Open Access
Add to BibSonomyAbstract
We propose using electronic equalization technology to allow components typically used in 2.5Gb/s systems to be used at 10Gb/s. We simulate the performance of links exploiting this concept and study the effect of receiver bandwidth on equalized systems in general. Links utilizing transmitters designed for 2.5Gb/s rates are experimentally demonstrated. Experiments also show that photo-receivers with 2.5 GHz bandwidths add minimal penalty when post-detection electronic equalization is employed.
©2003 Optical Society of America
1. Introduction
Over the last several years it has become economically advantageous to increase the data rate of systems from 2.5Gb/s to 10Gb/s. While 10Gb/s systems generally require more expensive components, such as wide bandwidth optical modulators and photo-receivers, the net cost-per-bit is still lower than transmission at 2.5Gb/s since only one-fourth as many channels are required at the higher data rate. Unfortunately, 10Gb/s systems are more susceptible to optical impairments such as polarization mode dispersion (PMD) and chromatic dispersion (CD). In many cases, the cost of optically compensating these effects is prohibitively expensive, as well as overly bulky. Optical impairments such as CD and PMD limit the transmission distance for a specified power budget. Alternatively, for a specified link length, these impairments reduce the power margin.
Electronic processing in the receiver has been used to compensate for optical impairments such as PMD, CD, and modal dispersion [1–6]. Since micro-chip sized electronic equalizers can be mass produced using standard semiconductor processing, they are very likely to become indispensable parts of next-generation optical communication systems. Moreover, electronic processing can also compensate for distortions occurring in the electrical domain. We propose that such equalization technology could also be used to relax the specifications of optical modulators and receivers down to levels traditionally used in 2.5Gb/s systems, thereby producing significant cost savings. Additional benefits such as lower driving voltages, reduced spectral bandwidth, and increased sensitivity may also be obtained.
In this paper we numerically model the performance of 10Gb/s optical links using various types of low-bandwidth transmitters and receivers. We predict that acceptably low penalties, and in some cases significant improvements, may be obtained by substituting low-bandwidth components and subsequent processing with an electronic equalizer, also known as an electronic dispersion compensator (EDC). We also show that the effect of receiver bandwidth on transmission quality, which is usually limited by balancing inter-symbol interference (ISI) on the low bandwidth side and added noise on the high bandwidth side, is modified by the addition of an EDC. This can be understood by noting that the EDC works to partially cancel ISI. Therefore, lower receiver bandwidths can be used with minimal added penalty or, in some cases, improved performance. Additionally, we experimentally demonstrate the use of Mach-Zehnder interferometer (MZI) and electro-absorption (EA) modulators designed for OC-48 systems, operating at OC-192 rates. A proto-type EDC is used to drastically reduce the penalty normally incurred by using such low-bandwidth transmitters. We also show that adding a 2.5GHz bandwidth filter after our photo-receiver causes minimal penalty after electronic equalization.
2. Results and discussion
2.1 Low bandwidth receivers
We first simulate the performance of optical noise limited systems which are dominated by amplified spontaneous emission (ASE) noise from optical amplifiers. A 223 -1 pseudo-random bit sequence drives the NRZ modulator at 10Gb/s. The system is modeled using the commercial program OPTSIM. For simplicity, we assume that the same 25GHz optical band-pass filter is used in every link. For a given transmitter and link length, the bandwidth of the photo-receiver (4th order Bessel filter) is varied. After detection the signal is either passed through or by-passed around an EDC. A 6 tap (two taps per symbol) feed-forward equalizer (FFE) is modeled with a 1 tap decision feedback equalizer (DFE) [1]. The Q-function is then determined by evaluating the noise on the ones and zeros of the digital signal. The Q is quoted in dB where a Q of 15.56dB represents a 10-9 BER. We use the definition QdB=20*log10(Qlinear). We checked the accuracy of the Q-function estimation by performing several simulations directly counting errors to find the bit error rate. The two methods typically differ by 0.1 to 0.4dB, due in part to error propagation effects in the DFE [7]. The Q estimation technique is thus reasonably accurate and will be used throughout this work.
Least-mean square (LMS) algorithm is a widely used technique in order to acquire and track the system parameters. LMS algorithm [8–9] can be seen as the adaptive implementation of the Wiener solution for a set of unknown parameters. Specifically it searches for the set of coefficients where the projection of the estimation error is orthogonal to the excitation vector. In our case the error is the difference between the detected digital data and the equalizer output while the excitation vector is the receive data.
Figure 1(a) shows the effect of varying the receiver bandwidth on the Q-function. For a Mach-Zehnder interferometer (MZI) modulator designed for OC-192 rates, the ideal receiver bandwidth is about 7GHz. The performance begins to drop quickly once the bandwidth is decreased below 5GHz. In contrast, the same modulator coupled with EDC shows much less sensitivity to receiver bandwidth. The performance is nearly constant between 2 and 12 GHz. These results show that it is not desirable to use OC-48 receiver components (which typically have bandwidths between 1.7 and 3 GHz) for 10Gb/s data rates unless combined with electronic equalization.
Fig. 1. Simulated results for transmission using an MZI modulator specified for 10Gb/s data rates (a) Top -Effect of receiver bandwidth on an ASE-noise limited system (b) Center - Effect of filter bandwidth after photo-detection using a wide-band receiver in a power-limited system (c) Bottom - Effect of receiver bandwidth when its noise properties are adjusted to give a constant sensitivity (1E-9) at about 3000 photons per bit (see text).
Figure 1(b) shows a similar plot but with electronic noise dominating the system. In Fig. 1(b), a wide-bandwidth (40GHz) photo-detector is assumed and the bandwidth of a Bessel filter after photodetection is varied. In Fig. 1(c), we change both the bandwidth and sensitivity of the receiver together so as to factor out the sensitivity parameter, making the assumption that the number of photons required per bit remain constant (about 3000 photons/bit). For instance, we assume that a 7.5GHz bandwidth detector has a sensitivity of -24dBm at 10Gb/s and a 3.75GHz bandwidth detector has a sensitivity of -27dBm at 5Gb/s. The OPTSIM program back calculates the noise properties of the receivers from the knowledge of their sensitivities at a specified bit rate. The electrical noise is assumed to come from a white-noise current. The receivers are evaluated at 10Gb/s data rates. Figure 1(c) shows that the ideal receiver bandwidth is reduced by adding EDC. Additionally, performance at the ideal bandwidth is significantly increased. By comparing Fig. 1(b) and 1(c), we can see that this increase in performance is due to the assumed increase in sensitivity as the bandwidth is reduced (a characteristic often seen in practice). This characteristic may make low bandwidth receivers, when used with an EDC, particularly well suited for power-limited links. Without EDC, the ISI added by moving to low (<3GHz) bandwidths outweighs the improved noise performance and causes significant power penalties.
Fig.2. Schematic of the general test system. The low-pass filter (LPF) and FFE (equalizer) may or may not be inserted in the system depending on the particular test. The clock input to the BERT is selected accordingly.
Fig. 3. Five tap FFE structure used in experiments with tap spacing at the symbol rate. Equalization and de-multiplexing are simultaneously performed.
In order to support these simulations, experiments were performed using a prototype 5 tap (1 tap per symbol) FFE equalizer with self-adaptive control algorithm [10]. The tap weights are digitally controlled with a 7-bit resolution. Unlike the simulations, no DFE stage is used. Figure 2 shows the schematic of the general test-bed. First we evaluate the effect of low-pass filtering after photo-detection. A MZI modulator designed for OC-192 rates is used. It is driven with a 223 -1 pseudo-random bit sequence. A 7.4 GHz PIN/TIA converts the optical signal into an electrical signal which is further amplified by an automatic gain control (AGC) post-amplifier with an estimated bandwidth of 8Ghz. The AGC has a differential output. One output is either sent directly to a bit-error ratio (BER) tester, or sent to the EDC for processing. The performance of the system is evaluated both with and without a 2.5 GHz low-pass filter (LPF) after the AGC. The LPF is approximately a fourth-order Bessel-Thompson type. The EDC de-multiplexes the input data stream into four OC-48 streams (see Fig. 3); one of which is sent to the BER tester for evaluation. When measuring the de-multiplexed data BER, a 2.5GHz clock generated in the FFE block is used to trigger the BERT. The other AGC output is sent (unfiltered) into a clock-recovery (CR) module. Because this CR module is designed for low ISI (typical) systems, it will not properly recover the clock after the LPF. However, we believe that a CR circuit designed to recover signals with high ISI would be able to function properly.
The results, shown in Fig. 4, show that without equalization the LPF destroys the performance of the link. Even at the reduced rate of 8.5Gb/s, the LPF causes significantly degraded performance. The penalty induced from filtering is highly dependent on the bit rate and increases very dramatically if the bitrate is further increased. However, the addition of EDC reduces the penalty to manageable levels even at OC-192 rates (9.953 Gb/s). The penalty between a 75km link using the unfiltered receiver in comparison to adding the LPF and EDC is only 0.8dB. Additionally, the degradation in sensitivity seen between operating at 2.5Gb/s versus adding the EDC and operating at OC-192 is about 3.4 dB (not shown in Fig.), with a 2.5G LPF used in both cases. This supports the premise that if a 2.5Gb/s receiver has a higher sensitivity (~ 5dB) than a 10Gb/s receiver, system performance may be improved by using the lower-bandwidth receiver in tandem with electronic equalization.
Fig. 4. Experimental data showing the performance with the 2.5GHz LPF inserted in the system before the FFE (solid lines). For comparison, the performance without the filter/FFE is also shown. Additionally, a data set with the filter and without the EDC is shown at a data rate of 8.5Gb/s - black dashed line (all other data is at 9.953 Gb/s). This demonstrates the poor performance the LPF causes without compensation.
Figure 5(a) shows eye diagrams after the band-limiting filter both before and after equalization. The eye before equalization is nearly closed due to the large amount of ISI added by the LPF. After equalization, the eye is clearly open. Figure 5(b) shows that, even at fairly high input powers, the eye after the LPF becomes fully closed when the data rate is increased to 10.664Gb/s. This rate is chosen because it is compatible with standard forward-error correction (FEC). Because the eye is fully closed, FEC would not be effective in this situation and could not be used instead of equalization.
Fig. 5. (a) Left - Eye diagram after the 2.5GHz LPF before (top) and after (bottom) the FFE. The received power is -19.4dBm. (b) Right - Eye diagrams after the 2.5GHz filter with -16dBm input power at 9.953 Gb/s (top) and 10.664 Gb/s (bottom). The fully closed eye at the FEC compatible rate suggests that FEC would not be an alternative to equalization.
2.2 Low bandwidth transmitters
We now investigate links using low-bandwidth transmitters. Figure 6 displays the simulated results of various links using transmitters modeled according to Table 1(a) when using the 10G receiver specified in Table 1(b). The attenuation and chirp parameter for the electro-absorption modulator is varied with driving voltage so as to meet the dispersion penalty specifications of a commercially available part (<2dB power penalty for a 640km 2.5Gb/s link). Shot noise and multiplication noise are now included in the receiver model (M=10,F=5). We show the received power required to generate Q=17 (BER=10-12) as a function of propagation distance in SMF fiber (D=17ps/nm.km). Notice that for links longer than 80km, the 2.5G MZI transmitter with equalization outperforms the 10G MZI transmitter without equalization (reference case). Also note that the use of the 2.5G EAM modulator with EDC causes a noticeable, but reasonable, penalty of less than 4dB in the back-to-back configuration. It is also less susceptible to chromatic dispersion than the reference case. The penalty for using the 2.5G EAM without an EDC (not shown) is over 10dB and is therefore not a reasonable option.
The performance of low-bandwidth transmitters is also evaluated experimentally. Table 2 shows eye diagrams for all the transmitters when detected with a 14GHz PIN receiver under back-to-back conditions. The high optical power (-1dBm) incident on the wide-bandwidth PIN detector effectively removes electronic noise from eye diagram, allowing one to clearly see the distortion generated by the transmitter. Obviously, there is a significant variation in transmitter performance. The MZI’s were used in conjunction with an external DFB laser near 1530nm. The EAM (Fujitsu FLD5F14CN) is integrated with a DFB laser which is also near 1530nm. A 223-1 pseudo-random bit sequence drove the modulator at the OC-192 rate. Note that, similar to the previous experimental results, a 10G PIN detector was used as opposed to the APD which was simulated. Therefore, the simulations and experiments differ by several dB in terms of sensitivity.
Fig. 6. Simulations showing the sensitivity results for various modulators at a BER of 1e-12. SMF fiber has a dispersion coefficient of 17ps/nm.km.
Table 1 (a):. Parameters used for various transmitters
Table 1 (b):. Parameters used for various receivers (without post-compensation)
Figure 7(a) shows the performance of the system using a 2.5Gb/s or a 10Gb/s MZI. The 10Gb/s MZI has a -20.3 dBm sensitivity at a BER of 10-9, which is within 0.1dB of the sensitivity of the photo-receiver measured at the manufacturer. Adding the EDC improves the performance by an additional 1dB due to mitigation of residual ISI (not shown in Fig.). The 2.5Gb/s MZI has a back to back sensitivity of -16.9dBm without equalization and -20.6dBm with equalization. Impressively, adding 75km of SMF fiber degrades the sensitivity after equalization by only 0.4dB. Similar to the simulations, the data suggests that replacing a 10G MZI with a 2.5G MZI and electronic equalization can actually improve system performance, especially for longer links. In fact, the experiments show an improvement even in the back-to-back case. This is most likely due to additional distortions occurring in the AGC amplifier which are partially compensated by the EDC.
Figure 7(b) shows the performance with a 2.5G EAM modulator. This modulator was designed for use in long-haul (600km) OC-48 links. As might be expected, without equalization the EAM is unsuitable for OC-192 data rates. However, the EDC reduced the back-to-back penalty with respect to the 10G MZI to less than 3dB. This is fairly consistent with our simulations. We note that in this case the receiver bandwidth is wider than the transmission bandwidth, and further performance improvements might be made by optimizing the receiver bandwidth appropriately. Because the clock-recovery circuit did not function properly when using this transmitter, we used the clock directly from the PRBS generator. This technique limited us from taking consistent data after propagating through fiber due to a slow drift in propagation time through the fiber, probably due to thermal variations. However, given the back-to-back performance measured and our simulated results, we expect this transmitter to be able to reach well over 40km with a properly designed clock-recovery circuit.
Table 2. Eye diagrams of the various modulators using a fast photo-detector (14 GHz) at high input power.
2.3 Low bandwidth transmitters and receivers
The previous simulations and experiments used low bandwidth receivers or low bandwidth transmitters. Here, we simulate using both. The transmitter parameters are the same as before, but the receiver is now the 2.5G receiver specified in Table 1(b). Due to the previously noted effects, there is actually a performance gain of several dB when using the 2.5GHz receiver instead of the 7.5GHz receiver.
Figure 8(a) shows the performance of the 2.5Gb/s MZI and 2.5Gb/s APD system as a function of received power for both the compensated and uncompensated cases. As demonstrated in Fig. 1, the performance of the uncompensated link is expected to be very sensitive to small changes in the link parameters, particularly the receiver bandwidth, while the equalized case is much more robust.
The 2.5Gb/s MZI and 2.5GHz APD can reach 120km with a BER <10 -12 at only -26.3dBm. If we assume 3dB allocation for connector losses and 0.2dB/km fiber loss, an MZI with 5dB insertion loss which is coupled to a 20mW laser will reach 120km with an over 4dB power margin. We note that the sensitivity of the 2.5G receiver at 10Gb/s when used with equalization is about 3dB worse than its defined sensitivity at 2.5Gb/s. This is consistent with our experimental results using the low-pass filter.
Fig. 7. (a) Top - Experimental data using OC-48 and OC-192 MZI modulators (b) Bottom - Data for the OC-48 EAM. The OC-192 MZI uncompensated case is shown as a reference.
Fig. 8. The effect of received power on performance (simulated) for transmission at 10Gb/s using (a) Top - 2.5G MZI and 2.5GHz receiver (b) Bottom - 2.5G EAM and 2.5GHz receiver. The unequalized system is inoperable.
The lowest cost transmitter/receiver pair simulated, which uses a 2.5Gb/s EAM and 2.5GHz APD photo-receiver, is depicted in Fig. 8(b). This configuration has a sensitivity of -25.4 dBm for a 80km link. This sensitivity can be easily reached by a 0dBm (modulated) output power transmitter. The power margin, assuming 0.2dB/km fiber loss and 3dB for connector losses, is over 6dB. Such a link would be unusable without electronic compensation (even in a back-to-back condition this case yields a closed eye without compensation).
3. Summary
Table 3:. Summary of penalties for various transmitter and receiver configurations
Table 3 summarizes the performance of various transmitters and receivers considered in this work. We have proposed using electronic equalization techniques in 10Gb/s bit rate communication links in order to relax the bandwidth specifications of modulators and photo-receivers down to levels typically reserved for 2.5Gb/s systems. Our simulations suggest that high quality links with acceptable penalties (for amplified links) or even improvements (for electrical-noise limited links), can be created. We experimentally tested the performance of several types of low-bandwidth transmitters and found that, as predicted, equalization substantially improved the performance. We also experimentally demonstrate that a 2.5GHz receiver bandwidth is adequate for OC-192 links if electronic equalization is employed, but not if FEC techniques are employed. Electronic equalization is therefore shown to have a significant influence on system design. The additional flexibilities and tolerances provided, as well as significantly reduced component costs, make the approach of combining equalization with low-bandwidth components very attractive.
References and links
1. H. Bulow, F. Buchali, W. Baumert, R. Ballentin, and T. Wehren, “PMD mitigation at 10Gbit/s using linear and nonlinear integrated electronic equaliser circuits,” Electron. Lett. 36, 163–164 (2000). [CrossRef]
2. J. C. Cartledge, R. G. McKay, and M. C. Nowell, “Performance of smart lightwave receivers with linear equalization,” J. Lightwave Technol. 10, 1105–1109 (1992). [CrossRef]
3. X. Zhao and F.S Choa, “Demonstration of a 10Gb/s transmissions over a 1.5 km long multimode fiber using equalization techniques,” IEEE Photonics Letters 14, 1187–1189 (2002). [CrossRef]
4. F. Buchali, H Bulow, W. Baumert, R. Ballentin, and T. Wehren, “Reduction of the chromatic dispersion penalty at 10Gbit/s by integerated electronic equalizers,” in OFC 2000 Vol 3 of OSA Proceedings Series (Optical Society of America, Washington, D.C., 2000), pp, 268–270.
5. D. Schlump, B Wedding, and H. Bulow, “Electronic equalization of PMD and chromatic dispersion induced distortion after 100km standard fiber at 10Gb/s,” in 24th European Conference on Optical Communication, Vol 1 of OSA Proceedings Series (Optical Society of America, Washington, D.C., 1998), pp. 535–536.
6. Moe Z. Win, Jack H. Winters, and Giorgio M. Vitetta, “Equalization Techniques for mitigating transmission impairments”, Optical Fiber Telecommunications 1V B, Academic Press2002, 965–997.
7. D. L. Duttweiler, J. E. Mazo, and D. G. Messerschmitt,“Error propagation in Decision-Feedback Equalizers,” IEEE Trans. Inform. Theory ,. IT-20, .490–497, (1974). [CrossRef]
8. W. A. Sethares, I. M. Y. Mareels, B. D. O. Anderson, and C. R. Johnson, “Excitation conditions for signed regressor least mean squares adaptation,” IEEE Trans. Circuits Syst. ,. 35, 613–624 (1988). [CrossRef]
9. A. Shoval, D. A. Johns, and W. M. Snelgrove, “Comparison of dc offset effects in four LMS adaptive algorithm,” IEEE Tran. Cir. Sys. II: An. Dig. Si. Pi ,. 42, (1995).
10. J. G. Proakis and M. Salehi, Communication Systems Engineering (Prentice Hall, 1994), 570–589
