Bluering Prototype System Results G. Hampson, W. Cheng, D. Humphrey, J. Bunton, P. Roberts, K. Bengston, R. Beresford, Y. Chen, R. Chekkala 22 nd September 2020 RadioNet Workshop: Future Trends in Radio Astronomy Instrumentation
Bluering Aims ● To develop a generic receiver ○ Using “RFSOC” = Radio Frequency System On Chip ○ Tailored RF front end to suit most types of astronomy antenna systems ○ Coax connected to enable high dynamic range RF front ends that can be operated with RFI ● RFSOC signal processing capabilities - a powerful DSP FPGA too ○ Can beamform at the antenna (if only limited # of beams) ○ Can calculate correlations, time pulsars, and RFI mitigation ● Integration results in a cost, size and power efficient receiver ● Optical connectivity to backend ○ Optical clocks and control ○ Optical data path out ● This presentation shows how far we have come ...
Bluering System View ● Each Bluering system contains a number of “Razorbacks” (containing the RFSOC), plus supporting equipment ● There are few central services required to operate the Bluering systems White AC-to-48V Rabbit Razorback Master (RFSOC) Razorback Cooling Coax (RFSOC) Fibre M&C # Server WR Slave (2 per (WR, antenna) 1GbE, * GbE) Optical Circuit Irukandji Link to DSP 19” Rack Backend Remote Systems
Bluering = Taipan + Razorback + Irukandji ● Taipan is a customisable RF module ● Razorback contains the Xilinx RFSOC plus: ○ Synthesizer for ADC sample clock and LO generation ○ DDR4 for transient data capture (raw, filterbank, beam, …) ○ MBO for 10GbE ring/star beamformer data, 1GbE, timing and calibration signals ○ Q/SFP for beam output (short or long distance) To beamformer/ ● Irukandji distributes correlator/ ○ Timing, Calibration, 1GbE (M&C) capture Network, Timing, RAZORBACK Calibration Synth Q/SFP 32 RF Chains Taipan 32 RF DC/DC Irukandji MBO RFSOC Optical Signals Beamformer (coax) 48/ Network 12V DDR4 48V (Isolated)
Razorback Board ● Clock and LO generation x8x2 and distribution ● RF inputs (Taipan) ○ 16 RF customisable modules (coax/RFOF, frequency, gain, etc.) ● Optical outputs ○ 40/100GbE QSFP ○ 10/25GbE SFP ○ 12x10GbE MBO ● RFSOC FPGA ○ 2560 channel oversampled filterbank ○ 2GSPS results in 781kHz channels (926kSPS) ● Optical inputs ○ 1GbE M&C (SFP) ○ Reference (125MHz) ○ 1 pulse-per-second ○ Calibration signal ● 8GB DDR4 memory ○ Also SDcard and 8GB iNAND ● 48V power
Taipan RF Card for MWA antenna ● Very simplified (!!!) RF signal chain diagram shown below ● Taipan powers each antenna’s LNA - no extra cables Taipan ● 22dB of sky brightness slope is partially compensated to extend dynamic range - a 12dB improvement ● All 12 RFSOC ADC bits processed in beamformer DSP ● The cost per RFSOC ADC input is halved by using a mixer architecture to sample both polarisations in the same ADC 125MHz Reference PLL PLL 875MHz 2000MHz Calibration Signal C LPF BPF RFSOC To Dual Pol Antenna (power over coax) ADC 12-bits C LPF to DSP Taipan RF board x16 for RFSOC
Taipan RF Module ● Taipan module is 101.5mm x 30.5mm ○ (purple hashed section in drawing) ● Several ports around the module provide RF, LO, control, 5V and ground ● 4-bits of attenuator control are provided for each polarisation ● Soldered in place Y-attenuator LO 5V-LO RF X 5V-RF Taipan 5V-LNA RF out (to final RF Y amplifier and ADC) Cal X-attenuator
MWA Taipan Module ● Prototype version ● Two low cost shields ($1) for first and second stage amplification ● Razorback contains final amp and balun
Taipan Signal Levels ADC Full Scale +1dBm ● RFSOC ADC performance ○ SNR is -150dBFS/Hz or 32dB -150+90 = -60dBFS/300MHz Average ADC SNR 60dB 17dB RFI ○ SFDR is 85dB ○ https://www.xilinx.com/support/docu ADC IM3 78dB ADC SFDR 85dB Sky noise Slope ADC IM2 85dB mentation/data_sheets/ds926-zynq- (300MHz BW) Sky 1MHz ultrascale-plus-rfsoc.pdf RFI IM2 spurious -50dBc RFI IM3 spurious -89dBc ● ADC similar to RF 25dB 32dB spurious Cal Signal ○ IM2 just as important as IM3 Sky (1MHz BW) ADC Noise ● Attenuator set to 17dB so ADC input level for 13dB 300MHz Sky is -33dBm ADC Spurious ○ Can still adjust up and down RFI Spurious further with attenuator depending on RFI
RFI measurements ● North of Sydney (Wamberal) ● Near Australia Telescope (Yarrie Lake) ● Greatest concern (in band) is FM radio ○ Seems to be everywhere - even the MWA site has FM at times Spatial filtering experiment with the Murchison Widefield Array - G.Hellbourg & I. Morrison, URSI GASS 2020
Razorback RFSOC Firmware ● “Standard” Bluering firmware and software ● Can be customised for any application Test Test Signals Packets Coarse Quantise, 2GS 10/100 16-RF Filter Select & Beamformer ADC GbE bank Packetise 1pps /BAT ADC ADC FB Sync Capture Capture 128MHz To ADC Spectrum SysMon PLL Statistics Maxhold monitoring Irukandji Attenuators Processor IO 1GbE & AXI IF
Bluering RFSOC Synchronisation High ECUO Calibration Isolation Firefly Taipan Resistive Rx RF RF Splitter Receivers LMX2594 125MHz 875 SC4PS-33+ RF MHz RF Splitter Synthesizer ECUO LMK04610 125MHz 125MHz Firefly Clock Jitter Rx Cleaner 125MHz LMK00304 LMX2594 ZU29DR 2GHz Differential RF ADC Clock 125MHz Synthesizer Clocks Buffer CVHD-950 Ultra-Low LMK00304 ZU29DR Sync Phase Noise Differential SysRef Pulse Oscillator Clock & Logic ECUO Sync 1pps Buffer ZU29DR Firefly Manchester Pulse Decode Rx Encoded
Filterbank Compute ● 2560 point filterbank, oversampled by 32/27, with 16x2560 FIR taps ○ Using sample clock 2000MHz results in 721kHz channels ○ Can be modified to suit your application ● Filterbank FPGA resources ○ LUTs : 122911 (29%) ○ DSPs : 2112 (49%) ○ 36K BRAMs : 568 (53%) ● Filterbank use about ~40W ○ Measure slightly less (data dependent)
Liquid Cooling ● EMI shielding combined with liquid cooling ● Encouraging results so far - no leaks - good performance
EMI Shielding ● A critical aspect of Bluering is EMI shielding as its only ~1m from an antenna. ● This box achieves >90dB shielding effectiveness in the single shielded side and >150dB in the double ○ Gets very hard to measure ● 48V power is filtered many times
First RFSOC ADC samples ● The RFSOC has an endless number of configurations - so getting the first ADC samples out is a particularly pleasing moment in time! ● Here an unfiltered 200MHz signal (that is locked to the 125MHz reference clock) is input into the ADC ● The location of the noise is ~right ○ RF in is -20dB and ADC SNR is 60dB, so 80dB about 40dB above noise floor ○ But processing gain reduce level of noise by 10log10(16384) = 42dB ○ So peak to noise floor is 40+42=~82dB ○ Measure 155-75=80dB so levels close to right for a first measurement
Upgrades being made now ● As with all prototypes there a few bugs! ○ Wrong values, wrong side, or just missing ● Major upgrades for the optics ○ Originally MBO + QSFP + SFP cages ○ Moving to quad QSFP - similar to Jimble (another CSIRO RFSOC receiver) ○ Enables more data output options ○ QSFP’s are lower cost than MBO now - as it's more of a COTS product ● RF connector ○ Moving away from ganged RF connector to SMA ○ Using a surface mount connector to improve RF over wider frequency range ● Taipan module ○ Originally soldered in and potentially difficult to replace ○ Now moving to MMBX solution to enable easy replacement/upgrades ● Second version available for Christmas
Prototyping Continues Many thanks to Wan, Keith and David
What are we planning at CSIRO? ● After completing the second revision of boards we intend to do installation at the Australia Telescope ● We have two MWA tiles (each MWA tile requires one Razorback) to install ● Hoping to use power and fibre connections available at positions along the array arms
Summary & Conclusions ● Only a small team of people work on Bluering - but it has come a long way ● Steady progress on RF side - greatest concern is FM radio ● RFSOC software and firmware has come along way ● First ADC results are very promising ● We still have a lot to do! ● Really looking forward to demonstrating all aspects of Bluering at the Australia Telescope https://i-love-png.com/images/octopus-tentacles-png-photos.png
Recommend
More recommend
