I. Stop "looking" at light, start "listening"
Traditional optical gas sensors are intuitive: shine a beam through a cell, see how much is absorbed, and let the remaining light hit a detector. NDIR works this way, and so does TDLAS. The only difference is how narrow the source is and how fast the detector is.
Photoacoustic Spectroscopy (PAS) turns this chain around. Instead of transmission measurement, it makes the gas itself the sound source. The process is as follows:
- IR light is intensity-modulated, flashing at a certain frequency onto the cell;
- Target gas molecules absorb photons of a specific wavelength and jump to an excited state;
- Excited molecules collide with surrounding molecules and release the energy as heat — this is vibrational-translational (V-T) relaxation;
- Inside the sealed cell, periodic heating causes the gas to expand and contract periodically;
- Expansion and contraction are pressure waves, i.e. sound waves;
- A tiny microphone on the cell wall picks up the sound pressure.
Throughout, the microphone only responds to the AC sound pressure at the modulation frequency and is completely blind to steady DC light. This yields a key property: zero baseline.
What is zero baseline? A dirty window, an aging source, a shifted optical path — in NDIR these slowly drift the transmitted intensity and bias the reading. In PAS, they only attenuate the acoustic signal overall; they never conjure a "false concentration reading". In other words, PAS has no optical zero-drift problem.
II. Amplifying the sound: acoustic resonant cavity
The photoacoustic signal is naturally very weak. In a micro-cell, a few-milliwatt IR beam may produce a sound pressure of only a few micro-pascals — orders of magnitude below the faintest sound a human ear can hear.
To reliably pick up this signal, two things are needed:
First, set the source modulation frequency to the acoustic resonance frequency of the cell. A common structure is the Helmholtz resonator — a sealed cavity with a neck. At resonance a standing wave forms inside and the sound pressure is amplified tens of times. This Q-factor amplification is the key that lets PAS extract a signal from micro-pressure.
Second, choose a modulation frequency away from environmental noise. Low frequencies carry HVAC fans, footsteps and traffic vibration; mid-to-high frequencies are relatively clean. A typical PAS system sets the modulation frequency from tens to hundreds of hertz — away from environmental noise, yet not so fast that relaxation can't keep up.
The latter half of the signal chain is standard lock-in amplification: the weak signal from the microphone passes low-noise pre-amplification and band-pass filtering, then a digital lock-in algorithm extracts the amplitude at the modulation frequency. That amplitude is, theoretically, proportional to gas concentration.
III. CO?'s special temper: relaxation compensation
PAS has a physical detail that other routes hardly worry about: relaxation time.
After absorbing a photon, a molecule does not instantly turn the energy into heat. Returning from the excited state and handing the excess energy to neighboring molecules takes time. If this relaxation time approaches or exceeds the light-pulse period, the energy hasn't become heat before the next pulse arrives — the signal attenuates and its phase lags.
CO? is especially sensitive here. Its V-T relaxation rate depends heavily on water-vapor content. Dry CO? relaxes slowly and signals poorly; add a little water vapor and the relaxation rate rises sharply. This means the same PAS system measuring the same CO? concentration in different humidity can output very different sound-pressure amplitudes.
So engineering must do one thing: embed temperature/humidity sensors to measure conditions in real time, then use an algorithm to compensate for the relaxation-induced signal change. This is not a nice-to-have calibration — it is a prerequisite for a PAS CO? sensor to work at all.
IV. The "chip-scale" of PAS
Early PAS was lab equipment: big cells, heavy sources, expensive microphones. What truly brought it into consumer and industrial markets is MEMS technology.
Today's chip-scale PAS sensors have all core parts miniaturized:
- IR source: MEMS thermal emitter, emits broadband IR when powered, millimeter-sized;
- Narrow-band filter: stuck directly on the source or cell window, capturing the target gas absorption band;
- Cell: a micro sealed cavity etched in a silicon substrate, including the acoustic resonance structure;
- Detector: MEMS microphone, a mature device originally developed for phones, used directly.
The entire photoacoustic transduction is compressed into a roughly cubic-centimeter package. Add an MCU, driver circuit and algorithm, and a postage-stamp-sized module can accurately measure CO?.
Compared with an equivalent NDIR module, PAS's size advantage is clear. NDIR needs a physical cell of a certain length to guarantee optical path, and is hard to shrink below 1 cm3. PAS's acoustic cavity can be more compact because the signal does not depend on optical path length but on optical power and cavity resonance.
V. Strengths and boundaries
PAS's strengths come down to three:
First, zero baseline. No optical zero drift, good long-term stability; window contamination only affects gain, not the zero point.
Second, small size. The acoustic cavity can be more compact than an optical cell, fitting wearables and battery-powered scenarios.
Third, controllable power. MEMS source works in pulses; with a low-power MCU and microphone, typical average power stays in the tens of milliwatts.
But PAS has its boundaries:
- Cross-interference remains. The narrow-band filter passband is on the order of tens of nanometers — not much better than NDIR. If several gases absorb in the same band, PAS cannot distinguish them by line resolution the way TDLAS can.
- Sensitive to ambient vibration and airflow noise. The inlet needs acoustic-impedance design to stop external noise coupling into the cavity.
- Relaxation effects need compensation. Especially for CO?, you must embed temperature/humidity sensors and run a compensation algorithm.
- Detection limit is bounded by the acoustic noise floor. For sub-ppb trace gases, PAS is currently less competitive than TDLAS.
VI. Applications: from indoors to the car cabin
PAS's most successful commercial landing is indoor CO? monitoring.
Fresh-air systems and demand-controlled ventilation (DCV) need indoor CO? concentration to judge occupancy density. The traditional solution is NDIR — stable, long-lived, reliable. But the physical size of NDIR modules limits them in more miniaturized devices. PAS fills the gap: at the same ppm-level CO? measurement, PAS can shrink the module to ~1 cm3, letting CO? sensors enter thermostats, smart speakers and wearables — product forms where an NDIR cell never fit.
Another fast-growing application is the car cabin. EVs pay increasing attention to cabin air quality; CO? sensors drive automatic recirculation switching and heat-pump fresh-air control. Automotive-grade PAS modules beat NDIR on size and vibration resistance.
Battery thermal-runaway early warning is another direction. In early thermal runaway, electrolyte decomposition releases CO, CH? and organic vapors. PAS can use several narrow-band filter channels to detect each gas and combine temperature/humidity for a comprehensive judgment. Note, however, that H? has no IR activity, so PAS cannot directly detect hydrogen and needs an electrochemical or catalytic combustion sensor as a supplement.
VII. Trend: where three routes meet
PAS will not replace NDIR, nor TDLAS. But the boundaries between them are blurring.
On one hand, PAS is pushing to higher sensitivity. With higher-power mid-IR sources, optimized cavity Q, wavelength modulation and 2f demodulation, PAS detection limits are approaching sub-ppm and even ppb levels, starting to erode the application space traditionally held by TDLAS.
On the other hand, NDIR is also miniaturizing. Through MEMS sources and thermal detectors, NDIR modules are shrinking too, though not yet as extreme as PAS.
The three routes share the same Beer–Lambert law; they differ only in how they treat "the part of the light eaten by the gas". NDIR measures how much is left, PAS listens to the sound after it's eaten, TDLAS stares at a single spectral line to see how it deforms.
For engineering selection there is no absolute winner — only a match to the constraint set: size, power, detection limit, cross-interference, maintenance interval and unit price. Whichever weight dominates, that route wins.