Nuclear Instruments and Methods in Physics Research A 614 (2010) 308 312 Contents lists available at ScienceDirect Nuclear Instruments and Methods in Physics Research A journal homepage: www.elsevier.com/locate/nima Development of a digital front-end electronics for the CdTe PET systems M. Nakhostin a,, Y. Kikuchi a, K. Ishii a, S. Matsuyama a, H. Yamazaki b a Department of Quantum Science and Energy Engineering, Tohoku University, Sendai 980-8579, Japan b Cyclotron and Radioisotope Center, Tohoku University, Sendai 980-8578, Japan article info Article history: Received 2 December 2009 Accepted 22 December 2009 Available online 5 January 2010 Keywords: Positron emission tomography (PET) Cadmium telluride (CdTe) Digital signal processing abstract We report on the development of a digital front-end electronics for high resolution CdTe PET systems. The energy and timing information are extracted through digital processing of the detector signals, digitized at the preamplifier stage. A cost effective preamplifier system and efficient digital algorithms for time pickoff and energy measurement are described. The effects of signals sampling rate and geometrical parameters of the detectors are explored. It is shown that with digitizers with 160 MSample/s sampling rates and 8 bit resolution, a timing resolution close to the analog results can be achieved. The advantages of the system include reasonable cost, short development time and ease of modification. & 2009 Elsevier B.V. All rights reserved. 1. Introduction In positron emission tomography (PET), direct detection of 511 kev g-rays by semiconductor detectors offers several advantages over conventional scintillator-based detectors. Advantages include excellent spatial resolution using pixilated detector arrays comprised of very small elements, excellent energy resolution and easy maintenance due to lack of photomultiplier tubes. Among the semiconductor detectors, Schottky cadmium telluride (CdTe) detectors appeared to be a promising detector for PET systems due to their high density (6 g/cm 3 ), high Z (50) and capability of operating at room temperature [1 3]. Our group has recently developed a very high spatial resolution small-animal PET system using Schottky CdTe detector arrays [1]. However, a CdTe PET scanner suffers from the huge number of readout channels [1 3]. To reduce the number of readout channels, we have recently developed a two-dimensional position sensitive Schottky CdTe detector with a position resolution of 1mm [4] and an effort is now underway to evaluate the performance of this detector in an actual PET system. For this purpose, a suitable front-end data acquisition system is required. Typical implementation of the PET front-end readout circuits relies on application specific integrated circuits (ASIC) designed to provide timing and energy information. However, commercial ASIC chips were found to be inapplicable for timing measurement with CdTe detectors, which require a different method of pulse timing due to the variation in the shape of detector signals. An alternative solution for the front-end Corresponding author. Present address: Department of Physics, University of Surrey, Guildford GU2 7XH, UK. Tel.: +44 1483 686113; fax: +44 1483 686781. E-mail address: M.nakhostin@surrey.ac.uk (M. Nakhostin). data acquisition is the digital processing of the sampled signals of detectors [5 8]. In this approach, detector signals are digitized by means of a free-running analog-to-digital converter (ADC), directly at the preamplifier stage, and the energy and timing information are extracted from the sampled waveforms by means of field-programmable gate arrays (FPGA). This paper reports on the development of a cost-effective charge sensitive preamplifier system and efficient algorithms for extracting timing and energy information from the sampled signals of the PET detectors. The effects of ADC sampling rate and geometrical parameters of the detectors are discussed as well. 2. CdTe PET detectors A schematic view of the CdTe PET detector blocks is shown in Fig. 1. The position sensitive Schottky CdTe detectors are tightly placed next to each other and, depending on the size of the detectors, several layers of detectors are stacked to reach the desirable detection efficiency. The Schottky CdTe detectors are fabricated by evaporating platinum and indium as electrodes onto surface of CdTe wafer [9]. Detectors are 1 mm thick and the slab of each detector is patterned with strips on one face. The strips are 1 mm wide and the pitch between the strips is 0.2 mm. Detectors with 10 10 and 20 20 mm 2 were examined in our tests. The first coordinate of a g-ray interaction point is determined by the strip from which the signal is originated and the second coordinate is determined by employing charge division technique along the strips, which are covered by a resistive layer of indium [4]. To reduce the number of position signals, the two ends of strips are connected to two chains of resistors and 0168-9002/$ - see front matter & 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.nima.2009.12.059
M. Nakhostin et al. / Nuclear Instruments and Methods in Physics Research A 614 (2010) 308 312 309 Fig. 1. Schematic view of the CdTe PET detector blocks. A stack of detectors is used to reach the desirable detection efficiency. The detector blocks are arranged in a ring geometry to serve as a PET detector. pulse division technique is employed to determine a strip carrying a signal. The position signals from the resistor chains are readout by the commercially available ASIC chips [10] and the digital method is used to extract the timing and energy information from the common electrode. 3. Front-end electronics 3.1. Preamplifier system The preamplifier system was constructed by using monolithic ICs. This offers low cost, spatial compactness and short development time of the system. Tests showed that there are only a few commercially available monolithic amplifier circuits that can be used for this task. We selected the FET-input Op-Amp, OPA656 from Texas Instrument Corp with the following features: low input bias current (2 pa), 500 MHz unity-gain bandwidth, low input noise (7 nv/ohz, and 1.3 fa/ohz), small size of 2.75 2.6 1.45 mm 3 and availability at low price. A diagram of the preamplifier circuit is shown in Fig. 2A. The charge sensitive preamplifiers are operated at R f =500 MO and C f =0.2 pf. An amplification stage is included in the preamplifier to adjust the amplitude of signals corresponding to the input voltage of the ADCs, which span 1 V. The Op-Amp, OPA847 from Texas Instrument Corp, was used for this stage. This Op-Amp combines very high gain bandwidth (3.9 GHz) with an ultra-low input noise (0.85 nv/ohz and 2.5 pa/ohz). A photograph of a prototype 4-channel preamplifier together with a CdTe detector and its holder is shown in Fig. 2B. The detectors are mounted on a flexible printed circuit (FPC), which is attached on a frame of fiberglass. The response of the preamplifier to a fast step pulse from a pulse generator is shown in Fig. 3. The preamplifier is connected to a 20 20 mm 2 CdTe detector and the test pulse is injected to the preamplifier through a capacitor. The rise-time of the preamplifier with this detector was measured to be 11 ns. The preamplifier noise for 10 10 and 20 20 mm 2 CdTe detectors Fig. 2. Top: a diagram of the preamplifier circuit. Bottom: a photograph of a 4- channel prototype preamplifier connected to a position sensitive CdTe PET detector. were measured as 6 and 11 kev, respectively. Although preamplifier noise is rather large, it has the advantage that detector and preamplifier can be integrated on the same FPC board on the detector holder. This minimizes the connectionrelated stray capacitance that was found to be quite large for connection through a FPC board. Fig. 4 shows the energy spectra of 22 Na measured with a 20 20 mm 2 CdTe detector for two cases of direct connection to the preamplifier and connection through a 5 cm long FPC board. It is seen that the direct connection of detector and preamplifier leads to a considerable improvement in the signal-to-noise ratio. 3.2. FPGA algorithms A major limitation in the timing performance of CdTe detectors is caused by the variation in the shape of detector signals due to the considerable difference in the mobility of electrons and holes. Variation in the shape of signals affects the time resolution by leading to a significant time walk when conventional timing methods such as leading edge or constant-fraction discrimination (CFD) methods are used. In analog domain, a general method for minimizing the effect of variations in the pulse shape of semiconductor detectors is to shape the signals with a small shaping time constant and determine the arrival time of pulses using the CFD method with an optimized delay on the CFD, corresponding to the so-called amplitude and rise-time compensation (ARC) mode of operation [11]. A digital implementation of this procedure is illustrated in Fig. 5 [8]: the preamplifier output signals are digitally shaped by a simple moving average filter (MAF1) whose smoothing power is chosen to not only remove the high-frequency noise but also attenuate the fast component of the signal. The smoothed signal is then subtracted from the original signal. Since the two signals only differ in the attenuated part of the original signal, subtraction leads to a signal formed by the fast
310 M. Nakhostin et al. / Nuclear Instruments and Methods in Physics Research A 614 (2010) 308 312 Fig. 3. The preamplifier response to a fast test pulse. The rise time of the preamplifier is 11 ns with a 20 20 1mm 3 CdTe detector. Fig. 4. The energy spectra of 22 Na measured with a detector directly connected to the preamplifier (black spectrum) and connected through 5 cm of FPC board (gray spectrum). component of the original signal that is used for timing measurements. The slope-to-noise ratio of the timing signal is optimized by applying a second moving average filter (MAF2) which is chosen to only filter out the high-frequency noise, while keeping the signal leading edge. The pulse shaping process is followed by a digital version of the analog zero-crossing constantfraction time discriminator. The timing signal is delayed for some time steps, amplified, inverted, and added to the original timing signal. This process transforms the unipolar signal into a bipolar pulse. The moment that the bipolar pulse initially crosses the time axis is marked as the arrival time of the pulse. To determine this moment, the maximum value of the bipolar signal is localized and all the signal samples before the maximum value are replaced by zero. Then, the time corresponding to the first sample below the zero level is taken as the arrival time of the pulse. In regard to energy measurement, first, a moving average filter is applied to the preamplifier signal to remove the high-frequency noise and then, the amplitude of the signal is determined by taking the difference between the maximum and minimum values of the signal stream. A critical issue concerned with the digital extraction of timing information from a sampled signal is the signal sampling rate. Even though there are ADCs that can sample up 1 giga sample per second (GS/s), such ADCs are very expensive and have very high power consumption. For these reasons, digital PET scanners are currently using ADCs with sampling frequencies of less than 200 MS/s. Such sampling frequencies result in a relatively long time interval between each sample, which is much coarse than the desired timing resolution. Therefore, some timing refinement must be made in order to obtain a reasonable timing resolution at low sampling rates. The interpolate algorithm has been shown to be very effective for this purpose [12]. In our algorithms, a simple linear interpolation is used to estimate the signal samples at sampling intervals of 1 ns. To test the numerical timing pickoff and energy measurement algorithms, an experimental setup including a CdTe detector, connected to the prototype preamplifier, a fast liquid scintillator (NE213), coupled to a Hamamatsu Photo Multiplier Tube (PMT) and a fast digital oscilloscope was used to collect waveforms from the detectors. A 22 Na positron source is placed between the two detectors and output signals of the preamplifier and the PMT are simultaneously digitized at a sampling rate of 1 GS/s and 8 bit resolution. The oscilloscope bandwidth is 1 GHz and the bias voltage of CdTe detector is 300 V. To compare the digital results with the results of the standard analog electronics, the preamplifier and PMT signals are also sent to a timing circuit using NIM modules. The oscilloscope is triggered using the NIM timing circuit and approximately 20 000 waveform pairs were stored to the hard disk drive of the oscilloscope. The data from the oscilloscope were then examined by a program written in
M. Nakhostin et al. / Nuclear Instruments and Methods in Physics Research A 614 (2010) 308 312 311 Fig. 5. Different steps of digital pulse timing: (A) original preamplifier signal, (B) signal after filtering with the MAF1, (C) timing signal (D) timing signal after noise filtering with MAF2, (E) timing signal after CFD shaping, (F) detection of pulse arrival time. Fig. 6. Energy spectra of 22 Na measured with two different sizes of CdTe detectors. (A) The energy spectrum of a 10 10 mm 2 detector (8%). (B) The energy spectrum of a 20 20 mm 2 detector (14%). In both cases the signals sampling rate is 166 MS/ s and detectors voltage is 300 V. MATLAB. The parameters involved in the timing algorithm of the CdTe detectors are the two moving average filters, MAF1 and MAF2, as well as attenuation fraction and shaping delay involved in the CFD shaping. The attenuation fraction of the CFD was set at 0.2 and the moving average filters and CFD delay were varied to reach the best time resolution. In order to evaluate the effect of digitizer sampling rate, the original data acquired at 1 GS/s, were recalculated as if they had been acquired with reduced sampling rate. The signals with reduced sampling rate are then interpolated to estimate the signal samples at sampling intervals of 1 ns. Such data were used for the simulation of the fast digitizer operation at 500, 250, 166 and 100 MS/s sampling. The results of energy and time resolution measurements for two different detector sizes of 10 10 and 20 20 mm 2 are shown in Figs. 6 and 7. The energy resolutions of the small and large detectors at 166 MS/s are, respectively, 7% and 14%, which are sufficient for a PET system. In regard to timing performance, at 1 GS/s sampling rate and for an energy threshold of 300 kev, the time resolution of the small and large detectors are, respectively, 8 and 14 ns FWHM, which are the same with the measurements with the standard analog circuits. It is seen that for the sampling rate 250 MS/s 1 GS/s the difference in the timing resolutions is not significant and at 166 MS/s sampling a slightly worse resolution compared to the case 1 GS/s is obtained. The time resolutions at 166 MS/s sampling rate are 10 and 17 ns FWHM, respectively, for small and large size detectors. A considerable degradation of the time resolution for the larger size detector is caused by the electronic noise due to the larger capacitance and leakage current of the detector. Since the number of signal channels and consequently cost of data acquisition system is a function of detectors size, a compromise should be made between the cost and timing performance. 4. Summary and conclusion This work analyzes the expected performance of a CdTe PET data acquisition system, which operates based on the digital processing of preamplifier signals. It was shown that solutions based on small size preamplifiers using commercial ICs represent a reasonable trade-off between cost and performance for a prototype PET system. The small size of preamplifiers enables to make a highly integrated detector and preamplifier system, leading to an acceptable level of electronic noise. Efficient
312 M. Nakhostin et al. / Nuclear Instruments and Methods in Physics Research A 614 (2010) 308 312 frequencies and it has been shown that, within the sampling range of interest (160 MHz), results are acceptable, as both the detectors provide enough resolution for the application of a 20 ns timing window. The results were obtained with a digitizer with 8 bit resolution and therefore a better resolution is expected for the prototype PET system, which is being constructed with 10 bit ADCs. The timing and energy performance of the system strongly depend on the size of detectors. Since the size of detectors determines the number of data channels, a detector size of 20 20 1mm 3 is a good compromise between the cost and performance. Acknowledgments This work was supported by a Grant-in-Aid for Specially Promoted Research no. 17002010 (K. Ishii) of the Ministry of Education, Culture, Sports, Science and Technology of Japan. References [1] K. Ishii, et al., Nucl. Instr. and Meth. A 576 (2007) 435. [2] G.S. Mitchell, et al., IEEE Trans. Nucl. Sci. NS-55 (2008) 870 876. [3] A. Drezet, et al., Nucl. Instr. and Meth. A 571 (2007) 465. [4] K. Ishii, et al., Proceedings of the Annual Congress of the European Association of Nuclear Medicine, Barcelona, Spain, October 10 14, 2009. [5] M. Streun, et al., Nucl. Instr. and Meth. A 486 (2002) 18. [6] R. Fontaine, et al., Proceedings of the 25th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, vol. 25, 2003, p. 998. [7] B. Joly, et al., Nuclear Science Symposium Record, vol. 3, IEEE, 2008, p. 4078. [8] M. Nakhostin, et al., Nucl. Instr. and Meth. A 606 (2009) 681. [9] K. Matsumoto, et al., IEEE Trans. Nucl. Sci. NS-31 (1998) 556. [10] RENA-3, NOVA R&D, Inc. USA. [11] G.F. Knoll, Radiation Detection and Measurement, third ed, Wiley, New York, 2000, p. 663. [12] L. Bardelli, et al., Nucl. Instr. and Meth. A 521 (2004) 480. Fig. 7. Results of digital timing with two different sizes of CdTe detectors. (A) Time resolution of a 10 10 mm 2 detector. At 166 MS/s sampling rate a time resolution of 10 ns FWHM at an energy threshold of 300 kev is obtained. The parameters for time pickoff are: MAF1=40, MAF2:=7 and CFD delay=25 ns. (B) Time resolution of a 20 20 mm 2 detector. A time resolution of 17 ns at an energy threshold of 300 kev is achieved. The parameters of digital timing are MAF1=40, MAF2=9 and CFD delay=28 ns. algorithms for timing and energy measurement were developed, which are easily implemented with FPGA technology. Timing performance has been analyzed for a wide range of sampling