10G Linear TIA in Long-Reach Multi-Mode Applications
By Lian Zhao, Senior Application Engineer and Ariel Nachum, Validation Test Engineer, Inphi Corporation

Long-reach multi-mode fibers (MMF) applications suffer from multi-path modal dispersion where different modes of light arrive at the receiver at different times, severely limiting the maximum bit rates and transmission distances. Increasing capacity requirements in the enterprise and data center networks require the upgrade of the network backbone from 1 Gbps to 10 Gbps, which is not possible with existing components over the installed base of legacy OM1 Multi-mode fiber. Since replacing the legacy OM1 fiber is too expensive, the installed base of legacy MMFs is severely bandwidth limited due to modal dispersion, making it extremely difficult to achieve 220m at 10 Gbps.

The upgrade of the enterprise backbone networks was initially deployed using LX4 technology, which uses four wavelengths running at 3.125 Gbps each. This technology is expensive and consumes significant amounts of power. As a result, the LX4 technology failed to reach mass adoption in high volume. Recently, however, a new technology has been developed to overcome this problem: the long-range multi-mode (LRM) serial link at 10 Gbps with its cost less than half the LX4 solution and with significantly lower power consumption. LRM uses the complex technology of Electrical Dispersion Compensation (EDC) to compensate for optical dispersion in the electrical domain and requires a linear transimpedance amplifier (TIA).

For LRM applications, the performance of the EDC is critical for a reliable optical-link performance. The most common architecture is feed-forward equalizer/decision feedback equalizer (FFE/DFE). Adaptive analog technology and Digital Signal Processors are two methodologies that are commonly used to enable FFE/DFE. Equally important for LRM, is the performance of the TIA, which must amplify the signal without introducing distortion. A linear TIA not only has all the normal limiting TIA performance, like small sensitivity and overload, but it includes unique linear functions, like Total Harmonic Distortion (THD), Waveform Distortion Penalty (WDP) and Relative Noise (RN).

Linear TIA’s Unique Parameters
Before linear TIAs, transimpedence amplifiers were limited both in linearity and range. The linear range of a TIA depends on its transimpedence (Zt). The upper 3 dB bandwidth (BW) measurement, for example, falls within the linear range. A measurement of –20 dBm, available on the latest TIAs, with the Zt of 10k, is considered small signal BW.

For linear TIAs, using an automatic gain control (AGC) circuit to make the linear range from sensitivity range to overload range is preferred. AGC is an adaptive system where the average output signal level is fed back to adjust the gain to an appropriate level for a range of input signal levels. Linear TIAs should remain linear from small signals (sensitivity level) to large signals (overload level). Figures 1 and 2 show the BW data using the same TIA at different input optical power levels.

The above figures show that the BW differs using different input power even when the BW is measured with the same device. Both the AGC circuit time constant vs. network analyzer sweep time set and the AGC circuit frequency dependency could possibly cause these significant peaks.

For some TIAs, the AGC time constant can be adjusted using external capacitance. With the capacitance range of 270 pF to 1 nF, the AGC time constant can be measured from 1-3 ms. The network analyzer sweep time can be adjusted to be as slow as possible. In this example, the sweep time was 1 s allowing the AGC time to be neglected. From the initial test results, no difference in the BW measurement is evident as the sweep time changes. Clearly, the AGC time does not cause the large BW signal peaking.

With frequency dependency, the test results are better when the TIA die uses an electrical signal instead of an optical signal. For an optical signal, the optical sine wave performance must be checked and adjusted using laser bias current and modulation current to guarantee the laser works in the linear range. For the electrical signal, the sine wave signal can be added directly to the TIA input using bias T. The results show that in the overload range, the output amplitude remains constant when the input signal is 1 GHz and 2 GHz sine wave. The output amplitude increases as the input signal changes to 4G and 5G at the overload current. These results indicate that the AGC circuit is frequency dependent, and match the BW curve in Figure 2, which starts to ramp after 2G.

The following setup (Figure 3) is used to measure the linear TIA BW with the AGC circuit on. Two wavelength signals are used: the 1310 nm at 100M sine wave frequency to turn on the AGC circuit and the 1550 nm optical signal to measure the linear TIA BW using the sweep process.

To verify that the set up provides for the BW measurement with the AGC circuit turned on, it is important to first verify that the 1310 nm signal at 100M changes the TIA gain. With the AGC circuit on, the BW is measured with the –20 dBm 1550 nm optical signal from the HP8703.

The sequence of events is as follows: the 1550 nm laser source is blocked, then the TIA input is changed to get different TIA output and the TIA AGC performance is verified. Figure 4 shows the 10G linear TIA working properly with the 100M sine wave under different inputs.

Figures 5 and 6 show the BW measurement using the above methodology in the same TIA at different input powers.

The methodology for the large signal BW is useful when checking the linear TIA BW with the AGC circuit on. With the Inphi ROSA test, the BW for both the large and small signals is consistent. In the following, BW refers to the small signal BW.

Total Harmonic Dispersion (THD)
To measure the linear TIA THD, a linear 1310 nm laser source is used to modulate with a 1 GHz sine wave. The laser bias and modulation can be adjusted to obtain appropriate signals. The following (Figure 7) is used to test the laser linearity. An EDFA may be needed to provide enough peak-to-peak overload input optical power, and the spectrum analyzer must be monitored to guarantee that only a 1 GHz sine wave signal is present without additional harmonics of the fundamental frequency. The procedure is the same if 2 GHz and 3 GHz sine wave signals are needed to modulate the laser.

After selecting the proper bias current and modulation current, the NEW FOCUS PD is replaced with the ROSA under test. From the spectrum analyzer, the fundamental signal (1 GHz), 2nd order harmonic (2 GHz) and 3rd order harmonic (3 GHz) signal power as P1, P2 and P3 are read (Table 1). THD is calculated using the following: THD=sqrt(P2+P3)/sqrt(P1).

Waveform Distortion Penalty (WDP) and Relative Noise (RN)
IEEE 802.3aq defined three different stressed optical signals as precursor pulse, post cursor pulse, and split symmetric pulse. These parameters and related PIE-D values (a metric used to assess EDC capability to equalize the link) are used to test the link performance using the linear TIA.

WDP is a deterministic dispersion penalty due to a particular transmitter with reference emulated multimode fibers and receivers. WDPi is the dispersion penalty of the TP3 test signal, and WDPo is the dispersion penalty measured at the ROSA output. The distortion contributed by the ROSA is determined by the following: dWDP= WDPo-WDPi Circadiant Hydra test system is a popular stressor generator used by LRM IC, module and system vendors, and ensures that difficult stressors are consistent in the semiconductor food chain. Figure 8 is a typical WDP measurement set up.

According to IEEE802.3aq, the simulated fiber stressor are set to 4.1 dB, 3.9 dB and 4.2 dB for pre-cursor, symmetric and post-cursor, respectively. It is better to measure the optical source WDPi for different stressors separately for improved accuracy.

WDP for a linear TIA is similar to TWDP for the transmitter. By setting the test signal pattern PRBS9, the optical power to –6.5 dBm, and the extinction ratio to 3.5 dB, the Hydra optical source WDPi can be calibrated using different stressors, and the ROSA WDPo can be measured (Table 2).

RN is the reciprocal of SNR for a signal. The test set up is similar to WDP, when the Hydra test system is being used (Figure 8), although it can be tested using a scope following the instructions given in SFF8431 for manual RN calculations. Where RNi is the relative noise of the test signal characterized using O/E curve, the relative noise contribution of the module is calculated using the following: dRN=sqrt(RNo^2-RNi^2) (Table 3).

Linear TIA 1348TA BW Relation with dWDP and dRN
Noise performance and bandwidth are affected by PD capacitance and bond wire inductance on the TIA input node. Therefore, the same TIA assembled by different firms using different photo detectors can differ in performance. Selecting two different ROSAs: ROSA1 and ROSA2, results in different bandwidths (Figures 9 and 10).

The different ROSA BWs affect the ROSA’s dWDP and dRN data. Table 4 shows the trend with using Inphi linear TIA 1348TA.

In this section, each of these parameters is unique for the linear TIA 1348TA, not for any TIA +EDC combined performance.

Linear TIAs with different vendor EDCs
The most important test in LRM applications is the optical comprehensive stressed test. Optical comprehensive stressed receiver sensitivity and overload can be done using a Circadiant Hydra where an electrical signal is created using a specified pattern and impaired by:

1. Gaussian low pass filter
2. Gaussian white noise source
3. ISI (pre-cursor, post-cursor and split symmetric-cursor)

The resulting electrical signal is converted to an optical signal using a linear electrical/optical converter. The optical attenuator is connected before testing the linear ROSA and EDC for comprehensive stressed receiver sensitivity and overload.

Recently, FFE/DFE (feed-forward equalizer/decision-feedback equalizer)-based EDC techniques for ISI (Intersymbol Interference) mitigation in optical networks have been commercialized. Today, SiGe process-based EDC is in production stage, while power consumption and price motivated IC companies design CMOS process based EDC. EDC integrated with CDR and EDC integrated Serdes target LRM markets with different packages, like SFP+ and next generation X2.

ROSA1 and ROSA2 samples were tested using four different EDC available from different vendors. The most difficult split symmetric stressor is used for comparison. Figure 11 shows the typical stressor test set up.

The block diagram (Figure 11) EDC+CDR is very general; some EDC and CDR are separate components, some are EDC+CDR integrated, and some are EDC+Serdes integrated with only a loop back circuit used. Table 5 shows the results of these tests at nominal conditions with a 3.3 V power supply for TIA and PD at room temperature and it shows that different BW ROSAs work differently with diverse EDC vendors even though the same TIA is used internally. For split symmetric stressed tests, EDC vendor A and C work well with low BW ROSA and high BW. Comparatively, Vendor B works better with high BW ROSA, while vendor D is partial to low BW ROSA.

The methodology used to implement these FFE/DFE-based equalizers can be divided into analog and digital. Theoretically, EDC, by using adaptive analog methods, can tolerate more ROSA noise and higher bandwidth ROSA. EDC, using the DSP method, prefers ROSA low noise and needs low bandwidth. With more solid EDC design using both analog and digital, the comprehensive stressed test results should be very similar when using different BW ROSAs. Figures 12 and 13 show the typical stressed test of 1348TA ROSA2 with vendor A symmetric stressed sensitivity and overload results.

Table 6 shows the updated EDCs with different BW of 1348TA ROSAs.

Conclusion
As EDC and linear TIA technology get more mature, LRM application will have a very bright future. In 2008, SiGe based EDC in X2 package has been shipped in volume (several ten thousand parts). In 2008, as CMOS based EDC and linear TIA become more solid, next generation LRM X2 (with low cost and power consumption) and SFP+ package will have more volume.

Linear TIA +EDC can be worked not only in LRM multimode fiber applications, but also in single mode fiber applications like metro and long haul. With EDC, 40 km-rated EML perform similar to an 80 km version EML. Carriers can extended their transmit distance with less change.
With the progress so far, it is foreseeable that linear TIA +EDC will soon achieve milestones for performances, power consumption and economics that can make it viable as a widespread technology solution for the market of 10 Gbps and beyond.

Inphi Corporation
www.inphi-corp.com
TXTLINX.COM156
Email this article to a friend!