.44
ABB-ASR (%)
59.86
65.63
67.46
68.63
69.39
Total Time (s)
495.75
544.93
720.61
1158.04
1550.88
Avg Time (s)
0.55
0.61
0.80
1.29
1.72
TABLE VI: Performance and Overhead Under Different EoT Sampling Counts T
Fig. 7: Comprehensive evaluation of EoT sampling count T : (a) Boxplot of cross-model transfer rate variance; (b) Trade-off between worst-case transfer rate and time cost.
Fig. 8: t-SNE dimensionality reduction visualization of hidden feature manifolds under different cross-architecture transfer paths.
Fig. 9: Dynamic evolution trajectories of cross-model gradient cosine similarity during the iteration process.
Fig. 10: Effective ASR degradation curves under extreme physical distortions: clock drift, phase rotation, and AWGN channels.
Fig. 11: Normalized time-domain envelope comparison between the clean DFTJ signal and the SSTA adversarial example.
Key Laboratory for Information Science of Electromagnetic Waves (MoE), School of Information Science and Technology, Fudan University, Shanghai 200433, China
Department of Electronics and Communication Engineering National Institute of Technology Calicut, India · Department of Electrical Engineering Indian Institute of Technology Madras, Chennai, India
Shanxi Normal University, Taiyuan, China · Guizhou University, Guiyang, China · Shanxi Key Laboratory of Cryptography and Data Security, School of Computer Science and Artificial Intelligence, Shanxi Normal University, Taiyuan 030031, China +15