1. Introduction

A laser warning receiver (LWR) is a passive optical detection technology that alerts a user when laser light is incident upon the receiver. LWRs have found military applications in the detection of LADAR systems, and civilian application in alerting drivers to the presence of laser radar detectors operated by law enforcement. Typically, these LWR systems employ direct detection (photon counting) methods for the identification of laser light, relying on the assumption that laser signals, operating in a narrow optical band, will be much brighter than the incoherent environmental background into which the field of view of the LWR is looking. While this assumption is likely valid for a laser directly incident on the LWR, a desirable performance feature would be the ability to detect a laser even before it hits the LWR. The sensitivity to detect laser light scattered from the primary beam via Mie scattering processes (as shown in Fig. 1) would enable such a capability, however Mie processes are highly-directional favoring forward scattering [1,2]. Therefore optical intensities at large scattering angles ($\theta _{SC} > 10^{\circ }$), and in an environment with natural daylight make direct detection of these scattering processes challenging for typical laser powers and sensor standoff distances [3]. Experiments have demonstrated the detection of atmospherically scattered laser light using direct detection receivers [2,4–8]; and, recently, the fundamental quantum limit to discriminating coherent laser light and incoherent noise light was demonstrated [9]. These measurements require ultra-sensitive detectors, narrow-band optical filtering and often operation at nighttime for sufficient noise suppression to perform detection of scattered laser light. Moreover, the requirement for narrowband optical filtering requires that the receiver know, a-priori, the wavelength of the laser for which it is searching, a burdensome demand.

 

Fig. 1. A shot-noise limited balanced coherent detection (BCD) receiver is used to detect the presence of a laser by collecting atmospherically scattered light. Even in the presence of strong broadband background noise the BCD can discriminate the presence of weakly scattered laser light.

Download Full Size | PDF

We present the results of an experiment that demonstrates coherent heterodyne detection of near infrared laser light scattered off-axis in a broadband daylight background. The receiver uses no optical filtering, and a non-cooperative, tunable laser source for the local oscillator for the coherent detection receiver. The signal processing algorithm developed for our laser detection system demonstrated nearly perfect detection with zero false alarms for an integration time $>40 \mu s$. Coherent detection of atmospherically scattered laser light is widely used in Doppler LADAR systems where the same laser that generates the signal light is used to implement the local oscillator to detect the laser light scattered from atmospheric particulates for measurement of wind fields [10,11]. This mono-static architecture using a single laser guarantees coherence and deterministic wavelength matching between the signal and local oscillator. In this work we implement a bi-static architecture, using non-cooperative lasers, and demonstrate that coherent detection can be implemented successfully to detect weak laser light, even after undergoing scattering, and reject incoherent background light.

2. Balanced coherent detection receiver

A diagram illustrating the physical sensing scenario and the design for balanced coherent detection (BCD) is shown in Fig. 1. A laser beam (with frequency $f_{S}$) propagating through an atmospheric channel interacts with atmospheric aerosols and undergoes Mie scattering. A small fraction of the laser light is scattered from the channel and received by an aperture which couples the scattered laser light, along with environmental noise light, into a single-mode optical fiber directing the light to a coherent receiver tasked with determining whether or not laser light is present in the channel. The BCD receiver mixes the signal light with a local oscillator laser (with frequency $f_{LO}$) in a 50:50 beamsplitter and the resulting mixture is directed to two photodetectors with matched responsivities which generate photocurrents proportional to the optical power detected. These two photocurrents are then subtracted from each other and the resultant current is sent to an amplifier with transimpedance gain ${GT}$, resulting in an output voltage $V_{S} = G\left (I_{1} - I_{2}\right )$. The balanced coherent detection receiver was first proposed in [12]. The coherence time of the broadband environmental noise light is vanishingly small and therefore there is no coherent interference with the narrowband local oscillator laser in the receiver on timescales relevant to electrical bandwidths, resulting in $V_{S} = 0$ at the output of the receiver. Received laser light, however, does generate coherent interference under the assumption that the optical bandwidth of the laser is narrow (within the electrical bandwidths of the BCD) and the wavelength is near the wavelength of the local oscillator (both within the electrical bandwidth of the BCD receiver). The concept for balanced coherent detection (BCD) was first introduced in [12] as a method to reject excess intensity noise in the local oscillator laser. A schematic of the experiment is shown in Fig. 2. A RIO Grande amplified, fixed frequency, continuous wave laser with wavelength $\lambda _{S} = 1549.646$ nm served as our signal laser. This laser is capable of generating $\sim 200$ mW of laser light and reports kHz-scale linewidths. A mechanical variable optical attenuator (VOA) was used to control the intensity of the signal laser. The signal laser was coupled to free-space through a fiber collimator (Thorlabs F220 APC-1550) and a half-wave plate was used to rotate the linear polarization of the light. The signal laser was then directed through a glass cell containing environmental particulates collected from both indoor and outdoor settings. The sample contains a mixture of environmental dust and pollen. The particulates in the jar were agitated by a motor such that it was approximated to be evenly distributed throughout the cell as the laser passed through the cell. The cell was positioned such that the beam passed directly through the center of the cell minimizing scattering due to interaction with the glass cell walls. We placed a fiber collimator (also Thorlabs F220 APC-1550) at an angle $\theta _{SC}=45^{\circ }$ relative to the signal laser propagation direction to collect signal laser light scattered from the particles in the cell. This fiber collimator couples free-space light with a mode field diameter of 2 mm into a single-mode optical fiber with $9 \mu m$ core. Laser light scattered from the dust particles in the cell, as well as background light, is collected by the fiber collimator and coupled to a single-mode optical fiber and directed into the signal input to the balanced coherent detection system. An optical isolator (Thorlabs IO-H-1550FC) was inserted into the setup to prevent the local oscillator from leaking through the receiver optical setup and into the free-space channel. The balanced coherent detection receiver was operated with a tunable laser source (Ando AQ-4321) with a tuning range $\lambda _{LO} = 1520 - 1600$ nm, a tuning resolution of 1 pm and reports 200 kHz linewidth. The local oscillator laser power was $900 \mu W$ ensuring that the BCD receiver operated at the shot-noise limit. The collected signal light was mixed with the local oscillator laser in a 50:50 fiber beamsplitter and the two outputs of the beamsplitter were passed through VOAs and into an amplified balanced detector (Thorlabs PDB 425C), the output of which was captured by an oscilloscope operating as the data acquisition system. The VOAs connecting the mixing beamsplitter to the balanced detector are used to match the path loss between the arms, as any small path loss difference will result in an amplified error signal at the output of the balanced detector. Alignment between the signal and receiver spatial modes was established using a single laser as both the signal and local oscillator, with the signal and receiver trained on a fixed scatterer to observe a homodyne signal when the scattered light was successfully coupled to the receiver. Once this alignment was established, the local oscillator input was replaced with the tunable laser source. The tunable laser wavelength was scanned to search for frequency alignment with the signal laser wavelength observing the beating signature in the BCD.

 

Fig. 2. Experimental setup: The signal laser is attenuated with a mechanical variable optical attenuator (VOA) coupled to free-space with a fiber collimator (FC). The signal laser is directed through a glass cell containing dust particles at the scattering site. Scattered laser light is collected by the receiver, coupled into fiber and directed through an optical isolator (OI) to the BCD receiver.

Download Full Size | PDF

The receiver collects light scattered by particles that are located within the intersectional volume ($V_{int}$) defined by the overlap between the spatial mode of the signal laser and the spatial mode of the receiver collimator. This intersectional volume was calculated to be $V_{int}\sim 9\cdot 10^{-3}$ milliliters. The range from this intersection location to the receiver was approximately 50 cm, limited by the geometry of our optical table. While we did not directly measure the particle size distribution in our sample, careful studies have shown that aerosols of this type (mixture of pollen and dust) has a mean diameter $\sim 10 \mu m$ [1] and mass distribution of $10 - 100$ nanograms/grain evaluating a broad spectrum of reference material (e.g. [13,14]). We estimate the average number of particles in $V_{int}$ by weighing our sample (800 mg) and dividing by the referenced particle mass (50 ng/grain), to be $\sim 150$ particles. Typical values for dust concentration in the atmosphere are between 1-10 $\mu$g/m³ [15] substantially lower than our concentrations. We chose high concentrations due to the limited acquisition time of our experiment to ensure that we would have scatterers in our intersectional volume while we acquired data. Signal traces were collected, comprised of 20 million data points sampled at 250 MHz, with a total length of 80 ms. We collected data for a variety of signal laser powers ranging from 7 dBm - 20 dBm. While we could not directly verify the scattered laser power we collected, we performed simulations of the Mie scattering process using the MiePlot program [16] and estimate that we collected $\sim 2 - 40$ pW of scattered laser light over this range of signal laser intensity. We also collected data with the signal laser extinguished so that we could generate measurements for both probability of detection ($P_{D}$) and probability of false alarm ($P_{FA}$) for the BCD receiver to detect laser light. An example data capture from the oscilloscope is shown in Fig. 3 illustrating the coherent interference between the signal and local oscillator laser as measured by the BCD receiver. The two lasers are separated in optical frequency by $\sim 8$ MHz, but over the duration of data capture (40 ms) the power spectral density shows that the frequencies of the two lasers drift relative to one another by $\sim 5$ MHz.

 

Fig. 3. The Fourier transform of a 40 ms capture of the signal and local oscillator mixing in the balanced coherent detection receiver. For this illustrative data both lasers were delivered to the balanced coherent detection system via fiber. The data illustrates the frequencies of the laser drifting relative to one another over a range of 5 MHz during this time period. (Inset) A small ($\sim 1 \mu s$) section of the time-dependent signal at the output of the balanced coherent detection receiver.

Download Full Size | PDF

As an independent test of the capability of the BCD to reject incoherent background light we coupled light from an incandescent bulb (Thorlabs QTH10) with high intensity in the wavelength band of interest ($\sim 0.5 pW$) into the receiver while detecting a laser signal of 25 fW. There was no measurable degradation in the output signal ($V_{S}$) from the BCD receiver, as it is insensitive to incoherent light.

3. Data analysis and results

We analyzed raw signal data collected by the oscilloscope from the BCD receiver to identify characteristics of the signal that could be used to determine if an active laser was in the field-of-view of the coherent receiver. When analyzing signals in the frequency domain, we observed that data obtained when the laser was active consistently exhibited peaks in the power spectral density (PSD) in the >10 MHz range. With sufficient acquisition time, these peaks are consistently $\geq 1$dB above the PSD floor and we used this feature of the measured signal as a sufficient statistic for laser detection.

To process the raw data from the BCD receiver we accounted for several features in the signal and noise floor. There was a divergence in the PSD at frequencies <1 MHz owing to excess noise in our local oscillator laser. Also, the electrical bandwidth of the balanced detector is 75 MHz and the oscilloscope has an internal low pass filter of 250 MHz, so the magnitude of the PSD rolls off at frequencies >100 MHz. Finally, the center frequency, spectral width and spectral shape of the signal peaks (example in Fig. 3) is random providing little a-priori information for automated peak detection. The signal processing procedure we implemented (1) high-pass filtered the signal with a cutoff of 1 MHz, (2) calibrated the spectral features at higher frequencies (>100 MHz) and (3) integrated the result to generate the variable ($\hat {\varrho }$) used as the decision variable to determine if a laser is present.

We processed data collected with signal laser power ranging between 7 dBm and 20 dBm. At each power setting, we collected 80 ms data bursts at 250 MHz sampling rate. To simulate shorter acquisition times we extracted segments from the measurements corresponding to acquisition time ranging from 4 $\mu$s to 40 $\mu$s. For each simulated acquisition time, we extracted as many independent segments (e.g. no overlap between samples) as possible from the 80 ms acquisition. Dividing the data in this way enabled us to generate 200 to 20000 independent samples for each laser power setting, depending on the simulated acquisition time. Additionally, we extracted segments of the same time lengths from a signal collected when there was no active laser present. All segments were processed using the signal processing chain described earlier, and the output of the chain was used to generate receiver operating characteristic (ROC) curves to enable us to study coherent receiver performance at different laser power settings and acquisition times.

The ROC curve plots the probability of successful detection of a laser ($P_{D}$) against the false-alarm probability ($P_{FA}$) for indicating a laser detection when no laser is present. To generate the ROC we pick a threshold value ($\varrho _{TH}$) for the integral of the PSD and for a given signal laser power, and acquisition time if $\hat {\varrho }>\varrho _{TH}$ we determine that a laser is present, and if $\hat {\varrho }<\varrho _{TH}$ we determined no laser was present. Our detection probability is computed as $P_{D} = P\left (\hat {\varrho }>\varrho _{TH}|\textbf {laser}\right )$ and our false alarm probability is computed as $P_{FA} = P\left (\hat {\varrho }>\varrho _{TH}|\textbf {no laser}\right )$. We repeated this process for each of the acquisitions for a given signal laser power, and swept the value $0< \varrho _{TH}<\infty$ to generate the ROC curves.

Sample detection results are shown in Fig. 4, which shows ROCs plotted for data collected with the coherent receiver with a 20 dBm laser active as simulated acquisition time varies. As was expected, detection accuracy increases as acquisition time increases. In fact, we found that as long as acquisition time was greater than 20 $\mu$s, we were able to achieve 99$\%$ detection accuracy or greater with at most with 1$\%$ false positive rate. Furthermore, at acquisition time of 100 $\mu$s or greater, we were able to completely distinguish all simulated acquisitions collected with the laser active from acquisitions when there was no laser present. Similar detection results were seen at all laser powers that we experimented with. In fact, we found that at 100 $\mu$s acquisition time or greater, we were able to determine whether or not the laser was active with all measurements taken during our experiments.

 

Fig. 4. (Left) Calculated receiver operating characteristic (ROC) curves for detection of 20 dBm laser with acquisition time ranged from 4 $\mu$s to 40 $\mu$s. The black dashed line sets the threshold for zero detection sensitivity. (Center) ROC plots on logarithmic scale to illustrate the deviation from near unity detection. The legend is shared with the (Left) plot. (Right) Area under the Curve (AUC) for the ROC plots as a function of signal laser power and integration time.

Download Full Size | PDF

Detection performance of the coherent receiver from all measurements across all powers and simulated acquisition times are summarized in Fig. 4. This plot shows the area under the curve (AUC) calculated by integrating the ROCs generated from processed data for each laser power at each simulated acquisition time. Laser detection improved at shorter acquisition times as the power of the laser increases. This is shown through the fact that AUC increases at lower acquisition times when laser power is higher. We suspect that this is due to the fact that more scatter passes through the receiver’s FOV as laser power increases, resulting in a stronger measured signal. At 1 $\mu$s acquisition time or longer, calculated AUC is >0.95.

4. Conclusions

The work presented here demonstrates that BCD can be used for the detection of laser light by detecting weakly scattered light from atmospheric particulates, such as dust. While our demonstration was performed in a laboratory, the intensity of the scattered laser light we detected (<50 pW) is similar to that measured in large-scale scattering experiments conducted in marine environments [1], where they measured radiance values of $\sim 1 nW/m^{2}$ received per Watt of source laser power for a $45^{\circ }$ scattering angle, as was configured in our experiment.

The BCD receiver constitutes an important capability in LWR technology as it rejects strong incoherent background noise light while simultaneously having sensitivity over a broad wavelength range given a tunable local oscillator. The results presented demonstrate detectability of a scattered laser signal within a 100 MHz optical bandwidth ( 0.8 pm), defined by the electrical bandwidth of the balanced detector. As we demonstrated excellent detectability in acquisition time $<20\mu s$ over this bandwidth, we can expect that with a tunable local oscillator laser scanning over a wavelength range of 10 nm (comprised of approximately $1.25\cdot 10^{4}$ of these frequency bins) would require <250 ms.

In this work, our detection algorithm employed a human designed signal processing chain. We are currently assessing data driven anomaly detection algorithms such as [17] to further improve laser detection performance. Additionally, the signal processing capabilities can be easily enhanced to extract parameters of the signal laser, such as wavelength, linewidth and frequency stability – which direct detection solutions cannot provide.