1. Introduction

With the development of high speed digital-to-analog converters (DACs) and analog-to-digital converters (ADCs), coherent detection optical orthogonal frequency-division multiplexing (CO-OFDM) has attracted a great deal of interest for optical transmission systems beyond 100 Gb/s or even 400 Gb/s [1–10 ]. The straightest way to increase spectrum effeiciency (SE) is to use high order modulation formats [8–10 ]. Utilization of higher order modulation formats results in higher optical signal-to-noise ratio (OSNR) requirement and an increase of implementation complexity in the transceiver. Therefore, transmission distance drops rapidly with high order modulation formats. Taking these factors into consideration, 16-ary quadrature-amplitude (16-QAM) modulation format is regarded as a practical candidate for the next generation 100G-beyond or 400G backbone optical networks [6, 7, 11 ]. Until now, the highest single band/λ OFDM transmission reported in [5] is 453.2 Gb/s/λ with compilicated bit and power loading technique. However, after excluding the proposed 47% FEC overhead, single band/λ OFDM net rate is far less than 400G.

For single band/λ 400G OFDM signal transmission, the total bandwidth (BW) of OFDM signal is quite large even with polarization-division multiplexing (PDM). The SNRs of high frequencies bins of OFDM signal are severely degraded. Pre-equalization can be applied to compensate for system components BW limitation induced high frequencies bins power attenuation [12, 13 ]. High peak-to-average power ratio (PAPR) is another drawback of CO-OFDM [1, 7 ]. Particularly, high PAPR is possible to cause serious nonlinear distortions in fiber link. Thus the PAPR of OFDM should be suppressed to extend transmission distance. In DFT-spread OFDM, each data symbol is carried by all in-band subcarriers, such single-carrier-like characteristic of DFT-spread OFDM signal exhibits lower PAPR and better power attenuation tolerance, while DFT-spread OFDM is a special case of OFDM as data symbols are still carried by orthogonal subcarriers. In our previous work regarding 256 Gb/s PDM-16QAM-OFDM signal transmission, DFT-spread OFDM is proved to outperform pre-equalized conventional OFDM for broad BW optical signal transmission [7].

To future improving large BW DFT-spread OFDM signal transmission performance, pre-equalization for DFT-spread OFDM signal transmission is realized and discussed in Ref [14]. In this paper we cover in detail for the combinations of DFT-spread and pre-equalizations technique, especially we discuss how to obtain accurate channel response for DFT-spread OFDM signal transmission. We discuss optimization of the size of IFFT/FFT for OFDM signal modulation/demodulation. Large size IFFT/FFT in OFDM signal transmission can improve the frequency resoulations during channel estimation, this will help to improve system receiver sensitivity. While lage size IFFT/FFT in OFDM signal transmission will also lead some disadvantages. First, large IFFT/FFT size OFDM is more sensitive to phase noise and frequency offset induced inter carrier interference (ICI), this will lead to the BER performance degradation. Second, the compution complexity will increase siginificantly when the size of IFFT/FFT increases. Taking both computation complexity and BER performance into consideration, there should be a optimized IFFT/FFT size for OFDM signal transmisson. During the experiment test, we found that the optimized IFFT/FFT size in this paper is 1024. From the following experimental results we can see that the performance of 65 GHz DFT-spread 16QAM-OFDM can be significantly improved after pre-equalization. In this paper, transmission of single band/λ 400 G net rate OFDM enabled by combination of DFT-spread technique and pre-equalization is achieved. We successfully transmit 8 × 520 Gb/s PDM-16QAM-OFDM signal generated from 80 GSa/s sampling rate DAC over 840 km SMF with the BER under soft decision forward error correction (SD-FEC) thershold (2.4 × 10⁻²) [15].

2. Principle of pre-equalization for DFT-spread OFDM

Principle diagram of 65 GHz DFT-spread 16QAM-OFDM with pre-equalization is shown in Fig. 1 . 65 GHz DFT-spread 16QAM-OFDM signal is generated by an 80 GSa/s DAC with 18-GHz 3-dB BW. In the transmitter, PRBS is first mapped to 16QAM, and then an optional M-point FFT is implemented in the transmitter and M is half of IFFT/FFT size N for OFDM signal modulation/demodulation, the outputs of FFT are mapped into the low frequencies sub-carriers of N-point IFFT input excluding 5 sub-carriers around DC, these null subcarriers are reserved for RF-pilot insertion for frequency offset and phase noise estimation [6, 7, 16 ], after cyclic prefix and TSs are inserted, down-sampling is done to generate 65 GHz 16QAM-OFDM signal, at last off-line OFDM symbols are uploaded into DAC for signal generation. Two parallel arms of I/Q modulator biased at null point are modulated by I/Q components of 65 GHz 16QAM-OFDM signal boosted by a pair of drivers. A self-homodyne detection (SHD) like structure [7, 12 ] is applied to get transceiver’s response. One laser is used as both the transmitter light source and optical local oscillator (LO) in the SHD system and no frequency offset exists. The linewidth of the laser should not be very large during transceiver’s response for pre-equalization to reduce the phase noise and a typical 100-KHz linewidth laser in coherent optical communication is used in transceiver’s response acquisition stage. In this case, frequency synchronization is not necessary during transceiver’s response acquisition. In the receiver, resampling, time synchronization, N-point FFT and M-point IFFT are implemented before we acquire the transceiver’s response. Training Sequences (TSs) can be generated with/without additional IFFT/FFT for DFT-spread during channel response acquisition. Without additional IFFT/FFT for DFT-spread, the samples in the TSs for transceiver’s response estimation are discrete vector symbols, while the samples in the TSs with additional IFFT/FFT for DFT-spread are analog samples [17]. From this point of view, TSs without additional IFFT/FFT for DFT-spread should show better performance in acquiring accurate transceiver’s response. When the IFFT/FFT size N for OFDM signal modulation/demodulation is set at 1024, the amplitude and phase of transceiver’s response estimated with TS without additional IFFT/FFT for DFT-spread are given in Figs. 1(a) and 1(b), respectively, while the amplitude and phase of transceiver’s response estimated with TS with additional IFFT/FFT for DFT-spread are shown in Figs. 1(c) and 1(d), respectively. Apparent fluctuations are observed in both amplitude and phase of transceiver’s response acquired by TS with additional IFFT/FFT for DFT-spread, and the fluctuations disappear in the transceiver’s response estimated with conventional TSs without additional IFFT/FFT. While in optical link, the transceiver’s response should be very smooth and intensive fluctuations should not appear, thus TS without additional IFFT/FFT for DFT-spread should be a better option to acquire transceiver’s response. Repeated averaging method (RAM) based channel estimation with TSs is used to obtain reliable transceiver’s response. In the RAM, all the transmitted conventional OFDM symbols can be used as TSs in the transmitter. The number of transmitted TSs in the RAM is 127 and the modulation format is QPSK. During estimation, TSs are first transmitted to estimate the transceiver’s response and then these estimated samples are averaged to suppress the random noise. The obtained pre-equalization coefficients is shown in Fig. 1(e).The pre-equalization process is implemented for data sub-carriers in the frequency domain after the transceiver’s response is acquired. Since the transceiver’s response acquisition is implemented in SHD based on the commonly-used 1-tap frequency domain equalizers in regular coherent receiver-side DSP, the proposed method in this paper without additional DSP is promising for system implementation.

 

Fig. 1 Principle diagram of pre-equalization for 65 GHz DFT-spread 16QAM-OFDM. Inset: (a) amplitude and (b) phase of transceiver’s response estimated with TS without additional IFFT/FFT for DFT-spread, (c) amplitude and (d) phase of transceiver’s response estimated with TS with additional IFFT/FFT for DFT-spread, (e) pre-equalization coefficients.

Download Full Size | PDF

3. Experimental setup

The experimental setup of dual-polarization 8 × 520 Gb/s 16QAM-OFDM transmission and coherent detection on a 75-GHz wavelength-division multiplexing (WDM) grid is shown in Fig. 2 . In the experimental intial testing stage, the IFFT/FFT size for OFDM signal modulation/demodulation is fixed at 1024. At the transmitter, eight 75-GHz frequcny spacing external cavity lasers (ECLs) with linewidth less than 100 kHz are divided into odd and even groups. The odd and even group of ECLs are first combined by two independent sets of polarization maintaining optical coupler, and then modulated by two I/Q modulators with 27-GHz 3-dB BW, respectively. During 65 GHz 16QAM-OFDM signal generation, a CP of 8 samps is used to resist PMD and residual chromatic dispersion (CD) caused by drift of center-wavelength and fiber length minor deviation during electrical dispersion compensation (EDC) [18]. Two 25 GHz 3-dB BW RF drivers are used to boost I and Q components of OFDM signal. The polarization multiplexing is realized by a polarization multiplexer [7], the delay between two polarizations is controlled to be an exact OFDM symbol time period (1/80 × (1024 + 8) × 8/13 ns = 7.94 ns) to genenrate a pair of time-interleaved TSs for de-multiplexing. One pair of TSs is inserted before every 125 16QAM-OFDM symbols. The pay-load bit rate of OFDM signal is 65 Gb/s × 1024/1032 × 125/127 × 4 × 2 = 507.84 Gb/s after excluding overheads. The generated odd-channel and even-channel PDM-16QAM-OFDM optical signals are combined by a 3-dB optical coupler and jointly boosted by an erbium-doped fiber amplifiers (EDFA). The generated 8-channel WDM signal is then launched into a re-circulating fiber loop, which consists of five spans of 84-km SMF-28. Each span has 18-dB average loss and 17-ps/km/nm chromatic dispersion at 1550 nm without optical dispersion compensation. An EDFA is used before each span to compensate for the fiber loss. After 5 × 84-km SMF-28 transmission, the optical signal passes through a programmable wavelength selective switching (WSS) to remove out-band amplified spontaneous emission (ASE) noise. An EDFA is used after the WSS to compensate for the switch loss in the loop. All the EDFAs used in the experimental setup has 5 dB noise figure. At the receiver, a tunable optical filter is used to select the desired channel. An ECL with linewidth less than 100 kHz is used as a LO for the selected channel. O/E detection of the signal is implemented with an integrated coherent receiver. The analog-to-digital conversion is realized in the real-time oscilloscope with 160-GSa/s sampling rate and 65-GHz BW. The captured data is then processed with offline DSP shown in Fig. 2. Two parameters of this RF pilot-tone based phase noise compensation scheme have a major effect on the receiver performance, including the BW of the electrical band pass filter (BPF) for RF pilot-tone selection and the pilot-to-signal ratio (PSR). For the 100-kHz linewidth ECLs used in this experiment, the optimal BW of BPF is 10 MHz [19]. The PSR is defined as PSR [dB] = 10 × log(P_RF / P_OFDM) where P_RF and P_OFDM are the power of RF pilot tone and OFDM symbol respectively. In the high PDR region, the relative OFDM signal power is too low and thus affects the recovery of the OFDM signal. On the other hand, in the low PSR region, the ASE noise in the RF pilot reduces the efficiency of the phase noise compensation. The optimum PSR is −13 dB in this experiment. Conventional OFDM symbols TSs without additional IFFT/FFT are transmitted for channel estimation and equalization. Intra-symbol frequency-domain averaging base channel estimation is applied in this experiment [7]. In this experiment, the BER is counted over 10 × 512000 bits (10 data frames, and each frame contains 512000 bits).

 

Fig. 2 Experimental setup. (ECL: external cavity laser; DAC: digital to analog convertor; PM-OC: polarization maintaining optical coupler; Pol. MUX: polarization multiplexer; OC: optical coupler; ATT: attenuator; TOF: tunable optical filter; ADC: analog to digital convertor; SW: switch). Insets: (a) optical spectra (0.1 nm) of 65 GHz 16QAM-OFDM without and with pre-equalization, 65GHz 16QAM-OFDM electrical spectra: (b) without and (c) with pre-equalization.

Download Full Size | PDF

4. Experimental results

The BERs versus OSNR of single polarization 65-GHz DFT-spread 16QAM-OFDM with 1024 IFFT/FFT size in OBTB without and with pre-equalization are measured and the results are shown in Fig. 3(a) . The pre-equaliztion procedure is implemented as described above. After pre-equalization, the BER performance has been improved significantly. More than 5-dB receiver sensitivity is achivevd after pre-equalization is applied at BER of 1.0 × 10⁻². 65-GHz 16QAM-OFDM optical spectrums (0.1 nm) without and with pre-equalization are inserted in Fig. 2(a). A little over-compensation is oberved in the high frequency after pre-equalization and this over-compensation just counteracts the high frequency attenuation of integrated coherent receiver. The corresponding electrical spectrums of received OFDM samples taken from the real-time oscilloscope are inserted in Figs. 2(b) and 2(c), respectively. As we can see the high frequency power attunations have been totally compensated after pre-equalization. The constellations of OFDM without and with pre-equalization with OSNR at 29.8 dB are inserted as inset (i) and inset (ii) of Fig. 3(a), respectively. The constellations of OFDM with pre-equalization become more concentrated. BER versus different IFFT/FFT sizes for single polarization 65 GHz DFT-spread 16QAM-OFDM signal with OSNRs at 19.1, 24.4 and 29.8-dB are also measured in the paper and the results are shown in Fig. 3(b), all the signals with different IFFT/FFT sizes are pre-equalized according to the principle described in section 2. From the experimental results we can see that the BER performance is constinously improved with the increase of IFFT/FFT size from 256 to 1024 at different OSNRs. The residual ICI after RF-pilot based phase noise and frequency offset compensation is very small, at this stage OFDM signal is not very sensitive to this residual ICI, while increasing IFFT/FFT to improve frequency resolution is an effective way to overcome ASE noise during channel estimation, thus BER performance improvement is achieved. No evident BER performance improvement is seen when the length of IFFT/FFT size increase from 1024 to 4096, in this stage OFDM signal becomes more and more sensitive to residual ICI with the increase of IFFT/FFT size. BER performance degradation arises when the IFFT/FFT size is set at 8192, ICI induced BER performance degradtion even overcomes the frequency resoulation improvement at this point. As the computional complexities of small IFFT/FFT size OFDM signal are lower, taking both computation complexity and BER performance into consideration, 1024-point IFFT/FFT is choosen in the rest of this paper.

 

Fig. 3 BER versus (a) OSNR for single pol. OFDM and (b) IFFT/FFT size M for OFDM signal

Download Full Size | PDF

The measured optical spectra (0.02 nm) of 8-channel WDM 65 GHz DFT-spread 16QAM-OFDM with pre-equalization on a 75-GHz grid in OBTB are shown in Fig. 4 (a) and channels from channel 1 to channel 8 are labeled. Figure 4 (b) shows the measured BERs versus OSNR of single channel and the 6th sub-channel in WDM case. No penalty is observed between single channel and WDM case. The constellations of dual polarizations signal in WDM case with OSNR at 36 dB are inserted in Fig. 4 (b).

 

Fig. 4 (a) Optical spectrum in WDM case and (b) BER versus OSNR of Ch. 6 in WDM.

Download Full Size | PDF

Figure 5 (a) shows the measured BER versus input power per channel of the 6th sub-channel of WDM after 840-km SMF transmission. The optimal input power per channel is measured to be 1 dBm. The constellations of 6th sub-channel after 840-km transmission are inserted in Fig. 5(a). Measured BER versus transmission distance of 6th sub-channel is shown in Fig. 5 (b). After 840-km SMF transmission, the OSNR of the 8 × 520Gb/s OFDM signal after 840-km SMF transmission is 20.1 dB and BER of 65-GHz PDM DFT-spread 16QAM-OFDM with pre-equalization is still under 20% SD-FEC limitation (2.4 × 10⁻²). For all sub-channels 65GHz DFT-spread 16QAM-OFDM, the measured BERs after 840-km SMF transmission are below 2.4 × 10⁻² and shown in Fig. 5(c). Optical spectrum (0.02 nm) of WDM signal after 840-km transmission is inserted in Fig. 5(c). After excluding 20% FEC overhead, the net rate of each sub-channel is 507.84 Gb/s × 1/(1 + 20%) = 423.2 Gb/s.

 

Fig. 5 BER versus (a) input power (Ch. 6) and (b) transmission distance; (c) BERs of all channels after 840-km SMF.

Download Full Size | PDF

5. Conclusion

In this paper, it is the first time to achieve single band/λ 400G OFDM signal transmission combination of DFT-spread technique and pre-equalization. We successfully transmitted 8 × 520 Gb/s PDM-16QAM-OFDM signal generated from 80 Gsa/s DAC with 1024 IFFT/FFT size for signal modulation/demodulation over 840 km SMF with the BER under SD-FEC limitation (2.4 × 10⁻²).