## Abstract

Simulations have been performed to compare the system capacity and power dissipation of NRZ, CAP-64 and 64-QAM-OFDM systems over FEC enhanced POF links using LEDs, for both unidirectional and bidirectional transmission. It is shown that CAP-64 outperforms NRZ and 64-QAM-OFDM in terms of system capacity and supports a record high 3.5Gb/s bidirectional and 2.1Gb/s unidirectional transmissions over 50m POF. The CAP-64 transceiver consumes similar power compared with NRZ whilst the 64-QAM-OFDM transceiver consumes about twice as much.

©2012 Optical Society of America

## 1. Introduction

Gigabit plastic optical fibre (POF) systems have been widely viewed as cost-effective solutions for short distance high speed communications [1, 2]. Recently, 1.25 Gbit/s transmission over 50 m POF using NRZ modulation and forward error correction (FEC) has been reported for LED based POF links [2], this representing the published best performance for NRZ POF systems as far as we are aware.

The main limitation for further improving the system capacity of a POF link system using an LED is the strong channel fading from both the transceiver and the transmission media. A straightforward way to increase capacity is to improve system spectral efficiency, allowing the same bit rate to be transmitted using a reduced bandwidth. Various studies have been carried out to investigate different advanced modulation formats such as pulse amplitude modulation (PAM) [3], optical orthogonal frequency division multiplexing (OOFDM) [4], and carrierless amplitude and phase modulation (CAP) [5] to increase the capacity of POF links. OOFDM and CAP have especially good potential [6], with OOFDM using mature digital signal processing (DSP) which features high spectral efficiency, excellent resistance to fibre dispersion and system flexibility. CAP potentially allows simpler implementation, and hence has the potential of being lower cost whilst still achieving high performance.

In this paper, we investigate and compare the performance of CAP-64 and 64-QAM-OFDM modulation schemes for FEC enhanced POF links using LEDs. The achievable system capacity and transceiver power dissipation of the system are used as the criteria to determine the best scheme for different practical applications. All schemes are compared with NRZ modulation. The paper is organized as follows: Section 2 describes the system architecture of various transceivers and presents the simulation parameters. Section 3 presents the results relating to system capacity and Section 4 analyzes the power dissipation for each transceiver. Finally Section 5 summarizes the paper.

## 2. System architectures

#### 2.1 CAP-64 transceivers

The CAP-64 transceiver is shown in Fig. 1(a) . In the transmitter, the input bit stream is first encoded with FEC and then converted into two parallel streams. In each parallel stream, the bits are mapped onto PAM-8 symbols and then pulse shaping is performed using a square root raised cosine filter [7]. The two shaped signals in the transmitter are combined to drive a MUX module and then the generated optical signal is transmitted over the POF link and is detected by a MUX module in the receiver.

The detected electrical signal in the receiver is processed in two parallel lanes by using feedforward equalization (FFE) and decision feedback equalization (DFE), which provides the dual-functionality of both channel equalization and demodulation for data recovery. The signal processing procedure after DFE in the receiver is the inverse of that in the transmitter.

The MUX module shown in Fig. 1 is dependent on transmission modes and two transmission modes are considered here:

- − Bidirectional transmission. In this case, a transmitter and a receiver co-exist at each end of the link. One way to implement the bidirectional system might be using the MUX module shown in the right inset in Fig. 1(a), which consists of a LED and a PD that is integrated in the module. The MUX module functions as a bidirectional multiplexer, where the LED is driven by the transmitter electrical signal and its output beam is launched into the POF through a mirror hole which has a similar size as the LED facet; while the received light is reflected by the mirror to the PD facet which has a size similar to that of POF (1mm radius). The detected signal is processed in the receiver in the same side. Therefore, the system can work in a half/full duplex manner. Obviously, the MUX module introduces loss due to partial reflection of the received optical beam.
- − Unidirectional transmission. In this case, only a transmitter or a receiver is located in each side and the MUX module is simplified into a sole LED (a sole PD) in the transmitter (receiver) side. As a result, MUX module loss in the bidirectional transmission mode does not exist in the unidirectional case.

#### 2.2 64-QAM-OFDM transceivers

The 64-QAM-OFDM transceivers, as shown in Fig. 1(b), involve dense DSP blocks [8] including symbol mapping using 64-QAM, power loading (PL), inverse fast Fourier transform (IFFT), cyclic prefix insertion, OFDM symbol serialization and DAC. The detailed descriptions of the OFDM modem have been presented in [8]. Note that the IFFT input satisfies Hermitian symmetry so that the output of IFFT is real-valued [8]. Likewise, OFDM is also studied for both unidirectional and bidirectional transmission. For both cases, the detected electrical OFDM signal is finally processed in the receiver OFDM modem with an inverse procedure to that adopted in the transmitter.

Compared with the CAP and OOFDM schemes, the NRZ system is the simplest and is analyzed to provide a reference for comparison. The NRZ architecture involves FEC/FEC^{−1}, LED and driver, the POF link, a PD and FFE/DFE identical to that used in the CAP link.

### 2.2.1 Restriction on power loading for 64-QAM-OFDM system

Variable power loading (PL) is an efficient way for the OOFDM system to compensate for channel fading [8]. Variable PL is performed prior to IFFT, as shown in Fig. 1(b), by multiplying a channel condition-dependent constant *p _{k}* to the

*k-*th subcarrier information symbol

*c*, as described in [8]. The overall frequency domain transfer function of the link including DAC/ADC, LED, POF and optical receiver corresponding to the

_{k}*k*-th subcarrier is denoted as

*H*. On the receiver side, the FFT of the received signal corresponding to the

_{k}*k*-th subcarrier is expressed as

*n*is the additive noise which is dominated by the receiver thermal noise and OFDM signal clipping and quantization caused noise, and

_{k}*r*is the received symbol for the

_{k}*k-*th subcarrier. Suppose the pilot-aided channel estimation for the

*k*-th subcarrier is ${\widehat{H}}_{k}$. According to [9], ${\widehat{H}}_{k}\approx {p}_{k}{H}_{k}$ is feasible by sending a number of pilot symbols and averaging the channel estimations. Therefore, via single-tap equalization, the equalized symbol is given by

It is clearly shown in Eq. (2) that if the subcarrier experiences channel fading (i.e., the normalized overall channel frequency response *p _{k}H_{k}* is less than one), the noise from the receiver is amplified upon single-tap zero-forcing equalization. Namely, noise enhancement with a factor of ${p}_{k}^{2}{H}_{k}^{2}$occurs. We have $SN{R}_{{}^{k}}^{\text{'}}=SN{R}_{k}{p}_{k}^{2}{H}_{k}^{2}$ with $SN{R}_{{}^{k}}^{\text{'}}$ and $SN{R}_{k}$are the signal-to-noise ratio (SNR) after and prior to equalization by taking into account PL. The BER of the OFDM signal using 64-QAM is expressed as [7]

*N*is the IFFT/FFT size. Theoretically speaking, one can choose${p}_{k}=1/{H}_{k}$ so that the noise enhancement factor is equal to one for each subcarrier, as shown in Eq. (3). However, this may not apply to practical implementations when the channel fading is strong. This is because as the total transmitted electrical power is fixed, the extra power assigned to strongly faded subcarriers is subtracted from other subcarriers. Published papers indicate that a few dBe power level difference between subcarriers via PL is practical [1,10]. Therefore, a PL constraint is introduced in this paper: the maximum power level ratio between subcarriers is set to be 12 dBe, namelyConsequently, when the channel response defined from the IFFT in the transmitter to the FFT in the receiver exhibits a roll-off less than 12 dBe, then noise enhancement penalty is zero; otherwise, the noise enhancement penalty is positive and increases as the roll-off grows.

#### 2.3 Simulation parameters

The NRZ, CAP-64 and 64-QAM-OFDM systems all use the same optical transceiver parameters as shown in Table 1
. Note that based on the reference receiver, a reference 300 Mb/s NRZ system has a sensitivity of −28 dBm at BER of 10^{−3} (and a sensitivity of −25 dBm at BER of 10^{−9}). The LED launch power is set to be 0 dBm, apart from the case where the POF length is 100 m, where it is assumed that several LEDs are used to achieve 3 dBm launch power. The total link power budget is 28dB (31dB for POF length of 100m) by using FEC(10^{−3},10^{−12}) where the BER threshold is 10^{−3}. There are also modulation scheme-dependent parameters: for CAP-64, two parallel decorrelated 2^{9}-1 PRBS data streams are used and the two square root raised cosine shaping filters have a roll-off coefficient of 0.5. For NRZ and CAP-64, the receiver equalizer contains a 20 tap T/4 spaced FFE and a 3 tap DFE. For 64-QAM-OFDM, an IFFT/FFT size of 64 and an adaptive cyclic prefix of ~12.5% are used, with 31 parallel decorrelated 2^{9}-1 PRBS streams being used as user data. The spectral efficiency for NRZ, CAP-64 and 64-QAM-OFDM are 1 b/s/Hz, 4 b/s/Hz and 5.3 b/s/Hz, respectively.

## 3. Results for transmission capacity

Figure 2
shows a plot of the achievable maximum system capacity versus POF length for all the systems shown in Fig. 1 using FEC (10^{−3},10^{−12}). The maximum bit rate is defined as the bit rate obtained when certain power margin (or unallocated penalty) is satisfied, and for bidirectional transmission mode, the bit rate refers to aggregated bit rate covering those in both directions. As shown in the inset of Fig. 2, the link power penalties comprise a noise enhancement upon equalization, a residual ISI due to imperfect equalization, the link loss and multilevel coding (and clipping and quantization (C&Q) for the OFDM system) induced penalty. The red bars are unallocated penalties, representing power margin, whose values are chosen to be + 1dBo to obtain the maximum bit rate.

Based on the power penalty analyses, Fig. 2 shows the system bit rates allowed for the systems shown in Fig. 1. For all the modulation schemes, the bit rate decreases with increasing POF length, simply due to increased fiber dispersion. For the same POF length, the bidirectional transmission supports a nearly doubled system bit rate compared with the unidirectional case. Taking the CAP-64 system for example, this supports 3.5 Gb/s and 2.1 Gb/s over 50 m POF for bidirectional and the unidirectional transmission, respectively. The reason for the former bit rate being less than twice of the latter one is due to the bidirectional multiplexer loss (3 dB assumed here). The unidirectional 2.1 Gb/s transmission over 50 m POF achieved by CAP-64 represents an improvement of 70% in capacity compared with the published best performance of 1.25 Gb/s NRZ signal over 50m POF unidirectional transmission [2]. As shown in Fig. 2, it is interesting to note that CAP-64 outperforms both NRZ and 64-QAM-OFDM for all POF lengths. The superiority of CAP over the other two schemes is mainly attributed to three physical mechanisms: First, NRZ unavoidably exhibits larger noise upon equalization due to its large signal bandwidth, though it has no multilevel penalty. Second, the C&Q causes significant extra penalty in the OFDM system compared with the CAP. Third, for the OFDM system, an increasing system bit rate requires larger DAC/ADC sampling rates, leading to stronger channel fading for OFDM subcarriers located in high frequency bands. As a result, it can be seen that the noise enhancement using zero forcing one-tap equalization becomes significant according to Eqs. (3)-(4). Because of these reasons, it is interesting to note that NRZ and 64-QAM-OFDM have similar performance.

## 4. System power dissipation estimates and comparisons

In this section, power dissipation estimates are made for the various system transceivers. A reference unidirectional Gigabit NRZ system is considered and its power dissipation is estimated based on typical power consumption of its constituent components measured in lab: 60mW for laser driver and LED, 50mW for photodiode and TIA amplifier, 70mW for CDR and 20mW for control circuits. The power consumption for FEC/FEC^{−1} is based on [11] where ~2mW at 80 MHz clock is reported in using an FPGA implementation, giving rise to 50mW here. The total power dissipation of the NRZ link is 250mW. For the bidirectional case, the total transceiver power consumption is simply doubled since it consists of two unidirectional NRZ transceivers.

For Gigabit unidirectional CAP-64 and 64-QAM-OFDM systems, their power consumption can be obtained by considering the power dissipation of the reference NRZ system together with that of other extra components required. As shown in Fig. 1(a), for CAP-64 the power dissipation includes 50mW for PAM-8 symbol map/de-map [12], 40mW for two transversal shaping filters which can be implemented by two *M*-tap (M>1) FFE equalizers, 70mW for receiver equalizers (20 taps FFE and 3 taps DFE), and 20mW for P/S and S/P convertors, resulting in a total power consumption of 380mW using the single POF link configuration. It is easy to show that CAP-64 based on bidirectional configuration needs ~760mW power consumption in total. For OOFDM incorporating 64-QAM, the main extra components are the OFDM modems in transceiver, as shown in Fig. 1(b). Based on 65nm CMOS technologies, the power dissipation for the OFDM modem including 64-QAM mapping/de-mapping and IFFT/FFT together with equalization is roughly ~500mW [13] at 25Gb/s bit rate, equivalent to ~20mW/Gb/s. The DAC/ADC operating at ~1GS/s requires about 100mW in total [14]. Thus the OFDM modem shown in Fig. 1(b) requires power consumptions of 220mW and 440mW for unidirectional and bidirectional cases, leading to total transceiver power consumptions 420mW and 840mW, respectively.

For transmission over 50m POF and considering the bit rate each system supports shown in Fig. 2, the power dissipations (defined as per Gb/s) estimated for various systems are normalized to that of reference NRZ system and plotted in Fig. 3 . For all modulation schemes, the power dissipation for the unidirectional system is lower than that for the bidirectional case. This is because the bidirectional multiplexer shown in Fig. 1 introduces loss leading to reduction in achievable bit rate. Figure 3 also shows clearly that CAP-64 offers similar power dissipation compared with that of NRZ system, for both unidirectional and bidirectional cases; while 64-QAM-OFDM doubles the power dissipations compared to that of NRZ for both unidirectional and bidirectional configurations.

## 5. Conclusions

Theoretical comparisons have been made on the system capacity and power dissipation of NRZ, CAP-64 and 64-QAM-OFDM POF links using LEDs with FEC for both unidirectional and bidirectional transmission. Results showed that CAP-64 outperforms NRZ and 64-QAM-OFDM and supports record high 3.5Gb/s bidirectional and 2.1Gb/s unidirectional transmission over 50m POF, with the latter bit rate representing a 70% improvement in capacity compared to the best published NRZ POF system performance to date. We also showed that CAP-64 consumes similar transceiver power compared with NRZ modulation whilst 64-QAM-OFDM consumes double that. This indicates that CAP-64 modulation offers great potential in terms of signal capacity and power efficiency for LED based POF links.

## Acknowledgments

This work was supported by EPSRC via the INTERNET project.

## References and links

**1. **B. Charbonnier, P. Urvoas, M. Ouzzif, J. L. Masson, J. D. Lambkin, Mo. O’Gorman, and R. Gaudino, “EU project POF-PLUS: Gigabit transmission over 50m of step-index plastic optical fibre for home networking,” OFC/NFOEC09, Paper OWR4.

**2. **C. Zerna, J. Sundermeyer, A. Fiederer, N. Verwaal, B. Offenbeck, and N. Weber, “Integrated PAM2 decision feedback equalizer for Gigabit Ethernet over standard SI-POF using red LED,” ECOC 2010, Paper We.6.B.4.

**3. **F. Breyer, S. C. J. Lee, S. Randel, and N. Hanik, “PAM-4 Signalling for Gigabit transmission over standard step-index plastic optical fibre using light emitting diodes,” ECOC08, Paper We.2.A.3.

**4. **S. C. J. Lee, F. Breyer, D. Cárdenas, S. Randel, and T. Koonen, “Real-time implementation of a 1.25Gb/s DMT transmitter for robust and low-cost LED-based plastic optical fiber applications,” ECOC2009, Paper 3.5.4.

**5. **L. Geng, R. V. Penty, I. H. White, and D. G. Cunningham, “FEC-free 50 m 1.5 Gb/s plastic optical fibre link using CAP modulation for home networks,” ECOC 2012, Paper Th.1.B.4.

**6. **J. L. Wei, J. D. Ingham, D. G. Cunningham, R. V. Penty, and I. H. White, “Comparisons between 28 Gb/s NRZ, PAM, CAP and optical OFDM systems for Datacommunication Applications,” IEEE OI 2012, Paper MA2.

**7. **J. J. Werner, *Tutorial on carrierless AM/PM* (ANSI X3T9.5 TP/PMD Working Group, 1992 & 1993).

**8. **J. L. Wei, C. Sánchez, R. P. Giddings, E. Hugues-Salas, and J. M. Tang, “Significant improvements in optical power budgets of real-time optical OFDM PON systems,” Opt. Express **18**(20), 20732–20745 (2010). [CrossRef] [PubMed]

**9. **W. Shieh and I. Djordjevic, *Orthogonal Frequency Division Multiplexing for Optical Communications* (Elsevier, 2010).

**10. **X. Q. Jin, J. L. Wei, R. P. Giddings, S. Walker, and J. M. Tang, “Experimental demonstrations and extensive comparisons of end-to-end real-time optical OFDM transceivers with adaptive bit and/or power loading,” IEEE Photonics J. **3**(3), 500–511 (2011). [CrossRef]

**11. **L. Biard and D. Noguet, “Reed-Solomon codes for low power communications,” J. Commun. **3**, 13–21 (2008).

**12. **J. Lee, M.-S. Chen, and H.-D. Wang, “Design and comparison of three 20Gb/s backplane transceivers for Duobinary, PAM4, and NRZ data,” IEEE J. Solid-state Circuits **43**(9), 2120–2133 (2008). [CrossRef]

**13. **P. A. Milder, R. Bouziane, R. Koutsoyannis, C. R. Berger, Y. Benlachtar, R. I. Killey, M. Glick, and J. C. Hoe, “Design and simulation of 25 Gb/s optical OFDM transceiver ASICs,” Opt. Express **19**(26), B337–B342 (2011). [CrossRef] [PubMed]

**14. **E. Alpman, H. Lakdawala, L. R. Carley, and K. Soumyanath, “A 1.1V 50mW 2.5GS/s 7b time-interleaved C-2C SAR ADC in 45nm LP digital CMOS,” ISSCC09, 76–78, (2009).