Title : Tuning into the Web: A Low-Cost Access Solution for Developing Countries
Authors : Ayush Pandey, Rohail Asim(New York University Abu Dhabi); Jean Louis K. E. Fendji (University of Ngaoundere); Talal Rahwan(New York University Abu Dhabi); Matteo Varvello (Nokia Bell Labs); Yasir Zaki (New York University Abu Dhabi)
Scribe : Yixiu Zhou (Xiamen University)
Introduction
Although Internet access has become essential for education, healthcare, employment, and information, billions of people remain offline because mobile devices, data plans, and network infrastructure are too expensive or unreliable. Existing solutions such as cellular networks, satellites, high-altitude platforms, and laser-based systems often require costly infrastructure or specialized equipment. Previous attempts to transmit web content over FM radio were mainly proof-of-concept studies based on simulations, short-range experiments, or lossless channels. They did not address real-world signal variability, user requests, content scheduling, caching, or long-term operation. This paper studies how existing FM radio infrastructure can be repurposed to provide low-cost web and AI access in developing regions.
Key Idea and Contribution
The authors propose WebFM, an end-to-end system that uses FM radio for downlink broadcasting and SMS for personalized uplink requests. Users send a webpage URL or an LLM prompt to the server through SMS. The server renders webpages as compressed WebP screenshots or obtains responses from an LLM, encodes the content into a loss-resilient file format, converts it into audio using OFDM, and broadcasts it through an FM radio station. Any compatible phone in the coverage area can receive and decode the broadcast.
WebFM includes several system-level contributions. It introduces a frame-based file format with metadata markers, checksums, and forward error correction to handle missing or corrupted data. It modifies LineageOS so that applications can access built-in FM tuners without root privileges, and it uses lightweight pixel interpolation to make webpage screenshots readable even when some data is lost. The system also caches frequently requested pages and proactively broadcasts popular pages and important internal links during idle periods. In addition, WebFM supports ChatGPT-like interactions through SMS, allowing users to submit prompts and receive model responses without direct Internet access.
Evaluation
The authors evaluate WebFM through controlled experiments and a six-week deployment at a real FM radio station in Cameroon with 30 participants. The system achieves a sustained downlink rate of 10 kbps. In the real-world deployment, most transmissions experience less than 20% loss when the received signal strength is above -90 dBm, and users generate 1,737 webpage requests and 2,936 GPT queries. The median battery consumption is about 4.1% per hour, while 72% of users rate the interface as intuitive and the content as useful. The results show that one FM frequency can support approximately 30 active users under similar traffic conditions, while caching significantly reduces the transmission queue. This result is significant because it demonstrates that widely available, low-cost broadcast infrastructure can provide practical access to web content and AI services in regions where conventional Internet connectivity is unaffordable or unreliable.
Q : Can WebFM still work if SMS services are unavailable?
A : Without SMS, WebFM cannot support personalized requests or interactive communication. However, the system can still broadcast popular webpages, news, and government information periodically over FM and cache them on users’ phones. Users can browse this pre-distributed content, although they cannot submit custom requests. Alternative uplink mechanisms could also be explored as a replacement for SMS.
Q : Why not deploy 3G networks or Starlink instead of using FM radio?
A : The main considerations are affordability and deployment cost. Satellite services may require expensive hardware and monthly subscriptions, while deploying cellular base stations in remote areas can be difficult because of geographic constraints and high maintenance costs. For example, although some regions of Cameroon have 2G or 3G coverage, users may still need to pay about $14 per month for 9 GB of data, which is unaffordable for many low-income users. WebFM provides a lower-cost alternative by reusing existing FM broadcasting infrastructure.
Q : How does WebFM verify the integrity and authenticity of broadcast content?
A : For privacy, the server can obtain a user’s public key during enrollment and encrypt personalized responses so that only the intended user can decrypt them. To prevent content spoofing, WebFM can use Ed25519 digital signatures to verify the authenticity of broadcasts. The signature adds approximately 64 bytes of overhead and reduces the data rate to about 8.6 kbps, but it provides a practical defense against forged content.
Q : Why does WebFM use an audio stream instead of RDS for data transmission?
A : RDS provides only about 730 to 760 bits per second, making the transmission of webpage screenshots extremely slow. WebFM uses OFDM-based audio modulation with multiple parallel subcarriers to transmit more data within a shorter audio interval. This increases the data rate to approximately 10 kbps, making it more suitable for webpages and large language model responses.
Personal thoughts
The most appealing aspect of this paper is its pragmatic design philosophy. Instead of waiting for entirely new connectivity infrastructure, the authors reuse FM radio stations, SMS networks, and commodity Android phones that already exist in many developing regions. The combination of broadcast downlink and SMS uplink is particularly interesting because it supports both shared content delivery and personalized requests. The real deployment in Cameroon also makes the work more convincing than a purely simulation-based evaluation.
However, WebFM is necessarily limited compared with the modern Internet. It mainly supports simplified webpage screenshots, has very limited interactivity, cannot support video or login-protected services, and may provide poor performance under weak or unstable FM signals. Privacy is also an important concern because SMS requests can associate a user’s phone number with requested content. Future work could explore stronger privacy mechanisms, better scheduling for larger populations, adaptive compression based on signal strength, and support for richer interactive content.