Passive all-sky imaging radar in the HF regime with WWV and the first station of the Long Wavelength Array



[1] We present a new passive, bistatic high-frequency (HF) radar system consisting of the transmitters for the radio station WWV and the dipole antenna array that comprises the first station of the Long Wavelength Array (LWA) or “LWA1.” We demonstrate that these two existing facilities, which are operated for separate purposes, can be used together as a unique HF radar imager, capable of monitoring the entire visible sky. In this paper, we describe in detail the techniques used to develop all-sky radar capability at 10, 15, and 20 MHz. We show that this radar system can be a useful tool for probing ionospheric structure and its effect on over-the-horizon (OTH) geolocation. The LWA1+WWV radar system appears to be especially adept at detecting and characterizing structures associated with sporadic-E. In addition, we also demonstrate how this system may be used for long-distance, OTH mapping of terrain/ocean HF reflectivity. Finally, we discuss the potential improvements in the utility of these applications as more LWA stations are added.

1 Introduction

[2] High-frequency (HF; 3–30 MHz) radars have been used for decades for several applications. Operating as both monostatic and bistatic radar systems, ionosondes have been and continue to be used to sound the ionosphere, probing the evolution of the ionospheric electron density profile. They remain valuable assets for both basic ionospheric research and as support systems for operational radars. So-called over-the-horizon (OTH) radars were developed after World War II to exploit the ionosphere as a “virtual mirror” that could reflect both outgoing and returning HF signals to track OTH targets. OTH radars remain a valuable long-distance surveillance tool used, for example, to monitor illegal narcotics trafficking in the Caribbean (see Headrick and Thomason [1998] for a more thorough discussion).

[3] As detailed by Headrick and Thomason [1998], the transmitter portion of operational OTH radars are particularly cumbersome and difficult to develop given the large size required for high-power HF transmitting antennas. Consequently, some work has been devoted to developing passive radar or passive coherent location (PCL) in the HF band. Such PCL systems use a matched filter-based technique to obtain range/Doppler information from particular targets using transmitters of opportunity. Efforts to develop and exploit these passive, bistatic radars have focused on the upper end of the very high frequency (VHF) regime, typically utilizing commercial radio and television broadcasts for the transmitted signals [see, e.g., Griffiths and Long, 1986; Howland, 1986; Gunner et al., 2003; Baker et al., 2005; Griffiths and Baker, 2005; Howland et al., 2005; Kulpa and Czekala, 2005]. It has also been demonstrated that such systems can be used to study ionospheric structure [e.g., Meyer and Sahr, 2004].

[4] However, relatively recently, there have been efforts dedicated to developing PCL techniques and systems in the HF regime [see, e.g., Ringer and Frazer, 1999; Thomas et al., 2006; Fabrizio et al., 2008; Xianrong et al., 2011]. The new long wavelength (> 10 m) radio broadcasting format, Digital Radio Mondiale, which is recently being adopted by radio stations in many countries, has been the focus of much of this effort [see, e.g., Thomas et al., 2006; Xianrong et al., 2011].

[5] This paper details the development of a novel HF passive radar system with two unique attributes. First, the transmitters broadcast a pulsed signal, allowing for radar applications without the need for sophisticated matched filtering techniques. Specifically, the transmitters used are those of the National Institute of Standards and Technology (NIST) radio station located near Ft. Collins, Colorado, call-sign WWV. WWV broadcasts the current time at five different HF frequencies, including a 5 ms pulse at the beginning of each second. Second, the receiver array is a relatively dense array of 256 dipole antennas capable of producing reasonably high fidelity all-sky images. This receiver array is the newly operational first station of the Long Wavelength Array (LWA), referred to as LWA1 [Taylor et al., 2012; Ellingson et al., 2013], located in central New Mexico. The full LWA will consist of 52 stations spread throughout the state of New Mexico designed to work as a large HF/VHF interferometer/telescope for imaging cosmic sources [Taylor et al., 2006]. Each station will be a roughly 100 m wide-phased array of 256 crossed dipole antennas. The unique capability of LWA1 to image the entire sky rather than point at one particular location makes it a particularly flexible radar receiving system.

[6] We will demonstrate that when used together, LWA1 and WWV constitute a unique all-sky passive HF radar imaging system that could potentially operate continuously and relatively cheaply. We describe the techniques involved (sections 2–3) as well as examples of possible applications of the system (section 4) and future capabilities as more LWA stations are added (section 5).

2 LWA1 and WWV

2.1 LWA1

[7] LWA1 is the first station of the planned LWA which will consist of more than 50 similar stations, acting together as an HF/VHF, interferometric telescope. Like LWA1, each station will consist of 256 cross dipole antennas capable of operating between 10 and 88 MHz arranged in a quasi-random pattern spanning roughly 100 m designed to minimize sidelobes. Each dipole antenna consists of a wire grid “bow-tie” mounted to a central mast with arms angled downward at a roughly 45° angle. This design maximizes sensitivity/throughput over the entire observable sky. For a detailed discussion of the antenna design, see Hicks et al. [2012]. LWA1 is currently being operated as an independent telescope/observatory located at 34.070°N, 107.628°W and is described in detail by Taylor et al. [2012] and Ellingson et al. [2013], including a description of the calibration of the array using bright cosmic sources and an additional “outrigger” antenna located roughly 300 m east of the array center.

[8] The antennas can be operated as a single-phased array with up to four simultaneous beams. Each beam can use up to two tunings, each with a passband having up to 16 MHz of usable bandwidth. Independent from the beam-forming mode, LWA1 also has a transient buffer (TB) mode that records the individual output from each antenna, allowing one to beam form to any point on the sky after the fact and making all-sky imaging possible.

[9] The TB can be operated in one of two modes. The first is a wideband mode (TBW) which captures 61.2 ms of the raw signal from each antenna at a sampling rate of 196 Msps with 12 bits per sample. With this data, one can use the full bandwidth accessible by the antennas. However, since the data volume is so large (> 9 GB per capture), it takes more than 5 min to write each capture to disk and thus cannot be run in a continuous way. Conversely, the narrowband mode (TBN) tunes each antenna signal to a specified central frequency with a sampling rate of up to 100 ksps and up to 67 kHz of usable bandwidth. The data rate for this mode is small enough that it can be run continuously for up to roughly 20 h before filling the available disk space. Examples of all-sky images of cosmic sources made using TBN data and the LWA1 Prototype All-Sky Imager [Ellingson et al., 2013] can be found at∼lwa/lwatv.html, which is updated in near real time as data are taken. An example of an image of the entire visible sky constructed from TBN data was also presented by Dowell et al. [2012].

[10] It is the two TB modes that are particularly useful for passive radar applications. Because they allow for all-sky imaging, any terrestrial transmitter whose direct signal, ground wave, or sky wave (ionospheric reflection) is observable from LWA1 can be detected and located with either TBW or TBN data. In cases where both the direct signal and the ground wave and/or sky wave are simultaneously detected (e.g., a local frequency modulation station), the transmitter signal can be obtained by beam forming toward the transmitter and this can be used as a matched filter for obtaining ranges for the ground and/or sky wave detections according to Meyer and Sahr [2004]. Alternatively, because LWA1 uses its own Global Positioning System clock to time stamp each TB sample, a transmitter with a known pulse-like signal structure like WWV (see section 2.2) can also be used for passive radar with good time-of-arrival accuracy (RMS variation of about 20 ns ≃ 6 m).

2.2 WWV

[11] WWV is an HF radio station that broadcasts the time from the Ft. Collins, Colorado area (40.7°N, 105.0°W; 770 km to the northeast of LWA1) at 2.5, 5, 10, 15, and 20 MHz. It has a “sister” station in Hawaii (22.0°N, 159.8°W), call-sign WWVH, which broadcasts the time at the same frequencies except 20 MHz. These stations and their signal patterns are described in great detail by Nelson et al. [2005]. In general, amplitude modulation of each signal is used to convey various types of information about the current universal time (UT). Here we briefly describe those aspects of the WWV signals that are relevant to passive radar.

[12] The transmitters for both WWV and WWVH are relatively powerful monopole antennas, each with an ERP (effective radiated power) of 10 kW for 5, 10, and 15 MHz and 2.5 kW for 2.5 and 20 MHz. The WWVH transmitters are directional with most of the power directed westward to minimize interference with WWV. Both broadcast voice announcements of the time at each frequency, modulated at either 500 or 600 Hz. They also broadcast a coded signal, modulated at 100 Hz, that transmits the UT of the current minute. There are also occasionally alarm signals transmitted, which are modulated at 2000 Hz.

[13] We have found that the amplitude modulation schemes used for the main portions of the WWV signals described above make for rather poor matched filters. Using LWA1 observations of the WWV sky wave (see section 2.3), we have attempted to apply the WWV signal as a matched filter to itself and found that it produces multiple peaks at different lags, making it difficult, if not impossible, to employ the same techniques used within typical PCL systems. This is basically the result of the relatively slowly changing amplitudes of the primary constituents of the continuous WWV signal, namely the main carrier, the 100 Hz time code, and the 500/600 Hz voice announcements. For instance, within the 100 Hz time code, date bits of zero or one are indicated using constant amplitudes over intervals of 200 or 500 ms, respectively. The main carrier and voice announcements have amplitudes that are similarly temporally stable as indicated by their spectral profiles which are just as narrow as that of the 100 MHz time code (see section 2.3 and Figures 1 and 2).

Figure 1.

An all-sky image made at 10 MHz from a single TBW capture (see section 2) showing (top left) the ionospheric reflection of WWV, (top right) the amplitude of the beam-formed signal toward the center of the sky wave marked with an × in the image to the left, and (bottom) the power spectrum of the beam-formed signal (lower) with grey-shaded regions representing the filter used to isolate the 5 ms, on-the-second pulse.

Figure 2.

The same as Figure 1, but at 15 MHz, showing the detection of a reflection from WWVH using the same TBW capture.

[14] The most useful part of both the WWV and WWVH signals for passive radar are the 5 ms pulses emitted at the beginning of each second (except on the 29th and 59th second of each minute). All other signals are suppressed 10 ms before the beginning of each pulse to 25 ms after the end of the pulse to help it stand out more. WWV pulses are modulated at 1000 Hz, and the WWVH pulses are modulated at 1200 Hz, making them distinct from one another. Thus, by applying the appropriate filtering techniques, one may isolate these pulses within signals received by the LWA1 antennas for different arrival times, corresponding to different ranges.

[15] The main drawback of using the 5 ms WWV pulse for ranging is that it provides little practical capability for determining Doppler speeds. Normally with a pulsed transmitter, one can use the pulses that have a common time of flight/group path as a time series that can be Fourier transformed to map the amount of power as a function of Doppler frequency. However, the 1 s cadence of the WWV pulses implies a Nyquist limit of 0.5 Hz. At 10, 15, and 20 MHz, this maximum Doppler shift corresponds to speeds of 7.5, 5, and 3.75 m s−1. While there may be OTH targets and ionospheric structures with Doppler speeds this low, there will likely be many more with higher speeds that will be aliased to lower Doppler frequencies by the coarse, 1 s sampling, and it would not be possible to distinguish between these two scenarios. Therefore, in practice, obtaining Doppler speeds with the WWV pulse it not feasible. However, when there is a single, dominant reflection of the WWV signal, the continuously broadcast carrier signal at the central frequency can be used to estimate the Doppler speed of the ionospheric structure(s) responsible for that reflection with reasonably good Doppler resolution. This will be addressed further in section 4.3.

2.3 Imaging WWV Signals With LWA1

[16] Since the LWA1 antennas have little if any sensitivity below 10 MHz, the bands where LWA1 and WWV overlap are 10, 15, and 20 MHz. Given the distance between WWV and LWA1, 770.0 km, the direct signal and ground wave are not detectable from LWA1. However, conditions in the ionosphere are nearly always such that a reflection of the 10 MHz signal is visible, and reflections at 15 and 20 MHz are sometimes detected. In addition, WWVH sky waves are often visible at 15 MHz and sometimes at 10 MHz. Given the large distance to WWVH (5236.0 km), these reflections are almost certainly “multihop” reflections, that is, they have bounced between the ionosphere and the ocean/terrain several times before arriving at LWA1. Relatively speaking, the WWVH signal is typically more prominent at 15 MHz.

[17] Example images of both WWV and WWVH sky waves are shown in Figures 1 and 2. Figure 1 shows an all-sky image made via total power beam forming at 10 MHz with 12.0 kHz of bandwidth from a single TBW capture. For such images, complex voltages were made using a fast Fourier transform (FFT) of the raw signals and each antenna's voltage was multiplied by a two-dimensional array of phasers before adding them together to compute the total beam-formed power over the whole observable sky. These phasers include corrections for cable losses, unequal cable delays, and other amplitude and phase errors measured for each antenna using observations of bright cosmic sources using the outrigger antenna mentioned in section 2.1. These corrections are described in more detail by Ellingson et al. [2013] and are included within the LWA Software Library [Dowell et al., 2012].

[18] The image displayed in Figure 1 is a combination of the total power from the north-south and east-west dipoles, yielding the total, Stokes-I power. For these images, the ordinate and abscissa are the so-called l and m direction cosines, defined in this case to be l = cose cosa and m = cose sina, where e is the elevation above the horizon and a is the azimuth measured clockwise from north. This projection was chosen because the LWA1 beam changes very little within the l,m plane. One should note that in this projection, elevation does not vary linearly with radius and that a source appearing extremely close to the horizon (i.e., inline image) may in fact be at a deceptively high elevation. For instance, the strong 10 MHz sky wave seen in the upper left of Figure 1 is at an elevation of 27°.

[19] Figure 1 also shows a time series of the amplitude of the signal beam-formed toward the approximate center of the 10 MHz sky wave seen in the all-sky image (marked with an ×). One can clearly see the 5 ms WWV pulse modulated at 1000 Hz with an arrival time of 2.97 ms after the beginning of the second, corresponding to a group path of 890 km. Because each TBW capture is set up to start at the beginning of each UT second, the 61.2 ms duration of each capture is more than enough time to detect such signals from either WWV or WWVH. The power spectrum of this beam-formed signal is also shown in Figure 1. The strongest part of the signal is the main carrier at 10 MHz. One can also pick out features mentioned in the description above (labeled in red) such as the 100 Hz time code, the 600 Hz voice announcement, the 1000 Hz pulse, and the 2000 Hz alarm code. Also shown is a shaded region one might use as a sideband filter to isolate the 5 ms pulse while avoiding the voice announcement and alarm codes, which is also wide enough to achieve a temporal resolution of roughly 1 ms.

[20] Figure 2 shows the same thing as Figure 1, but at 15 MHz, highlighting a detection of a WWVH sky wave. The all-sky image shows some contribution from “second-hop” signals from WWV, or “ground side scatter” [e.g., McNamara et al., 2008], received signals that reflected off the ionosphere, then off the ocean/terrain, then back off the ionosphere to LWA1, which are distributed around the horizon. Because of these second-hop signals, the beam-formed signal shown in the upper right panel is not as clean as that shown in Figure 1 for WWV at 10 MHz, but the 5 ms pulse is still clearly visible. In this case, the arrival time is significantly larger, 19.8 ms (or 5952 km), indicative of the more distant origin in Hawaii. In addition, the pulse has more cycles than the 10 MHz signal (six versus five), consistent with WWVH's pulse modulation of 1200 Hz. This can be seen more clearly in the power spectrum where the peaks from the 5 ms pulse are centered at ±1200 Hz instead of ±1000 Hz.

3 LWA1+WWV Bistatic Radar

[21] The examples shown in Figures 1 and 2 demonstrate the potential for using WWV and LWA1 as a unique bistatic HF radar imager, capable of probing the ionosphere and OTH targets using the ionosphere as a virtual mirror. Here we describe the methods we have developed to do just that using both TBW and TBN data.

3.1 TBW and Multifrequency Radar

[22] As described above, the LWA1 TBW mode offers a unique capability to make total power images from signals spanning the entire available frequency range, 10–88 MHz, all observed simultaneously. The sacrifice one makes for this wideband coverage is that one can only get 61.2 ms of data every 5 min or more. Fortunately, each TBW capture is timed to start at the beginning of each UT second, making it possible to detect reflections from the WWV 5 ms pulse within this short 61.2 ms window.

[23] To isolate reflections of the 5 ms WWV pulse within TBW data, we have used the following multistep approach. First, the format of the data dictates that one read in almost the entire 61.2 ms capture as part of any postprocessing effort. A TBW binary data file is broken up into frames, each consisting of 400 samples from a single antenna. The frames are not organized by antenna ID or time; they are written out on a kind of “first-come-first-serve” basis. Because of that, all of the data for a particular antenna might be contained within the first few percent of the file, while for another, one may have to parse the entire file to retrieve all of its data. In addition, this pattern is different for each file (see Ellingson [2007] for a more detailed description of the TBW system).

[24] Because of this file structure, it is more efficient to use an FFT to filter the data for each antenna as it is read in, yielding a time series of complex voltages for each antenna once the whole file has been read. In practice, for each antenna, an FFT is applied every 1024 samples, and the resulting voltages for the frequencies closest to 10, 15, and 20 MHz (9.953, 14.93, and 19.91 MHz) are saved, giving a time series of 11,718 voltages for each band, each with a bandwidth of 4.98 MHz.

[25] Following this, an FFT is applied to each of these time series and the resulting complex spectrum is filtered with two sideband filters illustrated in Figure 1. An inverse FFT is then applied to each sideband to yield a new complex voltage time series with a temporal sampling of 1.004 ms within which the 5 ms pulse appears as a square pulse rather than a sinusoid. For each point in this time series, a total power beam-formed image is made over the whole sky (see section 2.3), separately for each sideband and each polarization. These four images are then combined to form a single image cube that gives the total power as a function of azimuth, elevation, and time of arrival/group path.

[26] We have used three sequences of TBW captures to help illustrate the typical output of this combined radar system. These were conducted between 00:05 and 13:25 UT, 6 May 2012, between 05:05 and 14:05, 7 May 2012, and between 19:55 and 15:05 UT, 30–31 July 2012 as part of telescope commissioning efforts. Within each sequence, the captures were spaced 10 min apart. Figures 3-5 show example image cubes for 10, 15, and 20 MHz, which are the first frames of three electronically available movies made from these TBW sequences. Each frame shows the total power, all-sky map of the 5 ms pulse at different times of arrival, converted to group paths from 153 to 7374 km with the horizon marked with a white circle. We note again that due to the projection used to display these images (see section 2.3), nearly all of the detected sky waves appear as if they are at or just above the horizon, even though the typical elevation is actually about 10° and varies from a few degrees to as high as 40°. We also note that the times of flight/group paths for the detected pulse reflections are too large for them to be ground waves from WWV.

Figure 3.

An example of an all-sky/group path image cube from the 10 MHz WWV signal with a single TBW capture. The white circle indicates the horizon/field of regard. The number above each panel is the corresponding group path in kilometers. This is the first frame of a movie of three separate sequences of such captures, which is available electronically.

Figure 4.

The same as Figure 3 but using the 15 MHz signal from the same TBW capture. This is also the first frame of a movie of three separate sequences of such captures, which is available electronically.

Figure 5.

The same as Figure 3 but using the 20 MHz signal from the same TBW capture. This is also the first frame of a movie of three separate sequences of such captures, which is available electronically.

[27] The relatively poor range resolution afforded by the width of the 5 ms pulse is evident within these image cubes. However, one can see that the all-sky approach, when combined with range information, allows one to readily distinguish among the first-hop WWV signal (16° east of north), second-hop WWV signals at various azimuths, and the reflection of the WWVH pulse (almost exactly due west). One can also see that while 10 MHz reflections are almost always visible from WWV and sometimes from WWVH, only WWVH is consistently visible at 15 MHz. Reflections of the WWV pulse are only detected some of the time at 15 and 20 MHz. This is not entirely unexpected given the large plasma frequencies and/or low angles of incidence needed to reflect 15 and 20 MHz transmissions between WWV and LWA1. The instances where the WWV signal is seen at 15 and 20 MHz are likely related to sporadic-E (Es), which will be discussed further below.

3.2 TBN and Continuous Radar

[28] While the multifrequency data available with TBW captures offer a method for crudely sounding the ionosphere, TBN observations allow one to monitor a particular ionospheric region/layer (or OTH target) at one frequency with much better temporal coverage (i.e., one sample per second versus one every 5 min or more). For those readers familiar with digisonde dipole arrays, in this context, the TBW and TBN modes are analogous to the digisonde “sounding” and “sky map” modes.

[29] In principle, one may analyze TBN observations using the same procedure described for the TBW data in section 3.1. However, the differences in file structures between TBW and TBN data make it more practical to use a different but related approach with TBN data. Since the data are already filtered in TBN mode before they are written out, what is recorded are complex voltages rather than raw signals. Typically, these are written out with a sampling rate of 100 ksps (16 bits per sample), but lower sampling rates are possible. TBN binary files are broken up into frames, similar to TBW files. However, within a TBN file, a frame consists of 512 samples from a single antenna and the data for all antennas are written out together for each time stamp (although not necessarily ordered by antenna ID). This makes it faster to read in just the time stamp of each frame until the first frame closest to the next UT second is identified. After this, due to the lower sampling rate, one can easily read all of the data for all antennas covering the next several tens of milliseconds into RAM, even with a modestly equipped desktop computer (e.g., 40 ms amounts to roughly 4 MB).

[30] We found that rather than using the FFT/filter/inverse FFT approach employed with the TBW data, it was faster/more efficient to apply a sliding window filter to these data as they were read in. Specifically, as the data are read in, they are up/down converted by ±1000 Hz, after which a 5 ms wide Hamming window is applied and the data within this window are averaged. The Hamming window was chosen to suppress contributions from higher frequencies and is stepped at 1 ms intervals to give comparable temporal sampling to the TBW approach. This is done for the first 40 ms of each UT second, yielding 40 up- and down-converted complex voltages (i.e., at the carrier frequency ±1000 Hz) for each antenna corresponding to 40 different times of arrival for the WWV pulse. Similar to the TBW approach, the up- and down-converted voltages are imaged separately and then combined for each time of arrival. In applying this analysis to actual TBN data (see below), we found that the time stamps in the TBN files were off by 10.24 ms, a result of the time stamps being assigned after the data had been passed through the hardware filters, which have a length of 1024 samples. All times of arrival within the TBN analysis are corrected for this effect.

[31] As a demonstration of this method, we have used a 15 min TBN observation at 15 MHz that started at 00:00 UT on 25 September 2012 and another 5 min test observation at 10 MHz beginning at 20:00 UT on 25 March 2013. The analysis described above was applied to both data sets, and we have made movies of the resulting image cubes available electronically. Note that, as mentioned in section 2.2, the WWV pulse is not broadcast on the 29th and 59th second of each minute, which causes a “blinking” effect within these movies with a cadence of about 30 s. At these times, only the main 10 or 15 MHz carrier and/or the voice announcements are broadcast, so that even though reflections are detected, the arrival times/group paths are relatively meaningless without the 1000 Hz-modulated pulse.

[32] The 15 MHz movie shows that the first-hop signals from WWV often have observed structure beyond a simple unresolved, point-like source. This additional structure commonly lasts for only one to a few seconds. However, there is a strong secondary reflection to the southwest of the main sky wave that appears at 565 s and remains present to varying degrees for about 110 s. A frame from the movie that prominently features this structure is shown in Figure 6. This highlights the benefits of using the TBN mode for probing ionospheric structure on smaller temporal scales. With TBW captures being spaced by a minimum of 5 min, a sequence of TBW captures could easily miss the formation and evolution of such structures. The same is true for most ionosonde systems, which typically use integration times on the order of a minute or more to boost signal-to-noise ratio. As with the TBW captures, second-hop signals are seen throughout the observation, which vary noticeably in both strength and location due to significant ionospheric variability.

Figure 6.

An all-sky/time-of-arrival image cube from 1 s of a 15 min TBN observation at 15 MHz starting at 00:00 UT on 25 September 2012. As in Figures 3-5, the white circle indicates the horizon/field of regard. The time of arrival from the beginning of the nearest UT second is given above each panel in milliseconds. This is a frame from a movie (roughly 2/3 from the beginning) that covers the entire 15 min TBN observation and which is available electronically.

[33] In contrast to the 15 MHz observation, one can see a single, fairly stable sky wave within the 10 MHz movie (the first frame of the movie is shown in Figure 7). There are some weak second-hop signals and no sign of reflections from WWVH. These differences may be largely the result of Es being present during the 15 MHz observations but not during the 10 MHz observations. This will be discussed further in the following section.

Figure 7.

An all-sky/time-of-arrival image cube from the first second of a 5 min TBN observation at 10 MHz starting at 20:00 UT on 25 March 2013. As in Figures 3-6, the white circle indicates the horizon/field of regard. This is the first frame from a movie that covers the entire 5 min TBN observation and which is available electronically.

4 Applications

[34] While the fixed broadcast frequencies used by WWV limit flexibility, the combination of continuously broadcasting, relatively high-power transmitters with the all-sky capability of LWA1 make the combined passive radar system an effective tool for exploring both ionospheric structure and a variety of OTH targets. Here we describe examples of the possible applications of the LWA1+WWV radar system.

4.1 The Nature of the Observed Sky Waves

[35] To adequately assess the potential of the LWA1+WWV radar system to observe ionospheric structure, it is prudent to first compare the results from our test observations with data from nearby ionospheric sounders. Fortunately, there is a continuously operating digisonde station roughly between LWA1 and WWV in Boulder, Colorado that has data available through the Digital Ionogram Database [Reinisch et al., 2004].

[36] To compare the results from the LWA1 observations to the Boulder digisonde data, we computed a maximum reflection height for each detected sky wave as inline image, where R is the group path and d is the linear distance between LWA1 and WWV (769.74 km). If the actual signal path deviates in any way from a simple triangular/virtual mirror trajectory, this approximation fails but the actual height cannot be larger than this value. We similarly compute an approximate corresponding critical frequency as fc,max≃2hmax/R.

[37] From each image cube, we compute R for the first-hop WWV signal as well as its azimuth and elevation using a kind of center-of-mass computation using the total power as a weight. These azimuths and elevations were also corrected for a known LWA1 pointing offset, which has been well constrained/characterized by Dowell and Grimes [2012]. Using our center-of-mass style calculation allows us to compute estimates for the uncertainties in the corrected sky position and group path using estimates of the 1σ error in the measured power for all the images within the cube. We did this by measuring the standard deviation in the center of each image within the cube where there are essentially no WWV or WWVH sky waves. This also allows for error estimates to be computed for hmax and fc,max.

[38] For each of the three TBW sequences (see section 3.1), we have grouped the observations in time according to the signal strength (assessed by eye) of the three frequencies, 10, 15, and 20 MHz. For each of these groups, we also obtained all data from the Boulder digisonde station. In Figures 8-10, we show images of the mean ionogram amplitudes (both polarizations) from the Boulder data as functions of frequency and range within each time period for all three TBW sequences. Within each time period, we have plotted hmax versus fc,max for all detected first-hop WWV signals, color-coded by frequency, with qualitative assessments of the signal strengths printed above each panel.

Figure 8.

Mean ionograms (both polarizations) from the Boulder, Colorado digisonde station for 6 May 2012 within time intervals chosen according to the WWV signal strengths at 10, 15, and 20 MHz as observed with LWA1 (see panel titles). Also plotted are the maximum heights and critical frequencies for the LWA1 observations of WWV at 10 (blue), 15 (cyan), and 20 (green) MHz.

Figure 9.

The same as Figure 8 but for 7 May 2012.

Figure 10.

The same as Figure 8 but for 30–31 July 2012.

[39] Perhaps the most relevant feature seen within these plots is the indication of Es from the Boulder digisonde during all of the time periods. In many cases, the peak reflected Es frequency (foEs and/or fxEs) is relatively high, up to ∼10 MHz. One can also see that hmax for both the 15 and 20 MHz signals is never larger than about 130 km with corresponding values for fc,max between 4 and 6 MHz, roughly consistent with the digisonde-measured Es properties. There are times when foEs from the Boulder digisonde is relatively large and no 15 or 20 MHz reflections were detected; there is one instance where the opposite is true (see the top right panel of Figure 8). This is the result of the patchy nature of Es and is consistent with the 15 and 20 MHz signals being reflected off dense structures or “clouds” within the Es layers.

[40] The 10 MHz detections are generally consistent with being reflected off the Es layers themselves, with values for hmax<130 km and fc,max near or at the blanketing frequency (fbEs). In the top left panel of Figure 8, one can see 10 MHz reflections with possibly larger heights. However, when compared with the ionogram, it appears that these may in fact be second-hop signals reflected off the same Es layer because their values for hmax and fc,max are roughly consistent with a reflection from the Es layer appearing within the ionogram at a range of about 250 km. Within the 30–31 July TBW sequence, there do appear to be some 10 MHz reflections that are from higher within the ionosphere, which are not related to Es (see the top left panel of Figure 10), but they seem to be the exception for these observing runs.

[41] For the 15 MHz TBN observation conducted in September 2012, the values of hmax were all <150 km but with relatively high corresponding values of fc,max (again, between 4 and 6 MHz). This implies that these sky waves resulted from reflections off dense Es clouds, as was observed with the TBW observations. This is not unexpected since these relatively dense, low-altitude structures provide virtually the only means to produce a one-hop reflection between LWA1 and WWV at frequencies as high as 15 and 20 MHz. The Boulder ionograms from near the same time period do show evidence of Es with some instances of large foEs (4–5 MHz) later in the evening.

[42] The 10 MHz sky waves seen in the TBN observation conducted in March 2013, however, do not appear to be from Es at all. Their values for hmax are between 200 and 350 km with corresponding values of fc,max between 4.5 and 6.5 MHz. This is roughly consistent with the Boulder ionograms around the same time, none of which show any signs of Es. Thus, it is clear that while Es is generally required to receive single-hop 15 and 20 MHz WWV signals at LWA1, the same is not true for 10 MHz signals, which can just as easily be reflected by normal, higher-altitude ionospheric layers.

4.2 Characterizing Sporadic E

[43] As demonstrated in the previous section, all of the 15 and 20 MHz sky waves observed with the LWA1+WWV radar result from dense Es clouds. In addition, nearly all of the reflections observed at 10 MHz within the three TBW sequences presented here are associated with Es layers. Thus, when combined with additional information from the Boulder ionograms, the unique all-sky observations of these sky waves can be used to characterize the Es structures responsible for these reflections.

[44] The most basic property to be constrained is the spatial distribution of the Es structures. Because of the patchy nature of Es, especially when dense clouds are present, the signal path through an Es layer can be quite complex. Thus, one cannot reliably locate a single reflection point for the signal with something like a tilted virtual mirror approximation or conventional ray tracing, even when the transmitter and receiver locations are known. However, since the LWA1 all-sky images offer excellent measurements of the sky position of the reflected signals (1σ errors ≲0.01°), if one has a good estimate of the height of the lower boundary of the Es layer, the location where the signal left the Eslayer and continued to LWA1 can be computed. The parameters automatically extracted from the Boulder ionograms indicate that the Es layers during our TBW observations where consistently at a height of about 110 km. With a typical thickness of 20 km, the bottom side of these layers was likely at an altitude of about 100 km.

[45] Using this assumed altitude and the measured sky wave elevations and azimuths (see section 4.1), we used spherical trigonometry to compute the location at which each signal left the bottom of the Es relative to LWA1 as arc lengths both perpendicular and parallel to the great circle between LWA1 and WWV. Note that at an altitude of 100 km, the arc length of this great circle is 782.1 km. These locations are plotted in the left panels of Figure 11 for each TBW sequence. The data are color-coded by frequency and only include those detections with hmax<140 km (>90% of detections). One can see that for the majority of cases, the data are clustered relatively tightly around the direct path between LWA1 and WWV but with a considerable spread along this great circle (∼300 km). There are also significant departures to the east during the 6 May sequence and to the west for the 7 May observing run, both closer to LWA1 than WWV by 100–200 km.

Figure 11.

(left) From all TBW observations where Es was detected (see section 4.1 and Figures 8-10), the location where the signal left the bottom of the Es layer (altitude of 100 km) plotted parallel (ordinate) and perpendicular (abscissa) to the great circle between LWA1 and WWV at that altitude. Dotted lines mark this great circle and the halfway point between the two locations. The same color coding is used as in Figures 10-15 (i.e., 10 MHz is blue, 15 MHz is cyan, and 20 MHz is green). (right) The root-mean-square (RMS) in the Es signal position among all three frequencies for TBW captures where Es reflections were detected simultaneously at 10, 15, and 20 MHz as functions of time.

[46] For instances where reflections were simultaneously detected at all three frequencies, we have plotted the RMS among the three frequencies for both Es positions. One can see that the dispersions are rather substantial, especially parallel to the great circle between LWA1 and WWV, indicating that while all the reflections are from Es, the three frequencies probe completely independent structures. This highlights a unique ability of the TBW mode of the LWA1+WWV radar. Conventional sounders can either perform frequency sweeps that take roughly 10–15 min to complete and observe different frequencies at different times or they integrate at a single frequency. Using the TBW mode, LWA1 can observe reflections simultaneously at several frequencies, yielding an instantaneous picture of the physical span of any detected Es structures.

[47] The TBN mode offers the ability to observe the evolution of these Es structures on relatively short time scales. We show this in Figure 12 where we have plotted the bottomside Es positions of the sky waves observed during the 15 min TBN test observation at 15 MHz. One can see that, again, the dispersion is largest parallel to the great circle between LWA1 and WWV. However, with the shorter sampling interval (1 s versus 10 min), one can see that the observed reflections often appear to come from one of a handful (∼17) of regions where the data are noticeably clustered. We have isolated these regions by eye with polygon regions of interest, as shown in the left panel of Figure 12. For each region, we have plotted a time series of normalized received power in the right panels of Figure 12. These seem to imply that there was a mixture of short-lived (∼20–100 s) and persistent structures among which the signal moved in a quasi-random pattern. However, the signal seems to have been most frequently reflected by a group of structures just south of the halfway point between LWA1 and WWV.

Figure 12.

(left) For the 15 min TBN observation conducted on 25 September 2012 at 15 MHz, locations of Es reflections (i.e., where the signal left the bottom of the layer) plotted in the same manner as Figure 11. Here groups of reflections were identified and isolated with regions of interest drawn by eye, plotted here as color-coded polygons. (right) The normalized received power as a function of time for reflections from each of the groups shown in the left panel with matching color schemes.

4.3 Doppler Speeds

[48] As mentioned in section 2.2, while the WWV pulse cadence is temporally too coarse for useful Doppler analysis, the continuously broadcast carrier frequencies offer this capability when a single, dominant sky wave is present. This is rarely true at 15 MHz for LWA1 since reflections from both WWV and WWVH are often seen together. While the imaging capability of LWA1 allows one to discriminate between these two sky waves, sidelobe confusion is still an issue and makes analysis of carrier signal reflections problematic at 15 MHz. In contrast, WWVH does not broadcast at 20 MHz and is weakly, if ever, detected at 10 MHz with LWA1. Thus, when a first-hop signal from WWV is observed at either of these two frequencies, it is (nearly) always the dominant sky wave, and the 10 and 20 MHz carriers can be used to measure the Doppler speed of the ionospheric reflection point.

[49] For TBW data, the total amount of time per capture is about 61 ms, implying that a spectrum made from such a capture can have a resolution of, at best, about 14 Hz. This is more than enough to resolve the carrier from the next closest spectral feature of the signal, the 100 Hz time code (see section 2.2 and Figure 1). Within a TBW capture, the carrier is typically tens of dB above the noise floor and more than 15 dB stronger than any other part of the WWV signal (see Figure 1). Thus, even though the spectral resolution is relatively wide (∼200 and 100 m s−1 at 10 and 20 MHz), the Doppler speed can still be measured with good precision because of the sheer magnitude of the signal.

[50] Consequently, for each TBW capture, we beam formed the signal toward the position of the dominant sky wave from the 10 and 20 MHz image cubes and computed temporal spectra from these similar to those shown in Figures 12. We then measured the peak Doppler frequency, fD, using the weighted mean average of frequencies among the brightest three, with the spectral power as the weight. Because these are first-hop signals within a bistatic radar system, the Doppler shift is almost entirely from vertical ionospheric motion, with the vertical Doppler speed given by 2vD=−cfD/f, where downward motion is negative.

[51] Figure 13 shows the measured Doppler speed as a function of time for all detected sky waves at 10 and 20 MHz from all three TBW sequences as functions of time. For reference, the peak received power and group path for each reflected signal are also plotted as functions of time. These confirm that the two frequencies probe markedly different Es structures as the 10 MHz reflections generally show upward motion and the 20 MHz reflections show the opposite for the 6 May and 30–31 June sequences. The only two 20 MHz detections during the 7 May sequence have similar upward speeds as the 10 MHz reflections detected at the same time. During the 6 May sequence, the speeds measured at 10 MHz appear to support the supposition that the sky waves with larger group paths were actual two-hop signals reflected off the same Es layer as their speeds are roughly twice that of the reflections with smaller group paths (i.e., they were Doppler shifted twice).

Figure 13.

For all TBW observations, the (top) peak power, (middle) group path, and (bottom) Doppler speed as functions of UT for all detections of a first-hop reflection of the WWV signal at either 10 or 20 MHz (blue and green points, respectively).

[52] For TBN data, one can continuously observe the carrier signal, which can yield a full Doppler spectrum with excellent resolution. We found that the carrier signal was isolated very well by first beam forming the signal toward the location on the sky of the dominant sky wave detected within the all-sky/group path image cubes. Then, every ∼0.1 s (1024 ms or 200 TBN frames) of the beam-formed signal was Fourier transformed (after applying a Hamming window), after which only the 0th frequency was kept (i.e., the signal was filtered down to a bandwidth of 10 Hz).

[53] As an example, we show the Doppler spectrum computed from such a carrier signal time series using a 10 s-wide sliding (Hamming) window for the 10 MHz TBN observation conducted on 25 March 2013 in Figure 14. This spectrum has a Doppler speed resolution of about 1.3 m s−1. For reference, we also plot the peak received power and group path for the sky waves from the all-sky/group path image cube as functions of time. One can see that while the received power and group paths do vary somewhat, the Doppler speed is consistently upward with a magnitude of about 1 m s−1, with some small variations. This is roughly consistent with the Boulder digisonde data at the time which measure Doppler speeds of about 0.8 m s−1 at heights of 200–300 km, close to the values of hmax estimated for these sky waves (see section 4.1).

Figure 14.

For the 5 min TBN observation conducted on 25 March 2013 at 10 MHz, the (top) peak power, (middle) group path, and (bottom) Doppler spectrum as functions of UT for the first-hop reflection of the WWV signal. In Figure 14 (bottom), the weighted mean Doppler speed at each time step is plotted as a white curve.

4.4 Ionospheric Structure and Geolocation

[54] The movies of the image cubes made with both the TBW and TBN data illustrate the significant amount of variability within the ionosphere as one can see the first-hop WWV signal move noticeably in the sky. This is likely due to density gradients at the altitude/layer where the signals are reflected, in this case, typically associated with Es, which are among the known limitations to geolocation precision for operational OTH radar systems [e.g., Headrick and Thomason, 1998].

[55] To illustrate this, we have used the TBW data at 10, 15, and 20 MHz to geolocate the WWV transmitters. We did this by using the sky positions and group paths determined from the image cubes for each frequency (see section 4.1) and assuming a virtual mirror approximation (see Headrick and Thomason [1998] and references therein). In other words, we approximated the ionosphere with a flat mirror, using the determined sky position and group path with the laws of cosines and sines to determine the angular separation between LWA1 and the transmitter on the surface of the Earth. The measured uncertainties in sky position and group path (again, see section 4.1) were propagated through this computation. While some of the final geolocations have relatively large errors, the typical 1σ uncertainties for significant first-hop detections were 0.007° and 0.01°, or 0.8 and 1 km for the latitudes and longitudes, respectively.

[56] The geolocation results, both with and without the LWA1 pointing correction applied (see section 4.1 and Dowell and Grimes [2012]) are plotted in Figure 15 for all TBW data. From these, one can immediately see that the measured positions contain a significant amount of scatter and that the LWA1 pointing error causes a systematic offset to the northeast (left). The final, corrected positions (right) have a mean that agrees very well with the known WWV location, within 0.03°, or 3.2 km.

Figure 15.

Geolocation of WWV using detections of sky waves at 10, 15, and 20 MHz from TBW captures using a standard virtual mirror approximation. The two panels show the results (left) before and (right) after the LWA1 pointing correction was applied (see sections 4.1 and 4.4).

[57] There is a significant amount of scatter among these positions with an RMS deviation from the expected WWV position of about 0.13° or 14.6 km, more than an order of magnitude larger than the typical measured uncertainty based on the noise in the all-sky images. As noted above, this scatter is likely from structure within the Es layers that work to invalidate the virtual mirror approximation and is consistent with previous measurements of HF bearing errors at midlatitudes [Tedd et al., 1985]. In total, there were 237 separate geolocation measurements, but the error in the mean position, 0.03°, is more than an order of magnitude larger than RMS/inline image, the result one would expect if the ionospheric errors in the positions were uncorrelated. This suggests that a significant portion of the structures responsible for these position offsets spans the typical temporal/spatial scales probed by these observations.

[58] To examine this on much shorter time scales, we have repeated the geolocation exercise using the image cubes for the two TBN observations. Figure 16 shows the results for the pointing-corrected geolocations. One can see a clear difference in the nature of the position errors between the Es-based reflections seen at 15 MHz on 25 September 2012 and those that reflected off normal ionospheric layers at 10 MHz on 25 March 2013. While both observations incorrectly geolocate the transmitter, the 10 MHz positions are relatively tightly clustered (RMS scatter of 0.068° or 7.6 km) around a single location, roughly 0.3° due south of WWV. We estimate that this is consistent with an ionospheric “tilt” of between 2.5° and 4°, with a normal vector pointing about 15° east of due south. This is consistent with tilts measured for the midlatitude ionosphere [Tedd et al., 1985]. The 15 MHz, Es-based positions, however, exhibit much more structure, as one may have inferred from the results discussed in section 4.2 and shown in Figure 12 and the general patchy nature of Es. Thus, while the accuracy of the Es-based, 15 MHz positions appears no worse than those derived from the more normal, 10 MHz reflections, their precision is significantly worse with an RMS scatter of 0.11° (12.2 km). This is also consistent with previous estimates of the effect of Es on apparent ionospheric titles/geolocation [Paul, 1990].

Figure 16.

Geolocation of WWV using detections of reflections at 10 and 15 MHz from two TBN observations conducted on 25 March 2013 and 25 September 2012, respectively, using a standard virtual mirror approximation. Note that the 10 MHz observation was 5 min long and showed no evidence of Es while the 15 MHz observation lasted for 15 min and the reflections were entirely from dense Es structures.

4.5 OTH Terrain Mapping

[59] As noted in sections 3.1 and 3.2, there are second-hop signals observable at all three frequencies and within both the TBW and TBN data. Given some simplifying assumptions, these can potentially be used to produce maps of HF reflectivity for the distant terrain and/or ocean. In order to locate the point on the terrain/ocean where the second-hop signals reflected, we are forced to make the assumption that both ionospheric hops occurred at the same altitude. This assumption allows us to compute the distance traveled by the WWV signal from the reflection point on the terrain/ocean to LWA1 by simply dividing the group path in half. Then, using the image cubes described in sections 3.1 and 3.2, one can map the reflected HF power using a simple virtual mirror approximation (i.e., without tilts). The results presented in section 4.1 suggest that ionospheric structure and variability will introduce geolocation errors of about 0.13°. We can assume that this is increased by roughly a factor of inline imageto 0.18° for second-hop data since these signals have reflected off the ionosphere twice. However, for this type of mapping exercise, this is an acceptable level of uncertainty given that the LWA1 beam is much larger. Specifically, the full width at half maximum of the LWA1 beam at 10, 15, and 20 MHz is 15.8°, 10.5°, and 7.9°, respectively.

[60] Something that is a concern for terrain mapping and not the analysis presented in section 4.4 is the effect of sidelobes associated with the LWA1 beam. These are apparent in all of the images presented in Figures 1-6. Since the analysis in the previous section focused on finding the centroid of the brightest first-hop signal on the sky, these sidelobes were of little concern. However, as we attempt to map the HF reflectance using the detected power from several different directions at once, confusion from sidelobes will have a significant impact.

[61] Within the field of radio astronomy, an effective iterative approach to deconvolution has been developed to mitigate the consequences of such sidelobes. However, these techniques have been designed to work with data from a “multiplying” rather than “adding” interferometer. In other words, they work on images made with visibilities, which are correlations between all possible pairs of antennas or “baselines.” For a baseline with antennas i and j, inline image, where V is the visibility, ϵ is the complex voltage, and the average is performed over a fixed time interval. Images made with such visibilities are equivalent to total power beam-formed images but with a constant “DC” term subtracted.

[62] Specifically, the intensity on the sky is related to the visibilities by the following

display math(1)

where l and m are direction cosines for a particular part of the sky (i.e., not necessarily the zenith as we have used here), inline image, and the baseline coordinates, u, v, and w are the differences between the positions of the two antennas of the baseline in a coordinate system such that w points toward the observation field center, u points eastward, and v points northward.

[63] Typically, visibilities are “fringe stopped” by multiplying them by exp(−2πiw), so that for a small field of view (i.e., n≈1), the w term can be ignored because n−1≃0. In this case, the image can be made by gridding the visibilities in the u,v plane and performing a two-dimensional FFT. For our LWA1 observations, the field of view is far from small. However, because the LWA1 antennas lie nearly within a plane and we have defined the field center of our images at zenith, the w term is still small in our case (i.e., w≃0). In addition, as one can see from Figures 1-6, all of the detected WWV reflections are near the horizon where n≃0. Thus, by not fringe stopping our computed visibilities, we effectively make the nw term in equation (2) negligible and can use a standard FFT-based imager with our LWA1+WWV visibilities.

[64] The scheme used to reduce sidelobe confusion within FFT-based visibility imaging is as follows. After the data are initially calibrated, an image is made and a deconvolution algorithm is applied. The most commonly used algorithm is CLEAN [see Cornwell et al., 1999], which models the image with a series of delta functions, called “CLEAN components,” convolved with the impulse response of the interferometer or “dirty beam.” The locations of the CLEAN components are determined iteratively by placing one at the location of the pixel with the largest absolute value and subtracting a scaled version of the dirty beam at that location at each iteration. This process is usually terminated at the first iteration when the pixel with the largest absolute value is actually negative. After this, the CLEAN components are convolved with a Gaussian beam with a width similar to the main lobe of the dirty beam and added back to the residual image.

[65] Following the application of the CLEAN algorithm, the CLEAN components are used as a model of the intensity on the sky to refine the visibility calibration. Within this self-calibration process, antenna-based phase corrections are solved for using a nonlinear fit (usually a gradient search; see Cornwell and Fomalont, 1999). Because an array with N elements has N(N−1)/2 unique baselines, this is generally an overly constrained problem, e.g., for LWA1, one has to solve for 256 phase corrections using 32,640 baselines. Amplitude corrections can also be solved for within this process, but it is usually safer to use phase-only self-calibration to avoid biasing the calibration toward the locations of the CLEAN components. After applying the resulting calibration, the visibilities are reimaged and CLEAN is run again. Because of improvements in image fidelity made possible by the self-calibration-determined phase corrections, subsequent applications of CLEAN are able to mitigate the effect of sidelobes to a greater degree. After CLEANing, one can repeat self-calibration with the new CLEAN-component model and continue to iterate until the process converges.

[66] To apply these techniques to the TBW and TBN data, visibilities were computed by correlating the complex voltages for each group path over the entire observing period. Visibilities for both polarizations and both sidebands (i.e., carrier frequency ±1000 Hz) were averaged together. For the TBW data, the correlations were computed using all TBW captures and for the TBN data, the entire 15 min observation was used. This was done separately for each frequency. For each time of arrival and frequency, the visibilities were imaged with three iterations of CLEAN and phase-only self-calibration. Additional iterations did not significantly change the appearance of the images produced.

[67] The final CLEANed images from the TBW data at each frequency and time of arrival are displayed in Figure 17. For the purposes of terrain mapping, only group paths between 1958 and 4666 km were used to minimize contamination by first-hop WWV signals and sky waves of WWVH. Similar images were made from the 15 MHz TBN data but are not shown here. The 10 MHz TBN observation from March 2013 was not used in this analysis because second-hop signals were faintly and inconsistently detected to the extent that they were unsuitable for terrain mapping. In Figure 17, the horizon is indicated in each panel with a grey circle. One can see that sidelobes have been effectively eliminated, particularly around bright sky waves. It is also apparent that despite the narrow range in time of arrival used, contamination from WWVH and first-hop WWV sky wave is still an issue. For the analysis to follow, polygon regions were drawn by eye around the first-hop WWV sky waves at 10, 15, and 20 MHz and around the WWVH sky waves at 10 and 15 MHz (recall, WWVH does not broadcast at 20 MHz) to exclude those regions from the terrain mapping process; they are shown in white in Figure 17. The origin of the strong source significantly beyond the horizon to the east in the 15 MHz is not entirely clear. However, it seems likely that since its strength roughly correlates with the WWVH sky wave that it is an aliased version of WWVH introduced by our FFT-based imager.

Figure 17.

Average all-sky images from all TBW captures at 10, 15, and 20 MHz for ranges consistent with second-hop signals from WWV, i.e., too large to be first-hop signals and too small to be from WWVH. Regions used to exclude contamination by first-hop WWV signals and WWVH reflections from terrain mapping are shown with white polygons.

[68] Within each truncated image cube (i.e., only group paths between 1958 and 4666 km), the sky position and range of each cubic pixel was used to compute a latitude and longitude, using the assumptions detailed above. The pixels where then binned into a latitude and longitude grid, and the mean power within each grid cell was computed to produce power maps for each frequency. These are shown in Figure 18 for the TBW data and in Figure 19 for the TBN data. For all maps, the regions excluded due to contamination by first-hop WWV sky waves are shaded in grey as are the WWVH-excluded regions for the 10 and 15 MHz maps.

Figure 18.

Terrain maps made from the TBW-based images shown in Figure 17. Regions excluded to minimize contamination by first-hop WWV signals and WWVH reflections are flagged in light grey.

Figure 19.

The same as Figure 18, but for the 15 min, 15 MHz TBN observation conducted on 25 September 2012.

[69] One can see that while the 5 ms width of the WWV pulse limits the resolution in the direction radially away from LWA1, there are many distinct features visible. One of the most striking is a strong signal originating from the Gulf of Mexico which is apparent in the 10 and 20 MHz TBW maps but not in either 15 MHz map. This could be indicative of Bragg scattering from waves in that region with spacings of about 7.5 m. Such waves would produce Bragg scattering at wavelengths of 15 and 30 m (20 and 10 MHz), but not at 20 m (15 MHz).

[70] There is a strong feature originating from the Pacific Ocean to the southwest of the Baja Peninsula within all of the TBW-based maps. The superior azimuthal resolution of the 20 MHz map shows that this feature may consist of two or more Bragg lines. What is likely a similarly strong Bragg line is seen near the Oregon coast in all the TBW maps. These are not seen in the 15 MHz TBN map. The 20 MHz map also shows evidence of substantial reflections from the Pacific Ocean west of California that were not observable at 10 and 15 MHz because of contamination from WWVH.

[71] Both 15 MHz maps as well as the 20 MHz TBW map show strong reflections from the midwestern and plains states, stretching from Minnesota/Wisconsin to eastern Texas. This is likely due to the general lack of rough or mountainous terrain in these regions that would tend to scatter any incident HF signals.

5 Discussion and Conclusions

[72] We have demonstrated that using existing technologies, developed and operated for other purposes, it is possible to construct a unique and useful, passive HF radar imaging system. Using the first of many stations of the LWA to observe reflections of signals from the NIST station WWV, we have shown that one can probe and monitor ionospheric structure, map HF reflectivity of land and sea over a large area, and potentially track OTH targets. The all-sky capability of LWA1 allows us to locate and track both single-hop and multihop signals from WWV in any direction under a variety of ionospheric conditions.

[73] Using preliminary commissioning data, we have been able to demonstrate the impact ionospheric phenomena, predominantly Es, have on OTH geolocation precision. We have shown that the structures associated with the detected Es layers persist in time and space such that averaging over many, relatively closely spaced observations does not improve geolocation accuracy as much as one may expect (i.e., by a factor of inline image). As more LWA stations are added, continued observations of WWV will allow us to establish a kind of coherence scale length, the separation beyond which ionospheric structures of a given type are uncorrelated. Beyond this scale, the geolocation errors added by different parts of the ionosphere will likewise be uncorrelated such that one may achieve a inline imageimprovement in accuracy by averaging the results from N stations independently observing the same OTH signal and spaced a minimum of the coherence length apart from one another. For example, by using 10 stations spaced in this manner, the RMS geolocation error reported here for 14.6 km would be reduced to 4.6 km and would be achieved in real time (i.e., by averaging over space rather than time).

[74] In light of the results presented here, having many LWA stations observing WWV would also be a powerful probe of ionospheric structure, especially within Es layers. The techniques described here allow one to map the spatial distribution of Es structures and to estimate their Doppler/vertical drift speeds. In particular, TBN observations at 15 MHz showed evidence of dense Es structures spaced by a few to tens of kilometers from one another, consistent with the observed properties of quasi-periodic echoes [see, e.g., Bernhardt, 2002]. These TBN observations were also shown to be a valuable tool for exploring the sometimes short lifetimes (as short as 20 s) of these dense Es clouds. Having multiple stations with independent lines of sight capable of observing simultaneously at 10, 15, and 20 MHz will yield maps of Es layer structure with unprecedented detail and scope at a relatively high temporal resolution.

[75] The ocean/terrain mapping capability of this system will also be enhanced with multiple stations by improving sensitivity and location accuracy. For instance, a proposed intermediate stage of the LWA consists of roughly 10 stations with a maximum baseline of 200 km. The angular resolution of this array would be 2000 times better, improving from 15.8° at 10 MHz for LWA1 to only 28.5 arcseconds. The full LWA will have baselines twice as long and therefore, the synthesized beam will be half as wide. In either case, this is much smaller than the smallest achievable resolution set by ionospheric variability of ∼0.1° (see section 4.4). However, by employing ionospheric calibration techniques commonly used within HF/VHF astronomy [see, e.g., Kassim et al., 2007], one may still be able to achieve subarcminute resolution images of reflecting structures with either the intermediate or full array.

[76] Thus, when combined with WWV, a larger array of LWA stations could be used to make maps of HF reflectivity that have superb azimuthal detail. However, the width of the WWV pulse (5 ms) limits the radial resolution of such maps (see, e.g., the 20 MHz map in Figure 18). A different transmitter would need to be employed (e.g., the relocatable OTH radar (ROTHR) near Freer, Texas) to produce maps with comparably high resolution in the radial direction.

[77] An alternative transmitter similar to that used with the Texas ROTHR may also provide the means to use LWA1 for other real-time applications requiring better Doppler capabilities than are allowed by the WWV pulse cadence. For instance, with a well chosen/placed transmitter, LWA would be ideally positioned to monitor wave activity and currents simultaneously in the Gulf of Mexico and the Pacific Ocean via Bragg scatter of HF signals.


[78] This research was supported by an appointment to the NASA Postdoctoral Program at the Jet Propulsion Laboratory, administered by Oak Ridge Universities through a contract with NASA. The authors would like to thank F. Schinzel and T. Pedersen for useful comments and suggestions. Basic research in astronomy at the Naval Research Laboratory is supported by 6.1 base funding. Construction of the LWA has been supported by the Office of Naval Research under contract N00014-07-C=0147. Support for operations and continuing development of the LWA1 is provided by the National Science Foundation under grant AST-1139974 of the University Radio Observatory program. Part of this research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration.