Printer Friendly

Digital Voice and Multiplexing.

The integration of voice into data networks brings together two technologies that have in the past been treated separately: data communications and telephony.

To better understand the t,echnical terms and how the two work together, it is best to start with the older of the two, telephony, to see how they are really based on the same facilities.

In the beginning, Mr. Bell's invention was strictly an analog device (based on continuously variable signal strength). As offered to the public, it is operated on a direct current that flowed in a single continuous loop through the instruments on both ends of the conversation. The local loop at each end, then, was indeed a loop, formed from the two wires in the famous twisted pair.

Originally the entire phone system, which was limited to local calls, was based on continuous copper wire--one pair for each customer. As phones became more common, the number of wires increased dramatically. Even the connections between central offices were solid copper wire, and there had to be a separate circuit for everyone calling at one time.

Pictures from the 1880s of practically any US city show the result: poles draped in hundreds of wires on every street. Early in this century the practical limit was reached. There was no more room for wires.

Frequency-division multiplexing (FDM) provided the initial solution to the problem. The phone company found that a twisted pair could carry far more than the limited range of frequencies of a single voice. With periodic amplification, a common circuit could transmit more than 100 kHz. In a series of trade-offs--mainly bandwidth versus distance between repeaters--the standards were set to work with then-current technology: one voice channel at 4 kHz maximum, or a multiplexing factor of 24 to 1.

The resulting 96-kHz analog signal was carried on copper wires and had to be amplified periodically. It is the nature of analog circuits to pick up noise. The signal fades between amplifiers but, unfortunately, the noise is fairly constant throughout the circuit. Thus each amplification of the signal also amplifies the noise and increases the noise in proportion to the voice signal.

Eventually, enough amplifications will make the noise greater than the original voice signal, which will be hard to understand.

As analog signals, each voice conversation has to be switched individually. That is, if multiplexed, it must be demultiplexed to be switched, then possibly multiplexed again for transmission to the next central office.

The development of digital signal processing promised distinct advantages. Channels could be switched in the multiplexed form, without demultiplexing the entire group of channels. Control and billing information are handled much more easily by digital computers than by their analog counterparts. But most importantly to the end user, the quality of digital signals holds up through repated amplification.

The nature of the digital signal is that it has discrete levels: 0, [plus]1, [minus]1, or some small number of fixed values. The differences between these levels is set to be much greater than the anticipated common noise on the circuit (noise doesn't know if a wire is carrying analog or digital signals, so digital circuits pick up noise too).

The advantage of digital in regard to noise arises because digital signals are not, strictly speaking, amplified; they are regenerated. The digital repeater looks at the incoming signal, which is the original signal diminished by distance plus noise. This signal most probably won't be at one of the discrete levels (such as, [plus]1), but by comparing the signal to pre-set thresholds, the repeater will decide which level it should be. That is the level sent on from the output side of the repeater.

This process therefore eliminates the effect of moderate amounts of noise. Only if the noise is so large that the repeater guesses the wrong level will an error be propagated.

The advantages of digital circuits prompted the phone companies to make them the new standard for most applications, starting about two decades ago. Again, based on the then-current technology, the standard set was the same as for FDM: one voice channel at 4 kHz maximum, or a multiplexing factor of 24 to 1.

But in this case, the analog voice-channel input had to be converted to digital form, a job performed by a codec (coder-decoder), which has the same generic function as a modern (modulatsor-demodulator), but in the other direction.

The 4-kHz input bandwidth is converted to a 64-kb/s data stream. "how" is buried in the silicon; the "why" is of more interest, because it clarifies other concepts.

The basic step in digital voice is analog-to-digital conversion, done by the codec. The latest codecs are on a single integrated circuit chip. They sample the voltage on the analog side, compare it to a series of fixed voltage steps established as internal standards, decide which step is closest to the input value, then put out a digital word to identify the height of that step.

The number of steps (expressed as the number of bits in the digital word; for example, 4 bits, 10 bits) is important because it determines the "quality" of the analog/digital conversion. More step means less distortion, and the human ear is usually sensitive to distortion--it "sounds bad." To get acceptable voice quality, there should be 12 bits from the conversion.

The frequency of sampling is another consideration. It turns out that a researcher named Nyquist proved that the codec should sample at least twice as fast as the top frequency of the analog signal. The digital words can then be converted back into the original analog signal.

Sampling less often introduces aliasing, a condition where the digital words could collectively represent more than one analog frequency. Aliasing is avoided by filtering the analog input to eliminate high frequencies--for voice, the cutoff is 4 kHz. Therefore: Maximum voice
 frequency 4 kHz
Nyquist multiplier x2
Sampling frequency 8000
Bits per sample x12


Digital voice bit rate 96,000 b/s

The 96-kb/s rate is needed for voice because the 12 bits worth of steps are needed. There must be enough dynamic range, and, at the same time, a "fineness" in the range of normal voice loudness to minimize distortion.

However, the fineness" isn't needed at either the very loudest or quietest ends of the spectrum. The clever phone people decided they could ignore many of the steps in the codec, as long as they kept the fineness where it counted--in the normal range.

By using a coding table (mu-law in the US, A-law in the rest of the world), it is possible to represent the 12-bit range with only 8 bits, concentrating most of the steps in the normal range. Now the factors become; sampling frequency (8 kHz) time bits per sample (8) equals a bit rate of 64 kb/s.

This coded or compressed version of digitized voice is known universally as PCM, for pulse-code modulation. It is the world standard.

Multiplexing 24 channels of 64 kb/s gives 1,536,000 b/s. To this the phone company added an allowance of 8,000 b/s for control and synchronization, for a total of 1.544 million b/s. This is T-1 in the US.

Elsewhere in the world, 30 voice channels, one synchronization channel, and one signaling channel are multiplexed for a total of 32 at 64 kb/s, or a T-1 rate of 2.048 Mb/s.

The device, consisting of codecs and a time-division multiplexer, that merges voice channels into a T-1 facility is known as a channel bank. Channel banks work only with PCM--THAT is, 24 channels of 64 kb/s each.

PCM is more than 20 years old. The compression technique of ignoring some of the step saves a third of the original bandwidth, but PCM still describes each step fully.

That approach was probably good for early equipment that might lose track of the signal unless brought back to the proper step with each word. More modern circuitry, however, can operate without that redundancy.

The technique Timeplex uses in the new voice option for its Link/1 multiplexer, delta modulation, transmits only the direction of change in the analog input. This means that only essential information is sent, without redundancy.

Delta modulation compares the input analog voltage with the "reference" voltage. If the input is greater than the reference, a "1" is sent and the reference is increased a step. If the input is less than the reference, a "0" is sent and the reference is reduced a step.

In CVSD, the data word is one bit. To maintain "toll-quality voice" that users are accustomed to, Timeplex samples 32,000 times per second. The bit rate is therefore 32 kb/s, or half the PCM standard.

The sampling rate can be anything with CVSD; voice is recognizable at 16 kb/s, and still understandable at 9600 b/s.

An emerging technique is adaptive differential pulse-code modulation. Rather than send 8 bits to describe the step fully, ADPCM sends 4 bits to describe the change from the last sample.

The sampling rate is the same, 8 kb/s, so the data stream is 32 kb/s, or half of PCM. At present, there is no world standard for ADPCM encoding, but one is expected.

The only other voice-compression technique of commercial interests is linear predictive coding. LPC is very high cost, always sounds synthetic (robot voice), but can work at data rates as low as 2400 b/s. It is used only where bandwidth is extremely expensive.
COPYRIGHT 1984 Nelson Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1984 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Flanagan, W.
Publication:Communications News
Date:Mar 1, 1984
Words:1592
Previous Article:Viewpoint: Local Officials Risk Loss of Authority Over Cable Television.
Next Article:Concerns Over Data Security Stimulate Countermeasures.
Topics:


Related Articles
How Digital Data Transmission Technology Is Determining the Direction of the Future.
F-T1: buy only what you need.
Corvette dealer races to savings on multi-use network.
Integrated data/voice saves Sager thousands.
OMNIA PIONEERS NEW ACCESS SOLUTIONS WITH NEXT-GENERATION ADM.
CISCO'S OPTICAL INTERNETWORKING STRATEGY SELECTED BY CHINA.
CISCO 12000 GSR SELECTED BY CHINA'S SERVICE PROVIDERS.
Efficient cross-connections.
ARRIS Q5 DMTS COUPLES ARRIS QAM AND CMTS TECHNOLOGIES.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters