This paper was first presented at the National Conference of Standards Laboratories '97, Atlanta, Georgia, USA, on July 28, 1997. Implementing Automated Oscilloscope Calibration Systems Presenter: Richard Roddis Wavetek Ltd. Norwich NR6 6JB, UK Phone +44 1603 404824 Fax +44 1603 483670 Author: Richard Roddis, Wavetek Ltd. Abstract Automating oscilloscope calibration is not easy. Most oscilloscopes have multiple input channels and require a wide range of calibration waveforms. In the past, routing appropriate signals to these inputs has meant the use of a complex computer-controlled signal multiplexer and a large number of interconnecting leads. Both the multiplexer and the leads contribute additional uncertainties to the calibration system. These uncertainties must be evaluated and documented in order to maintain traceability. This paper describes an alternative approach to automated oscilloscope calibration, using fully characterized active heads that deliver all the necessary calibration waveforms directly to the oscilloscope's inputs. Because each head carries its own calibration data, all uncertainties are automatically included in the calibration system's uncertainty calculations. Introduction The increasing accuracy and complexity of modern oscilloscopes, coupled with the requirement of quality standards such as ISO9000 to have them routinely calibrated, is proving a major challenge for calibration laboratories. Although a high proportion of oscilloscopes with bandwidths below 100 MHz still remain uncalibrated, over 90% of the new breed of 400 MHz to 500 MHz scopes (now recommended as the standard workhorse oscilloscope for bench-top use), and virtually all oscilloscopes above 500 MHz, are calibrated. In addition, the vast majority of oscilloscopes sold today are digital-storage models. Because of their higher accuracy these digital storage scopes are increasingly being used to make critical measurements in real production processes, making the need for routine calibration even more important. With manual scope calibration requiring the skills of an experienced calibration engineer and typically taking several hours to complete, calibration laboratories face serious throughput problems. The length of time taken to calibrate modern high-performance oscilloscopes is not only the result of their functional complexity. It has also been a consequence of the lack of fully automated calibration systems. 1
Waveform Diversity Meets Scope Cal Needs Calibration of even basic oscilloscope functions, such as vertical and horizontal gain, vertical and horizontal bandwidth, timebase accuracy and trigger sensitivity, requires the use of a large number of calibration waveforms. Vertical and horizontal gain are normally calibrated using lowfrequency (typically 1 khz) square waves or DC levels. Square waves with amplitude accuracies of around 0.25% have traditionally been used for analog oscilloscopes because they allow DC offsets to be calibrated out. DC levels with accuracies better than 0.1% have been introduced to meet the higher accuracy needs of digital-storage oscilloscopes, many of which now interpolate to 14-bit vertical resolution. As oscilloscopes begin to offer higher resolutions, better accuracies will be required. To calibrate oscilloscope bandwidth a frequency response test using leveled sinewaves is normally used. This involves adjustment of the sinewave amplitude at a low reference frequency so that the trace occupies a given number of graticule divisions (typically 6 divisions), after which the frequency is increased until the displayed amplitude falls by 29% (typically from 6 to 4.2 divisions). The frequency at which this occurs is the oscilloscope's 3dB point (see Figure 1). Although it provides a direct measurement of bandwidth that can be checked against the scope manufacturer's published specifications, leveled sinewave testing provides no information about the phase response of the scope's input amplifiers. The observed 3dB point cannot therefore be used to predict the oscilloscope's risetime capabilities when it is subjected to a step (high-speed edge) input. Amplitude set to 6 divisions at reference frequency (typically 50 khz, 1 MHz or 6 MHz). 3-dB bandwidth indicated by upper frequency at which amplitude falls to 4.2 divisions. Figure 1 3dB bandwidth testing using leveled sinewaves For this reason, some oscilloscope manufacturers also include a separate pulse-response test in their calibration procedures. Because a 150 ps risetime pulse contains frequency components up 2
to and beyond a 1 GHz scope's 3 db point, the ability of the scope to faithfully display the pulse is a good indication of both its gain and phase response. Pulse performance is normally assessed by measuring the risetime, overshoot, undershoot and ringing that appears in the trace (see Figure 2.) Ringing Overshoot Undershoot 10% to 90% Risetime Figure 2 Typical pulse response waveform and aberrations Although leveled sinewave testing is not a good indication of pulse performance, pulse response testing does provide a good indication of bandwidth. The bandwidth can be deduced from the observed 10% to 90% risetime using the equation:- Bandwidth (MHz) = C / (t r1 2 - t r22 ) where t r1 is the observed trace risetime in nanoseconds, t r2 is the risetime of the calibrator output in nanoseconds, and C is a constant that depends on the high-frequency roll-off of the oscilloscope's input amplifier. This roll-off constant is typically 350 or 450, the appropriate value often being found in the oscilloscope's servicing manuals. If you use the above equation to determine bandwidth, it is important that the risetime of the calibrator output is accurately defined. For example, if a calibrator with a specified output risetime of 300 + 0 /-100 ps is used to determine the bandwidth of a 1 GHz scope, the result could be in error by as much as 300 MHz. The calibrator output risetime needs to be much more tightly specified for this equation to be of use. Wavetek's Model 9500 calibrator, for example, has its actual risetime displayed on the calibrator screen. This risetime is specified to within a few picoseconds. 3
In addition to the above analog waveforms, oscilloscope calibration also requires the use of precision timing markers to calibrate timebase accuracy. The triangular timing markers traditionally used to calibrate analog oscilloscopes are not, however, suitable for calibrating digital storage types. This is because a digital storage oscilloscope's asynchronous sampling technique may miss the peak of the triangle, making it difficult to align the timing markers accurately. Crystal controlled squarewaves or rectangular pulses are necessary for calibrating a digital storage scope's timebases, and can also be used to calibrate analog scopes. Additional waveforms are required to calibrate other scope functions such as trigger sensitivity and TV sync separation (for example, TV sync separation requires a composite-sync baseband video signal). To meet ISO9000 recommendations, which encourage the traceable verification of all instrument ranges and functions, calibration waveforms should be available at amplitudes that can exercise all an oscilloscope's VOLT/DIV ranges and multipliers. This typically means that the dynamic range of calibration waveforms should extend from a few mv to ±200V pk-pk. Problems with Connecting Cables Existing automated calibration systems typically employ many different pieces of equipment (DC calibrators, calibration waveform generators, RF signal generators, power meters, pulse generators, video waveform generators, etc.) to generate the necessary waveforms, and then route these via a signal multiplexer to the oscilloscope under test. This multiple-instrument arrangement not only increases the cost, complexity and size of the system (it is likely to occupy most of the space available in a 6-foot high equipment rack), it also introduces significant errors into the overall uncertainty equation, all of which must be traceably identified and taken into account. The multiplexer itself, and the cabling used between the calibration sources and the multiplexer and between the multiplexer and the oscilloscope, can cause significant errors. For example, as little as 1 metre of coaxial cabling between the output of a 150 ps risetime pulse generator and an oscilloscope's input connector will introduce waveform aberrations that will add to those caused by the oscilloscope you are trying to calibrate. Reflections caused by the inevitable mismatch at either end of the cable (i.e. between the cable and the oscilloscope's input and the cable and the pulse generator's output) will arrive back at the oscilloscope input some 10 ns to 15 ns after the original edge. As a result, the reflected signal aberration may be superimposed on the overshoot and ringing you are trying to measure at a point where the oscilloscope's aberration specifications are tight (see Figure 3a). In such circumstances, it is almost impossible to determine how much of the overshoot and ringing is due to the cable and how much is due to the scope. To make matters worse, cables are susceptible to damage and contact degradation (for example, where the cable terminates inside the connectors), which can further complicate the situation by causing additional mismatches and consequential reflections. Oscilloscope inputs never provide perfect 50 Ω loading over their entire bandwidth (their typical VSWR can be as high as 1.6), and it is therefore almost impossible to prevent signal reflections from occurring. The only way to overcome the potential problem they represent is to keep the signal path between the pulse generator and the oscilloscope input as short as possible, preferably no more than 60 to 70 mm in length for a 150 ps risetime pulse. In this way, reflections arrive 4
back at the oscilloscope input very close to the trace's initial overshoot and ringing, where the oscilloscope's aberration specifications are widest (see Figure 3b). As a result they represent a much smaller proportion of the overall accuracy specification, and test uncertainty ratios are maintained. The oscilloscope's input amplifier is now subjected to all the energy contained in the pulse edge during a single short burst, rather than receiving some of it in short bursts as reflections run up and down the cable. Figure 3a Long connecting cables result in reflections appearing where scope aberration specifications are tightest Figure 3b Short signal path results in reflections appearing where scope aberration specifications are widest, maintaining test uncertainty ratios. 5
The source and load mismatch at either end of connecting cables can also cause problems when delivering leveled sinewaves to an oscilloscope's inputs. One metre of coaxial cable between a sinewave generator and a scope input with reactive mismatch VSWRs of 1.2 at each end will result in the sinewave amplitude at the scope input changing by up to 0.9% every time the sinewave frequency hits a multiple of the cable's 1/4 wavelength. For this 1-metre cable, the sinewave amplitude would ripple up and down five times during a 0 to 500 MHz frequency scan (see Figure 4). When driving pulses or squarewaves into an oscilloscope input, connecting cables also cause an effect called 'dribble-up' a loss that increases the amount of time taken for the waveform to settle to its final amplitude. ERROR (%) 1.0 CABLE LENGTH = 1 METRE SOURCE OUTPUT VSWR = 1.2 SCOPE INPUT VSWR = 1.2 0.0-1.0 100 200 300 400 500 FREQUENCY (MHz) Figure 4 Amplitude changes due to cable length and VSWR Active Heads Ease Reflection Problems All the requirements of high-speed pulse generation and delivery can be met by the use of active pulse-forming heads that plug directly into the oscilloscope's input connectors. These heads can generate high-speed edges within a few millimetres of the scope inputs and deliver them via a carefully controlled transmission line. Unlike a connecting cable, this transmission path is fully protected against accidental damage. The active head principle is not new, and has been used extensively by Tektronix in commercial high-performance pulse generators and by NIST in its reference pulse systems. The 150 ps return-to-zero pulses generated by the active heads of Wavetek's Model 9500 Oscilloscope Calibrator are illustrated in Figure 5. Figure 6 illustrates how the accurate 50-Ω source impedance of these heads absorbs reflected signals caused by an oscilloscope's input mismatch (the input mismatch used in this test was the same as that for Figures 3a and 3b). 6
3-volt falling edge (t f = 141 ps) Aberrations at 2%/div Figure 5 The 150ps return-to-zero pulses generated by the active heads of Wavetek's Model 9500 oscilloscope calibrator (measured over a 22 GHz bandwidth using a HP54750) 7
Figure 6 The effectiveness of the accurate 50-Ω source impedance of Model 9500 active heads in absorbing reflections from an oscilloscope input mismatch (cf. Figures 3a and 3b) If high-bandwidth amplitude sensing circuitry is also built into the head, the amplitude of leveled sinewave outputs can be accurately controlled very close to the oscilloscope input, reducing the amplitude ripple that occurs with interconnecting cables due to source and load mismatch. The amplitude of DC and squarewave outputs can also be 4-wire sensed at the oscilloscope input. The use of an active head has further advantages. For example, if the attenuators required to generate low-level calibration signals are incorporated into the head, the head itself can be driven with high-level signals. This dramatically reduces the effect of any interference (for example, common-mode noise, microphony or RFI) that is picked up by mainframe to head cabling. Active heads, especially when more than one can be driven at the same time, are also the key to fully automated oscilloscope calibration. Activation of individual heads to drive the oscilloscope with all the necessary calibration waveforms and trigger signals can be performed by the calibrator, eliminating the need for an external multiplexer. In addition, active heads can also include 50-Ω terminators that are switched in whenever the active heads need to drive leveled sinewaves or pulse waveforms into an oscilloscope that has high impedance inputs. The ability of Wavetek's Model 9500 Oscilloscope Calibrator to measure an oscilloscope's input resistance and capacitance means that high impedance inputs can be detected automatically. Heads Ease Support Problems The use of active heads can also improve the serviceability of an oscilloscope calibration system when it becomes necessary to recalibrate the calibrator against higher order metrology standards. 8
Traceably calibrating 1.1 GHz leveled sinewaves or 150 picosecond risetime edges requires sophisticated equipment which is beyond the reach of most users. Calibrators that generate these waveforms entirely within their mainframe unit will therefore need to be returned to a suitably equipped calibration laboratory for recalibration, resulting in considerable system downtime. However, for calibrators that generate these signals within active heads, only the heads themselves will require sending away for recalibration. DC and low-frequency functions contained in the mainframe unit can be calibrated to sufficient accuracy on-site, often only requiring the use of a long-scale length DMM. If a spare head is substituted for the one that is away being calibrated, system downtime can be eliminated. In Wavetek's Model 9500 Oscilloscope Calibrator, each active head carries with it its own calibration data. This data is interrogated by the Model 9500 each time a new active head is plugged into it so that the head's calibration uncertainties can automatically be included into the overall uncertainty equation. Conclusion By choosing the right equipment, fully automated oscilloscope calibration is not only achievable, it can also be made simple. The availability of fully integrated calibrators that generate all the required waveforms simplifies system design and reduces support costs. Accurate delivery of these waveforms to the oscilloscope's input connectors via active heads reduces systematic errors and conforms to best-practice metrology techniques. References Bargroff, Keith P, 'New Techniques in Automated, High Performance, Oscilloscope Calibration', Cal Lab, Volume 3, No. 3, May/June 1996, p17 to p22. Roddis, Richard, 'Innovative Oscilloscope Calibrator Architecture', NCSL Proceedings, 1996, p467 to p476. Crisp, Peter, 'Significant Advances in the Provision of Traceability to Oscilloscope Calibrators', NCSL Proceedings, 1996, p391 to p395. Johnson, Mitch, 'Simple Tests for Evaluating Digital Oscilloscopes', Cal Lab, Volume 4, No. 2, March/April 1997, p28 to p32. 9