The Story of the Broadband Phone

This is a technical history of AT&T Labs Cambridge's Broadband Phone project, which seemed to garner a disproportionate amount of publicity during its short life. I'll try to show here that it wasn't all hype.


The Broadband Phone grew out of our hugely successful VNC software for remote access to graphical user interfaces, as well as earlier work on networked media endpoints.

The first VNC client was an embedded flat-screen device called a VideoTile and, from the outset, the VNC team were keen to develop embedded VNC servers (so that faceless equipment could export a GUI), and embedded viewers (so that simple terminals could display complex, arbitrary graphics). They believed that a stateless terminal architecture was the best way to design large-scale multimedia systems, especially as bandwidth was becoming cheaper.

One idea they came up with was a telephone, with a touchscreen in place of the keypad. The screen could display directory information from a server, perhaps with pictures of subscribers. Touching someone's picture would place a call (this concept was dubbed dial-by-face). When the lab was acquired by AT&T in January 1999, Sandy Fraser (then Chief Scientist at AT&T Labs) was impressed by the idea and suggested we develop it further.

At that time, AT&T intended to expand into broadband access and revamp their network as a unified IP-based system, so the name Broadband Phone seemed appropriate.

Early prototypes

[Libretto phone] [Prototype]

The first mock-up of a Broadband Phone was made in early 1999 from a Toshiba Libretto palmtop PC, broken back on itself and glued into an oversize phone case. It ran a VNC viewer; demo graphics were provided by Tcl/Tk on an Xvnc server. It wasn't a real phone (the handset was dead!) but nobody seemed to notice.

The next prototype was fully working, made from a PC104 card, sound card, Flash card adaptor, home-made ringer, LCD and touchscreen, all stuffed into a modified phone case. It booted Linux from the Flash, and ran VNC viewer and proprietary voice-over-IP software. Four of these were built, and widely demoed; such as to an invited gathering of industry leaders including Bill Gates and Steve Jobs in July 1999.

A revised prototype used a custom PCB to eliminate most of the internal spaghetti. 10 were made, and installed as permament demos at AT&T sites in the US.

Having no buttons, just a touchscreen, the phone's user interface was completely 'soft' and could be reconfigured (e.g. by a hotel chain, PBX owner or service provider); and because all applications ran on the server, any changes or upgrades could be made centrally.

The Broadband Phone is born

[Broadband phone] [Translucent orange phone] [Scribble application]

For the trial version of the phone, we contracted professional designers. The electronics were designed by OptionExist, based around a 200MHz StrongARM processor, again running Linux. The elegant slimline case was designed by Dickinson Associates (Rick Dickinson is better known for designing the Sinclair ZX81). Of the final model, 250 were made.

On the server, we decided that Tcl/Tk/Xvnc was not the best platform for large-scale development. We developed a thin server library in C++, which generated native VNC graphics without the need for a server-side frame buffer. We had fun writing applications and demos in this environment. Dial-by-face became a reality, integrated with our Active Bat location system.

Another landmark application was Shared Scribble: when a phone call was established, the parties could doodle together on the touchscreen (like PC whiteboard tools, but much simpler to use). Other two-party apps included a piano/stylophone and a chess game.

MPEG and motion-JPEG video decoders and more sophisticated audio were implemented on the device, to support music and video based applications. However, the Broadband Phone was never a videophone - it lacked a camera.

The Broadband Phone was demonstrated in June 2000 at a Royal Society public exhibition in London and Edinburgh; everybody who saw it liked it. In July it was exhibited again, to Cambridge University engineering alumni. Prince Philip was impressed by the uncluttered user interface.

By August 2000 we had a working 70-phone system, with a phone on everyone's desk in our Cambridge lab, plus some broadband VPN outposts. Further trials were planned.

The Broadband Phone comes of age

At the end of 2000, we expected the Broadband Phone project to spin out as a startup company (a route that has worked very well for ORL innovations in the past). Development continued apace, new applications (some more useful than others) appearing frequently. Server software was reorganised, and security features added to the protocol.

[iPAQ (spuriously rebranded)]

Our endpoint software was ported to run on iPAQ handheld PCs and generic Linux and Windows platforms. The iPAQ client, with wireless LAN, was particularly neat.

Telephony was converted to use SIP, with signalling implemented on the server (the device remained a "thin client"). A SIP stack and proxy were written from scratch. Outside calls were gatewayed by a Cisco router, via some extraordinary hardware and software hackery to keep everything on speaking terms. We encountered a steep learning curve, finding out all about SIP, telephony and ISDN, about which we had originally known nothing.

End of the project

2001 was a bad year to be seeking venture capital for a computer/telephony startup, but some progress was made. Then around July, the spin-out was abandoned.

In November, AT&T announced that they intended to sell or close the Cambridge laboratory to cut costs. A sale could not be arranged, and the lab closed in April 2002.

Picking up the pieces

Could the Broadband Phone have succeeded in the real world? Probably, though it would have needed some refinement...

Whilst we could have relied on a soft phone, we felt it was important for reliability and usability, to have a dedicated terminal. But our device proved expensive to make, largely due to its gorgeous colour TFT screen. It had some minor problems which limited the audio quality: rather unfortunate for a telephone! We also discovered that Linux was not the best operating system for low-latency duplex audio. Further work would also have been needed to make our server truly scalable, to support the thousands of devices we had envisaged.

Nevertheless, the principles of the Broadband Phone sytem were ideas whose time had come. Broadband access and VoIP are (slowly) replacing analogue lines; The use of SIP, H.323 and thin-client IP telephones in businesses (as a PBX replacement) is mushrooming; Handheld devices with UMTS or Bluetooth are opening up new avenues for wireless telephony with graphics, and a thin client approach could make sense here; Cable set-top boxes have similar architectures to our system; and thin client devices are now becoming available for domestic music and video streaming.

Nick Hollinghurst
1 May 2002