Sub-nanosecond risetime pulse generators with high repetition rates are necessary for gauging an oscilloscope's step response, time-domain reflectometry (TDR), and differential TDR. They're also effective for measuring the reverse recovery of diodes as well as the response and propagation delay of cables, transistors, amplifiers, and comparators. Time-domain studies like these generally require bipolar pulses, with a separate low-jitter pulse capable of triggering an oscilloscope or other test equipment.
This circuit produces such a trigger, along with balanced ECL outputs, at a pulse-repetition frequency of 1 MHz (Fig. 1). The AD8611 high-speed comparator is wired as a square-wave oscillator. Using the 1-kΩ potentiometer, positive feedback via a 1-MHz crystal is adjusted to yield a 50% duty cycle. While the complementary output provides a fast step-pulse to trigger an oscilloscope, the other output triggers a MAX9691 high-speed comparator. This comparator delivers balanced, ECL-level transitions that are specified at a 500-ps risetime. All components, including three BNC-connectors at A, B, and E, were mounted on a small PC board (Radio Shack #276-150) employing surfboards for the SO-8 comparators.
If the RG-58A/U cable network is connected using the switch S, the double shorting stub produces a short pulse at C. A single, inverted, delayed step-pulse is created, which follows the incident step-pulse toward C. The 6-dB attenuated step is divided at the cross-joint. It first splits into three equal transmitted pulses and one equal-and-opposite pulse that's reflected back toward the attenuator. Each pulse carries one quarter of the power, or half the voltage of the step.
At each shorted end, the pulse reflects with a sign reversal and returns to the cross-joint. Here, it splits again into three transmitted pulses and one equal-and-opposite reflected pulse. This reflected pulse, however, is canceled exactly by the equal-and-opposite pulse transmitted across the cross-joint from the other stub. So the result at C is that the primary step-pulse, which is one-quarter of the comparator's voltage, is followed by an equal-and-opposite step. The relative delay caused by the stub's round-trip time defines the length of the short pulse observed at C.
Two BNC T-adapters (i.e., Radio Shack #278-112) make up the coaxial cross-joint. Using a bandsaw, a 90° "V" is cut into the back of one of the T-adapters at a 45° orientation. Doing so makes it possible to fit a fourth BNC port, cut similarly from the other adapter. This fourth port is soldered to the adapter's V-cut using a low-heat iron. Finally, the brass center-conductor is threaded into the cross-joint center-conductor. A short length of 1-72 UNF threaded rod, cut from a small machine screw, is used for the threading process. The completed cross-joint is then plugged into the four RG-58A/U cables (Fig. 1, again).
These outputs were measured on a 1-GHz-bandwidth, homebrew sampling oscilloscope with a Tek 503 X-Y oscilloscope for the display (Fig. 2). The high pulse-repetition frequency of 1 MHz allows a high scan rate of 50 Hz. As a result, a live, flicker-free display is produced. Each nanosecond of equivalent time displays 200 samples, corresponding to an interval of 5 ps/sample. The sampler was triggered by a terminated RG-58A/U cable using the attenuated AD8611 complementary output at E.
While Trace A depicts the MAX-9691's ECL-level output driven by coupling capacitors, Trace B shows the complementary output. Transition durations are about 500 ps, in agreement with the component specification sheet. Trace C illustrates the pulse after the shorting-stub network, connected by switch S.
For this 2-ns pulse, 20-cm long stubs were employed. But different stub lengths will yield other pulse durations, including sub-nanosecond pulses. Transition trace widths on this photograph are less than 0.10 of the 200-ps smallest divisions. This indicates that the system jitter is less than ±10 ps.
Due to frequency-dependent attenuation of the line, the 40-ns delay line commonly used in a sampling oscilloscope input limits the device's bandwidth. As an alternative, a 40-ns delay line may be installed between the two comparators, becoming part of the 390-Ω:50-Ω attenuator. The AD8611 will then provide a low-jitter trigger at E, which precedes the MAX9691's pulses by 40 ns. In turn, the delay line may be eliminated from the sequential-sampling oscilloscope, thereby improving its bandwidth.
For a general-purpose pulse generator, it may be beneficial to have pulses shorter than those shown. With TDR tests, however, it's better to have steps whose transition durations are commensurate with the risetime of the pulses implemented in the final system. Super-fine structure details are then automatically filtered out by the transition duration. Therefore, they will not obscure the TDR traces. A faster component, such as the Motorola MC100EL32D flip-flop, can be substituted for the MAX9691 in order to produce shorter pulses.