By Topic

IEEE Quick Preview
  • Abstract



Recently, coherent optical OFDM (CO-OFDM) transmission beyond 100-Gb/s, including the 1-Tb/s experiment, has been demonstrated [1], [2], [3], [4], [5], [6]. However, the signal reception in those reports is processed offline, circumventing the real-time implementation of CO-OFDM reception, which is crucial for practical applications. Optical OFDM real-time transmitters [7], [8] and receivers [9], [10], [11] have recently been reported, but the data rate is low for backbone network application. With the bandwidth limit of state-of-the-art silicon technology, it is difficult to improve data rate by increasing the sampling rate of analog-to-digital converters (ADCs) beyond 100 GSa/s [12]. To alleviate bandwidth constraint of ADCs, multiband OFDM has been proposed, and it has been demonstrated that a part of the OFDM spectrum can be carved out and detected at a fraction of the overall data rate [1], [10]. Yang et al. demonstrated the first multiband real-time CO-OFDM reception up to 54-Gb/s [10]. However, in all of the real-time CO-OFDM experiments [9], [10], chromatic dispersion or polarization-mode dispersion (PMD) effects have not been considered. In this paper, we show multiband real-time CO-OFDM reception at a data rate of 110-Gb/s based on a field-programmable gate array (FPGA) with an individual subband at 3.33-Gb/s. The multiband CO-OFDM signal is successfully received and recovered after 600-km recirculation loop transmission and 400-ps differential-group-delay (DGD). To the best of our knowledge, this is the record real-time data rate for coherent OFDM reception with some realistic transmission distance and PMD effects.



The experimental configuration of multitone generation and real-time coherent optical MIMO OFDM (CO-MIMO-OFDM) reception is shown in Fig. 1. The transmitted data stream consists of Formula$2^{15} - 1$ pseudo-random bit sequence (PRBS). It is first mapped onto three OFDM subbands. Each subband contains 46 subcarriers of 4-quadrature-amplitude modulation (QAM) and 18 unfilled subcarriers. Twenty one unfilled subcarriers are placed at each end of the central subband, and 11 unfilled subcarriers are at left end of the left subband and right end of right subband. This allows the subbands to be eventually distributed when patched with other tones. The three subbands together with 64 unfilled subcarriers are converted to the time domain via inverse Fourier transform (IFFT) with size of 256. One eighth of the cyclic-prefix ratio is used, resulting in an OFDM symbol size of 288. The three OFDM subbands were converted into analog form by an arbitrary waveform generator (AWG) at 6 GS/s. The AWG outputs repetition of the OFDM blocks or frames, each containing 504 OFDM symbols. Within each frame, the first and third OFDM symbols serve as timing and frequency offset estimation sequence, and the second and fourth are left blank. In the next 32 training symbols, odd ones are normally modulated with transmitted data, while even ones are unfilled to form a pattern of alternative polarization launch after polarization multiplexing emulator. The two side subbands are pre-equalized to compensate the frequency roll-off of AWG, as shown in Fig. 1 (see the inset next to AWG). The baseband OFDM signal is then directly up-converted into optical domain through an optical I/Q modulator. At the output of the modulator, the optical signal is fed into a recirculating frequency shifter (RFS) to generate 11 uncorrelated OFDM bands (see the inset at the output of RFS). The RFS consists of a closed fiber loop, an I/Q modulator, and optical amplifiers to compensate the frequency conversion loss. The I/Q modulator is driven with two equal but 90° phase shifted 6-GHz RF tones through I and Q ports to induce a frequency shifting to the input optical signal [13]. The recirculating loop enables replicating multiple copies by one step per circulation if the initial optical OFDM signal is supplied. Because of extremely large loop delay introduced between the neighboring bands, each individual band is completely uncorrelated. The net data rate for each subband is 3.33-Gb/s and, thus, 10-Gb/s for each band. The total net data rate becomes 110-Gb/s after bandwidth expansion via RFS. The signal is fed into a polarization multiplexing emulator to simulate two independent transmitters and then coupled to a recirculation loop comprising 100-km standard single mode fiber (SSMF) and a two-stage EDFA to compensate the loss. It is coupled out from the loop after six circulations and fed into a PMD generator to add 400-ps PMD into the signal. The optical spectrum of CO-OFDM signal after RFS and polarization multiplexing emulator is depicted in Fig. 2.

Figure 1
Fig. 1. Experiment setup for real-time CO-MIMO-OFDM reception with 600-km transmission and 400-ps DGD.
Figure 2
Fig. 2. Optical spectrum of 110-Gb/s CO-OFDM signal at back-to-back transmission.

At the receiver side, we use direct optical-to-RF down-conversion to detect optical OFDM signal. The principle of signal processing procedures is the same as [9]. Each time, only one subband is detected and others are filtered out by two 575-MHz anti-alias LPFs before ADCs. The detected RF signals are then sampled with four high speed ADCs at 1.5 GSa/s with 7-bit resolution. The signal is transmitted via low-voltage differential signaling (LVDS) interface into FPGA through 1 : 2 multiplexed outputs, which lowers the rate down to 750 MSa/s. The multiple inputs are received and demultiplexed into four channels at 375 MSa/s in the FPGA for further processing. After all the OFDM signal processing, the recovered data are compared with transmitted ones in FPGA and the errors are counted. This error count, together with transmitted OFDM symbol numbers, is sampled by SignalTap II (an embedded logic analyzer from FPGA vendor Altera) debugging module and transported via Joint Test Action Group (JTAG) interface, which is a common standard for testing integrated circuits and printed circuit boards using boundary scan to PC for BER collection.

In the 600-km transmission experiment, an enable signal from pulse generator is required to indicate valid signal for BER counting due to use of recirculation loop for long-haul transmission emulation, as shown in Fig. 3. This signal is imported into FPGA chip to enable or disable the timing synchronization module to generate timing signal, which is indispensable for the rest of signal processing modules. When the timing signal is disabled, the BER is not being calculated, namely, only the BER for the OFDM signal after 600-km transmission is collected.

Figure 3
Fig. 3. Enable signal indicating valid data for FPGA signal processing.


The CO-MIMO-OFDM receiver architecture can be divided into nine stages: 1) timing synchronization; 2) frequency offset synchronization; 3) CP removal; 4) FFT; 5) phase estimation for training symbols; 6) channel estimation; 7) Jones Matrix inversion to recover two polarization signals; 8) phase estimation for payload symbols; and 9) symbol decision, error accumulation, and BER computation. Some stages of signal processing are discussed in detailed below:

  1. Timing synchronization. In this stage, the OFDM symbol is properly delineated to avoid intersymbol interference. A time-domain preamble [14] that is partitioned into four segments is used. To improve the accuracy of frequency offset estimation two training symbols are used for timing and frequency estimation. The timing metric of real-time data is shown in Fig. 4(a).
  2. Frequency synchronization. Frequency offset between signal laser and local laser must be estimated and compensated before further processing. Fig. 4(b) demonstrates the estimated frequency offset from two training sequences. According to the scheme in [9], at timing estimate points, the calculated values are estimated frequency offsets. The data used for frequency offset compensation is an average of these two values to improve accuracy.
  3. Channel estimation. The channel matrix Formula${\mmb H}$ is estimated by sending 32 OFDM symbols using alternative polarization launch. They are generated by filling the odd symbols with normal transmitted data, while leaving the even symbols blank. After the polarization multiplexing emulator, the training symbols form a pattern of alternative polarization launch for two consecutive OFDM symbols, as depicted in Fig. 5. The four elements of channel matrix Formula${\mmb H}$ is computed as follows: the two elements of the first (second) column of Formula${\mmb H}$ can be estimated using odd (even) number of the pilot symbols [9]. The signal processing algorithms used in the other modules have been elaborated in [9].
Figure 4
Fig. 4. (a) Timing metric for one polarization CO-MIMO-OFDM signal. (b) Frequency offset estimate for the CO-MIMO-OFDM signal.
Figure 5
Fig. 5. Time-domain representation of the dual-polarization OFDM block including training symbols for timing and frequency synchronization, channel estimation, and payload. “ Formula$x$” and “Formula$y$” represent two polarization components.


The average BER performance of 11 bands 4-QAM CO-OFDM signal at back-to-back transmission up to 110-Gb/s with and without 400-ps PMD is demonstrated in Fig. 6. The polarization in the system is left free running, and no manual polarization control is used during measurement. The inset shows a typical constellation diagram for the detected CO-OFDM signal. Each point in this figure is obtained by averaging over 30 OFDM transmission blocks. The combined laser linewidth is about 100-kHz. A BER of Formula$10^{-3}$ can be observed at an OSNR of 22 dB (ASE noise bandwidth of 0.1 nm) for 4-QAM. It is noted that the difference of BER performance for two polarization signals at same OSNR is less than 0.5 dB, and the OSNR penalty for a 400-ps DGD is smaller than 0.5 dB at Formula$10^{-3}$. We also measure the performance after 600-km recirculation loop transmission. The launch power for the transmission is +5.3 dBm. Fig. 7 shows the BER performance for the 11 tones at the reach of 600 km with and without 400-ps DGD, and it can be seen that all the tones can achieve a BER better than Formula$2 \times 10^{-3}$, which is the FEC threshold with 7% overhead. It is noted that there is no observable penalty for 400-ps DGD. The BER performance for this real-time 110-Gb/s CO-MIMO-OFDM transmission is limited by two factors: i) the noise accumulation for the edge bands that have gone through most of the frequency shifting and ii) the large phase drift due to long OFDM symbol length by using relatively low sampling rate of 1.5 GSa/s, which can be avoided by using ADCs with higher sampling rate. Nevertheless, our demonstration has achieved the real-time characterization of CO-OFDM signal after 600-km transmission and relatively large DGD.

Figure 6
Fig. 6. Average BER performance of 11 bands CO-OFDM signals at back-to-back transmission with and without 400-ps PMD.
Figure 7
Fig. 7. BER performance for individual OFDM tones after 600-km transmission.


A 110-Gb/s real-time coherent optical MIMO OFDM reception after 600-km transmission and 400-ps DGD based on FPGA has been demonstrated. The multiband OFDM signal is generated with a recirculating frequency shifter producing 11 tones, each comprised of three subbands, and the BER performance of each tone is better than the 7% FEC threshold of Formula$2 \times 10^{-3}$. The overall spectral efficiency is a modest 1.67 b/s/Hz due to large a band-gap between the 33 subbands.


Corresponding author: S. Chen (e-mail:


No Data Available


No Photo Available

Simin Chen

No Bio Available
No Photo Available

Yiran Ma

No Bio Available
No Photo Available

William Shieh

No Bio Available

Cited By

No Data Available





No Data Available
This paper appears in:
No Data Available
Issue Date:
No Data Available
On page(s):
No Data Available
INSPEC Accession Number:
Digital Object Identifier:
Date of Current Version:
No Data Available
Date of Original Publication:
No Data Available

Text Size