Back in the early 80s, when was working for Rolm, having left engineering and begun a two-year tour as general manager of Office Systems Division, I used to spend a lot of time speaking at conferences. We had a lot of circuit switching data communications products and the promise of future centralized packet switching products, and I often was scheduled opposite a local area network advocate. Bob Metcalfe, inventor of the Ethernet, who was on the same circuit, used to say that there are two kinds of people in the data communications world: those who begin their presentations by drawing a box, and those who begin by drawing a line. I was a box-drawer, and an advocate of switching. Bob was a line-drawer, and a proponent of distributed contention for a single piece of copper. Switching won the local data communication battle (you’d have a hard time buying an Ethernet hub except at the very lowest end of the market, and the yellow Ethernet coax is just a memory), but the winning technology wasn’t circuit switching and it didn’t do the PBX people any good.
During my talks, I would tweak the data communications people by pointing out that the vast majority of all the data communications in the world was carrying voice. Today, the situation is reversed: voice accounts for but a tiny fraction of all the data slewing around the globe. With the packet switched Internet Protocol dominating data communications, it makes economic sense for voice to try to hitch a ride. Thus is born Voice over Internet Protocol, or VoIP.
The promises of VoIP are drastically reduced voice communications expenses and a wealth of features not practical with circuit-switched voice. However, voice and IP is an awkward marriage. The problem is latency, the delay between your phone the phone of the person you’re talking to, added to the delay of the return trip back to you. If you remember talking over satellite links, a fixture of the 80s international telephone network now almost completely replaced by fiber optics, you know what I’m talking about. In the case of satellite communication, the delay is mostly due to signal propagation at the speed of light from the ground station on your end of the call up to the satellite in a geostationary orbit 22,000-plus miles above the earth down to the ground station on the other end of the call, plus the trip up to the satellite and down coming back. In IP telephony, the latency mostly comes from buffering as the packets wend their way through the Internet. Unless you’re using a satellite connection to get to your ISP, the latency of your Internet connection is a couple of hundred milliseconds, or less than a quarter of typical satellite voice call latency.
Latency causes several problems in telephone calls. The first is just the delay; no matter how good everything else is, having almost a 1 second round-trip time is a recipe for both talkers trying to talk at once. The second is electrical instability caused by echo which can cause effects varying from hearing your own voice delayed to oscillation. The cure for this is sophisticated digital signal processing in the form of echo cancellation. It is effective, but it makes problem one worse, since most of the time it turns the circuit off if both parties are talking simultaneously. In order to avoid the audible ills of satellite calls, most experts suggest that, for successful voiceover IP, latencies be kept to under 200 ms.
Because of the latency, VoIP is, in an engineering sense, an unnatural act. It only makes sense to attempt it because the Internet has become a cheaper way to move bits than the historically-common circuit-switched highways, and because IP telephony offers a host of features not traditionally available.
I thought I’d give VoIP a try. I’ll report on my experiences later.