Evaluating the vertical noise characteristics of an oscilloscope

Introduction
　　All oscilloscopes have an unwelcome characteristic: vertical noise present in the oscilloscope's front end and digitizing process. Measuring system noise will degrade your actual signal measurements, especially when measuring low-level signals and noise. Since oscilloscopes are broadband measurement instruments, the wider the oscilloscope bandwidth, the higher the vertical noise in most cases. Although engineers understand the vertical noise characteristics of an oscilloscope when purchasing an oscilloscope, these characteristics should be carefully evaluated because they can affect signal integrity in several ways. Vertical noise can:
　　1. Introduce amplitude measurement errors
　　2. Introduce sin(x)/x waveform reconstruction uncertainty
　　3. Introduce timing errors (jitter) as a function of the slew rate of the input signal edges
　　4. Cause observable undesirable "fat" waveforms

　　Unfortunately, not all oscilloscope manufacturers provide a vertical noise specification/characteristic in their data sheets. Even when this specification is available, it is often misleading and incomplete. This article compares the vertical noise characteristics of oscilloscopes with bandwidths ranging from 500MHz to 1GHz manufactured by Agilent, Tektronix, and LeCroy. It also describes useful techniques for making more accurate noise and interference measurements on low-level signals in the presence of relatively high levels of measurement system noise (oscilloscope noise).

What is Noise and How Should It Be Measured?
　　Random noise, sometimes called white noise, is theoretically unbounded and follows a Gaussian distribution. Unbounded means that the more data you collect in a noise characterization measurement, the higher the peak-to-peak excursion you will get due to the inherent randomness of the noise. For this reason, random phenomena such as vertical noise and random jitter should be defined and measured using rms values ​​(standard deviation). Table 1 shows the rms noise floor measurements of four competing 500MHz bandwidth oscilloscopes. Each oscilloscope was terminated with 50Ω and set to acquire waveforms using the highest sample rate specified for each oscilloscope with no signal connected.
　　See also Appendix A for the RMS noise floor measurements of competing 1GHz bandwidth oscilloscopes.
　　It is common to think of an oscilloscope’s “baseline noise floor” as the noise level when the oscilloscope is set to its most sensitive setting (lowest V/div). However, many oscilloscopes on the market today have reduced bandwidth characteristics at their most sensitive V/div setting. As mentioned earlier, an oscilloscope is a broadband instrument, and higher bandwidths generally have higher noise floors. So when you compare the baseline noise floor characteristics of each oscilloscope at its most sensitive V/div setting, you are likely comparing a lower bandwidth oscilloscope to a higher bandwidth oscilloscope, which is not an apples-to-apples comparison. The baseline noise floor of the same bandwidth should be compared at the most sensitive V/div setting of each oscilloscope, where the full bandwidth is available.
　　Many oscilloscope evaluators make the mistake of testing only the baseline noise floor characteristics at the scope’s most sensitive setting and assuming that this noise level applies to all V/div settings. There are actually two noise components inherent in an oscilloscope. The first is a fixed noise level that is primarily contributed by the oscilloscope’s front-end attenuators and amplifiers. The baseline noise floor at the scope’s most sensitive V/div setting is a good approximation of this noise component. This noise component dominates at the most sensitive setting, but is negligible when the scope is used at less sensitive settings (higher V/div). The second noise component is the
　　relative noise level based on the scope’s dynamic range, which is determined by a specific V/div setting. This noise is negligible when the scope is at its most sensitive setting, and it primarily affects less sensitive settings. Although the waveform does not appear to be very noisy at the scope’s high V/div settings, the actual noise level can be quite high, as you can see by comparing the noise levels measured at 1V/div and 10mV/div in Table 1. This relative RMS noise contribution for the Agilent MSO6054A is approximately 2% of the V/div setting. The relative RMS noise contribution for 500MHz bandwidth oscilloscopes from Tektronix and LeCroy is 3% - 4% of the range.
　　Once you have determined the fixed noise contribution (approximately the baseline noise floor) and the relative noise contribution, you can use the square root of the sum of squares formula to estimate the amount of noise at the intermediate V/div settings. From the noise measurements in Table 1, you can see that the Agilent MSO6054A has the lowest overall noise performance at most V/div settings.

Measuring Peak-to-Peak Noise
　　Although using RMS gives the best results for evaluating and comparing noise, people often want to measure and compare peak-to-peak noise. After all, peak-to-peak excursion is what you see on the oscilloscope screen, and it causes the largest amplitude error in real-time/non-averaged measurements. For this reason, many oscilloscope users prefer to compare and measure peak-to-peak noise. Since random vertical noise is theoretically unbounded, you must first establish a criterion for how much data to collect and then use that criterion to obtain a peak-to-peak noise measurement. Table 2 shows peak-to-peak noise measurements for four 500MHz oscilloscopes collecting 1M points of digitized data. See also Appendix B for peak-to-peak noise measurements of a competitively priced 1GHz bandwidth oscilloscope.
　　Note that because the TDS3054B (10k points) has limited memory depth, making peak-to-peak noise characterization measurements for 1M acquisition points is a very difficult task. To obtain a total acquisition of 1M points at each V/div setting, the instrument accumulates approximately 100 acquisitions with infinite persistence. The other oscilloscopes tested have deeper acquisition memories and can collect 1M data points in a single acquisition.
　　Since a particular 1M data point acquisition (a group of acquisitions for the TDS3054B) can produce peak-to-peak measurements that are either higher or lower, we repeated the 1M point peak-to-peak noise measurement 10 times for each V/div setting. The measurements were then averaged to give a “typical” peak-to-peak noise figure for an acquisition of 1M data points. As this table shows, the Agilent 6000 Series oscilloscope has the lowest overall peak-to-peak noise level (based on 1M data points) at the full bandwidth V/div setting. The 500MHz bandwidth oscilloscopes from Tektronix and LeCroy have much higher peak-to-peak noise levels at most settings.
　　While it is tempting to set the various oscilloscopes to the same time/div and then use infinite persistence mode to collect data for the set amount of time, such as 10 seconds, you should be aware that the peak-to-peak noise test does not use this more intuitive approach. Not only are the memory depths significantly different, but the update rates are also significantly different. For example, if you start with default setup conditions and then set the Tektronix TDS5054B and Agilent MSO6054A to 20ns/div, the Tektronix oscilloscope will acquire and update waveforms at a rate of approximately 30 waveforms/second. Because the Agilent 6000 Series with MegaZoom III technology has an extremely fast waveform update rate, it will update waveforms at a rate of approximately 100,000 waveforms/second. This means that if you collect 10 seconds of infinite persistence waveforms, the Agilent oscilloscope will collect approximately 3000 times more peak-to-peak noise measurement data. As mentioned earlier, due to the random and Gaussian nature of random vertical noise, peak-to-peak noise will increase as more data is collected.

Measuring Noise with Probes
Most oscilloscopes come with 10:1 passive probes that provide 600MHz system bandwidth (for 600MHz or higher oscilloscopes). Higher bandwidth oscilloscopes may also achieve higher bandwidths with active probes. Whether you use a passive or active probe, the probe itself will add additional random noise components. Today’s digital oscilloscopes can automatically detect the probe’s attenuation factor and readjust the oscilloscope’s V/div setting to reflect the signal attenuation introduced by the probe. So if you are using a 10:1 probe, the V/div setting indicated by the oscilloscope will be 10 times the actual setting inside the oscilloscope. That means if the oscilloscope is set to 20mV/div with a 10:1 probe connected, the actual setting of the input attenuator and amplifier in the oscilloscope will be 2mV/div. This means that because the baseline noise floor is amplified 10 times, you will observe a higher noise level relative to the height of the screen. If you are making critical low-level signal measurements, such as measuring power supply ripple, you should consider using a 1:1 passive probe. Also, if the oscilloscope bandwidth is limited at the more sensitive V/div ranges, it is

important to understand the attenuation factor for a particular probe, since this bandwidth limitation may also apply to higher V/div settings.
　　When you use an oscilloscope set to its most sensitive V/div setting, the oscilloscope's inherent random noise can obscure actual signal measurements. However, you can use certain measurement techniques to minimize the effects of oscilloscope noise. When you are measuring power supply ripple and noise levels, you will probably want to use the most sensitive ranges. First, try using a 1:1 probe as described earlier, rather than the standard 10:1 passive probe that comes with the instrument. Second, if you are measuring the rms noise of the power supply, the measurement results also include the noise contribution of the oscilloscope and probe system, which can be quite high. However, by carefully characterizing both the signal (power) and the measurement system, you can subtract the measurement system noise component and get a more accurate estimate of the actual power supply noise (rms).
　　Using an Agilent 6000 Series oscilloscope with a DC offset of approximately 4.7V, Figure 1 shows a power supply noise measurement made with a 1:1 passive probe at a 10 mV/div setting. Note that the documentation for 500MHz and 1GHz Tektronix and LeCroy oscilloscopes states that the input signal offset cannot be greater than ±1V when connected to a 1:1 passive probe and settings below 50mV/div. This means that when making noise measurements on a 5V supply with a Tektronix or LeCroy oscilloscope, you must use AC coupling due to the oscilloscope's DC bias limitations. However, if you must use AC coupling due to the oscilloscope's DC bias limitations, the DC component of the supply will be removed from the result, and you will not be able to make accurate measurements.
Using an Agilent oscilloscope with a 1:1 passive probe, we measured a noise of about 1.5mV RMS on a noisy 5V supply. Figure 2 shows a noise characterization of the measurement system noise using the same 1:1 passive probe. With the probe ground lead connected directly to the probe tip, the system noise measured at the 10mV/div setting is approximately 480V RMS. This oscilloscope/probe noise measurement is higher than the noise figure shown in Table 1 (250 V RMS) because of the additional noise component added by the 1:1 probe used. In addition, we used a 1MΩ input termination instead of the original 50Ω termination (used for the baseline RMS noise measurement in Table 1). Now using the root sum of squares formula to subtract this measurement system noise component, the noise of this power supply is approximately 1.4 mV RMS. [page] While
　　this particular power supply measurement may include deterministic/systematic interference/noise components in addition to the random components, if the deterministic components are not related to the oscilloscope’s auto-triggering, this technique can be used to subtract the measurement system error components and obtain a very close approximation of the total RMS noise of the power supply.
　　Individual deterministic/systematic components of interference, such as switching power supplies or digital system clock interference, can also be accurately measured in the presence of high random measurement system noise. You can trigger on the suspected interferer on a separate channel of the oscilloscope, repeatedly acquire the input signal, and average out all random and uncorrelated noise and interference components contributed by the oscilloscope/probe and input signal. The result will be a high-resolution measurement of the specific interference component of the power supply, even if you set the oscilloscope to a very sensitive V/div setting, such as 2 mV/div as shown in Figure 3. In addition, accurate measurement of the average DC component of the power supply requires the oscilloscope to have sufficient DC offset range (which only Agilent oscilloscopes can provide). Using this averaging measurement technique on the same noisy power supply signal, we measured that the system 10MHz clock (green trace below) introduces approximately 4.9mVp-p interference. To find all deterministic (non-random) interference and ripple, you need to make multiple average measurements with various suspected interferers as the trigger source of the oscilloscope.



Observing “Fat” Waveforms
　　Some oscilloscope users believe that digital storage oscilloscopes (DSOs) have higher levels of random vertical noise than older analog oscilloscopes. This conclusion is based on the fact that the traces on a DSO are generally wider than those on an analog oscilloscope. However, the actual noise level of a DSO is no higher than that of an analog oscilloscope. With analog oscilloscope technology, the extreme values ​​of random vertical noise are either very dim or not visible at all because the signal extreme values ​​occur so rarely. Although engineers generally think of an oscilloscope as a two-dimensional instrument that displays voltage versus time, there is a third dimension because analog oscilloscopes use scanning electron beam technology. The third dimension uses trace intensity modulation to display the frequency of signal occurrence, which means that analog oscilloscopes actually hide, or visually suppress, the extreme values ​​of random vertical noise.
　　Traditional digital oscilloscopes lack the ability to display the third dimension (intensity modulation). However, some of today’s newer digital oscilloscopes have an intensity gradation capability that more closely approximates the display quality of older analog oscilloscopes. Agilent’s newest 6000 Series oscilloscopes with MegaZoom III technology feature the highest intensity grading in the oscilloscope industry, mapping 256 levels of intensity to the XGA display. Figure 4 shows a low-level 10MHz signal captured at 100% intensity at a 10mV/div setting. This screen represents an older digital oscilloscope display without intensity grading capability. Without intensity grading, the oscilloscope display shows a “fat” waveform with extreme peak-to-peak noise. But the “thickness” of the relatively low input signal (about 50mVp-p) measured at the 10mV/div setting is primarily due to inherent oscilloscope noise — not input signal noise. Figure 5 shows the same 10MHz signal, but now with the intensity adjusted to 20% to better mimic an analog oscilloscope display that naturally rejects extreme noise. We can now observe a more “clean” waveform without the influence of the oscilloscope’s inherent noise at a relatively sensitive V/div setting. In addition, we can now see waveform details, such as the "wiggle" on the positive peak of the sine wave, which was previously obscured by the relatively high oscilloscope noise level when viewed at constant intensity (100%).
　　For a more detailed discussion of the benefits of oscilloscope display quality, download Agilent Application Note 1552, "The Impact of Oscilloscope Display Quality on the Ability to Find Signal Anomalies."
　　If you are acquiring a repetitive input signal, you can instead remove random signal noise from your measurement system by averaging the waveform, as shown in the example in Figure 3. For real-time/single-shot applications (where repetitive averaging cannot be used), some oscilloscopes offer a high-resolution acquisition mode. With this technique, you can filter out high-frequency noise and interference components from a single-shot acquisition through DSP/digital filtering, increasing the vertical resolution to 12 bits, at the expense of the bandwidth of the measurement system.






Conclusion
　　When evaluating various oscilloscopes for purchase, be sure to carefully consider the inherent noise characteristics of the oscilloscope. Not all oscilloscopes are created equal in this regard. The vertical random noise of an oscilloscope not only degrades measurement accuracy, it can also affect the quality of viewing digital signals. When you evaluate the noise characteristics of an oscilloscope, it is important to carefully set up the oscilloscope under test under the same measurement criteria, which include the same bandwidth oscilloscope, the same V/div setting (with full bandwidth), the same sampling rate, the same memory depth, and the same number of acquisitions.
　　As shown in this article, the Agilent DSO/MSO 6000 Series and 54830 Series Infiniium oscilloscopes have the lowest overall noise characteristics compared to other 500MHz - 1GHz oscilloscopes in the industry. In addition, the Agilent 6000 Series oscilloscopes with MegaZoom III technology provide the highest resolution display quality with 256 levels of brightness, which can be used to observe the random extremes of the suppressed oscilloscope's inherent noise.
　　You can use a variety of measurement techniques, such as math, waveform averaging, DSP filtering, and display intensity grading, to minimize or even eliminate measurement system noise components, thereby more accurately measuring low-level random and deterministic noise components in your system.
　　Although this article focuses on noise measurement comparisons between 500MHz and 1GHz bandwidth oscilloscopes, the principles apply to any bandwidth oscilloscope—higher or lower. In fact, Agilent’s higher bandwidth 12GHz DSO81204A oscilloscope has the lowest inherent internal measurement system noise of any oscilloscope in this bandwidth range, with noise levels not much higher than 1GHz oscilloscopes currently on the market. Agilent has also been able to achieve lower measurement noise performance levels due to the higher levels of integration achieved using low-power integrated circuit (IC) technology.
　　It should also be noted that only a very limited number of samples were taken in the evaluation of random vertical oscilloscope noise. All oscilloscopes selected for testing were current products from their respective manufacturers. We tested only channel 1 because it is the channel most commonly used by engineers. Although we cannot guarantee that the measurements described here are typical, we believe that these results are representative of current products from various oscilloscope manufacturers.

Glossary
Baseline Noise Floor: The RMS noise level measured at the oscilloscope’s most sensitive V/div setting 
Sin(x)/x: Reconstruction A software filtering characteristic that reconstructs the sample waveform with a higher data resolution, more accurately representing the original unsampled input waveform that meets the Nyquist criterion Noise
Floor: The RMS noise level measured at each V/div setting of the oscilloscope
Random Noise: Unbounded noise that follows a Gaussian distribution Dynamic
Range: The full-scale range of the analog-to-digital converter of a digital storage oscilloscope (DSO). It depends on the V/div setting of the oscilloscope and typically varies from 8 divisions peak-to-peak (full screen) to 10 divisions peak-to-peak (full screen + 20%) in most oscilloscopes Peak-to-
Peak Noise: The peak-to-peak noise of an oscilloscope based on certain criteria such as time, number of acquisitions, and/or acquisition memory depth
RMS Noise: Random noise measured as a standard deviation
Infinite Persistence: A common display mode of a digital storage oscilloscope (DSO) that accumulates and displays All acquisitions to show the worst-case deviation of the signal
Gaussian distribution: a typical bell-shaped statistical distribution
Deterministic: a systematic error/noise source that is bounded
Trace intensity modulation/grading: the oscilloscope display intensity at a specific time position varies with frequency
DSP: digital signal processing
MegaZoom III: an Agilent patented technology that provides trace intensity grading, fast waveform update rates, and responsive deep memory