What defines performance in high-speed digitizers?

Today’s advanced, PC-based digitizers for high-speed data acquisition are necessary for myriad applications – ranging from signal analysis, lidar, ultrasound imaging and non-destructive testing, to communications, particle physics, mass spectroscopy and more. This article highlights how digitizers are unique and describes the unique performance parameters engineers should use when selecting a high-speed digitizer.

Digitizers cater to in-process signal testing

While digitizers and Digital Storage Oscilloscopes (DSOs) both use Analog-to-Digital Converter (ADC) chips as their core electronics they are different tools. The DSO is optimized for the probing and visualization of signals that are typically encountered on an electronic engineer’s test bench during circuit development or first article approval of a PCB or cable. The signals are stored in the DSO acquisition memory and displayed on the DSO screen.

A digitizer can also display signals, but it is optimized for the acquisition of signals with high fidelity and at high repetition rates where the near real-time detection of small signal changes is critical. In Automated Test Equipment (ATE) for example, the digitizer typically acquires signals from an investigating probe, like an ultrasonic transducer or a photodiode. Further, unlike the unknown signals on a test bench, a digitizer typically acquires signals whose baseline characteristics are known but which experience small but critical deviations over time.

Also, while digitizers are available as independent box instruments (like DSOs) most modern digitizers are highly integrated components of larger high-speed systems.

Widely used tech platforms help digitizers solve complex problems

Today’s highest-performance high-speed digitizers are installed on a shared electrical bus, such as a PXI or VXI bus within a dedicated instrument chassis or a PCI Express (PCIe) bus within most modern PCs. Modular digitizers may be integrated with other modular instruments under the control of a custom software application to create custom automated test equipment.

A 4-channel, 16-bit, 1000 MS/s digitizer card is shown in Figure 1. On the bottom of the card is the PCIe edge connector used to install it in a PCIe slot within the host PC. Once installed, the card’s SMA signal connectors emerge from the back of the PC chassis.

Fig. 1. Gage RazorMax PCI Express (PCIe) digitizer card with 4 input channels, 16-bit vertical resolution. Source: Gage Applied

Specifications used to differentiate digitizers

Digitizer companies generally advertise three key specifications: channel count, vertical resolution and maximum sampling rate to indicate performance. In this section we will review these and describe why they are by themselves insufficient.

Channel count

The channel count is simply the number of separate electrical signals that a digitizer can acquire. Generally, a digitizer is equipped with only a single acquisition sequencer, so that signals from all channels, while different, are all acquired in an identical fashion – namely, using the same sampling rate and trigger.

Vertical resolution (Bit depth)

Vertical resolution, specified in bits, reveals the number of discrete digital levels into which the vertical input voltage range of a digitizer is divided. This number of levels is equal to 2^B, where B is the number of bits of resolution. For example, an 8-bit digitizer divides the vertical input range into 2⁸ = 256 discrete levels. Accordingly, an 8-bit digitizer is unable to acquire vertical signal features that are smaller than one 256th of the input range. By contrast, a 16-bit digitizer divides the vertical input range into 65536 levels, which proportionately increases the sensitivity to signal features by a factor of 256.

Almost all DSOs use 8-bit ADCs, which are usually sufficient for visualizing signals. Smaller signals may be amplified, and larger signals may be attenuated using the DSO’s broad selection of vertical input. Figure 2 shows how bit depth impacts signal clarity.

Fig.2. Figurative illustration of the relative accuracy of a 4-bit, an 8-bit digitizer ((typical of DSOs) and the 16-bit depth characteristic of the higher-end digitizers. Source: Vitrek

The value of high resolution is evident in applications where there are signals with high dynamic range—that is signals combing both high and low amplitude components. Examples of high dynamic range signals include exponential decay and ultrasonic, radar, or lidar signals that contain echoes from different-sized reflectors. High-amplitude components often inhibit excessive signal amplification, so high resolution is the only means of seeing low-amplitude echo and decay components.

Signal deviation detection needs more than vertical performance

In the process of converting an analog signal into a digital one, distortion can occur. Figures 3a and 3b show a pure sine wave, together with a sine wave that has been compromised by the addition of broadband signal noise and by signal distortion. Distortion is shown as attenuation near the input range limits, which is the typical precursor to signal clipping.

Fig.3a. Illustration of a pure sine wave (black) and one that has picked up noise (red). Source: Vitrek

Fig. 3b. Illustration of a pure sine wave (black) and one that has suffered distortion (red). Source: Vitrek

Consequently, the true measure of the vertical performance of a digitizer is not its nominal vertical resolution but the Effective Number of Bits (ENOB). ENOB is the effective resolution that is achievable in practice accounting for signal distortion and random noise introduced by the digitizer. A digitizer’s ENOB and other Dynamic Parameters are generally measured in the frequency domain.

In contrast to DSOs, dedicated digitizers — such as those on modular platforms like PCIe or PXIe — are usually optimized for the rapid acquisition and analysis of small changes in familiar signals. While providing lower maximum sampling rates, digitizers typically offer vertical resolutions of 12-, 14-, and 16-bits. The 16-bit RazorMax PCI Express (PCIe) digitizer in the example (Figure 1, above) exhibits the industry’s best ENOB of up to 12.1 bits.

 Absolute vs. Relative Accuracy

There is an important distinction between the absolute and relative accuracy of digitizers. The absolute accuracy of a digitizer describes how close its measured voltage values correspond to true absolute voltage reference standards. By contrast, relative accuracy specifies the fidelity of the shape of the acquired waveform with no reference to absolute reference standards. Using onboard calibration techniques, a high-speed digitizer may achieve absolute accuracies of an order of 0.1% of the full-scale input voltage range. In most digitizer applications, however, users are principally concerned with relative accuracy, which is specified by the Dynamic Parameters. The fidelity of a signal acquired by a digitizer device may be compromised by three distinct factors:

1. Addition of random noise by the digitizer to the acquired signal.
2. Distortion of the acquired signal by the digitizer.
3. Irregularities in uniformity of the time intervals between samples acquired by the digitizer arising from imperfections in the ADC clocking signal.

Maximum sampling rate

The sampling rate, usually measured in Mega Samples per second (MS/s), refers to the number of samples acquired by a digitizer per unit time. Nyquist sampling criterion requires the sampling rate to be at least twice the maximum frequency that a signal contains.

The Importance of Frequency Response

While the sampling rate is a key digitizer specification, an often-overlooked element of its time-domain performance is its frequency response curve. This curve is measured by acquiring a fixed amplitude sine wave with a digitizer and determining apparent measured sine wave amplitudes over a broad frequency range.

The frequency response curve for the RazorMax PCIe digitizer is shown in Figure 4. Frequency response curves function to show the analog input bandwidth, which is the frequency beyond which the digitizer attenuates a sine wave signal by 3dB or more. The bandwidth can be viewed as the -3dB roll-off frequency of the low-pass filter that a digitizer effectively presents to an input signal.

A common digitizer rule-of-thumb requirement is that the maximum frequency component within a signal acquired by a digitizer must be less than the digitizer’s bandwidth. This rule is often explained using the simplifying assumption that the digitizer behaves like a brick wall filter (more specifically a brick wall low-pass filter) that passes everything in the Pass Band below the bandwidth frequency (no attenuation) and that blocks everything in the Stop Band above the bandwidth frequency (infinite attenuation).

Such abrupt brick wall roll-off is not a good representation of the gradual roll-off of the curve of Figure 4. Consequently, determining the maximum usable signal frequency for a digitizer is a more nuanced decision than suggested by the common rule of thumb.

In the example of Figure 4, although a 200MHz signal is well below the 300MHz bandwidth of the digitizer, there is still an attenuation factor of -1dB, which corresponds to an amplitude decrease of about 11%. This significant attenuation may or may not be acceptable to the user, depending upon the application.

Fig. 4. Frequency response curve of the Gage digitizer shown in Fig. 1. The vertical red line indicates the 600 MHz analog input bandwidth (or -3dB roll-off frequency) of the digitizer. Source: Vitrek

Notwithstanding that the frequency response curve of Figure 4 is not a brick wall, it does indeed indicate a very good frequency response. The Pass Band shows a very flat response, and the attenuation is constant within 0.2dB up to a 100MHz signal frequency. As the curve rolls off, it shows only small residual oscillations of below 0.1dB. Poorly designed digitizers will exhibit large dB-order oscillations throughout their frequency response curves.

Design engineers must optimize the propagation of high-speed signals on a digitizer circuit board. Since signal wavelengths are of a similar scale to circuit board trace lengths, they must carefully control the electrical impedances and lengths of electrical circuit traces, stray capacitances, and other elements. These measures avoid the creation of distorting signal reflections, which result in large oscillations within the frequency response curve.

It is worth noting that the Nyquist sampling criterion employs and suffers from a similar simplification as the digitizer bandwidth rule-of-thumb. The Nyquist criterion requires 2X oversampling of the maximum signal frequency (or signal bandwidth) under the assumption that a signal has a brick wall spectrum, with no frequency components above the signal bandwidth frequency. A real signal will roll off more gently than a brick wall, so determining the required sampling rate becomes more nuanced. Often 5X or 10X over-sampling factors are employed to ensure that high-frequency signal components beyond the signal bandwidth are accurately acquired.

Conclusion

To maximize flatness in the Pass Band of the frequency response, engineers might maximize digitizer bandwidth so that higher signal frequencies could be acquired without attenuation. However, a digitizer’s bandwidth and ENOB are antipathetic –as one improves, the other degrades. Higher bandwidth reduces the low-pass filtering effect upon a signal. This lets through more high-frequency noise, which in turn degrades the ENOB.

Consequently, digitizer engineers must determine the optimal operating point that provides both sufficient ENOB performance and sufficient bandwidth. In addition to knowledge of its ENOB, consideration of its frequency response curve is paramount to the optimal selection of a digitizer for a given application.

————————————-

Andrew Dawson, engineering manager & application specialist, Vitrek, LLC.