Brands
Latest top stories
Technology

InterDigital tests AI to stop video freezing at the edge of mobile coverage

19 November 2026

InterDigital is testing how AI could keep mobile feedback usable in weak coverage and help phones and base stations from different suppliers interpret it. Photograph by Mike van den Bos on Unsplash.

 

A video call can freeze even when a phone still shows several bars of signal. Often, the culprit lies not in the signal reaching the handset, but in the return journey.

To keep a connection stable, a smartphone constantly sends back a brief status report describing how the base station’s signal is arriving across different frequencies. The network uses this feedback, known as channel state information (CSI), to aim and shape its outgoing signal. When a user drifts toward the edge of a mast’s coverage and that return report degrades, the network loses its bearings and the video breaks up.

At the Brooklyn AI-Native 6G Summit earlier this month, the US research firm InterDigital unveiled two early-stage prototypes using artificial intelligence to shore up that fragile exchange. Organised by Nokia and NYU Wireless from 3 to 5 November, the event showcased pre-trained models tested under controlled conditions.

Doug Castor, InterDigital’s vice-president and head of wireless research, outlined the work in written responses to MoveTheNeedle.news on behalf of the engineering teams behind the demonstrations.

 

Keeping feedback alive as the signal fades

 

In the first project, researchers from InterDigital, the electronics firm Keysight and the University of Texas at Austin used AI to collapse three conventional transmission steps into a single learned process. Their system compresses the handset’s feedback report, shields it against errors and converts it into a radio signal all at once. Crucially, the method, called joint source and channel coding and modulation (JSCM) , requires no extra airtime, spectrum or battery-draining transmit power.

Castor described how the feedback changes as the signal fades. “As reception worsens, the report can lose detail gradually, instead of failing abruptly. At very weak signal levels, both JSCM and the conventional approaches eventually fail.”

In the Brooklyn demonstration, physical radios ran the AI software while test equipment simulated fading signal conditions. Researchers compared live video delivered using a conventional 5G precoder, which shapes the outgoing signal,  with video delivered using a precoder guided by the AI-based feedback. While the traditional video became badly distorted, the AI-assisted stream kept playing with only occasional glitches. In simulations, the system matched conventional data-loss rates using a return signal roughly a tenth as strong.

 

Getting rival networks to speak the same language

 

Even when a feedback report reaches the base station intact, the network still has to decipher it. If one AI model compresses the measurements on a handset and a different supplier’s model reconstructs them at the mast, the two systems can end up encoding identical radio conditions in mutually incomprehensible ways.

Working with Arizona State University, InterDigital tackled this by applying a shared training principle that keeps the models’ internal representations structurally aligned. A small batch of calibration samples is then used to learn a lightweight translation layer between one vendor’s encoder and another’s decoder.

The approach avoids the need for a single industry-wide reference model. As Castor put it, “With our research, vendors could train independently without sharing their proprietary models or large training datasets.”

Unlike the simulated radio conditions of the first trial, this prototype ran live over the air with a moving test device and real-time processing. In controlled indoor tests, the aligned models achieved roughly 93% of the download data rates managed by models trained jointly from scratch; a marked improvement over running separately trained models without translation.

 

From laboratory prototype to global standard

 

The research sits at the heart of InterDigital’s business model. Founded in 1972 and listed on Nasdaq, the company develops wireless and video technologies and licenses its patents to device makers and network operators. Its smartphone licensing programme alone brought in roughly $589m of its $834m total revenue in 2025.

For a patent licensor, dictating how handsets and masts exchange data is as much a commercial question as an engineering one. The International Telecommunication Union is currently drafting the global framework for 6G, known as IMT-2030, while the standards body 3GPP is studying how to embed AI into future networks.

Castor noted that rolling out the techniques would largely rely on software updates to phones and base stations, alongside agreed rules for exchanging feedback. Yet hardware demands remain open, as does the question of whether the features could arrive via upcoming 5G-Advanced upgrades or will have to wait for 6G.

For now, Castor offered no commercial timetable. InterDigital’s next step is to test how both methods hold up outside the lab, measuring their drain on processing power and battery life, and seeing how real-world hardware imperfections affect the signals as users move between good and patchy coverage.

 

 

 

Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.

👉 https://buymeacoffee.com/movetheneedle.news