64
R. Lulli et al.
spectrometer is limited to 50–60 MHz in a continuous acquisition mode, and to
100 MHz for non-continuous acquisition mode. The host system is a PC in stock
from a past project composed by a liquid-cooled Intel quad-core i7-3770 K CPU,
16 GB Double Data Rate 3 (DDR3) RAM, and two high-end GPU NVIDIA GeForce
GTX 680 with 4 GB of GDDR5 each [29].
Intensive tests have shown that this device can reach real-time spectra frequency
resolution of 1 Hz for a 50 MHz bandwidth single-channel signals. In non-real-time
applications, the system is only limited by the Picoscope bandwidth reaching spectra
frequency resolutions of 1 Hz over a 100 MHz signal bandwidth.
Evaluating the possible application fields of this instrument, we realized that the
input bandwidth limit of this spectrometer is too narrow for the requirements of actual
INAF radio telescopes. With a total component cost of e3,000, this spectrometer can
be useful for narrow bandwidth applications like NEO tracking or as a spectrometer
for reduced bandwidth radiotelescopes like the ones used in amateur or educational
applications.
7.3.5 The Second Final Project
In 2018, after careful market researches, we were able to identify a DAQ board
suitable to equip the final system. The Spectrum M4i.2210- ×8 PCI Express DAQ
board [37] can acquire signals data with a maximum sample rate of 1.25 GS/s on a
single channel at 8-bit resolution (about 48 dB of dynamic range) with maximum
signal bandwidth up to 500 MHz. Its software drivers let it works in continuous
streaming with optimized data transfer to avoid bottlenecks in data transfer due to
inefficient buses. Indeed, data transfers are performed in the background (DMA)
without overloading the CPU and the PCI Express bus.
Starting from this DAQ board, and with the support of the NewSpec hardware,
which in the meanwhile returned available, we built another spectrometer following
the second pilot project experience (Fig. 7.11).
The components used for this spectrometer are the host system composed by
a liquid-cooled Intel quad-core i7-3820 CPU, a 16 GB DDR3 RAM, a high-end
motherboard that supports PCI Express bus with 40 lines, and by two high-end GPU
Nvidia Titan Xp with 12 GB of GDDR5 each [31]. As we illustrated above, the size
of the GPU memory is a crucial parameter because the dimension of a single block
FFT analysis with CuFFT software library [23] depends on memory availability.
From the tests we carried out with our software, 12 GB of GPU memory allowed us
to perform FFT with 1 Hz resolution of signals with a bandwidth over 450 MHz.
Lab tests have shown that this spectrometer can reach a resolution of 1 Hz on a
bandwidth of more than 450 MHz in real-time. The overall cost of this instrument is
about e10,000, however, most of it (around 60%) being taken up by the DAQ board.
R. Lulli et al.
spectrometer is limited to 50–60 MHz in a continuous acquisition mode, and to
100 MHz for non-continuous acquisition mode. The host system is a PC in stock
from a past project composed by a liquid-cooled Intel quad-core i7-3770 K CPU,
16 GB Double Data Rate 3 (DDR3) RAM, and two high-end GPU NVIDIA GeForce
GTX 680 with 4 GB of GDDR5 each [29].
Intensive tests have shown that this device can reach real-time spectra frequency
resolution of 1 Hz for a 50 MHz bandwidth single-channel signals. In non-real-time
applications, the system is only limited by the Picoscope bandwidth reaching spectra
frequency resolutions of 1 Hz over a 100 MHz signal bandwidth.
Evaluating the possible application fields of this instrument, we realized that the
input bandwidth limit of this spectrometer is too narrow for the requirements of actual
INAF radio telescopes. With a total component cost of e3,000, this spectrometer can
be useful for narrow bandwidth applications like NEO tracking or as a spectrometer
for reduced bandwidth radiotelescopes like the ones used in amateur or educational
applications.
7.3.5 The Second Final Project
In 2018, after careful market researches, we were able to identify a DAQ board
suitable to equip the final system. The Spectrum M4i.2210- ×8 PCI Express DAQ
board [37] can acquire signals data with a maximum sample rate of 1.25 GS/s on a
single channel at 8-bit resolution (about 48 dB of dynamic range) with maximum
signal bandwidth up to 500 MHz. Its software drivers let it works in continuous
streaming with optimized data transfer to avoid bottlenecks in data transfer due to
inefficient buses. Indeed, data transfers are performed in the background (DMA)
without overloading the CPU and the PCI Express bus.
Starting from this DAQ board, and with the support of the NewSpec hardware,
which in the meanwhile returned available, we built another spectrometer following
the second pilot project experience (Fig. 7.11).
The components used for this spectrometer are the host system composed by
a liquid-cooled Intel quad-core i7-3820 CPU, a 16 GB DDR3 RAM, a high-end
motherboard that supports PCI Express bus with 40 lines, and by two high-end GPU
Nvidia Titan Xp with 12 GB of GDDR5 each [31]. As we illustrated above, the size
of the GPU memory is a crucial parameter because the dimension of a single block
FFT analysis with CuFFT software library [23] depends on memory availability.
From the tests we carried out with our software, 12 GB of GPU memory allowed us
to perform FFT with 1 Hz resolution of signals with a bandwidth over 450 MHz.
Lab tests have shown that this spectrometer can reach a resolution of 1 Hz on a
bandwidth of more than 450 MHz in real-time. The overall cost of this instrument is
about e10,000, however, most of it (around 60%) being taken up by the DAQ board.