Definition and closed-loop boundary
A biofeedback adaptive aroma regulation system feeds physiological or behavioural signals into control software that changes the timing, intensity, duration or formulation channel of fragrance release. A typical chain includes wearable or contactless sensors, signal processing, state estimation, decision rules, a scent actuator and a human interface. It differs from an ordinary timer because output changes in response to a measured state rather than following only a fixed schedule.
The word “biofeedback” does not by itself establish a medical use, therapeutic effect or regulatory status. Some products merely use heart rate, movement or a sleep estimate to automate the environment; they do not present validated physiological feedback to a user as part of training. Documentation should separately state what is measured, how that measurement becomes a control decision, what is released, and whether the intended use is comfort, research or medical support. A closed loop alone cannot demonstrate a sleep or stress benefit.
Sensor inputs and data quality
Inputs may include heart rate and variability, breathing, motion, skin temperature, electrodermal activity, sound, bed presence, room temperature and humidity, airflow and time. Wrist optical signals depend on fit, movement, skin conditions and contact. Bedside radar, microphones or pressure sensors can mix a bed partner, pet, fan or position change into the data. A reading from one sensor is not a complete physiological state and does not automatically establish a sleep-stage diagnosis.
The system should expose valid signal, low confidence, missing data and offline-sensor states, and record quality rather than emit only a conclusion. Calibration, placement, clock synchronisation, sampling interval, packet loss and firmware updates can change classification. If an algorithm consumes sleep stages from a consumer device, those labels remain estimates from a particular algorithm. Validation must address the device, software version, posture and intended population; a “deep sleep” label from different brands is not an interchangeable measurement.
Decision model and personal calibration
Control logic may use thresholds, rules, a personal baseline or a machine-learning model. The design needs a defined target, decision window, minimum confidence, output ceiling, lockout interval and exit condition. A brief heart-rate change, for example, may reflect movement, a dream, room temperature, alcohol, illness or artefact and should not immediately be interpreted as a need for more fragrance. The model must not treat association as causation or repeatedly increase output while chasing a metric in an uncertain state.
Personalisation should begin with a no-output or minimum-output baseline. Fragrance tolerance and actuator behaviour should first be checked while the user is awake and able to intervene before adaptive rules are introduced gradually. Records should retain the triggering signal, confidence, issued command, execution result and manual override. Reassess the baseline after material software changes, a replacement sensor, illness, medication changes, travel or a major shift in sleep schedule. A setting that appeared suitable before is not permanently valid or transferable to another person.
Scent actuator and exposure control
The actuator may use a fan cartridge, warming, waterless atomisation or a water-based appliance that expressly permits fragrance. A software command is not the dose in the room. Delivery also depends on consumable viscosity and remaining volume, nozzle condition, airflow, room size, doors and windows, surface adsorption and residue from earlier scents. Olfactory adaptation can make odor seem weaker without showing that airborne material has disappeared, so perceived fading alone must not trigger automatic escalation.
Set limits for each event and night, a minimum interval, maximum continuous run, fragrance-change conditions and a dependable stop. Formulations must not be mixed without compatibility confirmation. Obstruction, an empty cartridge, leakage and pump failure require direct checks rather than inference from a report that no odor is noticed. Fragrance must not deliver medication or serve as a smoke, gas, oxygen, apnoea or other safety alarm; those purposes require dedicated and validated equipment and response paths.
Fault state, human control and shared rooms
Low confidence, signal loss, an algorithm fault, network outage, clock error, overheating, leakage or unknown actuator state should default to no new release, not maintained or compensatory output. A local stop must work without a phone, cloud service or account, and restart behaviour after power loss must be explicit. Alarm, manual override and automation priorities should be fixed so that a remote schedule cannot restart output after a user has stopped it.
One person's sensor data cannot decide exposure for everyone in a shared bedroom. Obtain agreement before activation and provide disablement or physical separation for people with fragrance sensitivity, asthma or migraine susceptibility and for rooms shared with children, pregnant occupants or pets. Irritation, coughing, headache, nausea or worse sleep calls for stopping output, ventilating and investigating. Persistent symptoms or a suspected sleep disorder need qualified assessment, not continued self-experimentation through algorithm changes.
Data, privacy and network dependence
Heart rate, breathing, motion, audio and sleep schedules may be sensitive health or behavioural data. A product should collect only what is necessary for a defined function and explain local and cloud processing, retention, account access, third-party disclosure, model-training use, export and deletion. A microphone or occupancy sensor may also capture non-users, requiring visible notice and appropriate permission in a shared space.
Transmission and storage need protection proportionate to risk, permissions should follow least privilege, and old device or account access must be revocable. Offline behaviour, essential control after cloud discontinuation, the security-update period and a vulnerability-reporting channel should be disclosed before purchase. A dashboard marked “off” still requires confirmation that the physical actuator stopped. Consent to data processing and consent to fragrance exposure are separate decisions and must not be collapsed into one checkbox.
Validation, evaluation and use limits
Evaluation should be layered: establish sensor accuracy and loss rates in the intended posture and population, assess state estimates and triggering rules, then test the end-to-end path from trigger to release, stop and room decay. Record false and missed triggers, latency, dose repeatability, recognition of obstruction and empty consumables, power and network failure behaviour, manual stop time, night noise and residue under different room conditions. Algorithm versions, thresholds and consumable formulations must remain traceable.
A claim about sleep onset, sleep stages, anxiety or another health outcome needs evidence for the complete product, actual control policy, intended population and a meaningful comparison. A study of one fragrance ingredient, sensor or algorithm is not enough to validate the combined system. Consumer data does not replace diagnosis of a sleep disorder, and the system must not independently change clinical treatment. Its defensible value is transparent, stoppable and testable environmental automation, not presentation of an uncertain physiological inference as a certain therapeutic effect.