N. Pingel, K. Rajwade, D.J. Pisano, D. Lorimer West Virginia University

Transcription

1 Brian D. Jeffs, R. Black, J. Diao, M. Ruzindanna, K. Warnick Brigham Young University R. Prestage, J. Ford, S. White, R. Simon, W. Shillue, A. Roshi, V. Van Tonder NRAO: Green Bank Observatory and Central Development Laboratory N. Pingel, K. Rajwade, D.J. Pisano, D. Lorimer West Virginia University NaGonal Science FoundaGon Award

2 FLAG SpecificaGons Overview 19 dual polarized elements,.7λ hex grid spacing 15 MHz instantaneous BW in the range of 1.3 to 1.8 GHz Cryogenically cooled LNAs, room temperature antennas 7 real- Gme beams with XX *, YY * & XY * pol. Many beams using post- correlagon beamforming (PCB) FPGA - GPU real- Gme digital receiver/correlator/ beamformer back end.

3 FLAG SpecificaGons Overview 5 coarse frequency channels 33 khz BW each, 15 Mhz total for PCB and real- Gme beamformer Fine polyphase filter bank: 32 channels, 9.5 khz BW each, 3 MHz total for PCB Commensal transient search and HI survey modes 9 arcmin beamwidth, 3 beam- width field of view diameter Element design opgmized for best sensigvity and average acgve array impedance match across beams

4 FLAG SpecificaGons Overview Digital back end: 5 ROACH II FPGAs for DDL digital fiber communicagon and PFB frequency channelizagon 5 high performance PCs 1 GPUS for fine PFB, real- Gme beam- former, and real- Gme correlator Ideal beamformer architecture for RFI miggagon

5 Phase I Commissioning Goals Tests over Jul 5 31 at Green Bank Outdoor Test Facility and on GBT First light demonstragon Hardware, firmware, and sojware integragon and funcgonal tests Full- up array, digital downlink, FPGA 2- HPC reduced BW 6 MHz PCB correlator IdenGfy and debug issues

6 Future Commissioning Goals Phase II: November / December 216 Phase I hardware configuragon (6 MHz BW, 2 HPC) Performance characterizagon for a range of HI and pulsar observagons Coarse channel operagon only Real- Gme beamformer tests EvaluaGon of sensigvity, T sys, beampalerns, and calibragon Phase III: Final commissioning, May / June 217 Full bandwdith back end: 15 MHz Commensal transient detector real- Gme beamformer and PCB fine channel HI observagon

7 FLAG Block Diagram Ant. Ant. LN A Cryostat LN A I- Q mix, ADC, Serialize & OpGcal Xmit ( 4) LO I- Q mix, ADC, Serialize & OpGcal Xmit LO Array aperture, Antenna elements, LNAs, Cryo system, Down converters Ch. 1 Ch. 8 Ch. 33 Ch. 4 Signal Transport: GBT prime focus to Jansky Lab fiber link 8 Fiber Digital OpGcal Rcvr Card 8 Fiber Digital OpGcal Rcvr Card ROACH II FPGA ( 5) ROACH II FPGA In Mezzanine I/F Slot 4 x 1 Gbe I/F Card 4 x 1 Gbe I/F Card F Engine: DDL deserializagon, boun- dary alignment, polyphase filter bank and 1 Gbe I/O 4 X 1 GbE 4 Gbe Ethernet Switch Melanox SX X 4 GbE ports Rack Mount PC 12 TB SATA RAID Disk Array 4 X 1 GbE 4 GbE System control and data storage CPU/GPU (Blade server + 2 nvidia GTX79) ( 5) CPU/GPU (Blade server + 2 nvidia GTX79) XB Engine: Correlator/ Beamformer, Spectrometer NRAO DDL System BYU Correlator Beamformer

8 Digital Back End

9 First- of- its- kind Architecture Packets from 5 ROACH IIs: 25 out of 5 channels from all 4 input ports Ethernet Switch Melanox SX X 1 GbE Hashpipe instance 2 Hashpipe instance 1 3 Channel SelecGon (k =... ) Channel SelecGon (k =... ) All 25 Coarse Channels, 7.5 MHz total BW All 25 samp/s Channels, 7.5 MHz total samp/s Fine PFB 32 point FFT, 256 tap 32 point Filter FFT, 256 tap 5 Selected Coarse Filter Channels Fine PFB Real- Gme Beamformer b k, Real- Gme j [n] = w Beamformer H k, j x k [n] b k, j [n] = Coarse/ Fine 5 Selected Coarse Channels 1 w H k, j x k [n] Correlator / Integrator Coarse/ (XGPU code) Fine x k [n] R k = 1 N 1 x k [n] x H k [n] (XGPU N code) n= Integrator c S k,( j, j ) = 1Integrator N 1 * c b k, j b k, j NS k,( j, j ) = N 1 1 * N = 3 for b k, ms N 1 j b k, j x k [n] Coarse: R k N = 1 3 for x k [n] ms x H k [n] N 16 Fine Fine: N = 4,75 n=for 5 ms Channels, Coarse: N = 3 for ms 1.51 MHz total samp/s 16 Fine Fine: N = 4,75 for 5 ms Channels, 1.51 MHz total samp/s n=1 Correlator / Integrator N 1 n=1 N = 3 for ms R k Channel SelecGon (k =... ) Coarse: 5 of 5 coarse channels Fine: Channel All 16 fine channels SelecGon (k =... ) RPost- k Coarse: CorrelaGon 5 to 25 coarse Beamformer channels f S k,( Fine: j, j ) = All 16 fine w H k, channels j R k w k, j c S k,( j, j ) R k Nvidia GTX79 Ti GPU, 1 of 2 per HPC 1 2 FITS Formaler Lustre Disk Array Storage To 4 other HPCs Nvidia GTX79 Ti GPU, 2 of 2 per HPC

10 First- of- its- kind Architecture PCB is the primary mode for all observagons: Array covariance, R k, is the output data product Correlator STI dump rates: T STI 1 ms: store R k for all channels T STI < 1 ms: store R k for fewer selected channels PCB permits: f S k,( j, j ) = w H k, j R k w k, j RevisiGng observagons with different beampalerns Recovery from bad calibragon, poor beams Arbitrarily dense beam spacing for radio camera AdapGve beamforming RFI miggagon ajer the fact!

11 First- of- its- kind Architecture PCB beamforming on coarse or fine channels Coarse channel correlator for beamformer weight calibragon, one weight vector w k per channel, k. Fine channel PCB for HI observagon Real- Gme beamforming spectrometer Runs always: commensal operagon with PCB Ultra fast dump Gmes for transient detecgon Coarse channels only, full 15 MHz BW. Can couple with correlator for rapid w k updates, real- Gme adapgve moving RFI cancelagon!

12 BYU s FLAG Array Element Design

13 Antenna Design Process Design Object FuncGon (field of view, bandwidth and sensigvity) Beamforming Algorithm OpGmizaGon (Several months computer Gme) 19- element array geometrical parameters Network Model ElectromagneGc Model (Finite Element Method) Reflector Model (Physical OpGcs Method)

14 19 Element S Parameter Test -5 Test HFSS -15 S11 (db) Test Picture Frequency (GHz) -5 Test HFSS S22 (db) -15 Manufacture Errors Frequency (GHz)

15 Numerical Results with Noise Model 19 element array feed on GBT reflector antenna (modeled performance): System noise temperature Inverse sensigvity

16 Commissioning Challenges Some dead elements, primarily due to problems with the DDL digital opgcal downlink OverheaGng, nighyme operagon only Limited observing Gme, just a few data sets Not able yet to nail down true T sys and sensigvity Verified operagon for most system components and got some early data Not bad for a first integragon experiment

17 Commissioning Challenges Some dead elements, primarily due to problems with the DDL digital opgcal downlink OverheaGng, nighyme operagon only Limited observing Gme, just a few data sets Not able yet to nail down true T sys and sensigvity Verified operagon for most system components and got some early data Not bad for a first integragon experiment

18 Results: CalibraGon Grid Grid centered on 3C295: ,465 Mhz Slow serpengne offset trajectory gives a fine grid of calibragon points Covariance matrix for each 5s STI R θ Used to calculate steering vectors, beamformer weights, and beampalerns Elevation Offset (degrees) R θ Scan Trajectory Azimuth Offset (degrees) R θ U = UΛ, U = [u 1,!, u M ] R off CalibraGon Steering vector a θ = u 1

19 Results: On- sky palerns per element 1Y 2Y 3Y 4Y 5Y Y 7Y 8Y 9Y 1Y Y 12Y 1 13Y 14Y 1 15Y Y 17Y 18Y 19Y MHz

20 Results: Beam Palerns Beam 1 Beam Elevation Offset (degrees) Elevation Offset (degrees) Max SNR beamformer 1391 MHz Beam Azimuth Offset (degrees) Beam Azimuth Offset (degrees) Beam Elevation Offset (degrees) Azimuth Offset (degrees) -4-5 Elevation Offset (degrees) Beam Azimuth Offset (degrees) Elevation Offset (degrees) Beam Azimuth Offset (degrees) Elevation Offset (degrees) Elevation Offset (degrees) w θ = R 1 off a θ b θ (φ) 2 = w θ H R φ w θ Azimuth Offset (degrees) Azimuth Offset (degrees)

21 Results: SensiGvity over FOV Formed Beam Sensitivity Map 9 Elevation Offset (degrees) S(θ) = 2k b 1 26 F s SNR(θ) SNR(θ) = w H θ R θ w θ w H θ R off w θ w H θ R off w θ m 2 / K Azimuth Offset (degrees)

22 Results: T sys vs Frequency, preliminary data Tsys (K) Minimum Tsys in FoV Assumes η a =.65 May be corrupted by sources in the only off- poingng available Higher due to fixable DDL bit errors and missing elements These hardware failures cause poor Frequency (MHz) illuminagon palern

23 Conclusions An encouraging first light and first look with FLAG SoluGon to DDL bit errors and longer observing windows needed for next commissioning session Data rate for PCB covariance matrix output to Lustre file store is easily sustainable PCB works! We are excited for its future promise.