Morning Digest — November 17, 2026
AI and wireless infrastructure are converging at every layer—from Q/V-band links and LEO sensing to optical AI fabrics, scientific agents, and tighter compute governance.
China tests a 6.75 Gbps bidirectional Q/V-band link
Chinese researchers completed a Q/V-band wireless experiment that reached a maximum bidirectional transmission rate of 6.75 Gbps on a single channel. The high-frequency band offers wide bandwidth for satellite feeder links and future 6G systems, where capacity is increasingly constrained in lower-frequency spectrum.
The result is a concrete technical follow-on to China’s newly stated goal of timely 6G commercialization by 2030. It also highlights the propagation and hardware challenges that must be solved before Q/V-band links become dependable parts of integrated satellite-terrestrial networks.
Source: Xinhua
Quebecor makes an unsolicited C$31 billion bid for Telus
Quebecor proposed a cash-and-stock combination with Telus valued at C$31 billion. The company argues that a larger Canadian-controlled operator could invest more aggressively in next-generation infrastructure while competing with global technology and satellite connectivity providers.
The transaction would radically reshape Canada’s concentrated telecom market and therefore faces difficult competition and government reviews. Quebecor says it would retain Telus’s brands and Vancouver headquarters and rely more heavily on Canadian satellite networks for resilience.
Source: Quebecor announcement via Canada Newswire
U.S. closes the remote-access loophole for advanced AI chips
The U.S. Commerce Department issued an interim rule requiring licenses for advanced-chip exports when the hardware could be used to provide remote computing capacity to China or other restricted countries. The policy expands enforcement beyond the physical destination of an accelerator to the customers accessing it through overseas data centers.
The change directly affects cloud operators and NVIDIA customers in third countries. It turns identity, tenancy, and workload controls into export-compliance obligations and could reshape where global AI capacity is built and who can rent it.
Source: NPR
Scientists report nearly doubling productivity with ChatGPT
OpenAI analyzed 580,186 scientific messages from 1,819 participants in its ChatGPT for Academic Researchers program. Across eight activities—from idea development and data analysis to coding and writing—participants estimated that one hour with AI available replaced about 1.9 hours of work without it.
Self-reported counterfactual time savings are not the same as independently measured productivity, but the scale and task-level analysis make the study useful evidence about how frontier models are entering real research workflows. OpenAI is also expanding free access and support to as many as 200,000 scientists worldwide.
Source: OpenAI
OpenAI embeds four independent safety assessors
OpenAI signed agreements with Apollo Research, METR, Transluce, and Goodfire for the first phase of an embedded-assessment program. The outside teams will receive deep access to internal models and systems to investigate alignment, safeguards, and gaps in existing evaluations.
The arrangement goes beyond commissioning a benchmark after release: assessors can study systems earlier and recommend new tests while models are still being developed. The key question will be how much independence, publication freedom, and deployment influence the external teams retain.
Source: OpenAI
Gemini 4 Carbon moves into scientific lab workflows
Google DeepMind says researchers are using Gemini 4 Carbon inside its Antigravity environment for multi-step scientific work. Reported examples include mathematically proving security properties in Signal’s cryptographic code, detecting subtle wet-lab errors and data anomalies, and finding 40 times more related protein sequences in large databases.
DeepMind is expanding the Antigravity program to 10,000 scientists with free Gemini 4 Carbon access. The important shift is from conversational assistance toward agents that inspect code, notebooks, raw observations, and scientific databases across a full workflow.
Source: Google DeepMind
Lambda adopts NVIDIA co-packaged optics for AI factories
Lambda is using NVIDIA Photonics co-packaged-optics switches to strengthen the networks connecting large GPU clusters. Moving optical engines next to switch silicon reduces long electrical traces, improving signal integrity while lowering power and latency compared with conventional pluggable optics.
As AI training jobs scale across thousands of accelerators, network failures and energy costs become first-order constraints. NVIDIA positions the technology across both scale-up fabrics inside computing domains and scale-out Ethernet links between systems.
Source: NVIDIA
Cloudflare open-sources its Kitesurf browser for AI agents
Cloudflare released Kitesurf, an experimental stateless browser engine built to run on Workers, under an open-source license. Developers can inspect the code, run it locally, or deploy their own instance while agents render pages and execute JavaScript without starting a full desktop Chromium session.
Purpose-built agent browsers could reduce memory and operational overhead for large fleets of web-automation workers. They also make isolation, reproducibility, and observability easier to design directly into the execution environment.
Source: Kitesurf
Research Radar
Risk-Aware Handover Management Approach for Non-Terrestrial Networks with the NTN-Connect Simulator
Mohammed Almekhlafi, Antoine Lesage-Landry, and Gunes Karabulut Kurt jointly optimize satellite association and transmit power under uncertain user rates, visibility duration, and per-satellite power budgets. The framework uses value-at-risk constraints and a low-complexity assignment method evaluated on OneWeb and Starlink constellations.
On the Starlink simulation, it increased the inter-handover interval from 0.37 to 1.47 minutes, cutting the handover rate by 75%. At a maximum-IHI operating point, it reduced total transmit power by 1.1 dBW, or 22%, while increasing user admission from 97.4% to 99.3%.
Source: arXiv:2611.21027
Early UAV Detection Using 5G NR LEO Downlinks for Passive ISAC
Mohamed Hamad, Nizar Zorba, Tamer Khattab, Hamid Saeedi, and Halim Yanikomeroglu use 5G NR LEO downlinks as passive illumination for UAV sensing. Their receiver selectively searches Doppler-rate hypotheses, issues early alerts, and continues refining delay and Doppler estimates.
In the simulated line-of-sight scenario, all configurations matched exhaustive search at 92.9% correct final detections with observed false-alert rates of 0.10–0.13%. The neural configuration reduced mean time to the first correct alert by 36.5% versus exhaustive search.
Source: arXiv:2611.20859
SiGMA: Site-Grounded Multi-Modal ISAC Data Generation for Environment-Aware Communications
Yijie Bian and collaborators introduce a data-generation pipeline that aligns OpenStreetMap and Overture data, procedural 3D scenes, cameras, LiDAR, positioning, FMCW radar samples, and ray-traced wireless channels around the same site and time-indexed actors. It can also create RIS component channels for configurable reflected propagation.
SiGMA improved zero-shot beam prediction on real DeepSense 6G measurements over an OSM-derived synthetic-data baseline. The work targets a core bottleneck in learning-based ISAC: gathering sensing and channel data that truly describe the same dynamic environment.
Source: arXiv:2611.20665
Source notes
IEEE Xplore and ACM Digital Library returned no newly indexed eligible wireless papers. Research Radar therefore uses verified arXiv records submitted on November 16, 2026.