Ten Measurements
The following 10 measurements define signal integrity. Master them, and you will become a guru of the art. Work on one measurement every quarter or every year until you fully grasp the relationship among circuit theory, simulation, and measurement. All of these problems harbor subtle difficulties, as well as sparkling gems of insight. If you already know the answers, I hope this outline inspires you to teach others.
Step response. Drive a capacitor with a 5Ω step source having a rise time of 1 nsec. Look at the step-response waveform in the time domain. Assume a basic resistance/inductance/capacitance- series electrical-circuit model for the capacitor and extract the circuit parameters. Plot the curve of impedance magnitude versus frequency that your model predicts. Using a sine-wave source, or S-parameter test set, measure the actual impedance versus frequency and see whether it matches your expectations.
(see "Operating Above Resonance," Electronic Design Magazine, Apr. 14, 1997)
Characteristic Impedance. Separately measure the inductance and the capacitance of a 1-foot section of coaxial cable at 10 MHz. Estimate the characteristic impedance as the square root of the ratio of inductance to capacitance. Using a time-domain reflectometer, measure the input impedance of a longer section of the same type of cable. Determine whether the two values match.
(see "Characteristic Impedance of Lossy Line," EDN, Oct. 3, 2002)
Dispersion. Terminate a 100-foot-long RG-58 coaxial cable. Using a rise time of 1 nsec, measure its signal delay. Then quadruple the cable's length and observe whether the delay quadruples. Learn about signal dispersion and the skin effect
(see "Risetime of Lossy Transmission Line" EDN Oct. 2, 2003)
Reflections. String together two 10-foot sections of RG-58 cable. Terminate the endpoint. In the middle, connect a 100-pF capacitor from signal to shield. At the head end of the structure, using a 10-nsec rise time, observe the reflections that the reactive load in the middle generates. Now, imagine a 2-in. PCB trace with a 1-pF load, operating at a rise time of 100 psec. Do you expect the same behavior?
(see "Potholes," EDN Nov., 1999)
Ringing. Inject a 1V p-p signal from a 50Ω source into a 10-foot coaxial cable with no termination. Use a rise and fall time of 100 nsec. Determine the largest signal you can create at the endpoint. Now make a T configuration, with 10 feet leading to the central T and a pair of 10-foot sections branching off from that section. Add a 39-pF load at just one of the branch endpoints. What is the largest signal you can now make, and what does this signal tell you about the unbalanced T configuration?
(see "To Tee or Not to Tee?" EDN Feb. 22, 1998)
Crosstalk. Select two adjacent traces on a working PCB. Inject a step waveform from a 50Ω source into the first trace, making sure to terminate its far end. On the adjacent trace, compare the measured crosstalk at the trace's two endpoints, with both terminated. Is the crosstalk the same at the two ends?
(see "Directionality of Crosstalk," Electronic Design Magazine, Aug. 18, 1997)
Loading. On a working PCB, disconnect a driver from its load. Record the output waveform with no load, with 50Ω to VCC, and with 50Ω to ground. Do any of the loads change the driver's switching speed and, if so, why?
(see "DC Loading," July 18, 2008)
Jitter. Capture a waveform showing thousands of edges from a reference oscillator. Put the waveform into a math-processing spreadsheet, such as Mathcad, Matlab, or Mathematica, and make a list of the exact time of arrival of each rising edge, interpolating between samples. Make one vector showing the delay, from edge to edge, of successive pulses. Make another vector showing the delay from each edge to the edge 100 cycles later. Plot histograms of the two vectors and determine which has the largest standard deviation.
(see "Jitter Characterization" Oct. 8, 2008, also "Jitter Capture," Mar. 10, 2010)
SSO noise. On a large BGA device, find an output that is programmed to stay at logic zero throughout this experiment and route that output to your scope. Activate the rest of the device and observe the crosstalk at the signal under test. What can you do in the BGA device to maximize the crosstalk? Do all of the other pins induce the same level of crosstalk at your point of measurement?
(see "BGA Crosstalk," Mar. 1, 2005)
Power supply noise. On a working PCB, strip off one bypass capacitor and use its mounting pads to connect a small coaxial probe. Observe the power-system noise at that location. Next, start disconnecting bypass capacitors. What happens to the power-system noise as you remove the capacitors, and what aspects of the noise change? Are those changes the same at other locations?
(see "Healthy Power," EDN Mar 30, 2000)