Author Affiliations
1Hefei National Laboratory for Physical Sciences at the Microscale and Department of Modern Physics, University of Science and Technology of China, Hefei 230026, China2CAS Key Laboratory of Microscale Magnetic Resonance, University of Science and Technology of China, Hefei 3006, China3Synergetic Innovation Center of Quantum Information and Quantum Physics, University of Science and Technology of China, Hefei 20026, Chinashow less
Fig. 1. The architecture of the high performance FFT spectrum analyzer. Two optional operating modes are designed using the reconfigurable FPGA resources.
Fig. 2. The block diagram of the down-sampling and reconstruction module. The input data are reconstructed by using N stages of data down-sampling and reconstruction, and the digital data after reconstruction are processed by the FFT module.
Fig. 3. The frequency spectrums before and after down-sampling and filtering. (a)–(c) The produced signal aliasing when performing down-sampling. (d)–(f) The signal aliasing suppressed effectively by the implementation of the multi-stage filters.
Fig. 4. Block diagram of customized software for the FFT spectrum analyzer.
Fig. 5. Spin noise measurements for alkali metal Rb. (a)–(c) FFT spectrums measured with coarse mode, data reconstruction, and fine mode, respectively. In the fine mode, the input data are processed successively by the down converting module, the multi-stage digital filters, the multi-stage data reconstruction module, and the FFT module.
Fig. 6. The plots of the signal-to-noise-ratio versus the time span of spin noise measurements: (a) and (b) with a 1/4 GSa/s sampling rate, (c) and (d) with a 1/16 GSa/s sampling rate, (e) and (f) with a 1/256 GSa/s sampling rate.
Fig. 7. The test results of measuring mixed signals with different frequency components. Utilizing the high performance FFT spectrum analyzer to obtain the FFT spectrums, the mixed signals can be measured with a high frequency resolution, and the signals aliasing can be suppressed.
Fig. 8. The plots of frequency response characteristics of the filter. (a), (b) The filters operating at a 1/256 GSa/s and a 1 GSa/s sampling rates, respectively.
Module | FFT | Reconstruction | Filter (one stage) | Down converter | Total time |
---|
Time occupation (single FFT spectrum) | FFT DAQ (This work) | 32.75 μs | 32.75 μs | 32.75 μs | 32.75 μs | 32.75 μs | | Software (Python) | 1.2 ms | 26.8 ms | 22.1 ms | 5.8 ms | 226 ms | | Software (C++) | 1.1 ms | 4.6 ms | 3.97 ms | 2.8 ms | 75.3 ms | Time occupation (1 Giga samples) | FFT DAQ (This work) | 1 s | 1 s | 1 s | 1 s | 1 s | | Software (Python) | 40.8 s | 876.8 s | 723.8 s | 190.8 s | 6898.8 s | | Software (C++) | 33.6 s | 140.4 s | 121.2 s | 85.4 s | 2298.1 s |
|
Table 1. Efficiency comparison between this work and the software based FFT.
Resourcea | FFT | Reconstruction | DDR3 controller | DDS | Filter | Data storage | Total occupation | Available |
---|
6-input LUT | 5182 | 36637 | 22324 | 91 | 35019 | 18267 | 128038 | 303600 | LUT RAM | 1257 | 25 | 3438 | 0 | 15582 | 9674 | 31281 | 130800 | Flip-flop | 7964 | 53627 | 15097 | 304 | 61735 | 23198 | 175896 | 67200 | Slice | 2003 | 12955 | 7100 | 84 | 14376 | 4259 | 44601 | 75900 | DSP | 31 | 4 | | 3 | 1160 | 0 | 1300 | 2800 | BRAM (36 kb) | 97 | 402 | 0 | 1 | 0 | 46 | 840 | 1030 |
|
Table 2. Resource occupation of the hardware accelerated FFT DAQ board.
Analyzer (Ref.) | This work | Iglesias et al.[19] | Agilent N9030A[20] | Tektronics RSA5000[21] | Li[27] | Crooker[17] |
---|
Na | 32768 | 4096 | 1024 | 1024–32768 | 128000 | 32768 | Bandwidth | 500 MHz | 100 MHz | 160 MHz | 110 MHz | 500 MHz | 1 GHz | Resolution | 119 Hz–30.5 kHz | 48 kHz | 383 kHz | 20 kHz | 7.8 kHz | 61 kHz | Spectrums/s | 30.5 k | 24.4 k | 292 k | 30.5 k–292 k | N/A | 61 k |
|
Table 3. Performance comparison among FFT DAQs.