The Collaboration for Astronomy Signal Processing and Electronics - - PowerPoint PPT Presentation

▶

the collaboration for astronomy signal processing and

The Collaboration for Astronomy Signal Processing and Electronics - - PowerPoint PPT Presentation

Jul 22, 2023 153 likes •696 views

The Collaboration for Astronomy Signal Processing and Electronics Research in 2017 Jack Hickish Radio Astronomy Lab, UC Berkeley jackh@berkeley.edu September 7, 2016 Outline Acknowledgements The Age of Digital Radio Astronomy Building DSP

slide-1

SLIDE 1

The Collaboration for Astronomy Signal Processing and Electronics Research in 2017

Jack Hickish

Radio Astronomy Lab, UC Berkeley jackh@berkeley.edu

September 7, 2016

slide-2

SLIDE 2

Outline

Acknowledgements The Age of Digital Radio Astronomy Building DSP systems for Radio Telescopes CASPER CASPER Hardware CASPER Hardware in Action Looking Forward Conclusions

2 of 41

slide-3

SLIDE 3

Acknowledgements

§ Local Organizing Committee

§ Michael Burke § Dave Hawkins § Ryan Monroe § Jonathon Kocz § Terry Filiba Schrager § Melissa Soriano § Cody Vaudrin

§ Busyweek-ers

§ Adam Isaacson § Jonathon Kocz § Wes New § Amish Patel 3 of 41

slide-4

SLIDE 4

The Age of Digital Radio Astronomy

4 of 41

slide-5

SLIDE 5

The Age of Digital Radio Astronomy

5 of 41

slide-6

SLIDE 6

More is more. Best is best.

§ More bandwidth § More field-of-view § More sensitivity § More resolution (time, frequency, spatial)

6 of 41

slide-7

SLIDE 7

Moore’s law in correlators

Figure: Credit: Mel Wright (https://casper.berkeley.edu/wiki/Videos)

7 of 41

slide-8

SLIDE 8

Moore’s law in spectrometers

8 of 41

slide-9

SLIDE 9

Radio-Astronomy DSP Parameters

§ Number of antennas (1 Ñ„ 1000) § Bandwidth (100 MHz Ñ 100 GHz) § Frequency channels („ 210 Ñ„ 230+) § Averaging period (None Ñ 10 s) § Many Tb/s interconnect, many complex Tops/s

“The data collected by the SKA in a single day would take nearly two million years to playback on an ipod.” https://www.skatelescope.org/amazingfacts/

9 of 41

slide-10

SLIDE 10

Building DSP systems for Radio Telescopes

§ Large data input rates § Large computation rates § Simple operations § Many common components between different telescopes § Highly parallel

10 of 41

slide-11

SLIDE 11

Radio Astronomy DSP Parameters

A large variety of Radio Astronomy instruments can be build from a small number of parameterised

§ Filters (F-Engines) § Correlation Engines (X-Engines) § Beamforming Engines (B-Engines) § Interconnect

11 of 41

slide-12

SLIDE 12

A Simple Spectrometer

12 of 41

slide-13

SLIDE 13

A Multi-Antenna System

13 of 41

slide-14

SLIDE 14

A Multi-Antenna System

14 of 41

slide-15

SLIDE 15

A Multi-User System

15 of 41

slide-16

SLIDE 16

CASPER

Center for Astronomy Signal Processing and Electronics Research

16 of 41

slide-17

SLIDE 17

CASPER

Center for Astronomy Signal Processing and Electronics Research Collaboration

16 of 41

slide-18

SLIDE 18

CASPER

Center for Astronomy Signal Processing and Electronics Research Collaboration Community?

16 of 41

slide-19

SLIDE 19

CASPER

“The primary goal of CASPER is to streamline and simplify the design flow of radio astronomy instrumentation by promoting design reuse through the development of platform-independent, open-source hardware and software.”

17 of 41

slide-20

SLIDE 20

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-21

SLIDE 21

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-22

SLIDE 22

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-23

SLIDE 23

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-24

SLIDE 24

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-25

SLIDE 25

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-26

SLIDE 26

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-27

SLIDE 27

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-28

SLIDE 28

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-29

SLIDE 29

CASPER

§ Simplify

§ Leverage industry standards (eg, Ethernet for interconnect) § Small number of custom [FPGA] platforms § Optimize for ease of use (not ops/watt, ops/rack unit) § Low knowledge-barrier for users

§ Re-use

§ General purpose hardware § General Purpose libraries § Modular, upgradable piecemeal § Flexible, scalable architectures 18 of 41

slide-30

SLIDE 30

Outline

Acknowledgements The Age of Digital Radio Astronomy Building DSP systems for Radio Telescopes CASPER CASPER Hardware CASPER Hardware in Action Looking Forward Conclusions

19 of 41

slide-31

SLIDE 31

BEE2 (Virtex 2 Pro) 2005—

20 of 41

slide-32

SLIDE 32

iBOB (Virtex 2 Pro) 2005—

21 of 41

slide-33

SLIDE 33

ROACH (Virtex 5 SX95T) 2009—

22 of 41

slide-34

SLIDE 34

ROACH2 (Virtex 6 SX475T) 2010—

23 of 41

slide-35

SLIDE 35

SNAP (Kintex 7 160T/325T/410T) 2016—

§ 600-1540 DSPs § 3 onboard

HMCAD1511 digitizers

§ 3x1 Gsps / 6x500 Msps

/ 12x250 Msps

§ 1 x ZDOK § 2 x 10 GbE IO § approx. $3k

24 of 41

slide-36

SLIDE 36

SKARAB (Virtex 7 690T) 2016—

§ 3600 DSP slices § 4 mezzanine card sites § HMC high-bandwidth

memory

§ up to 16 x 40 GbE

interfaces

25 of 41

slide-37

SLIDE 37

MX175 (Virtex 7 690T + 2xHMCAD5831 ADC) 2016—

§ 2 onboard 26 Gsps

digitizers

§ Same FPGA as

SKARAB

§ ? 4 x 40 GbE IO

26 of 41

slide-38

SLIDE 38

SNAP2 (Kintex Ultrascale KU115) 2017—

§ 160 Gb/s IO § Expansion card up to

« ∞ Gb/s

§ FMC interfaces – some

ADCs in development

§ FMC - ZDOK adapter § 5520 DSP Slices § Est. $15k

27 of 41

slide-39

SLIDE 39

COTS options—

§ NetFPGA-SUME § Virtex 7 690T § 3600 DSP Slices § 4 x 10 GbE § DDR3 & QDR § $6995 § Alpha Data

ADM-PCIE-7V3

§ Virtex 7 690T § 3600 DSP Slices § 2 x 10 GbE § 2 x 8 GB DDR3 § $3200 § Alpha Data

ADM-PCIE-KU3

§ Kintex U.S. KU60 § 2760 DSP slices § 2 x 40 GbE § 2 x 8 GB DDR3 § $2795

28 of 41

slide-40

SLIDE 40

COTS options—

DINIGROUP: “Uncle of Godzilla’s Bad Hair Day”

29 of 41

slide-41

SLIDE 41

GPUs

Fig. credit: Trusted Reviews

30 of 41

slide-42

SLIDE 42

"Switches are free"

31 of 41

slide-43

SLIDE 43

Outline

Acknowledgements The Age of Digital Radio Astronomy Building DSP systems for Radio Telescopes CASPER CASPER Hardware CASPER Hardware in Action Looking Forward Conclusions

32 of 41

slide-44

SLIDE 44

World Domination

33 of 41

slide-45

SLIDE 45

World Domination

Spectrometers: Fly’s Eye, GUPPI, CASPSR, BPSR, GAVRT, SERENDIP V.v, HiTREKS, Skynet, RATTY, cycSpec, C-BASS, HIPSR, KuPol, VEGAS, ALMA Phasing Project, Leuschner, R2DBE, DSN Transient Observatory, VGOS, AVN-Ghana, COMAP

34 of 41

slide-46

SLIDE 46

World Domination

Correlators & Beamformers: KAT7, PAPER, ATA, LEDA, ARI, MAD, Medicina FFTT, GMRT, MITEoR, AMI, MeerKAT, FLAG, BIRALES, Starburst, AMiBA, EOVSA, SWARM, HERA

35 of 41

slide-47

SLIDE 47

How? In a nutshell

36 of 41

slide-48

SLIDE 48

Outline

Acknowledgements The Age of Digital Radio Astronomy Building DSP systems for Radio Telescopes CASPER CASPER Hardware CASPER Hardware in Action Looking Forward Conclusions

37 of 41

slide-49

SLIDE 49

Not all roses

§ Documentation is too sparse § Simulink is intuitive, but it is sloooooow (and enraging). § Severe version compatiblity headaches. § Poor version control support. § Diverging development at different institutions § Unit tests § Where are my new boards?!?!

38 of 41

slide-50

SLIDE 50

The Future (at Berkeley)

§ More co-ordination of developers (retreats in France?) § More documentation. I actually promise. § Faster paths to supporting new hardware § Standardizing beyond the toolflow (pipelines, control software, etc.) § In need of people to help. § Please help.

39 of 41

slide-51

SLIDE 51

Conclusions

§ Getting everyone using [approximately] the same hardware /

software has been a huge victory.

§ There’s loads of cheap hardware around (not just FPGAs)! § COTs hardware is our focus in Berkeley. § The CASPER toolflow is often liked in principle, but less so in

practice. Hopefully we’ve made some steps to improve the lives of

users and developers.

§ Eager for fresh blood

40 of 41

slide-52

SLIDE 52

Thanks

Thanks (and enjoy the workshop) once developers have met you, it’s much harder for them to ignore your emails.

41 of 41