IELTS Academic Reading Practice Test 205

For most of the twentieth century the telephone network was designed to do one thing: carry the sound of the human voice. A modem exists to make that network carry something it was never intended to carry. The word is a contraction of 'modulator' and 'demodulator', and it describes the two halves of the device's job. Modulation turns a stream of digital bits into a pattern of tones that can travel down an ordinary telephone line. Demodulation does the reverse at the other end, turning the tones back into bits a computer can read.

The modem was not invented for the internet, or for personal computers, but for the Cold War. In the 1950s the United States Air Force began building SAGE, a vast air-defence system that would link radar stations across the continent to a small number of very large computers. The sites were hundreds of miles apart, and engineers at Bell Telephone Laboratories, the research arm of the American telephone monopoly, were asked to send digital information between them over existing lines. Their answer was the Bell 101 of 1958, which transmitted at 110 bits per second — slow today, but adequate for the short messages the system required.

The first modem to be widely manufactured was the Bell 103 of 1962, which ran at 300 bits per second. Its method was frequency-shift keying. One frequency represented a binary one and another a binary zero, with a separate pair of frequencies for each direction of travel so that both ends could transmit at once — a property engineers call full duplex. Frequency-shift keying is robust against noise and simple to build, but it is also wasteful, using a large slice of the available bandwidth to convey a single bit. Speeds based on it could not rise much beyond a few hundred bits per second.

In these early years the telephone company owned the network and controlled what could be attached to it, so modems were leased rather than bought. The equipment also had to cope with a telephone wired to the wall. The solution was the acoustic coupler, a cradle with two rubber cups into which the user pressed the handset. A loudspeaker in one cup played the tones into the telephone's microphone, and a microphone in the other picked up the incoming ones. Couplers were portable but slow, and easily confused by noise in the room. A ruling by the Federal Communications Commission in 1968, the Carterfone decision, allowed equipment from other companies to be connected to the network, and the market opened up.

By the early 1980s modems had become consumer products. Computer owners used them to call bulletin board systems, the small, often amateur-run servers where messages, games and software were exchanged. Because the line was occupied for as long as the connection lasted, and calls were charged by the minute in many countries, users learned to read messages offline and to compress their files. A culture of text-based communication took shape around the limitations of the equipment.

The Hayes Smartmodem, released in 1981, changed the way modems were used. Earlier devices required the user to dial by hand and then, at the right moment, to switch the modem into data mode. The Smartmodem could dial, answer and disconnect under the control of the computer itself, using a set of instructions that began with the letters AT and soon became an industry standard. Software could now call a list of numbers unattended, and the modem began to move inside the computer rather than sitting beside it as a separate appliance.

Meanwhile, engineers worked on speed. A telephone line passes only a narrow band of frequencies, roughly those of the human voice, so tones had to be packed more densely. Frequency-shift keying gave way to quadrature amplitude modulation, in which a single symbol is encoded as a combination of amplitude and phase, carrying several bits at once. Trellis coding added redundancy so that errors could be corrected, and echo cancellation made it possible to send data both ways over a single pair of wires without the streams drowning each other. Standards agreed during the 1990s raised the ceiling from 14,400 to 28,800 bits per second.

The last great leap of the dial-up era exploited a peculiarity of the network. By the late 1990s most telephone systems converted speech into digital form before carrying it, taking 8,000 samples a second and encoding each as eight bits — a raw rate of 64,000 bits per second. One bit in each sample was needed for signalling, leaving 56,000. If the only conversion happened at the exchange, and the link to the customer's home stayed analogue, a modem could in principle reach that figure. The V.90 standard of 1998 did so, but only downstream; in the other direction the rate remained much lower.

Dial-up's reign ended with broadband. Digital subscriber line technology used the same copper pair but sent data at frequencies far above those of the voice, so that a telephone call and an internet connection could share one line and the connection could stay open permanently, while cable networks carried data over the same lines that delivered television. Neither approach depended on an audible handshake, and the familiar screech of two modems negotiating vanished from homes. The word survived, however: the box that connects a house to a fibre or cable service is still called a modem.

A few communities, and a great many machines, still rely on the old technology. Dial-up remains a fallback where nothing else is available, and simple modems are built into equipment that needs to send small amounts of data cheaply.

Every image taken with a digital camera or a phone leaves a record, and not only the one the photographer intended. A single photograph is really a small package of data: a grid of numbers describing brightness and colour, a header containing the time, the camera model and often the location, and instructions telling software how to display it. Digital forensics is the discipline that reads these traces. It asks two questions that are often confused with each other. The first is whether an image is authentic — whether it shows what it claims to show, unaltered. The second is where it came from: which device produced it, and when. Answering either question rarely depends on a single test; it depends on several small, independent indications that agree.

The simplest tool is the hash function. A cryptographic hash takes a file of any size and reduces it to a short string of characters, a kind of digital fingerprint. MD5 produces a 128-bit value, usually written as 32 characters; SHA-256 produces 256 bits. Any change to the input, even to a single bit, produces an entirely different output, and it is practically impossible to construct two different files with the same value. An investigator who copies a suspect's hard drive can therefore show that the copy is identical to the original. But the technique has limits. A hash says nothing about which of two files was made first, or whether either has been edited. Re-saving a photograph, even without altering a pixel, changes its hash completely. And the guarantee of uniqueness is not absolute: researchers showed in 2004 that two files could be constructed with the same MD5 value, a collision.

Images themselves carry subtler evidence. Every sensor is manufactured with microscopic irregularities, so some pixels respond slightly more or less strongly to light than their neighbours. The result is a fixed pattern, unique to that sensor, known as photo-response non-uniformity. It is invisible in any single photograph because it is buried in noise. But if an analyst averages a hundred images from the same camera, the random noise cancels out while the fixed pattern remains. Comparing that pattern with a suspect image can link a picture to a specific device, even when its metadata have been removed. The technique has been used to match photographs posted online with a seized camera.

Compression leaves further marks. Most photographs are stored as JPEG files, which exploit the fact that the eye is less sensitive to fine detail in colour than in brightness. The image is divided into blocks of eight by eight pixels, each block is converted into a set of frequencies, and the high-frequency components are rounded off — quantised — more coarsely than the low ones. Every camera and every editing program uses a slightly different quantisation table, so the pattern of rounding is a kind of signature. If part of an image has been pasted in from elsewhere, that region may carry a different signature from the rest, or may show the double compression of a file saved twice. A related technique, error level analysis, highlights areas that respond differently when the image is compressed again. It is useful for directing attention, but it is a heuristic rather than proof.

Metadata are the easiest traces to read and the easiest to falsify. The header of a JPEG file may record the camera's make and model, the exposure settings and the exact time, and phones often add the coordinates of the place where the picture was taken. Editing software may rewrite these fields, and anyone can change them with a text editor. Investigators therefore look for evidence in the scene itself: the length and direction of shadows, which indicate where the sun stood at a particular time of year; the weather recorded at a nearby station; the architecture in the background; or the reflection of the photographer in a window.

Sound and video offer their own traces. Mains electricity hums at a nominal fifty or sixty hertz, but the actual frequency drifts slightly as demand and supply change across a grid. The drift is recorded in any audio containing the hum, and because grid operators publish detailed records of it, an analyst can sometimes match a recording to a particular time and place. In video, researchers look for features that are hard to fake: the pattern of blinking, the way skin changes colour with each heartbeat, and the geometry of shadows. None of these tests is decisive on its own.

The weaknesses of these methods matter as much as their strengths. Compression, resizing and re-saving destroy many traces, and a determined forger can launder an image by photographing it on a screen or printing and scanning it. Freely available tools can strip metadata or add a false sensor pattern. More fundamentally, courts have become more demanding about the software analysts use: a technique must be shown to give reliable results at a known error rate, and many machine-learning systems cannot easily explain how they reached a conclusion.

Research now moves in two directions. One is towards better automated detection, using neural networks trained on large collections of manipulated and genuine images. The other is towards prevention rather than detection. An industry consortium has developed a standard, known as content credentials, that allows a camera to sign an image at the moment of capture and to record every later edit, so that provenance can be checked rather than reconstructed. If such systems become widespread, the forensic question may shift from 'has this been altered?' to 'who has handled it, and in what order?'

In a hospital in Seattle, a patient in a headset throws snowballs at penguins while nurses clean a wound. For those few minutes, by the patient's own account, the pain is somewhere else. Scenes of this kind have made virtual reality one of the most photographed ideas in medicine, and have produced a steady stream of claims that the technology will transform mental health care. Something real is happening here, but rather less than the publicity suggests.

A virtual reality system consists of a head-mounted display with a screen for each eye, sensors that track the position of the head, and headphones. Because the images change as the wearer moves, the brain accepts the scene as a place rather than a picture. Researchers call this sensation presence, and it is the variable that matters most. Presence varies from one person to another and from one system to another, and a cheap headset paired with a phone may produce very little of it.

The best-established use of the technology is exposure therapy. A patient with a phobia is encouraged to approach what frightens them in graded steps and to stay in the situation until their predictions of disaster are contradicted. A virtual version has obvious practical advantages. The therapist can produce a lift, an aeroplane cabin, a crowded lecture theatre or a spider on demand, can make it harder or easier, and can stop it at any moment. Someone who would refuse to board a real aircraft may agree to try a virtual one, and the session needs no travel. Trials have reported benefits for fear of heights, fear of flying, public speaking anxiety and post-traumatic stress.

Pain control works on a different principle. The virtual world gives the patient something demanding to do. Attention is a limited resource, and the more of it an absorbing environment claims, the less remains for the signals arriving from a wound. The same logic has been applied to physiotherapy, where exercises can be dressed up as games, and to dentistry, where a child may sit still for a cartoon world that would not hold their interest on a screen.

Stroke rehabilitation has attracted similar interest. Recovery depends on repeating movements, which is tedious and often discouraging. Virtual tasks can give the repetition a point, and can be adjusted as strength returns. A patient who finds it hard to lift an arm twenty times may willingly do so to knock down a wall. The evidence here is thinner than for anxiety, however, with many small trials and few large ones.

Anyone reading the research literature soon meets its weaknesses. Systematic reviews and meta-analyses generally report moderate benefits, but the individual studies are usually small and short, and patients cannot be blinded to the fact that they are using a headset. Comparison groups differ widely from one trial to the next, and few studies follow patients beyond a few months, so the durability of any improvement is unknown. Publication bias is likely, since negative results are less likely to be submitted or accepted. Early enthusiasm may also inflate results: novelty probably accounts for part of the improvement seen in first trials.

What does the work is not the headset but the psychological process it engages. Exposure helps because expectations of catastrophe are not confirmed; distraction helps because attention is occupied elsewhere. Once that is understood, the technology looks less like a treatment and more like a delivery system. Where a simpler and cheaper method produces the same process — a well-constructed programme of home exercises, say — a headset adds expense and little else. The risk is that the excitement of the new will lead clinics to buy equipment before the evidence justifies it.

Practical obstacles remain. Headsets are becoming cheaper, but cost is not the main barrier. A minority of users feel nauseous or headachy in a headset, and the effect is worse when the images lag behind the wearer's movements. Shared equipment raises questions of hygiene, and someone has to set it up, maintain it and train staff to use it. Patients also differ in how readily they take to the equipment. Headsets record where the wearer looks and how the body moves, which makes the data they generate sensitive.

None of this means that virtual reality has no place in treatment. For a narrow range of problems it does something a clinic cannot easily do: it supplies a controlled, repeatable version of a situation that is dangerous, expensive or impossible to arrange, and it provides immersive distraction during painful procedures. But it is a delivery mechanism, not a therapy, and nothing in the evidence suggests that headsets will replace therapists. Money spent on trials and on trained staff would do more good than money spent on hardware.