Sub-divertor fuel isotopic content detection limit for JET and its impact on ICRF core heating and DTE2 operation^a

The importance of resolving isotopic concentrations at the ~1% level during plasmas with ICRH can be seen in figure 1. When the hydrogen concentration is reduced from 2.5% to below 1% (here based on divertor spectroscopy) in high NBI power H-mode discharges, second harmonic deuterium ICRF power absorption becomes dominant over the fundamental hydrogen minority one [6, 7]. In such conditions, deuterium ions (both from the thermal background and in particular from the injected NBI sub-population) are accelerated to high energies, as seen by the neutral particle analyser (NPA) measurements for deuterium at E  =  300 keV. These fast D⁺  ions not only lead to a considerable enhancement of the neutron yield with respect to the standard H minority case with similar Ti (not shown) but, in combination with the fast H⁺  ions (that are also accelerated to significant energies due to their small fraction) cause full stabilization of the sawtooth activity during the high ion cyclotron resonance heating (ICRH) power phase in the given example (JET pulse 89192). Note that the total radiation is similar in the two cases showing that equivalent high-Z impurity screening is achieved in both discharges.

Zoom In Zoom Out Reset image size

In related studies, a series of similar H-mode discharges in JET-ILW with varying H injection, varying the H/(H  +  D) ratio from 23% down to 2%, exhibited a correspondingly continuous increase in the core T_e, confirming that electron heating (after collisional redistribution) is maximized for low H minority concentrations (~2%). Since electron heating with ICRH has proven beneficial for high-Z impurity control in JET-ILW, the core W content (as measured by soft x-ray) is also reduced at the lowest hydrogen levels [7, 8].

The above examples would imply that the ability to measure, and ultimately control, the fuel isotopic content down to the 1% concentration level is important for optimizing the performance of a given ICRH scheme in fusion devices. This is particularly important when using ³He minority heating or for inverted ICRH heating scenarios, where a small overshoot of the minority concentration can have a strong impact on the wave absorptivity and on the fate of the RF power absorbed in the plasma [9]. Similarly, the so-called H–(³He)–D three-ion scheme needs fine control of the minority ³He ions' concentration at very low levels (<1%) to be efficient [10].

From a DT diagnostic standpoint capability, this case is also of interest regarding the present upgrade of the JET sub-divertor gas analysis diagnostic (to be briefly described in section 4). In addition to improvements in the fuel isotopic content measurement, the upgrade will also provide for potential He detection down to a 0.5% level, thanks to using spectrally resolved measurement for the He I emission, previously done with a filtered detector approach [11]. This detection limit for He is compatible with estimates of anticipated ⁴He concentration from DT-produced alpha particles.

The ability to have a fusion fuel isotopic content detection limit at or even below the 1%-conc level is also of great importance to the envisioned pre-DTE2 (TT) and post-DTE2 (DD) campaigns, in limiting the D/(D+T) content for the former, the T/(D  +  T) in the latter, thus effectively minimizing fusion neutron production for these phases [12]. JET has contractual limits on the vessel activation due to the total 14 MeV neutron fluence over the lifetime of JET (2  ×  10²¹ 14 MeV neutrons, of which 0.3  ×  10²¹ have been used in DTE1). The 14 MeV neutron production is measured for each pulse and the vessel activation is computed for the vessel composition. The in-vessel activation levels are measured during shutdowns. Thus, each campaign at JET has a 14 MeV budget, staying inside the total 14 MeV limit for the lifetime of JET. In this scheme, pre-campaign estimates are based on having (worst case) 1% D/(D+T) for TT campaigns, providing an estimate for the number of high power pulses that can be done during the campaign. For DT campaigns, the pre-campaign estimates use 50:50 DT concentrations for neutron estimates. Thus, the significance of the low end, fuel isotope detectability is most relevant in the TT and DD campaigns and, in fact, this is an area where a lower detection limit than 1% could be of value.

In the present study, the DTE1 OGA data [4] were revisited with focus on the uncertainty of the measurement on the low concentration end. The data plotted in figure 2 are the same OGA data plotted, e.g. in figure 5 of [4], however with the error bars now added and displayed with the data. The latter are from the error analysis carried out as part of the original fitting of the spectral profiles, which produced the isotopic concentration data for [4] and has since continued to automatically process data for this diagnostic.

Zoom In Zoom Out Reset image size

In the automatic analysis, the error bar on the isotopic ratio is derived from the error bars on fit parameters corresponding to the H and D spectral intensities, which are taken to be the square root of the diagonal elements of the covariance matrix, as is standard for χ² fitting. The signal on individual pixels is weighted by the inverse of the square root of the signal. It can be seen in the present figure that as the 1% T/(H+D+T) tritium concentration level is approached, the uncertainty in the measured concentration approaches 100% of the measurement value (figure 2). This suggests a 1% lower limit to the isotopic content measurement with this technique, assuming that the error bar calculated with this method corresponds to an interval of confidence large enough for the measurement to be reliable (see discussion in section 4.2).

Insight to the causes of the substantial relative uncertainty in the low concentration data can be most simply gained by observation of an actual, optical spectrum from this set of data. Figure 1(b) shows Penning-excited optical spectrum from the DTE1 published set, in this case for ~11.5% T and ~2.6% H concentrations, to illustrate the challenge of extracting the low concentration of these isotopes in a deuterium background. Firstly, it is noted that the overall intensity level (in number of CCD detector counts) is quite small. The D_α peak minus the continuum is barely above 100 counts. The saturation limit of the CCD detector was 32 000 counts. Thus, the full dynamic range was far from being utilized. It has been found relatively recently that the inherent property of Penning- (and the related magnetron-) type discharges to produce thin film coatings greatly contributed to low spectral line intensity levels. The Penning plasma discharge, which is in itself a relatively low emission source, can contribute to a progressive reduction in the signal by depositing carbon-based coatings onto the viewport window, through which its plasma light emission is collected. That, in turn, reduces light transmission through the window and therefore the amount of light that can be collected for spectroscopic analysis. This degradation of the S/N then directly impacts the detection limit, as will be seen.

Furthermore, this reduced light level can affect the detection limit of the tritium detection in a more complex way that the hydrogen detection. With state-of-the-art, passive optical spectroscopy used for DTE1, one can see in figure 2(b) that the H minority species is spectrally well separated from D. As such, it seems to be affected primarily by the relatively low spectral line signal over the random noise. On the other hand, the close spacing of the D and T components means that the low S/N can also compound the error of the non-linear fitting, which is necessary to distinguish the T from the D component (as also seen in figure 2(b)).

It is worth noting that a similar (~1%) isotopic detection limit was concluded in a recent, laboratory-based study, using H as a proxy for T in H/D mixtures, for an equivalent sub-divertor gas analysis system designed and prototyped for ITER [5]. The system included the same plasma excitation source as was used in DTE1 and until presently (i.e. the now obsolete, commercial, Alcatel CF2P Penning cold-cathode gauge) albeit with a different differential pumping arrangement with respect to the main analysis region. The fact this limit persisted, even though much better S/N levels could be obtained in the laboratory study with large integration times (>~1 s) suggested that additional mechanisms might also be playing a role in preventing H detection below the 1% concentration level. More indications of this will be seen in the next section, examining a more recent JET dataset with low concentrations of H in mainly D plasmas.

It is finally noted that the mentioned, historically used, automatic fitting program basically fits the spectral profile of each isotope Balmer-α line (i.e. H_α 6562.79 Å, D_α 6561.03 Å, and T_α 6560.44 Å) to an empirical model using two (nominally co-centric) Gaussian profiles [13]. Although the model has been interpreted as corresponding to the emission from both ('cold') atomic species and from ('hot') molecular dissociation species, no actual atomic or molecular physics is involved in the fitting of the lines. An ongoing effort, in parallel to the present study, is attempting to further improve spectral fitting by introducing atomic and molecular physics into the fitting program [14].

The present study also examined measurement uncertainty in a large set of data from more recent experimental campaigns, particularly focusing on comparison of the OGA data to those from divertor spectroscopy, i.e. measurement directly in the divertor plasma (with the local plasma providing in this case the excitation source for the emission). This dataset was taken from a database specific to ICRH physics studies. The database contains 1254 measurements with a hydrogen concentration varying from 0.75% to 14.75%. Each one is averaged over 1 second (or four spectral scans of 0.250 s integration time) and taken during the quasi-steady state in each discharge (both in terms of the isotopic composition and the plasma parameters, including the ICRH heating power).

In figure 3, ${{\left[{\rm H} \right]}_{{\rm Spec}}}$ is the hydrogen concentration $\frac{\left[{\rm H} \right]}{\left[{\rm H} \right]+\left[{\rm D} \right]+\left[{\rm T} \right]}$ as measured by divertor spectroscopy and ${{\left[{\rm H} \right]}_{{\rm OGA}}}$ is the same quantity as measured by the Penning-OGA. The figure shows the relative difference between the two measurements of the hydrogen concentration: $\frac{\delta \left[{\rm H} \right]}{\left[{\rm H} \right]}=\frac{{{\left[{\rm H} \right]}_{{\rm OGA}}}-{{\left[{\rm H} \right]}_{{\rm Spec}}}}{{{\left[{\rm H} \right]}_{{\rm OGA}}}}$, as a function of ${{\left[{\rm H} \right]}_{{\rm OGA}}}$, for low hydrogen concentrations (<15%). The full set of data from both diagnostics, plotted together with the error bars, as found in the processed data files, can be found in [15].

Zoom In Zoom Out Reset image size

Figure 4 shows, over the same data set and again in relative values, the uncertainty in the hydrogen concentration measurement specifically by the Penning-OGA: $\frac{\varepsilon \left[{\rm H} \right]}{\left[{\rm H} \right]}$. As with figure 2, the uncertainty is taken directly from the error bars produced by the standard JET automatic fitting program, which computes the isotopic ratios from the spectral profiles, and are stored in the JET processed data files.

Zoom In Zoom Out Reset image size

It is observed that $\frac{\varepsilon \left[{\rm H} \right]}{\left[{\rm H} \right]}\propto \frac{1}{\left[{\rm H} \right]}$ $({\rm which}\,{\rm implies}\,{\rm that}\,\varepsilon \left[H \right]\sim$${\rm Constant}$). That is to be expected, since the uncertainty (absolute value) of the spectral line fit, from which in turn are deduced the concentrations of the different species, is (for a given spectral resolution) essentially dependent on the S/N ratio; the latter in turn depends on the overall intensity of the fitted spectrum, but not on relative concentrations of the isotopic components. It is seen on this figure that $\frac{\varepsilon \left[{\rm H} \right]}{\left[{\rm H} \right]}\sim 1$ ${\rm for}~{{\left[{\rm H} \right]}_{{\rm OGA}}}\sim 0.01$. This is like in the case of the DTE1 T concentration, where the error bar in the T concentration was seen to exhibit a ~1%-concentration uncertainty in the measurement as the 1% concentration level is approached.

On the one hand, the similarity of apparent detection limits is not surprising, given that the same plasma source is used for the excitation of the measured species and the same spectrometer and detector of the recording of the spectra. On the other hand, a study of the spectrometer settings for this latter set of post-DTE1 data shows that the entrance slit of the spectrometer was increased by a factor of four, from 50 to 200 µm, to offset additional loss of light due to deterioration of window transmission. As the (likely to be mostly) amorphous carbon coatings had partially formed during DTE1, replacement of the windows was not trivial, and the replacement has only recently taken place as part of the upgrade discussed in section 4. What is important to note is that, with a 200 µm slit, the D and T components of the Balmer-α line would no longer be resolved (either by the present, high-resolution spectrometer or by any state-of-the-art upgrade). While this adjustment allowed for sufficient signal levels for resolving the hydrogen (proportionally increasing the amount of light admitted into the spectrometer from a ~1000 µm diameter optical fiber) it would have been detrimental to resolving the tritium in deuterium plasmas (with instrumental function width increasing roughly in proportion to the slit width above the ~50 µm setting). Therefore, window replacement and, most importantly, a window coating mitigation scheme were indeed critical system upgrade needs ahead of DTE1.

Looking back at figure 3 and taking the divertor spectroscopy measurements as the reference values for the isotopic content, one would expect that the relative difference $\frac{\delta \left[{\rm H} \right]}{\left[{\rm H} \right]}$ should be roughly bound by $\left\langle \frac{\delta \left[{\rm H} \right]}{\left[{\rm H} \right]} \right\rangle \pm \left\langle \frac{\varepsilon \left[{\rm H} \right]}{\left[{\rm H} \right]} \right\rangle$ which is what the green, dashed lines indicate in figure 3. As a caveat, it is also noted that the divertor spectroscopy is sensitive to the strike point position which could also be contributing to the spread in the data shown in figure 3.

The average value $\left\langle \frac{\delta \left[{\rm H} \right]}{\left[{\rm H} \right]} \right\rangle \sim 0.47$ (the blue solid line in figure 3) suggests a 'systematic' effect, in addition to the random error due mostly to S/N limitations. For the higher isotopic concentrations, this ('systematic') relative difference becomes insignificant and the Penning-OGA measurement tends to be in good agreement with main plasma diagnostics, especially after plasma-wall equilibration, as discussed in section 5. This difference, which has most impact at the low isotopic contents of interest in the present study, is presently attributed to isotopic exchanges processes with the metallic surfaces of the Penning cold-cathode gauges, which is used as the emission source for this spectroscopic technique (see e.g. discussion on surface interaction in the plasma source in [5]). Direct, laboratory studies of these interactions, using the same plasma sources and deploying baking under controlled conditions, as well as a model to describe these processes, are presently under way [16].

On the other hand, the random error is manifestly due to limitations in light emission from the OGA source and light losses from source to detection and can only be reduced by improved light emission or collection and by the mitigation of light transmission losses. As will be discussed later in the paper, an upgraded, multi-sensor (including shielded mass spectrometers) gas analysis diagnostic for DTE2 is aiming to reduce light losses for the OGA part of the system, partly by mitigation of window coatings produced by the Penning plasma source. Divertor species-concentration analysis—and especially the OGA part - remains a critical fuel isotope diagnostic as JET DTE2 is approached. It is also likely the main way to diagnose and control the isotopic concentration in ITER, especially at these low concentration levels. This is due to the increased challenge of fitting divertor spectroscopy spectra, in the presence of significantly more spectral line broadening, as well as more complex reflected light issues in ITER's metal wall environment (see e.g. discussion and references also provided in [5, 17]).

An upgraded, multi-sensor, diagnostic system has been recently completed to provide for JET DTE2 a more complete, plasma pulse-relevant, divertor gas analysis capability than was available for DTE1. This upgraded diagnostic system (partly described in [11]) is referred to, within the JET-DTE2 project, as the KT5 Upgrade [18].

Figure 5 illustrates a key feature of the present upgrade, aimed to alleviate gradual S/N deterioration through losses in light transmission in its Penning-OGA measurement. The scheme involves the use of a metallic mirror, designed to prevent direct line-of-sight between the Penning plasma source and the window through which the light is collected. It is aimed to prevent the deposition of thin coatings (Penning plasmas are akin to magnetron sputtering plasmas and thus efficient sources of coating deposition) directly onto the window. Impact of thin film coatings on light reflection from a metallic mirror was not expected to be as significant. In fact, an accelerated aging test, using this new, mirror-based configuration, with the same geometry and under similar gas and pressure conditions on a vacuum test-stand outside the JET enclosure (also the same spectroscopic measurement system) found no detectable loss of signals over the equivalent of 200 typical JET plasma pulses (figure 6).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In addition to the Penning-OGA measurement, this upgrade also includes mass spectrometers aimed for plasma pulse resolving operation, which was not available in DTE1. In fact, for DTE1, only the OGA measurement was able to provide gas species concentration information during the pulse (mass spectrometers were available but only post-pulse gas analysis and vacuum integrity characterization). That simpler configuration was known, until recently, as the KT5P diagnostic (the 'P' for Penning). Lower pressure conditions, as needed for the mass spectrometers to operate, will be made possible by separating the magnetically shielded mass-spectrometers from the OGAs via a motorized, variable aperture [18] as also illustrated in figure 5.

The spectroscopic detection system is also being currently upgraded with aim to improve D/T resolution and the overall dynamic range. In support of this hardware effort, an evolving, advanced spectral fitting program at CEA-IRFM, featuring full error statistics capability, has been run for the present study in simulation mode The aim is to evaluate what aspects of a hardware system configuration would most improve isotopic detection (especially at super-low concentrations) with particular focus on resolving T from D, as this is more dependent on fitting the spectral profiles than detecting the hydrogen.

Figure 7 shows a simulation for conditions relevant to the DTE1 data showed earlier (figure 1). The simulation first produces a theoretical D_α/T_α profile, for a given concentration and with each isotopic component comprising two Gaussians ('cold' and a 'hot' components) according to [13]. Random noise is then added (here the noise level typical of the DTE1 data is used). Subsequently, a large number (~10 000 in this example) non-linear fits are carried out (with the noise generated at random). The statistics then provide the upper and lower bounds of the $\pm 95\%$ confidence interval in the resulting fitted results (figure 7(a)). That means that, for an input tritium concentration ${{\left[T \right]}_{0}}$, and for given signal intensity and noise level, the measurement will have a probability 95% to be in the range $\left\{{{\left[T \right]}_{0}}-{{\varepsilon}_{{\rm lower}}},{{\left[T \right]}_{0}}+{{\varepsilon}_{{\rm upper}}} \right\}$, where ${{\varepsilon}_{{\rm lower}}}$ and ${{\varepsilon}_{{\rm upper}}}$ are plotted in figure 7(a). This high value $\delta _{{\rm lim}}^{{\rm Stat}}=95\%$ was chosen because of the large sensitivity to the low values of the minority gas concentration of the neutron generation in the DD and TT phases and of the ICRF core heating in the JET DTE2 campaign.

Zoom In Zoom Out Reset image size

In this case, the simulation was carried out for three levels of peak intensity. At the same random noise level this gives, correspondingly, three levels of S/N; this can be also thought of as three levels of use of the full dynamic range of the detector system. A simulated profile for 19% T concentration is shown in figure 7(b) for a poor S/N case, much as was the case of the DTE1 data, for comparison with figure 2(b). The S/N has clearly great impact on the detection limit. In fact, other cases ran with this simulation examined sensitivity to other instrument features, such as the linear dispersion (the spacing of the detector pixels for a given spectrometer instrument function). But these were not as important as the S/N. Looking, for instance, back at figure 7(a), for the condition that give only ~200 counts (cts) peak intensity, the error in T detection is already reaching 100% of the concentration value at 7%-conc. With medium signal level of about 2000 cts, one gets down to 0.7% concentration before the uncertainty reaches 100% of the value. This supports the strong emphasis put on ways to optimize light transmission/collection, as well mitigate the progressive deterioration thereof, due to coating of the windows by the Penning discharge, as discussed earlier in this section.

Although they have comparable S/N values, the limiting values (defined by $\frac{\varepsilon \left({{\left[T \right]}_{{\rm lim}}} \right)}{{{\left[T \right]}_{{\rm lim}}}}\sim 1$) deduced from figures 2(a) and 7(a) differ substantially (0.01 and 0.07, respectively). This is not a contradiction: the difference between the two estimates comes from the difference in the definition of the error bars. For the data presented in section 3, they are calculated from the uncertainty on the fit parameters and thus cannot be straightforwardly translated in terms of interval of confidence (However, it is recalled that these come from the already processed data, in the JET database, and from the automated data analysis procedure, which continues to be used to this date.) In contrast, in statistical study described in the present section, the error bars (i.e. the confidence interval) come from the consideration that the 10 000 values of ${{\left[T \right]}_{{\rm fit}}}$ obtained for a synthetic spectrum, to which a random noise has been added, exhibit a Gaussian distribution around the value ${{\left[T \right]}_{{\rm syn}}}$ used to produce the synthetic spectrum. Using this property, one could derive a relation between any chosen confidence interval (i.e. ${{\delta}_{{\rm lim}}}$) and the corresponding limit value ${{\left[T \right]}_{{\rm lim}}}$, i.e.: $\frac{1}{{{\left[T \right]}_{{\rm lim}}}}\cdot er{{f}^{-1}}\left(1-2{{\delta}_{\lim}} \right)={\rm Constant}$ (see appendix). This relation can then be used for estimating the confidence interval corresponding to the error bars of figure 2(a), yielding:

$\delta _{{\rm lim}}^{{\rm DTE1}}=\frac{1}{2}\left(1-erf\left[\frac{\left[T \right]_{{\rm lim}}^{{\rm DTE1}}}{\left[T \right]_{{\rm lim}}^{{\rm Stat}}}er{{f}^{-1}}\left(1-2\delta _{{\rm lim}}^{{\rm Stat}} \right) \right] \right)=\frac{1}{2}\left(1-\right.$$\left.erf\left[\frac{0.01}{0.07}er{{f}^{-1}}\left(1-2\times 0.95 \right) \right] \right)\sim ~0.59$.

In other words, for the data displayed in figure 2(a), there is only a 59% probability that the correct value is within the range of the fitted value  ±  the error bar from the fit. Such a probability is too low for a reliable estimation, e.g. of the neutron generation in the DD and TT phases or of the ICRF core heating at low minority gas concentration. This justifies the choice of a much higher value $\delta _{{\rm lim}}^{{\rm Stat}}$ for estimating the required performance of the Penning OGA system to be used in DTE2 (see section 4.1).