Configure your real-time oscilloscope for optimum jitter analysis, in 8 easy steps.
Real-time oscilloscopes must be configured properly to make accurate jitter measurements. We present here a step-by-step procedure to manually setup your instrument for measuring jitter of all types, which you can apply to any real-time oscilloscope, regardless of manufacturer. Although you can purchase specialized jitter-analysis software from the manufacturer to auto-configure your instrument using a one-button or wizard-type approach, the software doesn't always produce the optimum configuration. Auto-configured setups should therefore be verified using the same procedure (below). To configure your instrument properly, perform the following steps in order.
Turn on the oscilloscope and restore the factory default setting. Then adjust the following items, and save the measurement configuration so you can easily recall it in the future.
High-speed oscilloscopes typically employ analog-to-digital converters (ADCs) with 8 bits (256 levels) of quantization. The voltage reported by the ADC equals the true signal voltage plus a quantization error. This error is essentially a rounding error, so to minimize it, we just need to reduce the voltage range captured by each quantization level. We do this by reducing the vertical resolution, or volts-per-division, setting. The goal is to use the full range of the ADC. For most oscilloscopes, this means adjusting the signal's waveform until it just fills the vertical height of the display. Some oscilloscopes, however, are designed to overfill the display slightly (contact the manufacturer to learn more). Just make sure not to saturate the ADC, since this will destroy the integrity of the waveform.
Figure 1 shows how the jitter measurements improve by simply reducing the vertical resolution from (a) 100 mV/div to (b) 54 mV/div for an example 36 MHz clock signal, and 3 types of jitter: time-interval error (TIE), period jitter, and cycle-to-cycle (C2C) jitter, reported in units of seconds peak-to-peak (pp) and root-mean square (RMS). For reference, Figure 1(a) shows the auto-scaled signal, which, incidentally, should never be used to measure jitter.
Figure 1. Illustration of jitter measurements in a 36 MHz clock signal using (a) auto-scaled settings, followed by (b) optimizing the vertical resolution, (c) optimizing the system bandwidth, then observing the effect of (d) increasing the record length.
In theory, the sampling rate must be at least two times the highest analog frequency present in the signal to avoid aliasing. In practice, the acquisition process requires oscilloscopes to sample at 2.5 to 3 times this frequency. A conservative rule of thumb is to set the sampling rate so that each edge is sampled at least 5 times. More is always better to minimize interpolation error when computing jitter. The downside to higher sampling rates is a smaller population of jitter measurements, (unless the memory depth can be increased). If the edge cannot be sampled at least 5 times using the highest sampling rate provided by the oscilloscope, sinc interpolation may be enabled to provide additional data points, at the expense of processing time.
If the oscilloscope bandwidth is set too low, the signal's edge rates will appear slow, causing amplitude noise to more efficiently convert into timing noise via amplitude modulation (AM) to phase modulation (PM) conversion. Alternatively, if the oscilloscope bandwidth is set too high, additional instrument thermal and shot noise will unnecessarily increase the instrument's jitter noise floor, which can introduce measurement error.
A common rule of thumb for measuring NRZ data is to set the bandwidth of the oscilloscope (plus probes, if used) to at least 1.8, and more preferably 2.8, times the bit rate. When measuring clock signals with analog output-voltage levels, set the bandwidth to capture at least the 5th harmonic. Clock signals with digital levels have significant spectral energy at much higher harmonics, and a bandwidth of 20 times the fundamental frequency is recommended.
You can set the optimum bandwidth in just a few seconds by measuring the rise time at the highest bandwidth, then lowering the bandwidth until just before the rise time changes more than 5% from its highest-bandwidth value. Figure 2 illustrates such an experiment for an oscilloscope having a maximum analog bandwidth of 12 GHz. The y-axis is the rise (and fall) time normalized to the value at 12 GHz and expressed as a percentage. The optimum bandwidth is observed to be 1 GHz. Using a higher bandwidth would raise the instrument's jitter noise floor; using a lower bandwidth would slow the measured edges down and increase jitter from AM to PM conversion. Figure 1(c) shows how the jitter values improve by reducing the acquisition bandwidth from 12 GHz to 1 GHz, which results in a lower instrument noise floor.
Figure 2. This plot shows how a 36 MHz clock (shown in Figure 1) rise and fall times change as the oscilloscope bandwidth deviates from 12 GHz. The rise and fall times are fairly constant moving from 12 GHz to a 1 GHz bandwidth, then increase rapidly for lower bandwidths. The optimum oscilloscope bandwidth for this device is therefore 1 GHz.
The threshold voltage is the vertical level used by the oscilloscope to determine where to measure jitter. Ideally this level is set to emulate the level used by receiver circuitry in an end application. The threshold voltage is the voltage level that causes the decision-threshold circuit in a receiver to change state when the input signal crosses it. For example, the threshold voltage for a differential signal is 0 V. The oscilloscope uses the closest sampled points on either side of this threshold to interpolate a crossing point at the threshold voltage, which is then used to measure jitter.
Set the threshold voltage as an absolute voltage, rather than as a percentage of the voltage swing. Figure 3 illustrates why. If the waveform (a) is amplitude modulated, (b) does not settle at logic high (or logic low), or (c) contains ringing or other artifacts, the 50% level (red markers in Figure 3) of the amplitude swing can vary or be offset from the level that a reference receiver (gray lines in Figure 3) would observe in a system.
Figure 3. For accurate jitter measurements, set the threshold voltage at an absolute level (gray line), rather than at 50% of the swing (red markers).
You will also need to set a hysteresis voltage (sometimes specified as upper and lower voltage thresholds) to prevent detecting false edges, which can occur if noise in the signal causes the threshold voltage to be crossed multiple times per edge. Set the hysteresis voltage slightly larger than the maximum voltage spike expected in the signal. You can estimate this voltage with an oscilloscope measurement. Simply setup the scope according to all the steps in this procedure (including all steps above and below), then either turn off the DUT's power or disconnect the DUT from the oscilloscope. Capture a waveform, then measure the maximum peak-peak voltage over the entire waveform. Add a little margin to this value, and use it to compute a hysteresis value that you can enter into the oscilloscope (e.g. either peak, or peak-peak, as appropriate for the particular oscilloscope you're using). Typically the default hysteresis setting is sufficient, unless the signal is very noisy.
Set the type of jitter to measure (TIE, period jitter, C2C jitter, etc.), along with the edges of interest (e.g. rising edges only, falling edges only, or all edges).
You can optionally apply software filters to the measured jitter values to model a system's response to a signal passing through it. The goal of the filter is to extract only the jitter that would be observed by the real system. For example, TIE is always filtered as required by high-speed serial standards. When applicable, set the filter characteristics according to an industry standard or system requirement.
Note that the oscilloscope itself acts as a brick-wall bandpass jitter filter. The upper (low-pass) corner frequency is set by the oscilloscope bandwidth. The lower (high-pass) corner frequency equals 1 divided by the acquisition time. In other words, the lower corner frequency equals the sample rate divided by the record length, where the record length is the number of samples acquired.
The lower corner frequency deserves special attention since it can greatly effect the measured jitter values. Suppose we acquire a jitter-free signal as shown by the blue curve in the bottom of Figure 4. Let's add phase modulation (i.e. jitter) to this signal. If all of the data acquired by the oscilloscope is displayed within 10 units of relative time (as shown in the bottom of Figure 4), then the lowest phase-modulation frequency ωn that can completely fit in this timeframe is 1 divided by 10 units of relative time. The red curves in Figure 4 show this noise frequency (top) and its effect on the signal (bottom). When the noise amplitude is positive, the phase-modulated signal (red waveform) leads the unmodulated signal (blue waveform), and when negative, it lags.
If we then cut the acquisition window in half, by only acquiring data up to a maximum of 5 units of relative time, then we would only observe half of the phase modulation's effect on our acquired signal. The point is, increasing the length of time that we observe a signal enables our measurement to observe lower-frequency noise, which can increase the jitter we measure when noise exists there.
Figure 4. Adding phase-modulation (top curve) to a jitter-free signal (bottom blue curve) creates a jittered signal (bottom red curve). To observe one complete cycle of modulation on the jittered signal, the oscilloscope's memory depth needs be large enough to capture 10 units of relative time in this example. If the waveform is acquired with 5 units of relative time, then only half of the modulation would be observed in the jittered signal.
Continuing the earlier measurements, Figure 1(d) shows how increasing the record length (i.e. memory depth) can increase the measured TIE values when lower-frequency noise is present in the signal or test environment. Notice also that period and C2C jitter remain constant with memory depth. This is because the definition of TIE jitter is able to detect low-frequency noise, whereas the definition of period and C2C jitter essentially filter out this low-frequency noise. Another consideration is that longer acquisitions of data increase the population of jitter data, which can statistically lead to higher peak-peak values (even though we didn't observe this in the Figure 1 data).
For TIE, the minimum required memory depth is the depth required to capture the lowest noise-frequency of interest to your application. For example, if the standard you are working with requires TIE frequencies to be analyzed between 10 kHz and 20 MHz, and your oscilloscope requires 40 GSps to capture at least 5 samples per edge, then the minimum required memory depth is 40 GSps × 10 kHz = 4 Mpts of data.
For period or C2C jitter, start with a small memory depth, then increase it until you see the value of jitter stay constant. To add a little margin, use a minimum memory depth slightly higher than this value. For N-cycle jitter, the minimum required memory depth is the depth needed to capture N continuous cycles.
Regardless of the type of jitter being measured, using the minimum required memory depth won't produce a large enough population to quantify jitter. The exact population depends on the application, but 1E+4 measurements is a good start for clock jitter (a lot more is needed for measuring jitter in data signals; refer to your high-speed data standard's documentation). To increase the population of jitter measurements, increase the memory depth much higher than the required minimum value, or enable the measurement statistics to be accumulated over multiple acquisitions of data, or both.
Table 1 Revision History