Gas puff imaging on the TCV tokamak

Passive imaging to study turbulence in plasmas has been employed since the early 1960s,¹ but it was only in the early 21st century that Gas Puff Imaging (GPI), as it is used today, was implemented as a diagnostic technique on Alcator C-Mod^2,3 and NSTX.^4,5 Today, it is routinely used in many devices to study plasma turbulence in the edge and Scrape-Off-Layer (SOL) of magnetic confinement devices.^6–10 The technique relies on the local injection of a small quantity of neutral gas into the volume to be probed [see Fig. 1(a)]. Mostly used to probe the edge and SOL region of tokamaks, the neutral gas becomes partially ionized by the plasma near the Last Closed Flux Surface (LCFS), where typical temperatures are in the range of 1–100 eV and electron densities range from 10¹⁸ m⁻³ to a few 10¹⁹ m⁻³. Local line emission in the visible spectrum, resulting from the electron impact excitation of the neutral gas cloud before its ionization, is collected along optical chords, which are tangential to the local magnetic field, providing a poloidal cut of the emissivity. Popular choices for the neutral gas are deuterium $D2$ and helium He, where the brightest visible spectral lines are the Balmer-α at 656 nm and the HeI triplet line at 587 nm.⁶ The measured brightness can be expressed as

where $n0(r⃗,t)$ is the local neutral density and $ne(r⃗,t)$ and $Te(r⃗,t)$ are the local electron density and temperature, respectively. f is proportional to the ratio of neutrals in the upper excited state of the radiative transition to the ground state and the lifetime of the decay.⁶ Measured line intensities thus reflect a combination of local electron density and temperature fluctuations that are generally assumed in phase. Disentangling the individual fluctuating terms is, however, not straightforward from GPI data alone. Neutral density fluctuations can be reasonably assumed slowly varying, both in space and in time, compared to the plasma turbulence and can be neglected over a large part of the SOL. The neutral mean-field model, can, however, break down well inside the LCFS, where $n0(r⃗,t)$ fluctuations can become relevant in determining the GPI signal.^11,12 For D₂ injections, additional molecular and dissociation effects must also be considered for quantitative analysis of the GPI line emission.¹³ 

Comparing the diagnostic to reciprocating or wall-embedded Langmuir probes, GPI provides 2D data with high spatial and temporal resolution primarily limited by the detector. This allows for the study of, e.g., turbulent filaments,^14–16 providing measurements of cross-field sizes and velocities, without requiring simplifying assumptions on the filament shape and velocity.¹⁷ Furthermore, the evolution and movement of individual filaments can be tracked and followed throughout the Field of View (FoV), which is not possible with probes as they generally rely on conditional averaging techniques.¹⁷ 

A novel feature of the Tokamak à Configuration Variable (TCV) is that multiple GPI systems can simultaneously collect data at several poloidal locations in the SOL (see Fig. 1). The versatility of this setup together with TCV’s exceptional shaping capabilities can be combined to obtain a wide poloidal coverage at several locations simultaneously: the Low-Field Side (LFS) upstream SOL, around the X-point region, on both the High-Field Side (HFS) and LFS, and the divertor region.

In this work, we first describe and characterize the GPI systems installed on TCV. In Sec. II, we present a few considerations made prior to the design of the hardware. In Secs. III and IV, the design and capabilities of the installed hardware are described. In Sec. V, a selection of first results obtained with the GPI systems on TCV are described and future possible upgrades are discussed before concluding in Sec. VI.

To study plasma dynamics without significantly perturbing them, the quantity of injected gas should be minimized, and the distribution of this gas should be optimized depending on the scope of the intended measurement and the capacity of the sensors. There are dependencies on the sensor itself; the geometry and optics of the system, generally characterized by an optical etendue; and the intensity losses due to the optical elements in the inline, such as lenses, vacuum windows, mirrors, and optical fibers. The quantity of light emitted by the neutral gas cloud depends on the local plasma density and temperature and the density of the injected neutral gas cloud.⁶ Furthermore, a compromise between the high acquisition frequencies of the order of MHz, required for the measurement of filaments with velocities of the order of 1 km/s, and the necessary Signal-to-Noise Ratio (SNR) is needed. Arrays of Avalanche Photodiodes (APDs) are often used for weak signals, despite the drawbacks of lower spatial resolutions (limited by the number of APDs), the need for more complex optics (which may include optical fibers), and temperature stabilization of the APDs. Fast acquisition Cameras (FCs) are also a popular choice as they provide a higher spatial resolution and simplicity of use although sensitivity acquisition frequency is lower than in APDs. In this section, we present some considerations and choices made during the diagnostic design.

The midplane GPI was the first to be installed at TCV. To ensure a usable signal, APD detectors were chosen due to their high sensitivity. With optical fibers of a diameter ϕ_f = 400 μm and a numerical aperture NA = 0.22, the etendue is estimated to be $E=(πNAϕf/2)2≈0.019mm2sr$⁠.

Neutral gas transport modeling with DEGAS2^18,19 was performed for typical SOL average temperature and density profiles of TCV L-mode plasmas to generate an estimation of the signal levels within the FoV. As in other systems,⁶ $D2$ and helium neutral gas were chosen. Deuterium has the advantage of having a low perturbative nature for a $D2$ majority plasma, and He, as there is little “background” (intrinsic) He line emission in TCV, yields a good SNR. Injection through four nozzles, vertically spaced by 1 cm, was considered in simulation to obtain a more homogeneous cloud in the poloidal plane. Two different nozzle shapes were investigated in this study: four conventional nozzles (simple orifices of 1 mm diameter) and four de Laval nozzles (such as those used in Ref. 20), which provide more collimated gas puffs. A more collimated puff would reduce the image-smearing from the toroidal spread of the gas cloud and increase the signal for a given total amount of injected gas. The de Laval nozzles were modeled with a half-angle spreading of 7°. Considering typical GPI injections^4,5 of 6 × 10²⁰ atoms/s on NSTX for plasma parameters similar to TCV, together with the volume ratio of NSTX and TCV (∼7),²¹ for simulations, we chose a flow rate of $∼6⋅1019$ atoms/s.

The de Laval nozzles showed ∼2x increased signal levels, compared to simple orifices. However, with an APD-based detection with a sensitivity of 0.1 mW/cm²sr, conventional nozzles also provided a usable level of brightness, corresponding to a SNR of ≳20, as shown in Fig. 2. In Fig. 2, we show the time-average brightness for both helium (a) and deuterium (b). It can be noted how helium yields a brighter signal, while a deuterium puff penetrates the plasma more, compared to a helium puff. In conclusion, considering the short distance between the nozzle and the main plasma on TCV (⁠$∼4−7$ cm) and the appreciable signals obtained in the simulations for both helium and deuterium, optimization of the gas collimator was not considered worthwhile. With a system similar to the one used in Alcator C-Mod, with Hamamatsu Photonics APDs and an etendue of $∼0.015mm2sr$⁠,²² we chose to install conventional nozzles that are significantly easier to implement and operate.

For the “X-point” (or Xpt) GPI systems, TCV’s recently installed baffles²³ provide a more closed divertor geometry that allows for the placement of nozzles through the baffle tips of the HFS and the LFS baffles. The baffle tip injections bring the gas source closer to the regions of interest, such as the near SOL, the separatrix, and the X-point, as can be seen not only in Fig. 1 but also in Figs. 3(a) and 3(c). A significant benefit of TCV’s shaping and positioning control coils may be used to displace a multitude of poloidal shapes at different positions, relative to the GPI diagnostic. The Xpt GPI can thus view above and below the magnetic X-point on the LFS or the common and private flux region of the inner divertor leg. The main design challenge of the Xpt GPI systems was to predict the line emission intensity and to identify the accessible FoV. In the proximity of the magnetic X-point and in the divertor volume, significantly more background light from deuterium is present due to increased wall recycling.²⁴ In comparison to the upstream SOL, the intrinsic D-α 656 nm background light is typically about two orders of magnitude stronger, as also seen on Alcator C-Mod.²⁵ We thus opted to use He as the locally puffed gas. Furthermore, as illustrated in Fig. 1, distances to the separatrix depend on the plasma geometry and can be significant, leading to a reduced toroidal localization of the gas and reduced spatial resolution. A simplified neutral fluid model, named the Zero Temperature model, has been implemented to evaluate the diagnostic feasibility and answer the previously mentioned challenges.

The motivation for the zero temperature model (compared to DEGAS2 in Sec. II A) is greater flexibility and faster computation times, suiting the evaluation of a larger range of plasma geometries. A detailed model description is given in  Appendix, where we also illustrate a bench-marking against DEGAS2 with a direct comparison of the midplane GPI. It showed good spatial and quantitative agreement for the neutral gas distribution. The resulting 587 nm line integrated brightness larger by two orders of magnitude while matching the 2D poloidal distribution well. The magnitude difference was traced back to a difference between openADAS²⁶ and Goto,²⁷ respectively, tabulated collisional–radiative cross sections, and Photon Emission Coefficients (PECs) used for the evaluation of the local emissivity (see the  Appendix).

For the X-point system, the zero temperature model is applied in Fig. 3 to two typical plasma geometries above (a)–(c) and below (d)–(f) the X-point, respectively. In both cases, the gas flux of Φ = 6 × 10¹⁹ at/s, the same as in the midplane simulation, was used with a single capillary nozzle. The model predicts a peak brightness of 4 mW/(cm²sr) above the X-point and 2.5 mW/(cm² sr) in the divertor. On a given flux surface, this represents a brightness reduction of 55% and 70% in these regions compared to zero temperature model predictions at the outboard midplane. Around the X-point, i.e., the region of normalized flux coordinate $ρΨ=(Ψ−Ψ0)/(ΨLCFS−Ψ0)∈[1,1.02]$ (Ψ₀ and Ψ_LCFS denote the poloidal magnetic flux on-axis and at the LCFS, respectively), the brightness decreases rapidly and appears too low for detection. This test provides a useful frame for the choice of the detector, as discussed further in Sec. IV B. The model predicts a large poloidal emissivity distribution of the order of 15 × 15 cm² and motivates a larger FoV of the sensor (black dashed lines in Fig. 3) compared to the midplane system.

The Full-Width-Half-Maximum (FWHM) of the Gaussian emissivity distribution S_eff is used as the parameter for the toroidal spreading and shows a strong dependence on the distance to the injection nozzle, taken with a gas divergence half-angle of typically 20°, very similar to experimental measurements for helium.²⁸ In regions with brightness $≥3$ mW/(cm² sr) above the X-point [Figs. 3(a)–3(c)], S_eff ≈ 11 cm, which is comparable to the midplane system. Closer to the separatrix and below the X-point (d)–(f), S_eff reaches up to 17 cm. The effect of the toroidal spreading on the spatial resolution is related to S_eff as Δx = S_eff tan θ, where θ is the misalignment angle between the GPI sight lines and the magnetic field line.⁶ This applies mostly in the poloidal direction and with θ ≈ 3–4°(as will be further discussed in Sec. IV C) results in Δx ≈ 0.8 cm for S_eff ∼ 11 cm but becomes significant for S_eff ∼ 17 cm with Δx ≈ 1.6 cm at the very edges of the view. This indicates the limiting structure sizes that can be detected with the system.

Overall, the model predicts that the proposed system would probe well the common-flux part of ρ_Ψ > 1.02 and, depending on the plasma geometry, also further inward. Obtaining sufficient brightness at the X-point itself is unlikely in the TCV baffle geometry shown in Figs. 1 and 3. A collimating nozzle, such as the de Laval nozzle, to reduce the gas spreading and spatial smearing in the baffle tips was not chosen because of a limiting versatility with plasma geometries and added construction complexity. The model has further been used to test the effect of the central axis orientation and the possibility of using multiple nozzles, similar to the midplane system. Three nozzles with angles of 0, ±15° to the horizontal direction generated a negligible extension of the poloidal distribution on the toroidal spreading though the toroidal spreading remains largely unchanged. Hence, such a solution shows little benefit in reducing the difficulty of spatial resolution, and the FoV of interest is largely covered by the single capillary nozzle.

A new gas injection scheme has been developed for both the midplane and the X-point GPI systems. The control is actuated independently and with internal feedback on the requested gas flow. They provide a controlled, stable neutral gas flow in the range of 10¹⁸–10²¹ particles/s. The valve control is implemented through a Beckhoff CX5120 embedded PC that accesses the immediate flow and controls the valve actuating voltage. An industry-oriented Programmable Logic Controller (PLC) offers several advantages over TCV’s legacy analog circuits: (a) A real-time OPC-based²⁹ EtherCat communication between the PLC and any other observer in the EtherCat network, such as the user, TCV’s state machine, and the MDSplus³⁰ data storage servers. This also allows us to modify the control scheme remotely without the need to substitute/redesign hardware components. (b) The modular design of PLC allows for and simplifies the integration of input/output analog components (such as further actuators for moving elements in the optics or the input of auxiliary control parameters) without having to modify the existing hardware/software. (c) The ability of monitoring the PLC parameters during operation to follow the evolution of the control states and variables in real-time. This facilitates debugging and allows for a rapid adaptation of the control parameters when optimizing the control.

In Fig. 4, we show the flow chart of the control system. This scheme, developed for all GPI systems, is becoming the standard for control at TCV and is being adopted for a number of other systems, such as gas injections and plasma heating.

Figure 5(a) illustrates the schematics of the injection hardware at the outboard midplane. The other GPI systems follow a similar design. From the left to the right, a 0.1 L volume is pre-filled with the desired gas (⁠$He$ or $D2$⁠) at ∼1 bar. It is isolated from the main source of gas with a manual valve (valve 1). This serves to limit the total amount of gas that can be injected into the vessel. Its volume is chosen such that after ∼100 typical puffs the pressure in the volume is reduced by less than 10% and that, should the valve become stuck open, the pressure rise can be handled by TCV’s turbo-molecular pumps. A piezo-electric valve (Key High Vacuum PEV-1), actuated by a voltage of up to 100 V, feeds gas into TCV. Downstream of the PEV-1, a pressure transducer (Kulite XCS-190-M), mounted in an ad-hoc developed and machined brass inset, measures the pressure at the outlet of the piezo-electric valve. Designed to be as close as possible to the outlet, this allows for measuring the instantaneous gas flux,^28,31,32 as discussed below. For the midplane system, the gas flows through a tube (∅ 1 mm, L = 35 cm; see Fig. 1) to the injection nozzle, where it feeds four orifices, each vertically spaced by 1 cm from the next. The Knudsen and Reynolds numbers of the system,³³ when using neutral helium or deuterium, are of the order of 10⁻³ and 10³, respectively.

For the two X-point systems (see Fig. 6), an outlet capillary tube (∅ 1 mm) takes the gas into the vessel through the tip of a baffle tile. The tube length from the piezo-valve to the injection location at the inner and outer baffle²¹ is $∼1.1$ and $∼1.25$ m, respectively. Both valves are located at the bottom of the vessel on the same port. The capillary stainless steel tubes (marked in blue) of 1 mm inner diameter are placed behind the graphite tiles, taking care to minimize 90° angles. Grooves in the TCV graphite tiles for the capillary tube prevent an electrical connection between tiles. At the injection locations, where the poloidal coordinates are listed in Table I, the capillary tube is inserted into the baffle tile and transitions to a horizontal duct of 1 mm diameter at a distance of 1 cm before the baffle tip.

During the plasma discharge preparation, the PLCs collect the gas request from MDSplus and await a TCV synchronous discharge trigger. At the first requested injection, the control systems operate an “unsticking” procedure: the valves receive a series of intermittent spikes in the voltage (0.5 kHz, 100 V) to open the piezo-valves, which tend to “stick” closed after being left dormant. When a pressure rise p_XCS is detected downstream by the XCS pressure transducer (typically after 2–3 pulses), the system switches to “run” mode, where the instantaneous flow Φ_M is deduced from p_XCS.^31,32 The relation between Φ_M and p_XCS was calibrated from the rate of increase in TCV’s pressure for constant values of p_XCS. The actuation of the valves thus happens on a measurement of the actual gas flow rather than a previous calibration of the valve voltage/gas flow relation. This makes the system independent of any change, e.g., due to temperature, or upstream pressure variations in the 0.1 L volume. The literature^31,32 reports that the relation between the pressure p at the inlet of a tube and its throughput flow Φ can be described by $Φ(p)=α[βpC+1]1/C−1$⁠. The latter recovers the analytical solutions for the choked flow (a.k.a. sonic orifice flow, Φ ∝ p) and Poiseuille flow (Φ ∝ p²) depending on the choice/fit of the parameter C. Our case (L_tube ∼ 0.1–1 m, ∅_tube ∼ 1 mm, and T ∼ room temperature) is in the transition region between the choked flow and Poiseuille flow. Third-order polynomials

were adopted to better fit the measured response. The fit parameters do, however, depend on the injected gas type, so a calibration for each of the used gases was performed.

From Φ_M(p_XCS), control can be actuated on either Φ_M or p_XCS. Assuming Φ_M is used for feedback control, at each time step k, the error ɛ_k = Φ_M − Φ_R is measured as the difference between the measured and requested flow. This is then minimized by a Proportional-Integral-Derivative (PID) control scheme:³⁴ Using the current value of ɛ_k, the voltage applied to the PEV-1 valve at the next time step k + 1 becomes

where ɛ_C is the cumulative error since the beginning of the control phase $(εC=∑j=0kεj)$⁠. Efficiency and stability of this correction procedure depend on the control parameters κ_p, κ_i, and κ_d that act to reduce the instantaneous, cumulative (integrated), and derivative error. The choice of the parameters can be estimated based on the system characteristics using control system techniques. For the midplane system, we obtain an optimum for κ_p = 0.6, κ_i = 0.09, and κ_d = 0.5.

Assuming He neutral gas at room temperature, the speed of sound (⁠$∼103$ m/s) yields a characteristic time (for a characteristic system size of L ∼ 1 m) of the order of $∼1$ ms. This is reflected in the overall response time of the system, as shown in the delay between the opening of the valve and the pressure rise (⁠$∼3−4$ ms) and between the pressure rise at the outlet of the valve and the GPI signal (⁠$∼3−4$ ms) and the length of pressure decay (⁠$∼50$ ms) in Fig. 5 (and Fig. 9). The control system runs at 2 kHz, which is largely sufficient to react to—and control—the gas flow. With the controller, the gas flow can thus be considered constant, and the requested and measured flows are stored in the MDS data server for the shot. Using such a controller, the gas flow may be considered approximately constant and the requested and measured flows are recorded in the MDS data server for every TCV discharge.

As experimentally verified by increasing/reducing the gas flow rate, in the range of 10¹⁹–10²⁰ at/s, the neutral gas density does not influence filament size and velocity estimates and may be, therefore, considered approximately non-perturbative for these measurements. As a compromise between sufficient brightness and a global plasma density perturbation, short gas puffs of typically 50–100 ms with a neutral gas flow in the range of 2 × 10¹⁹–6 × 10¹⁹ at/s, depending on the plasma density, were chosen. For these flow rates, no significant effect on the global plasma can be detected in density and temperature measurements, and the SNR is acceptable (see Sec. IV).

Due to the variety in geometry, plasma conditions, and requirements at the different poloidal locations, the midplane and X-point viewing and light collection systems differ substantially. Each is described in detail in terms of design, differences, and capabilities. The main properties of the midplane and X-point GPI systems described in this section are summarized and compared in Table II.

At the outer midplane, as shown in Figs. 1 and 7, light is collected approximately tangentially to the local magnetic field from a FoV of 5 × 4 cm². Two mirrors [mounted on an optical port shared with the Electron Cyclotron Emission (ECE) diagnostic^35,36] steer the collected light through two achromatic doublets that, after the passage through the port window of the vacuum vessel, is focused onto a rectangular array of 120 (12 × 10) optical fibers. The fibers have a core diameter ∅_f 400 μm and a numerical aperture NA = 0.22 and are of 8.3 m long.

A He glow, lasting several minutes, is performed on TCV between plasma discharges. To protect the mirror closest to the plasma (mirror 2 in Fig. 7) from damage, it is automatically retracted for this cleaning discharge and, in general, not deployed, unless GPI data are solicited. Although the mirror is even when deployed, in the tile shadow of the adjacent wall protection tile, it remains subject to wear from residual interaction with the plasma (e.g., arcing or plasma disruptions). It must be either polished or replaced regularly to retain performance, and this happens approximately once every year. Strong localized arcing damage (see Fig. 8) is not an immediate problem since the mirror is not at a focal point of the optical relay. Even strong local damage results, therefore, in a minor general loss of intensity at the detector with little spatial dependence in the FoV.

Four sets of 30 fibers each are held in a precisely machined coupling plate such that light from each fiber is incident onto one, and only one, detector element of the 12 × 10 APD array. This consists of four 32-element APD arrays (Hamamatsu S-8550, quantum efficiency $∼0.8$⁠), which are actively cooled to 15 °C. Between each fiber set and APD array, one of the four 2.0 mm thick interference filters serve to select the desired emission line. The interference filter used for D_α detection has a 658 nm central wavelength with a 10 nm bandpass. The HeI line emission is passed through a filter with a 587 nm central wavelength with a 10 nm bandpass. These custom-made, thin, filters were supplied by Andover Corp. This scheme is a 12 × 10 fiber adaptation of a 9 × 10 fiber version used on Alcator C-Mod.²² 

In the passage through the whole system (two mirrors, two achromatic doublets, the port window, the optical fibers, and the filters), only ≈34% of the collected light is lost. The effective etendue for each detector element/pixel is $∼0.0192mm2sr$⁠. After amplification, the signal is digitized at a sampling rate of 2 MHz by D-tacq ACQ132-32 CPCI digitizers (see Fig. 9).

The system is absolutely calibrated by illuminating with a known light source (a lab-sphere from GHz Optik) through the complete system (starting from mirror 1). This is performed approximately twice a year during TCV vacuum openings, which is used to monitor any deterioration in the system. Every year of operation has, to date, resulted in an ∼10% loss of sensitivity. This loss can be mostly ascribed to the wear of the second mirror that, as described above, is substituted every other year.

The adjustment of the FoV location and the spatial calibration is performed by back-illuminating the fibers. The optics focus the light onto a target in the vacuum chamber at the image plane in front of the gas nozzles. The spot size (⁠$∼4$ mm) and locations in which the lines-of-sight intercepts the focal plane (the ϕ = constant plane in front of the nozzles) are measured to within an uncertainty of 2–3 mm.

As the measurement of interest relies on the fluctuations in brightness resulting from the fluctuations in plasma density and temperature, the SNR can be defined as the ratio between typical signal excursions due to the passage of filaments during the gas puff and the standard deviation of the signal before the gas puff. Values in the range 30–150 are generally obtained depending on the plasma density and shape, the distance of the measurement from the nozzle, the injected neutral gas, and the neutral gas density. For typical measurements, helium yields a better signal (SNR ∼ 20–80) compared to deuterium (SNR ∼ 10–60) since for helium, the background deuterium plasma does not affect the measurement as much. The angle between the Lines Of Sight (LOS) and the local magnetic field (0 < θ < 8) has been minimized during the design of the diagnostic and is therefore usually neglected when operated in the correct plasma helicity. A discussion on the effect of the alignment of the LOS to the magnetic field lines can be found in Sec. IV C.

The error in filament size measurement due to the finite spatial resolution has been estimated by comparing the known and measured sizes of filaments in synthetic data. We estimated the sizes of a set of 2D elliptical Gaussians (10⁵ realizations for each size), randomly distributed and orientated in the field of view of GPI. For a synthetic filament comparable to filaments typically measured at TCV,^15,37 the size is, on average, underestimated by an error of <5% to <1% for small (10 ≲ δ ≲ 15 mm) and large (δ ≳ 25 mm) filaments, respectively. As typical filaments move with velocities ≲5 km/s, the acquisition frequency of 2 MHz is largely sufficient to resolve their dynamics. In Fig. 10, an example sequence is shown, where every sixth–eighth image of the normalized GPI signal $S(x⃗,t)=(I(x⃗,t)−<I(x⃗)>1ms)/(<I(x⃗)>1ms)$ is plotted. $<I(x⃗)>1ms$ is the moving average for a 1 ms window, computed for each channel. In Fig. 10, we show the major axes (red solid cross) of the ellipse fits on the 2σ contour level (red dotted line). σ is here the average standard deviation of $S(x⃗,t)$ for the views outside the LCFS. The poloidal cross-field size of the filaments can be estimated, for example, by measuring the maximal size of the FWHM contour in the direction of the poloidal magnetic field (turquoise thick line) or the vertical extension of the FWHM contour passing through the center of the ellipse fit (vertical black line). As can be seen in Fig. 10, the typical dynamics, shapes, and sizes of filaments are well resolved by the diagnostic.

For the X-point GPI systems, a magnetically shielded Phantom v2012 FAST high speed (400 kHz, exposure time 2.1 µs at a resolution of 126 × 96 px²) was selected for its large pixel size of 28 × 28 μm², giving better low light performance and the short intra-image readout time of 0.3 µs. The camera was mounted onto the Multispectral Advanced Narrowband Tokamak Imaging System (MANTIS),³⁸ as shown in Fig. 11, installed on the lower tangential port of sector 8 of TCV. The light is relayed from the port by several mirrors and concentric lens-based relay optics (90% transmission) to a HeI 587.5 nm filter (2 nm bandwidth, 95% transmission) in front of the camera. This allows for the camera to be placed farther from TCV (⁠$∼2$ m), reducing the magnetic field influence yet avoiding the poor spatial resolution of an equivalent fiber optic relay system. This solution allows for the continuous operation of MANTIS while using only one of its 10 spectral channels for fast imaging of HeI line emission and the possible later addition of more high-speed cameras recording other spectral lines. This could include employing helium line ratio techniques to obtain insights into 2D filament densities and temperatures.

A significant advantage is the adaptability of the Region Of Interest (ROI) that, for a CMOS camera, is related to the acquisition frequency and the number of active pixels on the sensor. Hence, this allows for the observation of the full divertor chamber, as illustrated in Fig. 11, at up to 155 kHz and with a resolution of 384 × 256. Without external helium injection, the maximum usable acquisition frequency is limited by the available background light to ∼25 kHz. However, if the HeI filter is removed, high light levels are obtained and unfiltered, fast imaging of the full divertor is attainable, as shown in Fig. 11(c). An absolute spectral calibration provides absolute emissivity measurements.

An 85 mm lens with a numerical aperture NA = 1.4 has been used. A precise focus on the gas puff position is dictated by the shallow depth of field. Each camera pixel collects light from a 1.9 × 1.9 mm² region in the poloidal plane intercepting the injection position with an estimated étendue of $≈2⋅10−4$ mm² sr (⁠$∼1.06$ m between the emission cloud and the pupil position). Due to flux expansion in the region around the X-point and divertor, filaments may be expected significantly stretched compared to observations at the outboard midplane that motivated a bigger FoV and higher number of pixels than with APDs.

The camera was rotated by 1.3° and 0.8° with respect to the optical axis of MANTIS in vertical and horizontal directions for the camera FoV viewing closer to the injection location and better matching the cloud extent. The misalignment between the LOS and the magnetic field lines is typically 5° and 3° in the poloidal and toroidal directions, respectively, for standard helicity (I_p same direction as B_Φ), as will be further discussed in Sec. IV C. The toroidal extent of the gas cloud along the magnetic field lines, estimated using the zero-temperature model (Sec. II B), limits the resolution to between 1 and 10 mm depending on the distance to the nozzle. Smearing due to the integration time of 2 µs can add up to a further 2 mm to this limit for cross-field velocities of 1 km/s such that a total precision is estimated to be typically 7 mm. The camera’s view in the poloidal plane is modeled using calcam ray-tracing³⁹ and typically results in an uncertainty of 2 mm for the position of each Lines Of Sight (LOS).

In the diagnostic setup described above, the upper part of the FoV looks at the underside of the LFS baffle such that the curved area in the top right of Fig. 11(d) is excluded. The corresponding pixels should not be employed. Pixels with low SNR, typically due to the neutral gas cloud distribution, are also excluded [left and bottom right in Fig. 11(d)]. As is common practice, images recorded on the camera are horizontally mirrored such that the LFS region is located to the right of the images [Figs. 11(c) and 11(d)].

The HFS system uses the same camera and cannot be used simultaneously with the LFS Xpt system. Here, the camera angles to the central optical axis are modified such that the FoV is shifted by 15 cm toward the HFS. Due to the smaller neutral gas cloud distribution in this region, the camera ROI can be further decreased (see also Fig. 1). This could also enable increased acquisition rates, however, with the cost of shorter exposure times and lower signal levels.

We now look further into the effect of the alignment between the GPI’s LOS and the magnetic field at the plane of the FoV. A misalignment in toroidal and poloidal directions can be differentiated, as illustrated for the midplane system for a typical plasma discharge in Fig. 12. When operated in the appropriate helicity for which the different systems were designed, i.e., non-standard (I_p and B_Φ in the opposite direction) for the midplane system and standard (I_p and B_Φ in the same direction) for the X-point system, the misalignment of the systems is less than 5°.

The effect of helicity on GPI measurements is shown in Fig. 13, where images of Conditional Average Sampled (CAS) filaments^15,40 are analyzed in identical L-mode plasma discharges except for the plasma current direction. In CAS,⁴⁰ each view of the 2D GPI window is averaged over all times at which the signal of a single, selected view (the trigger) exceeds a certain threshold (2.5 standard deviations in this case). This way, one gets the average shape of the structures passing through the selected trigger view. The standard helicity is shown in (a) and (c) with the other helicity in (b) and (d). Hence, the optimized case for the midplane system is (b) and (c) for the Xpt system. The long dimension ɛ_l of the filaments is over-estimated by ∼50% in the non-optimized helicity. In both cases, the filament is nearly aligned with the flux surfaces at the outboard midplane. However, above the X-point, the tilt angle of the filament is altered and stretched vertically. Furthermore, the smearing of the filaments results in a 25% lower peak brightness in both systems.

This effect can provide a rough estimation of the toroidal gas cloud extent at the measurement location using the pitch angle difference between standard and non-standard helicity of $≈10$° at the outboard midplane and $≈5$° at the LFS X-point location. This yields estimates of the half-width toroidal extent of 3.4 and 5.7 cm, respectively, and agrees well with the estimate from Sec. II B.

Comparing identical discharges performed once in long and once in short outer baffle TCV configurations²¹ allows for assessing the impact of distance of the nozzle to the measurement region on the Xpt system observations. The short outer baffles are 5 cm shorter, resulting in an equivalently longer distance from the nozzle to the separatrix (with the same plasma shape and location). In the case of the plasma shown in Fig. 13(c) with short outer baffles, the filament long edge length is reduced by another 5 mm and the CAS filament intensity increases by 40% for the longer baffles. Minimizing the distance between the nozzle and the region of interest is important, and quantitative studies of intensity and sizes, as well as the future design of future gas injection systems for GPI, should take these effects into account.

Data from the GPI systems at TCV have already been essential for several scientific findings at TCV. These include the discovery that SOL filamentary turbulence is strongly suppressed in the upstream LFS SOL for highly negative triangularities of Han et al.,¹⁴ the identification of the propagation regime of filaments and their parallel extension into the divertor,¹⁵ and the existence of divertor-localized filaments.¹⁶ 

Ongoing research includes the development of a machine learning algorithm for tracking filaments,⁴¹ studies on the effect of divertor geometry on upstream turbulence, a study on the origin and parallel extension of divertor-localized filaments, and an estimate of the contribution of these filaments to the global cross-field transport for different conditions. For many of these, the availability of GPI measurements at several poloidal locations in the SOL is essential.

Upgrades of the diagnostic currently under development include the implementation of a multi-spectral GPI acquisition system for helium line ratio methods¹⁰ for density and temperature determination (currently at the proof-of-principle phase for TCV). The installation of other nozzle configurations (in particular, de Laval nozzles) is also investigated, which would reduce the spread of the neutral gas cloud in the X-point systems and thus improve the SNR and spatial resolution in that region. In the foreseeable future, as fast cameras increase in their sensitivity and resolution, all GPI acquisition will likely switch from APDs to more user-friendly and flexible fast camera systems.

In this work, we presented the design and operation of the gas puff imaging diagnostics at TCV. Particular attention was given to pre-design studies and a novel gas injection system.

In combination with TCV’s shaping capabilities, the GPI systems can simultaneously probe and explore several regions of the plasma. These include the outboard midplane LFS SOL, the X-point region (both low-field and high-field side), and the divertor leg region. Data is acquired at spatial (sub-cm) and temporal (∼μs) resolutions that suffice to resolve the filament dynamics. Filaments at TCV can thus be studied in detail and with little ambiguity, allowing for the determination of key quantities, such as their size, shape, velocity, parallel extension, and frequency. In terms of SOL and divertor studies, for both experiments and theory/code validation, the GPI systems have already provided new perspectives and possibilities.

First scientific results have also been, albeit briefly, described, and upgrades to the diagnostics, both currently being developed and planned for the future, have been listed.

The authors wish to thank Dr. Federico Pesamosca and Dr. Federico Felici for their help in the development of controller software, the personnel of the mechanical workshops for their work on the hardware, Dr. Josh Stillerman for invaluable help in the acquisition system, and the whole staff personnel at EPFL-SPC, MIT-PSFC, and the TCV team for making this work possible. This work was supported, in part, by the Swiss National Science Foundation. This work was also supported, in part, by the US Department of Energy under Award Nos. DE-SC0020327 and DE-SC0010529. This work was carried out within the framework of the EUROfusion Consortium, funded by the European Union via the Euratom Research and Training Programme (Grant Agreement No. 101052200—EUROfusion). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Commission. Neither the European Union nor the European Commission can be held responsible for them.

The authors have no conflicts to disclose.

N.O. and C.W. contributed equally to this work.

N. Offeddu: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). C. Wüthrich: Conceptualization (lead); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). W. Han: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Software (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). C. Theiler: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). T. Golfinopoulos: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). J. L. Terry: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). E. Marmar: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). C. Galperti: Conceptualization (equal); Investigation (equal); Methodology (equal); Software (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). Y. Andrebe: Conceptualization (equal); Methodology (equal); Software (equal); Supervision (equal); Writing – review & editing (equal). B. P. Duval: Conceptualization (equal); Methodology (equal); Supervision (equal); Writing – review & editing (equal). R. Bertizzolo: Conceptualization (equal); Methodology (equal); Software (equal); Supervision (equal); Writing – review & editing (equal). A. Clement: Conceptualization (equal); Methodology (equal). O. Février: Data curation (equal); Formal analysis (equal); Supervision (equal); Writing – review & editing (equal). H. Elaian: Conceptualization (equal); Methodology (equal). D. Gönczy: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Software (equal). J. D. Landis: Conceptualization (equal); Methodology (equal). TCV team: Conceptualization (equal); Data curation (equal); Validation (equal).

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

For the evaluation of GPI diagnostic capabilities of the X-point system, a simplified fluid model to simulate a helium thermal gas puff and the associated HeI line emission, called the Zero-Temperature model, has been developed. As the name indicates, it does not correspond to a kinetic model with a velocity distribution. The injected particles are rather assumed to all be at thermal sound speed, i.e., $vns⃗=γkBTmner̂$⁠, with k_B being the Boltzmann constant, T = 298 K being the gas ambient temperature, m_n being the neutral particle mass, and γ = c_p/c_v being the isentropic expansion coefficient dependent on the puffed gas species (γ = 5/3 for He). This assumption on the neutral velocity appears reasonable when looking at distributions used in kinetic codes, such as DEGAS2.^18,19

The evolution of helium neutrals is computed using a fluid continuity equation approach in steady-state conditions. As opposed to D₂, no dissociation effects need to be considered for helium. Solely the ionization of neutral helium through electron impact is considered as a neutral helium sink (S_ion). Recombination and charge-exchange reactions have been neglected due to the small reaction cross sections.⁴² Furthermore, also neutral–neutral collisions and elastic scattering with plasma electrons and, last, the re-absorption of light emission by the plasma along the LOS are neglected. The latter is reasonable for the 587 nm HeI line emission in typical SOL conditions.⁴² The model evaluates the stationary continuity equation $∇⃗⋅Γn⃗(x⃗,t)=−Sion(x⃗,t)$⁠, with $Γn⃗$ being the flux of the neutral density, along each line of simulation in a cone, as shown in Fig. 14. This equation in spherical coordinates, $Γn⃗‖er̂$⁠, becomes

where $nn(x⃗,t)$ is the neutral particle density and ⟨σ_ionv_e⟩ is the local ionization cross section as a function of electron density n_e and temperature T_e. The evaluation in spherical coordinates allows us to analytically solve the neutral equation along each (θ, ϕ)-line, as shown in Eq. (A2). The simulation cone is typically defined in a domain θ ∈ [0, θ₀], with θ₀ = 89° being the beam divergence angle,

The boundary condition n_n,GP is defined on the spherical surface at r₀ = w₀/sin(θ₀) just outside the nozzle to avoid the r = 0 singularity, where w₀ is the width of the nozzle outlet. We evaluate the Knudsen number of the gas flow at the nozzle outlet to be K_n = λ/d ≈ 0.05, with λ being the particle mean free path in the tube and d = 1 mm being the inner tube diameter.⁴³ Hence, we approximate the boundary condition by a viscous flow and a cosine law angular distribution as the boundary condition⁴⁴ has been implemented. This latter can be written, for a specified neutral gas flow Φ [part./s], as

After solving the neutral density distribution, the coordinate system is transformed to Cartesian machine coordinates (R, Z, Y), with Y being the toroidal direction, because it simplifies greatly the analysis hereafter. The emissivity ɛ (or emission intensity) [photons/(cm³ s sr)] can be locally obtained in the machine dimensions as

where PEC is the Photon Emission Coefficient of excitation of neutral helium. The different cross-sections both for ionization and PEC for the helium triplet line (unresolved dataset pec96#he_pju#he0.dat) are taken from openADAS²⁷. The model requires only information on the local, time averaged electron temperature and density as input that can be provided through reciprocating probe or Thomson scattering data from the experiment, in 1D or 2D, and a toroidal symmetry is used for reconstructing the 3D information. Alternatively, plasma parameters can be input from 2D or 3D simulations.

Finally, the emissivity is integrated along the LOS of the detector to obtain the brightness I [mW/(cm² sr)] as follows:

The LOS integral is performed over the toroidal coordinate Y, assuming that the camera sight-lines are perfectly tangential. In this case, the integral over the LOS can be approximated as the product of the emissivity in the injection plane times the FWHM $Seff$ of the Gaussian shape of the gas cloud in toroidal direction.

A direct benchmarking to the kinetic neutral model DEGAS2 allows us to evaluate the model performance. An identical plasma background and neutral helium flux as in Fig. 2 were chosen to simulate the midplane GPI setup with the zero-T model to directly compare with the DEGAS2 simulations in Sec. II A. As a result, Fig. 15(a) shows a very good agreement between the two models for the neutral helium density along the central puffing axis. The quantitative difference represents at most 20% in the SOL. Starting at 1 cm inside the separatrix, differences start to occur presumably because of the fluid approximation breaking down. The emissivity in Fig. 15(b) also shows a good agreement between the models when the PECs are taken from the Goto model,²⁷ which is also what is used in DEGAS2. An emissivity larger by a factor $∼2$ is observed if the openADAS database²⁶ is used instead for the calculation of the HeI triplet line emission as a function of neutral helium density and the plasma parameters.

For a direct comparison and for illustration of the differences in the CRM database for the HeI line emission, Fig. 16 shows the 2D (LOS integrated) brightness using the Goto database for the PECs. This result can be directly compared to Fig. 2(a) provided by DEGAS2 and displays a good agreement for both the spatial distribution of the brightness and the magnitude.

In this version of the model, diagnostic errors due to the LOS misalignment are not taken into account. However, this can be included using a rotation of the gas cloud with respect to the toroidal axis corresponding to the misalignment between the sight-lines and the magnetic field lines. This would be of particular interest for the usage of time-resolved plasma turbulence information from f.ex. GBS real size turbulence simulations.⁴⁵ Future improvements of this model could include a kinetic representation of the injected neutrals (i.e., a velocity distribution) and the consideration of reaction decay time scales of photon emission⁴² for time-resolved applications. Simulating the HFS Xpt GPI would require including re-absorption of HeI light by the plasma since the LOS pass through the plasma for a significant distance.