1. Introduction

In 3D modelling and sensing [1–4], lidars are essential sensors for providing 3D images with high-precision range information at long distances [5–9]. By measuring the time-of-flight (ToF) between transmitted and received light, object ranges can be determined [10,11]. In typical pulse and frequency-modulated continuous-wave (FMCW) lidars, repetitive pulses or linearly chirped waveforms are used to probe the surroundings. Inevitably, by emitting unspecific and repetitive waveforms, interference between multiple lidars, ambient light, or malicious jamming can result in detection failures or performance degradation [12–14].

To mitigate the effect of interference and jamming, correlation-based lidars, such as random-modulated continuous-wave (RMCW) lidars, have been proposed. Using a pseudorandom code modulated CW laser as a transmitting laser source, an RMCW lidar was first proposed and investigated by Takeuchi et al. in 1983 [15]. In RMCW lidars [15–19], high-speed electronics and electro-optic modulators are needed to modulate the laser light with pseudo-random bit sequences. Compared to pulse lidars, which generally have a higher peak power, RMCW lidars often require a longer integration time to achieve a sufficient signal-to-noise ratio (SNR) for successful detection. However, this limits the detection throughput and may not be suitable for applications such as autonomous vehicles, drones, sport analyses, and real-time 3D modelings that require high-speed detection.

To reduce the need for high-speed modulators and electronics, continuous-wave (CW) chaos lidars based on the nonlinear dynamics of semiconductor lasers have been proposed and studied [20,21]. With aperiodic and uncorrelated transmitted chaos waveforms, chaos lidars have low range ambiguity and are resistant to interference and jamming. To improve SNR and detection throughput while still complying with the Class-1 eye-safe regulation [22] for practical applications, a 3D pulsed chaos lidar system emitting chaos-modulated pulses has been studied [23]. Employing a fiber-Bragg grating (FBG) as a filter to enhance the energy efficiency and a master oscillator power amplifier to increase the peak power, a high-speed 3D chaos lidar system with sub-centimeter range precision and 100 kHz detection throughput has been demonstrated [24–26].

Although chaos lidars have good immunity to interference and jamming due to the aperiodic nature of the chaos-modulated pulses used, to better match the broad bandwidth of laser chaos (several GHz) and the limited detection bandwidth of the avalanche photodetector (APD) (hundreds of MHz) to have better energy efficiency, filters such as an FBG have to be used to prevent amplifying those high-frequency modulations exceeding the detection bandwidth [25]. In such a setup, because the wavelengths of both the laser and FBG are temperature dependent (typically 20 pm/$^\circ$C for the laser and 2 pm/$^\circ$C for the FBG), the output power and the stability of the system are, however, sensitive to temperature variation.

Another way of generating chaos-modulated pulses using a gain-switched semiconductor laser subject to optical feedback has also been studied [27]. Although implementing the light module with a gain-switched laser can eliminate the need for optical pulse generators or time-gating devices (such as an acousto-optical modulator, electro-optical modulator, or semiconductor optical amplifier), in such a scheme, the feedback delay time has to be carefully adjusted and synchronized relative to the pulse repetition interval (PRI) of the gain-switched modulation [28]. For a PRI greater than 1 $\mu$s to have sufficient peak power in detection, for example, a fiber longer than 100 m has to be used in the delay loop, which limits system integration.

To generate efficient random- or chaos-modulated pulses for lidar applications without the use of a temperature-sensitive FBG and a long delay fiber for synchronized optical feedback, in this study we investigated the generation of gain-switched random-modulated pulses using a delayed self-homodyne interferometer (DSHI) [23,29–31]. As intensity instability and wavelength drift occur at different stages of a gain-switched pulse [32–34], we found that, by sending it into a DSHI, beat modulations with random or down-chirped oscillations can be generated at different stages of the pulse. We studied the temporal and spectral characteristics of the gain-switched random-modulated pulses under different DSHI delay lengths ($L$), switching currents, and pulsewidths. We quantified the SNR, range precision, and cross-correlation between consecutive pulses of pulses generated under different operational conditions. To show that gain-switched random-modulated pulses are suitable for lidar applications, we established a random-modulated pulse lidar using a gain-switched semiconductor laser with a DSHI to demonstrate high-precision 3D imaging.

2. Experimental setup

Figure 1(a) shows the schematic setup of a gain-switched semiconductor laser (LD) with a DSHI. To generate random-modulated pulses, we used a pulse driver (Aerodiode CCS-std, rising time 5 ns) to gain-switch a single-mode distributed-feedback (DFB) LD (Shengshi Optical SWLD-155005P8-FC/APC-PM, threshold current: 35 mA) with different switching currents and pulsewidths at a pulse repetition frequency (PRF) of 100 kHz. The LD output was coupled in and out of the DSHI through two 50:50 fiber couplers (FC). To extract the instantaneous wavelength variation of the LD’s wavelength across the entire stage of each gain-switched pulse, we used a tunable laser (TL) (Yenista Optics, Tunics T100S, 400 kHz linewidth) as the local oscillator to beat with the LD output. In the DSHI, the optical path length in one arm was fixed and in the other arm was varied by adding optical fibers with different lengths. After the DSHI, we used two FCs to split the light into four channels where the power of each channel was controlled by an optical variable attenuator (OVA). In one channel, we used an optical spectrum analyzer (OSA) (Advantest Q8384, wavelength resolution 10 pm, measurement interval 1 ms) to monitor the optical spectra of the LD output. In another channel, we used a high-speed photodetector (PD) (Newport 1544-B, 12 GHz bandwidth) followed by an oscilloscope (OSC 1) (Agilent Technologies DSO-X 92504A, 25 GHz bandwidth, 80 GS/s) with a broad detection bandwidth and high sampling rate to record the original waveforms of the LD output.

 

Fig. 1. (a) Schematic setup of random-modulated pulses generation based on a gain-switched semiconductor laser with a delayed self-homodyne interferometer (DSHI). LD: laser diode; FCs: fiber couplers; OVA: optical variable attenuator; APD 1 and APD 2: avalanche photodetectors; PD: high-speed photodetector; OSC 1 and OSC 2: oscilloscopes; OSA: optical spectrum analyzer. (b) Schematic of the reference and signal waveforms from APD 1 and APD 2 with relative delay $\tau$ and their cross-correlation trace.

Download Full Size | PDF

While APDs with large gain are essential in lidar applications to detect the scarce backscattered light from the diffuse targets, their detection bandwidths are, however, limited to only a few hundred MHz. To characterize the performance of the random-modulated pulses generated in correlation-based lidar applications, we sent the other two channels to two APDs (APD 1 and APD 2) (Thorlabs APD430C, 400 MHz bandwidth) followed by another oscilloscope (OSC 2) (Tektronix MSO58, 500 MHz bandwidth, 1.25 GSa/s). In a correlation-based ranging scheme, one of the waveforms detected (APD 1) is used as a reference and the other (APD 2) is used as a signal. Through searching the time lag $\tau$ of the peak in the cross-correlation trace, as illustrated in Fig. 1(b), the relative path difference between APD 1 and APD 2 (or the range of a target) can be measured by multiplying $\tau$ with the speed of light $\it {c}$ [23,24]. Although acquisition with broader bandwidth and higher sampling rate can surely benefit range resolution and precision in detection, we intentionally used the APD 1, APD 2, and OSC 2 with moderate specifications to investigate the more realistic bandwidth-limited condition in practical lidar applications.

3. Instantaneous frequency detuning in a gain-switched semiconductor laser

To show how the central wavelength or instantaneous oscillation frequency of the LD is varied during the gain-switched modulation, Figs. 2(a) and 2(d) depict the optical spectra of the gain-switched LD at switching currents of 50 mA and 100 mA with a pulsewidth of 500 ns. As can be seen, with a long integration interval of 1 ms, the spectra are substantially broadened compared to the linewidth of 0.011 nm when the LD is in its CW mode. The broadening is attributed to the wavelength drift that occurred at different stages of the gain-switched modulation when the carrier density of the LD is varied [32–34]. To obtain the instantaneous frequency of the LD, we coupled the light from the TL as a local oscillator beating with the LD output. As indicated by the dashed arrow lines, the center wavelengths of the TL were set to 1550.98 and 1550.99 nm, while the center wavelengths of the LD in its CW state at switching currents of 50 and 100 mA were 1551.03 and 1551.37 nm, respectively.

 

Fig. 2. Optical spectra of the gain-switched semiconductor laser with switching currents of (a) 50 and (d) 100 mA and pulsewidth of 500 ns. The dashed arrow lines depict the central wavelengths of the TL at 1550.98 and 1550.99 nm, respectively. (b) and (e) show the waveforms of the gain-switched pulse beat with the TL. The insets show the oscillations of part of the waveforms. (c) and (f) show their instantaneous spectra obtained from wavelet analyses. The color of the spectra denotes their normalized amplitudes.

Download Full Size | PDF

By beating the LD output with the TL, Figs. 2(b) and 2(c) and 2(e) and 2(f) show the waveforms and their instantaneous spectra from wavelet analyses obtained at switching currents of 50 and 100 mA, respectively. The insets in Figs. 2(b) and 2(e) show the oscillations of part of the waveforms. As an alternative to windowed Fourier transforms, wavelet analyses provide resolutions both in time and frequency to reveal the instantaneous oscillation frequencies as the pulses evolve [35,36]. In the wavelet analyses, we used a second-order Chebyshev high-pass digital filter with a cut-off frequency of 100 MHz to remove the square-wave modulation in the waveforms from the gain-switching. These spectra show the instantaneous beat frequencies between the LD and TL, where the color denotes their normalized amplitudes.

As can be seen in Figs. 2(b) and 2(c), by abruptly varying the carrier density with gain-switched modulation [32–34], the beat frequencies spike at the beginning of the pulse and then gradually decrease in a down-chirped trend as the LD stabilizes toward its CW state. The LD almost reaches its CW state when the modulation is terminated after 500 ns. As we increased the switching current to 100 mA, as shown in Figs. 2(e) and 2(f), the beat frequencies spike and fluctuate significantly in the first 50–70 ns, where an instantaneous bandwidth of a few GHz is observed. As the pulse evolves, the detuning clearly shows a down-chirped trend, and the LD takes a relatively longer time to stabilize toward its CW state. At the end of the pulse, the beat frequencies spike again when the carrier density is abruptly reduced to terminate the pulse. Note that, although the wavelength of the TL can be arbitrarily chosen, here it was set so that most of the beat frequencies throughout the pulse can fall within the detection bandwidth of the PD for better observation. By interfering these pulses with time-dependent frequency variations in the DSHI, depending on the switching currents and relative delays between the two self-homodyne paths, waveforms with random or chirped-like beat frequencies at different stages of the pulses can be generated.

When there is no relative delay between the two paths, there will be no beat frequency between the pulses of the two paths, because at any instant there is no detuning. As the delay increases, the detuning increases, so that in general waveforms with higher beat frequencies can be generated after the DSHI. From the steeper down-chirped rate shown in Fig. 2(f), higher beat frequencies are also expected when operating at a higher switching current.

4. Characteristics of gain-switched random-modulated pulses generated under different operational parameters

In Figs. 3–5, we show the waveforms and spectra of the gain-switched pulses after the DSHI with different $L$, switching currents, and pulsewidths. The waveforms are two consecutive pulses truncated from pulse trains with repetition intervals of 10 $\mu$s. Their spectra are derived from wavelet analyses and the color denotes their normalized amplitudes.

 

Fig. 3. (a) Waveforms and (b) spectra of gain-switched random-modulated pulses generated with $L$ of (i) 0, (ii) 20, (iii) 140, and (iv) 200 cm at a switching current of 120 mA and a pulsewidth of 100 ns, respectively. Waveforms are two consecutive pulses truncated from pulse trains with repetition intervals of 10 $\mu$s. Their instantaneous spectra are obtained from wavelet analyses, where the color denotes the normalized amplitudes.

Download Full Size | PDF

 

Fig. 4. (a) Waveforms and (b) spectra of gain-switched random-modulated pulses generated with switching currents of (i) 50, (ii) 70, (iii) 120, and (iv) 180 mA at a $L$ of 200 cm and a pulsewidth of 100 ns, respectively. Waveforms are two consecutive pulses truncated from pulse trains with repetition intervals of 10 $\mu$s. Their instantaneous spectra are obtained from wavelet analyses, where the color denotes the normalized amplitudes.

Download Full Size | PDF

 

Fig. 5. (a) Waveforms and (b) spectra of gain-switched random-modulated pulses generated with pulsewidths of (i) 50, (ii) 100, (iii) 150, and (iv) 500 ns at a $L$ of 200 cm and a switching current of 120 mA, respectively. Waveforms are two consecutive pulses truncated from the pulse trains with repetition intervals of 10 $\mu$s. Their instantaneous spectra are obtained from wavelet analyses, where the color denotes the normalized amplitudes.

Download Full Size | PDF

Figures 3(a)(i–iv) and 3(b)(i–iv) show the waveforms and spectra of the gain-switched pulses after the DSHI with $L$ of (i) 0, (ii) 20, (iii) 140, and (iv) 200 cm for a switching current of 120 mA and a pulsewidth of 100 ns, respectively. When there is no relative delay in the DSHI, as shown in Figs. 3(a)(i) and 3(b)(i), there is no detuning between the two DSHI paths at any instant. Consequently, no beat oscillation is found in the waveforms of the pulses generated after the DSHI. The amplitudes of the reddish regions around 120 MHz in the spectra observed at the beginning of the pulse are actually very small before normalization. This is attributed to the unstable oscillation at the onset of the pulse.

When we increase $L$ to 20 cm to introduce relative detuning between the two pulses in the DSHI, as shown in Figs. 3(a)(ii) and 3(b)(ii), waveforms of random modulations and spectra with broad bandwidths from the beating of the two frequency-detuned pulses are observed. The range of the spectra is limited by the respective cut-off frequency of the high-pass digital filter and the detection bandwidth of the APD used. Note that, the pulsewidth does increase slightly (increase 1 ns for every $L$ of 20 cm in fiber) after passing the DSHI with a relative path delay of $L$.

When $L$ is further increased to 140 cm, the detuning between the two pulses becomes even larger. As shown in Figs. 3(a)(iii) and 3(b)(iii), the oscillation frequencies in the middle of the pulses exceed the detection bandwidth, where only weak modulations are generated. Before the termination of the gain-switched pulses, strong oscillations are observed where the detuned frequencies are lowered and revealed within the detection bandwidth. As shown in Figs. 3(a)(iv) and 3(b)(iv), when $L$ is increased to 200 cm, the detuned frequencies are even higher and the amplitudes of modulations are reduced due to the detection bandwidth limit.

Figures 4(a)(i–iv) and 4(b)(i–iv) show the waveforms and spectra of the gain-switched pulses after the DSHI with switching currents of (i) 50, (ii) 70, (iii) 120, and (iv) 180 mA at a $L$ of 200 cm and a pulsewidth of 100 ns, respectively. As can be seen in Figs. 4(a)(i) and 4(b)(i), with a low switching current, the beat frequencies of the random-modulated waveforms are relatively lower at around 150 MHz. When the switching current is increased to 70 mA, as shown in Figs. 4(a)(ii) and 4(b)(ii), the detuning becomes higher and the beat frequencies become higher accordingly. Note that for the two consecutive pulses shown, their waveforms and instantaneous spectra are not identical. This dissimilarity prevents them from interfering with each other if they reach the detector simultaneously.

As the switching current is further increased to 120 mA, as shown in Figs. 4(a)(iii) and 4(b)(iii), the beat frequencies increase and exceed the detection bandwidth in the middle of the pulses. Further increasing the switching current to 180 mA, as shown in Figs. 4(a)(iv) and 4(b)(iv), the beat frequencies are well above the detection bandwidth and almost no modulation is generated.

For the waveforms and spectra shown in Figs. 5(a) and 5(b), we varied the pulsewidths from (i) 50, (ii) 100, and (iii) 150 ns to (iv) 500 ns, respectively. The $L$ is fixed at 200 cm and the switching current is 120 mA. To have the same average power, the magnitudes and peak power decrease as the pulsewidth increases. As shown in Figs. 5(a)(iv) and 5(b)(iv), for the pulsewidth of 500 ns, the beat frequencies decrease from approximately 400 MHz to 100 MHz, revealing the instantaneous detuning between the two DSHI pulses. For this gain-switched laser, it reaches its CW state in approximately 500 ns, when the wavelength is almost stabilized. For shorter pulsewidths, as shown in Figs. 5(a)(i–iii) and Figs. 5(b)(i–iii), the modulation in the waveforms and the corresponding instantaneous spectra were simply terminated by the gain-switched modulation before the laser reaches its stabilized state.

Although a gain-switched laser with a DSHI under different $L$, switching currents, and pulsewidths can generate pulses with different modulation waveforms, as shown in Figs. 3–5, for applications such as random-modulated pulse lidars, the generation of waveforms with a large SNR, good range precision, and low cross-correlation between consecutive pulses is of particular interest [25]. Figures 6(a–c) show the SNR, precision, and cross-correlation peaks to consecutive pulses ($\mu _p$) of the pulses generated under different $L$ and switching currents with a pulsewidth of 100 ns, respectively. The SNR is defined as the ratio between the standard deviation of the modulation amplitude in the gain-switched pulse waveform (signal, S) to the standard deviation of the noise in the off region of the pulse (noise, N). To evaluate the precision under different conditions, we calculated the correlation traces between the waveforms of the same pulse received by APD 1 (reference) and APD 2 (signal). By extracting the time lags of the cross-correlation peaks from the traces and calculating the relative path lengths they represent, the precision is obtained by taking the standard deviation of 100 measurements. To estimate the possible interference between consecutive pulses generated at different times, we calculate the cross-correlations of a pulse with the next 100 consecutive pulses of the pulse train. We take the average of the normalized values of these 100 cross-correlation peaks and denote it by $\mu _p$. With a lower $\mu _p$, or when pulses are much different from each other, range ambiguity and detection interference can be mitigated when the detector received multiple pulses simultaneously. In contrast, with a higher $\mu _p$, or when pulses are similar to each other, receiving multiple pulses with similar waveforms at the same time could jeopardize valid detection [26,37–39].

 

Fig. 6. Mappings of (a) SNR, (b) precision, and (c) $\mu _p$ of the pulses generated under different switching currents and homodyne delay length $L$ at a pulsewidth of 100 ns. Mappings of (d) SNR, (e) precision, and (f) $\mu _p$ of the pulses generated under different switching currents and pulsewidths at a $L$ of 200 cm.

Download Full Size | PDF

Note that when calculating SNR, precision, and $\mu _p$, we used a second-order Chebyshev high-pass digital filter with a cut-off frequency of 100 MHz to remove the square-wave modulation in the gain-switching waveforms. Although a native range resolution of 12 cm is deduced from the 1.25 GSa/s sampling rate used, we employ Spline interpolation [40] to further improve precision when searching for time lags of the cross-correlation peaks. All correlation windows are set to be equal to the pulsewidths.

As can be seen in Fig. 6(a), when $L$ is shorter than approximately 20 cm, the beat frequencies between the two DSHI pulses are still small. Being filtered by the high-pass digital filter used with a cut-off frequency of 100 MHz, the SNR is generally low. For moderate $L$ and switching currents, SNRs above 16 dB (reddish) are observed over a wide region of the parameter space. For an $L$ too long or a switching current too high (toward the upper right corner), the SNR decreases as beat frequencies gradually exceed the detection bandwidth.

Governed by the Cramer and Rao lower error bound [24,41,42], the ideal precision is inversely proportional to the signal bandwidth, SNR, and integration time or sampling points. As can be seen, the precision shown in Fig. 6(b) has a good correlation with the corresponding SNRs shown in Fig. 6(a). In a broad region, a precision below approximately 1 mm (bluish region) is obtained.

The $\mu _p$ shown in Fig. 6(c) measures the similarity of the pulses and quantifies the correlation of the consecutive pulses generated at different times. Notably, when $L$ is short, the pulses of the two DSHI paths maintain good correlation, thus, stably generating similar waveforms in the consecutive pulses in the gain-switched pulse train. When $L$ is increased, the relative phases at any instant between pulses become less correlated and thus the similarity in each generated pulse is reduced. As shown in Fig. 6(c), increasing $L$ generally reduces $\mu _p$. As for the switching current, the LD biased near the threshold is not as stable as when it is driven at a higher switching current. At a low switching current, the random fluctuation especially at the onset of the pulse attributes to the variation of both the waveforms and spectra in the consecutive pulses and results in lower $\mu _p$. As can be seen, increasing the switching current generally increases $\mu _p$.

Figures 6(d–f) show the SNR, precision, and $\mu _p$ for pulses with $L$ of 200 cm under different pulsewidths and switching currents, respectively. As can be seen in Fig. 6(d), when the pulsewidth increases, the SNR decreases as the peak power decreases. When the switching current increases, especially when the pulsewidth is short (approximately 100 ns in the lower right corner), the SNR decreases as the beat frequencies gradually exceed the detection bandwidth.

In Fig. 6(e), the precision generally shows a good correlation with the SNR shown in Fig. 6(d), except when the pulsewidths are shorter than approximately 70 ns (lower left corner). In this region, the pulsewidths are too short, so the precision worsens due to the limited sampling points [42]. For the $\mu _p$ plotted in Fig. 6(f), longer pulsewidths and lower switching currents help to reduce the correlation between consecutive pulses. For long pulsewidths and low switching currents (upper left corner), $\mu _p$ lower than approximately 0.4 (blueish region) are obtained.

While high SNR, good precision, and low $\mu _p$ are generally favorable for lidar applications, the $L$, switching current, and pulsewidth have to be carefully chosen and optimized depending on each specific experimental realization. For the generated pulses shown in Fig. 4(a)(ii) (parameters marked with the star symbol in Fig. 6), a favorable SNR of 16.76 dB, a good precision of 1.28 mm, and a low $\mu _p$ of 0.35 are simultaneously obtained. While the $\mu _p$ of the gain-switched random-modulated pulses may fall short or be on par with the chaos-modulated pulses generated by laser feedback dynamics [25], the precision of the gain-switched scheme is noticeably better. More details on the precision are discussed in Fig. 8.

5. Demonstration of 3D imaging using a gain-switched random-modulated pulse lidar

To demonstrate its performance and feasibility, Fig. 7 shows the experimental setup of a gain-switched random-modulated pulse lidar system. The architecture and devices used in the light source module are identical to those shown and described in Fig. 1(a), where the laser output is sent to a DSHI to generate the gain-switched random-modulated pulses. Part of the pulses is sent to APD 1 to be used as the reference and the remaining is sent to an Erbium-doped fiber amplifier (EDFA, GIP CGB1E3128001A) to increase the emitting power. Under a PRF of 100 kHz, there is no significant waveform distortion after the EDFA [24]. In the optical transceiver module, the amplified random-modulated pulses are collimated and then scanned by a two-axis microelectromechanical systems (MEMS) mirror (Mirrorcle A8L2.2). We used two 2-inch Fresnel lenses to collect the backscattered light from the target and detected it with APD 2 as the signal. Both the signal and reference waveforms are captured and recorded by the oscilloscope (OSC) (Tektronix MSO58, 500 MHz bandwidth, 1.25 GSa/s) in the signal acquisition and processing module. We used a personal computer (PC) to calculate the cross-correlations and to control the MEMS controller for system synchronization. A trigger is sent from the MEMS controller to the function generator (FG, Agilent 81150A) and then to the pulse driver for pulse emissions and to the OSC for waveform acquisitions.

 

Fig. 7. Experimental setup of a gain-switched random-modulated pulse lidar. LD: laser diode; FC: fiber coupler; APD 1 and APD 2: avalanche photodetectors; EDFA: Erbium-doped fiber amplifier. FG: function generator. OSC: oscilloscope. PC: personal computer.

Download Full Size | PDF

Figure 8 shows the range precision at different SNRs of a diffused target located at 1 m. While the range resolution is only useful when two or more partially opaque targets are being emitted and detected at the same time, in most practical scenarios, range precision is a more important metric to evaluate the performance of a lidar system. For comparison, the precision of the gain-switched homodyne scheme investigated in this study (blue) and an optical feedback scheme based on laser dynamics studied previously in Ref. [25] (black) are plotted. The dashed curves are their fittings. The operational parameters of the gain-switched homodyne scheme used are as follows: $L$ of 160 cm, switching current of 70 mA, and pulsewidth of 100 ns, which are selected to have both a high SNR and an optimal precision simultaneously. For both schemes, the pulsewidths and average power received by the APDs are set to be equal.

 

Fig. 8. Precision of random-modulated pulses of the gain-switched homodyne scheme (blue) and chaos-modulated pulses of the optical feedback scheme (black) under different SNRs. The dashed curves are their fittings.

Download Full Size | PDF

As can be seen, while both schemes show similar trends, the gain-switched homodyne scheme performs better in terms of precision. For a good SNR of more than 12 dB, the optimal precision of the gain-switched homodyne scheme is approximately 1 mm, while that of the feedback scheme is approximately 3 mm. The precision can be viewed as a measure of the consistency of the time lag in the cross-correlation trace. Stable and repetitive waveforms in nature could have better precision than random or chaotic ones. By closely examining the waveforms of the gain-switched random-modulated pulses that have better precision, we found that they are mixtures of both random and chirped oscillations. While the random oscillations part contributes to lower $\mu _p$ for better interference resistance similar to that of the chaos-modulated pulses from the feedback scheme, the part containing consistent chirped oscillations in every consecutive pulse contributes to better precision. Depending on the applications and detection scenarios, as depicted in Fig. 6, modulated waveforms with different random and chirped oscillations constitutions, and thus with the corresponding precision and $\mu _p$, can be generated by selecting the appropriate operational parameters.

In Figs. 9 and 10, we demonstrate high-precision 3D imaging using the gain-switched random-modulated pulse lidar system. Figures 9(a) and 9(b) are the front and side views of a 4 cm diameter table tennis ball positioned 75 cm away from the lidar as the target. Scanning the emitted light using the MEMS mirror and detecting 100x100 pixels ranges in a frame, Figs. 9(c) and 9(d) show the front and side views of the 3D point clouds colored by their SNRs. For comparison, Figs. 9(e) and 9(f) also show the result obtained with the chaos-modulated pulse lidar based on optical feedback [25]. As can be seen in Figs. 9(c) and 9(e), with the same average output power of the EDFA (10 dBm), the highest SNRs of 17.6 and 16.1 dB are obtained near the center of the ball for the gain-switched and optical feedback schemes. While the SNR varies with the reflectance and range of the target, the scanning center along the detector line of sight typically has higher SNR from the better coupling. Having better precision at all SNRs, as shown in Figs. 9(d) and 9(f), the depths (ranges) measured by the gain-switched scheme are more precise and most of the point clouds congregate along the ideal surface of the ball depicted by the dashed semicircle. Consequently, the gain-switched scheme shows better performance in applications that requires high-range precision, such as 3D modeling.

 

Fig. 9. (a) Front and (b) side views of the table tennis ball as the target. (c) Front (d) and side views of the 3D point clouds obtained with the random-modulated pulse lidar based on the gain-switched homodyne scheme. (e) Front (f) and side views of the 3D point clouds obtained with the chaos-modulated pulse lidar based on the optical feedback scheme. The color of the point clouds denotes their SNRs. The dashed semicircles depict the ideal surface of the ball.

Download Full Size | PDF

To show the feasibility of the random-modulated pulse lidar based on the gain-switched homodyne scheme, in Fig. 10 we demonstrate the 3D imaging of a human face under eye-safe conditions. Figures 10(a) and 10(b) show the front and side views of the target, which is located 4 m away from the lidar system. The field-of-view of the scanned area is set at 5 degrees and the image resolution at 100x100 pixels. By scanning the target at a pixel rate of 100 kHz, the front and side views of the acquired point clouds are plotted in Figs. 10(c) and 10(d). The color denotes the SNRs of the point clouds, where the highest SNR of 19.0 dB is obtained between the eyebrows. The background point clouds are filtered for image clarity.

 

Fig. 10. (a) Front and (b) side views of the model’s face located approximately 4 m away from the lidar system. (c) Front (d) and side views of the 3D point clouds obtained with the random-modulated pulse lidar based on the gain-switched homodyne scheme. The color of the point clouds denotes their SNRs.

Download Full Size | PDF

Notably, even with the low reflectance and the diffusivity of human skin, a 3D image with a clear outline of the face is successfully captured. Detailed features such as the forehead, eye sockets, nose, lips, and chin are distinguishable. As a result, we have demonstrated the feasibility of the proposed random-modulated pulse lidar for practical applications.

6. Conclusion

In conclusion, we have proposed and characterized the generation of random-modulated pulses based on a gain-switched semiconductor laser with DSHI. Without the need for a temperature-dependent FBG to improve energy efficiency, a long optical fiber to provide synchronized feedback, and lossy time-gating devices or external modulators to generate pulses, the proposed scheme is more resistant to temperature variation and can be more easily integrated for practical applications. Favorable for lidar applications such as autonomous vehicles, drones, sport analyses, and real-time 3D modelings, waveforms simultaneously achieving high SNR, good precision, and low $\mu _p$ can be generated under various operational parameters. For a good SNR of more than 12 dB, the optimal precision of the proposed gain-switched homodyne scheme is approximately 1 mm, which is approximately three times better than the optical feedback scheme. Compared to the commercial ToF pulse lidars that have precision in the centimeter range using repetitive pulses, the proposed scheme is not only resistant to interference but also capable of providing depth information with greater detail. By establishing a random-modulated pulse lidar based on the gain-switched homodyne scheme, we successfully demonstrated 3D imaging of a table tennis ball and 3D profiling of a human face with good precision.