An Introduction to Deep Learning for the Physical Layer
Open accessThe DOI matches its Crossref record; 90 of the 92 references it cites were found in a registry or an index, and 2 are not checked yet; 8 of the 32 sources its claims cite were checked for support.
Abstract
We present and discuss several novel applications of deep learning for the physical layer. By interpreting a communications system as an autoencoder, we develop a fundamental new way to think about communications system design as an end-to-end reconstruction task that seeks to jointly optimize transmitter and receiver components in a single process. We show how this idea can be extended to networks of multiple transmitters and receivers and present the concept of radio transformer networks as a means to incorporate expert domain knowledge in the machine learning model. Lastly, we demonstrate the application of convolutional neural networks on raw IQ samples for modulation classification which achieves competitive accuracy with respect to traditional schemes relying on expert features. This paper is concluded with a discussion of open challenges and areas for future investigation.
Verified 2026-11-19 · 2 references not checked yet
Overview
CoChat Summary
TL;DR
The paper introduces deep learning for the physical layer by treating a communications system as an autoencoder that jointly learns transmitter and receiver implementations end-to-end. It also extends this idea to multi-user interference channels, proposes radio transformer networks (RTNs) to inject expert domain knowledge, and shows CNNs on raw IQ samples can outperform expert-feature modulation classifiers.
Key findings
- A learned autoencoder for a (7,4) system at rate R = 4/7 achieves the same block error rate (BLER) as a Hamming (7,4) code with maximum likelihood decoding, while an (8,8) autoencoder outperforms uncoded BPSK over the full E_b/N_0 range, indicating learned joint coding and modulation.
- For the two-user AWGN interference channel, the autoencoder matches time-sharing for (1,1) and (2,2), but achieves gains of about 0.7 dB for (4,4) and 1 dB for (4,8) at a BLER of 10^-3.
- An autoencoder (8,4) with an RTN over a multipath fading channel with L = 3 Rayleigh taps outperforms differential BPSK with maximum likelihood sequence estimation and a Hamming (7,4) code, whereas the plain autoencoder struggles to match it; RTNs also show faster training convergence.
- For modulation classification on 1.2M sequences of 128 complex-valued IQ samples across ten modulation schemes and SNRs from -20 dB to 18 dB, the CNN outperforms a boosted expert-feature classifier by around 4 dB in the low-to-medium SNR range, with almost identical high-SNR performance; a single tree is about 6 dB worse at medium SNR and 3.5% worse at high SNR.
Methods
- The transmitter, channel, and receiver are represented as one deep neural network trained as an autoencoder: one-hot message input, dense layers, normalization layer for transmit constraints, additive noise channel layer, and softmax receiver output, trained with categorical cross-entropy using SGD/Adam at learning rate 0.001 and fixed E_b/N_0 = 7 dB.
- The autoencoder concept is extended to multiple transmitter-receiver pairs in a two-user interference channel, with dynamic loss weights α_t per mini-batch to balance the two cross-entropy losses; performance is compared against 2^{2k/n}-QAM with time-sharing.
- RTNs combine a learned parameter estimator, a differentiable parametric transform, and a discriminative network; the paper demonstrates phase-offset compensation and multipath inverse filtering. For modulation classification, a VGG-like CNN with 324,330 trainable parameters is trained on the RML2016.10b dataset.
Limitations
- The paper states that scalability to large message sets M is a major challenge: for k = 100 bits, training complexity is prohibitive because the autoencoder must see at least every message once.
- The paper states that optimal data representations, loss functions, training SNR, and hyperparameters are not well understood; training at one SNR does not necessarily generalize to others, and no common communications benchmarks or datasets comparable to MNIST or ImageNet exist.
- The paper states that end-to-end learning over real channels requires system identification because backpropagation assumes a known channel transfer function; complex-valued neural networks are not supported by current DL libraries. Evident from the text: the reported results are simulation-based, use small (n,k) settings, assume known channel models during training, and the paper explicitly leaves detailed performance comparisons for various channel types and parameters to future work.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures











Verification
Claims the paper makes
6 Valid · 4 Not valid · 18 Not checked · 65 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“For instance in [22], it was shown that training with a single codeword is sufficient since the structure of the code is embedded in the NN through the Tanner graph.”
What the sources say
- Supportedcites [22] [22] E. Nachmani, Y. Be'ery, and D. Burshtein, 'Learning to decode linear codes using deep learning,' in Proc. IEEE Annu. Allerton Conf. Commun., Control, and Computing (Allerton) , 2016, pp. 341-346. - source
The Introduction says the decoder uses a Tanner graph representing the code and that one codeword suffices for training. Section III explains the precise condition: over a binary memoryless symmetric channel, the decoder’s error rate is independent of the transmitted codeword, so training can use noisy versions of a single codeword. (LLM, conf=0.91)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We study the use of NNs on complex-valued IQ samples for the problem of modulation classification and show that convolutional neural networks (CNNs), which are the cornerstone of most DL systems for computer vision, can outperform traditional classification techniques based on expert features.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Training at low SNR for instance does not allow for the discovery of structure important in higher SNR scenarios.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Another advantage of RTNs is faster training convergence which can be seen from Fig. 10 that compares the validation loss of the autoencoder with and without RTN as a function of the training epochs.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Similarly, Fig. 4b shows a (4 , 2) system which leads to a rotated 16-PSK constellation.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
39 more claims
Create a free account to see all 93 claims and to verify the 4 references still open.
- Probably not a substantial claim. 29
- It may state two separate claims as one. 24
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A. Autoencoders for end-to-end communications systems · p. 5
“Similarly, Fig. 4b shows a (4 , 2) system which leads to a rotated 16-PSK constellation.”
§ A. Autoencoders for end-to-end communications systems · p. 6
“Figure 4: Constellations produced by autoencoders using parameters ( n, k ) : (a) (2 , 2) (b) (2 , 4) , (c) (2 , 4) with average power constraint, (d) (7 , 4) 2-dimensional t-SNE embedding of received symbols.”
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“dense/fully-connected layers and terminated with a dense softmax layer for our classifier (similar to a VGG architecture [55]).”
BackgroundAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 4×
“Training was done at a fixed value of Eb/N0 = 7 dB (see Section IV-B) using Adam [46] with learning rate 0.”
BackgroundMethodInfluentialA Mathematical Theory of Communication
Shannon, C. E. (1948) · Bell System Technical Journal · doi:10.1002/j.1538-7305.1948.tb01338.x · Cited by 88,486
Scikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 1×
“11, we compare the classification accuracy of the CNN against that of extreme gradient boosting7 with 1000 estimators, as well as a single scikit-learn decision tree [56], operating on a mix of 16 analog and cumulant expert features as proposed in [53] and [57].”
Method(2014) · Cited by 33,419
(2008) · Journal of Machine Learning Research · Cited by 33,051
Human-level control through deep reinforcement learning
Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. (2015) · Nature · doi:10.1038/nature14236 · Cited by 30,357
(2007) · Electronics: A First Course · doi:10.4324/9780080477527-64 · Cited by 24,923
Pan, S. J., & Yang, Q. (2010) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2009.191 · Cited by 23,387 · cited 1×
“learning [64] is one appealing candidate for adapting an endto-end communications system trained on a statistical model to”
BackgroundMultilayer feedforward networks are universal approximators
Hornik, K., Stinchcombe, M., & White, H. (1989) · Neural Networks · doi:10.1016/0893-6080(89)90020-8 · Cited by 21,728 · cited 1×
“Third, it has been shown that NNs are universal function approximators [13] and recent work has shown a remarkable”
Background
Research works printed but not linked: 2
- [35] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning . MIT Press, 2016. -
- [52] J. Estaran et al. , 'Artificial neural networks for linear and non-linear impairment mitigation in high-baudrate IM/DD systems,' in Proc. 42nd European Conf. Optical Commun. (ECOC) . VDE, 2016, pp. 1-3. -
Cited by
Items 301–400 of 2,995
Wang, H., Wu, Z., Ma, S., Lu, S., Zhang, H., Ding, G., & Li, S. (2019) · IEEE Access · doi:10.1109/access.2019.2903130 · Cited by 67
“Recent works [12], [13] have interpreted an end-to-end wireless communication system as an autoencoder.”
BackgroundA Deep-Neural-Network-Based Relay Selection Scheme in Wireless-Powered Cognitive IoT Networks
Nguyen, T.-V., Tran, T., Shim, K., Huynh‐The, T., & An, B. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3038907 · Cited by 67
“To avoid overfitting, L 2 regularization is adopted in the DNN model by penalizing the weights forcing them to take smaller values, thus preventing notably small and large values for any weight [15].”
BackgroundMethodDeep Learning based Semantic Communications: An Initial Investigation
Xie, H., Qin, Z., Li, G. Y., & Juang, B.-H. F. (2020) · Globecom · doi:10.1109/globecom42002.2020.9322296 · Cited by 65
“It has been demonstrated that such an E2E system outperforms uncoded binary phase shift keying (BPSK) and Hamming coded BPSK in terms of BER [7].”
BackgroundLearning for Detection: MIMO-OFDM Symbol Detection Through Downlink Pilots
Zhou, Z., Liu, L., & Chang, H.-H. (2020) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2020.2976004 · Cited by 75
“For instance, an auto-encoder is introduced in [8], [9] to conduct”
MethodD²-JSCC: Digital Deep Joint Source-Channel Coding for Semantic Communications
Huang, J., Yuan, K., Huang, C. Z., & Huang, K. (2025) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2025.3531546 · Cited by 95
“It can be proved that the commonly used NN layers, such as fully connected and convolutional layers, as well as activation functions like ReLU and Sigmoid are Lipschitz continuous [37], [38].”
Deep Learning for Improving Performance of OOK Modulation Over FSO Turbulent Channels
Darwesh, L., & Kopeika, N. S. (2020) · IEEE Access · doi:10.1109/access.2020.3019113 · Cited by 69
“The authors of [12]–[14] suggested using DL for signal detection in order to replace the receiver in OFDM wireless communication systems.”
MethodCustomer Churn Prediction Modelling Based on Behavioural Patterns Analysis using Deep Learning
Agrawal, S., Das, A., Gaikwad, A. K., & Dhage, S. (2018) · International Conference on Smart Computing and Electronic Enterprise (ICSCEE) · doi:10.1109/icscee.2018.8538420 · Cited by 69
“[15] T.”
BackgroundTargeted Adversarial Examples Against RF Deep Classifiers
Kokalj-Filipović, S., Miller, R., & Morman, J. (2019) · ACM Workshop on Wireless Security and Machine Learning (WiseML) · doi:10.1145/3324921.3328792 · Cited by 64
“The research in the area of RF-based DL of the PHY layer is still embryonic [17].”
BackgroundHybrid Beamforming/Combining for Millimeter Wave MIMO: A Machine Learning Approach
Chen, J., Feng, W., Xing, J., Yang, P., Sobelman, G. E., Lin, D., & Li, S. (2020) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2020.3009746 · Cited by 68
“A channel autoencoder-based deep learning technique for optimizing the transmitter, receiver, and signal representation was first introduced in [40], but without considerations for large-scale MIMO HB systems.”
MethodAn Adaptive and Parameter-Free Recurrent Neural Structure for Wireless Channel Prediction
Zhu, Y., Dong, X., & Lű, T. (2019) · IEEE Transactions on Communications · doi:10.1109/tcomm.2019.2935714 · Cited by 67
“, or viewing the communication system design from a new paradigm as an autoencoder in deep learning [7].”
BackgroundMert, İ. (2020) · International Journal of Hydrogen Energy · doi:10.1016/j.ijhydene.2020.11.161 · Cited by 64
Andrews, J. G., Humphreys, T. E., & Ji, T. (2024) · IEEE BITS the Information Theory Magazine · doi:10.1109/mbits.2024.3504521 · Cited by 117
Real-Time and Embedded Deep Learning on FPGA for RF Signal Classification
Soltani, S., Sagduyu, Y. E., Hasan, R., Davaslıoğlu, K., Deng, H., & Erpek, T. (2019) · IEEE Military Communications Conference (MILCOM) · doi:10.1109/milcom47813.2019.9021098 · Cited by 70
“Examples include spectrum sensing [10], MIMO detection [11], channel estimation and signal detection [12], physical layer communications [13], jammer detection [14], stealth jamming [15], [16], power control [17], signal spoofing [18], and transmitter-receiver scheduling [19].”
BackgroundIntrusion Detection for IoT Devices based on RF Fingerprinting using Deep Learning
Bassey, J., Adesina, D. S., Li, X., Qian, L., Aved, A. J., & Kroecker, T. S. (2019) · International Conference on Fog and Mobile Edge Computing (FMEC) · doi:10.1109/fmec.2019.8795319 · Cited by 75
“Unlike identifiers at other layers such as MAC addresses and International Mobile Subscriber Identity (IMSI) numbers, RF features are very difficult to impersonate [3], [4].”
BackgroundShi, Y., Lu, X., Niu, Y., & Li, Y. (2021) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2021.3064843 · Cited by 62
Two-Dimensional Convolutional Neural Network-Based Signal Detection for OTFS Systems
Enku, Y. K., Bai, B., Wan, F., Guyo, C. U., Tiba, I. N., Zhang, C., & Li, S. (2021) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2021.3106039 · Cited by 67
Reinforcement Learning-powered Semantic Communication via Semantic SimilarityPreprint
Lu, K., Li, R., Chen, X., Zhao, Z., & Zhang, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.12121 · Cited by 89
Is Semantic Communication Secure? A Tale of Multi-Domain Adversarial Attacks
Sagduyu, Y. E., Erpek, T., Ulukuş, Ş., & Yener, A. (2023) · IEEE Communications Magazine · doi:10.1109/mcom.006.2200878 · Cited by 68
“…communications , where the encoder encompasses channel coding and modulation operations, the decoder encompasses demodulation and channel decoding operations, and the sole goal is the same as conventional communications, namely reconstructing messages (in form of symbols) at the receiver [10].”
BackgroundML-Based Massive MIMO Channel Prediction: Does It Work on Real-World Data?
Shehzad, M. K., Rose, L., Wesemann, S., & Assaad, M. (2022) · IEEE Wireless Communications Letters · doi:10.1109/lwc.2022.3146230 · Cited by 62
Jiang, Y., Kannan, S., Kim, H., Oh, S., Asnani, H., & Viswanath, P. (2019) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc.2019.8815400 · Cited by 79
“Using deep learning based methods for channel coding has received sufficient attention since its inception [27] [28].”
MethodIntelligent Massive MIMO Systems for Beyond 5G Networks: An Overview and Future Trends
Elijah, O., Rahim, S. K. A., New, W. K., Leow, C. Y., Cumanan, K., & Geok, T. K. (2022) · IEEE Access · doi:10.1109/access.2022.3208284 · Cited by 78
A Deep Learning Based Channel Estimation Scheme for IEEE 802.11p Systems
Han, S. H., Oh, Y., & Song, C. (2019) · International Conference on Communications · doi:10.1109/icc.2019.8761354 · Cited by 63
“The noise cancellation effect also arises in the estimated CFRs H̃i(k) due to the compression and reconstruction process in the latent space of the AE [9].”
BackgroundA Novel Attention Cooperative Framework for Automatic Modulation Recognition
Chen, S., Zhang, Y., He, Z., Nie, J., & Zhang, W. (2020) · IEEE Access · doi:10.1109/access.2020.2966777 · Cited by 64
“VGG architecture, containing series of narrowing convolutional layers followed by fully-connected layers and terminated with a dense softmax layer, is leveraged in [22].”
BackgroundMethodInfluentialLearn to Compress CSI and Allocate Resources in Vehicular Networks
Wang, L., Ye, H., Liang, L., & Li, G. Y. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2020.2979124 · Cited by 66
“Inspired by its remarkable performance, the wireless community is increasingly interested in leveraging machine learning for the physical layer and resource allocation design [16]–[24].”
BackgroundSaidutta, Y. M., Abdi, A., & Fekri, F. (2021) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2021.3078489 · Cited by 71
Tan, J., Zhang, L., Liang, Y., & Niyato, D. T. (2020) · IEEE Transactions on Communications · doi:10.1109/tcomm.2020.2971212 · Cited by 62
“By now, some work has been done to optimize wireless networks using DL, e.g., transmitter-receiver design [25], channel estimation [26], [27], radio signal classification [28], and spectrum sensing [29].”
BackgroundDeep Reinforcement Learning for Router Selection in Network With Heavy Traffic
Ding, R., Xu, Y., Gao, F., Shen, X., & Wu, W. R. (2019) · IEEE Access · doi:10.1109/access.2019.2904539 · Cited by 62
Ahmad, I., Shahabuddin, S., Sauter, T., Harjula, E., Kumar, T., Meisel, M., Juntti, M. J., & Ylianttila, M. (2020) · IEEE Industrial Electronics Magazine · doi:10.1109/mie.2020.2979272 · Cited by 64
“…challenges Low High High High [45], [62], and [63] Caching and memory High High Medium Low [23] and [39] Network traffic control High High Medium Low [27], [28], and [30] System complexity High High Medium Low [51] and [54] *Challenge levels or severity represented by low, medium, and high.”
BackgroundDeep Learning-Based Autoencoder for m-User Wireless Interference Channel Physical Layer Design
Wu, D., Nekovee, M., & Wang, Y. (2020) · IEEE Access · doi:10.1109/access.2020.3025597 · Cited by 69
“The two-user (two pairs) AE model introduced in [9], where”
BackgroundMethodInfluentialDecentralized Inference With Graph Neural Networks in Wireless Communication Systems
Lee, M., Yu, G., & Dai, H. (2021) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2021.3125793 · Cited by 66
“Recently, a large number of works turn to various ML techniques to facilitate different applications in wireless communication systems, such as physical layer design [2], [3], [4], [5],”
BackgroundGraph Neural Networks for Channel Decoding
Cammerer, S., Hoydis, J., Aoudia, F. A., & Keller, A. (2022) · IEEE Globecom Workshops (GC Wkshps) · doi:10.1109/gcwkshps56602.2022.10008601 · Cited by 61
Self-citationA Learning Approach for Physical Layer Authentication Using Adaptive Neural Network
Qiu, X., Dai, J., & Hayes, M. H., III (2020) · IEEE Access · doi:10.1109/access.2020.2971260 · Cited by 58
“Learning-based approaches provide many advantages over existing physical layer security and authentication [32], and artificial intelligence algorithms have made their way into intrusion detection and confrontation systems [33].”
BackgroundRestuccia, F., & Melodia, T. (2019) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom.2019.8737459 · Cited by 62
“Among other issues, physical-layer modulation recognition through deep learning has received significant attention in the last two years [15, 22–24, 51, 52].”
BackgroundDeep cascading network architecture for robust automatic modulation classification
Weng, L., He, Y., Peng, J., Zheng, J., & Li, X. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.05.010 · Cited by 58
RoemNet: Robust Meta Learning Based Channel Estimation in OFDM Systems
Mao, H., Lu, H., Lu, Y., & Zhu, D. (2019) · International Conference on Communications · doi:10.1109/icc.2019.8761319 · Cited by 62
“Therefore, the 2n-complex-valued vector of Y is cast to 2n-real numbers by considering one half as the real part and the other half as the imaginary part (see [20]), thus the input number corresponds to the number of real parts plus the number of imaginary parts of pilots in OFDM blocks.”
BackgroundMeta-Learning to Communicate: Fast End-to-End Training for Fading Channels
Park, S., Simeone, O., & Kang, J. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053252 · Cited by 56
“Denoting as Uh(θ(0)) = θ(0) − η∇L̂h(θ(0)) the local updated vector for channel h, MAML tackles (7) via SGD through the update (9), which can be computed as θ(0,i+1) ← θ(0,i) − κ ∑ h∈H (Jθ(0)Uh(θ (0,i)))∇Lh(Uh(θ(0,i))) ≈ θ(0,i) − κ ∑ h∈H (I − η∇2L̂h(θ(0,i)))∇L̂′h(Uh(θ(0,i))), (8) for iterations i = 1, 2, . . ., where J…”
BackgroundMethodInfluentialCai, Z., Ma, W., Wang, X., Wang, H., & Feng, Z. (2022) · IEEE Transactions on Reliability · doi:10.1109/tr.2022.3178707 · Cited by 57
“In addition, O’Shea and Hoydis [12] proposed some new applications of deep learning in the physical layer, and the recognition of the signal individual is discussed.”
BackgroundNaeem, M., Pietro, G. D., & Coronato, A. (2021) · Sensors · doi:10.3390/s22010309 · Cited by 67
“The research contribution is an extension of joint optimization [286] of physical layer representation as well as the encoding-decoding processes as a single end-to-end task.”
MethodMachine Learning for Smart Environments in B5G Networks: Connectivity and QoS
Alsamhi, S. H., Almalki, F. A., Al-Dois, H., Soufiene, B. O., Hassan, J. A., Hawbani, A., Sahal, R., Lee, B., & Saleh, H. (2021) · Computational Intelligence and Neuroscience · doi:10.1155/2021/6805151 · Cited by 66
Bair, S., DelVecchio, M., Flowers, B., Michaels, A. J., & Headley, W. C. (2019) · ACM Workshop on Wireless Security and Machine Learning (WiseML) · doi:10.1145/3324921.3328785 · Cited by 55
“While these techniques are still used, Deep Neural Networks (DNN) can also be leveraged to help Spectrum Sensing [12, 16, 17, 22–24] for scenarios in which the underlying likelihoods have no closed form solution or the expert features best suited to the problem are unknown.”
MethodA Survey on Deep Learning Based Channel Estimation in Doubly Dispersive Environments
Gizzini, A. K., & Chafii, M. (2022) · IEEE Access · doi:10.1109/access.2022.3188111 · Cited by 68
Deep learning based transceiver design for multi-colored VLC systems
Lee, H., Lee, I., & Lee, S. H. (2018) · Optics Express · doi:10.1364/oe.26.006222 · Cited by 55
Deep learning-driven wireless communication for edge-cloud computing: opportunities and challenges
Wu, H., Li, X., & Deng, Y. (2020) · Journal of Cloud Computing Advances Systems and Applications · doi:10.1186/s13677-020-00168-9 · Cited by 74
“The results of training deep learning models under certain SNR conditions are often not suitable for other SNR ranges [10].”
BackgroundInfluentialOnline Meta-Learning for Hybrid Model-Based Deep Receivers
Raviv, T., Park, S., Simeone, O., Eldar, Y. C., & Shlezinger, N. (2023) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2023.3241841 · Cited by 75
“This class of methods includes the approach of training a DNN using data corresponding to an extensive set of expected channel conditions, which is referred to as joint learning [23], [39].”
BackgroundMethodMessage Passing Meets Graph Neural Networks: A New Paradigm for Massive MIMO Systems
He, H., Yu, X., Zhang, J., Song, S., & Letaief, K. B. (2023) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2023.3321667 · Cited by 72
“…extracting representative features from data, deep learning (DL) has been recently utilized in the physical layer design of wireless communications [19], [20], [21], [22], such as mmWave channel estimation [23], [24], channel state information (CSI) feedback [25], and data detection [26], [27],…”
Detection of Rogue RF Transmitters using Generative Adversarial Nets
Roy, D., Mukherjee, T., Chatterjee, M., & Pasiliao, E. (2019) · IEEE Wireless Communications and Networking Conference · doi:10.1109/wcnc.2019.8885548 · Cited by 55
“Another study for modulation classification using raw IQ samples was presented in [12].”
BackgroundVu, T. X., Chatzinotas, S., Nguyen, V., Hoang, D. T., Nguyen, D. N., Renzo, M. D., & Ottersten, B. (2021) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2021.3052973 · Cited by 68
“Recently, the use of machine learning (ML) in communications systems has attracted much attention [12]–[24].”
Demir, Ö. T., & Björnson, E. (2019) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2019.2959913 · Cited by 54
“For further details on deep learning, please refer to the references [27], [28].”
BackgroundMethodViterbiNet: Symbol Detection Using a Deep Learning Based Viterbi Algorithm
Shlezinger, N., Eldar, Y. C., Farsad, N., & Goldsmith, A. J. (2019) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc.2019.8815457 · Cited by 60
“Due to the large amount of recent works on this topic, we only discuss a few representative works; detailed surveys can be found in [6]–[8]: The work [9] used deep learning for decoding linear codes.”
BackgroundTeam Deep Neural Networks for Interference Channels
de Kerret, P., Gesbert, D., & Filippone, M. (2018) · IEEE International Conference on Communications workshops · doi:10.1109/iccw.2018.8403662 · Cited by 56
Gaiarin, S., Ros, F. D., Jones, R. T., & Zibar, D. (2020) · Journal of Lightwave Technology · doi:10.1109/jlt.2020.3033624 · Cited by 67
“improved performance or shorter training time [6].”
BackgroundInfluentialWireless Avionics Intracommunications: A Survey of Benefits, Challenges, and Solutions
Park, P., Marco, P. D., Nah, J., & Fischione, C. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3038848 · Cited by 59
“d-to-end learning of communications systems is an interesting novel concept for learning full implementations of transmitter and receiver by considering a specific performance metric and channel model [174]. The basic idea is to treat the entire communication system as an autoencoder and to model transmitter and receiv…”
MethodDeep learning aided OFDM receiver for underwater acoustic communications
Zhang, Y., Li, C., Wang, H., Wang, J., Yang, F., & Mériaudeau, F. (2021) · Applied Acoustics · doi:10.1016/j.apacoust.2021.108515 · Cited by 56
Sharma, A., & Lashkari, A. H. (2024) · Computer Networks · doi:10.1016/j.comnet.2024.110984 · Cited by 75
A Comprehensive Survey on Machine and Deep Learning for Optical Communications
Amirabadi, M. A., Nezamalhosseini, S. A., Kahaei, M. H., & Chen, L. R. (2025) · IEEE Access · doi:10.1109/access.2025.3569559 · Cited by 67
Shi, J., Li, Z., Jia, J., Li, Z., Shen, C., Zhang, J., & Chi, N. (2023) · Journal of Lightwave Technology · doi:10.1109/jlt.2023.3236400 · Cited by 53
“End-to-end learning-based techniques, such as autoencoders (AEs), have been demonstrated to result in performance gains in a number of areas, including simulation and experimentation cases [23], [24], [25], [26], [27].”
BackgroundDeep Learning for CSI Feedback Based on Superimposed Coding
Qing, C., Cai, B., Yang, Q., Wang, J., & Huang, C. Z. (2019) · IEEE Access · doi:10.1109/access.2019.2928049 · Cited by 55
“Recently, the deep learning (DL) based physical-layer technique shows its promising prospects in wireless communication system [3]–[9], [17]–[21] and the comprehensive overview could be found in [18]–[20].”
BackgroundData-Driven Symbol Detection Via Model-Based Machine Learning
Farsad, N., Shlezinger, N., Goldsmith, A. J., & Eldar, Y. C. (2020) · IEEE Statistical Signal Processing Workshop (SSP) · doi:10.1109/ssp49050.2021.9513859 · Cited by 61
“Recent years have witnessed growing interest in the application of DNNs for receiver design; see detailed surveys in [7], [8], [9], [10], [11].”
BackgroundSpatial-Temporal Hybrid Feature Extraction Network for Few-Shot Automatic Modulation Classification
Che, J., Wang, L., Bai, X., Liu, C., & Zhou, F. (2022) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2022.3196103 · Cited by 54
“3196103 proposed in [10], it was trained on raw IQ samples for modulation classification.”
DeepWiPHY: Deep Learning-Based Receiver Design and Dataset for IEEE 802.11ax Systems
Zhang, Y., Doshi, A. S., Liston, R., Tan, W.-T., Zhu, X., Andrews, J. G., & Heath, R. W. (2020) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2020.3034610 · Cited by 61
“In [2], they formalize this intuition in”
BackgroundInfluentialDeep Learning for the Gaussian Wiretap Channel
Fritschek, R., Schaefer, R. F., & Wunder, G. (2019) · International Conference on Communications · doi:10.1109/icc.2019.8761681 · Cited by 62
“they can be easily plotted in 2-d. Our training procedure is divided into three phases. In the first phase, we only train the encoder and the decoder of Bob with the standard cross-entropy, as done in [3]. Here, the NN learns a signal constellation to combat the noise in an efficient way dependent on the power constrain…”
BackgroundMethodInfluentialWireless Fingerprinting via Deep Learning: The Impact of Confounding Factors
Cekic, M., Gopalakrishnan, S., & Madhow, U. (2021) · Asilomar Conference on Signals, Systems and Computers (ACSSC) · doi:10.1109/ieeeconf53345.2021.9723393 · Cited by 57
“Modern CNNs learning directly from I/Q data include [37, 38] for modulation classification, and [39, 40] for device fingerprinting.”
MethodSharma, H., & Kumar, N. (2023) · Physical Communication · doi:10.1016/j.phycom.2023.102002 · Cited by 67
Deep Learning-Aided Off-Grid Channel Estimation for Millimeter Wave Cellular Systems
Wan, L., Liu, K., & Zhang, W. (2021) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2021.3120926 · Cited by 54
“Recently, machine learning, specifically deep learning (DL), has emerged as an effective tool for wireless communication [32]–[34].”
BackgroundMulti-User Semantic Communications for Cooperative Object Identification
Zhang, Y., Xu, W., Gao, H., & Wang, F. (2022) · IEEE International Conference on Communications workshops · doi:10.1109/iccworkshops53468.2022.9814491 · Cited by 55
Agadakos, I., Agadakos, N., Polakis, J., & Amer, M. R. (2020) · IEEE European Symposium on Security and Privacy · doi:10.1109/eurosp48549.2020.00028 · Cited by 60
“O’Shea and Hoydis [70] identified some crucial problems such as the non-holomorphic nature of common activation functions, which was addressed in [16].”
BackgroundInfluentialSemantic Communications for Speech Recognition
Weng, Z., Qin, Z., & Li, G. Y. (2021) · Globecom · doi:10.1109/globecom46510.2021.9685250 · Cited by 60
“Inspired by the end-to-end (E2E) communication systems developed to address the challenges in traditional block-wise commutation systems [9], [10], different types of sources have been considered on E2E semantic communication systems.”
MethodEnd-to-End Learning of Geometrical Shaping Maximizing Generalized Mutual Information
Gümüş, K., Alvarado, A., Chen, B., Häger, C., & Agrell, E. (2020) · Optical Fiber Communication Conference · doi:10.1364/ofc.2020.w3d.4 · Cited by 52
“This also connects the end-to-end AE learning approach in [6] to early works on”
BackgroundA vision of 6G – 5G's successor
Lu, Y., & Ning, X. (2020) · Journal of Management Analytics · doi:10.1080/23270012.2020.1802622 · Cited by 77
Deep Learning-Based Robust Precoding for Massive MIMO
Shi, J., Wang, W., Yi, X., Gao, X., & Li, G. Y. (2021) · IEEE Transactions on Communications · doi:10.1109/tcomm.2021.3105569 · Cited by 65
Wang, X., Hua, H., & Xu, Y. (2020) · IEEE Access · doi:10.1109/access.2020.2978253 · Cited by 61
“In addition, DL technology has gradually gained more attention in the field of wireless communications [18], such as channel estimation [19], [20], modulation identification [21], [22] and channel state information (CSI) feedback [23], [24].”
BackgroundBinary signaling design for visible light communication: a deep learning framework
Lee, H., Lee, I., Quek, T. Q. S., & Lee, S. H. (2018) · Optics Express · doi:10.1364/oe.26.018131 · Cited by 51
Deep Learning Aided Method for Automatic Modulation Recognition
Yang, C., He, Z., Peng, Y., Wang, Y., & Yang, J. (2019) · IEEE Access · doi:10.1109/access.2019.2933448 · Cited by 57
“from a modulated signal without designing manmade features [21].”
BackgroundRF Fingerprinting and Deep Learning Assisted UE Positioning in 5G
Butt, M. M., Rao, A., & Yoon, D. (2020) · IEEE Vehicular Technology Conference · doi:10.1109/vtc2020-spring48590.2020.9128640 · Cited by 51
Geometric Constellation Shaping for Fiber-Optic Channels via End-to-End Learning
Jovanovic, O., Ros, F. D., Zibar, D., & Yankov, M. P. (2023) · Journal of Lightwave Technology · doi:10.1109/jlt.2023.3276300 · Cited by 59
“End-to-end learning, which was introduced in [8], utilizes an autoencoder (AE) [9] to optimize the communication system and it has gathered traction as a method to approach GCS due to its versatility to the employed channel model.”
MethodModulation-Constrained Clustering Approach to Blind Modulation Classification for MIMO Systems
Tian, J., Pei, Y., Huang, Y., & Liang, Y. (2018) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2018.2879370 · Cited by 50
“In [31], convolution neural network (CNN) with raw IQ samples is used to classify different modulation schemes.”
MethodOn the Performance of Neural Network Residual Kriging in Radio Environment Mapping
Sato, K., Inage, K., & Fujii, T. (2019) · IEEE Access · doi:10.1109/access.2019.2928832 · Cited by 50
Binary MIMO Detection via Homotopy Optimization and Its Deep Adaptation
Shao, M., & Ma, W. (2020) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2020.3048232 · Cited by 58
“In particular, we have seen a variety of emerging deep network applications for communications, such as end-to-end communication system design [19], multiuser power control [20], and, as our main interest, MIMO detection [21–27].”
BackgroundModel-Based Machine Learning for Joint Digital Backpropagation and PMD Compensation
Butler, R. M., Häger, C., Pfister, H. D., Liga, G., & Alvarado, A. (2020) · Journal of Lightwave Technology · doi:10.1109/jlt.2020.3034047 · Cited by 50
“The combination of linearities and nonlinearities in ANNs allows for the learning of complex functions.”
BackgroundInfluentialDeepBeam: Deep Waveform Learning for Coordination-Free Beam Management in mmWave Networks
Polese, M., Restuccia, F., & Melodia, T. (2021) · ACM International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing (MobiHoc) · doi:10.1145/3466772.3467035 · Cited by 63
“Specifically, deep learning is being used to address challenging problems such as modulation recognition, radio fingerprinting, and many others [29, 30, 55, 56].”
MethodGao, F., Lin, B., Bian, C., Zhou, T., Qian, J., & Wang, H. (2021) · IEEE Transactions on Communications · doi:10.1109/tcomm.2021.3110301 · Cited by 58
“Recently, deep learning (DL) [17] has been applied to a large variety of problems in wireless communications for modulation recognition [18], channel estimation [19], [20], signal detection [21], [22], channel equalization [23], CSI feedback [24], [25], and end-to-end transceiver [26]– [28].”
BackgroundSignal Processing Techniques for 6G
Mucchi, L., Shahabuddin, S., Albreem, M. A. M., Abdallah, S., Caputo, S., Panayırcı, E., & Juntti, M. J. (2023) · Journal of Signal Processing Systems · doi:10.1007/s11265-022-01827-7 · Cited by 60
“In addition, there is a lack of datasets to benchmark and compare the performance of ML models and algorithms [166].”
BackgroundHardware Implementation of Neural Self-Interference Cancellation
Kurzo, Y., Kristensen, A. T., Burg, A. P., & Balatsoukas‐Stimming, A. (2020) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2020.2992370 · Cited by 57
“use of neural networks (NNs) to augment or replace a range of signal processing tasks in communications systems [13]–[18].”
BackgroundGNN-Based Beamforming for Sum-Rate Maximization in MU-MISO Networks
Li, Y., Lu, Y., Ai, B., Dobre, O. A., Ding, Z., & Niyato, D. T. (2024) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2024.3361174 · Cited by 69
“Recently, some researchers have developed the deep learning (DL)-based signal processing approaches [6], where the traditional optimization algorithms were approximated by neural networks (NNs).”
BackgroundA Neural Receiver for 5G NR Multi-User MIMO
Cammerer, S., Aoudia, F. A., Hoydis, J., Oeldemann, A., Roessler, A., Mayer, T., & Keller, A. (2023) · IEEE Globecom Workshops (GC Wkshps) · doi:10.1109/gcwkshps58843.2023.10464486 · Cited by 97
“It was shown later in [4], that a similar architecture also permits pilotless communications when combined with end-to-end learning [2].”
BackgroundSelf-citationAI-Enabled Intelligent Visible Light Communications: Challenges, Progress, and Future
Shi, J., Niu, W., Ha, Y., Xu, Z., Li, Z., Yu, S., & Chi, N. (2022) · Photonics · doi:10.3390/photonics9080529 · Cited by 53
End-to-End Fast Training of Communication Links Without a Channel Model via Online Meta-Learning
Park, S., Simeone, O., & Kang, J. (2020) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc48557.2020.9154322 · Cited by 49
“The neural encoder and neural decoder, parameterized by φT and φR, respectively, have the same structure introduced in [13] with decoder being equipped with a Radio Transformer Networks (RTN).”
MethodPhysical Layer Communication via Deep Learning
Kim, H., Oh, S., & Viswanath, P. (2020) · IEEE Journal on Selected Areas in Information Theory · doi:10.1109/jsait.2020.2991562 · Cited by 69
“As opposed to updating V and W simultaneously, as in [42], we alternate the training of V and W to prevent converging to a bad local optimum [43], [44].”
BackgroundEnd-to-End Deep Learning for Long-haul Fiber Transmission Using Differentiable Surrogate Channel
Niu, Z., Yang, H., Zhao, H., Dai, C., Hu, W., & Yi, L. (2022) · Journal of Lightwave Technology · doi:10.1109/jlt.2022.3148270 · Cited by 54
Building Occupancy Detection and Localization Using CCTV Camera and Deep Learning
Hu, S., Wang, P., Hoare, C., & O’Donnell, J. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3201877 · Cited by 51
“Deep learning has engendered a shift for the communication and networking field, such as jointly optimizing transmitter and receiver components for the physical layer [11], and accurately classifying mobile encrypted traffic to obtain highly-valuable profiling information [12].”
BackgroundMaglogiannis, V., Shahid, A., Naudts, D., Poorter, E. D., & Moerman, I. (2019) · IEEE Access · doi:10.1109/access.2019.2902035 · Cited by 49
“The article in [32] discusses several applications of deep learning for the physical layer.”
BackgroundHo, T. M., Tran, T. D., Nguyen, T. T., Kazmi, S. M. A., Le, L. B., Hong, C. S., & Hanzo, L. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1907.10102 · Cited by 63
“PHY layer components, such as channel coding and signal detection, result in low adaptability to network dynamics. The aforementioned challenging problem can be overcome by applying ML techniques. In [189] and [190], O’Shea et al. propose to interpret a communication system as an autoencoder carrying out a high-accura…”
MethodOnline Energy-Efficient Power Control in Wireless Networks by Deep Neural Networks
Zappone, A., Debbah, M., & Altman, Z. (2018) · 2018 IEEE 19th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc.2018.8445857 · Cited by 48
“Overviews on deep learning applications to wireless communications have recently appeared in [12, 13].”
BackgroundPu, G., Liu, R., Yang, H., Xu, Y., Hu, W., Hu, M., & Yi, L. (2023) · Laser & Photonics Review · doi:10.1002/lpor.202200363 · Cited by 52
Deep Learning-Based Channel Modeling for Free Space Optical Communications
Chen, W., Zhang, M., Wang, D., Zhan, Y., Cai, S., Yang, H. X., Zhang, Z., Chen, X., & Wang, D. (2022) · Journal of Lightwave Technology · doi:10.1109/jlt.2022.3213519 · Cited by 53
“In addition, autoencoder-based end-to-end (E2E) learning as a powerful and prospective global optimization solution has been successfully applied in both wireless and optical communications [4], [5], where differentiable channels are required [6].”
BackgroundEnhancing Least Square Channel Estimation Using Deep Learning
Gizzini, A. K., Chafii, M., Nimr, A., & Fettweis, G. P. (2020) · IEEE Vehicular Technology Conference · doi:10.1109/vtc2020-spring48590.2020.9128890 · Cited by 58
“Recently, deep learning (DL) has drawn attentions for its great success in computer vision, automatic speech recognition, and natural language processing [5].”
BackgroundDNCNet: Deep Radar Signal Denoising and Recognition
Du, M., Zhong, P., Cai, X., & Bi, D. (2022) · IEEE Transactions on Aerospace and Electronic Systems · doi:10.1109/taes.2022.3153756 · Cited by 53
“In addition to utilizing the preprocessing time-frequency information in the aforementioned literature, the authors in e.g. [9], [10], [11] identified the radio signal directly from the raw time-series sequence.”
Srinivasan, M., Song, J., Grabowski, A., Szczerba, K., Iversen, H. K., Schmidt, M. N., Zibar, D., Schröder, J., Larsson, A. G., Häger, C., et al. (2023) · Journal of Lightwave Technology · doi:10.1109/jlt.2023.3251660 · Cited by 59
“AE-based deep NN architecture was proposed initially in [32] in the context of wireless links.”
Amirabadi, M. A., Kahaei, M. H., & Nezamalhosseini, S. A. (2021) · IET Optoelectronics · doi:10.1049/ote2.12060 · Cited by 48
Machine Learning in 6G Wireless Communications
Ohtsuki, T. (2023) · IEICE transactions on communications · doi:10.1587/transcom.2022cei0002 · Cited by 41
“We first introduce end-to-end learning of communication systems through neural networked-based autoencoders [11].”
Metadata
- Authors
- O'Shea, T., & Hoydis, J.
- Affiliations
- Nokia (France) (FR)
- Venue
- IEEE Transactions on Cognitive Communications and Networking
- Published
- 2017-12
- Volume
- 3
- Issue
- 4
- Pages
- 563-575
- ISSN
- 2332-7731 (ISSN-L)
- OpenAlex
- W2734408173
- Semantic Scholar
- 78989616eeeac55b202e3e4205225e7135054185
- Cited by
- 2,980
- References
- 69
- Citations by year
- 2026: 307 · 2025: 327 · 2024: 305 · 2023: 368 · 2022: 432 · 2021: 409 · 2020: 406 · 2019: 322 · 2018: 95 · 2017: 9
- FWCI
- 122.29
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE Transactions on Cognitive Communications and Networking (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Wireless Signal Modulation Classification
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Wireless Communication Techniques
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Wireless Communication Technologies
- Keywords
- end-to-end reconstruction, domain knowledge integration, convolutional neural networks, automatic modulation classification, IQ samples
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-11-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.