The RML2025 Series dataset is a benchmark collection specifically designed for evaluating the generalization ability of automatic modulation classification (AMC) model in real communication environments with significant distribution differences.
We simulate a variety of physical impairments commonly present in real-world communication systems. These impairments are combined in a structured manner to generate domain shifts with hierarchical complexity, offering a more challenging and realistic benchmark for evaluating cross-domain AMC methods.
These impairments, including various combinations of noise, multipath fading, carrier frequency offset (CFO), sampling rate offset (SRO), and Doppler shift, introduce significant distributional shifts and structural damage in the received signals, thereby posing considerable challenges for robust modulation classification.
Each dataset in the RML2025 Series contains:
- 11 modulation types:
8PSK, AM-DSB, AM-SSB, BPSK, CPFSK, GFSK, PAM4, QAM16, QAM64, QPSK, WB-FM - 20 SNR levels:
From -20 dB to +18 dB in steps of 2 dB - 220,000 total samples per subset
(1000 samples × 11 modulations × 20 SNR levels)
Each sample consists of 128 complex-valued IQ points, and all samples are energy-normalized to ensure consistency across channel conditions.
The RML2025 Series includes 7 subsets, grouped by channel type and impairment severity:
| Channel | Subset Name | Impairments |
|---|---|---|
| AWGN | AWGN |
Gaussian noise only |
| Rician | Ri1 |
Rician fading |
| Rician | Ri2 |
+ CFO + SRO |
| Rician | Ri3 |
+ stronger multipath + Doppler |
| Rayleigh | Ray1 |
Rayleigh fading |
| Rayleigh | Ray2 |
+ CFO + SRO |
| Rayleigh | Ray3 |
+ stronger multipath + Doppler |
The following table summarizes the specific impairment parameters across different subsets:
Channel Impairment Summary
- MDS (Hz): Maximum Doppler Shift
- Delay (ms): Multipath delay profile
- Gains: Corresponding amplitude weights of multipath components
- CFO / SRO: Carrier Frequency Offset and Sampling Rate Offset; specified as mean (std)
All Rician and Rayleigh channels use fixed delay/gain profiles with increasing Doppler, CFO, and SRO levels for more realistic simulation of mobile or non-stationary conditions.
The following table summarizes the impairment configurations for each dataset variant:
| Dataset | MDS (Hz) | Delay (ms) | Gains | CFO / SRO |
|---|---|---|---|---|
| AWGN | / | / | / | / |
| Ri1 | 4 | [0, 0.9, 1.7] | [1, 0.8, 0.3] | / |
| Ri2 | 4 | [0, 0.9, 1.7] | [1, 0.8, 0.3] | 50 (std 0.01) |
| Ri3 | 30 | [0, 0.05, 0.12, 0.2, 0.23, 0.5, 1.6, 2.3, 5] | [0.8, 0.8, 0.8, 0.7, 0.6, 0.4, 0.4, 0.4, 0.3] | 50 (std 0.01) |
| Ray1 | 1 | [0, 0.9, 1.7] | [1, 0.8, 0.3] | / |
| Ray2 | 3 | [0, 0.9, 1.7] | [1, 0.8, 0.3] | 50 (std 0.01) |
| Ray3 | 30 | [0, 0.05, 0.12, 0.2, 0.23, 0.5, 1.6, 2.3, 5] | [0.8, 0.8, 0.8, 0.7, 0.6, 0.4, 0.4, 0.4, 0.3] | 50 (std 0.01) |
- MDS: Maximum Doppler Shift
- CFO / SRO: Carrier Frequency Offset and Sampling Rate Offset
- All time delays and gain values are normalized and represent the tap profiles in multipath fading scenarios.
You can download the RML2025 Series datasets from the following sources:
- Google Drive: Download Link
- Baidu Netdisk: Download Link (Extraction Code:
fire)
with open(path, 'rb') as file:
Xd = pickle.load(file,encoding='bytes')
snrs,mods = map(lambda j: sorted(list(set(map(lambda x: x[j], Xd.keys())))), [1,0])
X = []
lbl = []
for mod in mods:
for snr in snrs:
X.append(Xd[(mod,snr)])
for i in range(Xd[(mod,snr)].shape[0]): lbl.append((mod,snr))
X = np.vstack(X)This README file was prepared with the assistance of AI tools to enhance clarity and quality. All technical content and dataset descriptions were reviewed and verified by the authors to ensure accuracy.