Pith. sign in

REVIEW 4 major objections 5 minor 81 references

SONIC: Cost-Effective Web Access for Developing Countries

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 ↗ Review JSON

Pith's one-line read SONIC claims that existing FM broadcast infrastructure plus SMS can deliver useful web content and LLM chat answers to ordinary Android phones, and backs this with a six-week, 30-user deployment in Cameroon sustaining 10 kbps downlink and…

desk verdict A real deployment of FM-broadcast web delivery, honestly reported; the abstract overstates some numbers and the novelty claim needs toning down. read the letter →

arxiv 2505.16519 v1 pith:ZPDBIGLZ submitted 2025-05-22 cs.NI

Ayush Pandey , Rohail Asim , Jean Louis K. E. Fendji , Talal Rahwan , Matteo Varvello , Yasir Zaki This is my paper · ORCID
classification cs.NI
keywords FMradiodatabroadcastSMSuplinkdigitaldividewebaccessfordevelopingregionsdata-over-soundOFDMAndroidtunerLLMover
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

SONIC is a complete data-delivery system that treats an FM radio station as a broadcast downlink and SMS as a narrow uplink, aiming to give people who cannot afford mobile data a usable slice of the Web. The paper's central claim is that this combination works in the field: a six-week deployment in Cameroon sustained 10 kbps, kept loss below 20% for most transmissions when signal strength was above -90 dBm, and drew regular use for both webpage requests and LLM chat queries. The authors argue this matters because 2.6 billion people remain offline, and FM receivers are built into many affordable Android phones while FM broadcast costs stay fixed regardless of audience size. If correct, the system shows that repurposing existing broadcast infrastructure is a viable interim path to connectivity, not just a laboratory demonstration.

What carries the argument

The system's load-bearing piece is an audio-over-FM chain built on OFDM, a multi-carrier modulation that splits the channel into narrow orthogonal sub-carriers: 92 sub-carriers centered around 9.2 kHz carry 10 kbps of coded data, modulated by an open-source audio modem library. On the send side, webpages are rendered headlessly, captured as 320-pixel-wide WebP screenshots, wrapped in a structured file format whose headers mark metadata, link maps, and payload frames, and then pushed through the station's audio path. On the phone, a modified open-source Android build opens the built-in FM tuner to apps, the app decodes frames with CRC32 checksums plus inner and outer forward-error correction, and lost pixels are filled by nearest-neighbor interpolation, exploiting the left-to-right structure of text. Requests travel by SMS; interactive hyperlinks are stored as click coordinates, and a scored push mechanism prefetches the three most likely next links during idle time.

What would settle it

Run the same nightly broadcast for one week at a second FM station, measuring per-frame loss against signal strength; if the majority of receptions above -90 dBm exceed 20% loss, or if any single day's transmission fails because station staff did not flip the switch, the paper's reliability claim does not generalize.

Watch

Extended reading notes

Core claim

The paper claims that pre-rendered webpage screenshots and LLM responses can be encoded as OFDM audio, broadcast over an FM station's off-peak hours, and decoded on unrooted Android phones using their built-in FM tuners, with SMS providing the request channel. In its Cameroon deployment, transmissions sustained 10 kbps; a majority of receptions with signal strength above -90 dBm lost less than 20% of frames, and mean decoding accuracy was 71%. Users made 1,737 URL requests and 2,936 query requests over six weeks, and the broadcast nature meant every tuned-in phone received every transmission, feeding a shared knowledge hub even for users who never sent a request. The paper further claims one FM frequency can serve roughly 30 active users in a seven-hour nightly window, and 300 users with ten frequencies.

Load-bearing premise

The system assumes the radio station's audio chain, including the mixer, transmitter audio input, and the phone's FM receiver, carries the OFDM-modulated audio with little distortion, and that station staff reliably switch to the data broadcast at 10 PM; the paper reports days were lost when the switch was missed or the transmitter volume drifted from 100%.

Editorial extensions

If this is right

  • If the deployment numbers hold, a single seven-hour overnight window on one FM frequency can clear the backlog of about 30 active users, and scaling to 300 users requires roughly ten frequencies, suggesting capacity can grow by adding channels rather than redesigning the system.
  • Because FM is broadcast, content requested by one user reaches every tuned-in phone; the knowledge hub turns this into a free discovery channel and preserves downlink anonymity, since the server cannot know who is listening.
  • LLM responses arrive faster than full webpages because they are smaller, but a single lost 500-byte frame can break an entire response, whereas webpage images tolerate partial loss; completion rates therefore depend on content type and signal strength.
  • Users with median signal strength above -70 dBm consistently exceeded 80% request completion, while users below -90 dBm often fell below 30%, so receiver placement and antenna conditions determine whether the service is usable.
  • Once the station and SMS gateway exist, serving additional listeners adds almost no marginal cost, making advertising-supported or SMS-premium business models plausible.

Reading between the lines

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

  • Editorial inference: the same off-peak broadcast channel could prefetch popular pages and LLM answers into phone caches, reducing later cellular usage to interactive or personalized requests only.
  • Editorial inference: since LLM responses lose value with a single lost frame while webpage images tolerate partial loss, rebroadcasting short LLM responses with added redundancy would likely lift completion rates more than increasing the overall bitrate.
  • Editorial inference: moving from the analog audio path to an FM subcarrier or a dedicated digital channel could raise throughput well beyond 10 kbps without changing the client's request-by-SMS model.
  • Editorial inference: the push priority score was fit on click data from one online study population, so its coefficients may shift for other regions; logging clicks on pushed links during deployment would let the metric adapt to local browsing habits.
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.

Referee Report

4 major / 5 minor

Summary. The paper presents SONIC, a system that delivers simplified web content and LLM-generated responses to low-end smartphones using FM radio as a broadcast downlink and SMS as an uplink. The authors describe the server and client architecture, a modified LineageOS build that exposes FM tuner audio to applications, image-based webpage compression with pixel interpolation, and a hyperlink-pushing heuristic. They report a six-week deployment at an FM radio station in Cameroon with 30 users, and claim a sustained downlink throughput of 10 kbps, less than 20% transmission loss for a majority of transmissions above -90 dBm, and strong user engagement across web browsing and ChatGPT interactions.

Significance. If the headline claims are supported, SONIC would be a genuinely useful contribution to low-cost connectivity in low-income regions: it repurposes near-ubiquitous FM radio and SMS infrastructure, and the six-week deployment with 30 real users is a substantial step beyond lab-only evaluations. The paper also makes a credible case that FM-capable phones are widely available and affordable in the target regions. The main concerns are not about the overall concept but about whether the abstract's quantitative claims are backed by the reported measurements.

major comments (4)
  1. [§6, Figure 14, and abstract] The abstract claims 'less than 20% loss for a majority of transmissions with signal strength above -90 dBm,' but the paper does not report the statistic that directly supports this claim. Figure 14 shows violin plots with bin widths labeled as percentages of total transmissions, but it does not state the median loss per bin or the fraction of transmissions with loss below 20% for the RSSI > -90 dBm subset. The text in §6 says data points are 'more concentrated below 20%' for the -80 to -50 dBm range, which is not a quantitative majority statement. Relatedly, 'loss' is not defined: it could mean lost frames, lost bytes, or failed transmissions, and the denominator is unclear. Please report the exact fraction of transmissions with loss below 20% among those with RSSI above -90 dBm, and define the loss metric precisely.
  2. [§3.1.1 and §6] The abstract and Section 1 describe 'a sustained downlink throughput of 10 kbps' as a demonstrated result, but Section 3.1.1 defines 10 kbps as the modulation rate of the Quiet OFDM profile (92 sub-carriers, 9.2 kHz center frequency). Section 6 contains no measured throughput or goodput data, such as bytes successfully delivered per hour or per transmission window. As written, the headline number is a configuration parameter, not a measured result. Please either remove 'sustained' and rephrase as the configured modulation bitrate, or add measured throughput/goodput from the deployment.
  3. [§1 and §6] Section 1 states that 'Mean decoding accuracy remained at 71% under real-world conditions,' but this metric is never defined and does not appear anywhere in Section 6. The evaluation instead reports transmission loss percentages and request completion rates. If 'decoding accuracy' refers to the fraction of successfully decoded frames, bytes, or transmissions, that definition and the supporting measurement must be provided; otherwise, the sentence should be removed from the introduction.
  4. [§6, Figure 16] The scalability analysis is based on 'FCFS queue simulations' whose methodology is not described. The paper does not state how the baseline traffic was scaled, how cache behavior was modeled, what service time was assumed per request, whether the simulated players experience transmission loss, or how the number of frequencies changes the service rate. Since the abstract and conclusion describe SONIC as 'scalable,' these simulation assumptions need to be specified, or the figure should be explicitly labeled as an illustrative back-of-the-envelope extrapolation rather than a simulation result.
minor comments (5)
  1. [Abstract and §1] The unit is written as '-90 dbM' and 'dBM'; it should be '-90 dBm' throughout.
  2. [§2] The sentence 'These results sheds light on the affordability of web access' should be 'These results shed light...'.
  3. [§6, Figure 14] The percentages printed beneath the RSSI bins are the proportions of transmissions, but the violin plot itself is not accompanied by a legend describing what the median, quartiles, or whiskers represent; please add a legend or a short caption explaining the plot elements.
  4. [§6, Figure 17] The term 'completion rate' is used but not defined; it should state whether a request counts as complete only when all frames are decoded, or whether partial content counts.
  5. [§4, Figure 10] The line 'over 95% of transmissions experienced loss rates below 10%' refers to ideal lab conditions, but this is easy to misread as a field result because Figure 14 is discussed later; consider adding an explicit qualifier such as 'under controlled lab conditions' in the sentence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; deployment results are empirical and self-contained, with only minor non-load-bearing self-citations.

full rationale

The paper's central results are measured deployment outcomes: sustained 10 kbps throughput, loss rates versus RSSI, completion rates, and user engagement from a six-week live FM station deployment in Cameroon. These are empirical observations, not quantities derived by fitting parameters and then re-reporting them as predictions. The pushing heuristic (score=0.68*w*h - 0.32*y) is transparently described as a logistic-regression fit from a Prolific user study and is used only as a prioritization heuristic for idle-time link pushing; it is not presented as a validated scientific prediction and it does not feed back into the throughput, loss, or engagement claims. The only substantive self-citation is reference [57], the authors' own prior CoNEXT work, which is used for the initial SONIC framework and for the pixel-interpolation benchmark: the paper states 'we adopt a lightweight approach proposed and benchmarked by [57] that provides consistently high content readability scores even at a 20% pixel loss rate.' That benchmark is an external prior user study from published work, not a fitted value of this paper, and it is not load-bearing for the deployment measurements. The paper also candidly discloses operational failures (inconsistent station switch-over, transmitter volume requirements, power outages, SIM blocking) that act as independent constraints rather than circular supports. No equation in the paper is defined in terms of its own output, and no 'uniqueness theorem' or ansatz is imported from the authors' prior work to force a conclusion. Any concern about whether the abstract's 'less than 20% loss for a majority of transmissions' is fully supported by the violin-plot medians in Figure 14 is a data-reporting or correctness question, not a circularity question. Overall, the derivation chain is self-contained against real-world external measurements, so the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The system relies on several domain assumptions about FM audio fidelity, Android FM access, server internet stability, and SMS uplink availability. The only explicitly fitted parameters are the link-pushing regression coefficients and hand-chosen image compression settings.

free parameters (4)
  • Link-pushing regression coefficients = area weight = 0.68, y-position weight = -0.32
    Fitted via logistic regression on 1,000 page interactions from 100 Prolific participants (Section 4). Used to prioritize which hyperlinks to push during idle periods.
  • Screenshot width = 320 pixels
    Empirically selected as 'a sweet spot where both the content layout and text remain comfortably readable' (Section 3.1.1).
  • WebP quality = 10%
    Chosen to significantly reduce file size while maintaining readability (Section 3.1.1). No quantitative readability-vs-quality analysis is reported at this paper's deployment site.
  • Frame size = 500 bytes
    Design choice for the SONIC file format's payload frames (Section 3.1.1). Used with crc32 and FEC schemes.
assumptions (5)
  • domain assumption The FM radio station's audio chain (mixer, transmitter, and receiver audio path) faithfully transports the OFDM-modulated audio signal.
    The whole system depends on this; the paper notes the transmitter output volume had to be at 100% and station staff had to correctly switch at 10 PM (Section 5).
  • domain assumption The modified LineageOS build can expose FM radio audio to third-party apps without root on a range of Android devices.
    Verified only on Xiaomi Redmi Go with Qualcomm Snapdragon 425 (Section 3.2); generalizability is assumed.
  • domain assumption The SONIC server has stable internet access to fetch webpages and call LLM APIs.
    In the deployment, MTN connectivity was unstable and occasionally unavailable for entire days (Section 5).
  • domain assumption Users have access to SMS service for the uplink.
    SMS is required for personalized requests; participants were given SIM cards with unlimited SMS bundles, which may not reflect real users' costs.
  • domain assumption Pixel interpolation recovers lost frames well enough for reading left-to-right text pages.
    Adopted from prior work [57]; relies on text pages being mostly left-to-right so nearest-left-pixel replacement works (Section 3.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of SONIC: Cost-Effective Web Access for Developing Countries." pith.science (2026). https://pith.science/paper/ZPDBIGLZ

@misc{pith2026250516519,
  author       = {Pith},
  title        = {Pith review of: SONIC: Cost-Effective Web Access for Developing Countries},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZPDBIGLZ}},
  note         = {Machine review of arXiv:2505.16519}
}
read the original abstract

Over 2.6 billion people remain without access to the Internet in 2025. This phenomenon is especially pronounced in developing regions, where cost and infrastructure limitations are major barriers to connectivity. In response, we design SONIC, a low-cost, scalable data delivery system that builds on existing infrastructures: FM radio for downlink broadcasting, and SMS for personalized uplink. SONIC is motivated by the widespread availability of FM radio and SMS infrastructure in developing regions, along with embedded FM radio tuners in affordable mobile phones. SONIC offers several innovations to effectively transmit Web content over sound over FM radio, in a reliable and compressed form. For example, we transmit pre-rendered webpages and leverage pixel interpolation to recover errors at the receiver. We further modify Android to offer a simpler deployment pipeline, supporting a wide range of devices. We deployed SONIC at an FM radio station in Cameroon for six weeks with 30 participants. Our results demonstrate a sustained downlink throughput of 10 kbps, less than 20% loss for a majority of transmissions with signal strength above -90 dbM, and a strong user engagement across both Web browsing and ChatGPT interactions.

Figures

Figures reproduced from arXiv: 2505.16519 by the authors.

Figure 1
Figure 1. Network performance of MTN, the best available mobile network provider at our deployment site in Cameroon. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Barriers to Web access from speed and cost. The [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. CDF of prices for android phones equipped with an [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (13 more)
Figure 6
Figure 6. Figure 6: shows the architecture of the SONIC server. The server runs as a Docker [35] container on a computer located at a radio station. It consists of several key components that work together to process user requests. The SMS Manager handles incoming messages, while the Scre…
Figure 7
Figure 7. Figure 7: SONIC file format. where the payload is simply appended to metadata, encoding webpages presents a significant challenge due to the limited data rates achievable over audio transmission—typically only tens of kbps (see Section 7). According to [16], the average mobile w…
Figure 8
Figure 8. Figure 8: SONIC client architecture. news and blogs, this metric prioritizes pages with top head￾lines, larger images, and prominent font sizes, ensuring that key content is already available when users attempt to follow links. Additionally, it penalizes links that are verticall…
Figure 10
Figure 10. Figure 10: CDF of loss percentage under ideal conditions. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: Likelihood of clicking on a hyperlink in a webpage [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: FM radio station in Cameroon, and SONIC server [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: Location map of the 30 SONIC users with their [PITH_FULL_IMAGE:figures/full_fig_p009_13.png]
Figure 14
Figure 14. Figure 14: Loss % as a function of RSSI range. Violin’s [PITH_FULL_IMAGE:figures/full_fig_p010_14.png]
Figure 15
Figure 15. Figure 15: CDF of loss % for GPT and URL transmissions. [PITH_FULL_IMAGE:figures/full_fig_p010_15.png]
Figure 17
Figure 17. Figure 17: RSSI distribution per user with their request completion rates. Users 13, 20, and 23 were removed as their devices [PITH_FULL_IMAGE:figures/full_fig_p011_17.png]
Figure 18
Figure 18. Figure 18: Top 10 categories of URL and GPT content re [PITH_FULL_IMAGE:figures/full_fig_p012_18.png]
Figure 19
Figure 19. Figure 19: User experience of SONIC on system reliability, [PITH_FULL_IMAGE:figures/full_fig_p012_19.png]
Figure 20
Figure 20. Figure 20: Distribution of user ratings (0–10) for top 50 Pak [PITH_FULL_IMAGE:figures/full_fig_p016_20.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

81 extracted references · 71 canonical work pages

  1. [1]

    https://developer.chrome.com/docs/ lighthouse

    Lighthouse. https://developer.chrome.com/docs/ lighthouse. 12

  2. [2]

    https://www.prolific.com/

    Prolific | quickly find research participants you can trust. https://www.prolific.com/. Accessed: 2025-04-12

    work page 2025

  3. [3]

    URL: http: //www.seleniumhq.org/about/

    Seleniumhq browser automation, 2017. URL: http: //www.seleniumhq.org/about/

    work page 2017

  4. [4]

    www.citiprogram.org, 2019

    Citi program - collaborative institutional training initia- tive. www.citiprogram.org, 2019. Accessed: 2022-05- 18

    work page 2019

  5. [5]

    https:// developers.google.com/speed/webp

    An image format for the Web., 2023. https:// developers.google.com/speed/webp

    work page 2023

  6. [6]

    https://github.com/quiet/ org.quietmodem.Quiet

    Android app built with Quiet which allows o pass data through the speakers of an Android device., 2023. https://github.com/quiet/ org.quietmodem.Quiet

    work page 2023

  7. [7]

    https://github.com/ jamesonrader/AudioQR

    AudioQR, 2023. https://github.com/ jamesonrader/AudioQR

    work page 2023

  8. [8]

    Expanding internet connectivity with stratospheric bal- loons, 2023.https://x.company/projects/loon/

    work page 2023

Show all 81 references
  1. [9]

    https://github.com/ggerganov/ ggwave

    GGwave, 2023. https://github.com/ggerganov/ ggwave

    2023

  2. [10]

    High altitude connectivity: The next chapter,

  3. [11]

    https://support.google.com/websearch/answer/ 9836344?hl=en

    Load mobile pages faster with Web Light, 2023. https://support.google.com/websearch/answer/ 9836344?hl=en

    2023

  4. [12]

    https://en.wikipedia.org/ wiki/MSN_Direct

    MSN Direct, 2023. https://en.wikipedia.org/ wiki/MSN_Direct

    2023

  5. [13]

    https://github.com/ quiet/quiet

    Quiet modem project., 2023. https://github.com/ quiet/quiet

    2023

  6. [14]

    https:// support.apple.com/guide/safari/hide-ads- when-reading-sfri32632/mac

    Safari Reader Mode, 2023. https:// support.apple.com/guide/safari/hide-ads- when-reading-sfri32632/mac

    2023

  7. [15]

    Starlink, 2023.https://starlink.com/

    2023

  8. [16]

    https:// almanac.httparchive.org/en/2022/

    Web Almanac, 2023. https:// almanac.httparchive.org/en/2022/

    2023

  9. [17]

    Accessed: 2025-03-

    Taara: A google x moonshot, 2025. Accessed: 2025-03-

    2025

  10. [18]

    Africa’s shifting media landscapes: Digital media use grows, but so do demographic divides,

    Afrobarometer. Africa’s shifting media landscapes: Digital media use grows, but so do demographic divides,

  11. [19]

    URL:https://x.company/projects/taara/

  12. [20]

    Audience listenership of fm radio: A case study of rural development in northern ghana.Cogent Arts & Humanities, 10(1):2184750, 2023

    Osman Antwi-Boateng, Muhammed Danladi Musa, and Mu-Azu Iddirisu Andani and. Audience listenership of fm radio: A case study of rural development in northern ghana.Cogent Arts & Humanities, 10(1):2184750, 2023. arXiv:https: //doi.org/10.1080/23311983.2023.2184750, doi:10.1080/2...

    2023

  13. [21]

    Ashigwuike, Ale Felix, and Farouq E

    Evans C. Ashigwuike, Ale Felix, and Farouq E. Shaibu. The impact of soil texture on path loss modelling of an fm signal using diffraction technique: A case study of prime fm radio nigeria.Euro- pean Journal of Engineering Research and Science, 4(4):56–63, April 2019. URL: http...

    2019 doi

  14. [22]

    Broadcastreceiver,

    Android Developers. Broadcastreceiver,

  15. [23]

    Signal coverage mapping of local radios

    Tina Baidar. Signal coverage mapping of local radios. Journal on Geoinformatics, Nepal, 2017. Accessed via Nepal Journals Online

    2017

  16. [24]

    Driveshaft: Improving perceived mobile web performance.arXiv preprint arXiv:1809.09292, 2018

    Ketan Bhardwaj, Ada Gavrilovska, Moritz Steiner, Mar- tin Flack, and Stephen Ludin. Driveshaft: Improving perceived mobile web performance.arXiv preprint arXiv:1809.09292, 2018

    2018 arXiv

  17. [25]

    Drivers’ warning application through personalized dsss-cdma data transmission by using the fm radio broadcasting infrastructure.IEEE Access, 11:11711–11731, 2023

    Radu Gabriel Bozomitu and Florin Doru Hutu. Drivers’ warning application through personalized dsss-cdma data transmission by using the fm radio broadcasting infrastructure.IEEE Access, 11:11711–11731, 2023

    2023

  18. [26]

    Batcomm: enabling inaudible acoustic communication with high-throughput for mobile de- vices

    Yang Bai, Jian Liu, Li Lu, Yilin Yang, Yingying Chen, and Jiadi Yu. Batcomm: enabling inaudible acoustic communication with high-throughput for mobile de- vices. InProceedings of the 18th Conference on Embed- ded Networked Sensor Systems, pages 205–217, 2020

    2020

  19. [27]

    Madhyastha, and Vyas Sekar

    Michael Butkiewicz, Daimeng Wang, Zhe Wu, Harsha V . Madhyastha, and Vyas Sekar. Klot- ski: Reprioritizing web content to improve user experience on mobile devices. In12th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’15), Oakland, CA, USA,

  20. [28]

    Deep learning for irregularly and regularly missing data reconstruction.Scientific Reports, 10(1):3302, Feb 2020

    Xintao Chai, Hanming Gu, Feng Li, Hongyou Duan, Xiaobo Hu, and Kai Lin. Deep learning for irregularly and regularly missing data reconstruction.Scientific Reports, 10(1):3302, Feb 2020. doi:10.1038/s41598- 020-59801-x

    2020 doi

  21. [29]

    Jsanalyzer: A web developer tool for simpli- fying mobile web pages through non-critical javascript elimination.ACM Transactions on the Web, 16(4):1–31, 2022

    Moumena Chaqfeh, Russell Coke, Jacinta Hu, Waleed Hashmi, Lakshmi Subramanian, Talal Rahwan, and Yasir Zaki. Jsanalyzer: A web developer tool for simpli- fying mobile web pages through non-critical javascript elimination.ACM Transactions on the Web, 16(4):1–31, 2022

    2022

  22. [30]

    A robust radiocommunication system for fm transmission based on software defined radio technology

    Radu Gabriel Bozomitu and ¸ Stefan Corneliu Stoica. A robust radiocommunication system for fm transmission based on software defined radio technology. In2022 IEEE 28th International Symposium for Design and Technology in Electronic Packaging (SIITME), pages 78–81. IEEE, 2022

    2022

  23. [31]

    Jscleaner: De-cluttering mobile webpages through javascript cleanup

    Moumena Chaqfeh, Yasir Zaki, Jacinta Hu, and Lakshmi Subramanian. Jscleaner: De-cluttering mobile webpages through javascript cleanup. InProceedings of The Web Conference 2020, pages 763–773, 2020

    2020

  24. [32]

    Accessed: 2025-03-22

    Cloudflare, Inc.Threat Intelligence APIs - Secu- rity Center, 2025. Accessed: 2025-03-22. URL: https://developers.cloudflare.com/security- center/intel-apis/

    2025

  25. [33]

    Global smartphone market share: Quarterly

    Counterpoint Research. Global smartphone market share: Quarterly. 2025. URL: https://www.counterpointresearch.com/ insights/global-smartphone-share/

    2025

  26. [34]

    Digital 2025: Cameroon, 2025

    DataReportal. Digital 2025: Cameroon, 2025. Accessed: 2025-03-26. URL: https://datareportal.com/ reports/digital-2025-cameroon

    2025

  27. [35]

    To block or not to block: Accelerating mobile web pages on-the-fly through javascript classification

    Moumena Chaqfeh, Muhammad Haseeb, Waleed Hashmi, Patrick Inshuti, Manesha Ramesh, Matteo Varvello, Lakshmi Subramanian, Fareed Zaffar, and Yasir Zaki. To block or not to block: Accelerating mobile web pages on-the-fly through javascript classification. InProceedings of the 12t...

    2022

  28. [36]

    Bufferbloat: dark buffers in the internet.Communications of the ACM, 55(1):57–65, 2012

    Jim Gettys and Kathleen Nichols. Bufferbloat: dark buffers in the internet.Communications of the ACM, 55(1):57–65, 2012

    2012

  29. [37]

    Speedreader: Reader mode made fast and private

    Mohammad Ghasemisharif, Peter Snyder, Andrius Auci- nas, and Benjamin Livshits. Speedreader: Reader mode made fast and private. InThe World Wide Web Confer- ence, pages 526–537, 2019

    2019

  30. [38]

    Amp is a web component framework to easily create user-first web experiences - amp.dev

    Google. Amp is a web component framework to easily create user-first web experiences - amp.dev. https: //amp.dev, 2019. Accessed: 2019-05-05

    2019

  31. [39]

    The llama 3 herd of models, 2024

    Aaron Grattafiori, Laurens Van Der Maaten, et al. The llama 3 herd of models, 2024. https://arxiv.org/ abs/2407.21783.arXiv:2407.21783

    2024 arXiv

  32. [40]

    Inc. Docker. Docker: Open platform for developing, shipping, and running applications, 2024. URL: https: //www.docker.com/

    2024

  33. [41]

    GSMArena - Mobile Phone Specifications, News, Reviews, 2025

    GSMArena. GSMArena - Mobile Phone Specifications, News, Reviews, 2025. Accessed: 21 Mar. 2025. URL: https://www.gsmarena.com/

    2025

  34. [42]

    A framework for im- proving web affordability and inclusiveness

    Rumaisa Habib, Sarah Tanveer, Aimen Inam, Haseeb Ahmed, Ayesha Ali, Zartash Afzal Uzmi, Zafar Ayyub Qazi, and Ihsan Ayyub Qazi. A framework for im- proving web affordability and inclusiveness. InPro- ceedings of the ACM SIGCOMM 2023 Conference, ACM SIGCOMM ’23, page 592–607, N...

    2023

  35. [43]

    Mobile wifi 4g router huawei e8372h-517 wifi modem

    Huawei Technologies Co., Ltd. Mobile wifi 4g router huawei e8372h-517 wifi modem. Online. Accessed: 25 March 2025. URL: https://sincereonetech.com/products/4g- router-huawei-e8372h-517-wifi-modem/

    2025

  36. [44]

    Near-ultrasound communication for tv’s 2nd screen services

    Soonwon Ka, Tae Hyun Kim, Jae Yeol Ha, Sun Hong Lim, Su Cheol Shin, Jun Won Choi, Chulyoung Kwak, and Sunghyun Choi. Near-ultrasound communication for tv’s 2nd screen services. InProceedings of the 22nd Annual International Conference on Mobile Computing and Networking, pages ...

    2016

  37. [45]

    The state of mobile internet connectivity report 2024, 2024

    GSMA. The state of mobile internet connectivity report 2024, 2024. Accessed: 2025-02-05. URL: https://www.gsma.com/r/wp-content/uploads/ 2024/10/The-State-of-Mobile-Internet- Connectivity-Report-2024.pdf

    2024

  38. [46]

    Muzeel: assessing the impact of javascript dead code elimination on mobile web per- formance

    Jesutofunmi Kupoluyi, Moumena Chaqfeh, Matteo Varvello, Russell Coke, Waleed Hashmi, Lakshmi Sub- ramanian, and Yasir Zaki. Muzeel: assessing the impact of javascript dead code elimination on mobile web per- formance. InProceedings of the 22nd ACM Internet Measurement Conferen...

    2022

  39. [47]

    Muzeel: A dynamic javascript analyzer for dead code elimination in today’s web.arXiv preprint arXiv:2106.08948, 2021

    Tofunmi Kupoluyi, Moumena Chaqfeh, Matteo Varvello, Waleed Hashmi, Lakshmi Subramanian, and Yasir Zaki. Muzeel: A dynamic javascript analyzer for dead code elimination in today’s web.arXiv preprint arXiv:2106.08948, 2021

    2021 arXiv

  40. [48]

    Tranco: A research-oriented top sites ranking hardened 14 against manipulation

    Victor Le Pochat, Tom Van Goethem, Samaneh Tajal- izadehkhoob, Maciej Korczy´nski, and Wouter Joosen. Tranco: A research-oriented top sites ranking hardened 14 against manipulation. InProceedings of the 26th An- nual Network and Distributed System Security Sym- posium, NDSS 20...

    2019

  41. [49]

    Chirp signal-based aerial acoustic com- munication for smart devices

    Hyewon Lee, Tae Hyun Kim, Jun Won Choi, and Sunghyun Choi. Chirp signal-based aerial acoustic com- munication for smart devices. In2015 IEEE Confer- ence on Computer Communications (INFOCOM), pages 2407–2415. IEEE, 2015

    2015

  42. [50]

    My traceroute (MTR)

    Matt Kimball. My traceroute (MTR). https:// www.bitwizard.nl/mtr/

  43. [51]

    Javascript dead code identi- fication, elimination, and empirical assessment.IEEE Transactions on Software Engineering, 2023

    Ivano Malavolta, Kishan Nirghin, Gian Luca Scoccia, Simone Romano, Salvatore Lombardi, Giuseppe Scan- niello, and Patricia Lago. Javascript dead code identi- fication, elimination, and empirical assessment.IEEE Transactions on Software Engineering, 2023

    2023

  44. [52]

    How the core web vitals metrics thresholds were defined, 2020

    Bryan McQuade and Barry Pollard. How the core web vitals metrics thresholds were defined, 2020. Accessed: 2025-04-16. URL: https://web.dev/articles/ defining-core-web-vitals-thresholds

    2020

  45. [53]

    Mtn cameroon official website, 2025

    MTN Cameroon. Mtn cameroon official website, 2025. Accessed: 2025-03-20. URL:https://mtn.cm/

    2025

  46. [54]

    Polaris: Faster page loads using fine-grained dependency tracking

    Ravi Netravali, Ameesh Goyal, James Mickens, and Hari Balakrishnan. Polaris: Faster page loads using fine-grained dependency tracking. In13th USENIX Symposium on Networked Sys- tems Design and Implementation (NSDI 16), Santa Clara, CA, 2016. USENIX Association. URL: https://ww...

    2016

  47. [55]

    Lineageos android distribution, 2024

    LineageOS. Lineageos android distribution, 2024. A free and open-source operating system for various de- vices, based on the Android mobile platform. URL: https://lineageos.org/

    2024

  48. [56]

    Openai api reference, 2024

    OpenAI. Openai api reference, 2024. Accessed: 2025- 03-24. URL: https://platform.openai.com/docs/ api-reference/chat

    2024

  49. [57]

    Sonic: Connect the uncon- nected via fm radio & sms

    Ayush Pandey, Rohail Asim, Khalid Mengal, Matteo Varvello, and Yasir Zaki. Sonic: Connect the uncon- nected via fm radio & sms. InProceedings of the 20th International Conference on Emerging Networking EX- periments and Technologies, CoNEXT ’24, page 41–47, New York, NY , USA,...

    2024

  50. [58]

    Maml: Towards a faster web in developing regions

    Ayush Pandey, Matteo Varvello, Syed Ishtiaque Ahmed, Shurui Zhou, Lakshmi Subramanian, and Yasir Zaki. Maml: Towards a faster web in developing regions. InProceedings of the ACM on Web Conference 2025, WWW ’25, page 727–739, New York, NY , USA, 2025. Association for Computing ...

    2025

  51. [59]

    Image inpainting based on deep learning: A review.Displays, 69:102028,

    Zhen Qin, Qingliang Zeng, Yixin Zong, and Fan Xu. Image inpainting based on deep learning: A review.Displays, 69:102028,

  52. [60]

    Speedtest® CLI, Internet connection measure- ment for developers

    OOKLA. Speedtest® CLI, Internet connection measure- ment for developers. https://www.speedtest.net/apps/cli

  53. [61]

    Backdoor: Making microphones hear inaudible sounds

    Nirupam Roy, Haitham Hassanieh, and Romit Roy Choudhury. Backdoor: Making microphones hear inaudible sounds. InProceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services, pages 2–14, 2017

    2017

  54. [62]

    Nearest neighbor value interpolation.International Journal of Advanced Computer Science and Applications, 3(4), 2012

    Olivier Rukundo and Hanqiang Cao. Nearest neighbor value interpolation.International Journal of Advanced Computer Science and Applications, 3(4), 2012. URL: http://dx.doi.org/10.14569/IJACSA.2012.030405, doi:10.14569/IJACSA.2012.030405

    2012 arXiv

  55. [63]

    A world wide view of browsing the world wide web

    Kimberly Ruth, Aurore Fass, Jonathan Azose, Mark Pearson, Emma Thomas, Caitlin Sadowski, and Zakir Durumeric. A world wide view of browsing the world wide web. InProceedings of the 22nd ACM Internet Measurement Conference, IMC ’22, page 317–336, New York, NY , USA, 2022. Assoc...

    2022

  56. [64]

    Huawei lte api

    Salamek. Huawei lte api. GitHub repository, 2025. Ac- cessed: 25 March 2025. URL: https://github.com/ Salamek/huawei-lte-api

    2025

  57. [65]

    A software- defined ultrasonic networking framework for wearable devices.IEEE/ACM Transactions on Networking, 25(2):960–973, 2016

    G Enrico Santagati and Tommaso Melodia. A software- defined ultrasonic networking framework for wearable devices.IEEE/ACM Transactions on Networking, 25(2):960–973, 2016

    2016

  58. [66]

    Dimension c-137, 2025

    Rick and Morty Wiki Contributors. Dimension c-137, 2025. Accessed: 24 Mar. 2025. URL: https://rickandmorty.fandom.com/wiki/ Dimension_C-137

    2025

  59. [67]

    Digital 2025 global overview report, 2025

    We Are Social and Meltwater. Digital 2025 global overview report, 2025. Retrieved on 18 March

    2025

  60. [68]

    Recovery of images with miss- ing pixels using a gradient compressive sensing algo- rithm.ArXiv, abs/1407.3695, 2014

    Isidora Stankovi ´c. Recovery of images with miss- ing pixels using a gradient compressive sensing algo- rithm.ArXiv, abs/1407.3695, 2014. URL: https: //api.semanticscholar.org/CorpusID:15997634

    2014 arXiv

  61. [69]

    Push-to-peer video-on-demand system: Design and evaluation.IEEE Journal on Selected Areas in Communications, 25(9):1706–1716, 2007

    Kyoungwon Suh, Christophe Diot, Jim Kurose, Laurent Massoulie, Christoph Neumann, Don Towsley, and Mat- teo Varvello. Push-to-peer video-on-demand system: Design and evaluation.IEEE Journal on Selected Areas in Communications, 25(9):1706–1716, 2007

    2007

  62. [70]

    Per- formance characterization of videoconferencing in the wild

    Matteo Varvello, Hyunseok Chang, and Yasir Zaki. Per- formance characterization of videoconferencing in the wild. InProceedings of the 22nd ACM Internet Mea- surement Conference, pages 261–273, 2022

    2022

  63. [71]

    A worldwide look into mobile access networks through the eyes of ami- gos

    Matteo Varvello and Yasir Zaki. A worldwide look into mobile access networks through the eyes of ami- gos. InTMA 2023 - Proceedings of the 7th Net- work Traffic Measurement and Analysis Conference. Institute of Electrical and Electronics Engineers Inc. doi:10.23919/TMA58422.20...

    2023

  64. [72]

    Revcast: Fast, private certificate revocation over fm radio

    Aaron Schulman, Dave Levin, and Neil Spring. Revcast: Fast, private certificate revocation over fm radio. In Proceedings of the 2014 ACM SIGSAC Conference on Computer and Communications Security, pages 799– 810, 2014

    2014

  65. [73]

    Speeding up web page loads with shandian

    Xiao Sophia Wang, Arvind Krishnamurthy, and David Wetherall. Speeding up web page loads with shandian. In13th USENIX Symposium on Networked Systems Design and Implementation (NSDI 16), pages 109–122, Santa Clara, CA, 2016. USENIX Association. URL: https://www.usenix.org/confer...

    2016

  66. [74]

    URL: https://datareportal.com/reports/ digital-2025-global-overview-report. 15

    2025

  67. [79]

    Springer Berlin Heidelberg, Berlin, Heidelberg, 1995

    Hans Wackernagel.Ordinary Kriging, pages 74–81. Springer Berlin Heidelberg, Berlin, Heidelberg, 1995. doi:10.1007/978-3-662-03098-1_11

    1995 doi

  68. [81]

    Worlddata.info: The world in num- bers, 2025

    WorldData.info. Worlddata.info: The world in num- bers, 2025. Accessed: 2025-04-08. URL: https: //www.worlddata.info/. A Pixel Interpolation Figure 20 contains the results from a user study conducted by [57] to benchmark the effectiveness of their pixel interpo- lation approac...

    2025

  69. [2015]

    Open Access

    USENIX Association. Open Access. URL: https://www.usenix.org/conference/nsdi15/ technical-sessions/presentation/butkiewicz. 13

  70. [2021]

    URL: https://www.sciencedirect.com/ science/article/pii/S0141938221000391, doi:10.1016/j.displa.2021.102028

    2021

  71. [2023]

    https://engineering.fb.com/2018/06/ 27/connectivity/high-altitude-connectivity- the-next-chapter/

    2018

  72. [2024]

    URL: https://www.afrobarometer.org/ wp-content/uploads/2024/04/AD800-PAP14- Digital-media-use-grows-across-Africa% 5ELJ-but-so-do-demographic-divides- Afrobarometer-29april24.pdf

    2024

  73. [2025]

    URL: https: //developer.android.com/develop/background- work/background-tasks/broadcasts

    Accessed: 2025-04-10. URL: https: //developer.android.com/develop/background- work/background-tasks/broadcasts

    2025

Pith tools

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