Although it may be nearly carefree, PCI really stands for peripheral component interface. Often, a PCI plug-in board is the best solution for data acquisition needs. Modern PCI boards come complete with self-installing software and truly deliver on the plug-and-play promise. In addition to ease of use, a very wide range of performance is available, both in terms of precision and functionality. One of the major benefits of PCI boards is the high speed with which they transfer data to the host PC.
Boards with conceptually the most straightforward operation digitize a number of input signals and output the results to the PC’s memory. Of course, the actual circuit complexity may be high if the board also provides signal conditioning, anti-alias filtering, and very fast conversion rates requiring large amounts of onboard memory.
The most general board architecture accepts both analog and digital inputs and generates analog and digital outputs. These so-called multifunction boards are ideal solutions to smaller-scale data acquisition and control problems. Some have onboard signal- processing hardware to ensure deterministic operation in control loops, even when running under Windows. Digital outputs include counter/timer operations as well as single bit lines, and additional functions such as a phase-locked loop are available.
A wide range of board types and their detailed characteristics are listed in the comparison chart that accompanies this article.
Analog Inputs
Signal Conditioning
Most single-ended (SE) signals share a common ground return path, but there are such things as SE floating signals referenced to a voltage other than ground. Balanced, or differential (diff), signals are related to ground only to the degree that their comcommon- mode voltage remains within some limited number of volts away from ground. The signal of interest exists differentially between the two equal-impedance source terminals.
You need to match the input characteristics of your data acquisition board to the signals being acquired. Many boards, especially those that multiplex several signals into one analog- to-digital converter (ADC) input, support either diff or SE channel configurations. Sometimes, this choice can be made on a signal-by-signal basis while other boards only support all SE channels or all diff channels. In addition, on some boards it is possible to connect the low side of a signal pair to a reference other than ground.Data acquisition boards may provide two or more voltage input ranges, either unipolar or bipolar. This means that the board contains switchable attenuation and perhaps gain that you must set up in your test software. High-speed boards usually have 50-Ω inputs and associated BNC or SMA connectors. Often, the input impedance can be changed via software from 50 Ω to 1 MΩ or higher, although on at least one manufacturer’s board, this selection requires changing the position of a soldered jumper.
Further signal conditioning in the form of anti-alias filtering is built into higher performance boards. For example, an analog filter may be combined with digital filtering computed by onboard DSP chips. Whether the board constrains the signal bandwidth or you do it separately ahead of the analog input, filtering is absolutely necessary to ensure alias-free analog-to-digital conversion.
And, don’t forget calibration. Digital autocalibration or self-calibration is the state of the art today. You no longer need to account for calibration constants in post-acquisition analysis software routines, nor are there any potentiometers to adjust on most boards.
While resolution of the ADC clearly determines the overall resolution of the board, the affect of the ADC on the board’s overall accuracy is not as obvious. Manufacturers that quote accuracy in terms of least significant bits (LSBs), for example, may be confirming that the ADC’s differential linearity is sufficiently good to ensure no missed codes, but they probably are not specifying overall board performance. Unfortunately, some data sheets don’t explicitly state percentage of reading when specifying accuracy. A combination of percentage of reading plus a very small percentage of range is a common approach. Beware of board specs quoted as percentage of range or percentage of full scale. For small signals, percentage of range is equivalent to a huge error.
Bandwidth
Virtually all board data sheets state the sampling rate although its range and resolution are not necessarily made clear. What may be surprising is that many manufacturers do not list a board’s bandwidth or interestingly state it as half of the sampling rate. Nyquist’s ×2 relationship applies to the bandwidth of the signal that can be reproduced from the digitized values and to the minimum sampling rate-to-input signal bandwidth ratio required to avoid aliasing.
Generally, the bandwidth of an input channel is not related to the sampling rate. An important exception occurs when a board uses sigmadelta ADCs. These devices perform digital filtering that determines bandwidth and is directly related to the sampling rate.
For most signals associated with sensors or transducers, the object is to oversample sufficiently fast to preserve the signal nuances, the small higher frequency perturbations that may help to distinguish one effect from another. Oversampling in this context refers to sampling several times faster than the minimum ×2 criterion.
However, for RF communications applications, it may be desirable to sample at a rate far less than the carrier frequency to capture the modulation waveform. In this case, the data acquisition channel bandwidth must accommodate the carrier frequency.
The sampling rate will exceed the modulation signal bandwidth by at least a factor of two but not the carrier frequency. For either oversampling or undersampling, you need to know the board’s bandwidth. If it’s not listed on the data sheet, ask the manufacturer to fully specify it. Make certain that it is adequate for your application. For example, bandwidth may vary greatly as a function of input amplifier gain.
Sampling
The actual signal sampling is accomplished in several ways. Generally, a very stable and accurate internal clock source drives a programmable clock divider, which supplies the sampling clock to the ADC. Many manufacturers refer to this signal as the pacing clock. Often, a board will accept an external clock input in place of the internal clock. This facility allows synchronization of ADC clocking to some external signal.
It can be important in some applications to understand how the external clock is treated internally, a detail not often explained on the data sheet. If the external input simply replaces the internal clock in all respects, synchronization of the ADC sampling rate is easily accomplished. Sometimes, however, an external input is retimed by the internal clock. In these cases, a large internal-to-external rate ratio may be required to minimize the clock jitter caused by retiming.
This type of clocking only provides approximate synchronization.
To help explain internal vs. external clocking as well as many other similar details, a block diagram of the board can be invaluable.
Unfortunately, many manufacturers do not provide one as part of the basic data sheet. If separate ADCs per channel are used in parallel, as is done for very high-speed boards, the pacing clock is the sampling clock.
This is not the case for boards that multiplex their input channels. Most of these boards do not attempt to time-align sampling across all the channels, for example by using multiple sampleand- hold circuits. Instead, the channels are sequentially sampled as a group by a fast pacing clock. This technique minimizes the circuit cost and the timing offset among the channels.
A separate memory maintains a so-called channel-gain list that controls the order in which the channels are scanned as well as the gain of each. Scanning in this way introduces additional terminology such as cycle time, that is, the time between completions of successive scans of the channelgain list.
It is common with this type of board to have a cycle time of perhaps 100 ms (each channel is sampled at a 10-Hz rate) and a 250-kHz or faster pacing clock rate. In fact, you may only have control of the cycle time while the pacing clock always runs at the full speed.
Typically, the specified sampling rate of a board with multiplexed inputs is the aggregate rate. For example, unless clearly stated otherwise, if the board has 16 channels, each one will be sampled only at the quoted rate divided by 16.
Memory
Many boards require a small amount of fast onboard memory to buffer the PCI data transfer process against the data input rate from the ADC. A temporary storage FIFO usually fills this need.
In contrast, if the sampling speed is faster than the PCI bus transfer rate, a large amount of onboard memory is needed to store complete acquisitions.
These forms of onboard memory are not related to the memory used to support analog outputs. Some data sheets list several headlines together, each of which may be an impressive specification. The problem is that the highlighted 500- kword memory, for example, may not be associated with the input channels but instead only supports the analog outputs.
Digital I/O
Digitizing a number of analog inputs is what most people think of when considering data acquisition boards. However, many boards acquire both analog and digital data, and correlating the two types is important.
For example, an analog signal may be affected by switching, such as when using a multiplexer.
Alternatively, you may be interested in determining how much effect a number of switching signals are having on one or more analog signals. In both cases, you need to know what the analog signals were doing when the digital signals changed. You need a board that minimizes the timing skew across both analog and digital channels by including the digital I/O channels and counter/timers in the channel-gain list.
Typically, the total number of I/O lines can be divided into several 8-b ports, each with a programmable I or O direction although some boards allow you to program the direction of individual bits. Part of the distinction between methods relates to the supporting hardware. Older or lower cost boards may use chips that restrict port direction while newer designs based on a custom ASIC are more flexible. All digital ports on a board do not necessarily operate at the same speed.
Drive capability is an important digital output specification. It determines the amount of additional circuitry, if any, that you need to provide to interface to the DUT. For example, if 10- to 20-mA drive current is available, you can operate solid-state relays to control highpower AC loads.
The speed at which digital inputs are sampled may be the same as the analog channel-gain cycle rate. However, some boards support faster sampling if only digital inputs are listed in the channel-gain memory.Very low-cost or older designs use software polling to sample the digital inputs. Clearly, this technique will not result in simultaneously sampled analog and digital channel data even if a fast PC is used. On the other hand, if a board’s analog performance addresses your application and you need only minimal additional digital I/O capability, then software polling may be a good low-cost approach.
Counters/Timers
If you want to generate a timing waveform, determine the frequency of a pulse signal, or simply count the number of times an event occurred, buy a board with counter/ timer functionality. In fact, you may require several of these circuit functions, but be careful how you count them.
Some manufacturers claim to have three counters/timers on a board, and they do, but one is needed to develop the internal sampling clock, and another is dedicated to scaling the external clock input. So, there is only one general-purpose functional block you can use in your application.
Or, a board may have several counters/timers of limited range although they can be cascaded. In this case, the real number of functions you can use depends on the size of the numbers you need to work with.
The onboard timing-clock speed determines the resolution available, for example, in a pulse-generation application. Few boards offer 100-MHz counter/ timer clocking, but even this rate supports only 10-ns resolution.
Analog and Digital Outputs
In many applications, a board’s analog outputs can replace arbitrary signal generators (Arbs) as well as classic sine-triangle-square function generators. The devil is in the detail, as they say. Output waveform fidelity and noise are affected by many factors including filtering; output digital-to-analog converter (DAC) resolution, accuracy, and glitch energy; grounding; shielding; and power supply noise rejection.
Waveform generation is supported by onboard memory. Some boards offer multiple configurations similar to those of real Arbs that allow segmentation, linking, and programmed numbers of repetitions. In addition, boards may have preprogrammed standard waveform and selectable output filtering capabilities. However, many lower cost boards simply provide basic analog outputs that may or may not be adequate for your application regardless of DAC resolution.
Digital outputs generally are not supported by onboard memory but instead simply are programmable. Those digital outputs driven by data stored in an onboard memory are termed vectored outputs by at least one manufacturer. This company also uses the vectored nomenclature to refer to input digital channels simultaneously sampled with analog inputs. A vectored digital output could be used to generate a fast arbitrary series of pulses, for example.
Triggering If you want to acquire data only when a certain event occurs, you need to consider the triggering facilities a board supports. Manufacturers with historic links to oscilloscope design may offer comprehensive triggering capabilities ranging from simple analog level and slope detection to sophisticated if/then/else logic together with programmable timing constraints.
Most boards can trigger an acquisition as a result of an external event, a software command, the detection of a digital value, or detection of an analog level crossing in a given direction. A common variation on these basic capabilities is triggering if a signal enters or leaves the band between a pair of values, called window trigger.
In many boards, the action initiated by the trigger is selectable. It can start data capture, which results in 100% post-trigger acquisition. It can end recording, resulting in 100% pretrigger data. Or, it can end acquisition after a programmable time delay or number of events, resulting in a mix of pre- and post-trigger data being acquired.
The amount of memory available to support these modes and how it is split between pre- and post-trigger data depend entirely on the particular board being considered. The total memory also can be segmented so that successive acquisitions use separate segments.
This feature is useful when the results from each pulse in a burst of laser pulses must be compared, for example.
Today and the Future
Special onboard capabilities available today include the following:
• High-resolution trigger event timestamping.
• Signal-conditioning bypass to achieve full ADC performance.
• External 10-MHz reference input.
• Transformer-coupled SE-to-diff inputs.
• A reconfigurable data processing unit for filtering, averaging, and analysis.
• A 2-GS/s sampling rate with up to 8-GB of memory.
• 24-b resolution via sigma-delta ADCs.
Several manufacturers cited achieving lower prices, greater functionality, and better reliability through the use of custom ASICs. In addition, improvements have been made to associated software that help to streamline system development. One unfortunate consequence of the large range of special features, however, is lack of detail in data sheets. Certainly, the new features are well explained, perhaps even with examples, but including this information often excludes basic operating details relating to triggering, external clocking, and bandwidth, for example.
In the future, data acquisition boards will continue to evolve. Plus, you can expect exciting changes as PCI-X becomes more widely adopted. This interface technology provides up to a 1-GB/s data transfer rate, which supports high-speed boards that also are low cost because they don’t require expensive onboard capture memory.
To View Data Acquisition Chart Click Here
To View Data Acquisition Chart Click Here
March 2005