RTC Magazine

Designed to address the performance and environmental limits now confronting VME64x, VITA 46 provides an open standard for hardware form-factor and electrical connectivity uniquely suited for high-bandwidth distributed processor systems. It provides several significant advances over traditional VME including its high-speed connector interface and, through the complementary VITA 48 standard, support for 2-Level Maintenance for in-the-field repair and replacement. Three particular application case examples highlight the most likely target applications for early use of VITA 46: these include radar systems, mission computer systems and small form-factor graphics display systems. Each of these applications highlights specific technological advantages of VITA 46 over the existing VME64x or CompactPCI systems that it will replace or enhance in heterogeneous systems. Table 1 shows a comparison between VME64x and VITA 46.

Radar Processor Systems

Radar systems comprise a classic “hard” problem for embedded multi-computing: they typically involve multiple channels of high-speed streaming input data. Significant data errors can occur if the streaming data is interrupted to allow processing to catch up. In radar systems, performance must be as close to real time as possible. As latency in the system increases, so does the risk of data loss. Because human operators use the processed data from these systems for targeting or navigation, the resulting data can be life-critical, making it all the more important to minimize latency as much as possible.

Today’s VME64x-based radar processing systems typically receive their preprocessed sensor input data through high-speed A/D boards or front panel I/O connectors such as FPDP. The initial stage of signal processing was formerly handled by ASICs, but it is now more commonly handled by an FPGA-based board on which a customer adds their own proprietary algorithms to the FPGA bit stream. The data is then pushed to a microprocessor-based multi-computer board such as one populated by multiple PowerPCs.

Another factor that makes radar processing in parallel systems especially complex is the need for “all-to-all communication” at certain stages of processing. During processing, radar data must undergo a “corner turn,” or a distributed matrix transpose. In a distributed corner turn every processor in the system has a piece of data that every other processor needs for the next stage of processing. In today’s VMEbus-based radar systems every processor in the matrix must wait its turn to get a chance to move its data across the bus. Not only does this stall the processing of the current block of data, but it also stalls the incoming data.

Switched fabrics directly address this problem by parallelizing the all-to-all data movement, thus minimizing both the processing stalls and the interruption to the input data streams. Switched fabrics were first introduced into the VME world in the mid-1990s and have grown significantly larger and faster in the intervening 10 years. StarFabric is one serial switched fabric currently used to connect embedded multicomputing products using existing VME64x backplanes. However, physical limits in the legacy VME64x connectors prevent further growth.

Prior to the development of VITA 46, radar systems had begun to confront fundamental performance limits. The two most serious limits being the amount of data that the VME signaling pins could sustain, and the amount of power that could be dissipated per board slot. These are both problems VITA 46 was designed to address. While the number of operations performed on incoming radar data has grown to a certain extent, the amount of incoming raw data has grown tremendously.

Radar data pipes have grown much bigger because the higher the frequency at which data is collected the better the resulting imagery. In addition, the increased availability of large quantities of data is driving even more demanding radar system applications such as moving target identification (MTI), Multi-Hypothesis Tracking (MHT) and Identify Friend-or-Foe (IFF). What’s more, there is greater use and deployment of radar sensors. The increased use of UAVs, for example, has increased the amount of data to be processed, while simultaneously, because of stringent constraints on payload size and weight, driving demand for smaller and smaller packages.

As the amount of data has increased, so has the number of processors needed to process the data. This increases the amount of power consumed and the amount of heat generated by a radar system. Unfortunately, the demand for processing power threatens to exceed the ability to cool the components in these systems. This has led some leading vendors to attempt to address the urgent need for higher processor densities with proprietary standards.