Delegates to the Third Planet Redesign Conference have agreed that temperature shall exhibit integer values instead of the continuous range it has had. This simple change will greatly enhance measurement accuracy.
By knowing beforehand that a temperature measurement must be a whole number of degrees, you can get exact results from an ADC. In contrast, the digital representation of a signal’s continuous analog value is only an estimate.
How well that estimate corresponds to the original signal depends on the ADC’s resolution, accuracy, differential and integral nonlinearities, sampling rate, bandwidth, and noise. Of course, an ADC does not have a solitary existence, being preceded by analog signal conditioning and followed by digital signal processing. Each introduces additional errors and uncertainties.
An ADC with signal conditioning forms the basis of a digitizer, and some modular digitizers when coupled with suitable software are sold as oscilloscopes. However, although the situation gradually is changing, digitizers most often are modular, and oscilloscopes usually are traditional bench instruments. This distinction typically relates to the way in which each type of device is used—oscilloscopes for troubleshooting and digitizers for data acquisition.
Both instrument types typically have a wide range of triggering capabilities—ways to select only certain kinds of events from the overall signal or to synchronize acquisition with an external trigger. And, especially in very high-speed modules, a significant amount of memory may be included to locally address fast data transfer.
As explained by Chris Gibson, senior product manager, data acquisition at VTI Instruments, “There still is value in benchtop instrumentation…. Having buttons and a display at your fingertips is convenient when troubleshooting. The lack of restriction on size/power/cooling allows for the ability to design for challenging application requirements. A lot of the measurement science, though, is in the [digital] processing, and the ability to build that processing into FPGAs allows vendors to shrink the overall footprint.”
He continued, “The trade-off in selecting a benchtop device is that it is not very scalable; to add channels, one must purchase additional boxes that increase the measurement footprint. With a modular instrumentation architecture, a measurement system can start with small channel counts and scale up easily through the addition of more modules. And with regard to the display/pushbuttons, there are many soft front panels these days that do an excellent job at replicating a physical panel.”
ADC DNA
ADC architectures vary considerably depending on the intended application. For low frequencies including audio, ∑∆ ADCs with up to 24-bit resolution often are used. At the other extreme, high-speed, 8-bit flash converters operate up to at least 10 GS/s and to much higher speeds when interleaved.
High resolution, low speed
Although Guzik Technical Enterprises is best known for its high-speed technologies, comments made by Lauri Viitas, director of business development at the company, are particularly relevant to the discussion of high-resolution ADCs. He said, “… a quick look at the industry shows that modular digitizers dominate the high-resolution market. Since the increased resolution is beyond what the eye can see on a screen, it is natural to process the information (via FPGAs or a controller) to get the measurement insight required by the application. Once this is the case, a faceless instrument is just as effective as a traditional instrument with a screen since the data is valuable only when processed anyway.”
This point of view may be entirely justified. For example, ADLINK Technology’s Zake Lin, product manager measurement and automation product segment, commented, “ADLINK’s PCIe/PXIe-9529 24-bit high-precision signal acquisition module with dynamic range up to 110 dB and eight input channels was chosen for multichannel vehicle navigation testing in China, measuring signal-to-noise ratio (SNR), THD+N, channel crosstalk, volume, primary frequency, and [additional] items. Eight 24-bit sampling analog input channels mean the 9529 can simultaneously measure two four-channel devices, and included LabVIEW drivers allow quick development of in-house testing programs.”
On the other hand, being able to directly observe small signal aberrations often helps an engineer identify the likely cause. A few 12-bit bench oscilloscopes have become available, and resolution up to 16 bits is provided by oversampling. And, National Instruments’ PXI-5922 is listed as a flexible resolution oscilloscope/digitizer, which trades off resolution for speed—24 bits at 500 kS/s to 16 bits at 15 MS/s.
In a 2010 tutorial, Analog Devices’ Walt Kester wrote, “The ∑∆ ADC is the converter of choice for modern voiceband, audio, and high-resolution precision industrial measurement applications.”1 The noise shaping inherent in this type of converter makes possible the high resolution provided by many models of ∑∆ devices. Sampling at a very high rate relative to the final output data rate causes the converter’s noise to be spread over a large bandwidth. Filtering the output eliminates much of this noise, improving the SNR. Oversampling in any kind of ADC has this benefit. However, in a ∑∆ ADC (Figure 1), the modulator further shapes the noise, as Kester explained, “… so that it lies above the passband of the digital output filter, and the ENOB [effective number of bits] is therefore much larger than would otherwise be expected from the oversampling ratio.”1
Courtesy of Analog Devices
To put things in perspective, 24 bits of resolution correspond to one part in 16,777,216, or 59.6 nV out of a 1-V signal. At room temperature, the rms noise generated by a 1-kΩ resistor is about 400 nV over a 10-kHz bandwidth. Therefore, an obvious caveat to designing with high-resolution ∑∆ ADCs is the use of low-noise techniques. Minimizing the bandwidth (∆f) helps a great deal because the noise is proportional to √∆f.
High-speed physical applications
Improved IC process technology together with ∑∆ ADC architecture evolution have allowed higher speed applications to be addressed. VTI’s Gibson explained, “… advances in ADC technology allow end users to get more granularity in the data that they are analyzing, but it pushes the volume of data to be processed to new levels. Our first-generation products were maxing out at 51.2 kS/s and 16 bits of resolution. Our latest digitizers now can sample every channel simultaneously at 625 kS/s and 24-bit resolution.”
The four-channel EMX-4350 PXIe digitizer Gibson referred to features true differential inputs and very high alias signal rejection—two factors contributing to the typical -125-dBFS spurious-free dynamic range (SFDR) specification and -98-dB THD rating from 20 Hz to 20 kHz. As a guide, since SNR and signal to noise and distortion ratio (SINAD) values are not listed in the datasheet, an ENOB of about 14.7 bits corresponds to -90-dB SINAD.
As described by Klaas Vogel, a consultant at Elsys, “Our products … are frequently used in the field of structural health monitoring through acoustic emission and ultrasonics. The purpose is to detect, store, and analyze surface displacements in rock, concrete, brick (buildings, bridges, mining, oil-drilling, etc.) of an internal, external, and/or seismic nature.”
The company’s TPCE-120-16-4x and TPCE-240-16-4x modular digitizers sample with 14-bit resolution at 120 MS/s and 240 MS/s, respectively, and 30 and 60 MS/s at 16 bits. Contrary to conventional resolution enhancement that averages 16 successive samples to add two bits of resolution, 16-bit resolution is achieved with only a factor of four speed reduction.
Increasing resolution by oversampling and averaging often relies on the inherent noise present in the system. For random uncorrelated white noise, the oversampling ratio = 2(2n) where n is the number of added bits. Dithering uses a signal that has been summed with the input to cause more ADC levels to be crossed in successive samples than otherwise would be. The signal can be the system’s own noise, or it can be a specially developed signal that takes advantage of the ADC’s specific architecture to enhance resolution by n bits but at a lower oversampling ratio. Dithering does not improve an ADC’s integral nonlinearity but can improve differential nonlinearity.
Keysight Technologies’ Jean-Luc Lehmann, high-speed digitizers product manager, commented on the smaller footprint of the recently introduced two-channel M9217A PXIe 16-bit isolated digitizer compared to the earlier L4532A LXI digitizer. In the latter, processing was done onboard while in the newer M9217A it is mostly done on the PXIe controller. Relative to the L4532A, the M9217A appears to have a little more noise, resulting in slightly lower SNR. The L4532A datasheet also discussed 2-MHz and 200-kHz noise filtering and listed typical ENOB values, but neither the filters nor the ENOB numbers are included in the M9217A datasheet.
Both the LXI and PXIe digitizers are intended for automotive and mechatronic applications and feature independently isolated input channels. The channels are protected to 400 V with the maximum input listed as ±256 V high-to-low. In addition, each channel’s low side can float as far as ±40 V off ground to accommodate differential signals. However, the datasheets do not state that the inputs are truly differential, and an isolated input is not the same as a differential input: The high and low sides have different characteristics—they are not balanced.
ADLINK’s PCIe-9814 12-bit digitizer simultaneously samples each of the four input channels at up to 80 MS/s and features a 40-MHz bandwidth. Version PCIe-9814P includes a PLL module for precise clock synchronization. In addition, this fully autocalibrating module provides 1 GB of onboard memory and selectable 10-MHz or 20-MHz digital filtering. As described by the company’s Lin, “The PCIe-9814’s FPGA-based 31st-order FIR digital filter supports noise reduction when signal content is 20 MHz or less…. The FPGA-based FIR digital filter performs much faster than on the host, with no host CPU bandwidth occupied.”
The 9814 datasheet includes values for SNR and THD, both of which vary little with either input range or input impedance, being approximately 64 dB and -73 dB, respectively. With these values and the relationship
S
SINAD = 20 log (——————-)
N+D
SINAD = 63.5 dB and ENOB = 10.25 bits measured with a 10-MHz signal having a -1-dBFS amplitude, sampled at 80 MS/s, and with no output filtering. With the 20-MHz output filter turned on, ENOB increases slightly to about 10.5 bits.
National Instruments makes no secret of leveraging technology developed by other companies. Often, this means using the latest Analog Devices ADCs, but for the 5-GHz, 12.5-GS/s PXIe-5186 development, the company partnered with Tektronix. As stated on NI’s website, “The analog front end and ADC ASICs incorporated in the 3-GHz bandwidth NI PXIe-5185 and 5-GHz bandwidth NI PXIe-5186 are state-of-the-art silicon germanium parts designed by Tektronix and used across the full suite of the Tektronix high-performance oscilloscopes.” Like many relatively high-speed digitizers, this device has eight-bit resolution—the same as many of Tek’s higher bandwidth scopes.
Very dense or fast modules
NI’s Ben Robinson, product manager-modular instruments, cited the company’s eight-channel PXIe-5171R reconfigurable oscilloscope (Figure 2) as an application of the compact and low-power Analog Devices 14-bit ADCs as well as the increased logic density and larger number of signal processing resources in the Xilinx Kintex-7 FPGAs. He said that including a user-accessible FPGA means that “… engineers and scientists can perform real-time processing and software-defined measurements on hardware, eliminating the need for complex post-processing operations.”
Courtesy of National Instruments
Guzik’s Viitas agreed, adding, “… [An] advantage of modular digitizers is that they are fundamentally designed for signal-processing applications. By placing powerful FPGAs onboard, coupled to a high-speed PCIe bus back to a controller or GPU, the user is given a new palette of options for placing signal processing where it is needed, an architecture that simply is not possible with traditional instruments.“
The AXIe format was developed to support large, high-power applications, such as the developing 5G standards, phased-array radar, and high-speed data acquisition in physics, which all require from tens to hundreds of channels. Viitas explained, “A modular solution is a perfect match because the user installs the number of modules needed to meet their application. The high-speed trigger and timing bus found in AXIe can be used to keep all the channels synchronized.”
Guzik’s ADC 6000 series 8-bit digitizers (Figure 3) provide one channel at 40 GS/s and 13-GHz bandwidth (6131), two channels at 20 GS/s (6082), or four channels at 10 GS/s each (6044). Up to 128 GB of onboard memory is available, enabling long data records. A PCIe Gen 2 link on the AXIe backplane ensures high-speed data transfer for further processing.
Courtesy of Guzik Technical Enterprises
Keysight has introduced the eight-channel 12-bit M9703B AXIe module (Figure 4) with up to 3.2-GS/s sampling rate, 2-GHz bandwidth, real-time digital downconversion, and a time-to-trigger interpolator. In addition, the B01 option adds real-time signal-processing capabilities. The company’s Lehmann said, “This system was developed for designing and testing satellite telecommunications systems where the main requirement was multichannel synchronization and FPGA programming capability.”
Courtesy of Keysight Technologies
The 32-channel M9709A 8-bit digitizer samples at up to 1-GS/s, accommodating signal frequencies up to 500 MHz. Up to 16 GB of onboard memory supports synchronous data acquisitions.
Summary
Modular digitizers are at the heart of data-acquisition systems and, with appropriate architecture and software, can perform well as oscilloscopes. Of course, a digitizer has an analog front end, and for best results, it needs to match your signal: differential or single-ended, voltage range, floating or grounded, 50 Ω or 10 MΩ, etc.
The digitizer’s ENOB combines accuracy, resolution, noise, and distortion so all of these factors must be optimized to result in a high ENOB. If a higher ENOB device is available that also suits the rest of your requirements, your measurements will be more accurate, although probably more expensive as well.
Whether a digitizer is used as a scope or DAQ system, trade-offs are necessary when you require high performance. As shown via the products that have been discussed, very high-resolution measurements are restricted in speed. Very high-speed measurements typically need more power and more volume in which to dissipate it. The highest speed modules cost more and typically don’t have a large number of channels.
And, if you already have developed part of a PXI/PXIe test system, multiple PXIe digitizer modules may be more economical than opting for a high channel-count AXIe module. However, there are many hybrid systems that mix formats to advantage and some hybrid chassis that support them.
Reference
- Kester, W., “ADC Architectures III: Sigma-Delta ADC Basics,” Analog Devices, MT-022 Tutorial, August 2010.
For more info
