Tests reveal importance of transceiver characteristics on OM-4 link performance

By Gerard Kuyt, Pieter Matthijsse, Denis Molin, Christoph Caspar, and Ronald Freund

Overview

With OM-4 fiber coming to the forefront, it is necessary to measure how well it performs in different applications. A recent test examined the performance of ten 10GBase-SR transceivers over such fiber.

There is significant research underway on the use of OM-4 multimode fiber (MMF) for 10-Gigabit Ethernet (10GbE) transmission in fiber links up to 550 m, and in shorter links to increase the power margin when many connectors are required.

With 50/125-μm MMF capacity growing, it’s necessary to examine its limits, particularly in light of the overwhelming growth in short-range singlemode fiber (SMF)-based systems in the local loop. A recent analysis indicated that for a single-fiber/single-wavelength OM-3 based system, the ultimate bit-rate length product limit is as high as 60 Gbps/km for transmission in the 850-nm window. Considering the current 0.3-Gbps/km bit-rate length product for 10GBase-SR systems, there is still enormous growth potential. Applying WDM on standard OM-3, OM-4, and other new “wide window” MMFs will increase this limit substantially. With that in mind, appropriate improvements in sources, detectors, modulation schemes, and mux/demux equipment will become necessary.

This article focuses on testing the system performance of 10GBase-SR transceivers applying OM-4 MMF in links up to 550 m, using cost-effective transceivers specified for a maximum distance of 300 m. A system test bed was set up for MMFs complying with the 10GBase-SR standard and specified for 550-m lengths (OM-4). Multiple links of 300, 550, and 750 m were tested, and transmitters/receivers were chosen from a batch of ten 10GBase-SR compliant transceivers.

Test setup

The basics of the test setup were rather simple (see Fig. 1). The major components were the transmitter, fiber link, variable optical attenuator (VOA), receiver, and a bit-error test apparatus. The fiber link, or fiber under test (FUT), had an LC connector on each end and was connected between a 2-m launching patch cord and the VOA. The output of the VOA was coupled to the receiver by another 2-m patch cord with LC connectors at each end.

Increasing the loss of the VOA causes the BER to increase. The average power impinging on the receiver was measured separately with an optical power meter to check the VOA settings. By doing this with and without the FUT, the effect of the fiber on the system performance could be determined. The difference between the back-to-back (BtB) curve and FUT results curve became the power penalty induced by the fiber link.

Although the general test setup was fairly conventional, several decisions had to be made. For example, commercially available transceivers could have been used to achieve results that apply to today’s practices. Other alternatives included designing a special transceiver/receiver pair or applying two separate transceivers on two different evaluation boards instead of a single looped-back transceiver. The evaluation boards took care of the electronics required to get the transceivers properly working in a nonswitch environment. They also provided auxiliary signals to monitor important parameters, such as laser heat-sink temperature.

Because of its presence in switches containing a series of tightly mounted optical cards, the effect of additional optical and electrical crosstalk was also studied. For the electrical crosstalk, the initial test pulse series was applied via an adjustable delay line. For the optical crosstalk, a second fiber connection was made (see Fig. 1). Different pulse series were used, ranging from the short pseudo-random binary sequence (PRBS) series 27-1 (with a relatively low number of pulse combinations) to the long PRBS series 231-1. The long series, due to the probability to have cascades of “1” bits and thus heat up the vertical cavity surface-emitting laser (VCSEL), is of particular interest for low-cost transceivers lacking thermoelectric elements.

It was observed that changing the setup parameters as mentioned previously measurably influenced the test result. As an example, the influence of electrical crosstalk (XT) was strongly dependent upon the phase of the applied electrical XT signal. Bit-error rate (BER) and receiver sensitivity were also influenced, even in lengths of MMF greater than 550 m. Thus, for reliable testing, the use of electrical XT should be avoided—although optical XT showed no noticeable influence. System performance in a test environment can therefore significantly deviate from behavior in real-life networks.

Each choice had similar impacts on BER, but performance changes could not always be attributed to a single cause due to combined parameter changes. However, test results on the setup made it clear that the dominant parameter was the choice of transceivers.

Transceiver characteristics

To investigate the general characteristics of commercially available transceivers, 10 different transceivers were purchased from three different suppliers. One supplier of four transceivers guaranteed that the optical units originated from three different subassembly suppliers.

It was discovered that various transceiver parameters have a significant impact on the BER test results, particularly the “encircled flux” that will be discussed later. First, since 10GBase-SR transceivers lack thermal stabilization for the VCSEL, temperature stability parameters were checked using a temperature-stabilized chamber to vary temperatures between 10° and 50°C.

The received power in the BtB configuration at the receiver input was measured as a function of transceiver temperature. Four of the transceivers displayed a significant power increase when heated. In parallel with the power test, the BER in the BtB configuration and the laser spectra were also measured at different temperatures. The BER varied up to three decades for the transceiver with the largest power change, while the main effect on the laser spectra was the typical wavelength shift of 0.07 nm/K. However, this characteristic will have a tremendous impact on the transmission over long fiber links due to the increased material dispersion effect.

Noise occurrence is another dominating factor for transmission performance. To get an impression, the eye pattern of each transceiver was measured in BtB configuration applying the Agilent 81495A receiver. Differences among transceivers were small, even at different temperatures. The worst case showed a signal-to-noise ratio reduction from 10.2 to 8.2 dB.

Transmitter encircled flux

In MMF transmission systems, the mode power distribution is of paramount importance for system behavior. In fiber test systems, this distribution can be controlled by special mode-content determining devices, such as filters and scramblers. This is not possible in 10GBase-SR compliant transceivers, so an “encircled flux” (EF) methodology was used. By definition, this is the encircled power as a function of the radius from the core center, measured for the near field that emerges from the short-launch fiber connected to the transceiver. The launch fiber should be a standard 50/125-μm MMF as applied in the fiber link.

EF is typically characterized by two points out of the full curve, such as the power fraction inside a 4.5-μm radius (η4.5μm) and the radius within which 86% of the total power is encircled (R86%). To measure the EF of the transceivers, the Arden MPX-1 was applied (see “Modal launch conditions—what the standards say” on page 23 in this issue for more on measuring EF). Averaging is performed on 500 frames; during the measurement, the fiber is shaken to further reduce effects of the near-field speckle pattern. Figure 2 shows the EF coordinates of the tested transceivers.

In 10GBase-SR systems, the EF is linked in via several routes. First, standard 10GBase-SR transceivers must meet specific compliance requirements. For example, the near field should not be concentrated too much in the core center, nor should it be distributed too much into the outer region of the core. Second, the specific EF coordinate values of a transceiver will influence the BER performance significantly, depending on the applied fiber.

The third route has at least as much impact but is indirect by the definition of the minimum calculated effective modal bandwidth (EMBc) of the fiber. This bandwidth is one of the major selection criteria for the OM-3 and OM-4 fibers and is calculated from the measured differential mode delay (DMD) curves of the specific fiber, as well as 10 additional sets of weighting factors. Each set of weighting factors corresponds with the modal power distribution as launched in the MMF by a fictitious VCSEL.

The EMBc is determined by the laser creating the worst case, i.e., the lowest calculated modal bandwidth. For fiber optimization purposes, it makes sense to both identify this “worst case laser” and investigate its impact on transmission performance.

The 10 EMBc lasers in the test can be considered representative of today’s commercially available lasers. As a reference, the EF coordinates of these lasers are also indicated in Fig. 2, achieved by recalculating the total power distribution in the fiber applying the 10 sets of weighting factors. Note that no measured EF approximates the extreme EMBc laser #1. Also, EMBc laser #6, with a strongly “filled” mode power distribution, is not even approached by one of the purchased transceivers. However, these two EMBc lasers are among the most critical lasers in calculating the EMBc of manufactured OM-4 fibers.

A goal of this investigation was to get feedback on the existence and improvement of certain artifacts in the DMD curve of the FUT. This curve can be considered as the fiber performance “fingerprint” and is the starting point for process improvements. By definition, this curve is related with modal group transmission at all radii, so it seems logical to pursue BER performance testing in which the modal power distribution can be modified. Modifying the EF coordinate values is a good starting point. Two issues included modifying the EF coordinates of a single transceiver and finding test lasers with EF coordinates that coincided with the extreme EMBc lasers found in two of the test transceivers.

A workable method was found by modifying the launching fiber (see Fig. 3). A special fiber was spliced to the original launching fiber and coupled with the transceiver. The “special fiber” can be chosen from a range of available fibers. Total effect is strongly dependent upon the EF coordinates of the applied transceiver. Coupling a transceiver that has a rather high mode volume with a quasi-SMF as the first section of the mode content conditioning launch fiber (MCF) resulted in a very high loss that made BER testing impossible. For three other test transceivers, however, the insertion loss with this fiber was acceptable.

An issue still under investigation is the impact of the EF modification on the other transceiver parameters. A tradeoff has to be made between the importance of reaching a specific EF coordinate value and the deterioration of the other transceiver parameters. EF modification obtained by giving preference to certain laser modes will increase mode partition noise due to the reduced noise averaging effect and can even affect the laser spectrum. Both parameters influence system performance and should be taken into account when judging the test results.