High-speed digital testing necessary for XFP modules

Th 144542

The XFP/XFI transceiver multisource agreement (MSA) has developed significant momentum in the telecom and datacom industries. Many transceiver manufacturers have announced XFP modules, and a number of communications IC vendors have announced devices in support of the standard. Perhaps most important, at least one key equipment manufacturer, Cisco Systems, has committed to the new interface. Unlike the other 10-Gbit MSAs, the XFP transceivers feature a 10-Gbit/sec differential I/O interface—the XFI or "ziffy" serial interface—instead of the 16-channel SerDes found in the 300-pin MSA and the four-channel XAUI used in XENPAKs, XPAKs, and X2s. Though the other MSAs will doubtless continue to be used for some time, many in the industry expect the XFP to become the dominant 10-Gbit optical interface for applications other than DWDM and long-haul transport.

XFP transceivers offer a compelling value proposition. First, they are designed to be multiprotocol, so they're able to support both of the established telecom standards—SDH STM-64 and SONET OC-192—and emerging applications such as 10-Gigabit Ethernet (10-GbE) and 10-Gigabit Fibre Channel (10-GbFC). Second, partly by moving the SerDes needed for multiplexing and demultiplexing signals out of the module, XFP transceivers are extremely compact. That offers network equipment manufacturers the ability to greatly boost the capacity of existing equipment and new designs with up to 16 full-duplex 10-Gbit/sec channels per line card. XFP modules are hot-pluggable, which enables customers to provision network equipment with just the capacity they need today and add more 10-Gbit/sec channels as needed to meet increased bandwidth demand. Finally, XFP modules offer the most economical 10-Gbit/sec optical interface with prices in quantity quoted as low as $500 each.Th 144542

Figure 1. The XFP module and compliance test board offers test points B´ and C´ for the XFI serial 10-Gbit/sec 100-Ω differential I/O. The applicable digital high-speed measurements are input stress sensitivity and jitter tolerance at B´. At C´, the tests include eye mask and jitter output.

Of course, the XFP MSA will fulfill none of these promises unless suppliers can ensure that the modules will perform reliably. Therefore, proper test and evaluation becomes a critical requirement. While the basic tests should be familiar to most, the XFP's architecture presents unique challenges to developers and manufacturers.

Multiprotocol-capable XFP modules can achieve higher manufacturing volumes through their ability to interface with SONET/SDH telecom equipment, Ethernet in the enterprise and MAN networks, and Fibre Channel SANs. To address multiple markets, XFP manufacturers have to be prepared to perform optical compliance tests to each of the related standards—either on each module or by supplying different model numbers for different applications.

High-speed digital SONET/SDH optical compliance tests include transmitter mask and extinction ratio as well as receiver bit-error-rate (BER) sensitivity testing and jitter tolerance tests. The 10-GbE and 10-GbFC standards require a transmitter mask test, vertical eye-closure penalty and optical modulation amplitude measurements, stressed receiver sensitivity test, and transmitter dispersion penalty test.Th 144543

Figure 2. This XFP differential XFI output mask test at C´ was measured on a new-generation specialized test set.

The XFI 10-Gbit/sec 100-W differential I/O used by XFP modules and compatible board interfaces pose significant signal integrity challenges and create new measurement requirements. The XFP MSA carefully specifies a board with short traces for module compliance testing (see Figure 1). Test point B´ is called the "host system output." That is the XFI differential input to the optical transmitter of the XFP module under test. The XFP standard calls for verifying that the module's electrical receiver operates with a maximum BER of 10-12 when tested with the specified stressed eye in combination with a jitter tolerance test.Th 144544

Figure 3. Jitter peak measurement at test point C´ of total jitter (TJ) and deterministic jitter (DJ) at 10.3125 Gbits/sec. XFP/XFI differential output jitter is specified to have a maximum DJ of 0.18 UI or 17.45 psec and a maximum TJ of 0.34 UI or 32.97 psec. This unit is failing TJ due to poor termination on one side of the differential output.

Test point C´ is the XFP module output. The output is an XFI differential signal from the module's optical receiver. The high-speed digital tests called for at C´ are eye mask (see Figure 2) and jitter output (see Figure 3). All tests require that both output and input signal paths be active during measurements at B´ and C´ and that measurements be made differentially. Coax measurement interconnect cables must be carefully managed to keep channel-to-channel skew below 5 psec. It is also important that all lines be correctly terminated. All eye-diagram measurements are made using clock recovery with a corner frequency of 4 MHz to trigger the oscilloscope.

The stressed eye is defined as an eye-diagram that complies with the specified eye template (see Figure 4). The stressed eye calls for 0.61 UI of total jitter (TJ) and 0.41 UI of non-data-dependent jitter. The specified stressed eye jitter is a combination of jitter present on the compliant data signal, intentionally added intersymbol interference (ISI), and sinusoidal jitter (SJ) added per the specified telecom or datacom template. TJ includes duty-cycle distortion, ISI, random jitter (RJ), and periodic jitter (PJ), which includes intentionally added SJ. Total non-data-dependent jitter is defined in the XFP MSA as TJ minus ISI. This definition is used to represent the portion of jitter that will not be affected by the equalization filter called for in the eye-diagram stressed mask measurements. This equalization filter for the eye-diagram measurements is intended to approximate the inverse response of the XFI channel.

The jitter present in the stressed eye signal is carefully defined in terms of TJ, data-dependent jitter, and bounded uncorrelated jitter (this is the sinusoidal jitter that is added). ISI and SJ are added to reach the required 0.61 UI of total jitter in the stressed eye. The XFP refers to the 10-GbE standard, 802.3ae, for setting up the required stressed eye test. The specified datacom and telecom jitter test templates call for a minimum applied SJ of 0.05 UI from 4 to 80 MHz (at least 4.8 psec for 10.3125 Gbits/sec; 5 psec for 9.953 Gbits/sec) peak to peak.

Once the stressed eye is constructed, BER performance to a (maximum) failure ratio of 10-12 is verified while stepping SJ across the levels specified by the datacom or telecom jitter template. Maximum applied SJ, at the lowest template frequencies (below 130 kHz for datacom, below 2 kHz for telecom), is 1.5 UI peak to peak for datacom applications and 15.2 UI peak to peak for telecom. That obviously means the stressed eye has even larger amounts of jitter at the lower frequencies, but the clock recovery in the XFP should track out this jitter, which is the point of the jitter tolerance test.

Mask testing is a straightforward test to perform on high-bandwidth oscilloscopes or one of the new instruments designed for transceiver testing (see Figure 2). The XFI output mask template is specified so that an output with the maximum allowed TJ would pass the mask test at the eye-crossing. However, the mask test is also called upon to assure that a BER failure rate of 10-12 will not be exceeded using margins or extrapolations, etc. Given the extremely low sample rates of high-bandwidth oscilloscopes, the practicality of assuring a BER of 10-12 is problematic.Th 144545

Figure 4. For this stressed eye input mask at test point B´, X1 is 0.305 UI for total jitter condition and 0.205 UI if total non-data-dependent jitter is measured.

One alternative measurement is to perform a jitter peak measurement on one of the above-mentioned new test sets (see Figure 3), which extrapolates BER performance based on samples taken at the full 10-Gbit/sec data rate—thousands of times more samples per second than high-bandwidth oscilloscopes. Jitter peak (also known as jitter bathtub) measurements also can verify that deterministic jitter meets the required specification of 0.18 UI maximum. The measurement in Figure 3 required 5 sec.Th 144546

Figure 5. This bit-error-rate (BER) contour measurement at test point C´ of an XFP module's 10.3-Gbit/sec XFI output took 31 sec. The BER measurement depth to 6.5-8 gives far higher performance assurance than traditional high-bandwidth oscilloscope mask testing.

A second alternative measurement to the standard mask template test is to use BER contour measurements (see Figure 5). Again, the measurement is executed with sampling rates far higher than a high-bandwidth oscilloscope could provide. That greatly reduces the risk that an undetected BER floor, which an oscilloscope will never see, will allow units to pass that fail to provide BER performance of at least 10-12. Both accurate jitter peak and BER contour measurements depend on extremely accurate delay adjustments. Accurate, high-resolution, monotonic, delay adjustments were not available with earlier-generation 12-Gbit/sec BER test sets.

The new XFP MSA promises to bring economical 10-Gbit/sec optical interfaces to the telecom and datacom industries. XFP's 10-Gbit/sec serial differential I/O, XFI, brings new signal integrity and test demands to device, transceiver, line-card, and backplane designers and manufacturers. New instruments bring accurate and high-speed measurements combined with advanced signal integrity analysis to 10-Gbit/sec testing.

Stressed eye construction requires careful measurements and adjustments of ISI and SJ magnitude. Assuring maximum failure ratios will be less than 10-12 requires sample populations far greater than conventional high-bandwidth oscilloscopes can provide. Careful design and manufacture and new IC designs with integrated emphasis and equalization as well as powerful new measurement and analysis tools will enable rapid adoption of XFP.

Charlie Schaffer is vice president of marketing at SyntheSys Research (Menlo Park, CA). He can be reached via the company's Website, www.synthesysresearch.com.

More in Transmission