A Search for Technosignatures from 14 Planetary Systems in the Kepler Field with the Green Bank Telescope at 1.15–1.73 GHz

We verified the validity of our data-processing pipeline by analyzing the monochromatic tone data and recovering the signal at the expected frequency. We also folded the pulsar data at the known pulsar period and recovered the characteristic pulse profile.

We unpacked the data to 4-byte floating point values, computed Fourier transforms of the complex samples with the FFTW routine (Frigo & Johnson 2005), and calculated the signal power (Stokes I) at each frequency bin. Signals with frequencies outside the range of the L-band receiver (1150–1730 MHz) were discarded. We used the GBT's notch filter to mitigate interference from a nearby aircraft surveillance radar system. Signals with frequencies within the 3 dB cutoff range of the notch filter (1200–1341.2 MHz) were also discarded.

Individual channels of the GUPPI instrument exhibited a mostly uniform bandpass response, with a few notable exceptions. We fit a 16-degree Chebyshev polynomial to the median bandpass response of well-behaved channels. We divided each power spectrum by this median response, which normalized signal levels across the entire bandpass.

We used a Fourier transform length of 2²⁰, corresponding to a time interval of Δτ = 0.336 s and yielding a frequency resolution Δf = 2.98 Hz. This choice of transform length and frequency resolution was dictated by our desire to examine drift rates of order 10 Hz s⁻¹ (Section 3.5). We stored about 450 consecutive power spectra, depending on the exact integration time of each scan, in frequency–time arrays of 2²⁰ columns and 432–451 rows (hereafter, "time–frequency diagrams," sometimes known as "spectrograms" or "spectral waterfalls"). The average noise power was subtracted and the array values were scaled to the standard deviation of the noise power.

Because radio signals experience time-variable Doppler shifts due to the rotational and orbital motions of both emitters and observers, uncompensated signals smear in frequency space on a timescale ${\rm{\Delta }}f/\dot{f}$, where Δf is the frequency resolution and $\dot{f}$ is the Doppler drift rate. The duration of a time series required to obtain a spectrum with resolution Δf is 1/Δf, such that the maximum drift rate that is observable without smear is ${\dot{f}}_{\max }=\pm {({\rm{\Delta }}f)}^{2}$. The maximum Doppler drift rates due to Earth's rotational and orbital motions are ∼0.15 Hz s⁻¹ and ∼0.034 Hz s⁻¹, respectively, at the maximum frequency of our observations and at Green Bank's latitude. The Doppler drift rates due to the emitters are unknown. We examined drift rates of up to ${\dot{f}}_{\max }=\pm {({\rm{\Delta }}f)}^{2}=\pm 8.88$ Hz s⁻¹, corresponding to accelerations of ±1.6 m s⁻², which admit a wide range of planetary radii, spin rates, orbital semimajor axes, and orbital periods. We applied a de-smearing procedure to compensate for the accelerations of both emitter and observer. Specifically, we implemented a computationally advantageous tree algorithm (Taylor 1974; Siemion et al. 2013), which enabled examination of 512 Doppler drift rates from 0 to 8.88 Hz s⁻¹ in linearly spaced increments of 0.0173 Hz s⁻¹. This algorithm reads the frequency–time arrays, then shifts and sums all powers corresponding to each of the 512 possible drift rate values, effectively enabling integration of the signal power over the entire scan duration without smear. Application of the algorithm with positive and negative Doppler drift rates resulted in two frequency–drift rate arrays of 2²⁰ columns and 512 rows for each scan.

We identified candidate signals with an iterative procedure. We searched for the element with the highest S/N in the frequency–drift rate arrays. The characteristics of this candidate signal (unique identifier, source name, scan number, scan start time, frequency at start of scan, drift rate, S/N, frequency resolution) were stored in a SQL database for subsequent analysis. Because a candidate signal would often be detected redundantly at multiple drift rate values adjacent to that with the highest S/N, we decided to keep only the instance with the highest S/N value. In order to do so, we blanked the frequency–drift rate arrays in a region of frequency extent $\pm {\dot{f}}_{\max }\tau$ centered on the frequency of the highest S/N candidate signal. We then repeated the procedure and searched for the element with the next highest S/N. All candidate signals with S/N > 10 were identified in this fashion and stored in the database. We counted 858,748 candidate signals, which amounts to ∼750,000 candidates per hour of on-source integration time in the useful frequency range of the GBT L-band receiver.

A signal from a distant source at rest or in uniform motion with respect to the observer exhibits no time variation in the value of the Doppler shift. Signals from extraterrestrial sources, unless cleverly compensated for a specific location on Earth, experience a nonzero Doppler drift rate due in part to the rotational and orbital motions of Earth. For these reasons, we categorized all signals with zero Doppler drift rate as likely terrestrial and eliminated them from further consideration. About a quarter (231,181) of the candidate signals were flagged on this basis, leaving 627,567 candidates with nonzero Doppler drift rates.

To further distinguish between radio-frequency interference (RFI) and genuine extraterrestrial signals, we implemented two additional filters. First, we flagged any signal that was not detected in both scans of the same source. This filter can rule out many anthropogenic signals that temporarily enter the beam (e.g., satellite downlinks). Second, we flagged any signal that appeared in more than one position on the sky. This filter can rule out many anthropogenic signals that are detectable through the antenna sidelobes. A logical AND was used to automatically flag candidate signals that remained for consideration after the rejection steps. Our rejection filters used the scan start times, frequencies, Doppler drift rates, and frequency resolutions stored in the SQL database to properly recognize signals from the same emitter observed at different times. These filters successfully flagged 617,410 of the remaining signals as likely anthropogenic, leaving 10157 signals for further investigation.

Overall, our rejection filters automatically eliminated 99% of the initial detections as RFI.

Several regions of the spectrum exhibit an unusually high density of detections (Figure 1). Most of these high-signal-density regions can be attributed to known interferers. We discarded all candidate signals in these regions, which we defined by the frequency extents shown in Table 2. For signals generated by global navigation satellite systems, we used the signal modulation characteristics, typically binary phase-shift keying (BPSK), to delineate the frequency extent. For satellite downlinks, we used the Federal Communications Commission table of frequency allocations. On average, the density of detections in the frequency regions ascribed to these satellites is ∼3500 detections per MHz of bandwidth. In contrast, the density of signals in regions of the spectrum that exclude these interferers is ∼500 detections per MHz of bandwidth.

Zoom In Zoom Out Reset image size

We flagged 9663 candidate signals out of 10157 as most likely due to these known interferers, leaving 494 signals for further consideration.

After we excised the known interferers listed in Table 2, we identified hundreds to thousands of detections with common time–frequency characteristics in narrow regions of the spectrum spanning a total of ∼9 MHz (Table 3). Because these classes of signals were each detected in at least six distinct pointing directions, they are almost certainly anthropogenic, and we flagged them as RFI. We flagged 456 candidate signals out of 494 as most likely due to RFI, leaving 38 signals for further consideration.

We detected strong interference in the radio astronomy protected band (1400–1427 MHz). The interferers are visible in histograms of our detections in two 9-MHz-wide regions of the spectrum (Figure 2). We briefly describe the characteristics of these interferers below.

Zoom In Zoom Out Reset image size

The interferers near 1396, 1400, 1403, and 1406 MHz are similar. They have a comb-like appearance in the frequency domain, with spectral features spaced every kilohertz. Each feature has a bandwidth of about 140 Hz and resembles the modulation of a double-sideband suppressed-carrier transmission. In addition to these features, strong carriers are observed at 1396.18, 1396.75, and 1402.94 MHz. Most of these interferers are in the radio astronomy protected band.

The region of the spectrum near 1401.5 MHz is characterized by a number of narrow lines that seem to cluster near 10 discrete frequency regions. They have small (mostly <0.2 Hz s⁻¹) positive or negative Doppler drift rates. Some of these lines exhibit somewhat erratic behavior as a function of time, perhaps indicating an unstable oscillator. The small Doppler drift rates suggest a terrestrial source.

The interferer near 1422 MHz generates a broad (∼270 kHz) region of increased noise power without distinct lines, making the identification of the Doppler drift rate difficult.

The region between 1423 and 1429 MHz exhibits some interferers that are approximately 70 Hz wide and reminiscent of those described by Siemion et al. (2013, their Figure 5). Others are time-variable signals (12–15 s periodicity) that are approximately 70 Hz wide. These characteristics are similar to those of some Air Route Surveillance Radars (ARSR) that track aircraft in all azimuthal directions at 5 rotations per minute, except that these systems are designed to operate between 1250 and 1350 MHz. If due to ARSR, it is unclear whether the interference is caused at the source or generated by intermodulation products in the GBT receiver system. The sinc-type appearance of the histogram (Figure 2) is reminiscent of the power spectrum of a BPSK waveform with 1 MHz bandwidth and compressed pulse width of 1 μs. Such waveforms are used to provide radar imaging with 150 m resolution (e.g., Margot et al. 2000). Some ARSR systems use nonlinear frequency modulation (NLFM) to provide a nominal range resolution of 116 m. Some NLFM schemes exhibit a sinc-like power spectrum. In addition to the periodic interferers, strong carriers with variable drift rates are observed near 1425.05 and 1428.18 MHz.

Figure 3 shows a graphical summary of the frequency regions that were excluded from our analysis.

Zoom In Zoom Out Reset image size

The 38 remaining signals represent 19 pairs of scans. The characteristics of the first scans are shown in Table 4. We generated time–frequency diagrams for all 19 pairs. We then searched our database for signals at similar frequencies and compared their time–frequency diagrams to those of our top candidates. This process revealed that all 19 candidates are detected in more than one direction on the sky, ruling out the possibility of an extraterrestrial signal. Examination of the time–frequency diagrams revealed groups of signals that can be attributed to the same source of RFI. We provide a brief description of the signals below, along with a few examples of time–frequency diagrams. To our knowledge, these interferers have not been reported in the literature, although they may of course have been detected in other searches.

The candidate signal near 1151 MHz (Figure 4) is a monochromatic signal with a substantial Doppler drift rate that is observed in at least eight distinct directions on the sky.

Zoom In Zoom Out Reset image size

The candidate signal near 1375 MHz has a complex time–frequency structure that is observed in at least one other direction on the sky.

The candidate signal near 1414 MHz (Figure 4) exhibits both high S/N and somewhat erratic frequency behavior. The rejection filter logic likely failed because the Doppler behavior of this signal is erratic.

The candidate signal near 1444 MHz is a low S/N monochromatic signal that is observed in at least four other directions on the sky.

Both candidates near 1453 MHz exhibit a set of intermittent narrow lines that are observed in multiple directions on the sky.

All five candidates near 1457 MHz are due to anthropogenic RFI and share the same characteristics, i.e., a monochromatic signal superimposed on a broadband, elevated noise level.

The candidate signal near 1461 MHz is a single line that is observed in at least two directions on the sky. Moreover, signals with similar lines that are offset in frequency by almost exactly 20 and 30 kHz are observed in other directions on the sky.

The candidate signal near 1467 MHz (Figure 4) exhibits a set of monochromatic lines. The strongest one exceeded our S/N threshold. It is observed in at least four directions on the sky.

The candidate signal near 1472.254 MHz is detected in at least three directions on the sky.

The candidates near 1472.262 MHz correspond to the same signal that is observed in at least two distinct directions.

The candidates near 1485 and 1501 MHz are monochromatic signals that are detected in at least two and six directions on the sky, respectively.

The candidate signal near 1623 MHz has a complex time–frequency structure that is observed in several directions on the sky. An approximately linear ramp with slope −150 Hz s⁻¹ appears then disappears with on and off durations of 5–10 s.

Although no extraterrestrial signals were identified in our analysis to date, we emphasize that our study encompassed only a small fraction of the search volume. The fraction of the sky that was covered in our search is 14 times the solid angle of the GBT beam. At the center frequency of our search, each beam has a solid angle of 0.015 deg². Considering all 14 sources, we covered 0.21 deg² or about 5 ppm of the entire sky. Our observations lasted a total of 5 minutes on each source, which is about 10 ppm of a terrestrial year. After the elimination of polluted bands, our useful bandwidth spanned almost exactly 300 MHz, which is a small fraction of the electromagnetic spectrum available for transmission.

We computed the Drake figure of merit (Drake 1984; Enriquez et al. 2017) that corresponds to our search parameters:

$$\begin{eqnarray}&&\mathrm{DFM}=\displaystyle \frac{{\rm{\Delta }}{f}_{\mathrm{tot}}{\rm{\Omega }}}{{S}_{\det }^{3/2}},\end{eqnarray} \tag{ 1 }$$

where ${\rm{\Delta }}{f}_{\mathrm{tot}}=300$ MHz is the total bandwidth examined, Ω = 0.21 deg² is the total area of the sky covered, and S_det = 10 Jy is the minimum flux density required for a detection (Section 4.3). In these units, we find DFM ≃ 2 × 10⁶. Referring to the values provided in Enriquez et al. (2017), our search is about 150 times larger than that of Horowitz & Sagan (1993) and 10 times larger than that of Gray & Mooley (2017). It amounts to about 2% and 12% of the recent large surveys by Enriquez et al. (2017) and Harp et al. (2016), respectively.

Attempting to place existence limits on the basis of SETI observations is a difficult exercise. Certainly, we can place limits only on the kinds of signals that we are looking for, not actual limits on the presence of civilizations. For instance, Enriquez et al. (2017) attempted to place a limit on the number of 100%-duty cycle transmitters (e.g., a radio beacon) and suggested that fewer than 0.1% of the stellar systems within 50 pc possess such transmitters. However, beacons operating at frequencies lower than 1.1 GHz, larger than 1.9 GHz, or in the 1.2–1.34 GHz range would be undetected in their (and our) search, which makes general claims about the number of beacons unreliable. In this spirit, we describe the types of signals that are detectable with our search, but we do not attempt to make inferences about the prevalence of radio beacons in the Galaxy.

For the detection of narrowband signals above a floor with noise fluctuations, the S/N can be expressed as ${P}_{r}/{\rm{\Delta }}{P}_{\mathrm{noise}}$, where P_r is the received power and ΔP_noise is the standard deviation of the receiver noise given by

$$\begin{eqnarray}&&{\rm{\Delta }}{P}_{\mathrm{noise}}=\displaystyle \frac{{k}_{{\rm{B}}}{T}_{\mathrm{sys}}{\rm{\Delta }}f}{\sqrt{{\rm{\Delta }}f\tau }},\end{eqnarray} \tag{ 2 }$$

with k_B Boltzmann's constant, T_sys the system temperature, Δf the frequency resolution, and τ the integration time (Ostro 1993; Naidu et al. 2016). The power received by a transmitter of power P_t and antenna gain G_t located at distance r is

$$\begin{eqnarray}&&{P}_{r}=\displaystyle \frac{{P}_{t}{G}_{t}}{4\pi {r}^{2}}\displaystyle \frac{{A}_{e}}{2},\end{eqnarray} \tag{ 3 }$$

where A_e is the effective area of the receiving station and the factor of 0.5 accounts for reception by a single-polarization feed. If powers received in both polarizations are added incoherently as in this work, ${\rm{S}}/{\rm{N}}=\sqrt{{n}_{\mathrm{pol}}}\times {P}_{r}/{\rm{\Delta }}{P}_{\mathrm{noise}}$, with n_pol = 2. The S/N is proportional to the product of factors that relate to the transmitter-receiver distance, transmitter performance, receiver performance, quantization efficiency, and data-taking and data-analysis choices, as shown in this expression:

$$\begin{eqnarray}&&{\rm{S}}/{\rm{N}}=\left(\displaystyle \frac{1}{4\pi {r}^{2}}\right)({PtGt})\left(\displaystyle \frac{{A}_{e}}{2{k}_{{\rm{B}}}{T}_{\mathrm{sys}}}\right)({\eta }_{Q}){\left(\displaystyle \frac{{n}_{\mathrm{pol}}\tau }{{\rm{\Delta }}f}\right)}^{1/2}.\end{eqnarray} \tag{ 4 }$$

The second factor is known as the effective isotropic radiated power (EIRP) and the third factor is the inverse of the SEFD, so we can rewrite the S/N as

$$\begin{eqnarray}{\rm{S}}/{\rm{N}} & = & 9{\left(\displaystyle \frac{100\mathrm{ly}}{r}\right)}^{2}\left(\displaystyle \frac{\mathrm{EIRP}}{{10}^{13}\,{\rm{W}}}\right)\left(\displaystyle \frac{10\,\mathrm{Jy}}{\mathrm{SEFD}}\right)\\ & & \times \left(\displaystyle \frac{{\eta }_{Q}}{1}\right){\left(\displaystyle \frac{{n}_{\mathrm{pol}}}{1}\right)}^{1/2}{\left(\displaystyle \frac{\tau }{1{\rm{s}}}\right)}^{1/2}{\left(\displaystyle \frac{1\mathrm{Hz}}{{\rm{\Delta }}f}\right)}^{1/2}.\end{eqnarray} \tag{ 5 }$$

The S/N is maximized when the frequency resolution of the data matches the bandwidth of the signal.

We used the Arecibo planetary radar as a prototype transmitter with P_t = 10⁶ W, G_t = 73.4 dB, and EIRP = 2.2 × 10¹³ W. The nominal receiver is the GBT with A_e = 5600 m², T_sys = 20 K, and SEFD = 10 Jy. The quantization efficiency of the four-level sampler is η_Q = 0.8825. Our data-taking and data-analysis choices correspond to n_pol = 2, τ = 150 s, and Δf = 3 Hz. With these values, we find ${\rm{\Delta }}{P}_{\mathrm{noise}}=3.9\times {10}^{-23}$ W and P_r = 3.1 × 10⁻²² W for a transmitter located at a distance of 420 ly (128 pc) from Earth. Such transmissions would be at the limit of detection with our minimum S/N threshold of 10.¹⁰

The target stars with known distances in our sample are located at distances that exceed 420 ly, such that a more powerful transmitter, a more sensitive receiver, longer integration times, or narrower frequency resolutions would be needed for detection. Transmitters with considerably larger EIRPs than that of Arecibo may be available to other civilizations. A transmitter with 1000 times Arecibo's EIRP would be detectable in our search from distances of up to 13250 ly. Current technology enables the detection of technosignatures emitted from a large fraction of the Galaxy (Figure 5). In such a vast search volume, there are billions of targets accessible to a search for technosignatures. In contrast, the search for biosignatures will be limited in the foreseeable future to a few targets in the solar system and to a few hundred planetary systems around nearby stars.

Zoom In Zoom Out Reset image size

One can use Figure 5 to evaluate the detectability of the Arecibo planetary radar by other civilizations. From an S/N standpoint, an increase by a factor of 10 in EIRP is equivalent to an increase in effective area by a factor of 10. In other words, the colored lines in Figure 5 also represent the detectability of Arecibo by remote antennas with effective areas that are 1–1000 times larger than those of the GBT or Arecibo, assuming similar system temperatures and search parameters (n_pol = 2, τ = 150 s, and Δf = 3 Hz). Equation (5) can be used to evaluate detectability with other transmitter, receiver, or search parameters. Typical planetary radar transmissions have a duty cycle of approximately 50%, and the maximum tracking duration for celestial sources at Arecibo is ∼2.5 hr. For observations of main belt asteroids and more distant bodies at opposition, the pointing direction over the entire tracking duration changes by less than 2 arcmin, i.e., less than the width of the beam for S-band (2.38 GHz) radar transmissions.