Persistent Wide-Area Maritime Surveillance Using Smallsats With MIMO Radar Beamforming

This article presents an application, using existing technology, that would implement persistent (uninterrupted) noncooperating maritime surveillance of ship targets over wide areas of the Earth's surface. Persistence is essential for detecting and monitoring fleeting events, such as piracy, cargo transfers at sea, human migrations, and ship capsizings. The application would employ active radars aboard a constellation of small satellites (smallsats) placed in geostationary orbit. It would be augmented with the employment of existing multiple-input–multiple-output beamforming technology in order to reduce, to practical levels, the number of smallsats needed. An example is presented in which such a constellation would provide L-band surveillance of seas adjoining southern Europe, including the Mediterranean, Adriatic, Aegean, and Black Seas. An investigation of ship detection in the presence of sea clutter is included. The major design challenges are assessed to include the minimization of aggregate payload launch mass and the minimization of the data processing burden. In addition, comparisons are made with alternative approaches. Finally, in view of the large scale of such a prospective deployment, a risk reduction plan is offered, along with suggestions for design variations.


INTRODUCTION
The remote surveillance, tracking, and identification of both civil and military ocean ship traffic have been a subject of considerable importance for many decades.Soldi et al. [1] provide an excellent summary of this subject and review the use of passive spaceborne sensors [e.g., electrooptic, radio frequency (RF), communication receivers] for the surveillance of broad-ocean ship traffic.They also include a discussion of the applicability of synthetic aperture radar (SAR) operating in low-Earth orbit (LEO).
However, none of these surveillance methods provides for persistent (uninterrupted) coverage.Cloud cover and darkness can thwart electrooptic surveillance.LEO SAR entails unacceptably long revisit intervals.Persistence is essential for the continuous surveillance of vessels, including those employed by terrorists, pirates, smugglers, traffickers, and other rogue mariners, as well as by naval forces, that may not be using the cooperative RF Automatic Identification System (AIS) established by international convention [2].It also permits the continuous monitoring of naval blockades as well as humanitarian naval corridors as they develop.
Of particular interest is the detection of fleeting events, such as human migrations, capsizings, cargo transfers at sea, or inadvertent contact with sea mines.
As is argued later in this article, the only practical way of providing the needed persistent and all-weather continuous surveillance is via a radar deployed in geostationary orbit (GEO).A prospective system, using existing technology, for such surveillance is described herein.
Due to the very large slant ranges from GEO, a radar configuration of sufficient size is necessary to attain the needed detection sensitivity and to preserve an acceptable angular resolution on the ocean surface.If a traditional dish reflector or phased array were employed, this would require a huge, undoubtedly prohibitive, structure.Instead, an array of identical separated small satellites (smallsats) can be employed to overcome this obstacle.(Smallsats have been in use for several decades; thousands have already been placed into orbit for various applications.)Figure 1 depicts such a prospective GEO deployment.The associated surveillance swath size and shape on the Earth's surface is determined by the size and shape of a single smallsat, whereas the resolution cells are defined by the physical extents of the smallsat array.
To augment such an array of smallsats, it is extremely advantageous to connect them coherently in a multipleinput-multiple-output (MIMO) beamforming configuration.This serves to greatly reduce the number of smallsats required, thereby limiting the cost and launch mass of the GEO constellation.(MIMO technology, described in [3] as well as in numerous other works, has been employed for various applications over roughly two decades.) Note that what is presented here is an outline for a prospective and untested system, rather than a detailed engineering design.The latter should be carried out as part of subsequent efforts by appropriate specialists, including those experienced in spacecraft engineering development.Thus, some of the quantitative parameters presently suggested should be regarded as approximate or tentative.
Regardless of the design details, an ambitious program is described herein, but one that could provide a powerful maritime surveillance capability.
The section "Background" provides a review of some of the relevant phenomenology and general technical aspects.We then present the section "Example" in which a surveillance swath is created over a maritime region including the Mediterranean, Adriatic, Aegean, and Black Seas.The section "Specific Technical Items" contains a more general discussion of various technical matters and tradeoffs.We conclude with the section "Summary." Author's current address: The author resides in Washington, DC 20007 USA (e-mail: weissman.ike@ieee.org).

PHENOMENOLOGY
We consider here, nominally, ships exhibiting radar cross section (RCS) values of þ20 dBsm and above.Note that this is purely a convenient reference level and does not preclude detection of lower RCS targets given a sufficiently large smallsat configuration and sufficiently low sea clutter levels.
For viewing at low grazing angles, the minimum size of vessels falling into this category can be estimated from the empirical formula presented in [4], which is based on measurements by the U.S. Naval Research Laboratory: where f is the radar frequency in MHz and D is the fully loaded ship displacement in kilotons.In the microwave region, the above-mentioned minimum RCS applies to vessels that are fairly small, about 150 tons for L-band operation, which is typically that of small or mediumsized yachts, or of the size of World War II patrol torpedo (PT) boats.Even smaller vessels may fall into this category if populated with migrants situated above deck.Skolnik [4] also presents RCS variations with aspect angle (angle between ship's heading and the horizontal component of the radar line-of-sight) for a typical case at X-band.A variation (less than 10 dB) with aspect angle, excluding direct port or starboard incidence, is indicated.
In [5], a different approach to estimating ship RCS was suggested, namely, one based on a ship's physical dimensions.Assuming a reflectivity of À10 dB, an RCS value of þ20 dBsm would then correspond to a physical area of about 1000 square meters, which would indicate somewhat larger ship classes than those corresponding to (1).However, as pointed out in [5], this ". ..does not take into account special conditions like small ships carrying on-board radar corner reflectors...," meaning perpendicular surfaces above deck, which could raise the RCS levels significantly.
In [6], an empirical relationship based on a regression of horizontally polarized C-band RADARSAT measurements was presented.That relationship relates ship RCS to vessel length, and yields a length of 21.2 m for a ship RCS of þ20 dBsm at moderate grazing angles.
Ideally, we assume that the ship Doppler velocities remain approximately constant over a coherent radar integration period using pulse-Doppler waveforms.This may be true in portions of the surveillance swath exhibiting calm seas.Practically, however, there will certainly be regions where such coherence is degraded by a component of quasi-random ship motion, due to rougher sea states (wave heights).This will cause growth in Doppler sidelobe levels and reduction of overall coherence time.The consequent decrease in signal-to-noise ratio (SNR) may require consideration of larger minimum RCS values   (though providing more frequent samples for noncoherent integration).
Sea clutter represents another important consideration since it also can limit the detection of small ships.Generally, sea clutter RCS increases with increasing sea state, increasing grazing angle, and shorter radar wavelengths.Also, the clutter RCS tends to be higher for vertical, as opposed to horizontal, beam polarization relative to the sea surface.

SMALLSATS
Due to their small size and weight, smallsats can be deployed in large numbers in one or a few launch operations.They can subsequently be spread out in space to form much larger effective phased-array apertures than those practical with conventional spacecraft antennas, thereby providing substantially finer angular resolution.
Smallsat configurations provide the additional benefit of graceful degradation; that is, unlike single high-value satellites, destruction of one or a few smallsats (such as by collision with orbital debris) need not terminate the mission.Malfunctioning smallsats can be readily replaced.Moreover, the configuration can be enlarged, as desired, in successive campaigns.We use 500 kg as the upper limit for the mass of a single smallsat, but the mass can be much lower for many applications.

MIMO
The consideration of MIMO beamforming for multiplying the number of virtual receiving elements has been the subject of investigations for roughly the last two decades.
As explained in [3], as well as in many other works, there are two different classes of MIMO radars: a) "statistical" MIMO radar, which uses widely separated antennas to observe target fluctuations as a function of aspect angle, and b) "coherent" MIMO radar, where antenna elements are sufficiently close to limit target echo spatial decorrelation.For the subject of this article, we consider only coherent MIMO radar.
One way of describing coherent MIMO radar operation is by comparing it to a conventional phased-array radar.For the latter, the same waveform is transmitted and received by each array element.In contrast, in a typical MIMO operation each transmitting element sends out a different waveform, meaning a separate modulation on a common carrier, to an overlapping swath area on the surface.Each receiving element receives and separately processes all transmitted waveform echoes.Ideally, the waveforms are orthogonalthat is, uncorrelated-to facilitate their separate processing by matched filters.
Figure 2 illustrates two MIMO configurations: The "baseline" configuration depicted is preferred in order to permit straightforward augmentation of the configuration, whereas the "picture frame" configuration is more compact in space.
For purposes of explanation, separate linear transmitting and receiving arrays are shown, but it is usually more advantageous to employ all smallsats for both receiving and transmitting.It is important to note that a virtual aperture element appears at the midpoint of each transmit and receive pairing [3].

SYSTEM CONSIDERATIONS
The great advantage in locating the smallsat-MIMO maritime surveillance radar configuration in GEO is due to its orbital rotation rate which, with proper adjustments, is the same as that of the Earth, thereby resulting in the persistent surveillance of the selected swath on the Earth's surface.This also means that, for a fixed swath, no mechanical or electronic scanning of the smallsat antennas is needed.
The surveillance swath extents are determined by the size and shape of each smallsat's antenna aperture, the radar frequency, the slant range from GEO, and the beam incidence angle on the surface.
The spacings between adjacent smallsats must be restricted to minimize grating lobes in the physical receiving beam pattern.As is well known, this can be roughly accomplished by choosing the spacing so that the grating lobes are placed near the nulls of the individual smallsat angular beam pattern, as given in [7].It turns out that application of Frank and Richards [7] can approximately satisfy this when adjacent smallsats, with practically sized antennas, are physically attached to each other.(This also helps to stabilize relative smallsat positioning and orientation variations within the configuration.) The total number of smallsats is then determined by the physical extents of the perpendicular linear arrays comprising the configuration; the latter are, in turn, set by the desired angular beamwidth (cross-range resolution) of the virtual aperture.
The two main drivers regarding the practicality of the system design are the aggregate smallsat mass to be launched into orbit and the digital data processing burden.Each of these can be affected by certain system design tradeoffs.
If all of the smallsats in a configuration are identical, with each both transmitting and receiving, the aggregate mass is then approximately the product of the number of smallsats and the mass of each one.
The mass of each smallsat is largely driven by its antenna size and the average transmitted power (and its prime power components).Since the detection performance in noise is governed, for a given target RCS and swath size, by the product of average transmitted power and receiver aperture area, the smallsat mass is roughly minimized by balancing the mass contributions of its antenna and that of its power chain.Such a balance can be affected by a choice of radar frequency, notwithstanding other factors in this choice.(For example, the choice of a higher frequency reduces the needed size of the smallsat antennas, but therefore dictates a higher average power.) It is obvious that the aggregate mass can be greatly reduced by also setting an appropriately larger minimum target RCS, as discussed further in the section "Implementation Variations." With respect to the other principal driver, namely that of the digital data processing burden, this is mainly dependent on the number of smallsats, the radar waveform, and the number of virtual beams created.
There are generally two stressing processing functions and both have to do with computational throughput: a) the pulse compression for the matched filtering for every orthogonal transmitted waveform in every receiver (each matched filter output corresponding to a particular element of the virtual array), and b) the target detection process with a pulse-Doppler waveform in each interrogated virtual radar beam.In both cases, the burden is mitigated by applying the fast-Fourier transform (FFT).
Additional mitigation can be achieved by limiting the virtual beams interrogated to those illuminating only the sea surface, and ignoring those impinging on land.Further relief can be attained by processing, for each beam, only those range cells within that beam (which are known a priori) and also by limiting the waveform's RF bandwidth.
With respect to the latter limitation, it can be seen that there is a tradeoff between the number of range cells to be processed (which decreases with decreasing bandwidth) and the tolerable sea clutter levels (which increase with decreasing bandwidth).
These driving considerations do not represent fundamental theoretical roadblocks, but may instead affect the amount of hardware required.

EXAMPLE
Figure 3 is a map depicting an example of a surveillance swath as seen from GEO and encompassing much of southern Europe, and including the Mediterranean, Adriatic, Aegean, and Black Seas.The location of the radar configuration above the equator is offset in longitude to reduce the grazing angles relative to the sea surface, in order to minimize the sea clutter without significantly increasing the slant range from GEO.
A radar frequency at L-band (0.25 m wavelength) is chosen, low enough to minimize clutter RCS and decorrelations due to ship motions but also high enough for avoiding a significant degree of ionospheric Faraday rotation.
We assume the case where all smallsats contain both a transmitter and receiver.To create the large surveillance swath (shown in Figure 3 as an approximate overlapping 3-dB illumination contour from each transmitter), we employ a smallsat individual aperture of 4.33 m Â 3.88 m. (This could be practically implemented with a nonscanning reflector.)Horizontal polarization would be employed, mainly to minimize sea clutter returns.
If we desire a cross-range resolution of 10 km on the sea surface, we would employ, for a "baseline" configuration, two perpendicular arrays of physical smallsats, each approximately 2 km in length.This is shown in Figure 4 along with a depiction of some of a smallsat's components.(The depiction is notional and does not represent an actual antenna design.Also not shown are vernier motors needed for smallsat orientations and alignments, as described further in section "Station Keeping.")In order to avoid receiving array grating lobes, the smallsats must be closely spaced, by a maximum determined by a smallsat's beamwidth (in other words, by the individual transmitting directive gain); in this example, the smallsats are thus attached to each other, as mentioned afore in section "System Considerations."A total of approximately 1020 smallsats is therefore needed.Note that, due to their fixed nonscanning employment, as well as their multiwavelength spacing, deleterious mutual coupling effects among the smallsats are minimized.
To limit the sea clutter return in each resolution "cell," we choose a radar pulse bandwidth of 0.1 MHz, to provide a downrange resolution on the sea surface of about 1500 m divided by the cosine of the grazing angle.(Grazing angles at several locations along the swath contour are indicated in Figure 3.) Multiple vessels that may appear within such a resolution cell can mostly be resolved on the basis of differing Doppler returns.
The aggregate transmitted power must be sufficient to assure an acceptable detection sensitivity.In this regard, we note that the RCS of ship targets, fortuitously, tends to be large and can cover a wide span, assumed here as ranging from þ20 dBsm to over þ50 dBsm depending on the ship size and displacement, as discussed in the section "Phenomenology."We rearrange the radar equation in order to calculate the aggregate average RF transmitted power requirement SP av,, to attain an integrated SNR of 10 dB: where the various parameters, and the values used, are contained in Table 1.
Note that the total losses (L) assumed in the table include the aperture efficiency and the two-way beamshape loss at the contour of the surveillance swath.
The calculation shows that an aggregate RF average transmitted power level of 6.1 kW is required.This corresponds to an average transmitted power of about 6 W per smallsat.
Figure 5 depicts an appropriate waveform, using pseudorandom binary-phase coding (indicated by pluses and minuses for the subpulses).In order to avoid rangeambiguous echoes arising from the entire Earth's surface viewed from the GEO position, a pulse spacing of about 62.5 ms is chosen.(Note that the precise pulse spacing should be slightly varied over successive integration intervals in order to sort out possible Doppler ambiguities.)This allows also for an individual pulsewidth of 1.28 ms, corresponding to a duty factor of about 2%.The resulting peak power requirement for each smallsat is, then, about 300 W, a value readily available at L-band from a single gallium nitride transistor.
The assumed 0.1-MHz pulse bandwidth corresponds to a pulse compression ratio of 128.For the binary-phase Waveform for illustrative case.
subpulses illustrated, a huge number (2 128 ) of pseudorandom codes is therefore available, from which 1020 of the most suitable orthogonal or quasi-orthogonal transmissions can be selected.
To assess the detectability of ship targets of interest in significant sea clutter, we consider a "worst case" combination of a lower-end ship RCS of þ20 dBsm and seastate 6 wave height.
Based on [8] and [9], and assuming target phase coherence over an 8 s integration period, a marginal signal-toclutter ratio (SCR) (roughly 2 dB) is attained at about 38 grazing angle by Doppler filtering, but a comfortable SCR (about 20 dB) is calculated at the lowest grazing angles along the swath contour.Of course, there would be a more favorable result (higher SCR) for larger ships or calmer seas anywhere within the assumed surveillance swath.On the other hand, the SCR could be degraded with shorter target coherence durations (which could occur in rough seas).In that case, a minimum ship Doppler velocity can be established above which a target effectively clears the clutter spectrum.(These clutter calculations assumed a 3-dB clutter Doppler spread of about 3 m/s at sea-state 6 [9].) It should be noted that the clutter levels referred to in [8] are nominal values and, as noted in [8], could be in error by as much as 5 dB at moderate grazing angles.Furthermore, these levels do not take wind direction into account.In [5] and [6], a more comprehensive model for sea clutter behavior is employed, together with an algorithm to determine ship detectability relative to a clutterbased constant false alarm rate threshold level.
We must also assure that cumulative range sidelobes and virtual beam sidelobes from clutter remain sufficiently below ship target echoes.If needed, the cumulative range sidelobes can be sufficiently reduced by moderate tapering of the spectrum of the subpulses, resulting in a slight increase of the subpulse and pulse durations.As for virtual beam sidelobes, we can again consider a worst case: namely a small ship with RCS of þ20 dBsm and located at the maximum grazing angle for the swath (about 38 ).The latter sidelobes originate from an accumulation of all beams along a ring defined by a particular projected range resolution cell.Calculations then show that we need the sidelobes from each virtual beam to remain below about À30 dB; this requires moderate tapering of the virtual array during the beamforming process.
Absent a detailed smallsat design, a close estimate of the aggregate mass of this illustrative configuration would be premature.However, based on an overview of other spaceborne antenna structures [10] and solid-state transmitters aboard smallsats [11], a goal of 50,000 kg for the aggregate payload does not seem unreasonable.(This estimate used 1.0 kg/m 2 for the antenna aperture and 5 kg per watt of average RF power for the power chain within a smallsat.)The need for multiple launches to deploy such an aggregate mass would likely be the case.Note, however, that this mass goal corresponds to the most stressing case considered, and less ambitious objectives (e.g., a minimum ship RCS of þ30 dBsm) would lower this aggregate mass considerably, as suggested in the section "Implementation Variations." A generic digital data processing architecture for the proposed concept is discussed in the section "Data Processing."For the example presented here, the important values are a) approximately 2 terabytes of storage for the fixed beamforming relationships; b) about 0.3 petaFLOPS throughput, using FFTs, for the 1020 Â 1020 in-phase and quadrature matched filters; and c) very roughly, 5 peta-FLOPS of throughput for the range-Doppler FFT for target detection in the restricted set of virtual beams, that exclude land areas.The storage amount indicated is commonly available.The throughput requirements cited are more difficult; however, a relatively compact 5-petaFLOPS graphics processing unit with a packaged weight under 200 kg, appears to be available [12].If the latter proves not to be adequate, two or more such units can be combined.
The GEO-to-Earth downlink data rate for this example is estimated to be approximately 160 MB/s.

STATION KEEPING
There are requirements, both formal and less so, for the location of the GEO configuration.To ensure sufficient separation among the numerous satellites, of different types, in GEO, a United Nations committee is responsible for assigning specific longitudes for their placement [13].Fortunately, a specific assignment for the currently considered configuration is not critical; effectively the same swath can be created even if the assignment is displaced by a few degrees longitude.
One effect that must be taken into account is a slight variation in gravitational forces due to the relative motions of extraterrestrial bodies that could affect the exact location of the configuration, and this must be compensated by the use of the thrusters attached.
Corrections are also required for perturbations of the exact location and pointing direction of individual smallsats.Slight relative variations in altitude can likely be compensated by small RF carrier phase adjustments.However, more generally, surface-based (preferably land-based) calibration sources should be included, with individual smallsat adjustments controlled by the on-orbit processor.

PRIME POWER
For almost all of a geostationary constellation's orbit, direct solar radiation is available for prime power (unlike Persistent Wide-Area Maritime Surveillance Using Smallsats With MIMO Radar Beamforming LEO orbits).Lelikov [14] reported an available prime power of 1350 W per square meter of solar panel.This assumes, of course, that the orientations of the solar panels are continually adjusted to face the sun.For example, a modestly sized panel of 0.2-m 2 area on each smallsat should then be able to generate about 270 W of prime power.Due to the perpendicular linear arrays in the assumed physical configuration, the panels would be effectively unobstructed, as illustrated in Figure 4.
For about 97% of the daily orbit, the radar configuration is clear of both the Earth's shadow and penumbra.Thus, stored energy should be used for prime power for close to 2590 s per daily orbit.For the example in the section "Example," a rechargeable battery capable of storing about 200 W-hours of energy should then be included on each smallsat if persistent surveillance is to be preserved.The mass of such a battery is reported to be less than 2 kg [15].

TARGET CLASSIFICATION
For the maritime surveillance application considered here, only targets on the Earth's sea surface are of interest.Large aircraft and orbiting satellites will inevitably appear within the field-of-regard from GEO.However, for any beam synthesized for the virtual array, a range measurement less than a certain beam-dependent value would be rejected as a target not located on the sea surface.That still leaves us with the need to filter out certain types of aircraft (e.g., helicopters, uninhabited air vehicles) flying at low altitude (that is, below a threshold set by the spread of range measurements in a particular beam).By setting variable RCS and Doppler thresholds, below and above which, respectively, targets would be rejected could be helpful in this regard.
A remaining question is to what degree ships themselves can be identified-or at least classified-by noncooperative means.Clearly, the method presented does not provide a direct imaging capability, except perhaps during sharp turns of a particular vessel.However, a number of other radar-derived features may help in this respect.These may include direct observables such as ship RCS, speed, direction, and, for architectures able to use much wider bandwidths, downrange radial extent.Also, the surveillance context can be revealing, such as a ship's geographic location and grouping with other vessels (e.g., a naval task force).

SHIP MOTION
A concern expressed in the section "Phenomenology" is the shortening of a target's coherence time over a radar integration period, particularly in rough seas.If it is desired to preserve, or even extend, the coherence time, more sophisticated detection algorithms may be needed.
One candidate for the latter is the micro-Doppler technique [16], as applied to ship motion.This technique is able to extract the main component of a target's motion spectrum which, in our case, would correspond to a ship's forward linear transit over an extended time period.

MUTUAL COUPLING
One concern that has been raised in connection with MIMO operation is that of mutual coupling among antenna elements.Any mutual coupling effects here, however, are probably less significant than those encountered in a conventional phased-array antenna.For one thing, adjacent smallsats here are spaced by many RF wavelengths because of the relatively small beamwidths needed to cover the desired swath.Second, impedance matching in the physical elements should be straightforward since the virtual array does not scan; instead, the virtual beams maintain the same relative pointing for the duration of any measurement.Frank and Richards [7] indicate that impedance matching is simpler for nonscanning antennas.

DATA PROCESSING
As mentioned earlier, the data processing burden is undoubtedly one of the major issues for engineering developers of the present application.The degree of this burden (in throughput, storage, or short-term memory) depends on the parameters of the smallsat-MIMO configuration.Generally, the burden increases with the swath size, the number of smallsats, the system bandwidth, and the number of pulses coherently integrated.
A conceptual architecture for the data flow is shown in Figure 6.
The signals received at each smallsat would first be digitized for both in-phase and quadrature (I and Q) components.(The A/D converter dynamic range must be sufficient to accommodate the uncompressed (narrow-band) single-pulse clutter levels.)Subsequently, the raw digital data would be formatted and downlinked for the main processing.
Nearly all of the intensive processing would take place at a facility on the Earth's surface, preferably at a land site, with other functions (formatting, smallsat alignment, station keeping, subarray formation if included, etc.) carried out in space at the processor included in the GEO physical configuration.
The surface hardware would first perform the FFT matched filtering for every transmit/receive pairing.The I and Q outputs would then be fed to the beamformer, in which each input (corresponding to a specific known transmit/receive pairing) is assigned to the appropriate element of the virtual array.(For a fixed measurement application, these assignments would be unchanging.) The final step would be the extraction, recording, and display of the entire target environment.For this, each virtual beam of interest would be processed separately, with the pulse-Doppler signal subjected to FFT processing to provide the target information for the range-Doppler cells within each beam for every successive coherent integration interval.
For "clean" display and recorded outputs, thresholding of the processed data is essential.Clearly, the placement of thresholds would vary from beam to beam, as well as with time-varying clutter levels.
One additional advantage of operating from GEO is that, unlike MIMO operations from LEO, the Doppler spread due to the surface echoes is negligible, thereby avoiding a multiplication of the number of matched filters to handle signals with different Doppler velocities.

RISK REDUCTION
The general approach seems to be practically feasible because of the use of MIMO in conjunction with the smallsat configurations.However, technical difficulties may arise during actual engineering development.To mitigate risks associated with such prospective difficulties, a staged approach is suggested.
Figure 7 lists some of the important technical investigations that should be carried out as part of this approach.These would be followed by fabrication and launching of a limited number of smallsats.Assuming successful operations of this limited configuration, the configuration would be enlarged and its products validated-and so forth.

ALTERNATIVE GEO RADAR APPROACHES
It is useful to briefly discuss some possible alternative approaches for providing the desired persistent wide-area maritime surveillance, which is the subject of this article.Several of the most obvious are discussed here.
The first would employ a straightforward phased-array non-MIMO configuration in GEO that would use a smallsat for each array radiating element, with a spoiled transmit beam and digital beamforming to create the receive beam cluster.This would avoid the burden and complexity of the data processing needed for a MIMO configuration.However, the number of smallsats for comparable L-band  performance would be huge.For example, as compared with the example presented in the section "Example," a total of approximately 64,000 smallsats would be required, each with the same antenna dimensions.This would result in a very large aggregate launched payload mass; for typical planar antenna specific mass values assumed in this article (roughly 1.0 kg per m 2 of aperture), an extremely large aggregate launch mass, close to one million kilograms, would result.The large number of smallsats would also increase the total production costs.
(This arrangement can be considered to be a "singleinput-multiple-output," or SIMO, smallsat configuration.)A different alternative approach would employ geosynchronous synthetic aperture radar (GEOSAR).It is discussed here because it has been suggested that it would be the equivalent of the approach in this article.
GEOSAR is a desirable and seemingly popular approach for shortening the revisit time to 24 h, compared to a conventional SAR in LEO.Apparently, no GEOSARs have yet been launched, but a number of GEOSAR programs are in the planning stages.They differ from geostationary deployments by introducing an inclination angle relative to the equatorial GEO orbits.This results in a satellite velocity relative to features on the Earth's surface, thereby permitting SAR operation.The ground track of the GEOSAR beam scans the observation sector in a "figure-eight" pattern, revisiting daily and with varying velocity along its ground track.
The vast majority of GEOSAR concepts are intended for land mapping, for which SAR integration times are as much as many tens of minutes in order to preserve the SAR resolution considering the very large range from geosynchronous orbit [17].An exception to this, intended for maritime surveillance, is presented in [18].However, the latter's cross-range SAR resolution, as limited by ship target coherence and dependent on the GEOSAR latitude excursion, is roughly comparable to that of the geostationary configuration described herein.
Hence, the required aggregate power-aperture products of the configurations are roughly the same, and the only practical approach for this with GEOSAR is to employ a large smallsat-MIMO arrangement, as with the subject geostationary configuration.(An approach, using a much smaller GEOSAR-MIMO constellation, but not applicable to the present case, is described in [19].) The GEOSAR alternative also suffers from certain relative complications.Associated with the daily "figureeight" surface projection is a varying velocity relative to the Earth's surface.This results in a varying cross-range resolution along the ground track for the fixed coherent integration intervals, thereby significantly adding to the data processing burden, as well as complicating interpretation of the measurements.Additionally, as a result of the constellation's changing latitudes, the antennas of the individual smallsats must be continuously adjusted mechanically in order to maintain the desired fixed surveillance swath.Presumably, this will have the effect of shortening the constellation's useful performance life in orbit.
Table 2 contains a qualitative comparison of the various configurations considered for persistent maritime surveillance.
Particular attention is directed to the column on the far right of Table 2. LEO SAR satellites have been contributing very valuable surveillance of ships and other targets for many decades, but it should be understood that a single LEO satellite cannot provide uninterrupted observation of any specific target within a surveillance swath.In fact, a typical revisit interval is a matter of calendar days, during which several successive ship crossings of the Mediterranean Sea, for example, can take place undetected.
It seems that the only way of achieving the desired persistence from LEO (or, for that matter, from medium-Earth orbit) is to provide enough satellites over the entire Earth (or at least everywhere between a pair of northsouth latitudes, if a prospective surveillance swath can be restricted to these).But these numbers would be huge and the aggregate launch mass of such a constellation would be enormous.
Consider, for example, the Sentinel-1A satellite, which has provided an abundance of C-band SAR data since 2014.It operates in a near-polar orbit at an altitude of 693 km, with a maximum instantaneous swath size of 410 km [20].Assuming a latitude range of interest between 30 and 45 , it turns out that roughly 2000 such satellites would be required.Combining this number with the approximately 1000 kg of mass for the single radar payload [20] aboard Sentinel-1A, we can estimate an aggregate mass of such an LEO constellation to be, very roughly, two million kilograms.

IMPLEMENTATION VARIATIONS
The illustrative example presented in the section "Example," although physically realizable, entails a number of fairly stressing design parameters.These can be mitigated by relaxing certain system parameters.
The most obvious adjustment is to relax the minimum RCS for ship detection, thereby reducing the required individual and aggregate average RF power transmitted, with a consequent reduction in payload launch mass.For example, by raising a minimum ship RCS specification from þ20 to þ30 dBsm, it is estimated that the aggregate mass can be reduced by more than 50%.(The reference ship displacement observed, using (1), would thereby be increased from approximately 0.15 kiloton to approximately 0.7 kiloton.) The potential benefits of such an application are substantial.However, although there appear to be no fundamental obstacles to such an implementation, there are significant design challenges, including the minimizations of the aggregate launch mass and the data processing burden.A risk reduction program is suggested to address these challenges by implementing the application in distinct stages.
For persistent maritime surveillance, there seems to be no better practical alternative to geostationary radar deployments.
Image licensed by Ingram Publishing

Table 1 .
Radar Equation Parameters Used