With Digitizers, The Little Bits Count

Separation of the elements that have traditionally composed a single instrument is a growing trend within the ATE industry. A good example of user-accessible modularity is the range of synthetic instruments currently being developed to address military test set obsolescence.

Rather than an RF signal analyzer having its own local oscillator, down converter, digitizer, analysis software, and display, these functions are provided as individual modules. They are combined to create an RF signal analyzer and can be recombined to create other instruments. Because each module is separately replaceable or upgradeable, obsolescence is largely avoided.

The distinction between an oscilloscope and a digitizer is one of the issues encountered when reducing a traditional instrument to its constituent parts. It’s clear that a DSO performs the digitizing function, but as Gleb Geguine, software manager at Gage Applied Technologies, explained, there’s more to it than that:

“While the internal components of a digitizer are very similar to those of a DSO, the two instruments are optimized differently. DSOs are used for visualization of unknown signals acquired during benchtop probing while digitizers accomplish the fast automated acquisition of signals whose broad nature is known but that under-go relatively small changes as some other system variable is altered.”

Because a digitizer’s rate is much faster than a DSO’s, Mr. Geguine highlighted the rate of waveform transfer to the host PC as another distinguishing feature. On the other hand, an internal DSO bus can be dedicated to fast memory access and data analysis, allowing a DSO to quickly produce and transfer a measurement result to a PC.

Both limited data transfer rates and the need for high-fidelity visualization affect the performance of PC-based scopes. Some manufacturers address this trade-off by transferring only the amount of data needed to update the available display pixels. The complete data file corresponding to an acquisition only gets transferred for archiving or if needed to perform a measurement not locally supported by the module.

Of course, you could simply use a much faster interface bus such as PCIe. With the advent of multicore processors, PCs have gained substantially improved data-processing capabilities. The combination of multicore and PCIe makes it practical to transfer even relatively large files in their entirety and shift the analysis burden to the PC.

The dedicated nature of PCIe lanes also simplifies high-speed data streaming because the bus is not shared. Digitizers may or may not stream data, but DSOs do not. Traditionally, scopes have operated in a repeated single-shot mode: trigger, acquire, display; trigger, acquire, display…. The reason for this is obvious if you try to look at a truly continuous streaming image rather than a sequence of related frames.

Certainly, some scopes can stream acquired data, but when they do, they are being used as digitizers. Although definitions of digitizers and scopes considerably overlap, digitizers have at least four distinguishing characteristics:
• Modularity
• Nonvisual application
• Fast data transfer to host PC
• Possible streaming data acquisition

Many products fall into a digitizer+ or DSO+ classification with enhanced functionality intended to address a primary market segment as well as secondary ones. To be certain that they are not excluding possible users, manufacturers label these instruments scopes/digitizers. A good example would be a PCIe-based module with streaming and on-board measurement capabilities. With appropriate software and a fast PC, it could perform as a digitizer or a DSO.

Classifying Digitizers by Resolution

You can segment the digitizer market in many ways, including by interface bus type, sampling rate, number of channels, bandwidth, and resolution. Resolution is a good metric to use because it is dictated by the digitizing application and largely determines speed.

Figure 1 was constructed from representative digitizer specifications and shows the approximate shape of the sampling rate-resolution boundary. At the left side, increased resolution is limited by noise: 24 to 26 bits is equivalent to a good-quality 7½-digit DMM. At the right side, extremely high sampling rates imply high power dissipation and expensive devices. Nevertheless, ADC specifications continue to improve each year so the performance boundary gradually shifts up and to the right.

Figure 1. Digitizer Resolution vs. Sampling Rate

Improvements in ADC technology are key to increased digitizer performance even though each new generation of ADCs may provide only incremental gains to what were already good specifications. Kars Schaapman, president of Applicos Measurement and Control, commented, “Designing very high-performance digitizers is quite a challenge because the latest ADC chips have such good specifications that it is difficult to maintain them through the ranging and signal conditioning stages. If resistor values do not fully match the op-amp properties or capacitors with insufficient THD performance are used, the chances are high that the total performance of the design will not surpass the specifications of a previous generation of ADC chips.”

Most recently, there has been an increase in the number of 14- and 16-bit products available with sample rates greater than 100 MS/s. On the other hand, the largest sales volume is for digitizers with 12- to 16-bit resolution and sampling rates up to 100 MS/s. Because there is little difference in price, 16-bit units have displaced many older 12-bit models for general-purpose, lower speed applications.

Whether you actually benefit from the additional resolution depends as much on signal conditioning external to the digitizer as it does on the digitizer itself. A 10-V signal input to a 16-bit ADC results in the LSB having a value of about 150 µV. If the signal contains millivolts of noise, either you must filter it to remove the noise or accept a result with perhaps 12-bit precision.

One clear advantage of higher resolution is greater dynamic range. This can be particularly useful if the signal includes a large and variable offset. In this case, the input to a 16-bit digitizer can be scaled to capture the expected signal with 12-bit resolution while accommodating an offset up to 16x the signal amplitude. Dynamic range is one reason that bioengineering often requires higher resolution.

Bustec’s Model 3424 24-bit ProDac Function Card provides eight differential or single-ended channels, each with a separate sigma-delta ADC producing samples at a maximum 216-kHz rate. Up to four cards can be mounted on a Model 3180 VXI Motherboard for large channel-count systems, or the 3424 can be mixed with other ProDac modules. The motherboard features a 320-MB/s internal throughput rate, and the digitizer includes 6-pole anti-aliasing Bessel filters.

The actual ADC clock rate is software selectable as a 32x, 64x, or 128x multiple of the output sample rate. A further decimation factor also can be selected as 0, 10, or 100. For both the ADC filter and decimation, the passband, stopband, and group delay are functions of the sample rate.

Louis Hsu, product manager for the Measurement and Automation product segment at ADLINK Technology, described a few typical uses for his company’s 12- and 16-bit digitizers. “We are focused on developing high-resolution and high channel-count digitizers. High resolution is especially beneficial for IF and ultrasonic signal acquisition applications where dynamic performance is a primary concern. Multichannel simultaneous acquisition is a typical requirement for electronic testing and military applications.”

Agilent Technology’s Darlene Carpenter, a product manager, said, “Our Models L4532A two-channel and L4534A four-channel 16-bit digitizers sample at 20 MS/s, a rate ideal for electromechanical test and measurement applications often seen in automotive, aerospace/defense, and industrial control environments. The combination of 16 bits and 20 MS/s also makes these instruments suitable for vibration analysis.”

Many digitizers include on-board measurement capability, but that is accomplished by a firmware algorithm after the signal has been digitized. The L4532A and L4534A inputs are isolated and rated to ±250 V to address specific test requirements such as high-side shunt measurements. Bandwidth typically is 20 MHz with a choice of 2-MHz or 200-kHz bandwidth noise filtering.

National Instruments’ (NI) Senior Product Manager for Digitizers/Oscilloscopes Rebecca Suemnicht described an optical coherence tomography (OCT) development that required 256 synchronized channels. “Although high channel-count digitizers are well known for being used in ultrasonic nondestructive test applications, a related area gaining momentum is medical imaging. Often, tens or hundreds of channels must be acquired simultaneously,” she explained. “Dr. Ohbayashi of Kitasato University in Japan developed an OCT cancer detection system using 32 NI PXI-5105 eight-channel, 12-bit, 50-MS/s digitizers.”

Applicos’ Mr. Schaapman discussed the importance of both high resolution and speed to his company’s digitizer customers. “Our digitizers are used in high-end applications where electrical performance is most important. Many of the users are in the semiconductor test industry. They measure and characterize DACs and other on-chip analog features. Here, an excellent dynamic performance in combination with high absolute accuracy is required,” he continued. “Our digitizers also characterize MEMS sensors that require good absolute accuracy but at the same time operate so fast that they need sample rates in the megahertz range.”

On-Board Data Processing

For many years, benchtop DSOs have included built-in waveform measurements. One reason for doing this was to increase instrument usefulness. In addition to viewing wave shapes, you also could determine signal parameters without having to transfer files to a PC for processing. A related reason was speed. In an automated test application, transferring measurement results is much faster than dealing with data from an entire acquisition.

Today, the latest generations of FPGAs provide a similar capability in digitizers. Those products classed as scopes/digitizers often include built-in measurements, mimicking the functionality of benchtop DSOs. In addition, some digitizers support custom algorithms.

There is a distinction between algorithms that distribute well across multiple parallel processing engines and those that don’t. The performance of the automated measurements usually found in DSOs may improve in a massively parallel environment, but it’s not necessary. On the other hand, some signal-processing techniques that simultaneously deal with a large number of information channels do require a massively parallel environment for highest efficiency. New FPGAs may contain hundreds of DSP slices that can be configured to address these applications much more effectively than even a multicore PC.

Anthony Hunt, vice president and CTO at Signatec, said, “This embedded processing on-board the digitizer plays a very significant role in signal intelligence-based applications. Digitizers now produce the frequency spectrum results as well as the desired narrowband measurements from within the captured wideband spectrum. These are processing-intensive tasks that in the past required very expensive additional pieces of hardware.

“In addition to the powerful processing capabilities of new FPGAs, their increased density has enabled Signatec to implement the two-channel Model EC14150 14-bit, 150-MS/s digitizer in the ExpressCard™ form factor.” He continued, “This is a miniature card module defined by the PCMCIA organization that can plug into standard laptop computers. We expect this format to become widely accepted in mobile computer-based signal capture applications.”

GaGe Applied Technologies manufactures many models of PC-based digitizer and oscilloscope boards. Several of these, such as the recently released BASE-8 CompuScope Digitizer, feature eXpert^™ on-board FPGA signal analysis functionality that includes FIR filtering, signal averaging, peak detection, FFT, or customized processing algorithms. Rather than transfer raw data to the host PC, the results of these operations are transferred, offloading the computations from the PC.

Data Storage and Transfer

Inherent in a digitizer’s maximum sample rate specification is the guarantee that there is somewhere for the data to go as fast as it’s being sampled. An amount of on-board RAM usually provides this function, but implementations range from a temporary FIFO capability to a full-blown segmentable memory system.

Data buses such as PCIe have reduced the need for large amounts of memory in most applications. At the same time, very high-speed ADCs must have a large local memory to achieve even a fraction of a second buffering.

Signatec’s Mr. Hunt said that recording to memory and transferring from memory to the host PC can occur simultaneously. The company calls continuous acquisition PC Record mode, indicating that the rates of the simultaneous read and write processes are the same. As an example, the dual-channel, 16-bit PDA16 Digitizer sustains continuous 160-MS/s data acquisition and transfer rates. PCI-X technology with a theoretical maximum 1.06-GB/s rate provides the required 640-MB/s transfer speed.

On the company’s dual-channel, 14-bit, 400-MS/s PX14400 Digitizer, the eight-lane PCIe x8 bus does limit streaming performance to 350 MS/s simultaneously on both channels. However, a single channel can stream at the full 400-MS/s rate. Signatec designs its digitizer buffers to hold about ½ second of data at the peak sampling rate.

Strategic Test offers six 14-bit digitizers that use the PCI-X bus. The same acquisition specifications are available on the PCIe bus as well. All cards include 256-MB memory as standard and can be upgraded to 4 GB. Bob Giblett, the company president, added that hardware options such as memory segmentation for radar applications, timestamping, and asynchronous digital I/O also are available.

ZTEC Instrument’s Boyd Shaw, director of sales and marketing, discussed the improvements the company made in its new range of M-Class Digitizers/Oscilloscopes. The PXI and PCI versions support bus mastering with 32-bit DMA data transfers at up to 66 MHz for a combined burst rate of 264 MB/s. Internally, the instruments have an 800-MB/s data bus rate and a 4,800-MIPS processing speed. These capabilities facilitate on-board measurements and computation of up to four math channels.

Up to 256 MB of memory are available in the M-Class products and can be segmented into as many as 32,000 parts. More acquisition modes also are supported, including normal, average, envelope, peak detect, high resolution, fast re-arm, and equivalent time sampling. Some modes, such as high resolution, require significant data processing.

According to ADLINK’s Mr. Hsu, “The company’s digitizers include an FPGA integrated with proprietary memory controller IP that uses SRAM to emulate a FIFO. With this architecture, the digitizers provide an 84-MB/s sustained throughput to system memory within the 32-bit/33-MHz PCI bus environment. A new range of four-channel, 16-bit digitizers, Models PXI-9816/26/46, is designed with 512 MB of on-board memory to allow longer signal acquisition periods.”

It’s not clear if these products are intended to continuously stream data to the host PC at acquisition rates low- er than the maximum 84-MB/s transfer rate. However, at higher sampling speeds across multiple channels, operation is constrained to be a repeated single-shot mode because of the PCI bus speed limitations.

Agilent’s L4532A and L4534A always operate in this way, the maximum data transfer rate via USB being 8 MB/s and 17 MB/s via Gigabit Ethernet. These products feature scope-like trigger modes and built-in measurements on a selectable part of the overall acquisition. The 32-MS standard or 128-MS optionally extended memory can be segmented into 1,024 records. Segmentation supports a burst acquisition mode that minimizes dead time.

Applying a Digitizer

Thorough specifications support digitizer application engineering. For example, Agilent’s L4532A and L4534A datasheet lists spurious-free dynamic range (SFDR), THD, SNR, signal to noise and distortion (SINAD), and effective number of bits (ENOB) for both 1-MHz and 10-MHz inputs at a range of input sensitivities. Similarly, ZTEC’s ZT410 series of 14-bit and 16-bit high-precision oscilloscopes includes SNR, THD, and SINAD for two representative frequencies.

ENOB is the number of bits resolution a digitizer actually provides under a certain set of circumstances and is directly related to SINAD. A single ENOB value corresponds to one set of conditions, such as a 1-V pk-pk 10.7-MHz signal on the 1-V input range with a 50-MS/s sampling rate.

where: A = signal amplitude
V = full-scale voltage

More generally, ENOB is best expressed as a graph of effective bits vs. another parameter.

Typically, resolution significantly degrades as frequency increases so ENOB vs. frequency is a very useful specification. Unfortunately, digitizer datasheets include it at only one or two points.

Figure 2 shows a plot of ENOB and SINAD vs. frequency for the Analog Devices Model AD9226 12-bit ADC running at a 65-MS/s sampling rate but with four different input configurations. For this part, the highest ENOB is achieved in the differential configuration. A 1-V span has an almost constant ENOB to 100 MHz but is about 0.6-bit lower than the maximum ENOB with a 2-V span.

The figure extends well beyond the first Nyquist zone to characterize the ADC for undersampling applications. Typically, digitizers and DSOs sample at rates greater than twice the highest frequency of interest. However, for many communications applications, the digitizer bandwidth can be the limiting factor. It must be high enough to accommodate the carrier plus modulation while the sampling rate need only be at least twice the modulation frequency.

The availability of diverse options is another factor that allows a wide range of applications to be more completely addressed. As an example, Strategic Test’s UltraFast Digitizer datasheets include an interesting page titled Possibilities and Options:
• The input impedance is selectable as 50 ? or 1 M?.
• Data can be acquired as a single record with variable pretrigger or streamed to the host PC.
• A variety of trigger conditions is available including edge, level, window, pulse width, and external.
• Segmented memory supports acquisitions related to successive triggers with zero dead time between each record.
• Gated sampling acquires data only when the gate signal meets preprogrammed conditions.
• An external clock or a reference clock can be used.
• A number of modules can be synchronized in several ways.
• Digital I/O signals are recorded and output synchronously with analog channels.

Summary

Resolution, sampling rate, number of channels, bandwidth, data transfer rate, and cost remain the basic specifications a digitizer must meet to suit an application. But in contrast to oscilloscopes, digitizer specifications are more comprehensive and the performance options broader. This means that the second-order differences among similar products can be important.

You should be able to narrow the field of possible choices to only the few that best match your needs. Selecting a product that supports custom data processing may improve your test system throughput as well as allow the same hardware to address different applications in the future.

March 2009