Notes on jitter and timing measurements with new tek scope



Yüklə 12.82 Kb.
tarix07.11.2018
ölçüsü12.82 Kb.

Notes on jitter and timing measurements with new tek scope


Given a week or so to play with a rental TDS6154C 15GHz oscilloscope, a few SMA cables & connectors were ordered and then I did a couple measurements to see how well it works.

Timing sensitivity


The TDS6154C has a “fast rise pulse” output. Two short (6-inch) SMA-to-BNC adapter cables were used plus a BNC “tee” in the middle to connect the pulse output to channel 1 as shown in Figure 1.


Figure 1 - using the Tek scope's own fast output

The “tee” was used to provide a connection point for an impedance discontinuity later. Figure 2 shows the signal as measured, plus the FFT of the waveform as calculated by the scope. The “fast rise pulse” output has a fundamental frequency of ~1.5kHz and so the fundamental peak is buried at the left edge of the FFT, out of view. Unfortunately I had the cursors in the wrong place to make this obvious in the photo, but the 10% - 90% rise time is 0.5 divisions, or 250ps. The specs on the output pulse are for a 200ps rise time, so we can estimate the bandwidth of the cabling system & scope as being 0.35/50E-12, or about 7GHz. The cables are 12GHz parts, but we have the BNC bits, so that’s reasonable.


A second measurement was made by inserting a BNC “barrel” into the “tee” center connection. The open barrel creates an in-phase reflection. Assuming typical RG-58 propagation velocity (0.66c, or 7.8E12 in/sec) and a barrel conductor length of 1.5in, the expected round-trip delay is about 385ps. Figure 3 shows a round-trip delay of 365ps, so that says the scope is working well for direct timing resolution.


Figure 2 - signal rise time and spectrum for setup of Figure 1



Figure 3 - Reflection caused by addition of barrel to tee

Jitter Measurement

The TDS6154C software package contains demonstration versions of the jitter analysis software that can be run a small number of times before they are disabled. The software should be good for one or two more invocations before it is exhausted. Once invoked it may be left running for days without counting as another license checkout.


To measure the jitter I re-used the fast pulse output but ran the signal through a relatively long piece of RG-58 cable. This was mostly for the convenience of being able to connect the same signal to both this scope and another for comparison, but the cable also adds more capacitance, smoothing the rise and making measurements slightly more difficult. The setup is shown in Figure 4. The SMA-to-BNC cables and the BNC “tee” are the same as used previously. The jitter package generates the usual statistics, measuring cycle-to-cycle time, duty cycle, etc., and providing min/max/average calculations. A histogramming feature is provided which allows easy comparison to other scopes.



Figure 4 - Setup for jitter measurement
Two problems were immediately noted. The first is typical. After hooking up the signal and manually adjusting the channel gain and horizontal sweep rate to reasonable settings, I then hit the “autoset” button in the jitter package to let the scope find its own gain & sweep speed. Multiple attempts from various starting positions all failed miserably, generating error messages saying that the scope was unable to find a gain/sweep setting that it could use to measure statistics. While disappointing, this was not a surprise; few scope auto-set features ever seem to work right. After resetting the scope manually for gain, threshold point and sweep rate, the statistics were easily captured.

The second functional problem is one of software. The jitter controls suggest that one should take data and then ask for a plot, but this doesn’t work. Doing so will only histogram the last few data points. If you want to see the histogram of all data, the plotting function must be invoked first, then you have to clear the accumulated data, and then start taking new data. The histogram builds as more samples are collected. The help files are, as you might expect, mute on the subject.


Figure 5 shows the histogram of jitter as collected over about 400 samples. The scope’s own test pulse has a trimodal distribution, indicative of a normal period plus a sideband caused by some other modulation. The error is about ±3ns out of 1ms. The histogram has controls to zoom in and out, and one place cursors on it, although I was unable to manually set the horizontal scale of the histogram to integer numbers of ns per division. Perhaps with more time this could be figured out, but after a short operational period it looks like the cursors are the preferred method of measuring where the peaks are.


Figure 5 - Histogram to cycle-to-cycle width as measured by Tektronix scope
To compare the results with that of another scope, I used the same piece of RG-58 to connect the TDS6154C’s test output to a LeCroy Wavepro 960 we have at ANL. This setup is shown in Figure 6. Following the same sequence, the auto-set function of this device failed just as miserably as the more expensive scope’s did. Once more manually setting everything up, a similar histogram was developed. The LeCroy scope’s histogramming function is slightly more user-friendly than that of the Tek device, however it still fails to display the histogram with any arbitrary horizontal units. However, a larger failing is that the LeCroy scope can’t use the cursors on the calculated waveform.
Display issues aside, however, it can be seen in Figure 7 that the histogram of the Wavepro 960 is, qualitatively, not significantly different than that of the TDS6154C. The peaks are in the same places and the outliers are in the same relative positions. However, the peaks are broader, indicating that the sampling clock of the Wavepro960 isn’t quite as clean as that of the TDS6154C; the expectation is that the width of the peaks is a vector sum of the jitter of the input signal and the jitter of the sampling clock of the scope.



Figure 6 - Setup for comparison jitter measurement using LeCroy Wavepro 960


Figure 7 - Jitter histogram from Wavepro 960

Vertical accuracy and range

A couple quick tests with open and 50-ohm terminated inputs shows that the TDS6154C has quite a bit of internal noise that does not scale with input range. The noise appears to be artifacts from the input filtering algorithm that’s used along with the ADC. The manual indicates there is an internal DSP that can be turned on and off to reduce this to some degree, but it’s bad enough that I cannot recommend this scope for accurate low-level voltage measurements.


In similar vein the maximum input range for the channel inputs is 1V p-p, and warnings about damaging the inputs are plentiful. While the scope certainly has bandwidth and timing, it is by no means a general-purpose instrument good for everything.

Additional comments


It would have been interesting to see what the jitter measurement software did with a very clean clock input with a faster edge transition, but I didn’t have one close at hand to try. Obviously the slower edge intentionally created by the use of the RG-58 limited the jitter resolution (peak width in histograms) obtained by either scope. For entertainment I tried to obtain jitter measurements of the TDS6154C’s test output on a TDS3054B digital scope but couldn’t resolve the jitter to better than 10s of nanoseconds.

Conclusions



The TDS6154C is a nice scope for timing measurements, but it really doesn’t outclass other instruments unless there is a need for jitter measurements at scales markedly less than ±1ns, or a necessity to measure timing in the less than 500ps realm. The real power of the device is based in its very long record length. It’s worth a month’s rent for those occasional times that a clock signal must be very carefully characterized but the cost/benefit analysis does not favor outright purchase.


Dostları ilə paylaş:


Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©genderi.org 2017
rəhbərliyinə müraciət

    Ana səhifə