Ever bought a perfume online with rave reviews and awesome packaging, only for it to arrive and the first whiff brings tears to your eyes ... tears of sorrow? Well, that may soon be a thing of the past, as scientists have figured out a way to transmit smells over the internet. Technically speaking.
Researchers at The Ohio State University and Zhejiang Laboratory in China have created a system that detects an odor in one place and reproduces a corresponding scent somewhere else.
The system consists of two devices: an electronic nose (e-nose) that sniffs out airborne chemicals at the sending end, and a device called Scentronic that releases fragrances at the receiving end. Machine-learning software and the internet connect the two. So far, the team has demonstrated the system with beer, cake, coffee, oranges and red wine.
We've long ago figured out how to capture and transmit sights and sounds. Specialized devices can even reproduce aspects of touch. Smells, however, have proved trickier. A camera records light falling on a sensor, while a microphone records changes in air pressure caused by sound waves. Both light and sound follow the laws of physics, a framework that works perfectly with electronics, allowing them to be digitally replicated.
A smell, however, is chemistry, comprising molecules floating through the air, often in complex mixtures. Their identities and concentrations determine what you perceive. An aroma has no straightforward equivalent to a photograph or audio file. So how do you capture one, and how do you convert what you've captured into digital signals? These are the questions the researchers set out to solve.
The e-nose takes the form of a circular circuit board carrying six metal-oxide semiconductor gas sensors. Their main sensing material is tin dioxide.
In ordinary air, oxygen molecules settle onto the sensor surfaces and draw electrons away from the material, increasing its electrical resistance. When volatile organic compounds (VOCs) from food or drink reach those surfaces, some react with the adsorbed oxygen and return electrons to the tin dioxide. Resistance drops, and the circuit registers a corresponding specific rise in voltage.
Ethanol from alcoholic drinks, limonene from citrus, and diacetyl from buttery aromas are some of the VOCs that can produce these reactions. Because each sensor responds differently, each source produces a distinctive voltage pattern across the array. Think of it as a chemical fingerprint, albeit a fairly blurry one.
Now, data that vaguely matches a scent to a voltage pattern is available. However, precise matching is necessary for the system to be useful. For this, the researchers turned to AI.
They collected 250 labeled recordings, 50 for each of the five foods and drinks, and measured the sensors’ changing voltages over 200 seconds per recording. They then trained and compared several machine-learning models to associate those patterns with the correct categories. Once the system recognizes a familiar pattern, it can assign a label such as "coffee" or "orange." A computer classifies the digitized signals, then relays the result through an internet-connected cloud server.
Now to the receiving end, where the Scentronic device lies. The device contains miniature chambers filled with pre-prepared scented liquids, each linked to its own release channel. A cotton filter draws liquid towards a piezoelectric atomizer, a tiny disc that flexes when electricity is applied. Driven at roughly 110,000 vibrations per second, it forces liquid through microscopic openings as a mist of fragrant droplets.
When a "coffee" instruction arrives as electronic signals, the receiver activates the channel loaded with a coffee-like fragrance. An ESP32 microcontroller controls the atomizer with rapid electrical pulses, adjusting spray frequency to regulate how much scent is released. It can also activate multiple channels together to create mixtures, varying their relative contributions by controlling each channel independently.
Importantly, the Scentronic device doesn't build the scent from scratch. It’s already preloaded with the five possible accents. The AI identifies the smell category, while the receiver follows programmed instructions to release the matching fragrances already loaded inside.
The team explored several Scentronic designs, including a lightweight wristband with two scent channels, a necklace with five arranged like pearls on a string, and a handheld five-channel version. The wristband can be raised towards the nose for a sniff, while the necklace keeps the emitters close to the face.
In controlled tests, the e-nose’s machine-learning models classified the five smell categories with accuracies ranging from 96% to 100%, depending on the method used. However, when 14 volunteers identified fragrances from the handheld Scentronic, they got about 73% right for individual smells and 71% for two-scent combinations. In a separate intensity-ranking test, accuracy was around 59%.
It should be noted that smell-generating wearables do already exist. In 2023, for example, researchers in Hong Kong demonstrated wireless VR interfaces that released preloaded fragrances, plus we've seen a face-mounted version capable of presenting over a dozen odors. The Scentronic team, however, combines electronic sensing, AI recognition, internet communication, and scent release into a remote-sharing system.
Of course, there are some sizeable catches. The e-nose currently recognizes only five categories, and the training used just one product from each. Scentronic also needs a matching fragrance loaded in advance for every smell it is expected to produce. It cannot sniff an unfamiliar perfume, discover its complete molecular recipe, and faithfully recreate it for an online shopper. With just five channels, the result is a recognizable stand-in for the original smell.
Still, there's room for that to be useful. One day, online shops could let customers sample scents before ordering – although the right fragrances would need to be available at the receiving end. A more practical application may be in virtual reality, where a game can ship with a dedicated Scentronic pack of smells that release according to gameplay. Although (again), the e-nose doesn’t necessarily have a role in this scenario.
The researchers even brought taste into the picture.
They built an experimental cup with five scent channels and five miniature electromagnetic pumps that drip flavoring liquids into water. When someone wearing a mixed-reality headset selects a virtual beer or coffee, the cup releases a matching aroma and adds the corresponding liquid for taste. They also demonstrated responses to beer, coffee, and their combination. The prototype shows how sights, sounds, smells, and tastes could work together.
“This has broad applications for patients with nutritional challenges and populations like astronauts (facing unappealing food supplements) or warfighters (consuming field rations), where enhancing perceived food quality can have critical nutritional/performance outcomes,” the researchers wrote. Although (one last time), couldn’t they just add the enhancements themselves? It's not exactly easy to identify a scenario that fits the system as it is. That said, it is a prototype that can be developed for many future applications.
Next up for the researchers are more sensitive sensors, bigger training datasets, additional fragrance channels, and better handling of complex mixtures. They also want more rigorous human testing and standardized methods for assessing medical smells. For now, those regrettable perfume blind buys will just have to continue.
The study is detailed in the journal Science Advances.
Source: American Association for the Advancement of Science via EurekAlert
