Hardware design of a signal acquisition and processing system

A signal acquisition and processing system was designed based on the PXIe bus architecture, as shown in Fig. 1(a). It adopts the structure of mother-daughter board that are connected through printed circuit board (PCB) connectors manufactured by TE Connectivity. The board was designed with four channels, only one of which channel was used for testing during the experiment.

Fig. 1Full-screen View

(Color online) Electronics design. (a) the signal acquisition and processing system board. (b) Motherboard

The daughterboard shown in Fig. 1(a), mainly includes an analog signal processing module and waveform digitization module. The analog signal processing module is composed of a SMA interface, AC coupling, impedance matching, overvoltage protection, amplifier, single-ended to differential, and anti-aliasing filtering. The output signals of the preamplifier are transmitted to the daughterboard through the SMA interface. The signals are then sequentially processed by AC coupling, impedance matching, and overvoltage protection. AC coupling can isolate the DC component of the signal, help compress the amplitudes of the piled-up signals, and suppress the drift of the baseline. Impedance matching uses a $50\text{ Ω }$ resistor that matches the characteristic impedance of the transmission line in parallel with the terminal load. Thus, the reflection coefficient of the terminal is 0, and the signal reflections at both ends of the transmission line to cause pulse waveform distortion are avoided. To avoid damaging the device owing to excessive input voltage, the overvoltage protection circuit uses a pair of clamping diodes to limit the signal amplitude to $\pm 5\text{ V}.$ A amplifier (LMH6703) ^[21] is used to adjust the signal amplitude within –1 – +1 V to maximize the use of the ADC's dynamic range and improve the signal-to-noise ratio of the ADC. The single-ended differential circuit converts a single-ended analog signal coming from the amplifier (LMH6703) ^[21] into a differential signal, which enters the waveform digitization module after anti-aliasing filtering and is converted into a digital signal through sampling, quantization, and coding.

The waveform digitization module includes high-speed ADCs and a clock jitter cleaner. Owing to the narrow pulse width and fast rising edge of the neutron signal described above, as well as the Campbelling integration accuracy requirement, a single-channel pipeline ADC (ISLA214P50) ^[22] with a sampling rate of up to 500 MPS, resolution of 14 bits, and high conversion rate is selected. The ADC can conduct sampling at a sufficient rate in each pulse signal for algorithm analysis. Furthermore, the ADC provides a 250 MHz associated clock. The input double data rate (IDDR) primitive of the FPGA is used for double-edge (rising and falling edges) sampling. This method reduces the FPGA’s processing time; thus, more data can be processed and a high time resolution can be achieved. To achieve the desired effective number of bits, a clock jitter eliminator is used to reduce the jitter of the ADC sampling clock. Here, the chip (LMK04828B) ^[23] of TI is selected, which can generate an ultralow jitter clock to meet the requirement of the ADC sampling clock.

The motherboard, which is shown in Fig. 1(b), mainly includes a core processing module, external trigger module, and bus interface. The core processing module mainly processes the ADC output digital signals. Considering factors such as speed grades, pins, resources, number of high-speed transceiver channels, and price, the xc7k325tffg900 ^[24] FPGA chip (Kintex-7) manufactured by Xilinx was selected. Important functions such as the pulse count algorithm of the pulse-counting mode, square integral algorithm of the Campbelling mode, and data buffering were all realized in the FPGA. The external trigger module receives and processes the external trigger signal taken from the EAST central control system and initiates the signal acquisition and processing system. The bus interface adopts a PXIe framework. It is designed based on the PCIe IP core of the Xilinx 7 series FPGA integrated block. Considering the 8 b/10 b encoding overhead, the theoretical unidirectional bandwidth can reach 32 Gb/s, which is sufficient to meet the current application of the NFM system and requirements of future upgrades. The GTX high-speed transceiver, PCIe hard core, and related IP cores are integrated into the FPGA. It greatly facilitates the hardware and FPGA programming and improves the integration of the board. In this study, the communication between the signal acquisition and processing system and upper computer is realized via a high-speed PXIe protocol in DMA mode. The DMA mode of the PXIe communication adopts a high-speed data packet transmission and is controlled through interruption.

FPGA design

A logical design block diagram of the FPGA used in the signal acquisition and processing system is presented in Fig. 2. It mainly includes the trigger control logic module, clock management module, jitter eliminator configuration module, ADC configuration and data preprocessing module, pulse-counting mode, Campbelling mode, first input first output (FIFO) module, and PXIe communication module with a DMA mode.

Fig. 2Full-screen View

Logical design block diagram of FPGA

The trigger control logic module realizes the hardware trigger function of the NFM system. If the signal acquisition and processing system receives the hardware trigger signal from the central control system, the NFM system starts to operate. The clock management module realizes a low-skew clock distribution based on high-speed buffers and routing and supports frequency synthesis and phase shift. It uses the 100 MHZ differential clock provided by the chassis backplane as the backplane triggering clock signal, which can be applied to the synchronous triggering mechanism of multi-board cards in the future. The clock jitter eliminator configuration module mainly utilizes the configuration of the clock jitter eliminator constructed via a serial peripheral interface (SPI) communication, such that the clock jitter eliminator outputs low-jitter differential 500 MHZ clocks as the sampling clock frequency of the ADC. The ADC configuration and data preprocessing module mainly complete the SPI communication and register mechanisms for the ADC. It also converts high-speed raw data into preprocessed data that can be recognized by the next module according to the defined rules. Then, the preprocessed data are divided into two parts. One part enters the pulse-counting mode to directly count the neutron pulses, while the other enters the Campbelling mode to obtain a Campbelling square integral value (hereafter called the Campbelling value). The two modes work at the same time. Their measurement results are acquired every 1ms and buffered in the FIFO module. Finally, the buffered data are transmitted at high speed to the upper computer for storage by the PXIe communication module with a DMA mode.

Key algorithm

Multi-point average linear calibration for Campbelling mode

When the neutron counts rate are low, the background noise occupies a larger proportion of the Campbelling values, which results in larger deviations in the Campbelling values. The counts rate measured by the pulse-counting mode can better reflect the neutron counts rate. When the neutron counts rate are in the middle region [ (approximately 40–500 counts per millisecond (cpms) ^[5,25]], the results measured by the pulse-counting and Campbelling modes can accurately reflect the neutron counts rate. Thus, this region is called the overlapping region in the dual mode ^[29,30]. When the neutron counts rate are high, the pulse signals will accumulate significantly because of the pulse signals with a certain pulse width. The neutron counts rate measured in the pulse-counting mode are lower than realistic counts rate. Fortunately, the Campbelling mode directly calculates the squared integral of the voltage signal, thus, its results will not be affected by the piled-up pulse and can accurately reflect the neutron counts rate. However, the Campbelling value is proportional to the neutron count rate value. Therefore, it is necessary to convert the Campbelling value into the neutron count rate via proper calibration.

As mentioned previously, there is an overlapping region in the dual modes. Thus, the Campbelling mode can be calibrated using the neutron count rate of the pulse-counting mode. The relationship between these two modes in the overlapping region is defined as follows: $\text{Campbell}=k\times \text{Count}+b.$ (10)

Here, $k$ is mainly determined by the neutron pulse waveform, amplitude, and number of signal sampling points, and $b$ reflects the contribution of background noise, which mainly depends on the magnitude of the electronic noise and various non-neutron background rays. Count and Campbell represent the neutron count rate of pulse-counting mode and Campbelling value of Campbelling mode, respectively.

Parameters $k$ and $b$ used in Eq. (10) are important for establishing reliable calibration results for the Campbelling mode. To accurately obtain the parameters $k$ and $b$, this study adopted a multi-point average linear fitting method. First, all the data with neutron counts rate ranging from 40 to 500 measured in the pulse-counting mode were extracted. Then, to suppress the random error in the Campbelling mode, the Campbelling values under the same count rate are extracted and averaged. Finally, the Campbelling average values and corresponding counts rate are linearly fitted using Eq. (10). In this study, shot 104218 (where $P_{\text{NBI}}$ is approximately 870 kW) was selected to calibrate the Campbelling mode. The results are presented in Fig. 4 and Eq. (11). The basic parameters of the shot are the plasma current ($I_{\text{p}}$ ~ 0.4 MA), central electron density ($n_{\text{e}}$ ~ $8.4\times \text{1}{0}^{19}{\text{ m}}^{-3}$), and loop voltage ($V_{\text{loop }}$ ~ 0.11 V) at the flattop. $\begin{array}{c}\text{Campbell}=2.7758\times {10}^5\times \text{Count}+3.5655\times {10}^8， R^2\\ =0.9911\end{array}$ (11)

Fig. 4Full-screen View

The linear fitting between the counts rate and average Campbelling values in the overlap region. The red point represents the average of all Campbelling values that correspond to a count rate in the pulse-counting mode.

In Fig. 4 and Eq. (11), the counts rate and corresponding Campbelling values in the overlap region exhibit a good linear relationship; subsequently, and $R^2$ reaches 0.9911. To verify the reliability of the calibration results further, this study uses Eq. (11) to convert the Campbelling values of shots No. 103609, No. 103939, and No.104250 into counts rate. The basic parameters of shot No. 103609 are the plasma current ($I_{\text{p}}$ ~ 0.4 MA) and loop voltage ($V_{\text{loop }}$ ~ 0.5 V) at the flattop. The basic parameters of shot No. 103939 are the plasma current ($I_{\text{p}}$ ~ 0.5 MA) at the flattop. Finally, the basic parameters of shot No. 104250 are the plasma current ($I_{\text{p}}$ ~ 0.4 MA), central electron density ($n_{\text{e}}$ ~ $3.2\times {10}^{19}{\text{ m}}^{-3}$), and loop voltage ($V_{\text{loop}}$ ~ 0.1 V) at the flattop. The counts rate converted by the Campbelling values are compared with the neutron counts rate directly measured in the pulse-counting mode, which is depicted in Fig. 5. In the overlap region, when neutron counts rate fluctuate, the change trends are consistent between the counts rate converted by the Campbelling values and those directly measured in pulse-counting mode. These values can overlap significantly. These results indicate that the calibrated Campbelling mode results can be used to calibrate different shot data with different neutron count rate fluctuations while attaining significant reliability.

Fig. 5Full-screen View

(Color online) A comparison between counts rate converted by the Campbelling values in the Campbelling mode and measured counts rate in the pulse-counting mode under the conditions of different shots.

Comprehensive test

In this study, to obtain the effective measurement range of the dual modes, the experimental data of different $P_{\text{NBI}}$ were analyzed. The dual mode measurement results of shot No. 101955 with a high level of injection power are presented in Fig. 6(a). The basic parameters of the shot are plasma current ($I_{\text{p}}$ ~ 0.4 MA), central electron density ($n_{\text{e}}$ ~ $7.2\times {10}^{19}{\text{ m}}^{-3}$), and injection power of NBI ($P_{\text{NBI}}$ ~ 3.5 MW) at the flattop. The red line represents the neutron counts rate measured in the pulse-counting mode, while the blue line represents the neutron counts rate converted by the Campbelling values through calibration. The neutron counts rate measured in the Campbelling mode were significantly higher than those in the pulse-counting mode. The reason for this is as follows: the neutron pulse signals are piled up in the case of high neutron counts rate, which results in smaller measured values in the pulse-counting mode; however, the piled-up pulse signals have almost no effect on the Campbelling mode.

Fig. 6Full-screen View

(Color online) Comprehensive tests with different injection powers of NBI. (a) The comparation between the converted counts rate in Campbelling mode and neutron counts rate directly measured in pulse-counting mode inside the high neutron count rate region; (b) The converted counts rate in Campbelling mode are used as the realistic counts rate for evaluating the deviations of the neutron counts rate directly measured in pulse-counting mode inside the high neutron count rate region; (c) The comparison between the converted counts rate in Campbelling mode and neutron counts rate directly measured in pulse-counting mode inside the low neutron count rate region; (d) The neutron counts rate directly measured in pulse-counting mode are used as the realistic counts rate for evaluating the deviations of the converted count counts in Campbelling mode inside the low neutron count rate region.

In addition, because the pulse-counting and Campbelling modes perform well in the case of low and high neutron counts rate, respectively, they can be used as realistic counts rate for evaluating the deviations of another mode. In this study, the shot data are further analyzed. When cpms is lower than 500 (including the overlap region), the counts rate measured in the pulse-counting mode can be used as realistic counts rate and those measured by the Campbelling mode can be used as measured counts rate. When cpms is higher than 500, the opposite is true, which is demonstrated in Fig. 6(b). In the overlap region, the counts rate converted by the Campbelling values through calibration are approximately equal to the realistic counts rate, which again indicates that the dual modes attain good working performance, and the calibration method and results of the Campbelling mode are sufficiently reliable. In the high-count-rate region, the deviation increases with increasing count rate and is approximately 19.8% as the count per millisecond reaches 1500. These results prove that the pulse-counting mode is unsuitable for high neutron counts rate.

The data for shot No. 104216 are selected to analyze the measurement effect of the Campbelling mode with low counts rate, as shown in Figs. 6(c) and 6(d). The basic parameters of the shot are the plasma current ($I_{\text{p}}$ ~ 0.4 MA), loop voltage ($V_{\text{loop}}$ ~ 0.1 V), and injection power of NBI ($P_{\text{NBI}}$ ~ 0.9 MW) at the flattop. When the neutron counts rate are lower than those measured in the overlap region, the counts rate in the Campbelling mode are higher than the realistic counts rate (counts rate measured in pulse-counting mode). This is because the counts rate measured in the Campbelling mode are interfered by non-neutron background such as noise, which will increase the neutron counts rate measured in the Campbelling mode and lead to inaccuracies in the measurement results.

In summary, when the neutron counts rate are lower than those measured in the overlap region, the results measured via the pulse-counting mode can be selected as realistic neutron counts rate. In the overlap region, the results measured by pulse-counting mode or Campbelling mode can be selected. However, it is best to select the results measured by the pulse-counting mode in the overlap region with low neutron counts rate, and select the results measured by the Campbelling mode in the overlap region with high counts rate to avoid greater deviations. When the neutron counts rate are higher than those measured in the overlap region, the results measured by the Campbelling mode can be utilized. Therefore, combining the dual modes expands the measurement range of the NFM system to ${10}^6$ cps.