Time-Domain THz Emission Spectroscopy

• Time-domain terahertz emission spectroscopy is a coherent method that uses an ultrafast pump to generate THz pulses and a synchronized probe to sample the emitted electric field.
• It differs from traditional THz-TDS by measuring the emitted field directly, enabling access to both amplitude and phase for analyzing ultrafast charge, spin, and plasmonic dynamics.
• Advanced implementations include on-chip, vector-resolved, and noise-aware acquisition strategies, offering high spectral resolution and precise mechanism discrimination.

Time-domain terahertz emission spectroscopy is a coherent, field-resolved methodology in which an ultrafast excitation generates a terahertz transient and a synchronized probe samples the emitted electric field waveform $E_{\mathrm{THz}}(t)$ as a function of delay. Fourier transformation of the recorded time trace yields the spectral amplitude and phase, so the technique accesses source dynamics through the emitted field rather than through intensity alone (Zhao, 2023, Shen et al., 2020). In this sense it is the emission-side counterpart of terahertz time-domain spectroscopy (THz-TDS): the pump creates THz radiation, the probe samples the emitted field coherently, and the waveform encodes ultrafast charge, spin, plasmonic, or near-field dynamics at the source.

1. Definition and measurement formalism

The central observable is the time-dependent THz electric field. In standard THz-TDS language, the recorded waveform is Fourier transformed according to

$$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$

so both amplitude and phase are retained (Shen et al., 2020). Because the measurement is field-resolved, the complex spectrum is obtained directly rather than reconstructed from intensity-only data, and the spectral resolution is set by the total scan window,

$\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$

This makes the time window itself a spectroscopic parameter (Zhao, 2023).

In emission spectroscopy, the conceptual structure differs from transmission or reflection THz-TDS. Transmission experiments relate an input waveform and an output waveform through a transfer function, whereas emission spectroscopy measures the field generated by the sample or source itself. A plausible implication is that the forward model must be written in terms of the emission mechanism, detector response, and noise covariance rather than as a simple ratio of input and output spectra. This distinction is explicit in time-domain maximum-likelihood analysis, which notes that the standard two-waveform transfer-function setup must be modified for emission problems (Mohtashemi et al., 2020).

A recurring misconception is that THz emission spectroscopy is fundamentally a power measurement. In the time-domain implementations discussed across the literature, it is instead a coherent electric-field measurement. That distinction underlies subsequent extensions to vector polarimetry, nanoscale waveform sampling, and uncertainty-aware model fitting (Zhao, 2023).

2. Sources, detectors, and canonical instrument architectures

A canonical implementation uses a femtosecond laser split into pump and probe arms. The pump generates the THz pulse, and the probe acts as a short temporal gate that samples the THz field at successive delays (Zhao, 2023). In photoconductive systems, the pump pulse creates carriers in a biased semiconductor, the applied field accelerates them, and the resulting ultrafast current transient radiates a broadband THz pulse (Shen et al., 2020). A representative photoconductive-antenna spectrometer was constructed with a 800 nm Ti sapphire femtosecond laser with 80 MHz repetition rate, a 45 fs pulse width, and as much as 400 mW power; two commercial PCAs with 34 um and 6 um gap size were used as the emitter and receiver, and the system was characterized by its absolute radiated THz power, spectral bandwidth, signal-to-noise ratio, dynamic range, and beam profile at the focal plane (Zhang et al., 2014).

Detection is dominated by two classical schemes, electro-optic sampling (EOS) and photoconductive sampling (PCS). EOS transduces the THz field through Pockels-effect birefringence in crystals such as ZnTe, GaP, LiTaO$_3$, GaAs, and DAST, whereas PCS converts the local THz field into a transient photocurrent in a photoconductive gap (Zhao, 2023). In PCS the detected current obeys the convolution form

$$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$

so the instantaneous fidelity depends on the carrier response $N(t)$ (Zhao, 2023).

Modality	Transduction	Representative characteristics
EOS	THz-induced birefringence via the Pockels effect	Broad bandwidth; limited by phonon absorption and phase matching
PCS	THz-driven transient photocurrent in a PCA	Electrical readout; bandwidth limited by carrier lifetime and probe pulse width
THz-ABCD	Air plasma detection through a third-order nonlinear process	No crystal phonon absorption; useful range roughly $0.2$ to over $30$ THz

The practical performance of an emission spectrometer depends strongly on collection geometry. A broadband room-temperature system based on an LT-GaAs PCA emitter and ZnTe EOS detector used backward THz collection from the same side as the incident pump beam rather than forward transmission through the GaAs substrate. Because GaAs has a phonon resonance around 8 THz and strongly absorbs above about 4 THz in forward collection, backward collection eliminated substrate absorption and enabled a usable spectral range of 1–20 THz with 0.075 THz resolution (Shen et al., 2020).

3. Emission mechanisms and source physics

The physical meaning of the measured THz waveform depends on the emitting current source. In photoconductive emitters, the THz pulse is the radiated signature of an ultrafast photocurrent launched by carrier generation and acceleration in the biased gap (Shen et al., 2020). In spintronic emitters, the waveform encodes ultrafast spin-to-charge conversion. For 3d/5d bilayers such as Co/Pt, NiFe/Pt, NiFe/Au:W, and NiFe/Au:Ta, femtosecond excitation launches hot spin-polarized carriers from the ferromagnet into the adjacent metal, and the inverse spin Hall effect (ISHE) generates a transverse current,

$$\mathbf{j}_c \propto \boldsymbol{\sigma} \times \mathbf{j}_s,$$

with the emitted THz field proportional to the time derivative of the net charge current: $$E_{\mathrm{THz}} \propto \frac{\partial J_c(t)}{\partial t}.$$ In these systems, THz emission from Co/Pt and NiFe/Pt is strong, whereas NiFe/Au:W and NiFe/Au:Ta are about 20× weaker; the analysis attributes this not to the spin Hall angle alone but to conductivity, spin transmission, spin-mixing conductance, and spin-flip rates (Dang et al., 2020).

Time-domain emission spectroscopy is also used to discriminate among competing source mechanisms. In RuO$$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$0/Py bilayers, THz emission was analyzed in terms of ISHE, electrical anisotropic conductivity (EAC), and inverse altermagnetic spin-splitting effect (IASSE). The field-even component under magnetic reversal,

$$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$1

isolates the non-magnetic EAC contribution, while the remaining field-odd part is interpreted primarily as ISHE (Plouff et al., 2024). For (001), (100), (110), and (101) RuO$$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$2 orientations, the measured anisotropy was consistent with ISHE plus EAC, and the study reported no evidence for IASSE in either as-deposited or field-annealed samples. The (101) orientation was distinctive because the coherent superposition of ISHE and EAC made the emitted THz polarization tunable from linear to elliptical by rotating the external magnetic field (Plouff et al., 2024). This serves as an example of how waveform symmetry, polarization analysis, and crystallographic orientation can separate magnetic and non-magnetic emission channels.

4. Spatially resolved, on-chip, and vector-resolved implementations

A major extension of the field is the migration from free-space, diffraction-limited detection toward nanoscale, on-chip, and full-vector measurements. In terahertz scanning tunneling microscopy, THz-driven field emission sampling (TDFES) retrieves time-domain THz waveforms directly inside the STM tunnel junction. A strong Gate pulse creates a short nonlinear emission window, a weak near-field pulse modulates the emitted current, and delay scanning reconstructs the waveform from the linear response around the strong gate bias, provided $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$3 (Li et al., 2023). The gate pulse can create a local voltage of over 10 V and field-emission currents in the tens of microamperes range, while remaining narrow enough to act as a sampling gate; for $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$4, the fitted gate pulse has $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$5, and the measured waveform-peak image width on Au(111) is $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$6 (Li et al., 2023). The same platform also measured THz carrier-envelope phase in the tunnel junction by using a home-built THz CEP shifter with phase shifts of $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$7 (Li et al., 2023).

On-chip architectures address the free-space diffraction limit by guiding picosecond or THz transients directly through nanostructures. A monolithically integrated LT-GaAs / GaAs-AlGaAs heterostructure used photoconductive switches to inject and detect picosecond pulses in a high-mobility two-dimensional electron system, enabling on-chip spectroscopy of plasmons up to ~400 GHz and electrostatic tuning of a gate-defined cavity (Wu et al., 2015). In a separate cryogenic platform, semiconducting photoconductive switches integrated with coplanar stripline transmission lines measured the optical conductivity of a 7.5-$$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$8m wide NbN film across the superconducting transition with a bandwidth of 200 GHz to 750 GHz, in a sample smaller than 2% of the Rayleigh diffraction limit (Potts et al., 2023). These on-chip systems show that emission and detection can be localized to individual lateral nanostructures rather than ensemble-scale free-space spots.

Conventional EOS is ordinarily projection-based and detects one field component at a time. A recent tensor-resolved EOS scheme replaced the linearly polarized probe with a circularly polarized probe and used a polarization-sensitive camera with integrated micro-polarizers at $$E(\omega)=\int_{-\infty}^{\infty} E(t)e^{-i\omega t}\,dt,$$9 and $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$0, permitting direct single-shot retrieval of the full in-plane vector field $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$1 from one acquisition (Lafreniere-Greig et al., 12 Jun 2026). The method reconstructed linear, circular, azimuthal, and radial THz fields, measured a helicoidal field rotation rate of $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$2, extracted a quartz birefringence of about $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$3 at 1 THz, and characterized a broadband waveplate using $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$4 and $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$5, with a broadband region where $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$6 (Lafreniere-Greig et al., 12 Jun 2026). In a related imaging direction, Fourier synthetic aperture with multi-angle time-resolved field measurements was shown theoretically and numerically to reconstruct spatial and temporal features of complex inhomogeneous samples with subwavelength resolution (Kumar et al., 2024).

5. Acquisition strategies, noise models, and parameter estimation

The conventional step-scan delay line is accurate but slow. Several alternative acquisition formats therefore modify how the THz waveform is sampled. Single-shot EOS based on phase diversity reformulates chirped-pulse spectral decoding using two complementary polarization outputs with transfer functions

$\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$7

and reconstructs the waveform by maximum ratio combining,

$\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$8

This removes the classical $\Delta \nu = \frac{1}{T_{\mathrm{window}}}.$9 resolution-window penalty of single-channel spectral decoding; experimentally it recorded free-propagating THz pulses over about 20 ps with high fidelity (Roussel et al., 2020).

A reflective-echelon single-shot spectrometer spatially encoded delay across the gate beam using a 20 mm × 20 mm mirror with 1000 steps, 20 $_3$0m step width, and 5 $_3$1m step height, yielding a total stretched temporal window of about 33 ps and an effective time window of $_3$2 ps with $_3$3 GHz spectral resolution (II et al., 2016). That system was specifically developed for optical-pump/THz-probe measurements in pulsed magnetic fields up to 30 T. At the opposite end of the speed spectrum, an acousto-optic delay based on an AOPDF produced gap-free scanning over a 12.4 ps time window with an 11.3 fs step size, a waveform refresh rate of 36 kHz, and a normalized signal-to-noise ratio of $_3$4 (Urbanek et al., 2016). THz asynchronous optical sampling similarly replaces the mechanical stage with two lasers of repetition rates $_3$5 and $_3$6; the review example $_3$7 GHz and $_3$8 kHz yields spectral resolution around 1 GHz and acquisition time around 0.1 ms (Zhao, 2023).

As instruments became faster and more specialized, statistical modeling of the time trace became increasingly important. Maximum-likelihood parameter estimation in THz-TDS models the measured waveform as

$_3$9

with time-dependent noise amplitude

$$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$0

The three dominant noise sources are detector/electronics noise, laser power fluctuation noise, and timing jitter (Mohtashemi et al., 2020). In repeated-waveform experiments, the fitted amplitudes were $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$1, $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$2, and $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$3 (Mohtashemi et al., 2020). A related signal-estimation framework corrected sub-sampling delay drift, delay-line speed variation, amplitude fluctuations, and periodic sampling errors before averaging; in a dry-nitrogen experiment, the SNR improved from about $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$4 dB for the raw average to $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$5 dB after drift correction and to $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$6 dB after drift plus amplitude correction (Denakpo et al., 2024). These methods are particularly relevant to emission spectroscopy because they preserve waveform fidelity and provide covariance-aware fitting when the emitted signal itself is the quantity of interest.

6. Applications, interpretive limits, and current directions

Broadband emission-side THz-TDS is already used as a spectroscopic tool in chemistry and materials science. A room-temperature backward-collection photoconductive system resolved twenty vibrational modes of polycrystalline adenosine at 2.0, 2.2, 3.0, 3.4, 4.2, 4.5, 6.2, 6.7, 7.2, 8.6, 9.6, 10.5, 11.4, 12.4, 15.7, 16.2, 16.8, 17.7, 18.2, and 19.2 THz, with agreement against published FTIR, Raman, and NIS data (Shen et al., 2020). In a metrological direction, reflection-mode THz-TDS on doped 4H-SiC simultaneously determined the charge carrier density of epilayers and substrates in a single measurement over a range of about $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$7 to $$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$8, and mapped wafer-scale inhomogeneity with a 1 mm lateral step size (Hennig et al., 17 Jun 2025).

The same time-domain philosophy is also being coupled to algorithmic classification. A reflection-mode chemical imaging system with plasmonic nanoantenna-enhanced photoconductive generation and detection achieved a peak dynamic range of 96 dB and a detection bandwidth of 4.5 THz in 3 s, then classified pulse-wise reflections with deep neural networks (Jiang et al., 3 Dec 2025). Blind testing across eight chemicals yielded an average classification accuracy of 99.42% at the pixel level, and concealed-explosive testing under opaque paper yielded an average accuracy of 88.83% (Jiang et al., 3 Dec 2025). The key methodological point was that the classifier operated on individual reflected pulses of 13 ps duration rather than on the entire waveform Fourier transform, thereby reducing sensitivity to pulse overlap, thickness variation, and packaging geometry (Jiang et al., 3 Dec 2025).

Interpretively, the field has also become a site for mechanism testing rather than only for materials fingerprinting. The RuO$$J(t) = e \mu \int_{-\infty}^{\infty} E_{\mathrm{THz}}(t') \, N(t' - t) \, dt',$$9/Py study is exemplary: anisotropic THz emission had been linked to IASSE in prior work, but the time-domain, orientation-dependent, and field-reversal analysis found that the data were instead consistent with only ISHE and EAC, casting further doubt on altermagnetism in RuO$N(t)$0 (Plouff et al., 2024). This underscores a general feature of time-domain terahertz emission spectroscopy: because it measures the emitted field waveform, its polarization, its magnetic symmetry, and its crystallographic dependence, it can adjudicate between competing microscopic current-generation models rather than merely report that THz radiation is present.

Taken together, the literature shows a coherent trajectory. The core measurement remains the time-resolved THz electric field, but the instrumentation now spans photoconductive emitters, electro-optic and photoconductive detectors, single-shot and rapid-scan acquisition, on-chip transmission lines, STM junctions, vector-field cameras, and uncertainty-aware inference. This suggests that time-domain terahertz emission spectroscopy has evolved from a free-space waveform readout into a broader framework for ultrafast source diagnosis, nanoscale field sampling, vector polarimetry, and model-discriminating spectroscopy across condensed-matter, chemical, and device-oriented settings.