close

Вход

Забыли?

вход по аккаунту

?

Paul Alexander, Peter Hall

код для вставкиСкачать
Design issues and implementation
challenges
Paul Alexander and Peter Hall
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Aim and scope
• Concentrate on the design issues for SKA1
пѓ� SKA1 AA-low is a (transformational) world leading instrument
пѓ� Essential to design for SKA1
пѓ� Consider how to transition to SKA2
• Identify issues which are independent of detailed design
пѓ� Then consider issues which drive detailed design
• Aim is to pose questions that we can aim to make progress on during
the course of this meeting
пѓ� Some questions should be answered
• General point:
пѓ� Transition from a research programme to an instrument project
means we need to retire questions with an accountable path of
how and why the decision was reached.
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA-low
• Excellent learning platforms in pathfinders
– LOFAR, MWA, ...
– Science, engineering, project management, operational lessons
• Why optimize SKA-low?
– Evolving science case
• Possible new specification optimization
– Pathfinders not scalable to SKA-1
• e.g. LOFAR x10 > SKA-1 budget
– Rapid technology changes
• Verify or change long-standing assumptions
– Cost optimization funds new capabilities
• More independent FoVs, increased time domain processing, ...
– Actual SKA site conditions impact SKA-low design significantly
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA-low design
• SKA-low is part of bigger SKA system
– Specifications flow from (updated) SKA Design Reference
Mission
– Performance/cost analysis must be done in SKA design
environment
– Cost must reflect “total cost of ownership”
– SKA environment must capture key AA-lo issues
• SKA operational model is critical to costing, e.g.
– Simultaneity of SKA-low & SKA-mid operations
• Data transmission, signal + post-processing
– Data archiving
– Site infrastructure constraints and costs
• Including energy availability and cost
– Support model, and lifetime costs (“maintenance”)
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Top level issues
Insensitive to detailed design
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Basic specifications
• A low-frequency sparse aperture array with A/Tsys of up to 2000 m2/K
пѓ� At what frequency is this optimised (100MHz?) ?
• Operating at frequencies between 70 and 450 MHz
пѓ� At what range of frequencies is this optimised
пѓ� How tight are the constraints both scientifically and technically?
• Array will be centrally condensed but some of the collecting area will
be in stations located out to a maximum baseline length of 100 km
from the core
пѓ� What fraction of the collector is on longer baselines?
пѓ� How large is the core?
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Possible trade-offs
(cost constrained design)
• Built area vs FoV
– More area, or more accessible and/or processed FoV?
• Accessible bandwidth vs sensitivity
– Fewer compromises in a narrower band array
• Accessible bandwidth vs polarization capability,
polarimetry performance
• Processed FoV, bandwidth vs other parameters
– Optimum investment in data transmission, DSP, computing
– Investment level as a function of time
• U-V coverage vs other parameters
– More stations are costly (e.g. infrastructure, correlation)
– Station numbers and size related to calibration strategy (esp.
ionospheric)
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Frequency range 6.5:1
What frequency range must the array elements be designed/optimised for?
•
Approach 1: Observatory
•
•
•
Approach 2: Observatory, but prioritising EoR
•
Design antenna for good performance in EoR frequency range
•
What is the EoR frequency range 70 – 200 MHz? What about foregrounds?
Approach 3: EoR instrument with observatory function
•
•
Aim for best “average” or “uniform” response across the frequency range
Optimise design for EoR frequency range
Approach 4: Identify the technical difficulties and relax frequency range
•
AAVP 2010
100-450 MHz is only 4.5:1
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Sky coverage
45 degree scan
SKA1
GC
30 degree scan
ALMA
Circumpolar
limit
•
•
•
Critical design driver for element
Observatory requirement – large sky coverage пѓ lower gain antenna larger
scan angle of 45 degrees. What is largest scan angle we would like?
Dedicated EoR experiment perhaps require smaller scan angle пѓ higher
gain antenna possible
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Aside
SKA1 specification is for an amazing
instrument
~ 1 order of magnitude in sensitivity
~ 2-3 orders of magnitude in survey speed
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Sensitivity requirement
Design specification: 2000 m2/K
Tsky (K)
Aeff (km2)
100 MHz
988
2.1
150 MHz
350
0.70
f
•
We will be building approximately a square kilometre of collecting area
•
What sensitivity do we require across the band?
AAVP 2010
•
Very dependent on the frequency at which the array becomes sparse
•
Major impact on element design
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Tailoring the AA system
10000
1000
Fully sampled AA-hi
100
Tsky
Aeff
Becoming sparse
10
Aeff/Tsys
1
100
AA frequency
overlap
Frequency (MHz)
AAVP 2010
Aeff / Tsys (m2 / K)
Sky Brightness Temperature (K)
Sparse AA-lo
f AA
f max
Dish
operation
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA1 sensitivity model
2500
2000 m2/K at 100 MHz
Trec = 60 K
AA sparse above 150 MHz
Sensitivity: Aeff/Tsys m2K-1
2000
1500
SKA1
MeerKAT
LOFAR
1000
ASKAP
eVLA
500
0
10
100
1,000
10,000
100,000
Frequency MHz
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Tsys across the band
Matching
f
Tsky (K)
100 MHz
988
150 MHz
350
180 MHz
221
210 MHz
150
240 MHz
106
400 MHz
29
• Trec important even at 200 MHz
• Dominant at upper end of band
• True low-noise LNAs still important
Challenges: “Matching” across the band to ensure
Trec dominated at upper end and Tsky at lower
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA1 survey speed
1E+09
2000 m2/K at 100 MHz
Trec = 60 K
AA sparse above 150 MHz
NB gives 100 sq degrees across band
100000000
Survey Speed : Sensitivity2*FoV A4K-2deg2
10000000
1000000
100000
SKA1
MeerKAT
10000
LOFAR
ASKAP
1000
eVLA
100
10
1
10
100
1,000
10,000
100,000
Frequency MHz
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Survey speed
• What survey speed do we require at fixed Aeff/Tsys?
• Direct implication for cost of correlator and post-correlator
processing
• See next section for possible trade off
• Upgrade path
• Increasing survey speed is perhaps easiest designed in
upgrade path for AA-low
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Data rate
•
Data rate and survey speed intimately linked
•
Review basic design equations
D
B
Ns Stations
Re-write in terms of FoV and total collecting area
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA1 data rates and configuration
• AA Line experiment 50 AA-low stations
• 100 sq degrees, 10000 channels over 380 MHz bandwidth
пѓ� 3.3 GS/s
• Issues
• What data rate can we process?
• Trade UV coverage (Ns) for FoV and hence survey speed (W)
• Line vs continuum requirements
• What is the longest baseline
• What temperature sensitivity do we need and on what scales
пѓ� Defines filling factor in the core
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA1 configuration
Ideally – do not design in these trade-offs
Need to consider evolution of processing
capability in designing configuration
пѓЁOr even repositionable antenna positions?!
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Station and element design
• One or two elements?
• Many aspects to this – see later
• How sparse can the station be?
• Side lobes even for a random configuration when very sparse
• Complicates imaging, and increases Tsys
• Station size?
• Increasing D пѓ reduced UV coverage, reduced
processing load, less complicated ionospheric
model, move DSP from correlator to station B/F
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Station design
R1
regular
triangular
sparse
R2
R3
thinned
Embedded element
pattern
AAVP 2010
circular
Random
minimum l / 2
random
Random
minimum 2 l
Nima and Eloy
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Configuration, station design
and SKA2
SKA Phase 1 Array Distribution
Mid (30%)
Inner (20%)
• Is SKA1 a subset of SKA2
Core (50%)
• Should we compromise the design
(and hence science return) of SKA1
to ease implementation of SKA2?
500 m
2500 m
180 km
Not to Scale
• Optimum SKA1 AA-low core may have f ~ 0.5 Dcore ~ 1km.
пѓ� SKA2 AA-low core is larger with f ~ 1
пѓ� Almost certainly need to reposition elements on SKA1 пѓ SKA2
Do not compromise design of SKA1 пѓ maximise science
return for SKA1 & accept additional cost in SKA2
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA information and data system
Grid science reduction and
visualisation
Science
proposal
Observation
definition
M&C
database
Global
and local
sky model
Data excision
Correlator
Data excision
Collectors
Monitor and Control system
Cloud store
Data product
distribution
Calibration
loop
Imaging
processor
Science
product
archive
TH_n
12 f ibre lanes
@10Gb/s each
12 f ibre lanes
@10Gb/s each
TH_1
e/o
Local Processing
e.g. Cal; pulsars
…..
Notes:
1. No control network shown
2. Up to 4 station processor systems can
be implemented in parallel
3. Data shown are raw, typ. get 80% data
Data
Visibility
routing processors
To Correlator
TL_1
TL_m
AAVP 2010
Station Processor n
Long distance drivers
1.0 GHz
analogue
Max 4 Station
Processors
10Gb/s f ibre
....
TL_0
Typical:
AA-hi tiles: 300
AA-lo tiles: 45
Total:
345
I/p data rate: 42Tb/s
…..
e/o
…..
e/o
…….
e/o
Tile
e/o
Processor
- lo
Inputs #:
1296
Channel rate: 120Gb/s
(raw)
Total i/p rate: 1.5 Pb/s
…..
…..
1.0-1.4GHz
analogue
Station
Processor
0
………...
o/e
o/e
o/e
o/e
o/e
o/e
o/e
o/e
Long distance drivers
e/o
…..
Tile
Processor
e/o
- hi
e/o
…….
Ae
Ae
e/o
e/o
e/o
e/o
e/
o
e/
o
e/
o
e/
o
…..
o/e
o/e
o/e
o/e
o/e
o/e
TH_0
Long distance drivers
Ae
Hierarchical
station beam
former
Design Issues and Implementation Challenges
Local
science
reduction
Paul Alexander, Peter Hall
Processing – how much
and where?
•
For a given sensitivity and survey speed we can decide where and how to
do the processing
пѓ� Beam forming vs correlation пѓ survey speed vs imaging fidelity?
•
Physical location of processing
пѓ� Physically distribute processing only if it leads to a reduction in data
rate
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Processing – how much
and where?
•
For a given sensitivity and survey speed we can decide where and how to
do the processing
пѓ� Beam forming vs correlation пѓ survey speed vs imaging fidelity?
•
Physical location of processing
пѓ� Physically distribute processing only if it leads to a reduction in data
rate – e.g. Station beamformer
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Specific Design and
Implementation Issues
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Element and communications
•
Can we cover band with a single element?
пѓ� Where are the compromises?
пѓ� Can we afford two elements?
•
Where do we digitise
пѓ� Link, power consumption, lightning protection ...
•
What is the communication link?
пѓ� Cost, calibratability and lightning protection
•
How is the element powered?
пѓ� Cost, sustainabilty, manufacturability and deployability
•
What is the element assembly and how are they deployed?
пѓ� Cost, sustainabilty, manufacturability and deployability
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Station B/F and correlator
•
Station B/F
пѓ� What is, and can we meet the power budget with an all digital
design?
пѓ� Do we deploy ASICs in the SKA1 design? If so what are the
timescales for development cycle.
пѓ� Note cost of Station B/F dominated by number of elements not
how they are deployed (e.g. Station size)
пѓ� Internal station correlation for calibration?
•
Correlator
пѓ� Is a software correlator possible or desirable for SKA1 or
commissioning?
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Post-correlator processing
What is our system concept for SKA1 processing?
o Is the post correlator processing a single Peta-scale
machine or machine designed for our data flow?
Subtract current sky
model from visibilities
using current calibration
model
пѓ� Our problem is highly parallel in places and we
UV processors
Grid UV data to form e.g.
W-projection
could deploy a “UV-processor”
UV data store
Major cycle
пѓ� Need to be sure of processing model to
Image gridded data
go down this route, but can deliver more
Astronomical
quality data
Flops cheaply
Deconvolve imaged data
(minor cycle)
Update
current sky
model
Solve for telescope and
image-plane calibration
model
Update
calibration
model
Imaging processors
пѓ� Single-pass algorithms will reduce cost пѓ do we
want to restrict ourselves in this way?
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Cost control
AAVP 2010
Item
Advantage
Challenge
One element
Single core, one RF
chain
Adequate performance
across band; sparcity at
high f
Larger station size
Reduce cost of
correlator,
infrastructure and postprocessor
Loss of UV coverage
ASICs deployed in DSP
Power reduction saves
on operating budget
Time to deployment,
commissioning harder?
Loss of flexibility
Custom processing
path
Maximise Flops for cost
Loses flexibility
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
SKA-low implementation
challenges
• Low capital cost
– N x 100,000 active antennas пѓ integrated, reproducible
– Strong incentive to incorporate Design for Manufacture early in
development cycle
• Low operating cost
– Easily dominates capital cost over life of SKA
– Reliability and maintainability are crucial
• Probably dominant aspect of designing “outdoor” portion SKA-low
– Robust system is essential
• Damage limitation strategies (lightning etc), intelligent and resilient processing
• Low deployment cost (next slide)
• Data processing and archiving prominent in SKA Observatory plan
• EMC
– SKA-low is especially vulnerable to poor EMC practices, or poor site
management with respect to RFI
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Deployment challenge
• 300,000 elements (or tiles) deployed over 2 years
– 1 element/tile every minute!
• Connectivity and commissioning need to keep pace
with deployment
• Parallel, industrialized deployment needed
– … and during pre-construction
• Substantial site specific and environmental issues
• “Design for deployment” essential
– Results in highly modular, maintainable design
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
AAVP 2010
Design Issues and Implementation Challenges
Paul Alexander, Peter Hall
Документ
Категория
Презентации
Просмотров
6
Размер файла
2 144 Кб
Теги
1/--страниц
Пожаловаться на содержимое документа