Hardware description

The hardware of the NGDS comprised three parts: a liquid scintillator detector, programmable digitizer, and computer. The liquid scintillator detector (such as EJ-309 and BC-501A, which have proven to be capable of PSD) served as a neutron detector. The digitizer provided the detector with a high negative voltage, facilitating signal acquisition under appropriate conditions. Real-time digital neutron-gamma discrimination firmware was implemented on the FPGA. The processed data were uploaded to the computer via ethernet based on the TCP protocol. Finally, the neutron-gamma discrimination results were stored and displayed via an online measurement software [see Fig. 1(a)].

Fig. 1Full-screen View

(a) Block diagram of the digital NGDS. (b) Typical waveforms of the detectors BC-501A and EJ-309

The port on the detector output directly to the digitizer without amplification. A high-speed 500-Msps, 12-bit analog-to-digital converter (ADC) was used in the NGDS for digitization. The FPGA adopted in this study was A SPARTAN-6 Xilinx; its main features are listed in Table 1.

The experiment was performed using the neutron-gamma source Am-Be241 (located at Chengdu University of Technology) and the HL-2M tokamak. Typical waveforms for BC-501A and EJ-309 are shown in Fig. 1(b).

Architecture of the FPGA firmware

The FPGA in our digitizer performed real-time pulse processing, which primarily involved: (1) increasing the signal-to-noise ratio (SNR), (2) pulse triggering with a specified voltage threshold, (3) pile-up rejection, (4) acquiring the baseline value of a pulse within a specified duration, (5) acquiring the peak value of a pulse within a specified length, (6) integral calculation for a specific length, (7) PSD calculation and analysis, and (8) neutron / gamma (n/γ) pulse height analysis with a specified channel number. The firmware can be adapted to various detectors, and different software packages can be designed to meet practical requirements.

The FPGA architecture is shown in Fig. 2. The NGDS functioned in the list and spectrum modes, which processed data differently and corresponded to pulse information calculation and data analysis, respectively. Meanwhile, the waveform acquisition firmware inside the FPGA acquired the original waveform for offline analysis independent of the two measurement modes. The triggering and sampling options were controlled to collect the original pulse signal, which was necessary for analyzing the signal features. The data generated by these three modes were independently packaged and sent to the computer for different functions.

Fig. 2Full-screen View

Block diagram of the FPGA architecture

When the clock frequency for signal processing was 500 MHz, satisfying the timing requirements of the FPGA in signal pipelining was difficult. Therefore, this study used a serial-to-parallel conversion pipeline to convert a 500-MHz, 12-bit data volume into four parallel 125-MHz, 12-bit data processes. Four-channel parallel processing was a reasonable choice.

The main digital signal processing was performed in the list mode, and the essential parameters for n/γ discrimination were calculated. The signal from one of the channels was calculated as a trigger that determined the starting point of the baseline calculation and peak detection. After the peak and baseline values were obtained, the peak was used as the trigger signal for integration. Each channel simultaneously performed real-time pulse processing, including baseline, peak, and integral calculations, and the final result was obtained after synthesis. The baseline calculation increased the SNR. After the PSD ratios were determined by the computer, the information was displayed using the software. Simultaneously, these parameters were statistically analyzed directly on the FPGA (performing the data analysis) and finally, the discriminated n-γ spectrum was obtained.

Thus, the timing requirements of the FPGA were also satisfied while implementing the 500-Msps sampling rate and utilizing the ADC performance.

Fast-trapezoidal shaper and pile-up rejection

The fast-trapezoidal shaper is a universal and flexible digital shaper that handles the tradeoff between the minimization of resolution and maximization of throughput by providing adjustable parameters for the peaking time and flat-top duration [23].

The fast-trapezoidal shaper module contains delay and subtraction units. The input signal was first divided into two branches (assigned different delays) and subtracted to form a shaped signal that was positive and then negative. The trigger signal was generated when the input signal exceeded the preset threshold [24].

At high count rates, the pulse pile-up effect became significant, influencing the pulse amplitude and reducing the resolution of the energy spectrum [25]. A pulse was generated by the trigger signal. If the width between the previous and the next pulse was less than a predefined value, the current pulse was considered a pile-up pulse to be rejected; otherwise, it was considered an acceptance pulse.

Baseline calculation and peak detection

Baseline calculations and peak detection were triggered by a pulse flag generated by a fast trapezoidal shaper.

Before the arrival of the leading edge, the average of the signal was obtained and used as the baseline value for the pulse signal. In the peak detection module, the peak detection length was controlled by a command to accommodate different detectors. The module output the minimum (peak) signal when the predetermined upper limit of the peak detection length was reached.

Integration and PSD calculation

The charge comparison method (CCM) is an effective PSD method for performing n-γ discrimination [26, 27].

In an environment comprising mixed neutron and gamma-ray pulses, the neutrons and gamma rays produce different secondary charged particles in liquid scintillator detectors. In scintillators, the components of light pulses excited by charged particles possess different decay times. Furthermore, these particles produce different waveforms of the output light pulse, especially with regard to the intensity of the tail component, which allow their identification. Neutrons have a higher slow-tail component than gamma-rays [16, 28].

The CCM can be implemented on an FPGA through a simple integration operation, which typically involves comparing the charge integration of a current pulse over two different time intervals. For practical calculations, the PSD is defined as the ratio of the short integral to the long integral at the trailing edge of the current pulse. The integration was divided into long-gate (I_L) and short-gate (I_S) integrations, depending on the integration length.

Integration was performed after peak detection. I_S begins at the peak point of the pulse (T_P) and extends to a position before the end of the trailing edge. I_L starts at the same position as I_S and possesses a greater integration width. The longer decay time of neutron pulses facilitates discrimination between neutrons and gamma rays. These two integral regions can be easily implemented and adjusted by sending commands to the FPGA.

The integral operation is followed by the PSD ratio calculation. The PSD ratio is defined as follows: $PSD=\frac{I_{\text{S}}}{I_{\text{L}}}\mathrm{.}$ (1) The PSD ratio calculation unit was implemented using an eight-bit divider that calculated the ratio of I_S to I_L for each channel in real time. Finally, PSD results were transferred to the PC via a gigabit ethernet link. A common FOM method was used to evaluate the n/γ-ray discrimination effectiveness [29-31]. The FOM is defined as: $FOM=\frac{\Delta X}{{\delta }_{\text{gamma}}+{\delta }_{\text{neutron}}}\mathrm{,}$ (2) where ΔX is the distance between the neutron and gamma ray peaks, while Δ_gamma and Δ_neutron represent the full width at half maximum (FWHM) of each corresponding peak. To calculate the FOM values for the analysis, the PSD values of n/γ were fitted with two Gaussian distributions [15].

The selection of I_S and I_L widths significantly affected the FOM. As shown in Eq. 1, if I_S width is too small at a constant I_L width, the deviation in the PSD value obtained is larger (greater impact of the noise received); furthermore, a significantly long region along the trailing edge will not be utilized. In contrast, if the I_S width is set too large (close to the I_L width), the final FOM value will be smaller (loss of neutron-gamma differences), yielding PSD values close to one. An excessive I_L width inevitably causes noise near the baseline, which results in larger PSD value deviations and increases the calculation workload and computation time. This adversely affects pulse discrimination calculations at high count rates. A small I_L width (close to the I_S width) results in poor FOM values.

Spectrum analyzer

The PSD unit and PHA are important components of data analysis. The PSD values were stored by the ratio analyzer module as a histogram with 256 channels, whereas a pulse height histogram (1–1024 channels for gamma rays and 1025–2048 channels for neutrons) was stored by the PHA. The PSD ratio was calculated using the divider and fed into the two PHAs for discrimination. In this process, each pulse signal classified by the PSD value was completed once by the digitizers, and the computer interface displayed the data from the PHA and PSD analyses. The size of the data and transfer rate depended on the upload interval, which maximized the speed of the FPGA chips and implemented high count-rate measurements while reducing the dependence on network speed.

Four-channel parallel processing

To meet the timing requirements of the FPGA under the conditions of signal pipelining, four 125-MHz parallel pipeline processes were used with 12-bit data. Four-channel parallel processing was performed for baseline calculation, peak detection, and integral calculation, as shown in Fig. 3. After the baseline values were calculated separately for each of the four groups in the baseline calculation module, the average value was again calculated in the baseline average unit. The result was used as the baseline value.

Fig. 3Full-screen View

Block diagram for four-channel parallel-processing

The signal of one channel (e.g., CH1) was calculated using the fast-trapezoidal shaper module to generate a trigger signal using over-threshold triggering. Within the peak detection module, each unit obtained the peak value of each pulse and the four peaks were compared in the maximum peak value unit to determine the peak value. Simultaneously, the output channel peak flag was used to determine the channel to which the maximum peak belonged, because of its importance for the subsequent parallel integration [see Fig. 4(a)].

Fig. 4Full-screen View

(Color online) (a) Four-channel parallel processing in trigger calculation and peak detection, showing the original waveform and parallel signals with points marked every 2 ns (500 Msps) and every 8 ns (125 Msps), respectively. (b) Four-channel parallel processing in integration, with the starting positions of the four channels (CH1-4) for integration in four situations (channel peak flags from 1 to 4)

After the baseline was subtracted, the maximum peak was used as the trigger signal for integration. The channel sequence unit controlled the order of integration between the four signals according to the value of the channel peak flag (from 1 to 4). Thus, the channel where the maximum peak is located is integrated first, as shown in Fig. 4(b).

In the integral calculation unit, the results of the four-channel integration were added to output the total short- and long-gate integration values.

Data transfer

Because the list mode was trigger-based, the amount of data and the transmission rate depended on the count rate of the source. In this process, more detailed particle information might be obtained; however, this was limited by the data transfer rate from the FPGA. This method relies on the network speed, which is unsuitable for measurements at high count rates (usually above 1 Mcps). The data transfer speed in the list mode was influenced by the count rate. One package (with a size of 27 bytes) corresponded to the calculated data of one particle containing the arrival time of the pulse, integration value, count value, and amplitude information. When the count rate was 1 Mcps, the transfer rate was approximately 210 Mbps, which created high demands on the ethernet hardware. High count rates also required the storage of more data.

In the spectrum mode, all particles were continuously counted over a period of time, and the results were displayed. The size of the two packets was 6962 bytes, and the analysis results were uploaded at equal intervals until the end of the measurement. The measurement software facilitated the viewing of the analytical processes using the time-trace function. The upload time was set from 1 to 100 ms to satisfy the different time precisions. Ethernet hardware with different parameters were selected according to actual application requirements. The transmission rates were approximately 53 Mbps and 0.53 Mbps at high (1 ms) and low (100 ms) time precision, respectively. This resolves the limitation of network speed for high-count-rate analysis.

Energy calibration

A Cs-137 gamma source was used to calibrate the detector energy and perform neutron gamma discrimination at the determined energy threshold. Calibration was performed by plotting the energy spectrum of Cs-137. The response of liquid scintillator detectors to gamma rays primarily refers to energy deposition during Compton scattering. The results of the calibration, that is, the distribution of 50,456 pulse heights with the channel position for BC-501A and 67,043 for EJ-309, are shown in Fig. 5.

Fig. 5Full-screen View

(Color online) Gamma energy spectrum of Cs-137 obtained using BC-501A and EJ-309

The Compton edge energy of Cs-137 gamma rays is 477 keV. The channel position at 75% of the Compton-edge maximum was used to calculate the calibration line [32, 33]. Thus, a correspondence was observed between the channel number and energy.

Parameter optimization

In practical applications, the different response signal characteristics of different types of scintillator detectors, necessitates the selection of the optimal parameters through a series of comparative experiments. The long- and short-gate integration lengths affect the final PSD ratios and directly determine the discrimination under the same measurement threshold. To explore the effect of different long-gate (short-gate) integration lengths on the discrimination effect and determine the optimal discrimination parameters of the detector, we used the waveform acquisition module inside the FPGA to acquire a large number of pulse signals (105,000 for BC-501A and 84,595 for EJ-309) for offline analysis in MATLAB.

Based on the typical results of previous experiments, various short-gate integration times were set with an equivalent electron energy of 300 keV_ee and a long gate integration time of 96 ns. As shown in Fig. 6(a), the FOM trends of the two detectors remain the same as the short-gate integration time increases. The maximum FOM values are obtained at 24 ns and 32 ns for EJ-309 and BC-501A, respectively. This is not significantly different from the FOM value at 24 ns; therefore, 24 ns can be set as a suitable short-gate integration length. Figure 6(b) shows that with a short gate integration time of 24 ns, the FOM values of the two detectors increase with the long gate integration time. The growth of the values on the two lines gradually flattens, and excessively long integration times increase the FPGA computation time, which hinders digital signal processing at high count rates. Therefore, the integration time of 96 ns was selected as the optimal value.

Fig. 6Full-screen View

Effects of (a) short-gate integral length and (b) long-gate integral length. (c) FOM measured at different energy bands

Through offline analysis of the waveform data, our experiments showed that 24 ns was a typical interval for a short-gate integral using these two detectors. Meanwhile, 96 ns was a typical interval for a long-gate integral that approximately covers the tail area of a pulse.

The measured FOM for the different detectors in the different energy bands are shown in Fig. 6(c). The FOM improves with increasing pulse height, resulting in a higher SNR.

Results for real-time pulse shape discrimination

The real-time PSD results were calculated after selecting the optimized PSD parameters. Figure 7 illustrates the discrimination effects of the data collected in the list mode. We measured 56,031 and 34,781 pulses using the detectors EJ-309 and BC-501A, respectively, at an energy threshold of 100 keV_ee.

Fig. 7Full-screen View

(Color online) Scatter plots of the PSD and pulse height with a 100-keV_ee threshold

The distinction between neutrons and gamma-rays becomes evident as the energy threshold increased, as shown in Fig. 8. The PSD has a double Gaussian distribution, where the data on the left side represent the neutron pulse and its corresponding Gaussian peak, while the data on the right side represent the gamma rays. The FOM indicates the quality of particle discrimination. As the FOM increased, the fitting of the double-Gaussian distribution improved. For the detector EJ-309, the FOM values were 1.3 and 1.5 for thresholds of 100-keV_ee and 300-keV_ee, respectively. Likewise, the FOM values for BC-501A were 1.1 and 1.3 for 100- keV_ee and 300-keV_ee, respectively. These values are poorer than those of EJ-309. When the upload interval, measurement time, threshold, and other parameters were set, the real-time software simultaneously displayed the PSD curve, energy spectrum, and current count rate of the two detectors.

Fig. 8Full-screen View

PSD distributions with thresholds of (a) 100 keV_ee and (b) 300 keV_ee using EJ-309. PSD distributions with thresholds of (c) 100 keV_ee and (d) 300 keV_ee using BC-501A

Software real-time discrimination display

The results obtained by both detectors were simultaneously recorded using the neutron-gamma discrimination software. Parameters such as the threshold, integration length, and measurement time were set. The software directly displayed the PSD statistical curves of the two NGDS. When the discrimination parameters of the neutrons and gamma-rays were set, their respective energy spectra were obtained at the end of the measurement (see Fig. 10). Simultaneously, a data traceback viewing function was designed in the software to review the energy spectrum information and PSD curve statistics at any moment (with a minimum time precision of 1 ms).

Fig. 10Full-screen View

(Color online) Real-time measurement software interface. (a) PSD curves of BC-501A (CH1) and EJ-309 (CH2). (b) n/γ spectra of BC-501A (CH1) and EJ-309 (CH2)

Performance of the NGDS

4.2.1

Waveform acquisition

The characteristics of liquid scintillation detectors must first be evaluated by acquiring and analyzing their waveforms. As shown in Fig. 11, the NGDS operates in waveform acquisition mode during plasma discharge. During the measurement within 1 μs, the piled-up pulses were well suppressed owing to the strong shielding conditions even at high count rates, indicating that the detectors were performing in a sufficiently high and dynamic voltage range. Simultaneously, the existence of a pile-up pulse indicates the necessity of a signal processing method at high count rates.

Fig. 11Full-screen View

(Color online) Recorded waveform over a 1- μs duration from discharge

4.2.2

Neutron time traces of different neutron detection systems

For comparison with the NGDS, an high-temporal resolution neutron flux measurement (HTRNs) system (including a ZnS detector), ²³⁵U fission chamber (FC), and ³He counter were placed next to the tokamak. HTRN_S system, ³He counter, and ²³⁵U fission chamber were developed for neutron yield measurements [18]. A series of measurements was obtained from different discharge numbers. Figure 12(a) presents the time traces of the neutron count rates (corresponding to two different discharge numbers) with a temporal resolution of 1 ms.

Fig. 12Full-screen View

(Color online) (a) Time traces of the neutron count rates from discharges #03207 and #03148. (b) Time traces of the neutron and gamma count rates from discharges #03207 and #03148. (c) Dead time and (d) FOM with different maximum count rates using EJ-309 and BC-501A

Overall, during plasma discharge, FC had the largest neutron count rate, which was approximately 2-4 times that of BC-501A during the same period. EJ-309 and HTRN_S had comparable neutron count rates, whereas those of BC-501A and ³He were lower. The neutron detection efficiency of EJ-309 was approximately twice that of BC-501A during the same period; furthermore, it had a better neutron detection efficiency than BC-501A. However, both exhibited weaker neutron detection efficiencies than FC. The neutron count rate of BC-501A from discharge #03148 reached a maximum of 162 kcps, whereas that of EJ-309 reached 230 kcps.

The neutron count rate obtained with the NGDS exhibited the same variational trend as those obtained with the HTRN_S, FC, and ³He. Thus, the NGDS is suitable for high-flux neutron measurements.

4.2.4

Neutron-gamma discrimination with varying count rate

Care must be taken because the measurement dead time of the NGDS increases with the maximum count rate, and the FOM value for n/γ discrimination decreases, as shown in Fig. 12(c)). This implies that discrimination worsens.

The PSD curves of detectors #03207 and #03148, shown in Fig. 13, indicate that the FOM values are 1.433 and 1.92 for BC-501A (a) and EJ-309 (b), respectively, in #03207. The FOM values are 1.054 and 1.296 for BC-501A (c) and EJ-309 (d) in #03148, respectively, indicating that the electronics is equally capable of discriminating between neutrons and gamma rays at high count rates. Simultaneously, the BC-501A has a low neutron-to-gamma ratio owing to the lack of lead shielding, and a lot of gamma-rays are identified. This may also have resulted in EJ-309 being discriminated better than BC-501A under the same discharge number.

Fig. 13Full-screen View

(Color online) PSD distributions using BC-501A (a) and EJ-309 (b) from discharge #03207. PSD distributions using BC-501A (c) and EJ-309 (d) from discharge #03148