PXI Express brings streaming to test

PXI Express can be a foundation for streaming-data instrument design using a chassis such as the PXIe-1065. Courtesy of National Instruments.

When instruments first began communicating along a bus, they could handle only the most basic data traffic. GPIB, for instance, topped out at about a 1-Mbps data rate. Further, experience showed that system overhead and bus sharing limited the rate for sustained data transmission to about 30% of peak bus bandwidth. As a result, instruments that were gathering data needed to either process it themselves or save it to a buffer for later transmission to another instrument for post-processing. Neither option was ideal.

For one thing, having the data-acquisition board process the data itself reduces the benefits of synthetic instrumentation, according to Robert Lowdermilk, Scientific and Engineering Fellow at BAE Systems. “The idea is to separate processing from data acquisition to allow use of the best available technology for each.” Lowdermilk noted that processors double in performance every 18 months while data acquisition doubles every three years, so a board that incorporates both functions faces continual redesign.

But the alternative—buffering data for post-processing—also has limitations, the most obvious being that you can’t perform real-time data analysis or interpretation. In addition, buffer sizes can be limited because of the cost of high-speed memory, and this narrows the time window that a test engineer can peek into. To configure an instrument to capture the right time window, you need to develop elaborate triggering mechanisms.

With the advent of the PXI bus, however, a third alternative for processing data became available. PXI data bandwidth can be as great as 132 Mbytes/s—enough capacity in some applications to allow the streaming of information from a data-acquisition module to a processing module. The ability to stream data minimizes the need for buffering and allows for real-time data processing.

Still, the instrument bus remained a limiting system element. Data-acquisition sample rates and bit widths that continue to climb and a trend toward multichannel data acquisition have kept the bus a bottleneck for high-end applications, requiring that instruments continue to buffer their data. Further, the rising acquisition rates mean either that users must increase expenditures on memory to increase the size of the buffer or that the capture window must shrink to prevent the buffer from filling so quickly.

The introduction of PXI Express (PXIe) and chassis that support it has turned the situation around. Each lane of a PXIe link runs at 2.5 GHz and can handle more than 300 Mbytes/s. Designers can bundle lanes together to achieve data rates to many gigabytes per second. Further, the links are point-to-point, so unlike with conventional PXI, the bus bandwidth of a PXIe link does not need to be shared among all the system’s data transactions.

Future systems promise to be even faster. The second generation of PXIe links, likely to be available in the next year or two, is expected to handle 5 Gbps per lane, and the third generation should handle 10 Gbps. But even with the current PXIe, test engineers have access to a data bus that can stream many channels of data at the fastest sample rates and bit widths available. The result, according to Lowdermilk, is that the CPU in a PXIe system is now the bottleneck.

Designers of test systems have two options for handling the streaming data that PXIe supports. One is to record the data to a low-cost, mass-storage system. The other is to increase the processing power available for real-time handling of the data stream.

Recording data in a mass-storage system, such as a redundant array of independent disk drives (RAID), differs from buffering the data in onboard memory. The cost of mass storage is much lower than SRAM, and capacities as high as a terabyte are attainable. Thus, the time window that can be affordably captured and examined expands from fractions of a second to potentially hours.

Having access to terabytes of data does create a challenge: How do you manage that much data? Faced with hours of recorded information, engineers will need tools for previewing the information and quickly accessing the details of interest. Still, the benefits are compelling. A large time window provides independence from the need to set elaborate triggering conditions; users simply set the recording in motion and wait for the fault or other critical event to occur. An additional advantage is that the recorded data includes the full lead-in to the event, making it easier to find the causes.

Another benefit is that captured data can be streamed back through the system in real time. This allows engineers to develop their tests with captured field data rather than synthesized signals, resulting in a more thorough evaluation. The ability to stream recorded data is also useful in the design of manufacturing-test systems, where it is used to reduce the per-pin memory needed in testers.

Fig. 1 Streaming data on PXIe can outstrip the ability to record data, although partitioning the data across a RAID array can increase storage speeds. Courtesy of National Instruments.

Given that PXIe streaming and data acquisition can now quickly overwhelm conventional processors, test engineers who need real-time data analysis will need to enhance their processing power. Eventually, processor performance will catch up to data rates, but in the meantime, some form of parallel processing will be needed.

To that end, BAE Systems is working with Phase Matrix and National Instruments to develop a 26.5-GHz synthetic instrument based on PXIe. The proposed synthetic instrument uses a 26.5-GHz downconverter from Phase Matrix and IF digitizers from NI to capture the data. BAE Systems is handling the data processing using FPGAs with PCI Express (PCIe) interfaces built in. The parallel structures of the FPGA allow design of a data-processing engine that can handle multiple samples simultaneously, thus keeping the required device clock speed manageable.

Whether processed in real-time or sent to mass storage, streamed data in a PXIe instrument represents a powerful capability in test instrument design. It opens applications once requiring proprietary instrument designs, such as military signal intelligence, to commercial off-the-shelf (COTS) equipment. It can free engineers from hours of effort developing triggering schemes by trial-and-error and reduce the memory cost of instrument design. For engineers with applications that can take advantage of it, streaming data is a valuable new technique.