Digital beamforming techniques applied to satellite-based AIS receiver

4 IEEEA&ESYSTEMSMAGAZINE JUNE2014

INTRODUCTION

The Automatic Identification System (AIS) is a transpon-der system for ships used to increase the safety at sea [1]. It operates in the very high frequency (VHF) band, and the transmitter regularly transmits the ship’s position, heading, speed, and the unique maritime identification number (plus other data). These data are received by ships in the vicinity and by base stations along the coast.

For open sea scenarios, where no AIS terrestrial coverage is available, a fleet of low Earth orbit satellites at an orbital altitude of 500–600 km equipped with a satellite-based AIS receiver (SAT-AIS) may effectively integrate the existing ter-restrial infrastructure and cooperate with it.

The vessels signal is Gaussian-minimum shift key-ing (GMSK), modulated over two VHF channels, and the transmission coordination is based on self-organized time division multiple access (SO-TDMA) that requires strict syn-chronization of vessels signals [1].

Unfortunately, vessel signals in the satellite field of view (FOV) are not all synchronized; the “self-organized area” (i.e., the area where there are no time slots conflicts) dimen-sion is constrained by the ship-to-ship communication range, which is typically on the order of 20 nautical miles in radius. Therefore, because the satellite looks at many adjacent self-organized areas at the same time, interference occurs. More-over, received signals suffer from different delays and loss, due to the spreading in path lengths, Doppler shift caused

by the satellite orbiting motion, and ionospheric influence (Faraday rotation of linear polarization) [2]. To mitigate the previously introduced problems, some techniques, such as digital beamforming (DBF) [3] or multiple-input multiple-output [4], should be applied.

This article investigates using DBF onboard AIS satellites with the objective to increase the signal-to-interference-plus-noise ratio (SINR) so that the number of decoded AIS mes-sages can be increased. A DBF system is based on an array of antenna elements acting as independent receivers that cap-ture radio frequency (RF) signals. These signals are convert-ed into two digital streams of baseband I and Q signals, and then they are weighted (by changing their amplitude and phase) so that when they are combined together (summed) they create a desired output (i.e., the AIS signal of interest with a high level of SINR).

The article presents an antenna array with a single-ele-ment radiation pattern optimized to mitigate interference. Based on this antenna array, static and adaptive DBF tech-niques have been compared in terms of SINR increase and system complexity (i.e., need for a large antenna, computa-tional complexity of an adaptive DBF technique, etc.).

AIS SATELLITE SCENARIOS

To understand and assess the main characteristics of the in-terference signals, a typical AIS scenario has been employed using a system simulator developed by the European Space Agency (ESA) [5]. This tool is able to model the full SAT-AIS system and related natural phenomena so that it is possible to characterize the radio environment at the satellite recep-tion antenna. The vessel distribution over the world can be simulated together with their dynamics and transmitters characteristics. The simulator inputs are satellite constella-tion and array antenna characteristics.

The simulator computes the orbit propagation of the sat-ellites and determines the vessel-satellite visibility. For every message received by each satellite of the constellation, the simulator computes a link budget considering channel loss-es, Faraday rotation, propagation delay, Doppler shift, etc. On the basis of the onboard antenna characteristics, informa-

Digital Beamforming Techniques Applied to Satellite-Based AIS ReceiverFabio Maggio, Tommaso Rossi, Ernestina Cianca, Marina Ruggieri University of Rome Tor Vergata

Authors’ current addresses: F. Maggio, T. Rossi, E. Cianca, M. Ruggieri, University of Rome Tor Vergata, Department of Elec-tronic Engineering, via del Politecnico 1, CAP 00133, Rome, Italy, E-mail: [email protected]. This work was supported by the European Space Agency (ESA) under Project “AIS End-to-End Testbed,” contract num-ber 4000103249/11/NL/AD. Manuscript SYSAES-2013-0168r was received September 19, 2013, revised November 15, 2013, and ready for publication December 20, 2013. DOI. No. 10.1109/MAES.2014.130168. Review handled by R. Wang. 0885/8985/14/$26.00 © 2014 IEEE

JUNE2014 IEEEA&ESYSTEMSMAGAZINE 5

tion concerning received signals power, frequency, time of arrival, and phase is made available by the simulator The fi-nal product of the simulator is the baseband signals, as seen at the output of every antenna of each satellite; these signal snapshots are sampled at 96 kHz [5].

To characterize the SAT-AIS scenario, a single satellite orbit of 90 minutes has been simulated through the previ-ously described tool; the satellite altitude is about 600 km. The analysis of the received AIS signal distribution in sev-eral scenarios showed that it is very likely that tens of signals are included in a single AIS time slot of 26.7 ms, when the satellite is passing areas with high traffic density, such as the Mediterranean Sea or the Northern Sea.

Figure 1 shows a very typical distribution of vessels (from the satellite perspective point of view), transmitting concurrently within the same time slot; the satellite is locat-ed over the Mediterranean Sea, and the number of vessels transmitting AIS signals within the selected time slot is 27.

Note that most of the transmitting vessels usually group together in clusters along the horizon line because the ho-rizon is viewed from the satellite with grazing incidence. Therefore, many SO-TDMA areas overlap each other, and the probability of receiving interfering signals from the same direction within a single time slot rapidly increases. These near-horizon signals are a serious problem for the decoding of AIS messages because

C they cannot be spatially resolved by the satellite array antenna; it is well known that no signal detection (SD), directions of arrival (DOA), or nulling algorithm can manage such a high number of signals coming from the same direction, particularly when the number of antenna “degrees of freedom” (i.e., complex weights, and, finally, array antenna elements and independent beams) is small [6];

C they are transmitted with a high gain of the dipole antennas onboard the ships, even if their path loss is high; and

C they cannot be discriminated in terms of Doppler shift or Faraday rotation; conversely, they exhibit different times of arrival.

Hence, near-horizon signals should be spatially filtered out by a proper designed antenna array, as the one presented in next section.

DEFINITION OF THE ANTENNA ARRAY

In this section, the main characteristics of an antenna array that can provide a solution to the previously introduced problem of near-horizon signals interference are identi-fied. It is known that an array antenna is the most used and well-established solution to realize a reconfigurable radia-tion pattern, together with the application of beamforming techniques. This kind of technique is able to create a mul-tispot synthetic coverage, which acts as a bank of spatial filters. Ideally, to obtain a highly selective spatial filtering, the region of interest should be covered by a high number of very narrow beams; nevertheless, the beam size is lim-ited by the physical array dimension, while the number of

Figure 1. Perspective view of the distribution of vessels transmitting within the same time slot.

6 IEEEA&ESYSTEMSMAGAZINE JUNE2014

Digita l Beamforming Appl ied to AIS Receiver

synthetic spots is actually the number of available beam-formers.

The objectives of the array antenna presented in this sec-tion are

C to filter out the near-horizon signals; and

C to provide enough degrees of freedom for the imple-mentation of DBF techniques able to increase the SINR.

In the analysis scenario, the signals coming from the ho-rizon of the satellite receiver FOV are placed at about 66 de-grees of elevation.

Therefore, an antenna array able to filter them out should have a gain that starts decreasing beyond 50 degrees in el-evation. A solution for the array radiating element can be proposed that places a null around 66 degrees in elevation.

In detail, an antenna solution based on two pairs of cross dipoles has been considered with the following characteris-tics:

C the dipoles length is about 900 mm;

C the interspace between a couple of crossed-dipoles is about 460 mm; and

C the distance between the two couples of crossed di-poles is about 1.8 m.

Each pair of cross dipoles is configured as an end-fire array, by means of quarter-wavelength spacing and a 90-degree phase shift, thus exhibiting a cardiod-shaped radiation pat-tern. The two end-fire arrays are spaced in such a way as to ensure a pattern null at 66 degrees in elevation. In this way, the boresight directivity of the overall radiating ele-ment slightly decreases, but this is not critical, because a very limited number of signals is received from nadir, due to the radiation properties of the antennas installed on-board the vessels (the vessel antenna element is a vertical dipole). Using cross dipoles, a dual linearly (or even circu-larly, if needed) polarized radiating element is obtained. A

total of 25 dual linearly polarized elements were arranged in a classical square-lattice planar array configuration (5 × 5), as depicted in Figure 2, with the elements spaced half a wavelength apart. The cross dipoles were rotated 45 degrees with respect to the main grid axes, thus the two linear polar-izations (P1 and P2) are rotated 45 and 135 degrees, respec-tively, referring to the x-axis. This array configuration has been selected as a tradeoff between the number of degrees of freedom required to cope with the AIS scenario complexity and the need for limiting the physical array size.

The theoretical radiation pattern of the array element was computed, not including mutual coupling effects, and is shown in Figure 3 in u-v antenna coordinates: as expected a deep null at 66 degrees in elevation can be noticed.

Due to the designed array size, 19 overlapped inde-pendent narrow beams (to be processed theoretically by 38 parallel beamformers, considering the two polarizations) completely cover the satellite FOV. The beams are placed on a regular hexagonal grid (in antenna u-v coordinates, the spacing is 0.32), as shown in Figure 4. In this way, the array antenna has enough “degrees of freedom” to cope with the high number of vessels that can be included in a single time slot. The multibeam configuration allows single or small groups of vessels to be isolated; hence, it is expected to im-prove the SINR and the detection probability. Note that the patterns shown in Figures 3 and 4 were computed theoreti-cally, on the basis of the geometry of the radiating element and the array configuration. Nevertheless, a first-level op-timization was carried out, tuning each cross-dipole length and spacing, plus the array excitation coefficients, by means of the method of moments, to assess more realistically the feasibility of the proposed antenna, with the desired radia-tion properties. It is clear that more sophisticated antenna-radiating elements could be selected to reduce the overall array size. Moreover, the problem of deploying such a large

Figure 2. Square array configuration (5 × 5).

Figure 3. Theoretical array element radiation pattern (P1 polarization).

JUNE2014 IEEEA&ESYSTEMSMAGAZINE 7

Maggio et al .

array onboard the spacecraft should be addressed, but all these issues are beyond the scope of this article.

DBF TECHNIQUES

The objective of this section is to provide guidelines for the choice of the most suitable algorithms for AIS satellite ap-plications and has not been thought as an exhaustive review

of all the existing DBF techniques. Two main types of DBF techniques can be found [6], [7]:

C static DBF, in which the number and properties of the antenna beams are fixed and do not vary with time; the different beams covering the satellite FOV can be syn-thesized either sequentially by a single beamformer or by a bank of parallel beamformers;

C adaptive DBF, in which the antenna weights are opti-mized according to suitable criteria, such as maximiz-ing SINR (and the detection probability).

Static DBF is obviously easier to implement, but, in prin-ciple, adaptive DBF could ensure better system performance. In both cases, the number of required system “degrees of freedom” is high because many transmitting vessels can be included in a single AIS time slot. Adaptive DBF techniques are based on the following assumptions:

C time slots scenarios are stationary, i.e., the spatial dis-tribution of the received signals does not change with-in each individual time slot (note that the scenario of the observed signals is expected to completely change from one time slot to the next); and

C received signals are uncorrelated.

Unfortunately, neither of these assumptions is satisfied for AIS satellite. As shown in Figure 5, due to the different delays experienced by the messages transmitted by the ves-sels within the same time slot, when observed by the satel-lite, a significant signal overlap between two adjacent slots occurs. This issue will be further discussed at the end of the Simulation Results sections.

In Figure 5, it can be noticed that only in the central por-tion of a given time slot, the scenario can be assumed as

Figure 4. The 3-dB contours of the 19 overlapped beams (for P1 polariza-tion only), in a realistic scenario; vessels transmitting within the selected time slot on Ch. 1 are shown in cyan, while Ch. 2 vessels are in red.

Figure 5. Overlapping among adjacent slots, when seen from the satellite.

8 IEEEA&ESYSTEMSMAGAZINE JUNE2014

Digita l Beamforming Appl ied to AIS Receiver

stationary, because all the received signals are certainly in-cluded, independently of their delays, and interference from the preceding slot does not occur.

Stationarity property of the scenario is important, be-cause most of the DBF adaptive algorithms need to measure a correlation matrix R, using a sequence of antenna output signals snapshots. Moreover, a certain amount of correla-tion exists among the AIS interfering signals included in a scenario, due to the common parts in the transmitted pack-ets (training sequence, start flag, end flag, etc.). Also, in this case, most of the adaptive algorithms do not work properly. For example, Capon beamforming [8], [9] cannot ensure op-timal SINR performance for a selected signal, because even the wanted signal is affected by the received power mini-mization process. On the contrary, some algorithms, such as minimum mean-square error [3], in principle, could use the known redundancy in the transmitted packets as a priori in-formation to optimize the antenna weights, matching each incoming interfered signal. This is, in practice, a very dif-ficult task, because the receiver should provide the beam-former with a reference signal, correctly positioned over

time, with the receiver already locked before the optimal weight estimation.

Moreover, many adaptive techniques, such as Capon beamforming, need a priori signal DOA information. DOA algorithms, such as multiple signal classification [10], in turn, require the estimation of the number of incoming sig-nals (SD). DOA and SD techniques are intrinsically time-consuming, and their reliability can be negatively affected by signals correlation. In addition, they need a high number of system “degrees of freedom” to work properly in an AIS time slot scenario, including many received signals.

Because the GMSK packets transmitted by the vessels are constant envelope signals, a “blind” beamforming tech-nique, such as the constant modulus algorithm (CMA) [3], [5] can be of great interest in an AIS application. Blind tech-niques do not require any prearranged estimation about the current scenario, such as SD or DOA, relying completely on some a priori known property of the incoming signals. CMA assumes that each received signal has a constant en-velope, iteratively optimizing the antenna weights and pro-viding the best approximation of a constant modulus signal at the beamformer output. Unfortunately, when many sig-

Figure 6. Vessel 4672 is detected by means of the main lobe of Beam 10, Pol. 2.

Figure 7. Vessel 33615 is detected by means of a sidelobe of Beam 8, Pol. 1.

Table 1.

Static Dbf Performance for the Test Scenario

Vessel Identification Longitude Latitude SINR Polarization

Beam Number Channel

4672 18.009 76.087 30.045 P2 10 2

33615 10.044 64.052 14.390 P1 8 2

39915 1.174 66.409 37.074 P1 15 1

67074 14.576 68.241 35.030 P1 13 2

JUNE2014 IEEEA&ESYSTEMSMAGAZINE 9

Maggio et al .

nals of similar power level interfere, as may be the case in an AIS scenario, it is not clear which signal CMA converges to. Thus, CMA can work properly only if there is a signal whose level is higher than the others, i.e., SINR is already higher than zero (this can be ensured if the array antenna is large enough to generate narrow beams). As a consequence, CMA is a good candidate for enhancing an already positive SINR; moreover, it can be useful to reach the receiver detec-tion threshold.

Finally, it should be pointed out that CMA, an iterative algorithm, may be incompatible with an onboard real-time beamforming operation, due to the strong computational ef-fort that is required; thus, signals received from the antenna elements should be carefully multiplexed and retransmitted to Earth stations for processing.

In conclusion, a static beamforming strategy using a set of as narrow as possible array beams is expected to be a good choice for implementing DBF. Then, the weights of the static beams could be used as starting points for an iterative opti-mizer based on CMA.

STATIC DBF SIMULATION RESULTS

The planar array antenna (5 × 5 ) described in Definition of the Antenna Array was used to assess the static DBF per-formance in the SAT-AIS scenario. Moreover, a comparison between a static DBF and a CMA blind beamforming tech-nique has been carried out. The static DBF was implemented for the 19 overlapped beams shown in Figure 4. Simulation results for a single time slot scenario are reported in Figures 6 and 7.

The scenario considered in the figures includes 50 trans-mitting vessels, 23 for Ch. 1 and 27 for Ch. 2; it concerns a time slot extracted from the 90-minute-long satellite track obtained through the simulation tool described in AIS Sat-ellite Scenarios. The simulation results are summarized in Table I; the four signals with positive SINR are shown; all other signals have negative SINR.

Figures 6 and 7 show the contour plots of the two beams (within the 19 generated by the DBF) that provide the highest SINR for the signals transmitted by vessels 4672 and 33615.

The deep null of the radiation pattern over the horizon line is clearly shown. Note that signals can be “detected” not only by the main lobe but also by a sidelobe (see Figure 7).

SINR statistics for the static DBF solution have been com-puted along a 90-minute satellite track. The simulation re-sults are reported in Figure 8; 6.7 dB is the SINR exceeded with a probability of 0.5.

To complete static DBF analysis a “detection test” has been performed. A simplified detection criterion was select-ed that foresees a 7-dB SINR threshold to be exceeded for decoding a signal. Figure 9 reports the map of “detected” (in green) and “undetected” (in red) vessels for the whole satel-lite orbit, using the 19-beam scanning strategy.

Note that many messages transmitted by the same ves-sel reach the satellite-based receiver during the satellite pas-

Figure 8. SINR statistics computed along a 90-minute sample satellite track (SINR mean value is 6.7 [decibels]).

Figure 9. Map of detected (in green) and undetected (in red) vessels along a 90-minute sample satellite track.

Figure 10. Map of detected (in green) and undetected (in red) vessels over Europe.

10 IEEEA&ESYSTEMSMAGAZINE JUNE2014

Digita l Beamforming Appl ied to AIS Receiver

sage (due to the frequent retransmission); a single vessel is considered detected if at least one of these signals reaches a 7-dB SINR.

In Figure 10, the zooming of detected signals is reported for the passage over Europe (the Mediterranean Sea or the Northern Sea are some of the most critical regions, where a great number of interfering signals are present).

Good SINR performance is reached in almost all of the satellite FOV, excluding the border of the satellite coverage, where the signals are filtered by the deep null of the ele-ments radiation pattern. It has been shown that the big prob-lem present in high-density ship areas (i.e., the high number of overlapped signals received from directions close to the horizon) can be solved by the “optimized” array solution. On the contrary, when the number of transmitting vessels is very low, in principle, it would be possible to receive also some of the signals coming for directions with high eleva-tion angles (as depicted in the left side of Figure 9 for the satellite passage over the Pacific Ocean), but due to the ra-diation pattern of the elements of the optimized array, these signals are filtered and cannot be decoded.

Note that good SINR values are obtained not only for vessels located close to the satellite nadir, but also a large fraction of signals are detected when both the number of re-ceived signals is large (as over the Mediterranean Sea) and when the number of these signals is low (as over the Pacific Ocean).

CMA SIMULATION RESULTS

The CMA adaptive DBF technique introduced at the end of DBF Techniques has been applied using the antenna array (5 × 5) described in Definition of the Antenna Array. These

simulations have been performed to make a comparison be-tween the performance of the static solution and the adap-tive one.

Note that the computational complexity of the CMA adaptive solution is higher than the one of the static, because CMA works in an iterative way to find the optimized DBF weights. In this framework, the adaptive technique simula-tion time slot scenarios have been reduced with respect to the static ones to avoid a very high computational load. In particular, simulations have been performed over 2 minutes of satellite propagation, using a 1-minute satellite passage over the Mediterranean Sea and a 1-minute passage over the Pacific Ocean. The former is a worst case, in which a lot of interfering signals are present within every time slot, the lat-ter is a best case, in which few interfering signals are present.

Figure 11. Sample test time slot for the CMA algorithm.

Figure 12. Array pattern after CMA computed optimal weights.

Figure 13. Array pattern with no weights adaptation (static solution).

JUNE2014 IEEEA&ESYSTEMSMAGAZINE 11

Maggio et al .

The SINR performance results are then compared with the one obtained using a static solution (within the same 2-min-ute satellite propagation time).

Note that this approach is absolutely sufficient to pro-vide a comparative analysis between the static and adaptive DBF solution (about 4,500 time slot scenarios are used for comparison), saving much computational time (the simula-tion of CMA DBF over the 1-minute satellite propagation takes more than 72 hours; hence, a simulation of a full orbit is prohibitive).

The CMA adaptive solution has been implemented us-ing the same number of synthetic beams of the static one, 19 (hence, the same number of beamformers).

To show how CMA works, a sample test time slot sce-nario has been selected, as shown in Figure 11: the satellite is crossing the Mediterranean Sea from Egypt towards Greece. Thirty-nine vessels are included in the selected scenario, 21 transmitting on Ch. 1 (in cyan) and 18 transmitting on Ch. 2 (in red).

An example showing how CMA works is given in the following, concerning the detection of vessel 12740, located at (Lon. 27.015°E, Lat. 38.827°N), pointed by a white right ar-row in Figure 12. The best SINR is obtained by receiving P1 polarization (45 degrees) of Beam 14.

The CMA test scenario array pattern simulation results are shown in Figure 12, to be compared with the nonadap-tive (static) case, Figure 13. From Figure 13, it can be not-ed that with the static solution, the vessel is conveniently detected by a sidelobe of Beam 14, although it should be pointed out that the other closest vessels signals are partially attenuated by polarization mismatch too. In addition, note that the array can exhibit an even higher gain performance towards directions where no vessel is located (e.g., the lobes located over the Sahara and the central Mediterranean area). Figure 12 depicts how CMA distorts the antenna pattern, modifying the array weights, to increase SINR; in this par-ticular case, a 14-dB SINR increase is obtained by CMA (with

respect to the static solution), and nulling of high P1 signals close to that of vessel 12740 can be noticed.

Note that the CMA adaptive technique should work with snapshots sampled in a stationary environment, i.e., the sce-nario is assumed as not changing during the selected time window; hence, all the signals transmitted within the time slot are surely present.

Because of the different delays experienced by AIS sig-nals, only the central part of the snapshots collected by the array antenna can be used for adaptive techniques process-ing.

In particular, the minimum delay, experienced by a sig-nal coming from the satellite zenith, is about 2 ms, while the maximum delay, experienced by a signal coming from the horizon, is about 10.27 ms (see Figure 5). The receiver sam-pling frequency is 96 kHz; hence, the “stationary snapshots” start from the sample ns = 0.01027 s * 96,000 Hz = 986. The AIS standard [1] foresees a transmission-ending guard time of about 2.7 ms; hence, the “stationary snapshots” end at about 26.024 ms (the transmission-ending guard time 24.024 ms plus the minimum delay of 2 ms). The “stationary snap-shots” end at sample ne = 0.026024 s * 96,000 Hz = 2498.

In the implementation of the CMA adaptive technique, a further margin has been used, employing signal “stationary snapshots” that start from sample 1,000 and end at sample 2,399 (within the single time slot, each processed signal has 1,400 samples).

CMA performance has been evaluated in terms of SINR, and a comparison has been carried out with the ones ob-tained using a static solution. The results are depicted in Fig-ures 14 and 15, for the worst-case scenarios (1-minute pas-sage over Europe) and the best-case ones (1-minute passage over the Pacific Ocean), respectively.

It is very interesting to observe how CMA works: this adaptive solution is able to increase signals SINR (with re-spect to a static solution), when SINR is equal or slightly

Figure 14. CMA adaptive DBF versus static SINR performance analysis for the worst-case scenarios (1-minute passage over Europe).

Figure 15. CMA adaptive DBF versus static SINR performance analysis for the best-case scenarios (1-minute passage over the Pacific Ocean).

12 IEEEA&ESYSTEMSMAGAZINE JUNE2014

Digita l Beamforming Appl ied to AIS Receiver

higher than zero (these results agree with the discussion on CMA reported in DBF Techniques). In this case, an increase of 5 to 15 dB can be noticed. On the other hand, when the SINR is lower than zero, the CMA performs worse than the static solution; because the CMA cannot find a signal to con-verge and none has a sufficient SINR.

In this context, CMA could be useful when the AIS re-ceiver does not have very good performance (i.e., the SINR detection threshold is high); in this case, a CMA DBF solu-tion would increase the number of detected messages of 5%–10%.

CONCLUSION

Different DBF solutions have been identified and analyzed with the aim to improve the performance of a satellite-based AIS receiver.

The proposed solutions are suitable for medium satellite platforms only in a long-term scenario because the antenna is a large planar array (5 × 5). Both static and adaptive DBF techniques have been simulated to identify performance in terms of SINR increase.

The static DBF approach has been proven effective, pro-vided the antenna element is carefully designed, to strongly mitigate the interference from the horizon. As a matter of fact, this is one of the main problems, particularly in the case of satellite passages over crowded areas (in terms of transmitting vessels). As far as the adaptive techniques are concerned, good results were obtained for CMA (better with respect to the static approach).

The DBF performance is also related to the number of synthetic beams to be created, hence, the number of beam-formers to be implemented. It has been shown that 19 is a reasonable number to obtain very good results, but this re-quires a high computational complexity.

Digital signal processing technologies are not yet ma-ture for onboard satellite applications; hence, from a tech-noeconomic point of view, it would be better to multiplex onboard the RF received signals and retransmit them to ground, where beamforming could be carried out. Never-theless, even if in a near-future scenario, the onboard digi-tal signal processing could become affordable, the compu-tational effort required by CMA would be very high, and an onboard implementation seems to be, in any case, un-feasible.

It can be concluded that DBF techniques can improve the receiver performance, but the specific algorithm choice should be also based on the analysis of the system complex-ity increase, for instance, the need for a large antenna and assessing the stowing, deploying, and high computational complexity of an adaptive DBF technique.

ACKNOWLEDGMENTS

This work has been performed during ESA “AIS End-to-End Testbed” project, whose prime contractor is Compagnia Generale per lo Spazio (OHB-CGS). The authors thank Nader Alagha and Emilano Re from ESA for their valuable reviews and suggestions to improve the quality of the work; moreover, ESA provided the AIS scenario simulations that have been used to design and test the DBF solutions. The view expressed herein can in no way be taken to reflect the official opinion of the European Space Agency. The authors would like also to thank Veronica De Perini and Andrea Di Cintio from OHB-CGS for their guidance in understanding the AIS system and for their help in the test of DBF proposed solutions.

REFERENCES

[1] International Telecommunication Union. Technical characteris-tics for an automatic identification system using time-division multiple access in the VHF maritime mobile band, recommen-dation ITU-R M.1371-4, April 2010, http://www.itu.int/dms_pubrec/itu-r/rec/m/R-REC-M.1371-2-200603-S!!PDF-E.pdf.

[2] Dahl, O. F. H. Space-based AIS receiver for maritime traffic monitoring using interference cancellation. M.S. thesis, Nor-wegian University of Science and Technology, Trondheim, Nor-way, 2006.

[3] Godara, L. C. Application of antenna arrays to mobile com-munications, part II: beam-forming and direction-of-arrival considerations. Proceedings of the IEEE, Vol. 85, 8 (Aug. 1997), 1195–1245.

[4] Arapoglou, P. D., Liolis, K., Bertinelli, M., Panagopoulos, A., Cottis, P., and De Gaudenzi, R. MIMO over satellite: a review. IEEE Communication Surveys & Tutorials, Vol. 13, 11 (2011), 27–51.

[5] Re, E., Boissinot, V., Ginesi, A., and Tobehn, C. A simple high precision method for extrapolating Sat-AIS system perfor-mance. In Proceedings of the 6th Advanced Satellite Multimedia

Systems Conference (ASMS) and 12th Signal Processing for Space

Communications Workshop (SPSC), Baiona, Spain, Sep. 5–7, 2012, 266–272.

[6] Litva, J., and Kwok-Yeung Lo, T. Digital Beamforming in Wireless

Communications. Norwood, MA: Artech House, 1996.[7] Li, J., and Stoica, P. Robust Adaptive Beamforming. New York:

Wiley-Interscience, 2006.[8] Capon, J. High resolution frequency-wavenumber spectrum

analysis. Proceedings of the IEEE, Vol. 57, 8 (Aug. 1969), 1408–1418.

[9] Stoica, P., Wang, Z., and Li, J. Robust Capon beamforming. IEEE

Signal Processing Letters, Vol. 10, 6 (June 2003), 172–175.[10] Stoica, P., and Nehorai, A. MUSIC, maximum likelihood, and

Cramer-Rao bound. IEEE Transactions on Acoustics, Speech, and

Signal Processing, Vol. 37, 5 (May 1989), 720–740.