InterGridNet: An Electric Network Frequency Approach for Audio Source Location Classification Using Convolutional Neural Networks

Christos Korgialas; Constantine Kotropoulos; Georgios Tzolopoulos; Ioannis Tsingalis

REVIEW 4 major objections 5 minor 35 references

InterGridNet: An Electric Network Frequency Approach for Audio Source Location Classification Using Convolutional Neural Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A single compact network, InterGridNet, classifies which of nine power grids an ENF recording came from with 92% accuracy on the SP Cup 2016 test set.

desk verdict A clean raw-waveform ENF classifier with 92% on SP Cup 2016, but the 96% fusion model it compares against is left out of the table and the automated routing step is never evaluated on its own. read the letter →

arxiv 2502.10011 v1 pith:IS3III3I submitted 2025-02-14 cs.SD cs.LGeess.AS

Christos Korgialas , Ioannis Tsingalis , Georgios Tzolopoulos , Constantine Kotropoulos This is my paper · ORCID

classification cs.SDcs.LGeess.AS

keywords electricnetworkfrequencygridlocationestimationaudioprocessingmultimediaforensicsconvolutionalneuralnetworksarchitecturesearchRawNetsourceclassification

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper's claim is that the power-grid hum embedded in an audio or power recording can identify which national grid the recording came from, using one compact neural network instead of a fusion of many classifiers. The authors build InterGridNet, a shallow end-to-end convolutional network that reads raw filtered waveforms and first sorts recordings into audio versus power and 50 Hz versus 60 Hz groups. On the Signal Processing Cup 2016 benchmark, the system classifies test recordings from nine grids plus a 'None' category with 92% accuracy. If correct, this means a single seven-million-parameter model can beat earlier feature-based systems and come close to a much larger five-classifier fusion, which is useful for multimedia forensics and verifying where a recording was made.

What carries the argument

The machinery is the shallow RawNet classifier combined with a two-stage data-routing pipeline. RawNet is an end-to-end convolutional architecture that takes raw waveform frames as input; here it is reduced to two residual blocks, a gated recurrent unit that pools frame-level embeddings into one ENF-level representation, and a softmax output layer, with topology and Adam hyperparameters chosen by neural architecture search. The companion mechanism is the automated group-routing rule: for each recording, the average spectrogram magnitude at the first three harmonics of 50 Hz and 60 Hz is compared, the weakest harmonic in each set is ignored, and the larger retained average selects the nominal frequency group.

What would settle it

Add a strong 50 Hz interference tone to 60 Hz grid recordings while leaving the ENF at 60 Hz; if the harmonic-grouping step misroutes these samples to a 50 Hz model, the reported 92% test accuracy should drop, showing that routing decisions are the weak link.

Watch

Extended reading notes

Core claim

The central claim is that a deliberately shallow end-to-end network, named InterGridNet, can geolocate ENF-bearing recordings by classifying the grid of origin at 92% overall accuracy on the SP Cup 2016 test set of 100 ten-minute samples spanning nine grids plus None. The system routes each recording into one of four groups—audio/50 Hz, audio/60 Hz, power/50 Hz, power/60 Hz—using a spectrogram-harmonic comparison, applies a 2 Hz bandpass filter around the nominal frequency, and splits the signal into 16-second overlapping frames. A NAS-optimized shallow RawNet with residual blocks and a GRU produces a per-frame softmax distribution; an entropy-based threshold rejects low-confidence frames as 'None', and majority voting over frames gives the recording-level decision. The paper reports that the same architecture without bandpass filtering drops to 72% accuracy, indicating that the ENF band itself carries the discriminating information.

Load-bearing premise

The whole pipeline depends on the automated grouping step correctly telling 50 Hz from 60 Hz and audio from power; if a test recording is sent to the wrong group, the classifier trained on a different distribution must handle it and the framework has no way to recover.

Editorial extensions

If this is right

A single 7-million-parameter classifier reaches 92% test accuracy on the SP Cup 2016 grid-location task, exceeding the 86–88% range of earlier SVM and histogram baselines.
The 20-point gap between filtered (92%) and unfiltered (72%) operation shows that isolating the 2 Hz band around the nominal ENF is the main contributor to classification success.
Per-frame entropy rejection plus majority voting lets the system label out-of-set recordings as 'None' with 100% practice-set accuracy, so the framework can abstain rather than force a wrong grid.
The routing into 50 Hz and 60 Hz groups prevents impossible confusions between grids of different nominal frequency, which is why observed errors are mostly 'None' rather than cross-frequency misclassifications.
InterGridNet's 92% is close to the authors' own 96% five-classifier fusion but with roughly 7 million versus 11 million parameters, suggesting raw-waveform end-to-end learning is a parameter-efficient route to the same task.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The 20-point accuracy gap between filtered and unfiltered operation is so large that the ENF band itself is likely the entire useful signal; a simpler front end that estimates instantaneous frequency directly might match the network with far fewer parameters.
The automated grouping step can be evaluated in isolation: because any recording misrouted to the wrong 50/60 Hz model is guaranteed to be out-of-distribution, measuring routing accuracy separately would reveal how much of the 8% test error originates before classification.
The rejection rule suggests an open-set deployment pattern for media forensics: recordings whose per-frame softmax entropy never clears the threshold are flagged as unknown rather than forced into the nearest grid, which is useful when a recording may come from an unmodeled grid.
The practice set contains only five samples per class, so the 90% validation accuracy and the tuned thresholds 0.8 and 0.75 are estimated from very little data; validating on a larger held-out set would strengthen the claimed generalization.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a reimplementable empirical paper that adapts RawNet to raw ENF waveforms for grid classification and reports 92% on the SP Cup 2016 test set. The work is new in a modest way, the write-up is clear, and the code is referenced. But the evaluation has soft spots you should know about before taking the headline number at face value.

What it does well: it is the first raw-waveform DNN for ENF grid classification, uses NAS to tune a shallow RawNet, and the preprocessing is described in enough detail to reproduce. The bandpass filtering ablation is nice evidence that the model is actually using ENF content (72% without filtering vs. 92% with it). The entropy threshold and majority vote for the "None" class are clearly explained.

The soft spots are real but mostly fixable. First, Table III omits the authors' own prior fusion model [11], which scored 96% on the same test set. They discuss it in the text and argue that their 7M-parameter single classifier is more efficient than an 11M-parameter CNN plus fusion, which is a fair trade-off. But the state-of-the-art claim in the table is incomplete without that row. Second, Table II has an internal inconsistency: grid G shows Audio=80%, Power=80%, but All=60%. That cannot be right unless the sample sizes are bizarre, and it makes me distrust the validation table. Third, the test set is only 100 recordings. A 92/100 result has a binomial 95% CI of roughly 84%–97%, so the difference from 96% is not obviously significant. The authors should report error bars or at least acknowledge the small sample. Fourth, and most important, the automated grouping step that routes each test recording to a 50 Hz or 60 Hz model is never evaluated. The paper describes the harmonic-magnitude rule for 50/60 Hz but does not report how often that routing is correct. If a recording gets misrouted, its true class is not in the softmax of the group model, so it cannot be classified correctly. The end-to-end 92% therefore includes routing errors, but we have no idea how many. A simple fix: report routing accuracy and give results with and without oracle routing. This would make the pipeline transparent.

Who is this for? People working on ENF forensics and raw-waveform classifiers for weak periodic signals. The central idea is reasonable and the methodology is not broken; the missing diagnostics are the main weakness. I would send it to peer review because it is a legitimate empirical contribution that can be made rigorous with revision, but I would not cite it in the next year unless I were working directly on ENF grid localization.

Referee Report

4 major / 5 minor

Summary. The paper introduces InterGridNet, a shallow RawNet-based classifier for electric network frequency (ENF) geolocation, evaluated on the SP Cup 2016 benchmark. Recordings are grouped into audio/power and 50/60 Hz categories, bandpass filtered around the nominal ENF, and framed into 16-second segments. Four group-specific models are trained with a neural architecture search; an entropy threshold and a majority vote identify the 'None' class. The authors report a validation accuracy of 90% and a testing accuracy of 92% on the 100-sample SP Cup 2016 test set, and they compare against several prior methods.

Significance. If the reported performance is robust, InterGridNet is a competitive single-classifier raw-waveform alternative to the authors' earlier 96% fusion model, with a smaller parameter count. The paper uses the only public benchmark in the field, provides a GitHub repository, and includes a useful ablation of bandpass filtering. The main limitations are the lack of routing diagnostics for the hierarchical grouping, the small test set, and an internal inconsistency in the validation table; these need to be addressed before the central accuracy claim can be fully assessed.

major comments (4)

[Section III-A and Section IV-A] The end-to-end 92% test accuracy depends on correct automated routing of each test recording into the four groups (audio50, audio60, power50, power60), but the paper reports no routing accuracy. The automated grouping described in Section III-A only distinguishes 50 Hz versus 60 Hz by comparing averaged harmonic magnitudes; no procedure is given for the audio-vs-power branch. If the test set's audio/power labels are used, this should be stated explicitly; if they are not, the pipeline is underspecified. Because the group models have output vocabularies restricted to their group's grids (n=3 for 60 Hz groups, n=6 for 50 Hz groups), any non-None recording routed to the wrong 50/60 group cannot be classified correctly. Thus the reported 92% is an upper bound on routing accuracy for the 90 non-None test items, and routing errors are not separated from classification errors.
[Section IV-A, Table II] Table II contains an internal inconsistency: for grid G, the Audio row reports 80%, the Power row reports 100%, but the All row reports 60%. If the All row is any aggregate of the two types, the combined accuracy for G cannot be below 80% (and with equal class counts would be 90%). This error undermines confidence in the validation results, which are used to tune alpha1 and alpha2 and to argue that testing accuracy aligns with validation accuracy. The table, and the underlying validation counts, should be corrected and reported.
[Section IV-B and Table III] The paper claims 'superiority' over previous works, yet Table III omits the authors' own fusion model [11], which reached 96% on the same test set, as acknowledged in Section IV-B. Because 92% is below that figure and the difference may be within statistical noise on 100 test samples, the state-of-the-art claim is not supported. Please either include [11] in Table III, or qualify the contribution as a parameter-efficient single-classifier alternative rather than an accuracy improvement. In addition, a binomial 95% confidence interval for 92/100 (approximately 85% to 97%) would place the comparison with 96% in context.
[Section IV-A (test sample size)] The test set contains only 100 recordings, and the overall accuracy is a single point estimate with no confidence intervals or significance testing. The differences between 92%, the 88% best prior method in Table III, and the 96% fusion model in [11] may not be statistically meaningful. Reporting per-class test counts, a confusion matrix with numerical entries, and an interval or exact test would strengthen the central empirical claim.

minor comments (5)

[Section III-B] The architecture description states that the softmax maps to a probability distribution over the 9 classes, but each group model has only n=3 or n=6 output classes. Please clarify that the output layer size equals the group's class count.
[Section III-A] The harmonic-based routing rule should specify the spectrogram computation parameters (window length, FFT size, hop) and the frequency range around each harmonic that is integrated, to make the procedure reproducible.
[Section V] The conclusion refers to 'InterGrid' instead of 'InterGridNet'; please correct the name.
[Section IV-A] The practice set is described as containing 50 samples (5 per class); if this means 5 per class per recording type, the total would be 100, not 50, and the validation accuracy numbers in Table II should be consistent with that count.
[Figure 5] The confusion matrices are not described numerically; please include the exact counts for each cell or a corresponding table, especially for the 'None' class predictions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 92% test accuracy is an empirical measurement on the external SP Cup 2016 benchmark, and no fitted parameter or self-citation is disguised as a prediction.

full rationale

The paper's central claim is an end-to-end accuracy measurement on an unseen test set from the SP Cup 2016 competition, not a quantity derived from the framework's own assumptions. Training uses the competition training split, validation thresholds α1 and α2 are tuned on the separate practice set, and test performance is reported on the augmented 100-sample test set. The NAS hyperparameter search optimizes model topology and optimizer settings on validation data, which is standard model selection rather than a prediction forced by construction. The 50 Hz / 60 Hz automated grouping rule in Section III-A is a preprocessing heuristic based on spectrogram magnitudes; it is an assumption about the test data's spectral characteristics, and the absence of reported routing accuracy is an evaluation-limitation issue, not a circularity. The comparison with the authors' earlier fusion model [11] is a benchmark comparison, and that citation is not used to justify the correctness of InterGridNet's own 92% result; the result is independently measured on the same public test data. No equation in the paper defines its output in terms of its input, no fitted parameter is renamed as a prediction, and no load-bearing claim reduces to a self-citation. Accordingly, the appropriate circularity score is 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central accuracy claim depends on two validation-tuned thresholds and a set of NAS-selected hyperparameters; these are fitted values rather than externally fixed constants. The method also assumes the ENF signal is a reliable grid fingerprint and that the practice set represents test conditions.

free parameters (3)

Entropy threshold alpha1 = 0.8
Tuned on the practice set to decide whether a frame's softmax distribution is confident enough to be assigned to a known grid rather than None; directly affects the final classification.
Majority-vote threshold alpha2 = 0.75
Tuned on the practice set; sets the minimum fraction of frames that must agree on a grid label for the recording-level decision, otherwise the sample is labeled N.
NAS-selected hyperparameters (learning rate, beta1, beta2) = See Table I, e.g., lr 6.5e-4 to 9.7e-4; beta1 0.96-0.98; beta2 0.992-0.998
Found via Keras-Tuner neural architecture search on validation data; these settings determine the trained model that produced the reported accuracy.

assumptions (3)

domain assumption ENF signals contain grid-specific, learnable patterns in raw waveforms
Central premise of the method; stated in Introduction and Section III-B.
domain assumption The practice set is representative of the test set for tuning alpha1 and alpha2
Thresholds are chosen on the 50-sample practice set and used unchanged on the test set (Section IV-A).
domain assumption The automated harmonic-based grouping correctly separates 50/60 Hz and audio/power recordings in testing
Described as a four-step spectrogram procedure in Section III-A, no error analysis is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of InterGridNet: An Electric Network Frequency Approach for Audio Source Location Classification Using Convolutional Neural Networks." pith.science (2026). https://pith.science/paper/IS3III3I

@misc{pith2026250210011,
  author       = {Pith},
  title        = {Pith review of: InterGridNet: An Electric Network Frequency Approach for Audio Source Location Classification Using Convolutional Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IS3III3I}},
  note         = {Machine review of arXiv:2502.10011}
}

read the original abstract

A novel framework, called InterGridNet, is introduced, leveraging a shallow RawNet model for geolocation classification of Electric Network Frequency (ENF) signatures in the SP Cup 2016 dataset. During data preparation, recordings are sorted into audio and power groups based on inherent characteristics, further divided into 50 Hz and 60 Hz groups via spectrogram analysis. Residual blocks within the classification model extract frame-level embeddings, aiding decision-making through softmax activation. The topology and the hyperparameters of the shallow RawNet are optimized using a Neural Architecture Search. The overall accuracy of InterGridNet in the test recordings is 92%, indicating its effectiveness against the state-of-the-art methods tested in the SP Cup 2016. These findings underscore InterGridNet's effectiveness in accurately classifying audio recordings from diverse power grids, advancing state-of-the-art geolocation estimation methods.

Figures

Figures reproduced from arXiv: 2502.10011 by the authors.

**Figure 1.** Flowchart of the proposed InterGridNet framework. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Number of audio and power recording frames in each grid. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Figures from the paper (3 more)

**Figure 3.** Figure 3: Spectrograms focused on the nominal ENF value for different grids. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]

**Figure 4.** Figure 4: Architecture of the proposed optimized shallow RawNet model. The operators utilized in the network include Conv1D(kernel size, strides, filters), [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Confusion matrices predictions on the testing set employing the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 33 canonical work pages

[11]

On spectrogram analysis in a multiple classifier fusion framework for power grid clas- sification using electric network frequency,

G. Tzolopoulos, C. Korgialas, and C. Kotropoulos, “On spectrogram analysis in a multiple classifier fusion framework for power grid clas- sification using electric network frequency,” in Proc. of the 13th Inter- national Conference on Pattern Recognition Applications and Methods . SCITEPRESS, 2024, pp. 91–99

work page 2024
[1]

Analysis of rolling shutter effect on ENF-based video forensics,

S. Vatansever, A. E. Dirik, and N. Memon, “Analysis of rolling shutter effect on ENF-based video forensics,”IEEE Transactions on Information Forensics and Security, vol. 14, no. 9, pp. 2262–2275, 2019

work page 2019
[2]

Leveraging electric network frequency estimation for audio authentication,

C. Korgialas, C. Kotropoulos, and K. N. Plataniotis, “Leveraging electric network frequency estimation for audio authentication,” IEEE Access , 2024

work page 2024
[3]

A robust RELAX-based algorithm for enhanced electric network frequency estimation,

C. Korgialas and C. Kotropoulos, “A robust RELAX-based algorithm for enhanced electric network frequency estimation,” in Proc. of the 13th Hellenic Conference on Artificial Intelligence (SETN). ACM, 2024, pp. 1–6

work page 2024
[4]

Electric network frequency optical sensing devices,

C. Moysiadis, G. Karantaidis, and C. Kotropoulos, “Electric network frequency optical sensing devices,” in Proc. of the 2023 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applica- tions. IARIA, 2023, pp. 1–6

work page 2023
[5]

Modeling and analysis of electric network frequency signal for timestamp verification,

R. Garg, A. L. Varna, and M. Wu, “Modeling and analysis of electric network frequency signal for timestamp verification,” in Proc. of the IEEE International Workshop on Information Forensics and Security (WIFS). IEEE, 2012, pp. 67–72

work page 2012
[6]

Error analysis of forensic ENF matching,

G. Hua, “Error analysis of forensic ENF matching,” in Proc. of the IEEE International Workshop on Information Forensics and Security (WIFS , 2018, pp. 1–7

work page 2018
[7]

Time-of-recording estimation for audio recordings,

L. Zheng, Y . Zhang, C. E. Lee, and V . L. L. Thing, “Time-of-recording estimation for audio recordings,” Digital Investigation , vol. 22, pp. S115–S126, 2017

work page 2017

Show all 35 references

[8]

A dynamic matching algorithm for audio timestamp identification using the ENF criterion,

G. Hua, J. Goh, and V . L. Thing, “A dynamic matching algorithm for audio timestamp identification using the ENF criterion,” IEEE Transactions on Information Forensics and Security , vol. 9, no. 7, pp. 1045–1055, 2014

2014
[9]

ENF based location classification of sensor recordings,

A. Hajj-Ahmad, R. Garg, and M. Wu, “ENF based location classification of sensor recordings,” in Proc. of the IEEE International Workshop on Information Forensics and Security (WIFS) , 2013, pp. 138–143

2013
[10]

ENF-based region-of-recording identification for media signals,

——, “ENF-based region-of-recording identification for media signals,” IEEE Transactions on Information Forensics and Security, vol. 10, no. 6, pp. 1125–1136, 2015

2015
[12]

Source location identification of distribution-level electric network frequency signals at multiple geographic scales,

W. Yao et al., “Source location identification of distribution-level electric network frequency signals at multiple geographic scales,” IEEE Access, vol. 5, pp. 11 166–11 175, 2017

2017
[13]

Feasibility study on intra-grid location estimation using power ENF signals,

R. Garg, A. Hajj-Ahmad, and M. Wu, “Feasibility study on intra-grid location estimation using power ENF signals,” 2021, arXiv:2105.00668

2021 arXiv
[14]

Geo-location estimation from electrical network frequency sig- nals,

——, “Geo-location estimation from electrical network frequency sig- nals,” in Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2013, pp. 2862–2866

2013
[15]

New forensic ENF reference database for media recording authentication based on harmony search technique using GIS and wide area frequency measurements,

M. M. Elmesalawy and M. M. Eissa, “New forensic ENF reference database for media recording authentication based on harmony search technique using GIS and wide area frequency measurements,” IEEE Transactions on Information Forensics and Security , vol. 9, no. 4, pp. 633–644, 2014

2014
[16]

M. H. J. Bollen and I. Y . H. Gu, Signal Processing of Power Quality Disturbances. John Wiley & Sons, 2006

2006
[17]

Frequency sensitivity and electromechanical propaga- tion simulation study in large power systems,

S.-J. Tsai et al., “Frequency sensitivity and electromechanical propaga- tion simulation study in large power systems,” IEEE Transactions on Circuits and Systems , vol. 54, no. 8, pp. 1819–1828, 2007

2007
[18]

Instantaneous frequency estima- tion and localization for ENF signals,

A. Hajj-Ahmad, R. Garg, and M. Wu, “Instantaneous frequency estima- tion and localization for ENF signals,” in Proc. of the Asia Pacific Signal and Information Processing Association Annual Summit and Conference, 2012, pp. 1–10

2012
[19]

I’m listening to your location! Inferring user location with acoustic side channels,

Y . Jeon, M. Kim, H. Kim, H. Kim, J. H. Huh, and J. Yoon, “I’m listening to your location! Inferring user location with acoustic side channels,” in Proc. of the World Wide Web Conference , 2018, pp. 339–348

2018
[20]

Exploiting spatial signatures of power ENF signal for measurement source authentication,

Y . Cui, Y . Liu, P. Fuhr, and M. Morales-Rodriguez, “Exploiting spatial signatures of power ENF signal for measurement source authentication,” in Proc. of the IEEE International Symposium on Technologies for Homeland Security, 2018, pp. 1–6

2018
[21]

RawNet: Advanced end-to-end deep neural network using raw waveforms for text-independent speaker verification,

J.-W. Jung, H.-S. Heo, J.-H. Kim, H.-J. Shim, and H.-J. Yu, “RawNet: Advanced end-to-end deep neural network using raw waveforms for text-independent speaker verification,” arXiv preprint arXiv:1904.08104, 2019

1904 arXiv
[22]

Location signatures that you don’t see: Highlights from the IEEE Signal Processing Cup student competition,

M. Wu, A. Hajj-Ahmad, M. Kirchner, Y . Ren, C. Zhang, and P. Campisi, “Location signatures that you don’t see: Highlights from the IEEE Signal Processing Cup student competition,”IEEE Signal Processing Magazine, vol. 33, no. 5, pp. 149–156, 2016

2016
[23]

Digital audio recording analysis–the electric network frequency criterion,

C. Grigoras, “Digital audio recording analysis–the electric network frequency criterion,” International Journal of Speech Language and the Law, vol. 12, no. 1, pp. 63–76, 2005

2005
[24]

Application of electrical network frequency of digital recordings for location-stamp verification,

M. Sarkar, D. Chowdhury, C. Shahnaz, and S. A. Fattah, “Application of electrical network frequency of digital recordings for location-stamp verification,” Applied Sciences, vol. 9, no. 15, p. 3135, 2019

2019
[25]

Improving location of recording classification using electric network frequency (ENF) analysis,

ˇZ. ˇSari´c, A. ˇZuni´c, T. Zrni´c, M. Kne ˇzevi´c, D. Despotovi ´c, and T. Deli ´c, “Improving location of recording classification using electric network frequency (ENF) analysis,” in Proc. of the IEEE International Sympo- sium on Intelligent Systems and Informatics , 2016, ...

2016
[26]

Advanced enf region clas- sification using units-sinspec: A novel approach integrating sinusoidal activation function and spectral attention,

Y . Li, T. Lu, G. Zeng, K. Zhao, and S. Peng, “Advanced enf region clas- sification using units-sinspec: A novel approach integrating sinusoidal activation function and spectral attention,” Applied Sciences , vol. 14, no. 19, p. 9081, 2024

2024
[27]

ENF power frequency data for location forensics,

A. Hajj-Ahmad, “ENF power frequency data for location forensics,” https://dx.doi.org/10.21227/H2159S, August 2016, Signal Processing Cup

2016 doi
[28]

Adam: A method for stochastic optimization,

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

2014 arXiv
[29]

InterGridNet: An electric network frequency approach for audio source location classification using convolutional neural networks,

“InterGridNet: An electric network frequency approach for audio source location classification using convolutional neural networks,” [Accessed: October 10, 2024]. [Online]. Available: https://github.com/ ckorgial/InterGridNet

2024
[30]

Exploring power signatures for location forensics of media recordings,

A. Triantafyllopoulos et al. , “Exploring power signatures for location forensics of media recordings,” University of Patras, Greece, Tech. Rep., 2016, Signal Processing Cup

2016
[31]

ENF based grid classification system: Identifying the region of origin of digital recordings,

R. Ohib, S. Y . Arnob, R. Arefin, M. Amin, and T. Reza, “ENF based grid classification system: Identifying the region of origin of digital recordings,” Criterion, vol. 3, no. 4, p. 5, 2017

2017
[32]

A novel ENF extraction approach for region-of-recording identification of media recordings,

M. El Helou, A. W. Turkmani, R. Chanouha, and S. Charbaji, “A novel ENF extraction approach for region-of-recording identification of media recordings,” Forensic Science International , vol. 155, no. 2-3, p. 165, 2005

2005
[33]

Exploring power signatures for location forensics of media recordings,

D. Despotovi ´c et al., “Exploring power signatures for location forensics of media recordings,” University of Novi Sad, Serbia, Tech. Rep., 2016, Signal Processing Cup

2016
[34]

Multi-harmonic histogram comparison,

C. Chow et al. , “Multi-harmonic histogram comparison,” Purdue Uni- versity, Tech. Rep., 2016, Signal Processing Cup

2016
[35]

Geographic location estimation from ENF signals with high accuracy,

H. Zhou, et al., “Geographic location estimation from ENF signals with high accuracy,” University of Science and Technology of China, Tech. Rep., 2016, Signal Processing Cup

2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [11]

On spectrogram analysis in a multiple classifier fusion framework for power grid clas- sification using electric network frequency,

G. Tzolopoulos, C. Korgialas, and C. Kotropoulos, “On spectrogram analysis in a multiple classifier fusion framework for power grid clas- sification using electric network frequency,” in Proc. of the 13th Inter- national Conference on Pattern Recognition Applications and Methods . SCITEPRESS, 2024, pp. 91–99

work page 2024

[2] [1]

Analysis of rolling shutter effect on ENF-based video forensics,

S. Vatansever, A. E. Dirik, and N. Memon, “Analysis of rolling shutter effect on ENF-based video forensics,”IEEE Transactions on Information Forensics and Security, vol. 14, no. 9, pp. 2262–2275, 2019

work page 2019

[3] [2]

Leveraging electric network frequency estimation for audio authentication,

C. Korgialas, C. Kotropoulos, and K. N. Plataniotis, “Leveraging electric network frequency estimation for audio authentication,” IEEE Access , 2024

work page 2024

[4] [3]

A robust RELAX-based algorithm for enhanced electric network frequency estimation,

C. Korgialas and C. Kotropoulos, “A robust RELAX-based algorithm for enhanced electric network frequency estimation,” in Proc. of the 13th Hellenic Conference on Artificial Intelligence (SETN). ACM, 2024, pp. 1–6

work page 2024

[5] [4]

Electric network frequency optical sensing devices,

C. Moysiadis, G. Karantaidis, and C. Kotropoulos, “Electric network frequency optical sensing devices,” in Proc. of the 2023 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applica- tions. IARIA, 2023, pp. 1–6

work page 2023

[6] [5]

Modeling and analysis of electric network frequency signal for timestamp verification,

R. Garg, A. L. Varna, and M. Wu, “Modeling and analysis of electric network frequency signal for timestamp verification,” in Proc. of the IEEE International Workshop on Information Forensics and Security (WIFS). IEEE, 2012, pp. 67–72

work page 2012

[7] [6]

Error analysis of forensic ENF matching,

G. Hua, “Error analysis of forensic ENF matching,” in Proc. of the IEEE International Workshop on Information Forensics and Security (WIFS , 2018, pp. 1–7

work page 2018

[8] [7]

Time-of-recording estimation for audio recordings,

L. Zheng, Y . Zhang, C. E. Lee, and V . L. L. Thing, “Time-of-recording estimation for audio recordings,” Digital Investigation , vol. 22, pp. S115–S126, 2017

work page 2017