David R Walt, Shannon E Stitzel, Matthew J Aernecke. American Scientist. Volume 100, Issue 1. Jan/Feb 2012.
For human beings, a good sense of smell might not be as coveted a trait as sharp eyesight or hearing. The 18th century philosopher Immanuel Kant singled out smell as the sense that was “most dispensable” because “the pleasure coming from the sense of smell cannot be other than fleeting and transitory.” But scents can be a significant source of information, and over evolutionary time a diverse suite of sensitive mammalian (and other animal) olfactory systems have arisen to tap into those data. Over the past few decades, researchers and engineers have created increasingly sophisticated electronic systems—artificial noses—to detect and identify numerous odors.
Artificial or electronic noses (also called e-noses) often mimic traits of the mammalian olfactory system. So why use an artificial nose at all? For one, biological noses can become desensitized after exposure to a smell, adapting to it as a new baseline of the environment. Artificial noses do not tire, and there’s no ethical dilemma about exposing them to dangerous situations or toxic substances while detecting explosives or nerve agents, or checking sewageplant odors. They can provide standard measures for foods, monitoring processes such as fermentation or checking for freshness. They can expand our sensitivity by analyzing medical samples for unwelcome microbes—for example, by sensing the volatile compounds that the microbes release. The advantages of scientifically measuring odors were evident to Alexander Graham Bell, who noted in a 1914 speech:
Did you ever try to measure a smell? Can you tell whether one smell is just twice as strong as another? Can you measure the difference between one kind of smell and another? It is very obvious that we have very many different kinds of smells … But until you can measure their likenesses and differences you can have no science of odor. If you are ambitious to found a new science, measure a smell.
It’s perhaps not surprising that it has taken so long to fulfill Bell’s request. Mammalian olfactory systems are very elaborate. They can detect thousands of different odors, both singular scents and complex mixtures, within seconds of being exposed to them and in concentrations ranging from saturated vapors to, in some cases, odorante present in the low parts-per-billion range. The olfactory system has powerful elements not present in other sensing systems. The process of sniffing an odor is the first step, and it prepares the whole system to receive a scent. Second, molecules that are inhaled into the nasal cavity are partitioned into the mucous layer on the inner surface, which may separate the various components in a mixture of scents. And most importantly, unlike virtually all other sensing systems, in which a specific receptor links to only one substance, a single receptor in the olfactory system can respond to many different odors. Mammalian genomes code for about 1,000 different types of olfactory receptors, but the system is able to detect far more odorante by using each receptor in this cross-reactive manner, combining the responses from all the receptors to create unique patterns.
Olfactory receptors, the membrane proteins that bind odorants and initiate the process of nerve signaling, at first glance appear to respond in a very nonspecific fashion. However, when the responses of all the activated sensor types are taken together, they create a complex response pattern. It is this pattern (rather than any one single receptor response) that encodes the identity of the odorant. The outputs from the cells containing each sensor type converge on the olfactory bulb, a nerve cluster at the connection of the nasal cavity and the brain. In this way the nerves sum their signals to increase the signal-to-noise ratio, strengthening the response. These patterns are then relayed to higher level processing centers in the brain, where they are perceived as a particular odor and stored in memory.
Mammalian olfaction is thus an incredibly complex but extremely sensitive and efficient system, and researchers have labored for years to mimic its capabilities and versatility in artificial systems. The components of an artificial nose have been designed to follow, in general function, the segments of the mammalian olfaction system. At their core, artificial noses, like their biological counterparts, monitor the responses from a collection of crossreactive sensors. They then use pattern recognition algorithms to connect these complex signals with a particular odor.
Even before electronic systems became available, scientists had been looking for ways to measure and quantify scents. The first olfactometer was created in 1888 by Dutch physiologist Hendrik Zwaardemaker. The apparatus consisted of a large test tube coated on the inside with an oily, odorous liquid. A small open tube was inserted into it, and the length of this smaller tube determined the intensity of the odor perceived by a person. Zwaardemaker did experiments’ using two open tubes to see which odors, when mixed, neutralized each other to the human’s perception. Later, the first practical uses of such olfactometers were to create substances that neutralized foul odors. The devices were also used to test the threshold at which a diluted odor (for example, farm waste or diesel exhaust) could be detected. These systems were sometimes called “mechanical noses,” and advanced versions are used today to present standardized odors to human testers. But such systems still require a human sensor and that person must make a subjective judgment about an odor’s intensity and other qualities. Electronic noses take the next step: They remove the human being from the process in the hopes of creating a reproducible measurement of scent.
Since the first description in 1982 of an artificial-nose system, advances in sensor design have increased the number of different mechanisms used in various devices. New data-mining techniques have also streamlined information extraction and improved the classification of odorante. Despite several approaches, artificial noses generally have three main components, which often are at least premised on the mammalian system.
The Building Blocks
The elements of an artificial nose are the vapor-delivery system, a sensor array and a pattern-recognition algorithm. Vapor-delivery systems introduce samples to the sensor arrays, and the majority are in a class called activeflow systems. Some of these sampling systems collect the air above an odorcreating substance, termed the “headspace.” Others work by diffusion and bubbling. Vapor delivery can also be done with a static system, designed to allow the array of sensors to come to equilibrium levels with a vapor, to obtain a steady-state response. The choice of an active or passive delivery system can significantly affect the reliability of signals obtained from a sensor array.
Once the vapor is delivered, the sensors can be designed to respond in several ways. Static responses make a single measurement once the sensors achieve equilibrium. Dynamic responses collect time traces from individual sensors in the array. These measurements contain more information than static responses, so this approach has proved to be more successful at discriminating odors. Detection limits have been improved by suinming the signals from hundreds of replicate sensors randomly distributed throughout an array, a process that mimics the biological system’s approach of having many olfactory-receptor nerve axons converge onto one glomerulus, a structure in the olfactory bulb of the brain.
Additional biomimicry concepts include one used by Nate Lewis and his colleagues at the California Institute of Technology, among other groups, of embedding a sensor array in an absorptive layer several centimeters long. This approach imitates the nasal mucosal layer and adds a spatiotemporal component to the array responses, which greatly improves the system’s ability to discriminate vapor mixtures. Our group at Tufts University has exploited the physical features of the canine nasal cavity in an artificial-nose design. We demonstrated with a scaled plastic replica that the highly complex flow environment found within the nasal cavity could be used to enhance the discriminatory abilities of a single sensor type.
The pattern-recognition technique used to interpret odor data is also critical to the overall performance of enoses. Some algorithms are parametric: They assume that the sensor responses are normally distributed and thus can be described by a probability-density function. Nonparametric approaches, on the other hand, do not make any assumptions about the distributions within the data, which can make them less efficient but allows them to be applied more generally.
A further distinction between methods is whether or not the algorithm is based on supervised or unsupervised learning. The former has a training stage, where a known set of input responses is used to develop a database of descriptors that can define the possible output classifications. Later, during classification, unknown sensor responses are compared to the database and are grouped by similarity to previously learned descriptors.
Unsupervised learning is more similar to how human olfaction works; there is no initial knowledge of how odors should be classified. The algorithm instead has the goal of finding new classes that define input responses. Artificial neural networks are among the approaches used in this type of algorithm.
All three of these components are vital to a successful artificial nose, but the heart of the system is the sensor array. It is essentially a transducer; it converts the presence of a vapor into a measurable signal. There are many diverse types of sensors, but most fall within three categories: electrical, gravimetric (based on mass) or optical. All of these types are used in currently available commercial systems. Some hybrid nose systems combine these sensors with gas chromatography or mass spectrometry. In the former, gas chromatography is used to separate the odor components in a first stage, and then the separated vapors are presented to a sensor array. The mechanism is a bit similar in concept to the nasal-cavity partitioning in biological olfactory systems.
Electrical devices used as gas sensors are most often based on transistors similar to those found in most electronic equipment, but e-nose transistors instead employ conducting polymers, or more recently, carbon nanotubes, as the elements that modulate current. Some of these sensors respond to changes in the conductivity or resistivity of an active layer in the device upon exposure to vapors. Others respond to a change in voltage.
The first reported artificial nose used a metal-oxide semiconductor sensor, which is composed of a film of metal oxide, a heating element and electrical contacts. The oxide material is most commonly tin, although zinc, titanium and tungsten oxides have also been used. These oxides are called ?-types, meaning that free electrons carry the electrical charge through the material. In air, oxygen adsorbs to the surface, extracting electrons from the oxide. These electron-depleted regions have low conductivity. When the adsorbed oxygen reacts with odor gases, it releases electrons back into the oxide, increasing conductivity. This reaction is dependent on temperature, so the sensitivities of the devices can be manipulated by changing operating temperature. The devices are relatively inexpensive, as they were one of the earliest commercially available gas sensors, but their operation at high temperatures makes them energy intensive. In addition, they are sensitive to water vapor, so humidity has to be controlled during their use.
Another gas sensor used is a fieldeffect transistor. The typical design for these transistors starts with a base of silicon that is p-type, meaning the mobile charge-carriers in the material are positively charged “holes,” or gaps where electrons can flow through. Two spots of ?-type silicon on top of the base are used to create a “source” and a “drain.” After an insulating layer is deposited, a “gate” material is put in place, and a positive electrical potential applied to the gate allows current to flow from source to drain. The gate voltage is set to maintain a constant current, but when a gas adsorbs onto the sensor, it alters the voltage necessary to maintain that current. These transistors can operate at lower temperatures than metaloxide semiconductor sensors, but they can be affected over time by baseline drift, a shift in the threshold voltage needed to maintain current flow.
Some types of conductive polymers can also be used for gas sensors; examples include polyacetylene, polypyrrole, polyaniline and polythiophene. Their conductivity is mainly due to electron movement along the “backbone” of the polymer (where its constituent monomers link together) rather than hopping across polymer chains. Typically, polymers with a more ordered structure have a higher observed conductivity. Although the mechanism by which these polymers respond to vapors is still not thoroughly understood, it is presumed that the vapors cause the polymer to swell, which disrupts the conduction pathway along the backbone. Conductive polymers are sensitive to some vapors in concentrations as low as 0.1 parts per million, but they are also sensitive to humidity and can be subject to baseline drift as the polymers oxidize over time.
Another approach is to blend insulating polymers with conductive materials, such as carbon black, graphite or metal powders, to create a composite conductive polymer. This mixture is deposited between two electrodes, and the conductive particles within the polymer form a network that conducts current between the two electrodes at a specific baseline of resistance. When the polymer is exposed to an odor, it swells, increasing the distance between conducting particles and, consequently, raising the resistance of the sensor. The advantage of these composites over pure polymers is the wide range of materials that can be used to make sensors, increasing the diversity of vapors that can be detected. But composites are subject to the same limitations of humidity and baseline drift.
The most recent addition to the field of electrical gas sensors are single-walled carbon nanotubes (some nanotubes have multi-layered walls). These graphite tubes, about a nanometer in diameter, conduct electricity along the axis of the tube. Structures made from the tubes resemble field-effect transistors, except that a random network of carbon nanotubes is used for the gate material. The molecules to be analyzed physically adsorb to the nanotubes’ surfaces and in the process change their conductance or capacitance (their ability to store energy). The sensitivities of the nanotubes can be altered by coating them with polymers or metals. One advantage of using nanotubes is that they respond equally well to compounds with low or high vapor pressures, so they could be most useful in detecting substances with low vapor pressures (such as chemicalwarfare agents). However, it is difficult to fabricate batches of identical nanotube sensore, which limits the technology.
By Other Means
Vapors can also be detected by their mass, using so-called gravimetric sensors. One such device is a quartz-crystal microbalance, which uses the pressure of acoustic waves for sensing. The sensor is usually a layer of single-crystal quartz that is piezoelectric, meaning that it generates an electrical current in response to pressure. The quartz is coated with a layer that absorbs odor molecules, and it is placed between electrodes that use an alternating current to generate a resonant sound wave in the quartz. When a vapor permeates the coating, it increases the mass of the film, which results in a proportional change in the resonance frequency of the quartz. Using these sensors, researchers have detected mass changes as small as 1 nanogram. But again, the devices are sensitive to both temperature and humidity. In a variation of this device, an oscillating acoustic wave can be transmitted along the surface of the quartz, rather than through the bulk. Such surface-acoustic-wave devices are operated at higher frequencies, but this feature means they also have higher background noise. An advantage is that they can readily be miniaturized, but they tend to have poor long-term stability.
Tiny cantilevers can also be used as sensors. A thin layer of material, typically silicon, is coated with a chemically selective layer. The difference in thermal expansion of the two layers makes the cantilever oscillate, like a diving board, at a fundamental frequency. When vapors enter the coating layer, its mass increases and thus changes the resonant frequency of the cantilever. Measurement of this change can be made optically by focusing a laser on the far end of the cantilever and measuring its reflection on a position-sensitive detector. Alternatively, a measurement can be made electrically using piezoresistance or capacitance. These cantilevers have a small active area, which makes them very sensitive; they can detect vapor concentrations down to the picogram range.
Optical methods can be used for sensing as well. Two of the most common are colorimetrie and fluorescent techniques, but other approaches have been developed using a wide range of platforms. Colorimetrie sensors detect a color change that occurs when a pigment interacts with a vapor. Such pigments, which change color when a vapor dissolves into them, have been developed by a variety of research groups. The first sensor of this type used a glass capillary coated with a dye film, the response of which was detected by a change in transmission of light through the film. Otto Wolfbeis and his colleagues at the University of Regensburg in Germany have demonstrated that thermal printer papers have reversible color changes in response to organic vapors such as ethers and alcohols. In this case, the response was detected by measuring a change in the absorbance of light at the specific wavelength of 605 nanometers.
Several groups have used metalloporphyrins, compounds with a ringbased structure that are the basis of many pigmented substances, such as red blood cells and chlorophyll. These compounds are stable and well characterized, and they can be modified by changing either the components in their ring structure or the metal atom bound in the center of the ring. Kenneth Suslick and his colleagues at the University of Illinois have developed an array of pigment sensors made by printing these dyes onto inert surfaces (paper or polymers). When the array is exposed to vapors, the interaction with the various dyes generates distinct color changes, which are recorded with a flatbed scanner. The color values for each spot of dye are measured and compared with the values before exposure, and the difference in color is calculated, creating a color-image map that can be processed with a pattern-recognition system. One advantage of colorimetrie arrays is that, unlike many other types of sensors, they can be designed to be insensitive to humidity. However, these sensors may be restricted to a single use as a result of compounds formed by the reaction of the vapors with the metalloporphyrins.
Fluorescence-based sensors enable the use of many different features to determine a sensor response, such as the lifetime of the fluorescence response, the emission wavelength, or the intensity and shape of the resulting emission spectrum. Such sensors have been demonstrated with polymers that are either intrinsically fluorescent or have dyes added to them. Huorescent systems require a light source to excite the sensor- often a laser, a light-emitting diode or a xenon arc lamp—as well as some kind of optical detector.
Our group has developed several types of fluorescence-based sensor systems. In the first generation of this “optical nose,” we deposited different dyeimpregnated polymers on the ends of separate optical fibers and then bundled the fibers together, so that each fiber had a unique dye-polymer combination. The other ends of the fibers were attached to a microscope and camera to record the fluorescence response.
When the bundle was exposed to a vapor, the gas penetrated into each polymer to a different extent, changing the local environment of the embedded indicator and resulting in a shift in fluorescence emission. The emission spectrum for each sensor was recorded for a two-second exposure to a pulse of vapor. The patterns of these complex temporal signatures across all sensors in the array could then be used to identify the vapors. Although such sensors are fast and give information-rich responses, the dye photobleaches. Over time, it is destroyed by light exposure.
Our second-generation optical nose was a rrviniaturized, high-density array of bead-shaped sensors. These polymer and silica microspheres are about 3 micrometers in diameter and can be easily prepared both in multiple types and in large numbers—one gram of the beads contains billions of individual sensors. The beads are placed into wells etched in the ends of optical fibers used for imaging. The fibers themselves consist of bundles of more than 40,000 strands. Each type of sensor exhibits a unique response pattern when exposed to a specific vapor, and these baseline data allow us to determine the position of each sensor type in the bundle. Many different arrays can be fabricated from the same microsphere mixture, allowing a response database collected on one array to be transferred to another, thus getting around the problem of photobleaching. We can extend longevity of the device further by masking off part of the array, measuring responses from only this section until it is photobleached, and then moving on to a fresh section of sensors.
Our first-generation device had 19 sensors on single-core fibers, but the bead-based arrays on imaging fibers have more than 40,000 sensors in an area less than a millimeter in diameter, which is the highest-density sensor array in any artificial-nose technology. The use of a large number of redundant sensors enables us to sum signals over multiple sensors, improving our signal-to-noise ratio. The main limitation is the difficulty of finding dyes with large and diverse shifts in emission wavelength.
On the Job
Commercially produced e-noses that utilize the technologies described here are now available and are being used for diverse tasks. In food and beverages, the primary focus is on assessing the quality of a product during manufacture, shipment or storage. Many of these applications are traditionally performed by highly trained human sensory panels, which can be subjective, no matter how conscientious the assessor may be. People are prone to fatigue too. E-noses may conquer those limitations.
Microbial fermentation plays a prominent role in the manufacture of many consumer goods. The types and amounts of compounds, both desired and undesired, detected in the headspace above a fermenting product depend on both the species and the growth stage of the microbes involved. Monitoring such volatile compounds can help determine when fermentation has proceeded long enough to produce a desired product without going too long and creating byproducts that can ruin a batch. It can also determine whether a batch has been contaminated by an unwanted organism. In one study, an e-nose containing eight metal-oxide semiconductor sensors was tested on fermenting tea leaves. After training on several batches, it was able to distinguish between four different process stages and correctly identify the optimal one. Other potential applications include helping with the manufacture of bread and yeast products; cheese; and pharmaceuticals, such as antibiotics, that are produced in culture.
Analyzing the headspace of a product can also link it to a geographic region of production, which can be useful in verifying its authenticity. For example, wine characteristics depend on the specific location, say a French province, vintage or even vineyard. If an e-nose were trained using quality assessments made by human sensory panels, it might eventually be able to make the same determinations from sampling the product’s headspace. In wine, the headspace is dominated by ethanol and water, so some of the more successful methodologies have removed these substances as a first step in sample preparation, allowing the more subtle yet highly characteristic components to elicit a sensor response. E-nose systems have been reported to discriminate between types of vinegars and between eight varieties of apricots from different geographical regions.
Spoilage, from bacteria or natural decay, changes the composition of volatile components released by a food. E-noses face tough competition from biological counterparts in this arena, as millions of years of evolution have rendered human noses extremely sensitive to spoiled food. Currently, the gold standard in identifying unhealthful bacterial contamination in food is culturing samples in a lab, which is both labor and time intensive. This time lag in monitoring means that most incidents of contamination are not detected until after shipment. E-noses using a variety of gas sensors have been used to detect spoilage in eggs, red meat, apples and fresh vegetables. They have also been used to screen for optimal ripeness in tomatoes and apples.
There are numerous nonfood applications for e-noses as well. For instance, the devices can be used to monitor in real time odorous or pollutant gases emitted from industrial plants. Devices near the source of pollutants can warn when the concentration rises above a certain threshold level. More compact devices might be worn by individuals sensitive to elevated levels. Two studies have found that e-noses can detect carbon monoxide and nitrogen dioxide at hazardous threshold levels. But unfortunately, they also found that water vapor destroys the sensors over time. Suslick’s research group was able to avoid this issue with their hydrophobic colorimetrie sensors, and their artificialnose platform was able to discriminate 20 toxic chemicals in amounts below exposure levels that are deemed harmful. The array they used was about a square centimeter and was read with a handheld scanner, so this device might someday be used like a radiation badge, worn throughout the day and scanned at various intervals to gauge the identity and level of exposure.
When trying to determine whether odorous compounds emitted from industrial or waste faculties are problematic, air samples are serially diluted and presented to a human panel until the people can no longer detect an odor. The level at which half of the panel cannot detect the odor is set as the threshold. But the process is costly, infrequent and time consuming, and as a result, it cannot enable preventative action. Studies of e-noses that targeted a landfill, a composting facility, a meat-rending plant and a wastewater-treatment plant demonstrated some success at detecting when odor concentrations exceeded acceptable levels. Another e-nose study was able to determine when diesel fuel had been released into a wastewater stream. These studies underscore the fact that appropriate training of the system is critical. Even though devices can be calibrated using samples rated by human testers, it’s not clear whether the human being and the machine are reacting to the same chemical in the sample. If subsequent samples don’t have the same combination of odors, the device may become inaccurate.
Lifesaving uses of e-noses are more obvious when it comes to the detection of explosives and nerve agents. Such a task is challenging, as an e-nose needs to be very selective, provide a high probability of detection, respond within seconds and have a low false-alarm rate. Most explosive and nerve agents have vapor pressures in the low parts-perbillion range, making the task even more difficult. Our optical-fiber fluorescence system was used to test for DNT (dinitrotoluene), a stand-in for TNT (trinitrotolulene), in the headspace above contaminated soil, and was found to detect the vapor at concentrations as low as 120 parts per billion. In the laboratory, our microsphere array-based e-nose system correctly identified a nerve-agent surrogate when it appeared in approximately 50 out of 17,700 vapor exposures over two weeks. This system can also detect diesel fuel, often a component in improvised explosive devices.
Lewis’s group has used composite polymer sensors to successfully detect DNT and nerve-agent surrogates. LaI Pinnaduwage and his colleagues at Oak Ridge National Laboratory in Tennessee have found that microcantilevers are particularly sensitive to certain explosives, including DNT, with detection limits in the parts-per-trillion range and a response time of a few seconds. Moreover, the responses were reproducible over a year of testing. A few studies have also shown the sensitivity of carbon nanotubes to nerve agents and explosives surrogates. The high surface area of the nanotubes, combined with their ability to undergo charge transfer when molecules adsorb to their surfaces, makes them a promising new material for sensing such vapors.
Finally, e-noses may find a place in medical diagnostics. As in food fermentation, culturing is often used to diagnose a microbial infection, a process that is quite reliable but also time consuming. Several researchers have shown that microbial growth in culture produces a mix of volatile compounds that varies by species and growth medium. Clinical studies report that commercially available enoses with conductive-polymer sensors can identify Helicobacter pylori (the causative agent of some stomach ulcers) in breath samples and Mycobacterium tuberculosis in sputum samples, as well as several microbes commonly found in urinary-tract infections.
Breath samples work well in e-nose diagnostics because air that originates from deep in the lungs comes from structures that are constantly exchanging volatile compounds with the blood. Several studies have used various e-nose sensors to measure the breath of patients with lung cancer and other respiratory conditions, and were able to distinguish them from healthy control participants. In one instance there was some evidence that the device could discriminate between disease stages. A second, chronic condition targeted by e-noses is diabetes, because the disease generates acetone. In two studies of the breath of diabetic patients, e-noses performed fairly well in distinguishing them from healthy controls. The headspace above other bodily fluids can also be used.
As more features of the mammalian olfactory system are integrated into artificial-nose technology, these devices will likely become more effective at some complex discriminating problems. The ultimate success of such systems will be their ability to adapt and perform at a level comparable to biological systems.
One of the biggest barriers to implementing e-noses is the lack of a clear large-scale application. Despite the fact that a variety of uses have been explored, these applications are often highly fragmented. In food processing, for example, every type of food is a different application that requires the sensor array to be trained, verified and validated. In addition, each application may only have a limited number of locations where an artificial nose can be deployed.
Another barrier is a user’s expectation of what artificial noses can do. If a complex mixture is presented to an artificial nose, it can be trained to identify the mixture, but it cannot identify all of the components. Unlike gas chromatography, artificial noses do not separate mixtures into their constituents. They cannot provide molecular-level identification, as mass spectrometry does. They are designed to be broadly responsive rather than compound specific.
Calibration is another issue. This property involves an extensive training regimen to ensure reproducible classification of vapors. A related aspect is the transfer of training, as it is extremely difficult to prepare two identical arrays, but it is too complex to separately calibrate each array. This issue must be solved if artificial noses are ever to be widely implemented.
Most artificial noses still have detection limits that do not approach the high sensitivities of mammalian noses. New sensing materials, such as molecular receptors^ as well as new detection modalities, will be needed to address the issue of sensitivity. In addition, sensor diversity must be improved. In this regard, efforts are underway using a strain of genetically modified yeast that has been engineered to express mammalian olfactory receptors on its surface.
Assuming some of these issues can be resolved, artificial noses offer analytical capabilities that alternative systems cannot provide. They have the intrinsic capability to detect all vapors and could give anticipatory responses. As such, we can envision a variety of additional applications. For example, a system for detecting chemical agents in public spaces, such as subway systems or government buildings, would be an ideal use for e-noses. One day, artificial noses may even be found in the home for applications such as air-quality monitoring, food freshness and food preparation (for example, to determine whether cookies are done baking).
Drawing on principles gleaned from the biological olfactory system has provided a new model for designing artificial sensing systems. As we learn more about the structure of odorant receptors and the underlying principles of biological olfaction, new design features can be implemented in artificialnose systems.