MASSIVE Overview

1
MASSIVE Overview

• Laurentiu Nicolae

2
A distributed radio telescope LOFAR
3
How does it work
Matlab algorithms
Input data - signal hidden in the noise
Output data - interference suppressed and signal
correlated
4
Design space exploration strategy
Requirements
System specification
System realization
System-level decomposition and exploration
System-level composition
Subsystem exploration
Subsystem validation
Integration and calibration
5
The DSP station model
6
The MASSIVE Exploration Framework
7
Topology specification
8
Scenario specification
9
Performance Model
W, , Size usage
Architecture errors With error margins
10
Topology manipulation

• Detection of several topology variations
• mapping modifications
• replacing and / or renaming elements
• subsystems derivation
• repartitioning

11
Automatic line mapping detection

• Automatic line mapping detection
• loosely based on Floyd Warshalls shortest
  path algorithm
• simplifies the mapping process a great deal

12
The Massive Exploration Framework

• The application and architecture constraints
  were introduced manually by the designer in the
  IVL
• We want to automate the process and link the
  constraints to the top-level requirements

13
Constraint derivation validation
14
Constraints processor
Given the application constraints and the
mapping, we can derive the component level
constraints and back-track all the way to the
architecture top level.
15
Constraints derivation example
e 1
16
LOFAR architecture topology
Dedicated Platforms
Uniform Node Cluster
17
LOFAR multiplicity of nodes
Uniform Node Cluster
Less flexibility 3 clusters topology
18
LOFAR arithmetic representation
Dynamic processing Limits ?
19
LOFAR reconfigure ability

• Hardcopy FPGA for subband filtering
• FGPA for the other processing intensive tasks

20
LOFAR system requirements
Bandwidth of sub bands Number of sub bands
Then 30 cost increase
Optimum for platform 128 sub bands 256 kHz
21
Thank you.