Network attached data acquisition leads to configuration freedomStory
February 10, 2009
Multichannel, low-frequency signal processing architectures used in test and measurement, vibration analysis, and acoustic processing (sonar) are undergoing revolutionary changes and diverging from equivalent high-frequency military applications such as imaging, signals intelligence, digital radio, or radar.
Recent implementations - particularly of sonar systems in larger submarines with their vast arrays of hull-mounted or towed acoustic sensors - have introduced the concept of network attached data acquisition. Data is acquired and packetized at groups of sensors, then passed via Ethernet to signal processing servers forming the main sonar signal processing computer. This freedom from direct attachment of sensors to signal processing creates opportunities to configure systems that are easier to integrate, perform better, and offer both form factor and processor independence.
High- and low-frequency applications
Directly attaching the A to D will continue in high-frequency applications, often using PMC/XMC mezzanine modules to provide compatibility with popular VMEbus, VPX, or CompactPCI systems. These use PCI-X or PCI Express to transfer data samples at very high rates, directly into the signal processor's main memory. High-performance processors such as Freescale's dual-core 8641D PowerPC CPU are often used in multicomputing configurations, performing floating-point FFTs on blocks of data as they are assembled into memory. In addition, FPGAs can be used for preprocessing where many repetitive algorithms must be performed in parallel on the incoming data. Data flow is an important characteristic of these high-performance systems, and switched fabrics form high-bandwidth connections between processing nodes. This creates a complex heterogeneous architecture that is difficult to implement, manage, and integrate, requiring specialized knowledge plus sophisticated modeling and testing tools.
On the other hand, for low-frequency applications, this level of complexity has been rendered unnecessary by the widespread availability of GbE, with its ability to transfer data at rates in excess of 30 MBps. It offers the bandwidth to support many low-frequency acoustic channels through a single cable. This allows colocated data acquisition with the sensors via self-contained data concentrators accessed and controlled by standard network protocols such as SNMP. By effectively using point-to-point data transfers, bus contention and determinism issues are eliminated. Separating the acquisition in this way offers many advantages. Data acquisition modules can be configured in purpose-built packaging to suit the application and its environment. They can be mounted close to the sensors and, by being remote from signal processing, will not be subjected to the noisy electrical environment of clocks, bus switching, or power supply chopping.
This data acquisition separation gives many more choices in architecture, form factor, and signal processor type. If the packaging meets the environmental requirements, off-the-shelf PCs, embedded workstations, file servers, and the familiar 6U format, PowerPC-based multicompute engines can be used to meet the needs of the application. A leading example of a network attached data acquisition system is daqNet, offered by GE Fanuc Intelligent Platforms (Figure 1). Packaged in a 1U high, 19-inch rack-mounted enclosure, daqNet incorporates up to 192 channels plus an FPGA for network protocol processing. It is also capable of front-end preprocessing.
(Click graphic to zoom by 1.9x)
Network attached devices
Network attached functionality of all types is growing rapidly. For example, Network Attached Storage (NAS) is a rapid growth area for shared network applications. In nonmilitary markets, instrumentation systems are migrating from the well-established VMEbus eXtensions for Instrumentation (VXI) standard to the newer LAN-based eXtensions for Instrumentation (LXI), based on similar concepts of network attached data acquisition. 10 GbE is the next evolutionary step, giving more latitude in network topology and enabling more low-frequency channels or a smaller number of video channels to be carried, for example. A sonar signal processor is a very tightly controlled network environment. Other instances of attached data acquisition such as instrumentation and analysis of a gas turbine aero engine might need the flexibility of a number of shared local network applications today, but could migrate to being entirely Web based.
The flexibility and ease of integration offered by network attached devices is changing the way military embedded systems are being architected and implemented. Multichannel, low-frequency data acquisition is already part of the inventory and looks set for extension into many other application areas as 10 Gb (and faster) Ethernet rapidly becomes more widespread.
To learn more, e-mail Duncan Young at [email protected]