As the sampling rate of analog-to-digital converters (ADCs) has continued to increase, it has become practical to test, measure, and analyze RF and high-speed serial data from communications interfaces like Sonet, Giagbit and 10-Gigabit Ethernet, Fibre Channel, Serial ATA (SATA), PCI Express (PCIe), RapidIO, and a batch of others. But as these speeds have increased, transporting and storing the data to be displayed or analyzed has become a major problem. The key to testing and analyzing is streaming a large volume of samples so noise, glitches, and intermittent anomalies can be detected usually by software analysis. But with sampling rates above 5 Gsamples/s, the amount of data to be generated is enormous. Doing the math, you can see how at 5 Gsamples/s you need at least 5 Gbytes of storage just to see a couple of cycles or bits of the signal to be analyzed. Where do you store all that sample data, and how do you get more storage space for longer intervals of analysis? And how do you transport that data from ADC to memory? Traditional instrumentation deals with this problem to a degree. PXI-based virtual instrumentation systems with a RAID system, though, may have the best answer. Traditional Instrumentation Typical standalone instruments like digital storage oscilloscopes with built-in computers solve the problem by transmitting the samples over a fast PCI bus to the internal memory, which is fast RAM. The fastest scopes have deep memory up to about 200 Mbytes. Some instruments have provisions for memory expansion, but there is a limit. For many applications, it still isn’t enough. Besides high-speed serial data, applications include RF spectral monitoring, long-duration digital fault detection, the generation of non-periodic waveforms, and IF/baseband recording and playback. In such applications, the traditional instruments are limited to finite acquisitions and generations with lengths dictated by the amount of on-board memory. While hard-disk storage is sometimes available, most drives aren’t fast enough to handle the streamed data. Many traditional instruments need an external PC to do the analysis, so they rely on buses and interfaces such as the General-Purpose Interface Bus (GPIB) and Ethernet to transfer the sample date. However, these low-speed buses simply cannot achieve a transfer rate suitable for streaming data from instrument to PC. In both cases, traditional standalone instruments can perform the analysis on segments of the data but are limited in their ability to analyze streamed data. PXI Potential One solution is to use modular virtual instrumentation based on the PXI standard. These instruments exist as modular segments plugged into a chassis featuring a fast PCI parallel bus and/or an even faster PCIe serial bus (Fig. 1). With a built-in PC, the sampled data is easily streamed via the PCI or PCIe bus to the PC and its RAM. Large RAM expansions are possible internally. The PXI Express bus provides up to 6 Gbytes/s of total system bandwidth and up to 2 Gbytes/s of dedicated bandwidth per slot. In PXI systems controlled by an external PC, data can still be streamed over cabled PCI Express up to 7 m between the chassis and PC. The modular nature of the PXI system and its expandability make it a better choice for some test and measurement applications. But even PXI has its limitations. RAID to the Rescue Deep memory (usually RAM) is the primary need for test applications requiring streaming data. But typically, you can only go so deep, even in PXI systems. Hard-disk storage is the answer if you could find a drive that could keep up with the data. The bad news is that hard drives are mechanical, so they have their limits. Even the faster 3.5-in. drives spinning at 7200 rpm or faster in some models can’t keep up with gigabyte/s data. The good news is that there is a way to use hard drives to provide the ultra-deep memory for gigabyte/s streaming data. It is known as RAID.
RAID means redundant array of independent (or inexpensive) disks (or drives). It is a technique for clustering two or more hard drives to increase storage capacity, providing reliability through redundancy, and/or increasing throughput. Hardware or the operating system and related software can provide RAID functionality. A RAID array can exist in a PC with multiple drives. Or, the RAID can be an external hardware box with multiple drives (Fig. 2). Most RAIDs are used on servers, but they also can be found in systems for video editing and distribution. RAIDs increase throughput by streaming the data sequentially to several drives. This causes the data to be distributed throughout multiple drives. Throughput rates of 2 Gbytes/s or more can be achieved. With individual drives whose storage capacity is available to 1 Tbyte (or 1 trillion bytes), RAID systems with capacities of up to 40 Tbytes or more are possible. And such systems are affordable. In PXI systems with an external PC, the RAID may consist of multiple internal disks or an external RAID unit. An external serial ATA (eSATA) or cabled PCIe interface is common on the enclosure. The PC is then linked to the PXI chassis with Cabled PCI Express. In some cases with an embedded PXI controller, a PXI Express peripheral module can be installed to interface to an external RAID box with Cabled PCI Express. Or, the RAID may be connected to the embedded controller via eSATA. There are multiple versions of RAID devices defined by a general industry concurrence. They’re known as Levels, and there are up to a dozen or so predefined Levels. For example, Level 0 RAID uses multiple drives and provides data striping—that is, the distribution or spreading out of the streaming data blocks over multiple drives (Fig. 3). Striping improves speed of storage but not reliability. Apparently, Windows transparently supports this mode of striping. With this system, you should make the stripe size as large as possible. Such a technique is ideal for PXI instrumentation, as it provides for fast throughput, but it doesn’t factor in redundancy for reliability. Another popular RAID is Level 5. In addition to striping, Level 5 implements a parity error detection scheme using all of the disk drives. Standard RAID devices are widely available from multiple sources and make an ideal addition to any PXI instrumentation system where streaming data storage and analysis are essential. Give ’em What They Want Engineers want streaming, but they haven’t been able to do it because of technology limitations. Many applications can’t sustain a sufficient sampling rate for lengthy acquisitions or generations. In these situations, engineers must compromise with a slow enough sampling rate for data to be transferred over the bus or by sampling at the necessary high speeds for short periods of time that onboard instrument memory allows. Neither sacrifice is desirable. High-speed streaming permits data transfers to or from an instrument at a rate high enough to sustain continuous acquisition or generation. This is accomplished by having a bus with sufficient bandwidth for overall data throughput and a system that allows the entire acquisition or generation waveform to be stored. High-speed data streaming of IF and baseband signals is a common application, since data collection is often desired at high rates for an extended period of time. Onboard memory limits traditional instrumentation. Fixed onboard memory is in the hundreds of megabytes, but a RAID system can be in the terabytes. Two main technologies enable high-speed data streaming: high bus bandwidth for faster data transfer and RAID for faster data writing/reading and additional storage. Instead of data being generated or acquired at the full rate of the instrument for milliseconds, people can literally stream data for hours. This ability to sustain continuous acquisition or generation is enabling new applications like record and playback of RF or IF signals (analog) and LCD testing (digital).