WHEN THE SENATE begins hearings tomorrow on the INF arms-control treaty, a crucial issue will be whether -- and how -- the United States can detect potential Soviet violations.

The subject is critical because both INF and the Strategic Arms Reductions Talks (START) now being negotiated in Geneva will require unprecedented verification technologies. Unlike previous arms accords, which involved watching large structures such as fixed missle silos and bombers, INF and START would restrict individual small missiles.

Consequently, America's traditional monitoring systems -- surveillance satellites and electronic intelligence -- will not be sufficient. Extensive cooperative measures will be needed. Trucks and railroad cars must be inspected; plant gates, grounds and fences watched; small weapons examined for nuclear content. And the INF pact requires new systems for continuous monitoring of missile-production facilities.

Exactly what kind of sensors the U.S. will place on Soviet soil is still being decided. But U.S. research on on-site inspection systems -- involving tamper-resistant fiber-optic seals, video alert and data-encryption systems, infrared surveillance arrays and more -- is already well underway, much of it at Sandia National Laboratories in Albuquerque, which AT&T runs for the Department of Energy.

Overcoming Mutual Mistrust

Problems abound when working in an adversary's territory. The verification system must ensure that each side can trust the authenticity of the on-site data, producing a paradox: The Soviets must be assured that information gathered is for verification only -- not espionage -- and that it agrees with the facts. Thus, data cannot be encrypted. But the U.S. must be confident that streams of data, traveling through open channels in Soviet territory, are not forgeries. Thus, encryption must be used.

Cracking that puzzle falls to Sandia's Gustavus J. Simmons, a mathematician with a footlong beard and a flattop coif who has been solving such brainteasers for 20 years. Simmons and his colleagues are perfecting a data system that guarantees integrity. The technique that the U.S. will propose to the Soviets is the least sophisticated of Simmons' schemes. {See box.} But it has already passed the scrutiny of codebreakers at the National Security Agency and has been discussed with the Soviets in the '70s during the Comprehensive Test Ban Treaty talks. Moreover, it has proven reliable at remote U.S. seismic stations in Norway that glean data from Soviet underground nuclear tests.

The system works by automatically attaching an authentication "word" to the output of a monitoring device such as a camera. The output, in the form of a long binary sequence (strings of 0s and 1s), is fed into a computer, which breaks the data up into small blocks of, say, 64 bits. The first block is encrypted with a secret key which produces a 64-bit cipher held in the computer's memory. As the second block of data arrives, each element in the cipher is matched with its corresponding element in the second data block. If the two elements are alike, a 0 is recorded; if different, a 1. This produces a new 64-bit number, which replaces the first cipher and is in turn encrypted with the key. This new cipher is then matched against the contents of the third block of data, which begets yet another new cipher and so forth. The end result, after processing an entire data stream thousands of bits long, is a final 64-bit cipher incorporating information about each bit of data in the whole stream. This final cipher, or authentication word, is appended to the unencrypted monitor output and sent.

Nothing in the process prevents the Soviets from scrutinizing the data while it is sent. And the U.S. can verify the result by running the received data through the same encrypting procedure using a copy of the secret key. If the final cipher generated matches the one that was appended to the original monitor output, then the data are genuine. Just as increasing the number of grooves in a housekey makes it harder to pick the lock, the more variables there are in a cipher key, the less the likelihood of cracking the code.

On-Site and On Guard

Shortly before the 1986 summit in Reykjavik where the USSR agreed to U.S. proposals regarding on-site INF monitoring, Roger L. Hagengruber, vice president of systems analysis at Sandia, got a phone call from the Pentagon. DOD wanted a full-scale test facility built to examine schemes for continuous monitoring of a Soviet weapons-production plant. It also wanted a working model of the site. The project was given top priority.

Within two months, the Sandia team produced a table-top model showing the section of a typical Soviet missile factory which includes the main portal. {See illustration.} The Pentagon displayed the model to officials from the White House, State Department and Congress, demonstrating how a missile-carrying truck triggers a suite of sensors to record weight and other data. One U.S. official called it a "good marketing tool" which helped policymakers visualize potential problems.

Both the U.S. and USSR have agreed that INF monitoring systems for the exits will include "weight sensors, vehicle sensors, surveillance systems and vehicle dimensional measuring equipment." In addition, "non-damaging image-producing" gear will be installed to examine contents of shipping containers and launch canisters. The goal is to devise a system that automatically collects and records data 24 hours a day. The monitoring system must be accurate enough to detect potential violations but work fast enough so traffic flow is not unduly impeded. And because deployment within Soviet borders precludes use of trade-secret equipment, engineers must create highly reliable systems composed mainly of commercially available gear.

One such device is a vertical and horizontal array of infrared sensors to measure rapidly the length and profile of various vehicles leaving the plant. Like radar, the system would send out its own energy beam to sense objects day or night and in adverse weather. For weighing, Hagengruber says commercial scales can be modified to assess a moving truck or to weigh loads on freight trains. Railroad cars in some respects are easy to monitor because they are constrained to tracks, have a fixed geometry and uniform velocities. But they also pose special problems because they may weigh much more than their freight and the cars may come in a mix of say gondolas and box cars. To skirt this problem, Hagengruber says they may negotiate that only certain types of train cars are allowed into the plant.

If a vehicle is large and heavy enough to be carrying a prohibited missile, its cargo will be examined by nondestructive imaging, most likely by X-ray sensors tuned to appropriate intensities. X-rays can take measurements and determine material composition and are generally hard to deceive. Manufacturers of rockets routinely use them to inspect solid propellants for cracks. For verification, however, the scans must occur faster than industrial applications, and probably be less intrusive too, says Hagengruber.

Sandia is also examining tamper-resistant seals that would reliably indicate if enclosures had been breached. In one such device, a loop of multistrand plastic fiber-optic cable is cut to desired length in the field. Its ends are put into a one-piece seal body which contains a serrated blade that randomly severs a portion of the cable fibers. The result is a unique "signature" of the uncut fibers. That pattern is photographed. If the fiber-optic loop is later released, the blade is designed to cut additional fibers and change the signature. During inspection, a second Polaroid shot is taken for immediate comparison with the original signature.

Warheads and Holograms

Authorities note that it is much easier to verify a ban than a residual force of, say, 100 missiles. Consequently, the START pact, because it seeks to halve levels of strategic warheads, will require more strict measures.

In addition, the two superpowers are discussing in Geneva how to limit nuclear-tipped cruise missiles on ships and submarines. Such controls pose special monitoring problems because the missiles are much smaller than other strategic weapons and because some are fitted with conventional warheads.

Researchers have been examining techniques to "tag" concealable mobile nuclear weaponry. The challenge is to design a system that permits counting for verification but does not allow targeting by the military.

Fred Holzer, deputy leader of verification at Lawrence Livermore National Laboratory, outlined further constraints during a 1986 interview. The tags must be tamper-proof and impossible to duplicate; and they must in no way interfere with the missile's operation, he explained. Moreover they must be designed so they cannot be used -- or even be perceived to be usable -- as a homing device.

Numerous schemes exist. For new missiles, tags might be installed at the production line. One possibility is to make a special mold with an intricate surface pattern for producing a tag. After the required number were produced, Holzer said, the mold could be broken. Another possibility, for new or existing weaponry, is to make a photomicrograph or acoustic hologram of a small patch on the missile. Each weapon examined could then be checked against a database of the fiber patterns of "legitimate" missiles.

Yet another option is to use a microchip tag that could be queried on inspection. The basic technologies that might be used are being employed by auto manufacturers including BMW, Fiat and Honda. BMW's assembly line uses chips coded to contain such information as paint color, options to be installed and so forth for each chassis. The chip is queried during assembly stages and the specified actions taken. Honda uses an intrinsic property, like fiber grains, to guard against piracy in spare auto parts. Other scenarios are akin to existing methods of satellite tracking of caribou. A U.S. satellite monitors free-roaming herds fitted with radio transmitters in northwestern Alaska to an accuracy of 0.8 kilometers. Holzer says that "these kind of techniques are being developed rather rapidly." A senior Administration official observes, however, that despite all the studies no practical tagging schemes had yet emerged.

START verification might include a plan to designate assembly areas to make missile production more transparent to surveillance satellites. But satellites cannot provide the sort of information that Soviet leader Mikhail Gorbachev mentioned in his summit farewell speech, when he shocked many observers by declaring that the USSR had a technique that would remotely "identify not only the presence but also the capacity of the nuclear warheads aboard such vessels."

If it exists, such a device probably emits a pulse of high-energy neutrons to induce a small amount of fission in any nuclear warhead. The pulse would have to be weak enough to prevent the degradation in the reliability of the nuclear weapons but strong enough to produce a recognizable signature of gamma rays or neutrons. But because of the rapid degradation of this signature in the atmosphere, such measurements must be made from close range. Moreover, shielding by lead or water could foil the inspection. More detailed schemes must be fielded for effective START verification.

A November 1987 report by the House Intelligence Committee was unanimous in saying that the Executive Branch provides "no central direction and prioritization of research and development to improve arms control monitoring capabilities." It placed the blame largely on the intelligence community.

Indeed many technologies for use in the INF treaty were developed for other purposes. Participants say some analyses, like whether inspections of suspect sites should be allowed, were done hastily.

Although INF negotiations began in November 1981, money for the major INF monitoring program started flowing several years later. The Department of Defense was the surprising source, including the international security policy branch formerly headed by Richard N. Perle, popularly known as a bete-noire of arms control.

In spite of budget constraints, Congress supplemented administration requests for verification research for fiscal 1988. Whether that results in new innovative techniques for monitoring the strategically sensitive START pact remains to be seen.