While today’s most sophisticated instruments can communicate through ISA, VME or VXI buses, the peripheral component interconnect (PCI) most often is the instrument bus of choice. The reasons for the growing popularity of PCI-based instruments are very simple:
The instrument bus is much faster than GPIB and ISA so experiments can run faster.
Multiple instruments can be housed in the same chassis, saving bench or rack space.
Multiple instruments can share resources such as the power supply, display screen, microprocessor and memory, resulting in lower costs.
The same instrument mainframe can be used for different experiments by changing the instruments-on-a-card or the software.
PC-Based Instruments
Before analyzing the characteristics and merits of this latest instrument bus, let’s take a brief look at the evolution of PC-based instruments. In the late 1980s, the rapid growth of the IBM personal computer (PC) market created a very powerful, yet low-cost, platform on which to build instruments. This produced a surge of new PC-based instruments such as oscilloscopes, waveform generators and voltmeters.
By 1995, these plug-in cards had grown into very powerful and sophisticated instruments capable of replacing stand-alone instruments in many applications. PC-based instruments not only were more cost-effective, they also featured very respectable specifications. Manufacturers such as Gage Applied Sciences use the latest A/D technology along with proprietary amplifiers to provide 12-bit oscilloscopes with greater than 65-dB signal-to-noise ratio and 70-dB spurious-free dynamic range.
The processing power of today’s Pentium and PentiumPro-based PC can be used to analyze data gathered by PC-based instruments to build sophisticated test systems, such as the ones required by the space program, the automotive industry, disk-drive manufacturing and medical imaging.
And PC-based instruments are easy to program. No more worries about GPIB commands and handshakes, no SCPI commands, no VXI messages—just simple C or BASIC programming in DOS, Windows 3.1, Windows 95 or Windows NT. Easy-to-use drivers for MatLab, LabVIEW, LabWindows CVI, HP VEE, DasyLab or Visual Designer also are available.
Until 1995, PC-based instruments operated on the ISA bus. ISA is a 16-bit asynchronous bus running at 8 MHz. While low cost and easy to design for, ISA is relatively slow, featuring approximately 2-MB/s data throughput.
PCI Bus
The only missing ingredient in making PC-based instruments a major force in instrumentation was a well-defined, high-speed bus for the IBM PCs. In 1995, the PCI bus started becoming the standard in all new Pentium machines.
PCI is a processor-independent, 32-bit synchronous bus running at 33 MHz. It features a multimastering capability, built-in plug and play, well-defined timing and a well thought-out architecture (Figure 1).
While PCI did not start out as an instrument bus, it has almost all the ingredients of one:
High sustained throughput up to 120 MB/s.
A choice of operating systems such as DOS, Win 3.1, Win 95, Win NT, QNX and Solaris.
Low-cost software tools.
Multimaster capability.
Bus mastering for data streaming from one instrument to another.
Simple programming.
Three or four slots standard, up to nine available using active backplanes.
Large supplier base, with approximately 100 new suppliers coming on-line in 1997.
Low cost without compromise on quality.
Since its inception, much has been said about the PCI bus. Many manufacturers use buzz words like plug and play, bus mastering, target and initiator devices and 132-bM/s throughput. So let’s analyze their meanings.
Plug and Play
PCI brings true plug-and-play capability to the PC platform. No longer do you have to play with DIP switches to set addresses on the add-on cards—the system configures itself.
At boot-up time, the PCI BIOS queries all add-on cards for the resources they need, such as I/O space, memory space or interrupt lines. All add-on cards must answer this query if they are to be deemed PCI compatible.
Once the BIOS knows the complete list of resources needed by the add-on cards, it configures the system by assigning mutually exclusive resources to different add-on cards. All add-on cards must accept the base addresses provided by the BIOS for communicating with their resources. In other words, PCI cards must not have DIP switches or jumpers that set specific I/O or memory addresses or interrupt lines.
Plug and play is an inherent part of the PCI specifications. Any add-on cards that do not adhere to its protocols are not PCI compliant and, if used, may cause system crashes.
Bus Mastering
Direct memory access (DMA) is a technique by which two system resources on a bus can transfer data between themselves without any CPU involvement. In the PCI bus specification, these two devices are the initiator and the target; that is, the initiator device takes control of the data, address and control lines of the bus and writes the data to the target device.
PCI implements DMA by allowing any device to become a bus master and write data to a target device, hence the term bus mastering. In plain English, bus mastering allows fast, 120-MB/s sustained data transfers across the PCI bus. A bus master is a hardware device that takes control of the data, address and control lines of the bus and transfers data to a target device.
A slave is a hardware device in a bus-based system that accepts data transfers coming from a master. A slave is almost always a target device.
A good analogy of slave and bus mastering is the telephone system in most businesses. Offices usually have a receptionist who answers incoming calls and directs them to employees.
The receptionist is the central processing unit (CPU) of the system, all the employees are target devices and the caller is the initiator. The slave mode does not require very sophisticated hardware, but does need considerable processing power (the receptionist’s time) and slows down the routing of the transactions (calls).
Some companies have direct inbound dialing (DID), a system in which each employee has a unique telephone number. By dialing an employee’s number, a caller can reach an employee without going through the receptionist.
DID is analogous to bus mastering. The initiator (caller) controls the bus (the telephone line) and connects with the target (employee) without disturbing the CPU (receptionist). Bus mastering is more efficient not only because the CPU (receptionist) can do other tasks while all the bus transactions (calls) take place, but also because it is faster.
Multimaster Capability
The PCI bus allows any of the system resources to become a bus master. For example, it is possible for an add-on card to be a master at one point in time and another add-on card to be the master at a later time. This provides the maximum flexibility in an instrument system so each instrument can use the maximum amount of system resources. Naturally, the CPU also is the bus master.
The obvious concern that arises in the case of a multimaster bus is how the bus is arbitrated. In other words, how does the system know which is the current bus master and how does a device become a bus master?
The arbiter always is with the motherboard or CPU card. Unfortunately, there is no clear specification for bus arbitration in PCI, so the exact schemes depend upon the motherboard or CPU being used. Nonetheless, bus arbitration is almost always handled in a safe and predictable manner, making PCI a very stable and well-defined system.
PCI Bus Throughput
All of us have seen claims made by PCI card manufacturers which tout data throughput of 132 MB/s. This figure is misleading. A total of 132 MB/s is the burst-mode data rate that lasts for no more than a few double words. Sustained data rates cannot be as high as 132 MB/s because no PCI device is allowed to burst for 100% of the time.
The maximum data throughput number is closer to 120 MB/s. This rate depends on both the motherboard and the PCI card. The motherboard must have the Intel Triton FX, HX or VX chipset for high-speed transfer. The PCI card must be capable of bus mastering and have enough data bandwidth on the add-on side of the PCI controller chip.
Not Everybody Will Need PCI
PCI is not essential for every PC-based test application. For example, if thermocouple measurements are made at a slow rate, there is no reason why a PCI-based instrument card must be used.
In fact, that is the beauty of a PC-based test system. It offers the high-speed PCI and the tried-and-trusted ISA bus on the same backplane and under the control of the same application software. An ISA-based thermocouple card measuring at the subhertz frequency can coexist with a PCI-based 500-MHz scope card.
Power of PCI Bus Instrumentation
Of course, there is no better measure of any instrument platform other than what it is capable of doing. A PCI-based instrument system can:
Digitize a signal at 50 MS/s with 12-bit resolution using a scope card.
Stream it directly to a quad C80 DSP processor.
Analyze the data using parallel processing on the DSP card.
Stream the result into an arbitrary waveform generator card.
Output the resulting analog signal.
Another example of a PCI-based instrument is:
Digitize a disk drive signal at 500 MS/s (2-ns/point) using a scope card.
Find the pulse width, the amplitude and overshoot on the signal.
Measure a DC voltage using a DMM card.
Record the values.
Output the same signal at half the speed using an arbitrary waveform generator at 250 MS/s.
PCI-based instruments can digitize analog inputs and stream the resulting data to a SCSI hard drive. One such product sustains an A/D and storage rate of more than 5 MS/s without any break in the data. Such performance is unique to PCI bus instrumentation.
Many other instrument systems can be configured with products from different vendors and run software under any operating system which supports PCI—which means virtually any operating system.
Conclusion
The high bus throughput and relatively low cost of PCI bus instrumentation make it a serious contender for test and measurement applications requiring multiple high-speed instruments to operate in conjunction with each other and a powerful microprocessor.
About the Author
Muneeb Khalid is a founder of Gage Applied Sciences and became the company’s president in 1990. He holds a bachelor’s of engineering degree from McGill University. Gage Applied Sciences, 5610 Bois Franc, Montreal, PQ Canada H4S 1A9
(514) 337-6893.
Copyright 1997 Nelson Publishing Inc.
May 1997
|