The Hawaiian Radio Hack That Became Every Wi-Fi Network on Earth
The Hawaiian Radio Hack That Became Every Wi-Fi Network on Earth
.webp)
Somewhere in your home right now, a device is doing something that looks, from the outside, like bad manners. It has data to send. It listens for a fraction of a millisecond. Hearing nothing, it simply transmits — no permission requested, no slot reserved, no guarantee that another device isn't about to do exactly the same thing. If the two transmissions crash into each other, both are destroyed, and both devices wait a random amount of time before trying again.
This is not a bug. It is the founding assumption of modern networking, and it was made deliberately, in 1971, by a professor on a Pacific island who could not afford telephone lines. The idea was considered so undisciplined that engineers spent the next decade arguing it could never be trusted with serious work. It won anyway. Here is how a wireless experiment in Honolulu became the wire in your wall — and then went back to being wireless again.
Four Islands and a Phone Bill Nobody Could Pay
When Norman Abramson left Stanford for the University of Hawaii in 1968, he inherited a geography problem. The university's campuses, community colleges and research stations were scattered across Oahu, Kauai, Maui and Hawaii Island, all within roughly 200 miles of the main Manoa campus, and all separated by water. The only way to reach a central computer from a remote terminal was a leased line or a dial-up connection — and in Hawaii that meant undersea circuits. According to the IEEE milestone record for the project, those connections were consuming the lion's share of the university's computing budget.
Abramson's instinct was that the whole premise was outdated. "The telephone network was something that was set in place at the end of the 19th century, and we were at the end of the 20th century," he later recalled in an oral history quoted in the same record. "We thought that radio communications might be a good way to start on it."
Radio was cheap. Radio was also, at the time, effectively forbidden: the FCC did not permit the commercial use of spectrum for this kind of application, and it rejected the project's initial request outright. The team got an experimental UHF allocation from the military instead, and funding through Project THEMIS, a Department of Defense program built specifically to seed research at universities outside the top tier. It was the largest research grant the University of Hawaii had received to that point.
The Protocol That Sounded Like Giving Up
Every terminal in the network shared one radio channel. Abramson's team spent three or four months, by his account, before landing on the answer: let them all talk at once and clean up afterwards. A terminal transmits its packet at a moment of its own choosing, then waits for an acknowledgement. If none arrives, it assumes another terminal stepped on it, waits a random extra interval — the randomness is essential, or the two collide again forever — and retransmits.
The system went live in June 1971, running two UHF channels at 407.350 MHz and 413.475 MHz, both at 9,600 baud. A machine the team called the Menehune, modelled on the ARPANET's Interface Message Processors, handled packets in and out of the campus mainframe. The first remote terminal sat in Abramson's house, about a mile away. By the end of that year there were four.
In his 1970 paper introducing the design, Abramson calculated what such a channel could actually carry. The answer was 1/(2e) — about 18.4% of the raw channel capacity. Adding synchronised time slots, a refinement analysed by Lawrence Roberts in 1972, doubles it to 1/e, roughly 36.8%. Both numbers look like failures. Against a leased line you cannot afford at all, they were infinite improvements.
The Memo That Put the Ether in a Cable
Two thousand miles away, Xerox's Palo Alto Research Center had the opposite problem: not distance, but density. PARC had built the Alto, a personal computer with a bitmapped display, keyboard and mouse, and it had built laser printers and file servers worth sharing. What it lacked was a way to connect them at speeds a modem could not touch.
On May 22, 1973, Metcalfe circulated an internal memo titled "Ether Acquisition." Its immediate purpose was a rename — he wanted the team to stop calling the project the "ALTO ALOHA Network." Its lasting contribution was the metaphor. He borrowed the luminiferous ether, the all-pervading medium 19th-century physicists had wrongly believed carried light, and applied it to a shared broadcast channel that was deliberately dumb and deliberately passive. Coaxial cable was merely the first ether. The memo explicitly anticipated telephone twisted pair, optical fibre, radio and even power lines taking its place. Every one of those eventually did.
Metcalfe made two changes to Abramson's rule, and both matter. First, carrier sense: listen before you transmit, and stay quiet if the cable is busy. Second, and more subtly, binary exponential backoff — after each successive collision, double the range of the random wait. The ALOHA retry interval was fixed, which meant that under heavy load a network could thrash indefinitely. Doubling the window makes the network back off harder the more congested it gets, and it is the reason Ethernet degrades gracefully instead of locking up. That mechanism is claimed in US Patent 4,063,220.
Metcalfe recruited David Boggs, and by November 1973 the Experimental Ethernet was running: Manchester-encoded baseband signalling at 2.94 Mbit/s — precisely half the Alto's clock rate, because the Alto's clock was the cheapest available source of timing — over a single shared coaxial conductor tapped by puncturing the cable. By 1975 there were 25 nodes on it, linking Altos, Data General Novas, printers, a mainframe and a gateway to the ARPANET.
Winning the Argument It Was Supposed to Lose
The objections were serious and, on their own terms, valid. Ethernet offered only probabilistic delivery. Its latency under load was non-deterministic, precisely because of the randomised backoff. Rival designs passed a circulating token that guaranteed each station its turn, and on paper they behaved far better when the network was busy.
Xerox began opening the design to industry in 1977. In September 1980 the "DIX" consortium — Digital Equipment Corporation, Intel and Xerox — published the 10 Mbit/s Ethernet 1.0 specification, universally known as the Blue Book. Submitted to the new IEEE 802 committee, it did not win outright: the committee published three mutually incompatible standards, Token Bus as 802.4, Token Ring as 802.5, and Ethernet as 802.3. The 802.3 standard was approved on June 24, 1983, and published on December 31, 1984 as ANSI/IEEE Std 802.3-1985.
What settled it was economics. Token Bus never found volume. Token Ring lost to a sequence of unglamorous advantages: multi-port switches, automatic speed negotiation, full duplex, higher speeds — and, decisively, twisted-pair 10BASE-T. After AT&T's 1982 deregulation, building owners frequently ended up owning surplus telephone cabling already run through their walls. Ethernet learned to use it. Xerox had 75 Ethernets at 40 sites serving over 1,350 Altos by 1981; by 1985 an estimated 500,000 Ethernet adapters had shipped worldwide.
| System | Medium | Raw rate | Access method |
|---|---|---|---|
| ALOHAnet (1971) | UHF radio, 407.350 / 413.475 MHz | 9,600 baud | Pure ALOHA random access |
| Experimental Ethernet (1973) | Single shared coaxial cable | 2.94 Mbit/s | CSMA/CD + binary exponential backoff |
| DIX Ethernet 1.0 (1980) | Thick coaxial trunk | 10 Mbit/s | CSMA/CD |
| IEEE 802.11 Wi-Fi | Unlicensed radio spectrum | Varies by generation | CSMA/CA + binary exponential backoff |
The Wire Won, Then Handed It Back to the Air
Here is the twist most retellings skip. The collision detection that defined Ethernet has been effectively dead on wired networks for years. Once hubs gave way to switches and links became full duplex, each cable carried exactly one transmitter in each direction. There is nothing to collide with. The CD in CSMA/CD is vestigial on a modern office LAN.
The ALOHA idea did not die with it — it emigrated back to radio, where the original constraint still holds. Wi-Fi cannot use collision detection because a station transmitting at full power cannot simultaneously hear a distant, weaker signal on the same channel, so 802.11 substitutes collision avoidance: listen, wait a random backoff, transmit, wait for an acknowledgement, and double the backoff window on failure. That is Abramson's rule with Metcalfe's stabiliser bolted on. Because stations may be able to reach the access point but not each other, Wi-Fi adds the RTS/CTS handshake to cover the hidden-node case — a problem that simply does not exist on a shared cable.
The pattern repeats wherever devices are many, cheap and mostly silent. LoRaWAN, the long-range low-power standard behind a great deal of industrial and municipal sensing, uses pure ALOHA essentially unmodified — which is why researchers keep publishing slotted overlays to claw back throughput as deployments densify. And on a mobile network, the very first message a handset sends when it wakes up and asks for a channel goes out on a contention-based random-access channel. Every phone call, every message, every session begins with a small, deliberate collision risk.
Abramson died in December 2020, having received the IEEE Alexander Graham Bell Medal in 2007 for his work on random multiple access. David Boggs died in February 2022, aged 71. Metcalfe received the 2022 ACM A.M. Turing Award for the invention, standardisation and commercialisation of Ethernet.
Almost every account of this story treats it as a lineage of inventions: radio begat coax, coax begat twisted pair, twisted pair begat Wi-Fi. We think that framing buries the actual transferable idea, which is not a technology at all but a decision about governance. Abramson and Metcalfe chose statistical arbitration over deterministic scheduling — they accepted failure as a normal operating mode and agreed to pay for it in retries rather than in coordination overhead. Token Ring genuinely was better behaved under load, and it lost, because it required every station to agree on something before anyone could speak. Ethernet required agreement on nothing. That trade keeps recurring far outside networking: optimistic concurrency beating row locks, eventual consistency beating distributed transactions, retry-and-backoff beating orchestration in nearly every large distributed system built since. The 1973 memo reads today less like a hardware proposal than an argument that when coordination costs more than repetition, the correct engineering answer is to let things collide and clean up afterwards.
What we still don't know: whether that trade holds at the scale now being built. Time-Sensitive Networking on the wired side and scheduled access in cellular are deliberately reintroducing the coordination Metcalfe removed, because some workloads cannot tolerate a random wait. Nobody has yet demonstrated that "let it collide" scales to tens of billions of low-power radios sharing unlicensed spectrum — and the LoRaWAN scalability literature suggests it may not.
The 18th-Century "AI" That Fooled the World for 84 Years — on how convincingly a clever mechanism can be mistaken for something else entirely.
Physics, Not Capital: The Week AI Ran Into the Power Wall — a modern reminder that shared resources have hard ceilings.
Sources
- Robert Metcalfe — "Ether Acquisition," Xerox PARC internal memo, May 22, 1973 (primary)
- Norman Abramson — "THE ALOHA SYSTEM: Another Alternative for Computer Communications," AFIPS Fall Joint Computer Conference, 1970 (primary)
- Robert Metcalfe and David Boggs — "Ethernet: Distributed Packet Switching for Local Computer Networks," Communications of the ACM, 1976 (primary)
- DEC, Intel and Xerox — The Ethernet 1.0 "Blue Book" specification, September 1980 (primary)
- ACM — "ACM Turing Award Honors Bob Metcalfe for Ethernet," March 2023 (primary)
- IEEE Engineering and Technology History Wiki — Milestone: Ethernet Local Area Network (LAN), 1973-1985
- IEEE Engineering and Technology History Wiki — Milestone: Demonstration of the ALOHA Packet Radio Data Network, 1971
- IEEE Spectrum — "ALOHAnet Introduced Random Access Protocols to the Computing World"
- Sensors (MDPI) — "Slotted ALOHA on LoRaWAN: Design, Analysis, and Deployment"
Historical figures and specifications verified September 5, 2026 against the primary sources listed above. GadgetGlow Bytes does not test vintage hardware; all technical details here are drawn from published documentation and archival records. GadgetGlow Bytes does not receive products from manufacturers for coverage.
.webp)
Comments
Post a Comment