1. Introduction

The coming beyond 5th-generation (B5G), 6G, Big Data Era (BDE) and Internet of Things (IoT) are the driving force of a lot of broadband wireless applications, such as smart city, smart home, wireless high definition (HD) stream, security and so on, as shown in Fig. 1. To follow this trend, the ultrafast photonic-wireless network is foreseeable to play a key role in the future to face the exponentially increasing global wireless net data traffic [1]. In addition, it also has great advantages of being compatible with the optical fiber network, facilitating wireless-over-fiber technology for long-distance distribution of wireless signals [2]. From the historical perspective, the comparison of the 3G/4G and 5G networks shows that the carrier frequencies for wireless communication have been increasing progressively to support the larger bandwidth demands [3,4].

 

Fig. 1. The application scenarios for next generation ultra-fast photonic-wireless networks.

Download Full Size | PDF

The frequencies below 60 GHz have been almost fully exploited [5], and in this context the photonic-wireless communication with carrier frequencies beyond 60 GHz such as millimeter wave (MMW) band has been exploited for the large available bandwidth [6,7]. The W-band (75-110 GHz) has attracted increasing interest as a candidate of MMW band to provide wireless communication links with high data rate transmission [8–14]. Among them, X. Pang et al. reported a hybrid optical fiber-wireless link by using photonic heterodyning up-conversion of optical 12.5 Gbaud polarization multiplexed 16-QAM baseband signal with two free running lasers [8]. J. Xiao et al. demonstrated a W-band optical-wireless transmission system with a bit rate up to 40 Gbit/s by adopting optical polarization-division-multiplexing quadrature-phase-shift-keying (PDM-QPSK) signals [13]. The two system-level schemes both consist of bulky discrete devices while requiring additional free running lasers as the optical local oscillators. However, further reducing the complexity of the transmitter and receiver and realizing integrated wireless communication system are highly desirable [2,15]. The on-chip light source-based intensity modulation and direct detection (IM-DD) photonic-wireless links with the advantages of low cost, small footprint and simple configuration are more desirable for practical broadband wireless applications and is becoming the trend of future photonic-wireless communication systems [16,17]. Pulse-amplitude-modulation with four levels (PAM-4) and discrete multi-tone modulation (DMT) have been proposed for the IM-DD systems [18] and the PAM-4 can simplify the transmitter since it only requires a 2-bit digital-to-analog converter (DAC).

In this paper, we report an experimental demonstration of a PAM-4/DMT-16QAM (quadrature amplitude modulation) photonic-wireless transmission system in the W-band based on a monolithically integrated mode-locked laser (MLL). First, an integrated MLL is used as a single light source for W-band wireless signal generation to simplify the transmitter. Compared to the traditional schemes of comb generation by using phase modulators driven by a RF signal in [3] and [6], the use of integrated MLL chip can simplify the system significantly. Second, we employ PAM-4/DMT-16QAM modulation to realize direct intensity modulation and use a mixer-based receiver for detection. Third, the two coherent optical carriers with 80 GHz spacing filtered from the integrated MLL are modulated with data simultaneously, to increase the signal-to-noise ratio (SNR) when beating in the W-band photodiode (PD), while saving an extra optical local oscillator (LO). Finally, the wireless link is free-space transmission without using any lenses to be more practical for actual application scenarios. In addition, we also investigate the IM-DD configuration by employing PAM-4 modulation and envelope detector (ED) in order to further simplify the system. The ED-based PAM-4 photonic-wireless system with a line rate of 10 Gbit/s is more suitable for the wireless applications with lower data rate, low cost and small footprint. For the scenario demanding higher data rate, the mixer-based PAM-4 and DMT-16QAM photonic-wireless system with a line data rate up to 38.1 Gbit/s are more desirable, although the cost and power consumption are also higher since an extra electrical LO at the receiver side is needed.

2. Concept of future embedded access networks vision

Since the W-band MMW carrier exhibits inherently limited transmission distance due to high atmospheric attenuation, it is mainly used to build high-speed wireless communication systems for access applications [19]. The vision of integrated MLL chip-based W-band photonic-wireless access network is illustrated in Fig. 2. The architecture of W-band photonic-wireless system will mainly provide fiber-wireless access networks for future indoor wireless communications.

 

Fig. 2. The vision of on-chip MLL-based W-band MMW embedded photonic-wireless access networks. The MLL-based structure for W-band MMW photonic-wireless access can converge the current optical and wireless networks for indoor communications. MLL: mode-locked laser, MMW: millimeter wave, CO: central office, WSS: wavelength selective switch, ONU: optical network unit, PD: photodiode.

Download Full Size | PDF

The scheme proposed in this paper support the scenario of an integrated MLL chip-based W-band MMW embedded photonic-wireless access network, by combining the frequency comb generation based on an integrated MLL, IM based transmitter, mixer or ED based receiver and lens-free wireless link. The integrated MLL is used to generate an optical frequency comb (OFC) with frequency spacing of Δf, as shown in Fig. 2. Then the generated OFC is launched into a wavelength selective switch (WSS-1) in order to select optical carriers with N*Δf spacing for different channels. For instance, two comb lines with Δf spacing are selected and equalized for channel-1, two tones with 2Δf spacing for channel-2 and so on. The two selected tones in each channel are both modulated with data and each channel can be modulated independently with different data. WSS-2 is used to combine all the channels for fiber transmission. Before sending into each optical network unit (ONU), WSS-3 is used to select different channels for different ONUs. In each ONU, the two optical carriers both modulated with same data beat in a broadband photodiode (PD) by photo-mixing to generate wireless signals. As shown in Fig. 2, to simplify the photonic-wireless access network, after the delivery of baseband optical signal over fiber, the optical self-beating method is utilized for optics-to-wireless conversion and all-optically generate a W-band MMW signal instead of the traditional all-electrical conversion by using an electrical mixer and a remote MMW LO. The proposed scheme is easier to be realized for practical applications by avoiding expensive MMW devices at the user end. In addition, the W-band PD with large available bandwidth is able to support the optics-to-wireless conversion with large bandwidth and realize the seamless integration of wireless and fiber-optic networks.

The employment of the integrated MLL chip for the frequency comb generation has the advantages of low energy consumption and small space occupation. To ensure capacity growth and support more ONUs, the wavelength division multiplexing (WDM) technique is employed to carry parallel optical baseband data channels and generate wireless channels with different carrier frequencies. This massive parallelization approach increases the demand for multi-wavelength laser sources. To address the issue of continuously installing more parallel laser sources, the OFC generated from the MLL becomes desirable, as it may replace a large number of parallel lasers by the individual comb lines. Furthermore, as the phase difference between two individual free-running lasers typically makes the beating MMW carriers unstable with residual frequency sweeping phenomenon, the phase-coherent comb lines can generate the MMW carriers with much improved stability and phase noise performance.

3. Experimental setup

Figure 3(a) shows the experimental setup. A monolithically integrated MLL is used to generate an OFC with 40 GHz frequency spacing, as shown in Fig. 3(b). The integrated MLL is a multi-section laser comprising an active and passive laser waveguide based on GaInAsP material system and operates in passive mode-locking regime with direct current (DC) biased gain and absorber section. The gain current (I_gain) and absorber voltage (V_SA) can be used to fine tune the repetition frequency f_pass of the passively mode-locked MLL to the desired frequency [20,21]. Here, we chose a set of operation parameters (in particular I_gain = 70 mA, V_SA = -1 V), and the corresponding repetition frequency is around 40 GHz. In addition, the MLL is with 6-MHz linewidth, low power consumption and small size in the level of a 1 Euro coin. According to [20,21], the integrated MLLs consist of a monolithic GaInAsP/InP-based multiple quantum-well (MQW) distributed Bragg reflector (DBR) laser chip integrated with either saturable absorber (SA) or electro-absorption modulator (EAM). Monolithic mode-locked DBR lasers have been fabricated as multi-section devices in an extended cavity configuration to generate a comb, based on the GaInAsP/InP material system. The lasers are mainly designed as ridge waveguide (RW) or (semi-insulating) planar buried hetero-structures (SIPBH).

 

Fig. 3. Experimental configuration of the on-chip MLL-based PAM4 photonic-wireless transmission system in the W-Band: (a) The experimental setup of the overall system. (b) The optical spectrum of the frequency comb generated from MLL. (c) The optical spectrum of two tones both with 10 GBaud PAM-4 data modulation after IM. (d) The picture of the actual wireless link. (e) The structure of the digital signal processing (DSP) routine. WSS: wavelength selectable switch, PC: polarization controller, IM: intensity modulator, AWG: arbitrary waveform generator, EDFA: erbium-doped fiber amplifier, OBPF: optical band pass filter, VOA: variable optical attenuator, PD: photodiode, LNA: low noise amplifier, LO: local oscillator, ED: envelope detector, DSO: digital sampling oscilloscope, IF: intermediate frequency.

Download Full Size | PDF

The generated 40-GHz spacing OFC is fed into a wavelength selective switch (WSS) and two comb lines with 80 GHz spacing are selected and equalized, which are both launched into an optical intensity modulator for data modulation. Here the WSS is used to select two tones with 80-GHz frequency spacing as carriers from the comb modes of the MLL, while filtering out the other modes to avoid the beating interference. A 65 GSa/s arbitrary waveform generator (AWG) is used to generate the PAM-4 and DMT-16QAM signals. It should be noted that by modulating the two optical tones with same data signals simultaneously, the SNR of the W-band wireless signal is increased and an extra optical LO is not needed, compared to the traditional scheme of using one tone for data modulation and the other tone for the LO [3]. The two optical carriers both modulated with the same PAM-4/DMT-16QAM data are amplified by an erbium-doped fiber amplifier (EDFA) followed by an optical band-pass filter (OBPF) to remove out-of-band amplified spontaneous emission (ASE) noise. A variable optical attenuator (VOA) is utilized to accurately control the optical power launched into the broadband photodiode (PD, XPDV 4120R, 3-dB bandwidth of 90 GHz) for heterodyne mixing. At the output of the broadband PD followed by a horn antenna, a single channel W-band wireless PAM-4/DMT-16QAM signal centered at around 80 GHz is generated and then emitted into a 0.5-m free-space wireless link without any lenses. The picture of the actual wireless link is shown in Fig. 3(d). Another horn antenna is used to receive the W-band wireless signal, and the received W-band signal is first amplified by a low noise amplifier (LNA, 75-110 GHz) with a 40-dB gain. Two receiver configurations are investigated to down-convert the W-band signal, mixer based receiver and ED based receiver, respectively.

When using the mixer-based receiver, the W-band signal after the LNA is down-converted to the intermediate frequency (IF) by a mixer operating in the W-band, driven by a 2-time frequency multiplied electrical LO. The IF output signal is amplified by a RF amplifier with 40 GHz bandwidth, and then sent into a broadband real-time digital sampling oscilloscope (DSO, Keysight DSOZ334A Infiniium Oscilloscope) with 80 GSa/s sampling rate and 33 GHz analog bandwidth, for analog-to-digital conversion, demodulation and communication performance analysis. When using the ED-based receiver, the W-band PAM signal after the LNA is directly detected and converted to the baseband domain based on envelop detection. The baseband output of the ED is amplified by a RF amplifier with 40 GHz bandwidth, and then also fed into the DSO for demodulation. The digital signals are processed and analyzed offline with a specifically designed digital signal processing (DSP) routine.

The structure of the DSP routine is shown in Fig. 3(e). After the data is mapped onto PAM-4 symbols, the PAM-4 signal is shaped by a Raised-Cosine (RC) filer to shrink the signal bandwidth. In the mixer and ED based setups, the roll-off factor of the RC filter is set to 0.01 and 0.5, respectively. The shaped signal is then resampled to match the sampling rate of the AWG. At the receiver side, the captured waveform is firstly resampled and clock recovery is implemented. In the mixer based setup, a digital down-conversion is performed with the electrical LO at 13.62624 GHz, which corresponds to the frequency of the IF output signal of the mixer. After the synchronization, a Volterra equalizer with a memory length of 40 and 95% reduced complexity is used to equalize the signal. The signal decision is then made for bit-error-rate (BER) counting. To generate the DMT data, 16-QAM signal is modulated on 2000 out of 8192 subcarriers. It is noted that the fast Fourier transform (FFT) size is configured as 8192, out of which the low-frequency 2000 subcarriers are chosen as payload considering the conjugate symmetry of DMT and the bandwidth limitation of AWG. The sampling rate for the DMT signal is 20.48 and 40.96 GSa/s for 10-GHz and 20-GHz double-sideband (DSB) signal, respectively. The overhead caused by cyclic prefix (CP) is 5%. With the CP added, the signal is resampled to match the sampling rate of the AWG. At the receiver side, the captured waveform is first resampled. When using the mixer based receiver, a digital down-conversion with 15.62624-GHz LO is employed corresponding to the IF signal of the mixer output. After the synchronization, channel estimation and compensation are applied for the DMT-16QAM demodulation. It should be noted that the mixer-based setup demands a smaller roll-off factor of 0.01 to be more desirable for the frequency response of the mixer IF bandwidth. In addition, at the receiver side, the signal from DSO is resampled to 60 GSa/s for further DSP. Since the digital clock is not the primary concern in this work, we applied an external referential clock from AWG to DSO.

4. Experimental results and discussions

In this section, we experimentally investigate the three cases: PAM-4/DMT-16QAM signal W-band wireless transmission using mixer-based receiver, and PAM-4 signal W-band wireless transmission using ED-based receiver. In each case, the corresponding bit-error-rate (BER) and electrical spectrum after down conversion are measured. The optical spectra of the OFC generated from the on-chip MLL and the two selected carriers modulated with PAM-4 signal are shown in Fig. 3(b) and (c). To evaluate the signal quality after the wireless transmission, the detailed error vector magnitude (EVM) on each subcarrier of the DMT-16QAM signal are measured.

4.1 W-band wireless PAM-4 signal transmission using a mixer-based receiver

Firstly, we investigate the case of using mixer-based receiver (75-110 GHz) to receive and down-convert the transmitted 10-Gbaud PAM-4 wireless signal. Figure 4(a) shows the BER performance for the 10 Gbaud PAM-4 signal after the wireless transmission, and the BER below the hard-decision forward error correction (HD-FEC) threshold of 3.8 × 10⁻³ with 7% overhead is successfully achieved, resulting a line rate of 20 Gbit/s and a net rate of 18.7 Gbit/s. The PAM-4 signal eye diagram measured at 2-dBm optical power is shown in the inset of Fig. 4(a). Figure 4(b) shows the electrical spectrum of the IF signal after the down conversion by using the mixer. The IF frequency is ∼13.6 GHz, and the mixer is driven by a 2-time frequency multiplied electrical LO with the frequency of 33.2 GHz. It should be noted that the IF frequency is set to be ∼13.6 GHz to avoid the signal-signal beat interference (SSBI) in the low frequency band.

 

Fig. 4. (a) The BER performance for the 10 GBaud PAM-4 signal W-band wireless transmission using a mixer based receiver. Inset: eye diagram at 2 dBm input optical power. (b) The electrical spectrum of the 10 Gbaud PAM-4 signal after the down conversion.

Download Full Size | PDF

The overall performance of the photonic-wireless transmission system is mainly limited by the conversion efficiency of the PD (0.5 A/W), the relatively high conversion loss (8-17 dB) of the mixer and free-space path loss (FSPL) without any lenses.

4.2 W-band wireless DMT-16QAM signal transmission using a mixer-based receiver

To further increase the overall capacity of the photonic-wireless system using intensity modulation and the mixer-based receiver, we employ more spectrally efficient modulation format of DMT-16QAM and investigate two modulation bandwidths of 10 GHz and 20 GHz. The BER performances for the two modulation bandwidths have been measured as a function of the input optical power fed into the PD. As shown in Fig. 5(a), the DMT-16QAM signal with 10 GHz bandwidth has achieved BER performance below the HD-FEC threshold of 3.8 × 10⁻³ with 7% overhead, resulting in a line rate of 19.05 Gbit/s and a net rate of 17.8 Gbit/s. As shown in Fig. 5(b), the DMT-16QAM signal with 20 GHz bandwidth has achieved the BER performance below the soft-decision FEC threshold of 2.7 × 10⁻² with 20% overhead (20%-OH SD-FEC) [22], resulting in a line rate of 38.1 Gbit/s and a net rate of 31.75 Gbit/s. The corresponding signal constellations measured at 4-dBm optical power for both cases are shown in the insets of Fig. 5(a) and (b).

 

Fig. 5. The BER performances of 10-GHz (a) and 20-GHz (b) W-band wireless DMT-16QAM signal transmission using mixer-based receiver. Insets: The constellations both measured at 4 dBm input optical power. (c) The optical spectrum of the selected two 80-GHz spaced tones both modulated with 20-GHz DMT-16QAM signal. (d) The electrical spectra of the 10-GHz and 20-GHz DMT-16QAM signal after the down conversion.

Download Full Size | PDF

The optical spectrum of the two 80-GHz spacing carriers both modulated with 20 GHz DSB DMT-16QAM signal is shown in Fig. 5(c). Figure 5(d) shows the electrical spectra of the 10-GHz and 20-GHz DMT-16QAM IF signals after the down conversion by using the mixer. The IF frequencies for the two cases are both set to be around 15.6 GHz, when the mixer is driven by a 2-time frequency multiplied 32.2-GHz electrical LO. It should be noted that the performance difference between the two DMT-16QAM signals with different bandwidths can also be found from the spreading clusters of the corresponding 16QAM constellations. With higher optical power launched into the PD than 3 dBm, BER performances cannot be further improved, which is mainly attributed to the saturation of the mixer in the receiver.

In addition, the detailed DSP chain for the DMT-16QAM system is as follows: The routine at transmitter includes PRBS generation, adding pilot symbol, uniform QAM mapping, QAM symbol symmetry, IFFT, adding CP (Cyclic Prefix), and resampled to match the AWG sampling rate. The routine at receiver consists of digital down-conversion, baseband signal resampling, synchronization, removing CP, FFT, channel estimation and compensation (using pilot symbols), QAM demodulation, and BER counting.

In addition, we investigate the detailed error vector magnitude (EVM) on each subcarrier of the DMT-16QAM signals for two modulation bandwidths (10 GHz and 20 GHz) using the mixer-based receiver, as shown in Fig. 6. For the DMT-16QAM signal with 20-GHz bandwidth, the signal quality of the subcarriers in the higher frequency band becomes much worse, which is consistent with the power distribution of the 20-GHz DMT-16QAM IF signal as shown in the electrical spectrum (Fig. 5(d)). This is mainly because the IF response of the receiver including the mixer and following RF amplifier becomes worse in the higher frequency band.

 

Fig. 6. Detailed error vector magnitude (EVM) on each subcarrier of the DMT-16QAM signals for two modulation bandwidths (10 GHz and 20 GHz) using the mixer-based receiver, and 2000 subcarriers are loaded with data for both cases.

Download Full Size | PDF

4.3 W-band wireless PAM-4 signal transmission using an ED-based receiver

To further simplify the configuration and reduce the cost and power consumption, we investigate an ED-based receiver to directly detect and down-convert the W-band signal to the baseband. In this case, 5 Gbaud PAM-4 signal is intensity modulated onto the two 80-GHz spacing tones. As shown in Fig. 7(a), the W-band 5 Gbaud PAM-4 signal after the wireless transmission has achieved the BER performance below the SD-FEC threshold of 2.7 × 10⁻² with 20% overhead, resulting in a line rate of 10 Gbit/s and a net rate of 8.3 Gbit/s. It’s noted that the ED based receiver does not need an extra electrical LO and is biased at zero voltage, thus having no power consumption. On the other hand, the ED based receiver requires higher optical power into the PD in order to achieve a good BER performance, indicating the mixer based receiver has better receiver sensitivity than the ED based receiver.

 

Fig. 7. (a) The BER performance versus the incident optical power launched into the PD of 5 GBaud PAM-4 data and ED receiver case. (b) The electrical spectrum of 5 Gbaud PAM-4 signal after ED.

Download Full Size | PDF

The electrical spectrum of the 5 Gbaud baseband PAM-4 signal after the down conversion using the ED is shown in Fig. 7(b). The performance of this IM-DD photonic-wireless transmission system is mainly limited by the mismatch of the working bandwidths between the PD-based transmitter (0-90 GHz) and ED-based receiver (90-140 GHz). We have selected the two coherent comb tones with 80 GHz spacing from the OFC, which can just match the working bandwidths of PD and LNA in the transmitter. However, the only available ED in the experiment has a working bandwidth of 90-140 GHz and degraded performance when working at 80 GHz. Therefore, it is expected that better system performance can be achieved by using an ED with the working bandwidth matched to the transmitter. In the future ONU applications, a PD-based transmitter with larger bandwidth and a frequency comb with smaller repetition will be used in the system to support more ONUs.

4.4 Discussions

Based on the results achieved in our experiments, it can be seen that the ED-based PAM-4 signal is more suitable for the wireless applications with low data rate, low cost and low power consumption. On the other hand, the mixer-based DMT-16QAM signal is more suitable for the system requiring higher data rate and longer wireless transmission distance, in spite of higher cost and power consumption. In addition, the noise figure of the W-band RF amplifiers and the layout of W-band antennas are also relevant for the overall performance of the W-band wireless transmission system. In the future work, some components with better performance will be employed in the experimental setup to further extend the wireless distance significantly, such as a PD with higher efficiency and a power amplifier at the transmitter side, a pair of lenses in the wireless link, and a LNA with higher gain at the receiver.

In terms of the DSP, we have tried to apply a sparse Volterra equalizer using a pruning algorithm which drops out the zero-weighted terms to reduce the computational complexity. The concept of this pruning algorithm is similar to the least absolute shrinkage and selection operator (LASSO) regularization [23], but implemented more straightforwardly. Based on our practice, 90% terms (complexity) can be reduced while the BER is only mildly sacrificed. More detailed description of this pruning process is given in [24] and in our previous work [25]. In addition, the Volterra filter has not been used in the DSP flow of the DMT-16QAM case.

5. Conclusions

We have demonstrated a lens-free W-band MMW photonic-wireless transmission of the intensity modulated 10 Gbaud PAM-4 signal and DMT-16QAM signal with 10 GHz and 20 GHz modulation bandwidth, using an integrated MLL chip in the transmitter and a mixer-based receiver. The integrated MLL acts as an on-chip single light source to generate photonic-wireless signal and simplify the transmitter. The two selected 80-GHz spacing tones are both modulated with the signal to enhance the SNR of the wireless signal generated by the beating of the two tones in the broadband PD. In addition, we also investigated the ED based receiver for the 5-Gbaud PAM-4 signal to realize the IM-DD wireless transmission system for further simplifying the system and reducing the power consumption, with a cost of lower speed and lower receiver sensitivity. Therefore, for various wireless applications with different requirements for the speed, distance, cost and power consumption, mixer based DMT-16QAM signal or ED based PAM-4 signal could be adopted to deliver the W-band wireless signal.