What are the factors that cause inaccuracy when measuring with an oscilloscope and not meeting the measurement range?

In English, we often use “comparing apples to apples” to indicate that we are making a fair comparison between similar things. On the other hand, if we say “comparing apples to oranges,” we mean the opposite.

Sometimes, despite our best efforts, our lab measurements can have an “apples to bruised apples” effect. In this month’s post, I’ll provide a relatively common example of this in “The Case of the Discrepant Scope Measurements.”

Initial Case

A few years ago I was supporting an oscillator customer and we had a problem with his measurements compared to mine. His period jitter measurements did not make sense when compared to the specs and the typical performance of the part.

I knew we were going to use similar boards, cables, terminations, test equipment, and attempt at least the same measurements. We even had very similar oscilloscopes. When I asked him to send me a waveform plot, I began to wonder if there was something wrong with his equipment. Bingo! His waveform was a fraction of mine, and I knew that was most likely the source of the difference.

Example Laboratory Measurements

To illustrate this, I took a standard 2.5V LVDS 100 MHz output clock and measured it using a high quality 1 GHz DSO or digital storage oscilloscope that supports measurement statistics (a higher bandwidth scope 1GHz is sufficient.) I took each measurement about 10K times, varying only the vertical scale, and tabulated the key results below. Note that I recorded the measurements below at a reduced vertical scale. The smaller the vertical scale, the larger the waveform is displayed.

The most noteworthy observation is that the standard deviation of all measurements decreases with vertical scale. The smaller the vertical scale, the larger the waveform displayed and the more accurate the measurement.

Their noise, my signal

I am particularly interested in period jitter, which is the standard deviation of the period measurement. If we plot period jitter versus vertical scale from the tabular values, we get the following linear graph.

The measured period jitter appears to be linear in scale with the samples. Also, there is no deflection at smaller values, which suggests that we are not close to the measurement "floor".

Before discussing the reasons for the measurement results, let's briefly review the upper limits of the screen for each scale selection and why we use them.

250mV/div option

This is the smallest displayed waveform option of the three instances. You will encounter this type of scaling when trying to measure multiple waveforms simultaneously, such as when comparing relative clock skew. The standard deviation of the period measurement, or period jitter, is 7.5 ps.

100mV/div option

This is the most common choice, as it is the default or automatic scale choice for a particular oscilloscope when measuring a single waveform. The period jitter dropped by more than half to 3.2ps. Great for browsing, but still, it turns out, not the best choice.

55mV/div option

The final selection maximized the waveform displayed on the oscilloscope without clipping. The period jitter dropped to about 2 ps.

Explanation of different range measurements

I intentionally did not make any sampling or time scale changes in these measurements in order to eliminate that aspect as a variable. The only change is that vertical scaling affects voltage noise. Voltage noise is converted to timing noise through the slew rate, which is the ratio of delta voltage to delta time, as shown in the figure below. The blue Gaussian curve is intended to represent a normal noise distribution about the decision threshold.

A DSO is a digital oscilloscope with a sampling ADC at the front end of the signal chain. Therefore, the voltage noise will be due to the DUT and the DSO's ADC quantization noise. This particular oscilloscope's ADC is nominally 8 bits. (The actual ENOB, or effective number of bits, may be lower, but the principle is the same.)

An 8-bit ADC has 256 unique codes or quantization levels. Therefore, we can compare the nominal quantization levels as follows. As we reduce the vertical scaling, the quantization noise decreases significantly.

The ADC's 8 bits are used for the entire vertical display of the oscilloscope, so anything less will use fewer bits and result in more quantization noise. This in turn creates more timing uncertainty. It is for this reason that oscilloscope manufacturers often recommend that users maximize the waveform on the screen to make the most accurate measurements. Doing so is the most conservative approach.