1. RadioNet FP7 Joint Research Activity
The UniBoard
RadioNet FP7 Joint Research Activity
Arpad Szomoru, JIVE
Contract no

2. Overview Some history, background Motivation Project setup
Current state
Future...

3. RadioNet FP7
RadioNet-FP7 has ~ 20 partners: all of the major radio astronomy facilities and the laboratories involved in technology development
10 M€ over 3 years
Has started January 2009

4. 3 decades of cooperatio, recording on tapes, central supercomputer, made to play tapes
TODO: is animation quick enough?

5. Miyun 50m
Irbene 32m
Sardinia 64m
Yebes 40m
Kunming 40m

7. EVN2015 roadmap
Science case:
Some highlights:
Nature of starburst/AGN in cosmological fields
The fate of black holes/radio quiet AGN
Jet physics close to the event horizon (VSOP2)
Determining star burst activity, resolving SNR’s
The accretion physics in transient radio sources
The detailed 3D kinematics of star formation
The nature of the ISM in active galaxies
Fundamental distances from astrometry
Pulsar astrometry
Monitoring spacecraft in the solar system

8. Next generation correlator
Requirements set by EVN2015 science case
Multiple data streams of Gbps
32 stations, or at least more than 16
Multiple bit representation
High bandwidth at higher frequency
Calls for a hundredfold more powerful correlator
Comparable to EVLA correlator
Similar size as some (other) SKA pathfinders
FPGA vs software
Software correlators could be intermediate solution
Distributed correlation attractive in Europe
WSRT APERTIF system; similar needs

9. Software correlators FX software correlation: EVLA
Small number of complex operations
Easy to implement in software
Better accuracy with floating-point
EVN MarkIV
16 stations, 8 bands, 4 pols,
32 spectral pts, 1 Gb/s
1.8 TFlops
EVLA
27 stations, 4 bands, 4 pols,
128 spectral pts, 32 Gb/s
~150 TFlops
So a top 10 supercomputer can do this!

10. APERTIF project Phased Array Feeds for 12 of 14 Westerbork 25 m dishes
8 x 9 dual polarized Vivaldi arrays
25 beams, 300 MHz bandwidth, full Stokes
Aeff/Tsys > 100 m2/K (Ntel=14 ha=0.75 Tsys=50 K)
8 deg2 Field of View
Frequency range: 1000 – 1750 MHz

11. APERTIF prototype APERTIF prototype One dish fully dedicated to PAF
Stand alone system (so far)
8 x 7 x 2 elements Vivaldi array
Dual polarisation
60 Receiving chains
Frequency range 1.0 – 1.7 GHz
30 MHz bandwidth
Element separation: 10 cm
1.5 GHz)
Data recording backend (6.7 s)
Output is full covariance matrix

12. Measured compound beam

13. First FPA synthesis image using Digestif and WSRT

14. First version (2006): CORFU
Or, correlator of the future
(or, a very nice island in Greece)
An investigation into:
Software correlator on a commodity hardware cluster
Grid-based distributed software correlator
Software correlator on integrated cluster, commodity or specialized (e.g. Blue Gene)
Hardware correlator using new technologies (e.g. IBM Cell Processor Architecture)
FPGA based hardware correlator
“Traditional” ASIC based correlator.

15. Second version (2007): Uniboard
A multi-purpose scaleable computing platform for Radio Astronomy
FPGA-based processing platform
Put as much computing power and as much I/O as possible on a as generic as possible board (Pogrebenko melt-down criterion)
Several applications:
Correlator (EVN, Apertif)
Pulsar machine
Digital backend
Participants:
JIVE, ASTRON, INAF, Bordeaux, Orleans, Manchester, KASI

16. FPGA based correlation
Several high-end FPGAs/board
Few Tops per board
Current EVN correlator on 1 board
Next generation correlator would require several racks
Power consumption of several 10s of kW
compared to MW range for supercomputers
Rapid high level development possible
e.g. Simulink (CASPER Berkeley)

17. Project setup
Hardware development + three separate applications (in fact four)
Concentrate efforts, minimize interdependencies
Ensure al least one FTE for duration of project at main institute
More than 50% matching contribution from partners
Hardware: ASTRON + BORD + KASI
Correlator: JIVE, ASTRON + BORD
Digital Backend: INAF + BORD
Pulsar Binning (+ RFI mitigation) : UMAN + UORL

18. Project status, activities to date
Most manpower in place
Regular telecons (started November 2008) to determine technical demands of different applications
Regular Jive-Astron meetings
Kickoff meeting in Dwingeloo February 26-27
Several draft design and use-case documents on wiki
Hardware design (nearly) finalised
Negotiations with Xilinx and Altera finished

19. Hardware selection Both Xilinx and Altera now use 40 nm Availability
Price
Layout

20. Correlator construction kit

21. And what about ExBox? Expandable Box for X-correlation
JIVE/ASTRON proposal to NWO
Build prototype FPGA EVN/APERTIF correlator
Using LOFAR expertise and technology
Backplane with several boards
Matching funds for UniBoard
First proposal deemed “excellent”
No money
Second proposal “excellent” as well
Funded at half the requested amount

22. Next 13 months In place: Hardware decision has been made
Funtional design document digital receiver
Hardware, software design and use-case documents
Realistic system design: flexible, multi-purpose, powerful, compact
Hardware decision has been made
Board design and implementation (2009)
Production first half 2010
Firmware design and (partial) implementation
Basic control software, first version correlator control software
Firmware repository

23. The future: possible applications
Next generation EVN correlator
APERTIF beamformer
APERTIF correlator
Digital receiver
Pulsar binning machine (EPTA)
Next gen. Korean e-VLBI correlator
Validation platform broadband digitization
Low frequency resolution correlator
Next gen. IRAM demonstrator correlator
Solar interferometer
Next gen. Chinese e-VLBI correlator
prepSKA correlator effort

24. Longer term How does UniBoard fit in with the SKA timeline?
2012 pathfinders complete
<2015 pathfinder science
>2015 phase 1 science