Worst is best in optical test

There are basically two approaches to testing. One is to create as perfect a stimulus as possible so any system measurements you make reflect only the system's behavior and not the stimulus. The other approach is to make the stimulus as bad as you expect it to ever be in the real world, then look to see if the system still behaves as it should.

Traditionally, optical communications system tests have taken the first approach. You work with as clean a signal as possible so you can understand how the system is operating. But the latest 10 Gigabit Ethernet specification, IEEE 802.3ae is now forcing optical test to adopt the second approach, at least in optical systems.

Instead of using ultra-pure signals for performance measurements, optical test is now using distorted signals like this "stressed-eye" to prove interoperability. Courtesy of Circadiant Systems.

The testing that the IEEE 802.3ae specification requires is called the stressed-eye (or dirty-eye) test. In this test, the receiver must continue functioning when the incoming signal suffers from four degradations:

• poor extinction ratio—the ratio of optical power levels for "1" and "0" can be as low as 3 dB;
• slow 0–1 transitions—creating a data-dependent jitter;
• vertical interference—amplitude modulation on top of the signal, creating a specified vertical eye closure profile (VECP); and
• additional time jitter—a frequency-based jitter error.

The effect of these degradations can be represented visually in an eye diagram, as shown in the figure . The stressed eye represents the expected worst case for 10 Gigabit Ethernet optical signals.

The approach is a departure from the method optical testing usually follows to ensure interoperability, according to John French, co-founder of instrument maker Circadiant Systems (Allentown, PA; http://www.circadiant.com). In optics, testing uses a "golden" standard to characterize the transmitter and receiver performance. If device performance falls within a specified range, it is considered good. The trouble is, French notes, this type of testing only works if the acceptable range is defined broadly enough to capture all possible behaviors that could cause system failure.

That's relatively easy to achieve if the field installation uses transmitters and receivers from the same vendors as were used in the test. When different vendors are involved, however, subtle variations from vendor to vendor may cause a system to work with one pairing and not with another. Over time, experience will dictate an acceptance range that will work with all vendor pairings. Until it does, designers can have fits trying to determine why devices that meet specifications in the lab are failing in the field. Stressed-eye testing seeks to eliminate the problem by using the worst case.

Performing a stressed-eye test requires that engineers have the ability to add controlled degradation to the transmitted optical signal. Because traditional optical testing has favored the best possible signal, controlled degradation is not a built-in ability of most optical test equipment.

Companies such as Agilent Technologies (Palo Alto, CA; http://www.agilent.com) and JDS Uniphase (San Jose, CA; http://www.jdsu.com) have met the need for 802.3ae testing by introducing stressed-eye test kits containing optical sources with controllable extinction ratio and the ability to insert externally generated amplitude modulation. These instruments, together with data pattern generators that have controllable jitter, form the stressed-eye stimulus for conformance testing. In most cases, then, the testing requires several pieces of equipment.

An exception comes from Circadiant Systems. The company's Optical Standards Tester (OST) combines all the required elements in a single package. The unit gives you control over optical parameters such as extinction ratio, signal-to-noise ratio, and optical power and also lets you insert errors into higher-level protocol signals.

Even though it is a change in approach, worst-case testing seems likely to take hold in the optical test industry. Charlie Schaffer, product planner at Agilent Technologies, notes that stressed-eye testing has been making strides toward ensuring interoperability in 10 Gigabit Ethernet. Circadiant's French is even more enthusiastic, saying, "It really seems like the right way to do testing." French notes that worst-case testing is now being adopted for 10 Gigabit Fibre Channel, and he expects even established standards such as SONET will be rewritten to include the approach.

Still, old ways die hard. One reason, Schaffer notes, is that some multi-vendor transmitter-receiver pairings work better than others. System designers that want to push the envelope of their system's performance will still want to test the system's absolute performance rather than its simple adherence to a specification.

Even if conformance testing is all a company desires, however, an integrated instrument may not always be the right answer.

Agilent's Schaffer points out that there is a significant installed base of equipment for optical test, and that adding a stressed-eye unit may be the right answer for many test departments. "Besides," adds Schaffer, "if you're doing design and verification, you need all that equipment anyway." As a result, companies such as Agilent are planning to stay with their multiple-instrument test configurations for now.