Industries / Technology / Telecommunications
Mobile networks
Mobile networks carry calls and data by radio between phones and base stations (antennas on masts and rooftops), and then through a core network, the computers that check who each user is and route their calls and data.
7
Milestones achieved, from 1946 to 2024.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1946
A phone in the car
On 17 June 1946 a driver in St. Louis made a call from a handset under his car's dashboard, over a system run by Southwestern Bell. ETHW, the IEEE History Center's wiki, calls it the first mobile telephone call. ETHW adds that a better scheme, cellular service, was conceived at Bell Labs in 1947 by Donald H. Ring but could not yet be put into practice.
Credit
AT&T and Southwestern Bell; a team including Alton Dickieson and D. Mitchell of AT&T, with H. I. Romnes of Western Electric
1979
Cellular networks open
A cellular network splits an area into small cells, each with its own base station, so the same radio channels can be used again in cells far enough apart and many more people can call at once. The ITU, the United Nations agency for telecommunications, says commercial networks opened in Japan in 1979 and in the Nordic countries in 1981; an IEEE Communications Society history names the Japanese system as NTT's, first used in Tokyo, and the Nordic one as NMT. ETHW calls the system that AT&T and Illinois Bell opened in Chicago in October 1983 "the first modern cellular system," so the sources count "first" differently.
Credit
NTT; the Nordic NMT operators; AT&T and Illinois Bell
1991
Digital calls with GSM
GSM is a digital standard: the phone turns speech into numbers before sending it. ETSI, Europe's telecom standards body, records that a GSM Memorandum of Understanding was signed in 1987, that the first GSM specifications were ready in 1988, and that the GSM committee moved to ETSI in 1989. On 1 July 1991 Radiolinja launched GSM service in Finland, and former prime minister Harri Holkeri called Tampere from Helsinki on a Nokia phone. Ericsson's history notes that Mannesmann in Germany told its ministry it was ready for GSM traffic the same day, and says who was first "depends on how the question is defined."
Credit
ETSI and the GSM signatories; Radiolinja; Mannesmann
2001
Third generation in Tokyo
NTT DoCoMo planned the full launch of FOMA, a 3G service on the W-CDMA radio standard, for 30 May 2001 and postponed it to 1 October 2001, ITU News reported. RCR Wireless reported that the service started on 1 October in the Tokyo area only, which DoCoMo called the world's first 3G mobile service. It had 10,400 sign-ups by 17 October, against a first-year target of 150,000, and early users met dropped calls, failed handovers between base stations and short handset battery life.
Lessonthe first network on a new standard can slip by months and still reach customers with faults they notice.
Credit
NTT DoCoMo
2009
4G with LTE
LTE (Long Term Evolution) is the 4G standard of 3GPP, a partnership formed in 1998, with ETSI among its founders, that now unites seven standards bodies. On 14 December 2009 TeliaSonera opened LTE service in central Stockholm and Oslo, saying that "as the first operator in the world, TeliaSonera launches 4G services commercially," AFP reported. The service carried only data at first, and the company said 4G phones would probably not arrive before 2011. 3GPP says that with LTE and 5G it became "the focal point for the vast majority of mobile systems beyond 3G."
Credit
TeliaSonera; 3GPP
2019
5G service begins
On 3 April 2019 South Korea's three carriers, SK Telecom, KT and LG Uplus, launched 5G, and AFP reported that South Korea became the first country with nationwide 5G service. Hours later Verizon began commercial 5G service in parts of Chicago and Minneapolis. Samsung's Galaxy S10 5G went on sale in South Korea as, in AFP's words, "the world's first available smartphone" with built-in 5G.
Credit
SK Telecom, KT and LG Uplus; Verizon; Ericsson, a supplier to SK Telecom and KT
2024
Shutting 3G strands phones
Operators close old networks as they move customers to new ones. In May 2024 Telstra put off closing its 3G network in Australia from 30 June to 31 August, the ABC reported, because some 4G phones are set to switch to 3G for calls to 000, Australia's emergency number, and their owners may not find out until an emergency. Communications Minister Michelle Rowland said the number of people at risk, which the government had earlier put at a million across Telstra, Optus and TPG, had fallen to about 400,000.
Lessonbefore a network closes, every device that still relies on it has to be found, including devices that look modern.
Credit
Telstra (operator); ABC News (report)
Watch for
AI computers in base stations
AI-RAN means running the radio access network, the base stations and their software, on graphics chips that can also run AI models. NVIDIA said on 28 October 2025 that it would invest $1 billion in Nokia, that the two would build "commercial-grade AI-RAN products" on NVIDIA's new ARC-Pro computer, and that T-Mobile would begin field trials in 2026. The announcement gives no measured gains.
Achieved whenT-Mobile or another outside party publishes field results for speed and energy use on live customer traffic.
Claimed by
NVIDIA and Nokia, October 2025
A first 6G standard
3GPP has concluded that 6G needs two releases of its specifications: Release 20 for studies and Release 21 for the standard itself. It notes that the ITU wants technology proposals for IMT-2030, the ITU's name for 6G, by early 2029, and full specifications by mid-2030 at the latest.
Achieved when3GPP completes Release 21 and the ITU approves it as an IMT-2030 technology.
Claimed by
3GPP, Release 20 page, updated after its March 2026 meetings
Slices that hold in crowds
A network slice is a share of a 5G network set aside, with its own performance targets, for one kind of traffic such as card payments. Ericsson and SoftBank report that SoftBank ran five slices at the 2026 Formula 1 Japanese Grand Prix at Suzuka over three days, and that "even during periods of congestion, there were no reports of service unavailability." They give speed gains over 2025, about 4 times for downloads on 5G standalone, but no latency figures.
Achieved whensomeone outside the two companies measures slice delay and failures at a crowded event and publishes the numbers.
Claimed by
Ericsson and SoftBank, Ericsson Mobility Report, June 2026
Open
01
Prove a 5G core's login code
Machine-checked proofs that the code an open-source 5G core uses to let a phone join the network does what 3GPP's security standard requires. The target is 5G-AKA (authentication and key agreement), the exchange in which the phone and the network prove who they are and agree on encryption keys, as written in Open5GS, an open-source 5G core in the C language. The standard is 3GPP TS 33.501, "Security architecture and procedures for 5G System."
The briefCredit · unclaimed02
An outside slice scorecard
A public scorecard of how well paid network slices hold up in crowds, measured from outside the operator: two identical phones side by side at a crowded event, one on a slice and one on regular service, each sending small timed messages every second.
The briefCredit · unclaimed
Open 01 · the brief
Prove a 5G core's login code
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Machine-checked proofs that the code an open-source 5G core uses to let a phone join the network does what 3GPP's security standard requires. The target is 5G-AKA (authentication and key agreement), the exchange in which the phone and the network prove who they are and agree on encryption keys, as written in Open5GS, an open-source 5G core in the C language. The standard is 3GPP TS 33.501, "Security architecture and procedures for 5G System."
Why it is still open
A 2024 study of seven open-source cellular cores, Open5GS among them, says relatively little research has looked at the security of cellular cores. Its automated code checks found predictable temporary user IDs and TLS certificate checks done wrongly, each in several cores. Checks like these find known bad patterns; they do not prove the code correct.
Built with what exists
The Open5GS source on GitHub (AGPL-3.0 license), including its AUSF and UDM network functions, which handle authentication; CBMC or Frama-C, open-source tools that prove properties of C code; the Tamarin prover, an open-source tool for proving security protocols; an AI coding agent to draft proof harnesses for a person to check (see the Notes "Proven in four weeks" and "Findings that cite notes"). AI agents that help a few people write proofs about real code are what bring this within reach of a small team.
First step
Build Open5GS, find the function in its AUSF that compares the phone's response with the expected one, and get CBMC to prove one property of it.
Done looks like
Proofs of at least five properties taken from TS 33.501, such as "the network accepts a phone only when its response matches the expected response," that run automatically on every new Open5GS release, with every counterexample reported to the maintainers and listed publicly.
Open 02 · the brief
An outside slice scorecard
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public scorecard of how well paid network slices hold up in crowds, measured from outside the operator: two identical phones side by side at a crowded event, one on a slice and one on regular service, each sending small timed messages every second.
Why it is still open
The crowd results published so far come from the operator and its supplier. In February 2026 Ookla, the speed-test company, and Ericsson demonstrated a method for measuring a single slice through Ookla's Speedtest app, but named no operator and published no results. The closest outside work, a 2026 Hungarian conference paper, sets out a method for measuring slices in high-traffic places and only briefly summarizes on-site results. A May 2026 per-slice delay dataset comes from a lab network built on Open5GS, not a commercial one.
Built with what exists
Two phones with slice access, from an operator that sells it or lends test SIM cards; iperf3 and ping run on Android through Termux; the Hungarian team's method; the arXiv team's open5gs-slice-measurement scripts with UERANSIM, an open-source simulator of phones and base stations, to rehearse in a lab without transmitting; Python for analysis.
First step
Run the measurement scripts on a simulated three-slice Open5GS network, then ask an operator that advertises slicing for two test SIM cards.
Done looks like
Measurements at 10 crowded events on at least two operators, giving the median and 99th-percentile delay and the share of lost messages for slice and non-slice phones, with all raw data published under an open license.