1. Introduction

The flourishing development of mobile networks and Internet of thing, such as social media, 5G mobile front haul and cloud computing, has pushed the capacity demands of data center from 100Gb/s to 400 Gb/s or even higher. It is widely confirmed that 4 × 100 Gb/s transmission scheme is an attractive way to realize 400 Gb/s or above signal transmission, meaning that 100Gb/s per wavelength transmission system is essential [1]. However, traditional electrical interconnect based on coaxial cable cannot satisfy 100 Gb/s or 400 Gb/s capacity demands. Alternatively, optical interconnect can realize ultra-high-speed signal transmission with lower cost and power consumption [2,3]. Intensity modulation with direct detection (IM/DD) is considered as a more attractive solution for short reach applications, such as access networks and optical intra-data center interconnect (Intra-DCI), due to its lower system cost and complexity [1–11]. To achieve higher transmission rate and better performance in IM/DD system, some linear and nonlinear equalization digital signal processing (DSP) algorithms are extensively studied, such as feed-forward with decision-feedback equalizer [9,12] and volterra equalizer [13,14]. In addition, some high spectral efficiency advanced modulation formants are applied, such as discrete multi-tone (DMT) [5–8], pulse amplitude modulation (PAM) [9,10] and carrier-less amplitude/phase modulation (CAP) [11]. Among them, DMT is considered as more effective technology due to its high spectral efficiency and high tolerance to chromatic dispersion (CD) and polarization mode dispersion (PMD) [15,16]. However, system bandwidth limitation induced high frequency edge SNR serious degradation is a common problem in high-speed communication system. The most straightforward and effective solution to deal with this SNR degradation is employing pre-equalization in transmitter. Unfortunately, the peak to average power ratio (PAPR) of pre-equalized DMT signal becomes even higher than conventional DMT signal. In this paper, discrete-Fourier-transform spread (DFT-spread) technique [4,15,16] is applied to reduce the signal PAPR and improve the bit-error-rate (BER) performance of DMT system, simultaneously.

To realize 100 Gb/s and beyond signal transmission with single optical carrier, the digital-to-analog converter (DAC) and analog-to-digital converter (ADC) with high sampling rate are widely employed. While some signal distortions such as clock-leakage can always be observed in high-speed DAC and ADC [9]. In single carrier signal transmission, a 2-tap least-mean-square (LMS) based Adaptive Notch Filter (ANF) is proposed to suppress the DAC clock leakage induced narrowband interference in our previous work [9], and 1.3 dB receiver sensitivity improvement is observed in 112 Gb/s PAM4 transmission system [9]. In multicarrier signal transmission, such as OFDM and DMT system, clock leakage can be effectively avoided by null-subcarrier reservation (NSR) in transmitter, in which the subcarriers affected by the leaked clock tone are left empty. However, in order to maintain the signal speed, subcarriers in higher frequency edge with poor SNR performance have to be utilized for data transmission which will degrade the system performance. It is obvious that the NSR scheme is not the best optional for DMT system to suppress the DAC clock leakage, so the ANF scheme is employed and compared with NSR scheme for clock leakage cancellation in this manuscript.

In this paper, 120 Gb/s DMT signal transmission and reception with narrowband interference cancellation for 2-km Intra-DCI is experimentally demonstrated. Digital pre-equalization and DFT-spread technique are jointly applied in transmitter to deal with the high frequency power fading and high PAPR of DMT signal. DMT signal transmission with NSR and LMS based ANF for narrowband interference cancellation are compared in this paper. The experimental results show that 1.2 dB and 1.8 dB receiver sensitivity improvements can be obtained for DMT signal in optical back-to-back (OBTB) transmission at the BER of 3.8 × 10⁻³ with optimized NSR and ANF for narrowband interference suppression, respectively. In addition, the required received optical power (ROP) of DMT signal with optimized NSR and ANF for narrowband interference cancellation at BER of 3.8 × 10⁻³ over 2-km single mode fiber (SMF) transmission are −6.5 dBm and −7.1 dBm, respectively. It is obvious that ANF scheme shows better performance for DAC clock leakage induced narrowband interference cancellation in DMT transmission system.

2. Experimental setup

The schematic of the experimental setup and the DSP blocks for 30-GHz DFT-spread DMT signal generation and reception are shown in Fig. 1(a). The 16-QAM DFT-spread DMT signal is generated off-line in MATLAB with pseudorandom binary sequence (PRBS). After that, additional 750-point DFT is applied to reduce the PAPR of DMT signal and improve the system BER performance. Then 5 deterministic DMT symbols are inserted as training sequences (TSs) for receiver synchronization and post equalization, and zero-forcing (ZF) based pre-equalization is implemented to reduce the bandwidth limitation induced high frequency power fading. After complex conjugation, signal transformation from frequency to time domain is realized by 2150-point inverse discrete Fourier transform (IDFT), and 32-point cyclic prefix (CP) is added for CD tolerance enhancement. The generated DMT signal is uploaded into a Fujitsu DAC with 86-GSa/s sampling rate and 16.7 GHz 3-dB analog bandwidth. Then the output signal from DAC is amplified by a single-ended 30 GHz driver with 20-dB gain. An Anritsu bias-tee G4N37 is used to optimize the bias of the electro-absorption modulated laser (EML) at −1.25 V according to the measured P-V curve of EML shown in Fig. 1(b). The DC-coupled DMT signal from bias-tee is modulated on an optical carrier at 1538.95 nm by the EML with 40 GHz 3-dB bandwidth, the output signal from EML is fed into single mode fiber (SMF). The optical spectra of generated DFT-spread DMT signal with and without pre-equalization are shown in Fig. 1(c) with red and black line, respectively. It is clear that the high frequency power fading is compensated by employing pre-equalization in transmitter. At the receiver, a variable optical attenuator (VOA) is applied to adjust the ROP and a Finisar photodiode (PD) MPRV1332A with 40 GHz 3-dB bandwidth is used to realize signal optical-to-electrical (O/E) conversion. After that, the detected signal is sampled by a Lecroy 80-GSa/s real time oscilloscope (OSC) and processed off-line in MATLAB. The off-line DSPs include synchronization, optional ANF based narrowband interference cancellation, CP removing, 2150-point DFT, addition 750-point IDFT, ZF-based post equalization and 16-QAM signal de-mapping. At last, the BER of DMT signal is calculated by simple direct error counting with 110 DMT symbols (750 × 110 × 4 = 330000 bits).

 

Fig. 1 (a) Experimental setup and DSP blocks for DFT-spread DMT system; (b) P-V curve of EML; (c) Optical spectra of DFT-spread DMT with and without pre-equalization.

Download Full Size | PDF

Figure 2 shows the data-carried subcarriers distribution of conventional DFT-spread DMT and DFT-spread DMT with NSR in transmitter. For conventional DFT-spread DMT, the effective symbols are carried on the subcarriers from X(1) to X(M) as shown in Fig. 2(a), and Hermitian Symmetry (HS) is performed to create real value DMT signal, then 2N-point (N > M) IFFT is employed to generate DMT symbol. For DFT-spread DMT with NSR scheme, several interfered subcarriers are set to null in transmitter as shown in Fig. 2(b). In order to find out the exactly position of those interfered subcarriers, TSs-based channel estimation and SNR calculation are applied in calibration stage. To obtain more accurate results, quadrature phase shift keying (QPSK) symbols are uploaded on the subcarriers of TSs. For SNR calculation, we first calculate the error vector magnitude (EVM) between ideal QPSK signal and the recovered TSs symbols, then the SNR = 1/EVM² is obtained. After obtaining the SNR distribution across the subcarriers, the interfered subcarriers are nulled to avoid the clock leakage in the transmitter. For NSR scheme as shown in Fig. 2(b), the effective symbols are uploaded on the subcarrier from X(1) to X(M + n), where n is the number of reserved null-subcarriers for avoiding the DAC clock leakage. It is clear that addition subcarriers from X(M + 1) to X(M + n) in higher frequency edge are used to carry data symbols to compensate the n-point NSR induced data rate reduction in NSR scheme. At last, the HS and IFFT are employed to realize DMT modulation.

 

Fig. 2 Generation structure of (a) conventional DFT-spread DMT and (b) DFT-spread DMT with NSR.

Download Full Size | PDF

3. Experimental results and discussion

In the calibration stage, DMT signal without DFT-spread processing is used for SNR calculation, as the calculated SNR of each subcarrier will turn to the same after additional DFT-spread operation. However, during the signal transmission stage, DFT-spread technique is employed to reduce the PAPR of DMT signal. The calculated SNR of conventional DMT signal without and with ANF in receiver are shown in Figs. 3(a) and 3(b), respectively. It can be seen that the SNR of clock leakage interfered subcarriers suffers seriously degradation in Fig. 3(a), and this phenomenon disappears in Fig. 3(b) in which ANF is applied to suppress the clock leakage. The calculated SNR of DMT signal with 10 and 15 NSR around the clock-leakage interfered subcarriers are shown in Figs. 3(c) and 3(d), respectively. The clock leakage induced interference can be totally suppressed if the number of reserved null-subcarriers is large enough. Figure 3(c) represents the scenario of the number of reserved null-subcarriers is not enough to avoid the clock-leakage, and the residual interference will still lead to SNR degradation seriously. From Fig. 3(d), we can see that the clock leakage induced interference is almost totally suppressed, while the SNR in high frequency edge is a little lower than that in Figs. 3(a) and 3(b). The main reason is that the additional subcarriers in higher frequency are uploaded signal to compensate the NSR caused transmission rate reduction as mentioned above. In fact, we have optimized the number of reserved null-subcarriers from one to twenty, and find the 15 NSR scheme has the best SNR performance. In the latter experiments, the number of null-subcarriers is optimized to be 15.

 

Fig. 3 Calculated SNR of (a) conventional DMT; (b) DMT with ANF in receiver; (c) DMT with 10 null-subcarriers; (d) DMT with 15 null-subcarriers in transmitter.

Download Full Size | PDF

The electrical spectra of received DFT-spread DMT signals are given out in Fig. 4. It can be seen that the bandwidth limitation induced high frequency power fading is fully compensated after pre-equalization. Figures 4(a) and 4(b) show the electrical spectrums of conventional DFT-spread DMT signal without and with ANF in receiver, respectively. It is obvious that the DAC clock-leakage is suppressed by ANF, and the bandwidth of ANF is optimized to 8.6 MHz as the principle described in [9]. The electrical spectrum of received DFT-spread DMT signal with 15 NSR in transmitter is given in Fig. 4(c), and the DAC clock leakage is successfully avoided by the NSR area.

 

Fig. 4 Electrical spectra of DFT-spread DMT signal; (a) with nothing and (b) with ANF in receiver side and (c) with 15 null-subcarriers in transmitter side.

Download Full Size | PDF

BER performance of 120 Gb/s DMT signal versus ROP is measured and given out in Fig. 5. In order to avoid the clock leakage induced system performance degradation, ANF, NSR and Chow Cioffi Bingham (CCB) algorithm-based bit-loading [17] schemes are experimentally discussed. Figure 5(a) gives out the specific bit allocation scheme according to the calculated SNR. In OBTB transmission, 1.2 dB and 1.8 dB receiver sensitivity improvements are obtained at BER of 3.8 × 10⁻³ as shown in Fig. 5(b) when optimized NSR and ANF schemes are applied to suppress the clock leakage, respectively. Figure 5(c) gives out the results of DMT signal over 1-km and 2-km SMF transmission. The required ROPs are −6.5 dBm and −7.1dBm at BER of 3.8 × 10⁻³ over 2-km SMF transmission assisted by NSR and ANF schemes, respectively. The constellations of received DFT-spread DMT signal with ANF scheme, with 15 NSR scheme and with no operations in OBTB transmission at −4 dBm ROP are also inserted as insets (i), (ii) and (iii) of Fig. 5(b), respectively. Furthermore, as the results shown in Figs. 5(b) and 5(c), it is obvious that the performance of CCB algorithm-based bit-loading DMT is worse than DFT-spread DMT with ANF and similar to DFT-spread DMT with NSR.

 

Fig. 5 (a) Bit-loading scheme according to SNR; (b) measured BER of 120Gb/s DMT signal versus received optical power in OBTB transmission and (c) over SMF transmission.

Download Full Size | PDF

4. Conclusion

In this paper, 120 Gb/s DMT signal transmission and reception in IM/DD system with narrowband interference cancellation are experimentally demonstrated for 2-km Intra-DCI. Digital pre-equalization and DFT-spread techniques are jointly applied in transmitter to deal with the high frequency power fading and high PAPR of DMT signal. NSR and LMS-based ANF schemes are compared for the DAC clock leakage cancellation. 1.2 dB and 1.8 dB receiver sensitivity improvements can be obtained at BER of 3.8 × 10⁻³ in OBTB transmission when optimized NSR and ANF schemes are applied to suppress the clock leakage. As well as over 2-km SMF transmission, the required ROPs are −6.5 dBm and −7.1 dBm at BER of 3.8 × 10⁻³ assisted by NSR and ANF schemes to avoid the clock leakage induced narrowband interference, respectively. It is the first time to find the ANF outperforms NSR scheme for clock leakage induced interference cancellation in DMT system.