In the most recent article about SIP Trunking, we mentioned that service providers are now starting to provide packet-based services for intercompany communications for voice and other multimedia communications. As such, we may be saying a semi-final farewell to the DS-0, a – or even the – fundamental building block for telecommunications networks for about 50 years.
Thus it seems appropriate to take a look back at how this particular 64Kbps “quantum” came into existence. And a lot of the credit goes to Harry Nyquist and Claude Shannon and their “sampling theorem.” Put simply, the theory states that in order to reproduce an analog signal accurately, the signal must be sampled at least twice as often as the highest frequency. (For more information about Shannon, we highly recommend the recent “Science Friday” feature “James Gleick On The History Of Information.”)
So here’s the story of the 64Kbps DS-0. Long ago, we had analog telephony. And “telephone quality” sound was bandwidth limited to roughly 300 to 3,300 hz. This was deemed to be good enough to recognize the caller while not taking too much bandwidth, thereby enabling the service providers to use frequency division multiplexing to aggregate multiple calls onto a higher frequency transmission facility.
Being good engineers, the folks at Bell Labs decided to give some margins (going to 4,000 kz) and use a sampling rate of 8,000 samples per second to replicate a single analog phone call.
But then there was the question of how many bits per sample. In reality, if all of the samples of the amplitude of the signal were evenly spaced, it took about 12 bits per sample to represent a “good” phone call. (This would have ended up with a fundamental rate of 96Kbps.) However, it turns out that more samples are needed at lower amplitudes than as higher amplitudes, so it was decided that using a technique called “companding” (compression and expanding), 8 bits per sample would be adequate.
Simple arithmetic. 8,000 eight-bit samples per second resulted in 64Kbps.
This building block, which now seems incredibly slow, became the fundamental building block for DS-1/T1/E1, DS-3/T3/E3, and other transmission speeds that for years were used for both voice and data communications.
So the irony is that a “T1” and “T3” circuits that formed the basis for so many data networks had nothing at all to do with optimal data processing speeds. Rather, they were the simple derivative of combining multiple 64Kbps DS-0 circuits, that happened to be 64Kbps because of the bandwidth of analog circuits being 300 to 3,300 hz.
Postscript. Back in the days when music buffs cared about things like frequency response, an audio system was deemed to be pretty good if it handled frequencies from 20 hz to 20,000 hz – which is the normal range of human hearing. So, with the sample rate having to be twice the highest frequency, we get to 40k samples per second. And the “standard” for CD-quality is 44Kbps. Conicidence? We think not. (By the way, the CD uses an encoding algorithm that’s basically one bit per sample.)
