Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
An Open Architecture, Server-
Based Solution for Next Generation
Electronic Warfare System
19 March 2015
Copyright 2015 Lockheed Martin Corporation: No part of this publication may be reproduced or transmitted in any form, or by
any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without
written permission from Lockheed Martin.
Greg Gannett
Outline
Electronic Warfare Overview
IR&D Backstory
System Architecture
Network and GPU Performance
Copyright 2015 Lockheed Martin Corporation 2
The EW Battlespace & RF Spectrum Challenge
Copyright 2015 Lockheed Martin Corporation 3
• Our Customers Need The Ability to Detect, Analyze, and React to Emitters in the Battlefield
• Modern EW Systems need to measure and report the entire RF Spectrum
• The RF spectrum is both wide and dense
• The emitters are diverse : commercial and military
• Emitters are becoming more agile with the landscape constantly changing
Previous approaches of selective listening on known emitters are not adequate
VHF UHF SHF EHF
300 MHz 100 MHz 4 GHz 6 GHz
S C X
8 GHz
Ku K
12.5 GHz 18.0 GHz 26.5 GHz 40.0 GHz
Ka
30.0 GHz
L
2 GHz 1 GHz
V
75.0 GHz
W
Example emitters are representative and randomly placed on RF spectrum
EW System Architecture - Basic
Copyright 2015 Lockheed Martin Corporation 4
RF
Time
Ampl
Pulse Detection &
Measurement
Clustering
Deinterleaving
Parameter Measurement
Emitter Tracking
Emitter Classification
FLOPS Breakdown
Digitized IF
Pulse Data Words
Signal Data Control Data
Other
IO Bandwidth Breakdown
RF Distribution
Hardware
Tuners Digitizer Pulse Detection &
Measurement
Signal
Processing
RF IF RF
Ant
Displays
Combat
System
Digitized IF Pulse Data Words
Analog Digital
Pulse Detection &
Measurement Type m
EW System Architecture – Scaled
Copyright 2015 Lockheed Martin Corporation 5
RF Distribution
Hardware IF
RF
Ant
Tuners Digitizers Signal
Processing Type x
RF Distribution
Hardware IF
RF
Ant
Tuners Digitizers Displays
Combat
System
RF Distribution
Hardware IF
RF
Ant
Tuners Digitizers
Contempory EW Systems will have multiple subsystems with specialized features
Signal
Processing Type y
Signal
Processing Type z
Subsystem 1
Subsystem 2
Subsystem 3
Pulse Detection &
Measurement Type n
Pulse Detection &
Measurement Type o
$$$$$
RF Distribution
Hardware IF
RF
Ant
Tuners Digitizers
Alternative: Digitize Once, Publish CDIF Data to Pulse and Signal Processing
Pulse Detection &
Measurement
Type m
Pulse Detection &
Measurement
Type n
Pulse Detection &
Measurement
Type o
Pulse Detection &
Measurement
Type m
Pulse Detection &
Measurement
Type n
Pulse Detection &
Measurement
Type o
$
Displays
Combat
System
Common Single Channel Sizes: 80 MHz, 375 MHz, 500 MHz
Single Channel of 500 MHz RF Digitized Data: ~2.7 GB/sec
I&RD Project Objectives
Key Tenants Envisioned for Next Generation EW System
Architectures
• Digitize the Spectrum and Distribute CDIF via a Flexible Digital Backbone
• Deliver new capability via mission and target-specific software applications
hosted on a common hardware solution
• Replace custom, proprietary interfaces with open, standardized interfaces to
facilitate rapid and affordable capability insertion
Server-Based Electronic Warfare Engineering Development Model
• Objective: To evaluate the ability of current & projected COTS technologies
to realize Next Gen EW Architecture Vision
Copyright 2015 Lockheed Martin Corporation 6
Validate key COTS technologies: RF digitization, high-
speed networks and high-speed computing environment
Migrate pre-existing FPGA and PowerPC based receiver
processing to Linux Servers and Intel & NVIDIA
accelerators
Demonstrate scalability of architecture to greater
bandwidths (wideband feasibility)
Validate Open Interface Data Distribution Services (DDS)
publish / subscriber services
Verify key TPMs: CDIF throughput, Processing
Bandwidth, Measurement Accuracy, Dynamic Range
Objectives
Enabling Technologies PCIe Gen 3.0
High Speed Network
General Purpose GPUs
COTS A/Ds Digitizers
Intel Xeon Multicore CPUs
•Open standard with accelerated performance road map •More configurable, higher
bandwidth than VME •Today: 7.88 GB/sec (8xGen 3) •Next Gen: 15.7 GB/sec (8xGen 4)
•Widely used open standard provides most bandwidth solution •User level multicast •Near term (2015) projection of
>100 Gbps •Today: 7GBps per channel •Next Gen: 12.5 GBps per channel
•Excellent GFLOP/watt ratio •Cost benefit over FPGA •High-order SW language •Vast CUDA libraries •Today: Kepler GPUs with ~5
GFLOPS per watt •Next Gen: Maxwell GPUs with
~12 GFLOPS per watt
•Excellent GFLOP/watt ratio •Supports numerous High-order
programming languages •Most mainstream target processing
environment •Today: 2600 Xeons •Next Gen: Xeons with Phi
Accelerator
•Easily scaled across general purpose HW hosts •Flexibility in configurations •Publish source I&Q data •Today: 12 bit @ 500 MHz •Next Gen: 16 bit @ 500 MHz
Architecture
Linux
Server
ADC
PC
Ie ADC
ADC
ADC
I/O
I/O
1 GHz IF
1 GHz IF
56
Gb
In
fin
iba
nd
Sw
itch
CDIF
RF Digitization High Speed
Network
High Speed
Computing
Linux
Server
PDWs
Linux
Server
PC
Ie
I/O
GPU
GPU
I/O
Linux
Server
PC
Ie
I/O
GPU
GPU
I/O CDIF
CDIF
Data Processing
Emitter
Reports
Spectrum
Display
1 GHz IF
Server-Based Electronic Warfare Demonstrator Advanced Development Model Overview
Copyright 2015 Lockheed Martin Corporation 7
100% COTS Architecture Provides an Open, Scalable, Upgradable and Affordable Solution
EW Advanced Development Model – First Generation
Next Generation EW System – Tactically Relevant, Multi-
Channel, Open Architecture EW Solution
Digitization Layer– COTS A/D Solution
Digitizes and Multicasts all CDIF Data to a
Scalable Number of Processing Nodes
Processing Layer – COTS GPGPUs Perform
Bulk of Real-Time Signal Processing with Multi-
Core CPU’s Completing the Balance High Speed Network– 56 Gbps per
channel Infiniband Network
Copyright 2015 Lockheed Martin Corporation 8
A/D
Digitization Tuners
Channelized
Receivers G
PDW
Clustering G
Deinter-
leaving Parameter
Measurement
Emitter
Processing &
Classification
Tune
Scheduling &
Control
Real-Time Integration Displays G
Track File
A/D Data (3x)
Tune Cmds PDW
Clusters PDWs Pulse Trains
Emitter Reports
Emitter Detections
C&D
Digitization Layer Receiver Layer Signal Processor Layer
Display Layer
Track File
Mainline Online Mainline Online
Server-Based Electronic Warfare ADM Software Overview
FDR Infiniband
GbE DDS (Pub/Sub)
Key
G GPU-based Processing
Multiple channels of Continuous
Digital IF (CDIF) are multi-cast over
an open, high speed network for consumption by multiple users.
Receiver, Signal and Data Processing is
hosted on Linux Servers featuring
NVIDIA and Intel accelerators supporting higher order programming languages.
IF inputs digitized via COTS A/D
converters (12 bit, up to 1.8 Gsps)
hosted on PCIe slots on a Linux Server.
Standardized open interfaces between software
modules are implemented via Data Distribution
Services (DDS) to enable real-time, reliable and
scalable data exchanges between publishers and subscribers.
The core EW processing thread, tested and proven
on an adjacent defense program, has been further
migrated to an open, scalable, maintainable and affordable architecture.
Copyright 2015 Lockheed Martin Corporation 9
Open Architecture and Modular, Software-Based Solution Facilitates Rapid Capability
Insertion in Response to New and Evolving Threats
Channelized Receiver GPU Implementation & Timeline
Copyright 2015 Lockheed Martin Corporation 10
Integer to Floating Point
Conversion
Pre-Sum (Polyphase)
FIR Filter
Receiver Pre-processor
Global Memory 2.7 **
2.7
Fast Fourier
Transform (FFT)
FIR Filter
Magnitude/ Phase
Magnitude Block Integrator
Phase Block Integrator
Constant False Alarm Rate (CFAR)
Pulse Detection
Xeon CPU
PDWs
CDIF
Kepler GPU
Phase Block Integrator
7.3
0.8
15.4 PDWs
0.1
** All Numbers are GB/sec per channel
2013 2014 2015 Quadro 4000 Telsa K20x Telsa K40
Telsa K8 Telsa K80
Architecture
trade studies
Initial SW implementation
using CUDA Libraries
Transition to hand
coded of Kernels
Development Timeline
Achieved Single 500
MHz Channel per GPU
Began transition to 2
channels per GPU
Achieved Two 500 MHz
channels per K40 CUDA 6.5 (10%
performance gain)
7.3
7.3
7.3
7.3
7.3
Benchmarks – 3 Generations of GPGPUs
Copyright 2015 Lockheed Martin Corporation 11
0.0 5.0 10.0 15.0
NoMod,10,Edge
NoMod,100,Edge
NoMod,1k,Edge
NoMod,10k,Edge
Wide PSK
Many
CW,Edge
CW,LFM
Many+CW
Many+CW,LFM
NoMod,20k,Edge
FFT
Mag Block
Phase Block
CFAR
Detect
Measure
CW
Tesla K20x (3/14) Design Goal
0.000 1.000 2.000 3.000 4.000 5.000 6.000
adc_out_matlab_1k_center_snr_30
adc_out_matlab_1k_edge_snr_30
adc_out_matlab_10k_center_snr_30
adc_out_matlab_10k_edge_snr_30
adc_out_matlab_fmcw_500_center_s…
adc_out_matlab_100k_center_snr_30
adc_out_matlab_100k_edge_snr_30
adc_out_matlab_cw_edge_snr_3
adc_out_matlab_bpsk_25ns_edge_sn…
adc_out_matlab_1M_center_snr_30
adc_out_matlab_1M_edge_snr_30
adc_out_matlab_many_snr_15_with_…
adc_out_matlab_many_snr_15
adc_out_matlab_many_snr_15_with_…
FFT
Block
CFAR
Detect
Measure
PDW
Tesla K40 (1/15) Design Goal
0.000 1.000 2.000 3.000 4.000 5.000 6.000
adc_out_matlab_1k_center_snr_30
adc_out_matlab_1k_edge_snr_30
adc_out_matlab_10k_center_snr_30
adc_out_matlab_10k_edge_snr_30
adc_out_matlab_fmcw_500_center_…
adc_out_matlab_100k_center_snr_30
adc_out_matlab_100k_edge_snr_30
adc_out_matlab_cw_edge_snr_3
adc_out_matlab_bpsk_25ns_edge_s…
adc_out_matlab_1M_center_snr_30
adc_out_matlab_1M_edge_snr_30
adc_out_matlab_many_snr_15_with…
adc_out_matlab_many_snr_15
adc_out_matlab_many_snr_15_with…
FFT
Block
CFAR
Detect
Measure
PDW
Tesla K-80 Performance – on one (of two) GK210 GPUs Design Goal
Kernel K40 K80 (per GPU)
Samples 13330000 13330000
Threads 1666250 1666250
Ops 1.14638 1.14638 Gflops
Time 1.058 1.185 ms
Ops/s 1084 967 Gflops
700 GLOPS typically advertised for Single Precision
cuFFT processing (FFT length of 4096)
Tesla K-80 Single Precision Performance (Constant Data Stream)
cuFFT 6.0 on K40c, ECC ON, 32M elements, input and output data on device
Real-time Instrumentation
Copyright 2015 Lockheed Martin Corporation 12
• Software monitors all network related
resources and provides real-time
feedback on current throughput
• Screen capture confirms Multicast
Infiniband of high bandwidth Digitization
Data
• Software monitors utilization of GPU
resources
• Color-codes various stages of the
Channelized Receiver processing over
time
Signal Processing Displays
Copyright 2015 Lockheed Martin Corporation 13
Both Core Signal Processing Software and Real-time Signal Processing displays hosted
on GPU processors
Summary
Conclusions
• The Enabling Technologies are moving fast driven by wide commercial /
non-defense markets
• We continue to be encouraged by results to date and anticipate continued
performance gains due to ongoing COTS evolution
Copyright 2015 Lockheed Martin Corporation 14
Next Steps
• The transition of this server-based
architecture to a tactical system is
underway
• Extend architecture to additional
applications across the RF/EW
enterprise
– Through integration of mission and
target-specific SW applications
– Through continual integration of next
generation COTS HW technologies
As the Technology Matures, More Systems will Adopt, Accelerating Development
1.0 – Complex Pulse Measurement - Background
Pulse Filtering and Channelizer Compensation
Summary:
• To achieve wideband coverage, the RF spectrum is divided into
channels and further sub-divided into sub-channels
• The division/sub-division of the RF spectrum creates conditions
that have to be accounted for various signal types in order to
accurately measure/characterize the RF environment
Min Freq Channel n Channel n+1 Channel n+2 Channel n+3 Channel n+4 Channel n+5 Channel n+6 Channel n+7 Channel n+8
Max Freq
SubChannel 1 SubChannel n . . . RF Sub-Channels & Channelization: created by
(early stage) signal processing
RF Channels: created by tuners/receivers.
Nominal Stable RF signal : No issues
Signal at sub-channel edge
Signal at channel edge
LFM Pulses at sub-channels edge
LFM Pulses at channels edge
Strong signal with corresponding rabbit ears
Pulse andduplicate
Duplicate at loweramplitude
Falling edge rabbit ears
Rising edge rabbit ears
Copyright 2015 Lockheed Martin Corporation
FFT Performance
Copyright 2015 Lockheed Martin Corporation 17