BY BRIG ASAY
Product Manager High Performance Oscilloscopes, Oscilloscope Product Div.
Agilent Technologies
www.agilent.com
The dramatic growth in mobile devices such as smartphones and tablet computers, along with the transmission of rich data content such as high-resolution images and video, continues to drive the need for more data handling capacity across the worldwide communications network. Multiple approaches for increasing data capacity are being employed, with the most familiar one being to increase the data rates of the signals used to carry the information. This in turn is driving high-speed digital designs into tens-of-gigabit-per-second rates, with optical research work occurring at the hundreds of gigabits/s and even terabit-per-second levels.
Fig. 1: LeCroy 10Zi and Agilent Q-Series ultra-high-performance oscilloscopes.
As data rates continue to increase, measurement systems must have sufficient bandwidth to accurately characterize the signals. For many designers of high-speed equipment, the measurement instrument of choice remains the real-time oscilloscope. However, for several years real-time oscilloscope bandwidth seemed plateaued at the 20-GHz level.
In 2009, LeCroy (now part of Teledyne) introduced the first 30-GHz oscilloscope. Agilent quickly followed with a 33-GHz family. Then, just two years later, both companies announced real-time oscilloscope bandwidths of greater than 60 GHz, with Agilent shipping 63-GHz products in July of 2012. In March of this year, Tektronix announced its intent to ship a 70-GHz oscilloscope sometime in 2014.
Essentially, real-time oscilloscope bandwidth has more than tripled in only three years, making it possible to accurately characterize data rates which previously could only be assessed using repetitive sampling techniques. Now the question must be asked: What technologies have enabled the significant bandwidth increases of real-time oscilloscopes to occur in such a short period of time?
Sample rate and bandwidth
On a first-order basis, a real-time oscilloscope’s measurement bandwidth is constrained by two core attributes: the actual analog bandwidth of the front end, and the sample rate of the analog-to-digital converter (ADC) system that samples the signal passing through that front end.
Fig. 2: Simplified oscilloscope signal path.
Starting with the latter of these two attributes, we know that to minimize measurement error due to aliasing, oscilloscope sample rates should be between 2 and 2.5 times the front-end bandwidth. While integrated ADC chips continue to get faster, commercially available 8-bit parts are currently limited to sample rates in the 60-Gsample/s range. So considering just the sample rate constraint on measurement bandwidth, an oscilloscope based on a 60-Gsample/s ADC would be able to achieve a maximum bandwidth of somewhere between 24 and 30 GHz (with aliasing becoming an issue as 30 GHz was approached).
Achieving 60-GHz bandwidth would require an ADC operating at 120 to 150 Gsamples/s. Such an ADC cannot be found on the commercial market today. As a result, oscilloscope manufacturers use a variety of techniques to combine multiple ADCs together into systems that can achieve higher overall sample rates. Agilent’s 90000 Q-Series, for example, combines eight 20-Gsample/s ADCs to achieve an overall sample rate of 160 Gsamples/s in order to properly digitize signals up to 63 GHz. The LeCroy 10Zi-A uses four 40-Gsample/s ADCs to achieve its 160-Gsample/s operation. Exactly how the ADCs can be combined will be described later.
But as mentioned above, adequate sample rate is not the only constraint. Oscilloscope manufacturers also face the challenge of achieving high front-end bandwidth, which is largely a function of the preamplifier (preamp) performance. The preamp of a real-time oscilloscope has multiple purposes. It presents a dc coupled 50-O termination impedance at the oscilloscope inputs, it provides part of the means for scaling the input signal for subsequent circuit blocks including the ADC, it provides anti-aliasing (band limiting) at the maximum sample rate, and it ultimately drives the ADC’s sampling circuits as well as the trigger circuits. The preamp must have more bandwidth than the input signal or it becomes a low pass filter and blocks high-frequency content. A key factor in obtaining high-preamp bandwidth is the speed of the semiconductor technology used to implement the circuit design.
The role of semiconductor technology
From an amplifier bandwidth standpoint, the most important specification of a semiconductor process is the cutoff frequency of the transistors fabricated in that process. The cutoff frequency is the frequency at which the output power of the device has fallen to a given proportion of the power in the passband, typically a factor of ½ or -3 dB. Two semiconductor processes often used for high-speed analog circuits are silicon germanium (SiGe) and indium phosphide (InP).
These processes currently produce transistors with cutoff frequencies in the 200-GHz range, enabling multiple transistors to be configured into amplifier circuits in the 30-GHz range. (Amplifier bandwidths are typically 5 to 10 times lower than the cutoff frequencies of individual devices because they require sets of transistors arranged in multiple stages, and in some cases make use of feedback topologies.) Today’s high-bandwidth oscilloscopes from LeCroy and Tektronix use commercial SiGe for their preamp circuits, while Agilent uses an in-house InP process.
Advanced architectures
The above discussion has been framed around the classical signal path design depicted in Fig. 2, that is, a straightforward flow from a preamplifier block to an ADC block. However, the need for higher oscilloscope bandwidths is outstripping what can be achieved with such an architecture, even when very fast semiconductor processes are used to realize the circuits. As a result, oscilloscope manufacturers must use alternative methods for achieving higher bandwidths and sample rates than are possible with individual preamp and ADC chips. Multiple approaches are possible for combining individual circuit blocks in order to achieve this goal; some approaches can be combined with one another and all of them involve some degree of manufacturer-specific know-how in the implementations.
In all of these techniques, the full oscilloscope bandwidth must be distributed across the individual ADCs in a way that enables the measured signal to be uniquely recoverable. The conventional method used for this is time interleaving, in part because it is the easiest to understand and implement. (While called “time interleaving,” this approach also has a frequency-domain interpretation.) The basic idea is to digitize the input signal with multiple time-skewed ADCs, each operating at lower sample rates.
For example, to double the overall sample rate, two ADCs can be interleaved by one half each ADC’s sample interval. Note that the bandwidth and sample rate of an ADC are not necessarily related. If the ADCs being interleaved each have a large enough bandwidth, then no other front-end processing is needed. In practice the ADCs used for ultra-high-performance oscilloscopes usually have less than the full oscilloscope bandwidth, and a sampler (or sometimes a mixer) prior to the ADCs is required. The sampler acts as a type of downconverter in this case – it spreads the signal’s spectral content across the two ADC’s lower individual bandwidths.
More recently, some of the highest-bandwidth oscilloscopes have employed variations of what are sometimes called hybrid filter bank (HFB) techniques. Basic HFB uses analog bandpass filters to select and redirect separate frequency bands to each preamp and ADC. Four ADCs can be combined, for example, by dividing the full oscilloscope bandwidth into four separate band-limited signals that are then mixed down to fit within each individual preamp and ADC’s lower bandwidth.
Note that the bandpass filters and mixers used in this technique perform the same function as the time-interleaving and sampling do in the previous technique. LeCroy uses the term “Digital Bandwidth Interleaving” (DBI) for its version of this general approach. The company also used DBI for its 30-GHz products introduced in 2009 and 60+-GHz products introduced in 2012. Agilent began shipping its “RealEdge” technology in 2012 when it moved from its previous 33-GHz products to the 63-GHz models of the 90000 Q-Series. The RealEdge architecture consists of a proprietary combination of the basic time interleaved and HFB techniques.
Two additional points should be made regarding these advanced techniques. First, by themselves, neither the time interleaved nor HFB approach has an inherent advantage in other important system attributes such as noise, power or cost. However, one may be better suited to existing hardware or software when oscilloscope manufactures desire to leverage existing technology into new products. Or one may be a better fit to a specific manufacturer’s area of technical expertise. The RealEdge hardware, for example, is largely implemented using Agilent’s in-house microwave circuit and packaging processes, including the previously mentioned InP semiconductor fab. These capabilities tap the company’s longstanding expertise in developing RF-oriented products such as spectrum analyzers and network analyzers.
Secondly, while this article has focused on hardware architectures, it should be mentioned that all of these approaches require significant software processing as well. In fact, the signal processing software is where some of the most sophisticated contributions of the advanced architectures are made.
While Agilent is shipping its 63-GHz product using its new technologies and LeCroy has demonstrated its 60+ GHz product at trade shows, few details are known about the architecture and performance of the future Tektronix high-bandwidth oscilloscope. Tektronix has indicated they will use a form of interleaving called “Asynchronous Time Interleaving,” which appears to be another variant of the advanced techniques discussed here.
Looking forward
Part of what makes the oscilloscope market fascinating is watching the manufacturers compete for higher bandwidths, while also striving to achieve very low noise and jitter. (Without explicit design focus, these two attributes typically get worse with higher bandwidth.) The current environment is anything but static. Semiconductor processes are moving to greater-than-300-GHz cutoff frequencies. LeCroy’s purchased by Teledyne last year came with indications that they will take advantage of that company’s InP fabrication facility. And customer demand for still more measurement bandwidth will continue to drive innovation. Oscilloscope manufacturers will thus continue to push the limits in the use of both new semiconductor technologies and advanced architectures and topologies.
At the same time, while the bandwidth competition makes for dramatic headlines, at the end of the day the core contribution of an oscilloscope is providing the most accurate depiction of the signal that is being measured, and this involves product attributes beyond bandwidth. Technologies such as RealEdge, Digital Bandwidth Interleaving, and Asynchronous Time Interleaving will often come with tradeoffs in these other attributes. Specifications such as noise floor, jitter measurement floor, and frequency response need to be considered based on the specific measurement application. These specifications involve a rich body of practice of their own, and are sure to show up in future articles covering the world of ultra-high-performance oscilloscopes.
http://cdn.teledynelecroy.com/files/pdf/lecroy_labmaster_10_zi_datasheet.pdf
https://docs.google.com/viewer?url=www.google.com/patents/US5568142.pdf
https://docs.google.com/viewer?url=www.google.com/patents/US5659546.pdf
https://docs.google.com/viewer?url=www.google.com/patents/US7219037.pdf
Advertisement
