…cutting deployment risk, time & cost 1
Introduction Today’s EW systems must meet threat environment that are diverse, deceptive & agile. To confront these challenges the systems must offer: • Ultimate performance for detection and classification • ability for rapid reconfiguration for handling ever changing threat scenarios • ability to handle different sensor-types and multi-mission requirements • ability to scale up to handle increasing processing load as sensor-count and signal bandwidth increases Most importantly, today’s EW system must reduce deployment risk, time and cost 2
Conventional Approach Falls Short The conventional approach of developing EW systems using bus (VME, PCI, etc.) based COTS boards from multiple vendors fails to meet the challenges: • It takes an enormous amount of time and money to select COTS boards from multiple vendors, choose a host platform, Operating system, device drivers and then integrate a system, develop application software and perform system testing – the process generally takes years and often results in time and cost overruns. • A critical issue is that it is often difficult to synchronize multiple bus based systems (in time and phase), particularly, as the sensor-count and signal bandwidth increases Architectural simplicity is key for meeting the demanding requirements 3
Simplicity is the Key An approach the circumvents the problems of the conventional approach is presented here Sensor Interface Processor Display (Modular & Scalable) (Scalable) Data Link(s) (High-Speed) Most EW systems, whether, Radar, COMINT, ELINT, Sonar can be partitioned into two parts: a sensor interface part & a processing part. Multi-core server based software (real-time) processing is ideally suited for most EW applications – they are inexpensive, readily available, upgradable and offers reconfigurable processing. Sensor interface subsystem is bus-less, scalable and incorporates one or more data links for transfer of pre-processed data (I & Q data for radio/radar applications) to and from the processor. Increasing channel count and/or signal BW may require additional data links. The questions are: what is an ideal data link and what should be its properties to simplify system development? 4
10 Gigabit Optical Network (Fiber) is an ideal data link Advantages • It is an open and evolving industry standard • The data transfer rate per fiber can be as high as 1 Gbyte/s • The use of optical fiber(s) allows data to be transferred over long distances, permitting antenna-level digitization • Better analog performance is achieved as there is no “noisy” computer bus in the analog section • It is linearly scalable which means that the throughput rate can be increased by adding more fibers • multiple fibers can be synchronized (important for MIMO, DF, Beamforming, etc.) • Multiple sensor data can be easily fused into a common processor • It is OS agnostic • It efficiently leverages the multi-core computer server technology The concept of 10 Gigabit Sensor Processing (10 GSP) is based on 10 Gigabit network-attached sensor interface unit & multi-core server based processing. The sensor interface unit may vary from one requirement to another but the concept is the same. 5
Common Configuration For Sensor Processing 10 Gigabit Network (Data) 10 GbE Network Attached Sensor Interface From / To Sensors Pre- installed 10 GbE NIC BRIDGE CORE CORE CORE CORE Off-The-Shelf Multi-core CORE CORE CORE CORE RAID Server RAID CONTROLLER 8 TB or 32 TB 1 GbE Storage (Control) User PC Multi-core Software Processing For Rapid Deployment & (GUI/Control API) Re-configuration Any system has two parts: Sensor Interface Part & Processing Part (Sever). One can move from one application to another by changing the software 6
Drinking From A Fire Hose Availability of multiple processors, each running at multi-GHz rate with full floating point precisions offers major advantages over FPGA based processing in terms of tremendous cost savings and rapid re-configurability. The challenge is how to program efficiently so each processor is optimally engaged to achieve real-time throughput rates. Input • Receives 10GbE data continuously in a ring Input buffer in the Input Server Module 10GbE Server • Receiving of data can be abstracted from the user Module • User creates their own processing thread based from a template class Multiple Pipes allow for load Output User Processing balancing • Users can add pre-processing Modules to Module process data before passing it to the Output Sensor Client Interface • The processing class is created based on a Device template class • Output Client transmits the DAC data over User Processing the 10GbE. Sending of the data is abstracted Module from the user PIPE 10GbE • Connection between modules and allows Output inter-stage data transfer synchronization Client • Handles all data transfer between modules Module (abstraction for the user) • Must be used to connect User designed A Framework For Multi-Core Software Processing processing modules to other modules 7
10 Gigabit Sensor Processing (10 GSP) Sensor Interface Processor (Modular with Minimal Display (Server) Pre-Processing One or More capability) 10 Gigabit Network(s) • • Scalable & Modular network Real-time multi-core software attached system processing • • Synchronized operation Processing, recording & playback can • Bus less for better SNR be combined • • Minimal pre-processing capability Port replication for future upgrade • for efficient software based Rapid upgrade or re-configuration • system implementation Low cost 10 GSP provides fast, scalable & synchronized systems for rapid deployment & fast upgrade Appendix A provides sample examples of actual deployed systems that highlights the advantages of the 10 GSP approach 8
Network Latency The 10 GSP concept discussed here is based UDP/IP protocol for high data pay load for high-speed operation. This protocol has lower latency than higher layer TCP/IP protocol. The latency is primarily dependent on the data packet size. For high speed, low interrupt rate operation a jumbo data frame (up to 56 Kbytes) is used. However, the user has total control on the data packet size. COTS bus based boards also has similar latency issues. For extremely low latency applications, FPGA based processing may be unavoidable. 9
Conclusions This paper has demonstrated the tremendous advantages of adopting the 10 Gigabit network as data backbone for developing sensor processing systems for EW applications. Properly designed network-attached sensor interface subsystem coupled with commercially available multi-core server can implement virtually any demanding EW system. This approach, termed 10 GSP (10 Gigabit Sensor Processing) offers a multi- core software based solution that offers rapid reconfiguration. In many cases to move from one application to another only the software needs to be changed. With sample examples of actual deployments (as illustrated in Appendix A), the paper has further demonstrated: 1. Multi-core software approach for real-time signal processing is a superior approach in terms of cost, performance & development time to FPGA based alternative 2. It is possible to achieve virtually Limitless scalability by building up from a modular base configuration for meeting the demand of higher channel count and / or signal BW 3. Multiple fibers can be operated synchronously for critical phase-coherent applications 4. It is easy to incorporate different sensor types by simply adding appropriate network attached sensor interface system to the same processing configuration In conclusion, the 10 GSP approach can drastically cut deployment time, cost & risk 10
Appendix A Sample Examples demonstrate 10 GSP Advantages 11
Demons nstr trati ation on of 1 10 GSP Advantag tages – Synchro chroni nize zed Opera rati tion on & Limitless Synchronization & Scalability Record / Playback @ 2 GBytes/s Over 2 Fibers Compact 1U Record / System Playback @ 8 GBytes/s Over 8 Fibers Record / Playback @ 4 GBytes/s Over 4 Fibers 2 x DTA-5000-SSD (46 TB) Stackable 2U System Synchronized Record/Playback Over Multiple Fibers @ Record Breaking Speed Typical throughput rate is 1 Gbyte/s per fiber. Multiple RAID servers can be stacked to achieve virtually any throughput rate. 12
Demonstration of 10 GSP Advantage - Antenna Level Digitization HF Conditioning DTA-3200H DTA-3200H DTA-3200H DTA-3200H DTA-2300S DTA-2300S DTA-2300S DTA-2300S HF To Baseband (I & Q) Long Range 10GbE fiber Processing / Recording 128-Channel Phased Array HF Radar • 90 dB SFDR • Optical 10GbE network allow long distance transmission with LR optics • Lengths up to 10km possible • A cluster of 16 or more channels can be placed in a single location -- improving sensitivity • Multiple clusters may be spaced out to match with the actual antenna placement • Digitized data can be transmitted to common recording / processing station • Reduces deployment cost and improves RF sensitivity 13
Recommend
More recommend
