1. Introduction

It is well known that the millimeter-wave (mm-wave) communication technology, with the advantages of huge bandwidth and low propagation loss, has been perceived as a promising candidate for the future wireless communications and space communications, providing various broadband services with good quality. It is a very hot topic to generate mm-wave signal based on photonics technologies recently [1–20 ]. The combination of advanced modulation formats and digital signal processing (DSP) makes it possible to modulate the data in the high dimensional space at the transmitter and overcome the optical impairments after the full information has been extracted at the receiver, respectively, which helps us benefit a great deal in both high spectral efficiency and high sensitivity [9].

In order to obtain the great benefits mentioned above, a stable high-order quadrature-amplitude-modulation (QAM) vector mm-wave signal generation scheme based on intensity modulator with photonic frequency doubling and unbalanced precoding was proposed in [19] very recently. In [19], we investigated and found that this scheme works well for all QPSK, 8QAM, and 16QAM modulation formats. But the carrier frequency is quite low in that paper, only 12 GHz, and the baud rate is just 1 Gbaud. Therefore, We will investigate 1~4 Gbaud vector mm-wave signal at the carrier frequency of 40 GHz in this paper.

Additionally, one balanced precoding technology was proposed and experimentally demonstrated in [20]. However, only relatively low-frequency mm-wave vector signal at 16 GHz with QPSK modulation is experimentally demonstrated. We will employ this balanced precoding technology to generate 8QAM and 16QAM vector mm-wave signal in this paper. According to our study, the balanced precoding is worse compared to the unbalanced precoding when employed in high-order QAM modulation formats due to the asymmetrical output and limited bandwidth of devices, and the rotation and overlapping of the randomly, asymmetrically scattered signal constellation points. Besides, DC component in unbalanced precoding case may help the linear up-convision, which could lead to a better performance.

2. Principle of pre-coded QAM vector mm-wave signal generation

Figure 1 illustrates the principles of pre-coded QAM vector mm-wave signal generation based on frequency doubling and precoding. As shown in Fig. 1, the continuous wave (CW) at the frequency of $f_{\text{c}}$, is modulated by the radio-frequency (RF) signal, which carries the data at the center frequency of $f_s$ and drives the Mach-Zehnder modulator (MZM). As derived from [17], when the MZM is biased at its minimum transmission point, after the square-law detection of the PD at the receiver, the RF current can be expressed by

 

Fig. 1 Principles of (a) pre-coded M-QAM vector mm-wave signal generation and (b) frequency doubling due to the square-law decetion. CW: continuous wave, MZM: Mach-Zehnder modulator, PD: photodiode.

Download Full Size | PDF

Therefore, in order to attain the amplitude and phase information of the vector signal after square-law detection directly, precoding is needed at the transmitter. The amplitude $K_2\left(t\right)$ can be derived from $J_1{}^2\left(\kappa \right)$, and phase $\phi \left(t\right)$ of the driving RF signal should satisfy

 

Fig. 2 Constellations of (a) original, (b) unbalanced precoded, (c) balanced precoded 8QAM signals, (d) original, (e) unbalanced precoded, and (f) balanced precoded 16QAM signals.

Download Full Size | PDF

The minimum distance between constellation points is the same in both balanced and unbalanced precoding cases, but there are more points in the balanced precoded signal constellation. Also, it is worth noticing that the constellation points are symmetrically distributed on the complex plane with a nearly equal probability in the balanced scheme, which makes the DC components in the baseband signal disappear.

When there is no fiber transmission before PD detection, the phase noise can be cancelled after PD detection, since the two first-order subcarriers used for heterodyne beating are from the same laser source and thus phase- and frequency-locked. However, when there is fiber transmission before PD detection, the phase noise after PD detection cannot be cancelled and will be enhanced with the increase of the fiber transmission distance, which will degrade the SNR performance [21].

In our experiment, a pseudo random binary sequence (PRBS) with a length of $3\times 2^9$ for 8QAM or $4\times 2^9$ for 16QAM is used as the information source, and then it is parsed into subsequnces each of length ${\log }_2(\text{M})$, and each sequence is mapped into one of the symbols of M-QAM. In balanced precoding scheme, it is necessary to give m in Eq. (2) a random value of 0 or 1 in order to make the constellation points distributed symmetrically with a nearly equal probability on the complex plane. Since in PRBS stream, zeros and ones have almost equal probability, a PRBS called m-controller can be used when performing the balanced precoding. After that, the precoded symbols are upsampled to match the ratio of the DAC’s sampling rate to the baud rate of the baseband signal. After filtered by a fifth-order Bessel low pass filter (LPF) with the cut-off frequency the same as the value of the baud rate, the baseband signal is up-converted to the passband RF signal. All operations mentioned above are completed by Matlab Programming. The RF signal is uploaded to an arbitrary waveform generator (AWG) with 80 GSa/s sampling rate to drive the MZM biased at its minimum transmission point. The total bitrate can be from 3 Gb/s to 12 Gb/s for 8QAM and from 4 Gb/s to 16 Gb/s for 16QAM.

3. Experimental setup and results

Here our experimental setup for high-order QAM vector signal generation at 40 GHz based on intensity modulator is shown in Fig. 3 . The CW from an external cavity laser (ECL) with the linewidth less than 100 kHz and the minimum power of 13 dBm is modulated by the aforementioned 20 GHz precoded RF signal by means of an IM. Figure 3(a) and Fig. 3(b) show the spectrum after 20 GHz up-conversion for balanced precoding 8QAM case and unbalanced precoding 8QAM case, respectively. There is no DC components in the balanced precoded signal before up-conversion. Therefore, strong 20-GHz carrier components are observed in Fig. 3(b) but absent in Fig. 3(a). The IM has a half-wave voltage ($V_{\pi }$) of 3.2 V, 3-dB bandwidth of ~36GHz and insertion loss of 4 dB. The 20 GHz precoded RF signal is amplified to 3.6$V_{p-p}$ by an electrical amplifier (EA) with the passband from 17~27 GHz, to drive the IM. The IM is biased at its minimum transmission point to realize the optical carrier suppression modulation. The precoded and modulated optical signal is then amplified by an erbium-doped fiber amplifier (EDFA) to compensate for the insertion loss and the modulation loss. After being transmitted over a 25-km single-mode fiber (SMF), the optical signal passes through a variable optical attenuator (VOA) for bit-error-rate (BER) measurement.

 

Fig. 3 Experimental setup for balanced/unbalanced precoded M-QAM vector mm-wave signal generation and direct detection with offline DSP, and spectra after up-conversion for 20 GHz (a) balanced and (b) unbalanced precoding 16QAM vector signal. AWG: arbitrary waveform generator, ECL: external cavity laser, EA: electrical amplifier, MZM: Mach-Zehnder modulator, EDFA: erbium-doped fiber amplifier, SMF: single mode fiber, VOA: variable optical attenuator, PD: photodiode, DSP: digital signal processing.

Download Full Size | PDF

At the receiver, the optical signal is detected by a square-law PD with 3-dB bandwidth of 50 GHz and converted into an ordinary M-QAM modulated RF signal with the center frequency at 40 GHz. Then, the electrical RF signal is sampled by a digital oscilloscope (OSC) with a sampling rate of 120 GSa/s and bandwidth of 45 GHz. The original M-QAM signal can be recovered from the 40 GHz electrical RF signal by means of advanced DSP offline. The structural level design of DSP includes resampling, intermediate frequency (IF) down conversion, constant modulus algorithm (CMA), cascaded multi-modulus algorithm (CMMA), frequency offset estimation (FOE), carrier phase estimation (CPE), decision and BER counting.

The optical spectra (0.02-nm resolution) at the output of MZM for balanced and unbalanced precoded 8QAM signal are shown in Fig. 4 , illustrating that the two first-order subcarriers are generated with 40 GHz frequency spacing in optical domain. Figure 5 shows the BER behaviours with respect to launched optical power into PD for 8QAM and Fig. 6 is for 16QAM case. Constellations are also given in Fig. 5(a) and Fig. 6(b), and the electrical spectrum of up-converted signal is given in Fig. 5(a). Obviously, the higher the transmission baudrate is, the worse BER performance is. Because the signal of higher baud rate occupies a wider bandwidth. For 1 GBaud, 2 GBaud and 4 GBaud 8QAM, the balanced precoding scheme needs 2.7 dB, 3.4 dB and 3.1 dB optical power penalty at BER of $3.8\times {10}^{\text{-}3}$(the threshold of Hard-Decision Forward Error Correction) compared with unbalanced precoding scheme, respectively, and for 1 GBaud and 4 GBaud 16QAM precoding scheme, 1.0 dB and 3.5 dB optical power penalty is required at the BER of $3.8\times {10}^{\text{-}3}$, respectively.

 

Fig. 4 Optical spectra at the output of MZM for (a) balanced precoded (b) unbalanced precoded 8QAM signals.

Download Full Size | PDF

 

Fig. 5 Performance of precoded 8QAM vector signals: (a) 1 Gbaud and (b) 2- and 4-Gbaud.

Download Full Size | PDF

 

Fig. 6 Performance of precoded 16QAM vector signals: (a) 1 Gbaud and (b) 4 Gbaud.

Download Full Size | PDF

According to the principles of precoding technique, it should be pointed out that there is no difference between balanced and unbalanced precoding in ideal circumastances. However, it can be found that the BER performance of balanced precoding technology is worse than that of unbalanced precoding technology when it is employed in high-order QAM modulation formats. This phenomenon can be explained as follows. Fisrtly, the constellation of balanced precoding is symmetrical and contains more points compared to the unbalanced one. In order to obtain the symmetry, the ideally symmetrical output of the AWG and other devices is needed. However, such accuracy of devices cannot be assured in fact. Secondly, high-order QAM signals with high baud rate have a wider bandwidth, which may exceed the bandwidth limitation of some devices. Thirdly, as shown in Fig. 2(c) and Fig. 2(f), any pair of points (one point and its symmetry point) will appear at the same location of the original M-QAM after the square-law detection of PD. But due to the degradation of the signals after transmission, one point becomes a mess of asymemetrical points randomly. Therefore, in the balanced precoding case, after random rotation and overlapping of the two masses of points, the constellation points of M-QAM scatter more dispersively, leading to a worse BER performance. In addition, the DC component in unbalanced precoding case may help the linear up-convision, which could lead to a better performance.

4. Conclusion

The generation of high-order QAM vector mm-wave signal with high bitrate based on intensity modulator with photonic frequency doubling and precoding has been investigated in this paper, and the comparison between balanced and unbalanced precoding technique has been made. We find that the performance of the balanced precoding technique is worse than that of the unbalanced precoding technique in high-order QAM modulation. This is caused by the asymmetrical output and limited bandwidth of devices, and the rotation and the overlapping of the randomly, asymmetrically scattered signal constellation points. Besides, in the unbalanced precoding scheme, the DC component may help linear up-conversion, bringing the better performance.