High-Speed 9.6-μm Long-Wave Infrared Free-Space Transmission With a Directly-Modulated QCL and a Fully-Passive QCD

Free-space optics (FSO) in the mid-infrared (mid-IR) contains rich spectral resources for future ultrahigh-speed wireless communications yet is currently under-exploited. Two atmospheric transmission windows at the mid-IR, namely, the mid-wave IR (MWIR, 3–5 μm) and the long-wave IR (LWIR, 8–12 μm), show great potential in supporting free-space communications for both terrestrial and space application scenarios. Particularly, the LWIR signal with a longer wavelength has high intrinsic robustness against aerosols’ scattering and turbulence-induced scintillation and beam broadening effects, which are the main concerns hindering the wide deployment of practical FSO systems. In this context, high-bandwidth semiconductor-based mid-IR FSO transceivers will be desirable to meet the requirements of low energy consumption and small footprints for large-volume development and deployment. Quantum cascade devices, including quantum cascade lasers (QCLs) and quantum cascade detectors (QCDs), appear promising candidates to fulfill this role. In this work, we report a high-speed LWIR FSO transmission demonstration with a 9.6-μm directly-modulated (DM)-QCL and a fully passive QCD without any active cooling or bias voltage. Up to 8 Gb/s, 10 Gb/s, and 11 Gb/s signal transmissions are achieved when operating the DM-QCL at 10 °C, 5 °C, and 0 °C, respectively. These results indicate a significant step towards an envisioned fully-connected mid-IR FSO solution empowered by the quantum cascade semiconductor devices.

(MWIR, 3-5 µm) and the long-wave IR (LWIR, 8-12 µm), show great potential in supporting free-space communications for both terrestrial and space application scenarios. Particularly, the LWIR signal with a longer wavelength has high intrinsic robustness against aerosols' scattering and turbulence-induced scintillation and beam broadening effects, which are the main concerns hindering the wide deployment of practical FSO systems. In this context, high-bandwidth semiconductor-based mid-IR FSO transceivers will be desirable to meet the requirements of low energy consumption and small footprints for large-volume development and deployment. Quantum cascade devices, including quantum cascade lasers (QCLs) and quantum cascade detectors (QCDs), appear promising candidates to fulfill this role. In this work, we report a high-speed LWIR FSO transmission demonstration with a 9.6-µm directly-modulated (DM)-QCL and a fully passive QCD without any active cooling or bias voltage. Up to 8 Gb/s, 10 Gb/s, and 11 Gb/s signal transmissions are achieved when operating the DM-QCL at 10°C, 5°C, and 0°C, respectively. These results indicate a significant step towards an envisioned fully-connected mid-IR FSO solution empowered by the quantum cascade semiconductor devices.
Index Terms-Free-space optics, long-wave infrared, quantum cascade detector, quantum cascade laser.

I. INTRODUCTION
F REE-SPACE optics (FSO) has recently emerged as a potential candidate to complement the radio technologies for both terrestrial and space communications, including groundto-satellite and inter-satellite communications. Extrapolating the applications and requirements of the fifth-generation (5G) mobile networks currently being deployed, one may anticipate at least two orders of magnitude higher data rate demand for the next-generation wireless communication technologies, i.e., the 6G [1]. Subsequently, an all-spectra communication paradigm facilitating cooperative free-space communications with both optics and radio has been proposed to offer the required ultrabroad spectral resources [2]. Among all the options spanning from the microwave to the ultraviolet (UV), the spectral region across the terahertz (THz) band, i.e., from 0.3 THz to tens of THz, where the mid-and far-infrared (IR) optics and This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ sub-millimeter-wave (MMW) radio overlap, has a relatively lower technology readiness level compared with other spectral windows [3]. Particularly, the two atmospheric transmission windows at the mid-wave IR (MWIR, 3-5 µm, 60-100 THz) and the long-wave IR (LWIR, [8][9][10][11][12][25][26][27][28][29][30][31][32][33][34][35][36][37] in the mid-infrared (MIR) region (wavelength range of 3-50 µm) contain a rich potential for FSO technologies and communication purposes, yet to be exploited. Compared with the MMW and sub-THz bands being heavily studied and discussed for 6G wireless communications, the MWIR and LWIR have almost two orders of magnitude lower atmospheric propagation attenuation with a tenfold broader fully unlicensed bandwidth [4]. On the other hand, compared with the 1.5-µm telecom band in the short-wave IR (SWIR, 1-2.5 µm), where most commercial FSO communications are developed, the MWIR and LWIR have much higher resilience against adverse weather, e.g., much lower sensitivity to particle scattering atmospheric turbulence effects [5]. Moreover, the mid-IR region shows a lower risk for eye safety [6].
An often-adopted approach to carrying out mid-IR FSO system demonstrations is using wavelength conversions [7], [8], [9], [10], [11], [12]. This approach utilizes the well-developed telecom transceivers and converts the signal wavelength between 1.5 µm and mid-IR based on difference frequency generation (DFG) before and after the FSO link. Recent studies with such an approach have reported very high data rates. For example, single-channel transmissions of 10 Gb/s in-phase and quadrature (IQ) modulated signals have been demonstrated [10], [11]. Moreover, with wavelength-and mode-division multiplexing techniques, MIR FSO transmissions with a total data rate of 300 Gb/s have been achieved at 3.4 µm [12]. However, the high-power consumption associated with the nonlinear wavelength conversion in such methods leads to energy deficiency and hardware complexity, obstructing practical development. A more attractive choice in the long-term would be directemission semiconductor lasers and photodetectors.
Quantum cascade devices, including quantum cascade lasers (QCLs) [13] and quantum cascade detectors (QCDs) [14], are based on inter-subband transitions to cover a wide spectral range from the MIR to THz. They appear as promising candidates due to their breakthroughs in a broad bandwidth, high-temperature operation, and low energy consumption [15], [16], [17], [18]. Particularly, directly modulated (DM)-QCLs have ultra-short carrier relaxation lifetime, leading to a high intrinsic modulation bandwidth and making the laser response over-damped, thus repressing the resonance frequency [19], [20]. In recent years, high bandwidth modulation of QCLs operated at room temperature has rekindled a new wave of interest in studying QCL-based FSO communication systems [21]. There have been several reports of free-space transmission demonstrations using DM-QCLs at room temperature [22], [23], [24], [25]. Among these works, the highest reported bitrate of 6 Gb/s is achieved in the MWIR region of 4.65 µm with a thermoelectric-cooled Mercury-Cadmium-Telluride (HgCdTe) photovoltaic (PV) detector operating at 200 K to assure the low-noise characteristics [25]. The LWIR range, which has a smaller energy bandgap than the MWIR, imposes more stringent requirements for the power and noise characteristics of the device [26], [27]. Before this work, the highest reported bitrate for LWIR FSO at room temperature with DM-QCL, to the best of our knowledge, was up to 2 Gb/s with a bit error rate (BER) below the 2×10 −2 soft-decision (SD) forward error correction (FEC) limit [28]. Lately, a 9-µm FSO data transmission based on an external free-space Stark modulator and an unbiased QCD without any active cooling was reported and up to 10 Gb/s non-return-to-zero on-off keying (NRZ-OOK) signals at room temperature was demonstrated [29], [30]. However, the external modulator-based FSO systems require extra LWIR optics and a more delicate beam alignment to collimate and gain the optimum power. These extra efforts are imperative to increase the modulation depth, thus achieving the required signal-to-noise ratio (SNR) at the detector. Therefore, a more efficient and straightforward candidate is DM-QCL which can accelerate commercial LWIR FSO solutions in the short term, whereas future maturity of external-modulation technologies may bring improved modulation bandwidth and quality in the long term.
In this paper, we extend our recent report on high-speed LWIR DM-QCL-based FSO demonstration [31] with more detailed system characterizations and transmission performance evaluations. For different temperatures from 10°C down to 0°C , we experimentally compare two modulation formats, i.e., NRZ-OOK and 4-level pulse-amplitude modulation (PAM4), and benchmark their respective maximum data rate against the 6.25% overhead (OH) hard-decision (HD) FEC limit [32]. We show that up to 8 Gb/s, 10 Gb/s, and 11 Gb/s PAM4 transmissions can be achieved at operating temperatures of 10°C, 5°C, and 0°C, respectively, with the 9.6-µm DM-QCL and an uncooled and unbiased passive QCD.
The rest of this paper is organized as follows: in Section II, we present the key characteristics of the DM-QCL and the QCD and describe the experimental configurations of the transmission system. Section III shows the measurement results of the LWIR transmissions at three different temperature values. Finally, we give our conclusions in Section IV.

A. Characteristics of the Quantum Cascade Components
The DM-QCL used in our experiment is a 3-mm distributedfeedback (DFB) ridge laser fabricated epi-down. In this way, the laser chip can be operated at room temperature with the help of a Peltier thermoelectric controller (TEC). The laser chip was soldered on a sub-mount and then wire bonded to a high-frequency coplanar waveguide. Fig. 1(a) shows the picture of the QCL mount and the zoomed view of the DFB-QCL chip. The output of the DM-QCL is a continuous wave centered at a wavelength of 9.6 µm with a few tens of milliwatts of output power. The device has a cut-off modulation frequency of around 3 GHz, beyond which the frequency response drops drastically. In our transmission system experiment, we use a Peltier TEC to stabilize the DM-QCL operation temperature at three different points, i.e., 10°C, 5°C, and 0°C. Fig. 1(b) shows the characterized L-I-V curve of the laser at the selected operating temperature values. One can observe that the DM QCL starts lasing around 560 mA injection current at all investigated temperatures and saturates at around 675 mA. It is noted that due to the large measurement steps around the threshold current, the slight differences between the thresholds at different temperatures couldn't be distinguished. For the same reason, the L-I curve for the case of 10°C also appears piecewise linear near the threshold, which is a measurement artifact. Lower operation temperature enables a higher output power, and the maximum achieved output power at 10°C, 5°C, and 0°C are 16.4 mW, 23.4 mW, and 30.5 mW, respectively.
The LWIR FSO receiver for this experiment is a fully passive GaAs/AlGaAs QCD without any bias or cooling, operating in a room-temperature environment (no air conditioning, temperature drifting between 25°C and 30°C). Fig. 2 shows the picture and the characteristics of the employed QCD. The detector was fabricated into a 55×55 µm 2 square mesa structure where the top contact is realized with an air bridge. Such a mesa structure is used to operate at high frequencies, compared with traditional mesa structures with larger sizes and long wire bonds usually used for static / low-frequency measurements. The air bridge is continued by a coplanar waveguide on the GaAs substrate, which is then connected through short wire bonds to a printed circuit board (PCB) coplanar waveguide equipped with a 2.92-mm connector. The QCD is based on a diagonal transition with a 45°polished facet that satisfies the polarization requirements for optical transitions [33]. This unipolar detector conducts in the photovoltaic regime; yet has s a very wideband frequency range with a first-order roll-off. This behavior is due to the fast energy relaxation of the electrons as the only charge carriers. Also, the asymmetry of the cascade region that acts as a pseudo-electric lead to a photocurrent. The cascade region is designed in such a way that the electron relaxation lifetime from one period to the adjacent one is estimated to be shorter than 10 ps. Therefore, the intrinsic bandwidth, in theory, can be in the order of 100 GHz [34]. In practice, the bandwidth is often limited by electrical parasitic capacitance in the mesa structure, packaging, and driving circuit. For this particular device, the cut-off frequency is around 6 GHz due to both the RF packaging and the impedance mismatch. Per the previous measurement, the detector has an almost flat frequency response up to its cut-off frequency [29]. Fig. 2(a) shows a picture of the QCD mount where the detector chip is wire-bonded to the coplanar waveguide. Fig. 2(b) shows the photocurrent characterization of the QCD at room temperature. The peak responsivity is measured at 141.6 meV photon energy, corresponding to 8.75 µm of wavelength. While at 9.6 µm, the detector responsivity performance drops to approximately 45%. Therefore, we can expect improved efficiency with a lightly shorter laser wavelength approaching the peak. As shown in Fig. 2(c), the photocurrent of the QCD has a linear relation with the injected optical power up to 50 mW with a responsivity of 4.5 mA/W at 9 µm at room temperature. In our transmission experiment, the highest optical power measured on the detector was 30.5 mW within the linear region. Fig. 3(a) shows the experimental setup of the LWIR FSO transmission system. We use a 64 GSa/s arbitrary waveform generator (AWG) to convert the offline generated digital signals into the analog domain. Two modulation formats are employed for system performance evaluation, i.e., NRZ-OOK and PAM4. Their symbols are mapped from a random binary sequence of >1 million unrepeated bit-length generated using the Mersenne Twister with a shuffled seed number. Then, the sequence was filtered with a root-raised-cosine (RRC) pulse shaping filter with a roll-off factor of 0.15 after optimization. To compensate for the bandwidth limitation of the system, we apply a static pre-emphasis finite impulse response (FIR) filter at the transmitter digital signal processing (DSP). The AWG output signal is amplified with a 30 dB-gain electrical amplifier (EA) to a peak-to-peak voltage of >2.5 V pp . The value of the driving signal varies between the modulation formats and different baud rates. The signal is combined with the bias current at an external bias-tee and delivered to the DM-QCL. Subsequently, at the DM-QCL output, we installed a ZnSe aspheric lens to collect and collimate the emitted laser beam. A picture of the transmitter configuration is shown in Fig. 3(b).

B. Transmission Setup of the LWIR FSO System
In this experiment, the FSO link distance was set to 25 cm to simplify the beam alignment and collimation. To achieve the maximum photocurrent value at the QCD, we placed an optical beam focusing element consisting of two cascaded lenses to focus the light on the detector to obtain the smallest spot size possible. Such an focusing configuration allowed us achieving a "best-effort" coupling efficiency into the detector, thus ensuring sufficient receiver SNR for the FSO transmissions. Fig. 3(c) shows the photos of the LWIR FSO receiver configuration. Before data transmission, we optimized the beam alignment using an optical chopper of 1 kHz and a lock-in amplifier while keeping the QCL unmodulated with continuous-wave output. The lock-in amplifier was connected to the QCD output, so that maximizing the detected signal at 1 kHz by tuning the optomechanics at both the transmitter and receiver indicates optimal alignment. After optimal alignment is achieved, the detected signal from the QCD output is amplified with a 30 dB-gain electrical low-noise amplifier (LNA) and collected with a 40 GSa/s real-time digital storage oscilloscope (DSO). Finally, the sampled digital signal is processed with a matched filter, timing recovery, and downsampling process based on maximum variance, a symbol-spaced decision-feedback equalizer (DFE) with 55-feedforward taps, and 55-feedback taps, and the BER performance is evaluated after the symbol-to-bit demodulation.

III. EXPERIMENTAL RESULTS
This section presents the system-level experimental results. In our experiment, we operate the QCL at three different temperature values, i.e., 10°C, 5°C, and 0°C, and explore the maximum supported data rates, respectively. Temperature   Fig. 4 shows the characterized end-to-end amplitude and phase response of the system at 10°C temperature, including the AWG, QCL, QCD, DSO, and all the electrical components that play a part in between. It's worth noting that for all temperature cases, the frequency response measurements were performed at optimal QCL bias points, taking into account the trade-off between SNR and linearity. For 10°C, the end-to-end 3-dB bandwidth is found to be around 400 MHz, and the 10-dB bandwidth is around 1.6 GHz. In addition, two static, narrow frequency dips are observed at 580 MHz and 950 MHz, induced by the external bias-tee and the impedance mismatch between the bias-tee and the QCL RF input. These frequency dips are equalized digitally with both pre-and post-equalizers. Fig. 4(b) shows the measured phase response, which stays smooth up to the 3-dB bandwidth and starts fluctuating at higher frequencies due to the influence of noise in the system, limiting the measurement stability and accuracy.

A. Results at 10°C
After the frequency response characterization at 10°C, we evaluate the system transmission performance with the two modulation formats, namely, NRZ-OOK and PAM4. At this temperature, maximum bit rates for each modulation format are explored against the 6.25%-OH HD-FEC limit. Moreover, because there is no variable attenuator in the experimental setup, we would face a lack of accuracy in tuning the optomechanics. Therefore, we keep the link setup untouched after optimizing the beam alignment and evaluate the performance of the system transmission by sweeping the bias current of the DM-QCL, which subsequently changes the received optical power at the detector. Understandably, the bias current point of the laser is also correlated to the modulation linearity. Fig. 5(a) shows the BER performance as a function of the laser bias for three achievable signal configurations: NRZ-OOK at 6 Gbaud and 7 Gbaud, and PAM4 at 4 Gbaud, and the top axis is the corresponding received optical power at the detector. The optimum point of the bias current was detected at 650 mA, yielding 14.5 mW received optical power at the QCD to attain a sufficient signal SNR and the linearity simultaneously. Around the optimal bias point, BER performances below the 6.25%-OH HD-FEC limit are achievable for all three test cases. The 20%-OH HD-FEC and the KP4-FEC thresholds are shown in the figure only for reference. Further increasing the PAM4 baud rate to above 4 Gbaud would degrade the BER performance beyond the 6.25%-OH HD-FEC threshold, limited by the receiver SNR due to insufficient output power from the QCL at this temperature, as previously shown in Fig. 1(b). Selected eye diagrams after the post-equalizer for all the three cases at the optimal bias point are shown in Fig. 5(b). Clear eye openings with negligible nonlinear distortions are observed.

B. Results at 5°C Temperature
We then decrease the QCL operating temperature to 5°C and perform the same system-level evaluation as at 10°C to explore the data rate limit. A higher data rate can be expected owing to higher output power from the QCL at a lower temperature. First, we characterize the system's end-to-end frequency response, and the measured results are shown in Fig. 6. One can observe only slight improvements in the 3-dB and the 10-dB bandwidth, whereas the same frequency dips remain at 580 MHz and 950 MHz due to unchanged hardware configuration. Due to the noise limit, similar phase fluctuation beyond the 3-dB bandwidth is again observed in Fig. 6(b). Fig. 7 shows the BER performance as a function of the QCL bias point. At this temperature, we manage to increase the baud rate of PAM4 to 5 Gbaud with BER performance below the 6.25%-OH HD-FEC limit. The optimal bias point throughout the current sweeping for PAM4 was found around 630 mA, corresponding to the received optical power of 15.9 mW. Fig. 7(b) shows the selected eye diagrams at different laser bias points for the 5 Gbaud PAM4 signal. It clearly indicates the compression of lower-amplitude levels at 600 mA for the PAM4 signal as it is too close to the QCL lasing threshold. On the other hand, increasing the bias current beyond the optimal point causes compression on the upper amplitude levels. It causes severe signal distortion at 680 mA, despite the high achievable SNR at this point. For the case of NRZ-OOK, as shown in Fig. 7(a), increasing from 7 Gbaud to 8 Gbaud fails to meet the targeted BER threshold of 4.5×10 −3 due to the limited bandwidth and SNR. Another observation is that the optimal bias for NRZ-OOK is higher than PAM4 due to its higher tolerance to modulation nonlinearity.

C. Results at 0°C Temperature
Finally, we characterize the system and evaluate the transmission performance at 0°C. As shown in Fig. 8, similar frequency response in both amplitude and phase is obtained as operating at 5°C. Fig. 9(a) shows the BER performance as a function of the laser bias for the two modulation formats. Due to the increased QCL output power, the highest achievable data rate can be further improved at this temperature. The optimal bias point was found between 630 mA and 640 mA, where both the 5.5 Gbaud PAM4 and the 8 Gbaud NRZ-OOK could achieve BER values that fulfill the 6.25%-OH HD-FEC requirement. Further increasing the PAM4 baud rate to 6 Gbaud fails to achieve BER performance below the targeted FEC threshold of 4.5×10 −3 BER. Fig. 9(b) shows the equalized signal eye diagrams at several selected bias points for both 5.5 Gbaud PAM4 and 8 Gbaud NRZ-OOK. Again, the trade-off between the signal SNR and modulation linearity can be clearly observed by looking at the eye diagrams, particularly in the case of PAM4. A higher coding-gain FEC configuration, 20%-OH HD-FEC, may be used for further data rate enhancement, but it is considered an unfavored option due to its increased latency and complexity.
The transmission results achieved at all temperatures benchmarked against different FEC thresholds are summarized in Table I. Comparing the obtained results across all the tested temperature and bias cases, we conclude that the transmission performance is primarily determined by the trade-off between the SNR and the modulation linearity, as the impacts due to other factors such as wavelength shift and QCD thermal dissipation are identified negligible per our evaluation. Finally, it is worth noting that enhanced transmission speed and performance can be well    expected by further decreasing the QCL operation temperature below 0°C to achieve higher QCL output power. However, it requires extra measures to prevent condensation on the QCL mount, which we will implement in our next-phase studies

IV. CONCLUSION
We demonstrated high-speed LWIR FSO transmissions at 9.6 µm wavelength with a directly-modulated QCL and a fully passive QCD without any cooling or bias voltage. The system supports up to 11 Gb/s bitrate with BER below the 6.25%-OH HD-FEC limit, verifying excellent bandwidth and linearity characteristics of both the DM-QCL and the QCD. This work is considered a significant milestone in the uncharted territory bitrate of >10 Gb/s LWIR FSO and shows the optimistic possibility of DM-QCL and QCD for the future ICT infrastructure and is a significant step towards a fully connected high-speed mid-IR FSO network, envisioned to be a promising alternative to the RF-based wireless systems.