Cooperation to compliance with SFP+
By Martin Rowe, Senior Technical Editor- October 1, 2008
Large data centers such as those at Facebook or YouTube have thousands of servers that must connect through switches and fiber-optic lines. These data centers need an ever-increasing supply of 10-Gbps links to keep up with traffic. To reduce costs and save space, network equipment manufacturers need to put more links onto line cards. Because of their size, SFP+ (Small Form-Factor Pluggable) optical transceivers let line cards hold more data lines than other transceiver modules.
SFP+ modules are smaller than other transceiver modules because CDR (clock data recovery) and SerDes functions have been moved off the module and onto the line card, or host (Figure 1). SFP+ modules are also unusual in that they have two variants, called limiting and linear. The limiting variant includes a limiting amplifier that resets the amplitude of the signal eye. Linear SFP+ modules don’t need the limiting amplifier. They send an analog representation of the incoming optical signal to the host where an EDC (electronic dispersion compensator) compensates for dispersion in the optical transmission and for losses in the electrical signal. “Why two variants?” explains the applications for limiting and linear modules.
Figure 1. SFP+ modules move more signal processing to the host board than other transceiver modules.
Both variants challenge engineers because they permit signals in the module to travel farther before they are restored to a clean form than do other optical transceiver modules. The longer path can add jitter and loss to the signal.
Testing SFP+ modules requires electrical and optical measurements. Because of the longer electrical signal channel, you must make measurements on the channel within the module and at the module’s input (for transmit) and output (for receive) ports. You must characterize the channel on the host, too. Optical measurements such as receiver sensitivity and transmit jitter and power are based on compliance to standards such as IEEE 802.3ae and 802.3aq (Refs. 1, 2). Optical measurements, therefore, are the same as those for other 10-Gbps modules.
The electrical measurements, however, differ from those for other modules, and they also differ for the limiting and linear SFP+ variants. The SFP+ specification defines test processes designed to ensure a product’s compliance to the specification (Ref. 3). Although some tests remain under development as of the May 8, 2008, draft, manufacturers are moving ahead with designs and production.
Jitter tests need definition
The test for jitter at the output of the receive-side limiting module and at the input for the limiting host is one test that needs defining before the SFP+ standard is finalized. George Noh, senior systems engineer and senior member of the technical staff at Vitesse, explained that the current SFP+ specification for the limiting receiver calls for a maximum total jitter of 0.7 UI (unit interval) and a maximum deterministic jitter of 0.42 UI. How you arrive at those numbers depends on whether you use an oscilloscope or a BER (bit-error rate) tester.
“One of the remaining open items in the SFP+ standard is how we want to measure jitter, and what limits make sense with that particular methodology,” noted Noh. “Some people prefer the dual-Dirac method using a BertScope from Synthesys Research or an Agilent J-BERT, while others prefer to use an Agilent DCA-J sampling oscilloscope. The dual-Dirac method uses low-probability jitter to fit Gaussian curves to two Dirac functions from which it measures total jitter, deterministic jitter, and random jitter at a particular BER. In contrast, the DCA-J tries to find deterministic jitter by subtracting extrapolated random jitter measured in the frequency domain from extrapolated total jitter at a particular BER. Our goal in the SFP+ committee is to specify the jitter methodology that better quantifies the host receivers’ weaknesses in a real system.”
Figure 2. A module compliance board provides optical and electrical test equipment with access to an SFP+ module’s ports. Photo courtesy of ClariPhy Communications.
Testing SFP+ modules and hosts requires the use of compliance test boards that provide test equipment with access to the module or host system under test. Compliance test boards provide standard points that provide access to modules and line cards regardless of manufacturer. Figure 2 shows where test equipment connects to a module compliance test board. The test board’s electrical signal path has known characteristics. Another option is to connect a vector signal analyzer to the module’s or host’s electrical transmit and receive ports and measure its S-parameters. Engineers at module makers may share their S-parameter measurements with manufacturers of line cards. Module manufacturers may also share design simulations with designers of line cards and PCBs (printed-circuit boards). Manufacturers of CDR, EDC, and TIA (transimpedance amplifier) devices also may share S-parameter data and design simulations with customers.
Figure 3. A host compliance board attaches to a host board, providing access for test equipment. Photo courtesy of ClariPhy Communications.
“Because of this shared responsibility,” said Rob Hannah, applications manager at Avago Technologies, “we must deliver simulations of our products to our customers.” Similarly, Vitesse, which manufactures limiting amplifiers, laser drivers, and transimpedance amplifiers, provides device data to module makers. “We swap hardware with module makers and test devices in both labs, and we analyze all signals,” said Noh. Noh will also work with line-card manufacturers because Vitesse manufactures CDRs and EDCs for line cards.
In order for this data sharing to be effective, all of the manufacturers must use identical test setups for test consistency. Figure 3 shows a host compliance test board with associated test equipment. (The SFP+ specification shows the host and module test setups in more detail.) Designers of host boards can characterize parameters such as transmit jitter and receiver sensitivity. A vector network analyzer can also measure S-parameters looking into a host. Line-card designers can use these measurements along with those provided by module makers to verify impedance matches before designing a host.
Figure 4. SFF-8341 defines test points (B, C, B', C') and measurements (Table 1) for the electrical paths of modules and host boards.
SFF-8431 defines several tests for transmitters and receivers, including two sets of receiver measurements that cover the limited and linear variants. The document also defines tests for hosts boards. Figure 4 and Table 1 highlight the measurements and test points. In addition, the standard defines a calibration procedure you should follow prior to making measurements. “Calibrate before testing” highlights the need to calibrate a test setup.
Optical tests on SFP+ transmitters are the same for both linear and limiting modules because both variants use the same transmitter circuit. Transmitter tests usually use an electrical signal such as a 9-bit or 31-bit pseudorandom bit sequence (PRBS9 or PRBS31) injected into the module through a module-compliance test board.
To perform optical measurements on transmitters, you need an oscilloscope with an optical input. Oscilloscopes that lack optical inputs require an optical receiver to convert the optical signal to electrical. Optical measurements include mask margin, extinction ratio, jitter (total, deterministic, and random), optical rise time and fall time, and RIN (relative-intensity noise).
Hannah noted that specialized oscilloscopes, commonly called digital signal analyzers or communications signal analyzers, measure these quantities through features not commonly available on general-purpose oscilloscopes.
|Examples of specialized equipment used to perform optical measurements on transmitters.|
Tests that don’t require oscilloscopes include TDP (transmitter dispersion penalty) for IEEE 802.3ae SR (short reach) and LR (long reach) applications and TWDP (transmitter waveform dispersion penalty) for IEEE 802.3aq LRM (long-reach multimode) applications (Ref. 4). TDP and TWDP quantify the degradation of signals caused by transmitters and fibers. TDP is a hardware setup that uses a reference receiver and BER tester to measure a transmitter’s effect on receiver sensitivity. TWDP tests use software to process a digitized transmitter signal. The software calculates TWDP based on calculated SNR (signal-to-noise ratio) from a simulated receiver and reference fiber.
To test whether the electrical signals transmitted by line cards comply with SFF-8431, you can use a host compliance board that inserts into the SFP+ cage on a line card. The compliance board’s SMA connectors provide access for an oscilloscope or a BER tester. Because the CDR circuit resides on the host, the transmitted electrical signals from the line card will travel along several inches of PCB trace before reaching the host compliance board. At 10 Gbps, the digital signals from the CDR will experience distortion. Thus, to perform compliance tests on the electrical signals sent by hosts, you must measure for amplitude, jitter, and BER.
Test methods for SFP+ receivers depend on the Ethernet standard to which a module and host board must comply. Limiting SFP+ modules need tests with an open eye per IEEE 802.3ae (Figure 5). Jitter, noise, and low amplitude add stress to the eye, but the eye opening will still be clearly visible. You must also add at least 0.3 UI of jitter in the signal. Under these and other conditions, a module under test must pass a 10–12 BER test.
Figure 5. Limiting SFP+ modules require receiver testing with an open eye input.
Linear modules, which produce analog outputs, may be less familiar to digital engineers than limiting modules. Figure 6 shows how dispersion in the LRM fiber causes adjacent light pulses to interfere with each other, resulting in ISI (intersymbol interference) and a closed eye at the receiver. SFF-8431 requires that you test linear SFP+ modules for compliance using three ISI test conditions—precursor, postcursor, and symmetric. These three conditions simulate wave shapes (Figure 7) that often appear as a result of dispersion in LRM fibers.
Figure 6. Model dispersion in optical fibers creates intersymbol interference that closes eye diagrams. Courtesy of Circadiant Systems.
Because of the closed eye, SFP+ modules can’t use a limiting amplifier and need an EDC. “Ultimately, you’ll see the EDC integrated into the host ASIC where its cost gets absorbed,” said Tom Lindsay, principal systems engineer at ClariPhy Communications and chair of the SFF-8431 committee. “Because an open eye isn’t needed, the linear TIA can potentially have lower bandwidth and thus a lower cost. Power and heat management are easier to implement on a host than on a module.”
Figure 7. Compliance tests for SFP+ modules on 10GBase-LRM fiber require three types of intersymbol interference.
To test a linear SFP+ module, use a module compliance test board. You must observe the distortion that a module adds to the outgoing electrical analog signal. Joey Thompson, executive VP at Circadiant Systems, noted that the amount of added distortion isn’t currently defined. Thus, it’s important to make sure that the line card’s EDC will be able to convert the incoming signal into digital form.
Testing the EDC circuit on a line card requires an electrical stressed-eye signal injected into the circuit. EDC evaluation boards, available from EDC manufacturers, let you test the device under stressed-eye conditions prior to incorporating them into a line-card design. SFP+ manufacturers need to test their modules with numerous EDCs and thus will use evaluation boards. Test equipment lets you create electrical or optical stressed-eye signals, letting you test the module, the EDC chip, or both as a system. To test an EDC once it’s on a line card, use a host compliance board to inject stressed-eye signals.
Although some of the SFP+ specifications and test requirements are still open to debate, manufacturers are pushing ahead with creating SFP+ products because of the economic benefits. The quest to lower costs for network operators continues.
|FOR FURTHER INFORMATION|
|Calibrated Jitter, Jitter Tolerance Test and Jitter Laboratory with the Agilent J-BERT N4903A, Application Note, Agilent Technologies, July 2006. cp.literature.agilent.com/litweb/pdf/5989-4967EN.pdf.|
|Foster, Guy, Dual-Dirac, Scope Histograms and BERTScan Measurements—A Primer, SyntheSys Research, September 2005. www.synthesysresearch.com/Literature/White_Papers/Dual-Dirac.pdf.|
|Measuring Extinction Ratio of Optical Transmitters, Application Note, Agilent Technologies, 2001. cp.literature.agilent.com/litweb/pdf/5966-4316E.pdf.|
|Precision Jitter Analysis Using the Agilent 86100C DCA-J, Product Note, Agilent Technologies, March 2007. cp.literature.agilent.com/litweb/pdf/5989-1146EN.pdf.|
|Thompson, Joey, and Earnest E. Bergmann, “Stress Tests in High-Speed Serial Links,” Digital Communications Test and Measurement, ed. by Dennis Derickson and Marcus Müller, Pearson Education, Boston, MA, 2008, Chapter 12.|