CASPER Development for GAVRT at Caltech Glenn Jones Aug. 03, 2008 2008 CASPER Workshop
Acknowledgements Xilinx – Generous FPGA and software donations Sandy Weinreb & Hamdi Mani – Feed measurement data
Useful stuff first! The Simulink scope is terrible!
gtkWave from Simulink for CASPER
gtkWave from Simulink for CASPER
Sick of ‘od - x’ for interpreting snapshot data?
gtkWave Snap – View SnapBRAM data
gtkWave Snap gtkgen(wave) $ gtkwave temp.vcd
Vector accumulator
Vector Accumulator
Utility blocks
G oldstone A pple V alley R adio T elescope 34 m telescope in southern California to be used by K-12 students to take data for astronomers. Currently being equipped with a novel ultra-wide-band radiometer designed at Caltech.
Wide band quad-ridge feeds 0.5-2 GHz 4-14 GHz Uncooled feed Feed and LNAs cooled LNAs cooled to 50K to 15 K
Receiver Noise Temperature Noise and Gain of Polarization X GAVRT HFF Front-End, DSS13 Pad, July 17, 2008 LFF X Pol Trcv and Gain 100 6 200 10.0 90 3 7.0 180 80 0 160 4.0 70 -3 140 1.0 60 -6 120 -2.0 Noise, K Gain, dB Gain, dB Trcvr, K Noise 100 -5.0 50 -9 Gain -8.0 80 40 -12 60 -11.0 30 -15 40 -14.0 20 -18 Noise 20 -17.0 Gain 10 -21 0 -20.0 0 -24 0 0.5 1 1.5 2 2.5 3 3.5 4 0 2 4 6 8 10 12 14 16 18 20 Frequency, GHz GHz
Crossover performance Nois e of HFF and L FF, X and Y Pol, from 0.5 to 5 GHz 200 180 L FF Pol Y HFF Pol Y 160 L FF Pol X HFF Pol X 140 120 Nois e, K 100 80 60 40 20 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Frequency, GHZ
Front end box 4 dual polarization pairs of receivers Input from 0.5 to 18 GHz Select bandwidth from 100 MHz / 500 MHz / 1 GHz / 2 GHz Downconvert to baseband.
Basic Receiver Architecture 2-14 GHz Feed I I LNA 1 GHz LPF 2 GHz BPF @ 22 GHz Q LNA 1 GHz LPF 0.5-4 GHz Feed 22 GHz 22-40 GHz Fixed LO Tunable LO The system consists of eight such receivers, arranged as four dual-polarization pairs.
Receiver modes The I and Q outputs can optionally be combined in a hybrid to form upper and lower sidebands. Additional filter options are also available. Currently only 8 of the 16 possible outputs are routed to the digital back-end. This will be upgraded in the future. DSS 28 Bandwidth Selections (MHz) Each IF converter provides the following bandwidths with center frequency from 1 to 15 GHz IF Filter I/Q or U/L Baseband Processed Image Comments Bandwidth Selection Bandwidth Bandwidth Rejection 2000 I/Q 10 -1000 2000 25 dB Needs 1 GHz I/Q spectrometer 2000 U/L 100-1000 1800 15 dB Needs 1 GHz spectrometer 2000 U/L 500-1000 1000 20 dB For 500 MHz spectrometer 400 U/L 120-520 400 50 dB For WVSR 400 U/L 270-370 100 50 dB For VSR
The digital backend 8x ADCs, 8x iBOBs 16x XAUI links to BEE2 2x 10 GbE links to Procurve switch ~20x 1 GbE to small cluster
Basic signal processing infrastructure iBOB iBOB Pol 3 Real ADC0 100 MbE 2Gsps Input Pol 1 Real 0-1 GHz ADC0 Input 2Gsps 100 MbE 8 Gbps 0-1 GHz XAUI0 16 Gbps 8 Gbps iADC 16 Gbps XAUI0 iADC Clk Clk PPS PPS Clock Splitter ADC1 8 way splitter Sampling Clock ADC1 8 Gbps BEE2 XAUI1 8 Gbps XAUI1 Clock Out Clock Clock in 100 MbE Out iBOB XAUI0 XAUI0 iBOB Corner FPGA Corner FPGA ADC0 100 MbE 2Gsps Input Pol 3 Imag ADC0 Input 2Gsps 100 MbE Pol 1 Imag 8 Gbps XAUI0 16 Gbps 0-1 GHz iADC XAUI1 XAUI1 0-1 GHz 16 Gbps XAUI0 8 Gbps iADC Clk Clk PPS PPS Pulse Distribution Amp PPS Pulse Distribution Amp XAUI2 XAUI2 ADC1 PPS ADC1 8 Gbps XAUI1 PPS XAUI3 XAUI3 XAUI1 8 Gbps Clock Out Clock Out iBOB iBOB XAUI3 XAUI3 ADC0 100 MbE 2Gsps Input Pol 2 Real ADC0 Input 2Gsps 100 MbE 8 Gbps XAUI0 0-1 GHz 16 Gbps iADC XAUI0 8 Gbps 16 Gbps XAUI2 XAUI2 iADC Clk Pol 4 Real Clk PPS 0-1 GHz PPS XAUI1 XAUI1 ADC1 ADC1 8 Gbps XAUI1 XAUI1 8 Gbps XAUI0 XAUI0 Clock Out Clock Corner FPGA Corner FPGA Out XAUI1 XAUI0 iBOB iBOB ADC0 100 MbE 2Gsps Input Pol 4 Imag ADC0 Computer 8 Gbps Input 2Gsps 100 MbE XAUI0 Pol 2 Imag 0-1 GHz 8 Gbps Cluster iADC XAUI0 0-1 GHz iADC Clk Clk PPS PPS 2x 10 GbE 24x 1GbE ADC1 Switch ADC1 8 Gbps XAUI1 8 Gbps XAUI1 Clock Out Clock Out
Thesis goals Build a unique instrument designed to take advantage of the wide bandwidth provided by the GAVRT telescope to measure the following: Detailed spectral characteristics of giant pulses from the Crab and other pulsars • Extensive statistics of giant pulses vs. frequency Nanostructure in giant pulses Dynamic spectra of pulsars with unprecedented bandwidth RFI performance in light of modern mitigation techniques • IQ imbalance correction
Thesis Goal: Detailed spectra of giant pulses What we want to look at: Crab Giant pulses vs. Frequency “Earlier, we had noted the potential spectral similarity between giant pulses from pulsars and that of the Sparker. It would be useful to determine the road- band spectrum of giant pulses, say from 1-m to 10-cm wavelength. In short, we are advocating the study of giant pulses from pulsars as convenient plasma laboratories that may further our understanding of the fleeting Sparkers. ” - Sri Kulkarni “Giant Sparks at Cosmological Distances?” From: Cordes et al. 2004 ApJ 612 375
Thesis Goal: Nanostructure in giant pulses What we want to look at: Crab Giant pulses vs. Frequency From: Cordes et al. 2004 ApJ 612 375 Hankins & Eilek, ApJ 670:693-701, Nov 2007
What we want to look at: Giant pulses vs. Frequency Hankins & Eilek, “Radio Emission Signatures in the Crab Pulsar.” ApJ 670:693-701, Nov 2007
Current limitations to giant pulse observations Multiple frequency observations have generally required simultaneous observation with many telescopes little data available Ultra-high time resolution has been limited by: Feed/receiver bandwidth Dispersed pulse is longer than memory buffer Lack of dedispersed trigger • More susceptible to RFI • SNR of dispersed pulse too low to trigger on
GAVRT Transient Capture Mode Raw data input rate: 16 Gbyte/s iBOB Pol 1 Real ADC0 Input 2Gsps 100 MbE 0-1 GHz Max data output rate: 2 Gbyte/s 8 Gbps 16 Gbps XAUI0 iADC Clk PPS Ping 4 way splitter Sampling Clock ADC1 Pong 8 Gbps XAUI1 BEE2 Clock Out XAUI0 XAUI0 iBOB Corner FPGA Corner FPGA ADC0 Dedispersion Computer Input 2Gsps 100 MbE Detector Accumulator Incoherent and Trigger Pol 1 Imag Small Vector Cluster XAUI1 XAUI1 0-1 GHz PFB 16 Gbps XAUI0 8 Gbps Align and Merge iADC Clk PPS Duplicate hardware for other receivers 2 GB XAUI2 XAUI2 Ping DRAM PPS 4 way splitter Pong Circular ADC1 Buffer 2 GB XAUI3 XAUI3 XAUI1 8 Gbps DRAM Circular Buffer XAUI0 Clock MbE 100 Out 2x 10 GbE Center FPGA 24x 1GbE Clock in XAUI1 Switch iBOB Pol 2 Real 2 GB XAUI3 XAUI3 ADC0 Input 2Gsps 100 MbE DRAM 0-1 GHz Circular XAUI0 8 Gbps 16 Gbps Buffer iADC XAUI2 2 GB XAUI2 Clk Align and Merge DRAM Circular PPS Buffer Duplicate hardware Ping for other receivers Dedispersion Accumulator Pong XAUI1 and Trigger XAUI1 ADC1 Incoherent Vector Detector XAUI1 8 Gbps Small PFB XAUI0 XAUI0 Clock Corner FPGA Corner FPGA Out iBOB ADC0 Input 2Gsps 100 MbE Pol 2 Imag 8 Gbps XAUI0 0-1 GHz 16 Gbps iADC RAM buffers are sufficient to store Clk PPS 1 second of raw voltages from Ping ADC1 Pong 8 Gbps XAUI1 2 chan * 2 pol * 4 Gsps Clock Out
Thesis Goal: RFI Performance Frequency domain IQ correction specialized for spectroscopy f 0 Corrected Spectrum f 1 FFT or PFB i(t) Filterbank (Real) f n C n = j if i(t) and q(t) were in c 0 perfect quadrature c 1 Looks like it requires n complex FFT or PFB multiplies and adds, but FFTs q(t) Filterbank are pipelined, so only requires (Real) 4-8 plus RAM for c n much more efficient than time domain for same level of image c n rejection
Front Back
Incoherent Dedispersion
Do you see the pulsar? No Dedispersion
BEE2 DRAM Circular Buffer DRAM0 - 1 GB XAUI0 10GbE XAUI1 Up to 20 Gbps DRAM1 – 1 GB 16 Gbps for 2Gsps@8bits 2^k sub- buffers per DIMM, k = 0…8 500ms to ~2ms @ 2 Gsps
Goals Spectroscopy/Polarimetry Currently (iBOB based): • 4096 ch spectrometer single pol • Dual 512 and Single 1024 ch fast dump for pulsar Goal: spectrometer with “zoom” mode • Need to add enable to PFB-FIR block and VACC block Hope to do RFI excision
iADC Stability Tests Preliminary!
-6dBm Anritsu, 3dB modulation
+0dBm Anritsu, 3dB modulation +0dBm noise, 20dB atten before ADC
-6dBm Anritsu, 3dB modulation
9dB modulation, 0dBm Anritsu
6dB modulation, 0dBm, +/-5% scale
6dB modulation -10dBm (10dB below total noise)
No Input
-6dBm Anritsu, 3dB modulation no noise
Goals Pulsar Observations Transient (giant pulse) capture • Incoherent trigger
Recommend
More recommend
