AIRBORNE RADAR AND SHIPBORNE SONAR: RECENT ADVANCES AND COMPARED SOLUTIONS Yves DOISY THALES Underwater Systems Sophia Antipolis, France Yves.doisy@fr.thalesgroup.com François LE CHEVALIER THALES Aerospace Division France Abstract The detection of slowly moving target in a strong backscattering background (clutter for radar, reverberation for sonar) is a challenge to both Airborne Radars and Surface Ship or Submarine Sonars communities. In this paper, after outlining the similarities and differences between the above and under-water domains, recent advances in the areas of waveforms design and beam-forming will be described. In a first part, more efficient wide-band Doppler waveforms for sonar will be introduced together with their advantages in terms of detection, and wideband coherent waveforms for radar moving targets detection will be presented. The second part will focus on the implementation of adaptive beam-forming for sonar and coloured transmission space time beam-forming for radar. Results on reverberation reduction obtained on a towed array sonar in CW mode will be discussed, and perspectives for coloured transmission radar systems will be outlined. Radar - Sonar - adaptive processing - beamforming - Space- Time adaptive processing - clutter - reverberation - antenna - Doppler - waveform Topics : Waveform Design, Beamforming and Signal Processing I. INTRODUCTION The detection of slowly moving target in a strong backscattering background (clutter for radar, reverberation for sonar) is a challenge to both Airborne Radars and Surface Ship or Submarine Sonars communities. In this paper, after outlining the similarities and differences between the above and underwater domains, recent advances in the areas of waveforms design and adaptive beam-forming will be described. While the sonar has been implementing digital adaptive beam-forming since more than a decade, radar is using coherent pulse trains Doppler waveforms which have only recently be transposed to the sonar domain. In a first part, more efficient wide-band Doppler waveforms for sonar will be introduced together with their advantages in terms of detection, and wideband coherent waveforms for radar moving targets detection will be presented. The second part will focus on the implementation of adaptive beam-forming for sonar and coloured transmission space time beam-forming for radar. Constraints in terms of stationarity and estimation duration for the covariance matrix will be outlined. These constraints are addressed by reducing the number of degrees of freedom, usually using beam space adaptive beamforming. Results on reverberation reduction obtained on a towed array sonar in CW mode will be presented, and perspectives for coloured transmission radar systems will be outlined. II. WAVEFORM DESIGN A. Wide-band Doppler waveforms for Sonar Mono-static active sonar are generally operated with a pulse repetition rate (or ping rate) equal or larger than the round trip travel of sound to the maximum range. They are unambiguous in range because otherwise, the reverberation induced by the latest transmitted signal would mask the longer ranges. An exception to this rule is Continuous Active Sonar, which will not be discussed here. For active sonars operated from a moving platform such as a surface ship, waveform design is driven by the coherence time which sets a limit on the duration over which the signal can be processed coherently at reception without losses in the replica correlation process. The main effects limiting the coherence time are usually platform motions induced by the surface of the sea, which produce accelerations affecting the signal both at transmission and reception. This duration depends on sea state, frequency of operation, type of sonar (Hull Mounted Sonars are more affected than Variable Depth Sonars) and to a certain extent on pulse type. It ranges typically from 1s to 10 s. On an other hand, the detection ranges in active surveillance are typically between 10 km and 100 km, which correspond to a round trip travel time ranging from 13s to 130s. Unlike the Radar case, this travel time is larger than the coherence time, and prevents the coherent exploitation of successive pings. When considering performance in reverberation limited environment, sonar waveforms can be sorted out according to the shape of their spectrum as discussed quantitatively in [3], and are summarized below. A.1 Flat spectrum waveforms: For waveforms with a wideband flat spectrum, detection performance is governed by bandwidth (or range resolution) and beam width, and is almost independent of target speed, because the transmitted bandwidth is in general much larger than the target Doppler shift. For these types of waveforms, target speed cannot be exploited to reject reverberation. Conventional widely used Hyperbolic Modulation Frequency waveform belong to this category, as well as Pseudo Random Noise, Binary Phase Shift Keying.
A.2 Narrow band waveforms: This category refers to Continuous Waveform pulses, consisting of a single frequency transmission of duration T, such that the width of its spectrum ( 1/T) is small compared to the Doppler shift of targets. Detection performance strongly depends on the (absolute) radial speed of the target, on the beam pattern of the array and on the own ship speed. For very slow targets ( Zone A on Figure A.1), the reverberation through the main lobe of the beam-pattern is at the same frequency as the target, and since the range resolution is very poor, the performance is poor. For target with intermediate radial speed ( Zone B ) the reverberation at the target frequency arises from an angular sector in the side lobes of the beam-pattern, and the signal to reverberation ratio is improved by the side lobe rejection (situation called endoclutter in radar). For very fast targets ( Zone C ), there is no reverberation in the target Doppler channel and the performance is very good ( exoclutter situation in radar applications). Most situations of operational interest correspond to the Zone B case (which is the case represented on Figure A.1) and despite the rejection of reverberation by the side lobes, the lack of bandwidth does not allow to bring significant detection advantage over HFM or wideband case. the motivation for adaptive beam forming discussed at the next section. In the case of a linear array and plane geometry (low grazing angles) the reverberation power at the output of conventional beam forming plus replica correlation is expressed as a two dimensional integral over all directions of the beam pattern times the frequency integral of the cross spectrum between the replica and the reverberation in the given direction [3]. The integral associated to the reverberation power is represented graphically on Figure 1 for case of a narrow band Weighted Continuous Frequency pulse of duration T, in the frequency u plane, where u=cos (bearing). On a platform moving at constant speed, the reverberation spectrum is dependant of the direction, and is centered about a straight line in this plane (in grey on Figure 1). The three zones A,B,C are indicated on the Figure. In the case of PTFM, refer to Figure 2. A.3 Comb spectrum waveforms: Comb spectrum waveforms contain several (P) narrow lines in their spectrum. They can be obtained by transmitting a periodic wideband signal (train of FM designated as PTFM, or other wideband pulses), or by superimposing CW signals at different frequencies, or by using a sine modulated phase. They combine doppler rejection of reverberation for moving target, and high bandwidth. Detection performances follow the same rules as narrowband CW waveforms, with an improvement of the signal to reverberation ratio. The issue of optimal spectral line spacing is related to the Doppler spread of reverberation induced by own ship speed V, as discussed in [3]. For a pulse train waveform consisting of N identical transmitted pulses over the duration T, each pulse BT having the bandwidth B, the spectrum contains P = lines N 4V spaced by the frequency interval N/T. We note: f = f0 the c Doppler spread of reverberation. As long as N/T > f, there is only one sector of space contributing to the power of reverberation through the side lobes. Since the range extension c of the reverberation cell is equal to: N, the number of sub 2 B pulses N should be reduced as much as possible. However, when N reaches the value N=T f, a second sector of space contributes to the reverberation, so that the benefit in range resolution of reducing N is compensated by an increase of the angular contribution. The number of contributing sector is S=ceil(T f/n), where ceil(x) is the nearest integer larger than x (Figure 2 corresponds to the situation where S=3). Therefore, the waveform should be designed according to N T f. If this condition id fulfilled, the detection gain of PTFM waveforms over: B CW is G cw = 10log( ). f HFM, in Zone B, with the same bandwidth is G ptfm = 10log(2θ 3) NS cste where NS is the beampattern sidelobe level (negative number) and 2θ 3 the 3dB beamwidth. Any effort to improve the side lobe level results in a direct gain on the performance. This is Figure 1. Reverberation power-cw Figure 2. : Reverberation power-ptfm
A.4 Experimental results: Figure 3 shows Doppler versus bearing plots of the energy at the output of beam forming followed by replica correlation, at the same range, for three waveforms of same duration T=4s. The transmission was omni directional, and the reception has been performed on a linear towed array of length close to 20m. The two upper plots correspond to CW at respectively 1088 Hz and 1919Hz, and the lower plot corresponds to a PTFM o28 pulses with a bandwidth of 700Hz. The three waveforms have been transmitted in the same ping. All the displays are signal normalized, so that the levels can be compared. On the CW displays, strong reverberation leaking through the sidelobes is clearly visible, while it is hardly visible on the PTFM display. related by: D a x V a = λ x c / 4. That relation means that many ambiguities, either in range or speed (or both), have to be dealt with, which in turn implies the transmission of successive pulse trains with different repetition frequencies, requiring more time to be spent on target for ambiguity and blind speeds removal. An alternative solution is obtained by increasing the range resolution, (or the instantaneous bandwidth) so that the moving target range variation (rangewalk) during the pulse train becomes non-negligible compared with the range resolution - which is equivalent to stating that the Doppler effect is varying across the whole bandwidth (compared with the Doppler resolution), and can not be considered as a mere frequency shift any more - : such radars may use bursts with low Pulse Repetition Frequency (no range ambiguities) wideband pulses such that the rangewalk phenomena during the whole burst is significant enough to remove the velocity ambiguity. It then becomes possible to detect the target and measure range and speed with only one coherent pulse burst. The condition is written, if N t is the number of pulses in the burst, T r the repetition period, V a the standard ambiguity speed [V a = l/(2t r ) ], F the instantaneous bandwidth, and δr the range resolution [δr = c/(2 F)]: λ NtVa Tr >> δr Nt >> δr 2 F0 Nt >> (typically 10) F For example, a burst of 60 pulses at 1 khz repetition frequency with 500 MHz bandwidth would be a possible candidate for non-ambiguous MTI detection at X band. pulses successive pulses / fixed target mobile target time (range) Figure 4. Wideband signal processing Figure 3. : Doppler bearing plots. The quantitative analysis performed on the data in [3] showed a reduction od 12 to 15 dbs in PTFM as compared to the two CW pulses. Wideband Doppler sensitive pulses in the form of train of wide band pulses can be considered as similar to Radar pulse train, with the difference that the successive pulses are adjacent, constituting one ping - and so the Doppler analysis is performed within the ping instead of ping to ping. Due to the high relative bandwidth in Sonar, the condition N>>f 0 /B is generally met, so that these waveforms are non ambiguous in Doppler (see following section). B. Wideband GMTI An essential limitation for standard radars comes from pulsed radar range-doppler ambiguity relation, which states that the ambiguous speed V a and the ambiguous range D a are Figure 5. Ambiguity functions (a: narrowband, 1/10000 bandwidth; b: wideband, 1/10 bandwidth) The coherent signal processing of such radars (whose range resolution is in the order of a few wavelengths, typically less than 10) involves (Figure 4), for each velocity hypothesis, a coherent summation of the received echoes (Fourier transform), after rangewalk compensation:
th th xr, t : received signal from p pulse, at t time sample Hypothesis : range t δr, speed V N 1 F0 2V 2π j p F c r TtδR, V = x VT e r, r, Γ t p p= 0 R δ with Γ( u) = nearest integer from u This processing leads to an ambiguity function (Figure 5) which does not exhibit the periodic ambiguities in Doppler, and provides the following advantages: Simultaneous detection of fixed and moving targets (SAR+GMTI), with the high resolution low PRF pulse train appropriate for SAR imaging ; High resolution range-doppler Classification: Figure 6 illustrates this possibility with the image of a hovering helicopter, with 50 cm resolution, where the main rotor and the tail rotor are clearly visible at ranges 5 m and 13 m (signals obtained by Electromagnetic modelling of a Puma helicopter); ECCM properties (spread spectrum signals, requiring specific interception for ELINT or ESM, and specific devices for simulation of the wideband Doppler compression effect) ; velocity (m / s) range (m) Figure 6. Helicopter HRR signature III. A. Sonar adaptive beamforming ADAPTIVE BEAMFORMING We discuss in this section the application of adaptive beam forming to active sonar in Doppler mode, considering either CW waveforms or Wideband Doppler waveforms. We restrict to the case of a linear receiving array aligned with its own speed. As mentioned in section A, in this mode the detection performances are related to side lobe rejection, and significant gains are expected from adaptive beamforming. The main challenge is to obtain enough independent snapshots for the estimation of the covariance matrix. Reverberation in Doppler mode has a spatial structure which depends on the frequency, and also on time as the spatial structure of the environment is not stationary. For an array of length L, the interference associated to reverberation in each Doppler channel can be considered as arising from an angular T f L sector of J = ceil( ) jammers, whose central direction N VT depends on the Doppler channel. The number of degrees of freedom required to cancel this interference is of an order 2 to 3J. The number of independent snapshots available depends on the mode and array directivity: in CW mode, each slice time of duration T (pulse length) will provide one independent snapshot and the number of independent available snapshots per jammer and per pulse duration is: VT S cw = 1 = J L in wideband Doppler mode (PTFM), each slice time of duration will potentially provide BT/N snapshots, and the number of independent available snapshots per jammer and per pulse duration is: BT BTλ S ptfm = 0 NJ 4L where λ 0 is the center wavelength. Taking into account the typical pulse durations, the covariance matrix estimation duration can hardly be extended further than a few pulses, and for adaptive beamforming to be effective, Scw and Sptfm should be larger than one. Wideband Doppler waveforms allow to increase this number by an order of magnitude with respect to CW. As the number of sensors on the array are several hundreds, the adaptive beamforming must be implemented in beamspace domain in order to reduce drastically the number of degrees of freedoms. It can be implemented either after conventional beamforming, by selecting the appropriate beams steered on the reverberation sectors as noise references (beam space beamforming), or at the output of conventional beams formed on selected subarrays from the full array (subarray beamforming). Figure 7 shows an example of Doppler - bearing plot of the energy at output of the conventional (left part of the figure) and adaptive (right part of the figure) processings for a sin2 weighted CW pulse. Adaptive subarray beamforming was implemented at the output of replica correlation on a triplet array. The array consists of triplets of hydrophones fitted within a hose towed by the ship. The acoustic length of the array is close to 20m, for an external diameter of 80 to 90mm. Each triplet consists of equally spaced hydrophones fitted on a circle in the section of the hose. The triplets are equally spaced over the length of the array. The sensor arrangement can be seen as a set of three identical and parallel line arrays fitted within the hose. In each Doppler channel, the three identical linear subarrays were selected from the main array. Conventional beams were formed on each subarray, and each identical triplet of primary beams were processed adaptively. The covariance matrix was estimated over several successive independent snapshots in time. The parameter VT/L was equal to 0.52. The reverberation leaking through the sidelobes is clearly visible in Zone B on conventional processing. It has been reduced almost to the level of the background noise through adaptive processing. Measured gain averaged to 18dB on the signal to reverberation ratio. Figure 7. : Doppler bearing plot. Left:conventional/Right:adaptive.
From a radar perspective, this processing can be described as an adaptive angular processing of monopulse channels - which is a simplified version of Space-Time Adaptive Processing (STAP), with adaptivity restricted to the spatial dimension, thus providing an efficient compromise between complexity and performance. B. Coloured transmission for surveillance radar 1) Digital Beam Forming () Standard digital beamforming is a procedure where wide angular sector instantaneous coverage is obtained with a wide beam illumination on transmit (transmission through one subarray), and directive beams are formed on receive through coherent summations of signals received on different subarrays, in parallel for each direction. Digital beamforming generally does not essentially change the power budget, compared to standard focused exploration, since the lower gain on transmit (due to wider illumination) is traded against a longer integration time (made possible by the simultaneous observation of different directions). Digital beamforming may provide specific advantages, such as a better visibility of short events (eg rcs flashes), and a higher Doppler resolution especially useful for identification purposes, or for detection of slow targets. However, for airborne applications, a severe limitation arises from the clutter spreading in Doppler, due to the wider beam on transmit (which is, anyway, difficult to obtain with active antennas): this leads to a poor minimum detectable velocity, and to a poor clutter rejection, since only half the dbs are obtained, compared to focused beam illumination. This is the main motivation for turning to coloured transmission. 2) Principles of coloured transmission The principle of coloured transmission consists in simultaneously transmitting different waveforms in the different directions, thus achieving space-time coding (Figure 8). On this figure, the coding is supposed to be a succession of sub-pulses, coded in phase or frequency, but any type of code can be used - for instance, transmitting different frequencies through the different sub-arrays could also be a possibility. The directivity on transmit is then recovered by signal processing on receive. Subarray #1 Subarray #2 Subarray #N w 11 w 12 w 1M w 21 w 22 w 2M w N1 w N2 w NM T r Figure 8. Coloured transmission w 11 w 12 w 1M w 21 w 22 w 2M w N1 w N2 w NM For signal processing on receive, the transmitted waveforms should be orthogonal, so that they can be separated from one another, on each receiving channel. It should be emphasized that the transmitted waveforms are still periodic, since that is a necessary condition for an efficient cancellation of long range clutter (e.g. mountains). The optimum processing then basically consists in the operations described on Figure 9 (essentially a coherent summation of the received samples, for each angle - Doppler - range hypothesis): Transverse filtering, for separating the signals received from the different transmitters; Digital Beam Forming on transmit (basically Fourier transform), coherently summing the transmitted signals, for each receiving antenna (i.e. each receiving channel); Digital Beam Forming on receive (again basically Fourier transform); P channels reception Matched filter T Sub-pulse For each transmitter, P receiving channels For each illuminated direction, P receiving channels For each illuminated direction, P directions on receive Delay 1 subpulse W Delay 1 subpulse Delay 1 subpulse x 1 x 2 x 3 x M (Separation of individual transmitters) y 1 y 2 y 3 y N-1 y N A (Beamforming on ) z 1 z 2 z 3 z N-1 z N T/R Imaging (P Channels Reception, N Channels Transmission) Figure 9. Optimum reception of coloured signals For increased performances in cluttered environments or adverse conditions, digital beamforming will preferably be performed with appropriate adaptive algorithms on transmit and/or receive. 3) Coloured transmission trade-offs and applications As a whole, these coloured transmission techniques can be described as providing angular instantaneous coverage (wide angular sector), traded against a larger instantaneous frequency bandwidth. Or they can be described, starting from the standard Digital Beam Forming as providing angular separation on transmit (hence better clutter rejection), traded against instantaneous frequency bandwidth For surveillance radars, these waveforms provide the solution to a classic dilemma: how to increase the Doppler resolution, needed for slow-targets detection and target classification, without widening the clutter spectrum? One possibility - actually very similar to the RIAS concept - is illustrated on Figure 10, where each sub-array transmits a different frequency carrier, coherently summed on receive as explained above. Depending on the specific system generation and reception constraints, it might be preferable to transmit different orthogonal codes (frequency modulations, phase codes, etc.), rather than different frequency carriers. It must be emphasized that in this case, each target is illuminated by the whole bandwidth, thus making it possible to use such modes for High Range Resolution analysis of the detected targets, or wideband MTI, as described in Part II (Paragraph B). Subarray #1 Subarray #2 Figure 10. Coloured transmission for surveillance 4) Interleaved scanning Another way to explore space is obtained by interleaved scanning, where successive pulses are sent is successive directions, thus interleaving different pulse trains - possibly with different frequencies, or different codes. An example is shown Figure 11, with 2 interleaved directions.
θ 1 distributed - which amounts to trading the Doppler ambiguities against increased blind zone at short ranges. θ 2 θ 1 θ 2 θ 1 θ 2 θ 1 θ 1 θ 2 θ 1 θ 2 θ 1 θ 1 θ 2 θ 1 θ 2 θ 1 Variant: orthogonal phase coding, rather than frequency coding Variant: adjacent pulses on transmit ( angle frequency coding) Figure 11. Interleaved scanning This scheme allows to trade a wider quasi-instantaneous coverage - and the possibility to implement adaptive angular processing, by coherently processing the signals received from the adjacent beams -, against a lower repetition frequency (and consequently more Doppler ambiguities) in each direction. It has no significant impact on the power budget: as for the previous coloured emission concepts, the loss in overall gain on transmit is balanced by a longer integration time on target. With that interleaved scanning concept, it becomes possible to implement any adaptive procedure on receive, with only one channel on receive, if the transmitted signal are identical (so that the samples received from the different direction can be coherently processed to extract angle information). Of course however, one has to take into account the fact that the samples are not taken simultaneously, so Doppler information has to be incorporated in the spatial filter. Such modes have been shown to be an effective way of implementing STAP (Space-time adaptive processing) with only one channel on receive: this is illustrated on Figure 12, with 3 interleaved directions. The result is of course an increased ambiguity in Doppler, and a loss of 5 db in signal to noise ratio (due to the apparent widening of the beam), but a performance comparable to the 3 channels standard technique for slow targets detection. Signal to noise ratio Target radial speed Figure 12. Interleaved scanning STAP 3 interleaved beams This mode, as shown on Figure 13, could provide efficient solutions for long range air-ground or air-air surveillance systems, preferably with adjacent pulses rather than evenly Figure 13. Interleaved scanning Surveillance CONCLUSION The few examples presented in this paper have shown that the mutual exchanges between radar and sonar are not over: sonar is now taking benefit of periodic waveforms similar in principle to radar pulse trains, while radar is progressing towards wideband detection of moving targets. Adaptive processing, initially implemented in passive sonar, has been extended to space-time processing for radar applications - which might in turn be considered for sonar applications. Various schemes of space-time coding on transmit ( coloured signals ) are now being considered for radar applications, and global intelligent sensor management will be required to take full advantage of the bandwidth and agility available on surveillance radar and sonar systems. REFERENCES [1] Guyvarch J.P., «Antenne Spatio-Temporelle à Codes de Phases Circulants. Colloque GRETSI 97, pp. 607-610, Grenoble, Septembre 1997. [2] Cox H., Lai H., Geometric Comb Waveforms for Reverberation Rejection. Proc. 29 th Asilomar Conf. Signals, Syst., Comput., Pacific Grove, CA, Oct. 29- Nov. 1 1995, pp.1185-1189. [3] Doisy Y., Deruaz L., van IJsselmuide S., Beerens S.P., Been R. : Reverberation Suppression Using Wide Band Doppler Sensitive Pulses. IEEE J. of Oceanic Engineering, Vol 33, n 4, Oct. 2008. [4] Calvary P. and Janer D., Spatio-Temporal Coding for Radar Array Processing. ICASSP 98, pp. 2509-2512, Seattle, 12-15 may 1998. [5] Le Chevalier F. «Principles or Radar and Sonar signal processing», Artech House, 2002. [6] Le Chevalier, F.; Savy, L.; «Coloured Transmission for Radar Active Antenna», International Conference on Radar Systems RADAR 2004, Toulouse, France, October 2004. [7] Le Chevalier, F.; «Smart beamforming and coloured signals for MIMO radars», Tutorial at the Third International Waveform Diversity & Design Conference, Pisa, Italy, June 2007. [8] Drabowitch, S. and Aubry, C. Pattern compression by space-time binary coding of an array antenna, AGARD CP 66, Advanced Radar Systems, 1969. [9] Dorey, J., Blanchard, Y., Christophe, F., Garnier, G.: Le projet RIAS, une approche nouvelle du radar de surveillance aérienne, L Onde Electrique, vol 64, N 4, 1978. [10] N. Levanon and E. Mozeson: Radar Signals, J. Wiley & Sons (Interscience Div.) New York, 2004. [11] Le Chevalier, F.; «Future concepts for electromagnetic detection», Aerospace and Electronic Systems Magazine, IEEE, Volume 14, Issue 10, Oct 1999. [12] Le Chevalier, F.; «Smart beamforming and wideband signals for ground surveillance», invited paper, International Radar Symposium, IRS 08, Wroclaw, mai 2008