<-- Last topic: Modulation Next topic: Frames, slots and channels --->

X Basic Digital Principles: Turning speech into digital

Okay, we've finished with the easy material. Now that you've read the introduction, something on digital history, modulation, a bit about standards, it's time to learn the basics: digital basics. Digital allows features analog schemes can't easily provide. Things like encryption, calling number identification, extension phone service and messaging. It's just a matter of adding more 0s and 1s to the data stream running between the mobile and the base station. Analog based systems by comparison can't easily expand.

But aside from more services for customers, the carrier also benefits. Calls use less bandwidth once digitized and compressed, allowing greater capacity in an already cramped radio spectrum. And an all digital wireless system promises complete compatibility with the landline telephone network. Before understanding how digital communications works we must first look at what makes it up.

A.Turning speech into electrical impulses

Speech is sound in motion. Talking produces acoustic pressure. Speaking into the can of a string telephone, for example, makes the line vibrate, causing sound waves to travel from one end of the stretched line to the other. A telephone by comparison, reproduces sound by electrical means. What the Victorians called "talking by lightning." A standard dictionary defines the telephone as "an apparatus for reproducing sound, especially that of the voice, at a great distance, by means of electricity; consisting of transmitting and receiving instruments connected by a line or wire which conveys the electric current."

Electricity works the phone itself: operates the keypad, makes it ring. Electricity provides a path, too, for voice and data to travel over wires. This gets confusing. Electric current doesn't really convey voice; sound merely varies the current. It's these electrical variations, analogs of the acoustic pressure originally spoken into the telephone transmitter or microphone, that represent voice. Get it? To sum up: 1) electrical current operates the telephone and 2) that electric current is varied by the telephone to communicate. More below.

 

The telephone is an electrical instrument. Speaking into the handset's transmitter or microphone makes its diaphragm vibrate. This varies the electric current, causing the receiver's diaphragm to vibrate. This duplicates the original sound. Take a look at this image to make this point much clear.

 

Speaking into an older telephone's transmitter causes the diaphragm, a thin metal sheet, to vibrate, varying the electric current. This up and down current, in turn, causes the receiver's diaphragm to vibrate, reproducing the original sound. Modern phones use electret microphones for transmitters and piezoelectric transducers (external link) for receivers but the principle is the same.

In wireless technology, a coder inside the mobile telephone converts sound to digital impulses on the transmitting side. On the receiving side it converts these impulses back to analog sounds. A coder or vocoder is a speech analyzer and synthesizer in one. Vocoders are in every digital wireless telephone, part of a larger chip set called a digital signal processor. Sound gets modeled and transmitted on one end by the analyzer part of the vocoder. On the receiving end the speech synthesizer part interprets the signal and produces a close match of the original. Keep following along.

Once converted by the telephone to electricity, normal speech, music, or tones, are all analog signals. Their electrical waveforms are like or 'analogous' to the sounds they represent. These sounds vary a telephone circuit's resistance, electrically representing speech with a continuous electromagnetic wave. But along with the good comes the bad. Analog voice transmission amplifies distortion along with the original signal. Like when you make a tape of a tape or a photocopy of a photocopy. Digital systems don't have that problem. Digital signals, for the most part, remain stable for the length of their travel. Why is that?

Digital signals are a mathematical or numerical representation of sound, with each sonic nuance captured as a binary number. Reproducing sound is as easy as reproducing the numbers. Extensive error checking schemes ensure that a wireless digital link stays intact, even when transmitted through the air. Let's see how digital signals are made and then later compressed.

B. Converting electrical impulses to digital signals -- voice coding

Converting sound to digital used to be easy to describe, however, with the newest techniques it's getting tougher. So let's first look at the old fashioned way of digitizing, before we complicate matters.

You've probably seen an analog signal wave. It's a rise and fall pattern, like what you see on an oscilloscope. By plotting its coordinates on graph paper, you know, A-2 , B-4, C-3, and so on, we could record its shape in a numerical or digital form. And the more coordinates we plotted the more accurate the record would become. Well, if we wrote down those plots and gave them to someone else, they could easily redraw the waveform and eventually reproduce the sound. And if we have digital signal processing technology, which we do, our coordinates of A-2, B-4, and C-3 could easily be converted to binary. See what I mean below?

 

The diagram above and the similiar one that follows are conceptual, don't worry about the plus and the minus, any plot, no matter above or below the median, can be converted to binary. The beautiful, stylized sine wave is from Jessica Koeppel's site: http://gratuitous.com/~jessica/

In T-1, the backbone of long distance telephone service, a caller's voice gets measured or sampled 8,000 times a second! That produces a highly accurate speech record, at least enough for landline telephones. In making a CD, by comparison, music gets sampled 44,000 times a second. Get what we mean by sampling? We take a numerical record of sound, with T-1, 8,000 times a second, and with a CD, 44,000 times a second. The more samples the more accurate our record.

As an aside, I find it odd that some audiophiles claim they can hear the difference between a song on a phonograph record and that same song recorded on a CD. How is it possible to distinguish between an analog record and a CD when sampling occurs at 44,000 times a second? Okay, and since I am rambling, how about that phonograph record? It is the perfect analog example: an entire song recorded in a single, long, continuous groove. No stops and starts or sampling like in digital. Even in silent periods the groove continues on, recording. See how the groove sort of resembles an actual sine wave? A record groove thus represents a continuous and ever varying wave. Analog!

 

See how a record grove represents a varying, continuous wave? This is totally different from digital. This graphic was from :http://members.chello.se/christer.hamp/phono/poliak.html

Back to sampling. This first step in digitizing is called pulse amplitude modulation or PAM.Amplitude refers to a signal's strength, the relative rise and fall that PAM takes measurements of. These levels, ranging from 0 to 256 in T-1, are plotted against time. How's that? To have a coordinate like those below you must have two magnitudes. The signal strength and the time it occurred. Once you have those you have a plot that can be put into binary.

 

After PAM takes its measurements, each sample gets converted to an 8 bit binary code word. Let's say one piece of conversation, a fraction of a second's worth, actually, hits a strength level of 175. It's now put into binary, transmitted by turning on or off an electrical current or light wave. Like sending Morse code. The bits 10101111, for example, represent 175. Voltage turned on or off. Since this second step encodes the previous information, it is called pulse code modulation or PCM. That's what the code in PCM stands for.

Putting the measured strength or amplitude into 8 bit code words is also called quantization. A name for both steps is called voice coding. And every code word generated is time stamped so it can be put back together in the order it was made. The result? The bottom line? Old fashioned pulse code modulation needs 64,000 bits (64kbs) every second to represent speech. Better ways exist for wireless. Oh, and make sure you don't confuse the sampling rate with the bit rate we just mentioned. A sampling of 8,000 times a second might result in a 64,000 bit a second signal but it all depends on what follows next.

STOP! Don't rush through. Do you really understand, at least enough to proceed with this article, PAM, PCM, voice coding, and quantization? If not, go back. Take five minutes. You'll learn better.

Does this help you visualize quantization better? It's another kind of waveform coding, different from PCM although similar. This and many other outstanding graphics are at Ericsson's site.

1. Better voice coding: VSLEP

PCM, invented decades ago, isn't efficient for digital wireless working. Radio frequencies are limited in number and size, yet demand for them keeps growing. Data must be sampled and then compressed more effectively to conserve bandwidth. In IS-54, now IS-136, the digital system we will look at later, voice traffic gets coded and compressed at the same time using a technique called VSELP. That stands for, hold your breath, Vector Sum Excited Linear Predictive speech compression. Of course. Voice is compressed down to 7.95 KBits/s, almost one sixth PCM's size. The circuit that does both the initial sampling and compression is called, as we mentioned briefly above, a voice coder, again, part of the digital signal processor or DSP. There's a number of tricks the DSP uses to crunch down speech and conserve bandwidth.

With VSELP, the coder models a speech waveform every 20 milliseconds. That helps immediately, at least compared to T1, which samples every 125 microseconds, piling up a lot of needless bits. And rather than copying the entire sound, VSLEP digitizes the voice's essential elements. It's used with digital sound processing techniques, along with proprietary algorithms owned by the chip maker. If modeling, rather than copying doesn't sound magical enough, hold on. "[I]f a speech segment gets lost over the radio channel, the VSELP decoder (on the receiving end) can 'repair' the effect through speech extrapolation."

In explaining how a GSM mobile encodes speech, Nathan Muller, in the Mobile Telecommunications Factbook, described the related technology called RPE-LPC. He says that "information from previous samples, which does not change very quickly, is used to predict the current sample. The difference between the predicted and actual signal, represent the signal." To put it another way, there's not much change between samples, since each takes place every 20 milliseconds. So, instead of transmitting full full samples each time, the digital signal processor sends only the change between samples. Get it? There's a little more, and then we'll move on.

Many coders support digital speech interpolation or DSI, which gains compression by filling in the gaps during speech pauses. It's said that silence makes up 60% of a conversation, consequently, DSI transmits only during voice spurts. Another active channel then uses the bandwidth during silent periods. Very efficient unless, as David Crowe points out, that speakers don't talk over each other. Don't get overwhelmed by the terminology. Just remember that coders and DSP make up a vital part of any digital wireless system, converting an analog signal to digital and back to analog again.

To wrap this up, let's make totally sure we understand the difference between a digital signal and an analog one. Sampling or quantization takes a lot of measurements. But it is not continuous, even at a hundred or a thousand times a second. There are always small gaps. These breaks, these starts and stops, differ an analog signal from a digital one. A digital signal is made up of discrete units but an analog signal is a continuous unit. Like the record I mentioned, remember?

(I know this animated GIF is annoying but I needed to show a continuous wave)

Digitized speech is a representative model of speech done in near real time. Let's discuss how digital transmission sends information -- inside frames.

<-- Last topic: Modulation Next topic: Frames, slots and channels --->

Text link to CellTowerInfo.com