I.

INTRODUCTION

Frequency Diverse MIMO Techniques for Radar

PIER FRANCESCO SAMMARTINO Joint Research CentreEuropean Commission Italy CHRISTOPHER J. BAKER The Ohio State University HUGH D. GRIFFITHS, Fellow, IEEE University College London It has been shown over several decades of radar research that the exploitation of diversity in a number of domains (space, frequency, time, polarization, and, recently, waveform) can provide increased agility, flexibility, robustness, and capabilities to the radar system. However this is often achieved either through efforts in system design, increased hardware complexity, or by employing additional resources. A conventional antenna array is considered with the intention of introducing, not major, but minor mismatches, in particular in the carrier frequencies and, eventually in the codes at the element level. The starting point of this analysis is the frequency diverse array (FDA), which has been demonstrated to generate a range-angle pattern. Through a reconsideration of the organization of the array, which we have termed the wavelength array (WA), a new pattern, orthogonal to that of the standard phased array, can be achieved. The bistatic combination of a WA and a receiver leads to the frequency diverse bistatic system (FDBS), which can be a significant application of this concept. In a second stage the analysis focuses on the effects of introducing waveform diversity in such a system. In particular, if the elements of an electronically steered array (ESA) simultaneously transmit a number of pseudonoise (PN) codes at slightly different carrier frequencies, the coherent summation of the codes gives rise to a waveform whose shape is a function of both angle and range. In fact this is the consequence of applying the multiple-input multiple-output (MIMO) technique to the FDA, which has the result of associating a waveform to each point range/angle of the space, with the possibility of recovering this information in receive after appropriate processing.

Manuscript received January 7, 2012; revised January 27 and June 13, 2011; released for publication March 13, 2012. IEEE Log No. T-AES/49/1/944348. Refereeing of this contribution was handled by M. Rangaswamy. Authors addresses: P. F. Sammartino, Joint Research Centre, European Commission, IPSC, Via E. Fermi 2749, TP 720, Ispra (VA), 21027, Italy; C. J. Baker, Department of Electrical and Computer Engineering, The Ohio State University, 752 Dreese Laboratory, 2015 Neil Ave., Columbus OH, 43210; H. D. Griffiths, Department of Electronic and Electrical Engineering, University College London, Roberts Building, Torrington Place, London WC1E 7JE, UK. c 2013 IEEE 0018-9251/13/$26.00

It is well known that the use of different frequencies in transmission can improve radar detection and the classification of targets, as it decorrelates clutter, provides a better estimate of a targets mean echo strength, and provides a sampling of the frequency response of the target [1]. In these examples the carrier frequencies are sufficiently separated such that the bandwidths of the signals do not overlap. In this way the electromagnetic characteristics of the signals are significantly mismatched so that it is possible to exploit the diverse information in order to improve the overall performance. In this work a different approach is considered by studying the effects in transmission and propagation when the signal is modulated on frequencies relatively close to one another. Thus, here it is not expected to have an increase of the performance due to the exploitation of frequency diversity since the frequency response is assumed to be the same for each transmitted antenna due to the negligible difference in the wavelengths. This concept has been explored in [2][9], and it is a brand new topic which is attracting interest from the technical community with applications in synthetic aperture radar (SAR) imaging [10, 11] and ground moving target indicator (GMTI) [12]. Theoretical results and experiments of the 1-way propagation pattern have been shown in [13], and they are consistent with one another. In its most simple form, one can envision an array antenna [14] where the different elements that comprise the array carry signals with differing frequencies. When the frequency shifts are negligible with respect to the carrier frequency, this constitutes the so-called frequency diverse array (FDA) [2], which is the starting point of the analysis in this paper. In this system the phases of the transmitted signals add constructively in certain regions of space, while destructively in others. This creates a range-dependent pattern whose amplitude and spatial distribution can be controlled by the frequency offsets and the number of elements of the array. Figures 1 and 2 are a first schematic outline of this concept. Whilst in the first part of this paper a development of the FDA concept is proposed, in the second part, the multiple input multiple output (MIMO) radar concept is combined with the FDA to illustrate how they can be used to create even further sensing capabilities. As is known, MIMO radars are receiving an increasing amount of attention in the literature as exemplified in [15], which comprises a collection of different developments on this technique. In reviewing this growing body of literature, it is clear that there are many and often very different but valid concepts that fall into the broad catchall that is coined by the term MIMO. In particular two different MIMO
JANUARY 2013 201

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

to be pointed out clearly that, as the waveforms are to be applied to radar systems, the requirement of orthogonality as intended in conventional communication systems, i.e., Z T wm (t)wk (t)dt = 0 (1)
0

Fig. 1. Sketch of FDA when each element transmits a pure tone at slightly different frequency.

systems have been designed as applied to radar. These are the waveform-diversity MIMO (WD-MIMO) and the space-diversity MIMO (SD-MIMO). Research on WD-MIMO radar systems has recently shown that it is possible to synthesize an arbitrary antenna array by starting from opportunely-spaced single elements by transmitting different waveforms at the element level and at the same time [15, 16]. In this case the spacing between the elements is relatively narrow, i.e., comparable with the wavelength. In its conventional implementation this technique requires each element to transmit its own waveform. Orthogonality of the transmitted waveforms is a requirement for allowing signal separation at the receiver. However it has

where wm (t) and wk (t) are two different waveforms limited between 0 and T and where is the conjugate operator, a necessary condition even if it is not sufficient. In fact other properties, such as mutual low cross-correlation, good Doppler tolerance, and good range resolution, are not only desirable but compulsory in order to ensure the correct operation of these systems to outperform, to some extent, the conventional electronically steered array (ESA). The WD-MIMO technique differs from the SD-MIMO radar system, which exploits the angular diversity to acquire independent measurements of the radar cross section (RCS) and increases, in a number of ways and under different points of view, the performance of the overall system [5, 17]. In the latter configuration this MIMO radar system is basically the same as a radar network where the transmitters and the receivers, which can be colocated in pairs or not, are generally widely separated from one another. It has been shown [17] that this MIMO system can outperform conventional, monostatic radars under different points of view even if the overall energy is kept constant and the complexity is increased reasonably. However the analysis in this paper does not take into account this kind of

Fig. 2. FDA spectrum occupation when band-limited signals are transmitted. 202 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1 JANUARY 2013

MIMO and focuses on the WD-MIMO, which is more suitable for combination with the FDA concept. In the rest of this paper, Section II provides an overview of the main characteristics of the FDA and of the WD-MIMO as developed so far, Section III describes the frequency diverse bistatic system (FDBS), which is a bistatic application of the FDA and allows a number of its drawbacks to be mitigated. In Section IV the fusion of the FDA and the MIMO

frequency. The antenna elements comprising the array are mutually spaced by the quantity 1 1 c 1 minfk g = N 1 = 2 k 2 2 f0 + (N 1)f (3)

i.e., the half-wavelength spacing is maintained albeit now as a function of the maximum transmit frequency. The expression of the pattern p(f , t, R0 , ) after coherent summation is as in (4), i.e.,

p(f , t, R0 , ) =

N 1 X k =0 N 1 X k =0

Rk 1 exp j 2 fk t Rk k Rk 1 exp j 2 (f0 + kf ) t Rk c

N 1 X k =0

= expfj0 g

1 fR0 d sin 2 fd sin exp j 2 kft k k k Rk c 0 c (4)

N 1 fR0 d sin fd sin expfj0 g X exp j 2 kft k k2 k R0 c 0 c

k =0

concept are presented. Discussion and conclusions are presented in Section V. II. BACKGROUND A. The FDA References [2] and [3] derive the pattern of an N -element antenna array where each single element transmits a sinusoid sk (t) sk (t) = expfj 2fk tg (2)

where Rk = R0 + kd sin , 0 = c=f0 R0 R0 0 = 2 f0 t = 2f0 t 0 c

(5)

and the approximation is due to the fact that Rk R0 . Although, in general terms, a closed form of the pattern in (4) cannot be written, a closed approximation exists when (N 1)f=c 1=0 , which is true as long as (N 1)f f0 . In this case the term k 2 fd sin =c in (4) becomes negligible, and the pattern can be written as

N 1 expfj0 g X fR0 d sin p(f , t, R0 , ) exp j 2k ft R0 c 0

k =0

d sin fR0 d sin sin N ft exp j c + (N 1) 0 c 0 = d sin R0 fR0 sin ft c 0

(6)

whose frequency fk differs from a starting frequency f0 by a quantity kf , i.e., fk = f0 + kf , with k = 0, : : : , N 1 (Fig. 1). For instance the number of array elements in the example in this paper is fixed at N = 9. Frequencies are equally spaced by f = 3 kHz starting from f0 = 8 GHz. It is then clear that the frequency offsets are a tiny fraction of the carrier

where

R c = 2fc t 0 c
N 1 1 X N 1 fk = f0 + f: N 2 k =0

(7)

and fc = (8)

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

203

Fig. 3. FDA. Snapshot of the one-way pattern at a time t = n1=f , n 4, minimum half-wavelength spacing between elements.

In Fig. 3 a numerical simulation of the beam pattern1 as in (4) is computed. Here range attenuation apart,2 the regularity of the pattern as a function of range and angle is shown clearly. In addition it has to be pointed out that this is a snapshot of the pattern in a precise instant of time, say t = n1=f , n 4, and it is important to remark that, whereas a CW is transmitted, the pattern propagates throughout all the ranges with time. This applies also to the other results in this paper, where CW signals are fed to the elements of the array. Equation (6) provides further details on the behavior of the pattern. It seems appropriate, fixing the angle , to remark on the following characteristics. 1) For a given instant of time, the pattern is periodic in range, and its peaks are spaced c=f from one another. For a given range the pattern is periodic in time, and its peaks are separated by 1=f from one another.
this point it is worth noting that, if the frequency shift at the element level is not present, (6) reduces to a conventional phased array pointing to 0 deg. The introduction of the frequency shifts at the element level has the effect of modifying the amplitude of the pattern, thus making it dependent, not only on the angle, but also ontime (in a carrierwave (CW) configuration) and range, i.e., it is an extension of the concept of conventional ESAs or phased arrays. 2 Whereas plotting short ranges would require a too high dynamic range and in turn affect the readability of some figures, ranges in this paper start from 50+ Km. Nonetheless the validity of the presented concepts still holds at shorter ranges, even if not reported herein.
1 At

2) The beamwidth at 3 dB is equal to c=Nf . So the more elements comprising the array, the sharper the peak and the greater the peak-to-sidelobe ratio, and the time of illumination (width at 3 dB) of a target is 1=Nf . 3) The arguments in round brackets of the sine functions can be rearranged as R 1 d f t 0 sin : c f 0 The latter part of this expression explains clearly why the pattern is a function of angle as well. At the same time it shows that the distance between the peaks of the pattern at = 0 and = =2 is (c=f )(d=0 ) in range and (1=f )(d=0 ) in time. Therefore the S-shape of the pattern is here shown clearly to be a function of the ratio d=0 and of the frequency shift of the antenna elements f . 4) Finally, if the frequency shift across the elements of the array is negative, the S-shaped amplitude of the pattern in Fig. 3 flips as well, as if positive shifts are applied starting from the last towards the first element of the array. This can also be derived by exchanging f with f and/or d with d in (6). Figure 4 clearly shows point 3 since the curvature in the S-shape of the pattern is mitigated by reducing the distance between the array elements. In this example the spacing has been taken to be equal to N 1 =8. Nevertheless the distance between the peak
JANUARY 2013

204

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

Fig. 4. FDA. Snapshot of the one way pattern at a time t = n1=f , n 4, octave-wavelength spacing between elements.

at 0 and those at 90 is still in the order of tens of kilometers, with a spacing between the elements which is far less that of a conventional antenna array. Because of the time-dependency (see (6)), the pattern slides in time, i.e., it illuminates all the ranges and angles of the space. Nonetheless this is performed in a way which is not the same as in conventional ESAs or phased arrays. All the ranges and angles are actually illuminated (3 dB beam) for a time 1=Nf and with periodicity 1=f . In the rest of the paper, we further develop the concept of applying a frequency shift at the element level, which shows the effects on the pattern when the spacing between the elements is changed or when different waveforms are employed. B. The WD-MIMO In a conventional configuration each element of a WD-MIMO radar system transmits a different waveform orthogonal to those of the others. Although the spacing between elements is not required to be uniform, in this paper we assume, in line with the previous section, to have N = 9 omnidirectional elements spaced at =2 in receive and transmit. This may not be the best configuration for a WD-MIMO system, which allows and consequently exploits a reduction of redundant paths [16], but it is a convenient geometry for the purpose of this paper.

It can be demonstrated [16] that, being possible to distinguish the phases related to each single path tx-rx, e.g., using orthogonal codes, this array is equivalent to a linear array twice as long, where one antenna only transmits and 2N 1 receive, with a triangular weighting through the virtual receiving antennas. The benefit of this technique is in moving the beamforming to receive only after collecting information from all the angles at the same time. However this is achievable at the cost of illuminating the entire space with an omnidirectional angle-dependent pattern. In fact if wk (t) is the generic transmitted band-limited pseudonoise (PN) code, the overall transmitted waveform w(t), at a distance R0 (in the far field) and at an angle from the reference point, can be expressed as N X Rk R0 1 w t exp j 2 f0 t w (t , ) = Rk k c
k =1

N d sin expfj1 g X wk (t ) exp j 2k R0

k =1

where = R0 =c 1 = 2f0 (t ) and the rest of the symbols are as in (6).

(9) (10)

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

205

Fig. 5. Example of three orthogonal and band-limited waveforms.

It seems worth highlighting that as wk (t) is a pseudorandom sequence, also w(t, ) is a PN code with a total power which is the sum of the power of the single codes, as wk (t) is orthogonal to wm (t), 8k 6= m. Figure 5 shows the amplitude of some of the transmitted waveforms (PN codes) at element level. When they are summed up in propagation, the phase shift due to kd sin = (see (9)) allows a waveform to be generated which has the peculiarity of changing from angle to angle. Figure 6 shows this concept (without taking into account the range attenuation, which is only an amplitude factor) by using the codes in Fig. 5. In particular subplots on the same row show the waveforms generated pointing at 0 and 45 , respectively. It is here shown clearly that the shapes of the amplitudes of the overall propagating waveform are the same when the angle is the same, even if the range changes (here 50 and 100 km). As seen in Section IV, range-dependency of the propagated waveform can be introduced by merging the FDA with the MIMO concept. III. THE FREQUENCY DIVERSE BISTATIC SYSTEM The analysis shown in this section starts from the FDA concept described in Section II-A. It is here shown that, through a different spacing of the elements of the array, it is possible to achieve a much more regular pattern which is orthogonal to that of an ESA in the range-angle plot.
206

A. The WA In order to avoid, or at least to lessen, the dependency of the range gain from angle gain, the elements of the array are arranged in a different configuration. This leads to what has been termed the wavelength array (WA). The main idea is to separate the elements of the array from a reference point by a distance related to the transmitted wavelengths. In other words, here it is assumed that the distances dk are proportional to k . It is important to highlight that the reference point is not active in transmission. This is shown schematically in Fig. 7. As can be seen here, the distance of each element of the array is evaluated with respect to a neutral point which is not active in transmission. As in the FDA case, each element keeps transmitting on its own carrier frequency and it is distant from the reference point of a quantity which is linearly dependent on the wavelength. In particular, assuming dk = Lk , the pattern p(f , t, R0 , ) can be written as
p(f , t, R0 , ) =
N 1 X 1 k =0 N 1 X 1 k =0

R exp j 2 fk t k Rk k

R + Lk sin exp j 2 fk t 0 Rk k exp j 2L sin j 2fk

X 1
N 1 k =0

Rk

R0 c

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

JANUARY 2013

Fig. 6. Propagated waveform at 0 and 45 at ranges R = 50 km and R = 100 km, WD-MIMO.

to k1 k = k1 (f=fk ) 14 109 m! As a consequence L should be big enough to ensure that the spacing between any two adjacent antennas dk = L(k1 k ) (13)

Fig. 7. WA concept; reference point is not active in transmission.

N 1 1 X exp j 2L sin j 2fk R0 k =0 N 1 k=0

R0 c

R0 c

expfj 2L sin g X exp j 2fk R0

R sin Nf t 0 c expfj (2 + c )g = R0 R sin f t 0 c

(11)

can be feasible in hardware manufacturing. The pattern derived here is shown in Fig. 8. This is a most unusual beam pattern which exhibits constant peak gain with angle at a particular range, i.e., the beam pattern in the space is orthogonal to that produced by conventional fixed phased arrays. In this case it is worth highlighting clearly the following: 1) The angle-dependency is found in the phase term only, whilst the amplitude depends on time, range, and frequency shift only and no longer on angle. 2) As in the previous case, as the amplitude is a function of time, this pattern slides in time across the ranges and at the propagation velocity of the carrier. B. The FDBS The concept of an FDBS arises when implementing the WA with a factor L, which guarantees the physical feasibility of the array, according to the considerations in the previous section. For this purpose let us assume s as the minimum manufacturable spacing between the antenna elements of a WA when using carrier frequencies spanning from f0 to fN 1 . According to the definition of the spacing between two generic consecutive
207

where 2 = 2L sin and where L is a necessary additional parameter to achieve practically-feasible distances between elements: in fact without L, the effective distance between two consecutive antennas would be equal (12)

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

Fig. 8. WA, snapshot of one-way pattern as seen from reference point with range attenuation and at time t = n1=f , n 4.

antenna elements, as in (13), the following holds dmin = minfdk1 dk g = minfLk1 Lk g

k k

= L(N 2 N 1 ) = Lmin :

(14)

Whereas for any array the minimum spacing between any two elements dmin must be greater than the minimum feasible spacing s, it is possible to write dmin = Lmin s which gives a constraint on L, i.e., L s : min (16) (15)

i.e., the receiver cannot be placed less than half the distance between two peaks (in the one-way pattern) from the transmitter. By applying the numbers in this paper, L 1:33 106 and DL 50 km. For instance here L = 1:33 106 and DL = 50 km are assumed. 2) Despite the high value of L, the spacing dk = L(k k+1 ) between two consecutive elements of the transmitting array is, in fact, constant and approximately equal to s.3 Because of 1) and 2), the transmitting part of the WA can be simplified into a conventional antenna array, with elements spaced by the same distance, each fed with a slightly different carrier frequency. Such an array is, in fact, an FDA. Nonetheless a viewer, placed at the reference point, which lies at some distance from the active array (according to (18)), keeps observing a pattern like that governed by (11) and shown in Fig. 8, i.e., a range-linear pattern. If now we reconfigure the geometry according to the considerations so far deployed (an equally-spaced transmitting array and a far viewer observing a range-linear pattern), the reference geometry can be displayed as in Fig. 9. In turn the remarkable distance that now lies between the FDA and the reference point leads to
3 In other words the difference 2 d = d k k 1 dk between the spacing of two consecutive active elements of the array is negligible. For instance here 2 dk 7:2 nm, 8k = 1 : : : N 1.

For instance, in the rest of this paper let us set s as half of the minimum wavelength employed in the WA, i.e., 1 s = N 1 : (17) 2 Under the constraints of (16) and (17), the following apply. 1) The reference point is at a distance LN 1 from the closest element of the array. When (14), (16), and (17) apply, it can be demonstrated (Appendix I) that the minimum distance DL between the reference point and the first transmitting antenna is 1 fN 2 c c DL 2 fN 1 f 2f
208

(18)

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

JANUARY 2013

The pattern pFDBS of an FDBS can be therefore written as pFDBS (f , t, R0 , 1 , 2 ) N 1 1 X 1 Rk + R0 = exp j 2 fk t R0 Rk k

k =0

1 = R0
Fig. 9. FDBS with omnidirectional receiver.

N 1 X k =0

1 Rk + 2R0 exp j 2 fk t : Rk k (21)

By introducing (19) and (20), respectively, in the phase and attenuation part of (21), this becomes pFDBS (f , t, R0 , 1 , 2 ) 2R sin Nf t 0 expfj3 + jc g c 2 R R (L, 2 )R0 0 sin f t c (22) where 3 = 2L cos 2 cos 1 : sin(1 2 ) (23)

a reconsideration of the expression for the pattern, and, in particular, the angle under which the target is seen cannot be considered constant. As a consequence the pattern in (11) should be rearranged, and the following two can be demonstrated (Appendix II): Rk = Rk R0 = Lk and cos 2 cos 1 sin(1 2 ) (19)

q 2 + L2 2 + 2L R sin Rk = R0 k 0 2 k q 2 + L2 2 R0 N 1 + 2LN 1 R0 sin 2 = R (L, 2 ): (20)

Now, given the pattern in Fig. 10 and a geometry like in Fig. 9, an immediate application of this concept consists in placing a passive ESA (Fig. 11) at the reference point rather than the omnidirectional antenna

Fig. 10. FDBS. Snapshot of normalized (2-way) pattern with range attenuation at time t = n1=f , n 4. SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR 209

transmitter and, consequently, are subject to different range attenuation, according to [18]. C. FDBS Developments: Windowing and Nonlinear Frequency Shift

Fig. 11. FDBS with ESA receiver.

(e.g. dipole) in Fig. 10. The overall pattern is the product of the 2-way FDBS pattern shown in Fig. 10 with the typical passive phased array pattern in reception (Fig. 12). The resulting pattern is reported in Fig. 13 (where a 9-element passive phased array pointing at 0 has been considered in reception, as in Fig. 12). This figure exhibits the peculiarity that most of the energy is concentrated in limited areas in range and angle. Nonetheless a careful reader may spot asymmetrical peak-to-sidelobe levels in angle (e.g. those at 90 and 20 ), especially evident for lower distances. These can be explained taking into account a geometry like that in Fig. 9 and (22), which show that all the points at a distance R0 from the receiver (L, ) from the are separated by the varying R 2

As in conventional antenna arrays, windowing allows the peak-to-sidelobe ratio to be improved, thus trading off the resolution (in angle and range in this case). Figure 14 shows the pattern of an FDBS after the application of a Hann window in tx and rx. If compared with Fig. 13, where no windows are applied, it is here shown that the beam has enlarged not only in angle (rx), as in standard ESA, but also in range (tx). As a result it is here shown that the sidelobes at 90 , that are the highest in Fig. 13 due to the reduced range attenuation, have been significantly mitigated so that, in practice, only sectorial areas are illuminated. It is also possible to apply nonlinear shifts to the active elements. This allows the shape of the range beam to be further modified, thus making it possible to control the sidelobes in a more sophisticated way. For instance the frequency shifts can be chosen to = 0, 1, 2, : : : 7, 9. In other be equal to kf , with k words whilst the first 8 shifts are the same as in the previous case, the last one differs by an additional f . Figure 15 shows the normalized pattern of the beams at 2 = 0 when linear shifts (blue plot) and nonlinear shifts (red plot) are applied.

Fig. 12. Phased array, pattern with range attenuation. 210 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1 JANUARY 2013

Fig. 13. Snapshot of pattern of FDBS with passive ESA at reference point and at time t = n1=f + R0 =c, n 4.

Fig. 14. Pattern in Fig. 13 after applying Hann window in tx and rx. SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR 211

Fig. 15. Comparison of main and sidelobes for linear and nonlinear shifts.

Whilst the mainlobe remains unchanged, the two patterns differ for the position of sidelobes and nulls. In particular the nulls of the blue pattern become sidelobes in the red pattern and vice versa. As a means of quantitative comparison, the integrated sidelobe ratio (ISLR) has been taken into account. This number has been evaluated by integrating the power of the sidelobes, up to a distance of c=4f , on the left and on the right of each peak and by dividing by the power of the mainlobe. The ISLR of the linear frequency spacing is approximately 18:6 dB, whilst in the nonlinear frequency spacing it is approximately 13:8 dB. Notwithstanding this, the most important remark in this section is that this technique provides an additional control in placing the nulls in range, and consequently, the radar designer can benefit from an additional degree of freedom in this sense. Even more different frequency shifts can lead to a pattern as in Fig. 16, which has been achieved by applying shifts equal to (0, 2, 4, 6, 8, 1, 3, 5, 7)f to an FDBS (with an omnidirectional receiver). The most evident drawback here is clearly that the linearity of the pattern in Fig. 10 is lost, but nulls in range and angle have been generated at specific locations and at the same time within the S-shape of the mainlobe. Consequently, in future work it may be worth investigating the best way of applying diverse frequency shifts to fully control the placement of the nulls of the pattern in range and angle.
212

IV. RANGE-ANGLE DEPENDENT WAVEFORM In this section the WD-MIMO and the FDA concepts are merged. As a consequence each of the 9 elements of the array transmits a code which is modulated on a slightly different carrier. We suggest the name of frequency and waveform diverse MIMO (FWD-MIMO) for this technique. The propagating waveform at range R0 and at angle can therefore be expressed as w(t, R0 , ) = N X R 1 wk (t ) exp j 2 fk t k Rk k
k =1 N

expfjc g X wk (t ) R0
k =1

R0 + kd sin exp j 2 fk t k (24) where fk = fk f0 and = R0 =c. Figure 17 shows the same signals used in the WD-MIMO configuration (Figs. 5 and 6) when the FWD-MIMO technique is used. As can be seen, the propagating waveform, which is the coherent sum of the single waveforms transmitted at element level, changes as a function of range and angle. This is not observed in the WD-MIMO technique (Fig. 6)
JANUARY 2013

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

Fig. 16. Additional example of pattern of FDBS when nonlinear frequency shift is applied at element level.

Fig. 17. Propagated waveform at 0 and 45 at ranges R = 50 km and R = 100 km, FWD-MIMO. SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR 213

which does not transmit over diverse carriers, and it is clearly the result of merger with the FDA concept. By assuming backscattering from a target, in a monostatic radar system configuration the received waveform at the hth receiver can be expressed as rh (t) = tgt
N X k =1

wk (t )

2R0 + (k + h)d sin exp j 2 (fk + fd )t k + nh (t ) (25)

where tgt comprises all the parameters of the 2-way propagation and backscattering (including range attenuation, as in the radar equation, = 2R0 =c), where fd is the Doppler shift, and where n(t) groups all possible nuisance signals. As a consequence the information within the received waveform can be correctly processed only if all the waveforms in transmission and the associated carriers fk are fully known, as shown in a few lines. It is, therefore, clear that this technique has applications for both radar and communication systems. In particular in the latter case, although the transmitted waveform propagates on an omnidirectional pattern, this can be decoded only by receivers with a priori knowledge of the parameters used in transmission. A. Signal Processing As in the conventional WD-MIMO case, the variety of possible waveforms generated by the FWD-MIMO technique (see (24)) does not make it feasible to build a bank of filters that match all the possible range and angle combinations, which may also be subject to ambiguities, as shown later in this section. As a consequence a possible processing approach can be as follows. Starting from the received echo rh (t) in (25) at the hth receiver, downconversion at an appropriate IF and digital conversion are the first steps to perform. This can be expressed mathematically as r [q] = [e
h j 2 (f0 IF)t h

where mp [q] is the filter matched to the pth waveform, W pp is the autocorrelation of wp [q] scaled by tgt and = IF + pf , inclusive of a constant phase term, f h p [q] aggregates all the kinds of nuisance, and n including nonperfect orthogonality between the codes and quantization. As can be seen from the last term of (27), after these steps, there is still a phase term depending on 2R0 =p . Whereas in the far field ej 22R0 =p is, in general, different from ej 22R0 =k , k 6= p, at this stage a range-compensation process has to be performed. Whereas each range R0 is associated to a delay , by knowing the time of transmission, it is possible to remove this term. In this sense the randomization of the phase that arises from ignoring the exact time of transmission provides the received signal with a sort of geometrical encryption, which prevents silent receivers from ignoring (or at least a very good approximation of it) in order to process the received waveform correctly. At the same time this term can provide additional mitigation of multipath or other sources of interference with a waveform corresponding to a range R1 6= R0 , due to the randomization of the phases whenever ej 2(R1 R0 )=p differs sufficiently from ej 22R0 =k . This can be enhanced even further by changing the starting phases of the transmitted signals on a pulse-to-pulse basis and is one of the possible ways to use additional techniques to vary the electromagnetic (EM) characteristics of the signals to give the opportunity to uniquely map the time/range/angle domain with the possibility of recovering the correct signals only as long as all the parameters of transmission (phases, codes, frequency shifts) are known. In removing this term the signal can be written as follows
h h p p r [q] = r [q]ej 22R0 =p "

(p + h)d sin =W pp [q] exp j 2 f d qT p h +n p [q]:

!# (28)

r (t)] t=qT

(26)

where T is the sampling time. Consequently the signal rh [q] has to pass through a bank of filters, each matched to one of the transmitted waveforms. Compensation for the residual frequency shift can also be performed, removing the IF as well, according to the radar applications. All together this can be written as
h p r [q] = rh [q] (mp [q]ej 2fqT )

Consequently, through a processing as in conventional WD-MIMO radar systems, it is possible to recover the direction of arrival of the target through beamforming in receive. This is made possible by compensating for the phase term (p + h)d sin 0 =p , expected when the echo arrives from 0 , similarly to conventional processing in ESAs.4 Eventually Doppler processing can be performed, after collecting an appropriate number of pulses, as in conventional radars. Figure 18 shows a schematic of the processing approach described in (26)(28).
4 In conventional ESAs = . However, whereas ' , as p 0 p 0 discussed in Section III-A, from a practical point of view, the phase can also be compensated according to (p + h)d sin 0 =0 as in this case the approximation d=0 ' d=p is acceptable.

=W pp [q] exp j 2 f d qT h +n p [q]

214

2R0 + (p + h)d sin p

(27)

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

JANUARY 2013

Fig. 18. Schematic of processing chain.

B. Cross-Correlation Properties In this section we discuss the cross-correlation properties of the waveforms in (24). In particular, taking the waveform at one range and one angle as a reference, this has to be compared against the other waveforms at different angles and ranges. This means that the cross-correlation is a function of the delay and the difference in range and angle. Notwithstanding the fact that the PN sequences used in this paper have not been studied for being Doppler tolerant, we also add an insight on the effect of the Doppler shift on the received signals. Whilst in radar theory the ambiguity function depends only on the delay t and the Doppler shift fd , dealing with range and angle mismatches leads us to consider a multi-dimensional cross-correlation function (which in turn relates to a multi-dimensional ambiguity function) C (t, R , , fd ), which arises from the attempt of distinguishing a reference waveform with all those generated by the FWD-MIMO technique. From an analytical point of view, this can be written as
C (t, R , , fd ) 1 = T

Z
0

w (t, R0 , , 0)w(t + t, R0 + R , + , fd )dt:

(29) Figures 1924 show this function when t, R , , and fd are equal to 0, in pairs and alternatively. The reference waveform w(t, R0 , , 0) has been chosen with R = 100 km and = 0 . To highlight

the periodicity of some of the following plots, we do not take into account the range attenuation due to propagation, according to (29). In particular Fig. 19 shows the peak of the cross-correlation as a function of angle and range, i.e., this can be expressed as C (0, R , , 0). As can be seen, range attenuation apart, this plot is the same as that in Fig. 3 and shows that the waveforms generated through the FWD-MIMO are not unique in the range-angle space but that they have the same periodicity as the FDA pattern. This could be predicted by taking into account that the shift in the carrier frequencies is the same as in the FDA. This means that the waveforms generated through this technique have a degree of correlation according to (6). Additional techniques to diversify the periodic pattern have been described in [9] for the FDA, and they can be directly applied to this technique as well. Figure 20 shows C (t, 0, , 0). In this case the behavior of the sidelobes in the delay-domain are the same as those of the cross-correlation of PN sequences. As a consequence a limitation of this technique is that, assuming that the mutual cross-correlation level of the single transmitted waveforms wk (t) have been designed below a specific value, it is not guaranteed that the sidelobes of C (t, 0, ) will be below that level. This is an additional issue which has to be taken into account by the waveform and radar designer. The sidelobes in angle, which cut the plot in t = 0 are the same as a conventional ESA, in line with the properties of the FDA/FDBS.
215

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

Fig. 19. Cross-correlation of reference waveform with others as function of range and angle (t = 0, fd = 0).

Fig. 20. Cross-correlation of reference waveform with others as function of delay and angle (R = 0, fd = 0). 216 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1 JANUARY 2013

Fig. 21. Cross-correlation of reference waveform with others as function of delay and range ( = 0, fd = 0).

Fig. 22. Cross-correlation of reference waveform with others as function of Doppler shift and delay ( = 0, R = 0). SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR 217

Fig. 23. Cross-correlation of reference waveform with others as function of Doppler shift and range ( = 0, t = 0).

Fig. 24. Cross-correlation of reference waveform with others as function of Doppler shift and angle (R = 0, t = 0).

218

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

JANUARY 2013

Finally Fig. 21 shows C (t, R , 0, 0). The behavior in the delay and range domains is the same as the previous cases. In particular the sidelobes in range are the same as those in Fig. 3 (range attenuation apart), and therefore they follow (6), with a repetition range of c=f (one way). In evaluating the Doppler effect on C (t, R , , fd ), a few pulses (H = 20) have been taken into account. The pulse repetition frequency (PRF) has been set to 3 kHz, i.e., the same as the frequency shift, according to considerations in [19] and to the Doppler shift to 500 Hz. Equation (29) has, consequentl,y been rewritten as follows C (t, R , , fd ) H 1 Z 1 X T+h=PRF = w (t, R0 , , 0) HT 0+h=PRF
h=0

w(t + t, R0 + R , + , fd )dt:

(30)

Figure 22 shows C (t, 0, 0, fd ). This result is in line with the cross-correlation of PN codes and can be improved, especially in the 500-Hz cut, if the employed PN codes have good Doppler properties. Nonetheless it is here clear that the Doppler shift of the target can be spotted as in conventional ESAs or radars in general. C (0, R , 0, fd ) is shown in Fig. 23. An important remark is that ambiguity in range is also preserved in range-Doppler analysis presented here. Finally Fig. 24 plots C (0, 0, , fd ). In conclusion it can be seen that the six plots in this section overall present ambiguities, especially in range. This is due to the periodicity of the peaks of the pattern in (6) and following. Whilst to some extent it seems possible to mitigate this issue, e.g., trading-off f so as to make the ratio c=f as big as desired or performing a sort of stagger on f , the radar designer will need to deal with these new kinds of ambiguities in the same way as he/she copes for instance with maximum unambiguous Doppler shift and maximum unambiguous range. V. DISCUSSION AND CONCLUSIONS In this paper a couple of major developments of the FDA concept have been presented. After focusing the analysis on the WA, which aims to stretch the S-shaped pattern of the FDA through a different arrangement of the elements of the array of antennas, the FDBS has been shown to allow in reception a pattern which is in fact orthogonal to that of a standard phased array or ESA. As a result of the transmission of multiple frequencies from each antenna element, a range-dependency of the pattern, rather than the usual angle-dependency, has been achieved, and it stands out clearly that the range-dependency is obtained in exchange of the angle resolution.

However, as the slight frequency diversity is applied in transmission only, receiving with a phased array can reintroduce the angular resolution, which allows the energy to be concentrated in particular areas. In addition, because of an intrinsic duality, it is important to highlight that the same pattern would be achieved if an ESA were to transmit and an FDA were to receive. However this is not the best configuration for FDA systems because, in effect, one of the peculiarities of this technique, i.e., illuminating at specific ranges only, is not exploited. It has also been shown that in FDBS/FDA systems, the pattern is controlled by a number of parameters, which are 1) the location of the receiver and the distance, 2) the frequency shift assigned to each element, 3) the distance between the elements in transmission, 4) the range, and, finally, 5) the angle. Most of these are in common with conventional phased arrays or ESA, as the FDA is in fact a development of these systems. However, because of the introduction of the frequency shifts, the effects on the pattern of an FDA/FDBS can differ widely. In addition the introduction of frequency shifts provides the radar designer with an additional degree of freedom. Possible fields that can benefit from a hybrid application of this technique are, for instance, SAR/ISAR imaging and interferometry, as much as sonar beamforming. It has also been shown that by introducing minor mismatches into the system at different levels offers the designer of future radar systems many new degrees of freedom that, hitherto, have not been fully explored. Additionally, with the advent of high-speed digital computation, these techniques can be exploited by using the technology of today. The challenge is more one of understanding the range of possibilities offered against their utility in important applications. In this case the potential of generating (using a pulsed version of the FDBS) a pattern orthogonal to that of a standard ESA is not only interesting for the development of LPI (low probability of intercept) radars in range, but also for other applications, such as controlling or at least mitigating echoes from unwanted ranges, as in low-grazing angle multipath. The FWD-MIMO concept, as a means to include a geometrical encryption in the transmitted waveform, has been presented. This technique is in fact the result of merging the WD-MIMO technique, which transmits different waveforms over the same carrier frequency, with the FDA technique, which transmits the same waveform over slightly different carrier frequencies. It has therefore been shown that the transmitted waveform changes in range and angle in accordance with the FDA pattern. It has not been reported here for brevity, but nonetheless, it is evident that the FWD-MIMO concept can be applied to a bistatic configuration as that of the FDBS, so as to avoid the
219

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

S-shaped pattern of the peak of the cross-correlation, as in Fig. 19. A final interesting remark is that, if the bandwidths of the signals in Fig. 2 are narrowed to a pure tone equal to f and if the transmitted waveforms are chosen within an appropriate constellation of symbols, the FWD-MIMO looks like an orthogonal frequency-division multiplexing (OFDM) radar [20, 21], in which the OFDM signal has been spread all over the elements of the array. Under this point of view, the FWD-MIMO can be considered a generalization of such systems, even if spreading the bandwidth across the antennas may also lead to a new potential for these systems. APPENDIX I Given the hypotheses as in (14), (16), and (17) in Section III-B, the distance between the reference point and the closer element of the WA is s DL = LN 1 min N 1 = N 2 N 1 N 1 c fN 1 1 c = 2 c c fN 1 fN 2 fN 1 1 = 1 2 1 2 1 fN 1 1 c fN 1
1 2 N 1

From the first 2 equalities of (33), Rk can be written as cos 2 cos 2 cos 1 Rk = Rk R0 = R0 1 = R0 cos 1 cos 1 (34) and through the last two equations of (33), R0 as R0 = Lk cos 1 : sin(1 2 ) (35)

Introducing (35) into (34) gives (19) in Section III-B. Equation (20) can be demonstrated by considering the law of cosines in Euclidean geometry. In particular, taking into account the same triangle as in the previous case 2 2 = R0 + L2 2 (36) + 2 : Rk k 2Lk R0 cos 2 Whereas cos((=2) + 2 ) = sin 2 and for the amplitude attenuation Lk LN 1 , this leads immediately to (20).
REFERENCES [1] Barton, D. K. Frequency Agility and Diversity. Norwood, MA: Artech House, 1977. Antonik, P., et al. Frequency diverse array radars. In Proceedings of the IEEE Radar Conference, Verona, NY, Apr. 2427, 2006. Secmen, M., et al. Frequency diverse array antenna with periodic time modulated pattern in range and angle. In Proceedings of the IEEE Radar Conference, Waltham, MA, Apr. 1720, 2007, pp. 427430. Antonik, P., et al. Multi-mission multi-mode waveform diversity. In Proceedings of the IEEE Radar Conference, Verona, NY, Apr. 2006. Sammartino, P. F. A comparison of processing approaches for distributed radar sensing. Ph.D. thesis, Dept. of Electronic and Electrical Engineering, University College, London, 2009. Sammartino, P. F. and Baker, C. J. The frequency diverse bistatic system. In Proceedings of the Fourth International Conference on Waveform Diversity and Design, Orlando, FL, Feb. 813, 2009, pp. 155159. Huang, J., Tong, K-F., and Baker, C. J. Frequency diverse array with beam scanning feature. In Proceedings of the IEEE Antennas and Propagation Society International Symposium, 2008 (AP-S 2008), San Diego, CA, July 511, 2008, pp. 14. Huang, J., et al. Frequency diverse array: Simulation and design. In Proceedings of the IEEE Radar Conference, Pasadena, CA, May 2009, pp. 14. Sammartino, P. F. and Baker, C. J. Developments in the frequency diverse bistatic system. In Proceedings of the IEEE Radar Conference 2009, Pasadena, CA, May 2009, pp. 15. JANUARY 2013

[2]

[3]

fN 2 =

fN 1

1 c f fN 1 fN 1 fN 1 fN 2

[4]

= = as in (18). APPENDIX II

1 fN 2 c 2 f fN 1 1 fN 2 c 2 fN 1 f (31)

[5]

[6]

Taking into account the geometry in Fig. 9, the triangle comprised by the k th transmitter, the omnidirectional receiver, and a generic point of the pattern front, the following applies (law of sines) R R Lk k 0 = = sin 0 sin sin + 2 1 2 2 (32)

[7]

[8]

where 0 = 1 2 . The set of equations in (32) can be written as Rk R0 Lk = = : (33) cos 2 cos 1 sin(1 2 )
220

[9]

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1

[10]

[11]

[12]

[13]

[14]

[15]

Farooq, J., Temple, M. A., and Saville, M. A. Application of frequency diverse arrays to synthetic aperture radar imaging. In Proceedings of the International Conference on Electromagnetics in Advanced Applications (ICEAA 2007), Torino, Italy, Sept. 2007, pp. 447449. Farooq, J., Temple, M. A., and Saville, M. A. Exploiting frequency diverse array processing to improve SAR image resolution. In Proceedings of the IEEE Radar Conference, (RADAR 08), Rome, Italy, May 2008, pp. 15. Baizert, P., et al. Forward-looking radar GMTI benefits using a linear frequency diverse array. Electronics Letters, 42, 22 (2006), 13111312. Antonik, P. An investigation of a frequency diverse array. Ph.D. thesis, Dept. of Electronic and Electrical Engineering, University College, London, 2009. Farina, A. Antenna-Based Signal Processing Techniques for Radar Systems. Norwood, MA: Artech House, 1992. Li, J. and Stoica, P. MIMO Radar Signal Processing. Hoboken, NJ: Wiley, 2008.

[16]

[17]

[18]

[19]

[20]

[21]

Bliss, D., Forsythe, K., and Fawcett, G. MIMO radar: Resolution, performance and waveforms. Presented at the Fourteenth Annual Adaptive Sensor Array Processing Workshop, MIT, June 67, 2006. Fishler, E., et al. Spatial diversity in radarsModels and detection performance. IEEE Transactions on Signal Processing, 54, 3 (Mar. 2006), 823838. Skolnik, M. I. Introduction to Radar Systems. New York: McGraw-Hill, 1981. Sammartino, P. F. and Baker, C. J. Frequency diverse array radars. Systems Engineering & Assessment Ltd, UK, Technical Report, Aug. 2006. Lellouch, G., et al. OFDM waveforms for frequency agility and opportunities for Doppler processing in radar. In Proceedings of the IEEE Radar Conference, Rome, Italy, May 2008, pp. 16. Donnet, B. J. and Longstaff, I. D. Combining MIMO radar with OFDM communications. In Proceedings of the 3rd European Radar Conference (EuRAD), Manchester, UK, Sept. 1315, 2006, pp. 3740.

SAMMARTINO, ET AL.: FREQUENCY DIVERSE MIMO TECHNIQUES FOR RADAR

221

Pier Francesco Sammartino received the Laurea in telecommunication engineering from the University of Rome La Sapienza and his Ph.D. in electronic and electrical engineering from the University College, London. He is currently a post-doc scientist with the Joint Research Centre of the European Commission in the field of radar systems. Until May 2009 he was an analyst in radar research with CEA Technologies in Canberra, Australia. His research interests include MIMO radars, digital beamforming, radar signal processing, electronically-scanned and frequency-diverse arrays, radar networks, interferometry, SAR imaging, and micro-Doppler techniques.

Chris Baker is the Ohio State Research Scholar in Integrated Sensor Systems at The Ohio State University. Until June 2011 he was the Dean and Director of the College of Engineering and Computer Science at the Australian National University (ANU). Prior to this he held the Thales-Royal Academy of Engineering Chair of intelligent radar systems based at University College London. He has been actively engaged in radar system research since 1984 and is the author of over two hundred publications. His research interests include, Coherent radar techniques, radar signal processing, radar signal interpretation, Electronically scanned radar systems, natural echo locating systems and radar imaging. He is the recipient of the IEE Mountbatten premium (twice), the IEE Institute premium and is a Fellow of the IET. He is a visiting Professor at the University of Cape Town, Cranfield University, University College London, Adelaide University and Wright State University.

Hugh Griffiths (M86SM90F99) holds the THALES/Royal Academy Chair of RF Sensors in the Department of Electronic and Electrical Engineering at University College London, England. From 20062008 he was the Principal of the Defence Academy of Management and Technology, Shrivenham. He received his M.A. degree in physics from Oxford University in 1975. Then he spent three years working in industry before joining the University College London, where he received a Ph.D. degree in 1986 and a D.Sc. (Eng.) degree in 2000 and where he served as head of department from 20012006. His research interests include radar and sonar systems, signal processing (particularly synthetic aperture radar and bistatic and multistatic radar), and antenna measurement techniques. He carried out some of the first experiments in passive bistatic radar. Dr. Griffiths has published over four hundred papers and technical articles in the fields of radar, antennas, and sonar. In 1996 he received the IEEE AESS Fred Nathanson Award (Radar Systems Panel Award). He has also received the Brabazon Premium of the IERE and the Mountbatten and Maxwell Premium Awards of the IEE. He is a Fellow of the IET, and in 1997, he was elected to Fellowship of the Royal Academy of Engineering. He serves as President of the IEEE Aerospace and Electronic Systems Society for 20122013. He has been a member of the IEEE AES Radar Systems Panel since 1989, serving as the chair from 20072009, and he chaired the working group which revised the IEEE Radar Definitions Standard P686 and reaffirmed the Radar Letter Band Standard in 2008. He is an IEEE AES distinguished lecturer.
222 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 49, NO. 1 JANUARY 2013