Pre-Print Only

Signal Waveforms and Range/Angle Coupling in

Coherent Colocated MIMO Radar
Olivier Rabaste, Laurent Savy, Mathieu Cattenoz Jean-Paul Guyvarch
ONERA, The French Aerospace Lab Thales Air Systems
BP 80100 3 avenue Charles Lindbergh
91123 Palaiseau Cedex, France 94628 RUNGIS Cedex - France

Abstract—In this paper, we consider the coherent colocated

MIMO radar framework. We show in this paper that the MIMO
ambiguity function presents a range/angle coupling for any
waveform family except for perfectly orthogonal families that
are unrealistic in practice. We then propose a general model
for intrapulse coded MIMO waveforms, and we exhibit different
waveform families within this model. For each coding scheme,
the MIMO ambiguity function is studied and the range/angle
coupling characterized. We also compute the corresponding
Cramer-Rao Lower-Bounds. Finally real measurements obtained
from a MIMO radar permit to confirm the shape of the ambiguity Fig. 1. Transmission beams: a) scanning pencil beam; b) wide beam; c)
functions for the studied waveforms. multiple simultaneous coded beams.

I. I NTRODUCTION focused beams, as shown on (Figure 1.c), allowing a wide

Modern radar systems, for medium or long range appli- angle domain to be explored instantaneously, without loss of
cations, are generally based on active antennas, transmitting transmit directivity as with classical wide beam illumination
the signals generated by agile waveform synthesizers, and of Figure 1.b. It is worth noting that this coherent MIMO as
adaptively processing the received echoes, for extraction of nothing to do with none-coherent (or statistical) MIMO, that
targets from clutter and identification of potential threats. mainly exploits spatial diversity on target, using well separated
Standard surveillance modes involve electronic scanning (multistatic, none-colocated) transmitting antennas [5], [6].
of a focused beam (Figure 1.a), successively exploring the Coherent co-located MIMO is the only mean for a radar to
directions of interest with a sequence of search waveforms obtain an ultra-wide angular beam on transmit (more than the
(e.g. High Repetition Frequency bursts for long-range air-air classical factor 3 or 4 compared to the focused beamwidth)
search, Low Repetition Frequency bursts for surface-based with good sidelobe and mainlobe properties (ripple, sidelobe
radars), and removing ambiguities through comparison of the level). Ultra wide beam are required for instance for transitory
received signals for the successive bursts at different repetition targets (helicopter pop-up, periscope), or slow small moving
frequencies and/or different wavelengths. Confirmation modes targets in competition with clutter and/or strong targets.
may also be interleaved between the standard scanning modes The coherent processing on receive is obtained via a general
for improved detection probabilities in directions where ele- matched filter in range, angles and Doppler. The corresponding
mentary detections have been obtained. MIMO ambiguity function which represents the output of the
Some modern radars also make use of a widened beam on MIMO processing for a given set of transmitted signals, can
transmit (Figure 1.b), allowing for longer illumination time be derived. It is an interesting tool to characterize different
and thus better extraction of targets, and combining multiple waveform families. This MIMO ambiguity function has been
receiving antennas for digital beamforming of focused pencil- extensively and theoretically defined and studied in [7], [3],
beams in parallel. This technique is for instance used for [8]. However most of the properties stated are obtained under
surface-based radar, with a wide elevation beam on transmit the assumption that the transmitted waveforms are perfectly
and so-called stacked beams on receive. orthogonal for any (angle-range-doppler) target hypothesis. In
This paper deal with an alternative to these basic scanning other words, in most cases authors do not consider the problem
or staring modes, called coherent co-located MIMO [1], [2], induced by the range sidelobes and/or the spreading of the
[3], [4]. It consists in space-time coding on transmit of each signal energy in the range/angle plane, even though these
individual transmitters (or sub-arrays), allowing to identify sidelobes may present a strong level. However these sidelobes
each of them by a convenient processing on receive, and are of critical importance for the radar detection problem.
hence to recover the angular directivity of the whole antenna. In this paper, we show that the MIMO ambiguity function
From a radar functional point of view, everything appears exhibits a range/angle coupling as soon as the transmitted
as if the directions were explored simultaneously by coded waveforms are not perfectly orthogonal, where the orthogo-

978-1-4673-5178-2/13/$31.00 © 2013 IEEE 157 RADAR 2013

nality is considered for any waveform and any time delay The optimal coherent MIMO processing on receive, where
values between the waveforms. Note that it is not possible we have skipped for simplicity the antenna gains, is given by:
to find a set of waveform that satisfies this orthogonality, so
that range coupling will always occur in practice. We then Beamforming on receive Beamforming on transmit
    
propose a general model for describing MIMO waveforms and NR −1 NE −1
T T
we present different possible families of waveforms that can be r(τ, ν, θ) = e−jxR,n k(θ) e−jxE,m k(θ)
used in coherent MIMO radars. For each family, we present the n=0 m =0
specificity of the observed range/angle coupling. It comes that,
× sn (t)s∗m (t + τ )e−j2πνt dt.
r
while for some waveform families the range/angle coupling
  
is approximately uniformly spread in the range/angle plane, Transmitter separation and range compression
for some other waveform families it may be concentrated (1)
in specific areas, even though the considered waveforms are Clearly this optimal processing can be decomposed into
all orthogonal for the zero time delay. The computation of three steps: first, at each reception antenna, application of a
Cramer-Rao Lower-Bound ellipses permits to quantify the filter matched to each transmitted waveforms (i.e. NE matched
distortion of the ambiguity mainlobe. Finally, in order to check filters applied to NR received data, producing NE NR output
the validity of the theoretical ambiguity function presented data streams); second a transmission processing that permits
here, a real experiment was conducted using a MIMO radar to retrieve the transmission phase; third a reception processing
demonstrator. The ambiguity functions obtained with these real that corresponds to the usual beamforming step.
measurements fairly match the theoretical ones. Inserting the expression of srn (t) in (1) and replacing ν − νc
II. C OHERENT MIMO AMBIGUITY FUNCTION by ν and τ − τc by τ provides the MIMO ambiguity function
given by [3]:
In this paper, we consider a transmitting array of NE
antennas and a receiving array composed of NR antennas.
N −1 

R
The position of the mth antenna of the transmitting array is A(τ, ν, θ, θc ) = e jxT
R,n (k(θc )−k(θ)) ×
denoted by vector xE,m while the position of the nth antenna n=0
of the receiving array is given by vector xR,n . We assume that E −1 N
N E −1
T T
the transmitting and receiving arrays are colocated. Of course, ejxE,m k(θc )−jxE,m k(θ) sm (t)s∗m (t + τ )e−j2πνt dt,
in a more general setting, subarrays could be considered in m =0 m=0
(2)
place of single antennas.
which is a function of four parameters (delay, doppler, target
A. Expression of the MIMO ambiguity function angle, reception angle) if only one angle direction (azimuth
In the coherent colocated MIMO framework, all elementary or elevation) is considered. If two directions were considered,
antennas (or subarrays) of the transmitting array transmit dif- it would become a function of six parameters. Note that the
ferent waveforms. Denoting by sm (t) the waveform assigned target angle and the reception angle cannot be summarized by
to the mth antenna, the signal transmitted by the array is only one angle parameter because in coherent MIMO, signals
N
E −1
received by two targets in two different directions differ.
T
s(t, θ) = E
gm (θ)ejxE,m k(θ) sm (t), Using the above vector notation, the ambiguity function can
m=0
be rewritten in the following way:
 
where θ is the considered direction, k(θ) is the wave vector, A(τ, ν, θ, θc ) = sH R (θ)sR (θc ) ×
gmE
(θ) is the gain of transmitting subarray m in direction θ 
T H −j2πνt
and the notation ·T represents the matrix transpose. If a target sE (θ) s(t)s (t + τ )e dt s∗E (θc ),
is present in direction θc with delay τc and doppler νc , then
the signal received on the nth antenna is: where xH is the hermitian transpose of vector x.
NE −1
 Since the reception processing is completely decoupled in
T T
srn (t) = gnR (θc )ejxR,n k(θc ) E
gm (θc )ejxE,m k(θc ) sm (t−τc )ej2πνc t , equation (2), we may consider only the following expression,
m=0 corresponding to the case of one single reception antenna:

where gnR (θc ) is the gain of receiving subarray n in direction
θc . Concatenating signals srn (t) from all receiving antennas Ae (τ, ν, θ, θc ) = sTE (θc ) s(t)sH (t + τ )e−j2πνt dt s∗E (θ).
in a single vector sr (t) = [sr0 (t), sr1 (t), . . . , srNR −1 (t)]T , (3)
and transmitted waveforms sm (t) in a single vector s(t) = For applications where the Doppler effect is negligible
[s0 (t), s1 (t), . . . , sNE −1 (t)]T , we get: within the pulse duration, it can be decoupled in the expression
  (3). This is the case for instance for the detection of slow
sr (t) = sR (θc ) sE (θc )T s(t − τc )ej2πνc t ,
moving targets. For simplicity, we restrict our analysis to that
where sR (θ) and sE (θ) are the steering vectors for transmis- approximation in the following, and (3) can be simplified
sion and reception, whose nth elements are given by: further to:
T T
(sR (θ))n = gnR (θ)ejxR,n k(θ) and (sE (θ))n = gnE (θ)ejxE,n k(θ) . Ae (τ, θ, θc ) = sTE (θc )S(τ )s∗E (θ), (4)

158
where W11 W12 W1Nc
antenna 0 F11 F12
... F1Nc
S(τ ) = s(t)sH (t + τ ) dt
antenna 1 W21 W22 ... W2Nc
F21 F22 F2Nc
is the matrix containing all the information about the auto and
cross correlations of the transmitted waveforms. .. .. Δt
. .
Other theoretical properties of the MIMO ambiguity func-
tion have been thoroughly presented and discussed in [3], [8].
WNE 1 WNE 2 WNE Nc
antenna NE − 1 FNE 1 FNE 2
... FNE Nc
B. Coupling effect between delay and angle parameters
t
From the expression of the MIMO ambiguity function (4), it Tp
can be seen that effects of delays and angles cannot generally
be decoupled, except for two specific cases: Fig. 2. General intrapulse coding scheme.

• S(τ ) = λs (τ )1NE ×NE where 1NE ×NE is a matrix of A. Intrapulse coding formalism
size NE × NE filled with ones and λs (τ ) is a correlation We consider here that the MIMO radar transmits a pulse
function. In that case, the ambiguity function becomes: train. Each antenna transmits its specific waveform during one
  pulse, and we assume here that this waveform is identical
Ae (τ, θ, θc ) = sTE (θc )1NE ×NE s∗E (θ) λs (τ ). from pulse to pulse. Therefore the orthogonality between the
different waveform arises only from the intrapulse coding.
This case arises when sm (t) = s(t) for all m, i.e. the Each pulse can be decomposed into Nc time slots or
classic phased array where all transmitted signals are “chips”. To each time slot and each antenna can be assigned
identical; it is of no interest in the MIMO framework. specific phases and frequencies. We therefore propose the
• S(τ ) = λs (τ )INE where INE is the identity matrix. In general model for intrapulse coding:
that case, the ambiguity function becomes:
N
c −1
  sm (t) = Wmp ej2πFmp t u(t − pΔt ),
Ae (τ, θ, θc ) = sTE (θc )s∗E (θ) λs (τ ).
p=0

This arises when all cross ambiguities are equal to zero where Wmp and Fmp represent the phase and the frequency
for all τ and all auto ambiguities are identical, i.e.: associated to the signal transmitted by the antenna m during
the time slot p. Δt represents the duration of one chip, and u(t)
sm (t)s∗m (t + τ ) dt = λs (τ )δm,m , represents the elementary waveform, that can be for instance
a simple rectangular pulse or a linear frequency modulated
signal (“chirp”). This expression permits to design at the same
where δm,m is the Kronecker operator. In other words, time waveforms with phase and/or frequency coding. Such a
it means that the transmitted waveforms are perfectly code is represented in figure 2. Note that the number of chips
orthogonal. Nc can be set as desired depending on the signal to transmit.
The second case is very interesting since it implies for instance For instance, if the transmitted pulse is a chirp, then we can
that an error in the estimated delay will not necessarily induce set Nc = 1 and use a chirp for elementary pulse u(t).
an error in the estimated direction. Unfortunately it cannot be Using the vectorial notation, the transmitted signal can be
exactly achieved in practice since it is not possible to generate expressed as:
NE perfectly orthogonal waveforms with identical autocorre- s(t) = (W ◦ F(t))u(t),
lations. It is however possible to design approximately orthog-
onal signals. Then, delays and angles may be approximately where ◦ represents the Hadamard product, W and F(t) are
decoupled again. The counterpart will generally be an increase matrices of size NE × Nc whose elements are respectively
in the sidelobe level. We will now present several possible provided by Wmp and ej2πFmp t , and
waveform families achieving this approximated orthogonality. u(t) = [u(t) u(t − Δt ) . . . u(t − (Nc − 1)Δt )]T .
We will now present different families of codes that can be
III. C OHERENT MIMO WAVEFORMS
used in MIMO radar.
In this section, we present first a general formalism for B. Circulating pulse (TDMA)
intrapulse coding of MIMO waveforms. Then we present
A first simple code is the circulating pulse, that consists in
several different coding schemes. For all the results presented
transmitting different cyclic permutations of the same code on
in the following, the parameters used for the simulations
the different antennas. This code can be defined by setting
have been set (unless differently specified) to: NE = 12
transmission antennas, Tp = 63.5 μs. W = C(s) and F(t) = 1NE ×Nc ,

159
 Circulating code (inter−code distance = 1), B = 2 MHz, theoretical signals FDMA, B = 2 MHz, theoretical signals
5 0 5 0

4 4
−5 −5

3 3
−10 −10
2 2

−15 −15
1 1

Range [km]
Range [km]

0 −20 0 −20

-1 -1
−25 −25

-2 -2
−30 −30
-3 -3

−35 −35
-4 -4

-5 −40 -5 −40
-40 -20 0 -40 -20 0
0 0

-10 -10

-20 -20

-30 -30

-40 -40
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
sin(θ)
sin(θ)

Fig. 3. Range - angle cut of the MIMO ambiguity function of the circulating Fig. 4. Range - angle cut of the MIMO ambiguity function for the FDMA
pulse for θc = 0 with a PN sequence of length Nc = 127. NE = 12 coding scheme with elementary bandwidth B = 166.67 kHz (i.e. total
transmission antennas. Tp = 63.5 μs, B = 2 MHz. One single antenna on bandwidth Btot = 2 MHz). NE = 12 transmission antennas. Tp = 100
receive. μs.

where C(s) is the circulant matrix built from signal code s coupling along the line τ = 1/(2Δf ) sin(θ), or, by replacing
containing Nc chips. the delay by the range, d = c/(4Δf ) sin(θ) where the range
This code fulfills the orthogonality condition for τ = 0 if is d = cτ /2 and c is the wave velocity. This coupling can
the code used presents good periodic autocorrelation features be easily seen in figure 4 that presents the MIMO ambiguity
(for instance a PN sequence) but not for non zero delays, function for the FDMA coding scheme with B = 166.67 kHz,
thus limiting the range resolution to that corresponding to the where Δf has been set to B. Interestingly we can notice that
bandwidth of s multiplied by the number of antennas. Cuts of the mainlobe width corresponds to the overall bandwidth used,
the resulting ambiguity function are presented in figure 3 for equal to Btot = 2 MHz.
the target position θc = 0. Let us notice that for that specific
code, the transmission directivity has been obtained at the price D. One phase code per antenna (CDMA)
of a loss in the range resolution. We have seen that perfect orthogonality is desired to
remove the range/angle coupling. Even though this perfect
C. One frequency per antenna (FDMA)
orthogonality cannot be obtained, it is possible to consider
Another simple case to consider is the case where all code families that present features close to this orthogonality.
antennas transmit signals at different frequencies. This strategy For instance, some classes of codes have been designed in
corresponds to the FDMA multiplexing in digital communi- digital communications to exhibit very good autocorrelation
cations. It corresponds to and crosscorrelations properties. This is the case for instance
T for the Gold codes [9]. Although the good properties of these
Nc = 1, W = 1NE ×1 and F = [0, Δf , . . . , (NE −1)Δf ] ,
codes are theoretically obtained for periodic correlations, they
where Δf is a frequency interval larger or equal to the signal remain interesting in the considered aperiodic case.
bandwidth, i.e. Δf ≥ B with B the bandwidth of u(t). For phase codes, the number of chips Nc is set so as to
Clearly the transmitted signals are orthogonal since they do provide a given desired bandwidth. Then the frequency matrix
not share the same frequency domains. But this is not enough is simply set to F = 1NE ×Nc while the phase matrix W
to get perfect delay - angle decoupling. Indeed, if the initial is provided by the considered phase codes. The ambiguity
phases are the same for the different signals, then the signal function for the Gold codes is presented in figure 5. We can
transmitted by the array at a given time instant t is the vector notice that for this class of codes, the range/angle sidelobes
are approximately uniformly spread in the range/angle plane.
[1, ej2πΔf t , . . . , ej2π(NE −1)Δf t ]T
Therefore approximate decoupling has indeed been obtained,
which corresponds to a direction θ = asin(2Δf t) when at the cost of a relatively high sidelobe level that may be
considering a linear array with antennas separated by λ/2. problematic in the presence of spread clutter.
Therefore this coding scheme resorts to a fast sweeping of
the different angular directions during the pulse duration. This E. Cramer-Rao-Lower-Bound ellipses
also means that the signal received in direction θ at a given The range/angle coupling discussed previously can also
time t is similar to the signal received in direction 0 at time be highlighted thanks to the well-known Cramer-Rao-Lower-
t = sin(θ)/(2Δf ), thus leading to a noticeable delay - angle Bound (CRLB) [10]. This CRLB is given by the inverse of

160
 Phase codes (Gold), B = 2 MHz, theoretical signals
5 0
100
TDMA
4 FDMA
−5 CDMA

3
−10
2
50
−15
1
Range [km]

0 −20

-1
−25

range (m)
-2 0

−30
-3

−35
-4

-5 −40 -50
-40 -20 0
0

-10

-20

-30 -100
-10 -5 0 5 10
-40 reception angle (deg)
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
sin(θ)

Fig. 6. CRLB ellipses in range and reception angle for the TDMA, FDMA
Fig. 5. Range - angle cut of the MIMO ambiguity function for the CDMA and CDMA codes discussed previously with same parameters.
coding scheme with Gold code of length Nm = 127. Nt = 12 transmission
antennas. Tp = 100 μs, B = 2.55 MHz.

antenna, may damage the orthogonality between the related

the Fisher information matrix J. In radar, each term Jkl of waveform and all others, and therefore induce an additive
this matrix can be computed from the ambiguity function by: ”noise” on the global ambiguity function. To measure the
possible effects induced by these defaults on the MIMO
∂ 2 |Ae (τ, θ, θc )|2 ambiguity function, we proceeded experimental tests with a
Jkl = −SNRout ,
∂γk ∂γl real MIMO radar: HYCAM.
where γk is one of the parameters τ , θ, θc , and SNRout
A. The HYCAM platform
is the SNR at the output of the MIMO processing. For the
sake of simplicity, we will consider only the two parameters A multifunction MIMO radar - named HYCAM - has been
τ and θ. In that case, J is a 2 × 2-matrix. Computing this designed and build by ONERA. The objective was to develop
matrix and the corresponding CRLB can provide insight on a system taking into account nowadays radar architecture but
the range/angle coupling for a given coding scheme. Indeed also foremost future radar concepts like MIMO. It operates
the CRLB matrix defines an ellipse corresponding to the in S band with a bandwidth up to 500 MHz. The antenna
minimal possible bound that can be achieved for estimating the arrays are composed of 12 electronically steerable columns
corresponding estimator. We present in figure 6 this ellipses for for the transmission and 16 reconfigurable mono-pulse capable
the three codes discussed in the previous sections. Note that columns for the reception. Up to 3 intermediate frequency
these ellipses are computed on the code after the band-pass signals can be simultaneously up-converted and provided to
transmission filter which permits to limit the bandwidth for the antennas through an optical rotary joint. Consequently, this
phase codes. Clearly the shapes of the ellipses differ between make it possible to work as real MIMO with the generation and
the codes. In particular the CRLB ellipse for the FDMA code transmission of 3 orthogonal waveforms. Once captured, the
shows a clear range/angle coupling. This coupling induces an received signals are downlinked through the optical fiber, then
increase of the minimal variance of the estimator in range and converted, digitalized and recordered for later data analysis.
angle that can be achieved for this specific coding scheme.
B. Sequential MIMO experiments
Among the three studed codes, the one producing the CRLB
ellipse with the smallest surface is the CDMA code, thus The aim of the experiments is to acquire data in the context
allowing the best estimate for single target estimation. On the of sequential MIMO transmission with HYCAM to a specific
contrary this code provides the highest sidelobe level, spread target simulator. The signals are generated by an Arbitrary
over the whole ambiguity domain, which prevent against its Waveform Generator (AWG). We want to test families of 12
use in strong clutter situations or for weak target detection. waveforms but since HYCAM cannot proceed more than 3
different signals simultanously, the transmissions and acqui-
IV. E XPERIMENTS ON A MIMO RADAR DEMONSTRATOR sitions are done sequentially: each waveform is transmitted
Many different effects of reality are not taken into account ”one by one” by the related antenna. A target simulator facing
in the theoretical data, especially material defaults like non- HYCAM at 250 meters has the ability to apply a delay and
linear phase of microwave transmit/receive chains and antenna a Doppler to the signal, such that it is possible to extract the
couling (inducing impedance mismatch) and also non-optimal transmitted signal from the clutter during the processing phase.
propagation like multipath. One threat with the MIMO is The successive received signal are calibrated and summed in
that a ”localized” issue, for example a defective transmission order to recover the equivalent MIMO situation.

161
 Circulating code (inter−code distance = 1), B = 2 MHz, HYCAM facing the target simulator Phase codes (Gold), B = 2 MHz, HYCAM facing the target simulator
5 0 5 0

4 4
−5 −5

3 3
−10 −10
2 2

−15 −15
1 1
Range [km]

Range [km]
0 −20 0 −20

-1 -1
−25 −25

-2 -2
−30 −30
-3 -3

−35 −35
-4 -4

-5 −40 -5 −40
-40 -20 0 -40 -20 0
0 0

-10 -10

-20 -20

-30 -30

-40 -40
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
sin(θ) sin(θ)

Fig. 7. Range - angle cut of the MIMO ambiguity function of the circulating Fig. 9. Range - angle cut of the MIMO ambiguity function for the CDMA
pulse with a PN sequence of length Nc = 127 obtained from real HYCAM coding scheme with Gold code of length Nm = 127 obtained from real
data. NE = 12 transmission antennas. Tp = 63.5 μs, B = 2 MHz. One HYCAM data. Nt = 12 transmission antennas. Tp = 100 μs, B = 2.55
single antenna on receive. MHz.
FDMA, B = 2 MHz, HYCAM facing the target simulator
5 0
assumed in most of the papers, it cannot be achieved in
4
−5
practice, so that range/angle coupling will always occur. We
3

2
−10 have therefore studied this feature for three different coding
1
−15 schemes that may take the form of a diagonal ridge in
Range [km]

0 −20
the range/angle plane, or relatively high sidelobes spread
-1
−25
over the whole ambiguity domain. We have also shown that
-2 a quantitative insight on the range/angle coupling can be
−30
-3 provided by CRLB ellipses. Measurements obtained from
−35
-4
a real MIMO radar have permitted to show a fairly good
-5
-40 -20 0
0
−40
match between theoretical ambiguity functions and measured
-10 ambiguity functions taking into account many transmission
-20
defaults. Finally, note that methods for generating new families
-30
of quasi-orthogonal waveforms with better range/angle cou-
-40
-1 -0.8 -0.6 -0.4 -0.2 0
sin(θ)
0.2 0.4 0.6 0.8 1
pling and/or sidelobe levels would be very relevant to produce
interesting MIMO ambiguity functions.
Fig. 8. Range - angle cut of the MIMO ambiguity function for the FDMA
coding scheme with elementary bandwidth B = 166.67 kHz (i.e. total R EFERENCES
bandwidth Btot = 2 MHz) obtained from real HYCAM data. NE = 12
transmission antennas. Tp = 100 μs. [1] J. Li and P. Stoica. MIMO Radar - Diversity means Superiority. In 14th
Annual Workshop on Adaptive Sensor Array Processing, 2006.
[2] J. Li and P. Stoica. MIMO Radar with Colocated Antennas. IEEE Signal
C. Results and analysis Processing Magazine, pages 106–114, 2007.
[3] C.Y. Chen and P.P. Vaidyanathan. MIMO Radar Ambiguity Properties
The same waveform families as the ones studied in section and Optimization Using Frequency-Hopping Waveforms. IEEE Trans.
III have been used for the experiments. The ambiguity func- on Signal Processing, 56(12):5926–59368, 2008.
[4] Y.I. Abramovich, G.J. Frazer, and B.A. Johnson. Noncausal Adaptive
tions measured from the acquired data are shown in figures 7, 8 Spatial Clutter Mitigation in Monostatic MIMO radar: Fundamental
and 9. They can be directly compared to theoretical figures 3, Limitations. IEEE Journal of Selected Topics in Signal Processing,
4 and 5 respectively. Despite some slightly higher sidelobes, it 4(1):40–54, 2010.
[5] E. Fishler, A. Haimovich, R.S. Blum, Jr. Cimini, L.J., D. Chizhik,
is visible that experimental ambiguity functions obtained from and R.A. Valenzuela. Spatial diversity in radars-models and detection
real data are close to the theoretical ones. In fact the discrep- performance. Signal Processing, IEEE Transactions on, 54(3):823 –
ancy is relatively small for the three waveform families, and 838, 2006.
[6] A. De Maio and M. Lops. Design principles of mimo radar detectors.
generally limited to some local changes in the sidelobe levels. Aerospace and Electronic Systems, IEEE Transactions on, 43(3):886 –
In particular their properties, such as range/angle coupling and 898, 2007.
sidelobe level, have been fairly well preserved. [7] G. San Antonio, D.R. Fuhrmann, and F.C. Robey. MIMO Radar
Ambiguity Functions. IEEE Journal of Selected Topics in Signal
V. C ONCLUSION Processing, 1(1):167–177, 2007.
[8] C.-Y. Chen. Signal Processing Algorithms for MIMO Radar. PhD thesis,
In this paper, we have studied the MIMO ambiguity func- California Institute of Technology, 2009.
tion. We have shown that, in its general form, it implies a [9] R. Gold. Optimal binary sequences for spread spectrum multiplexing.
IEEE Trans. on Information Theory, 13(4):619–621, 1967.
range/angle coupling, unless the transmitted waveforms are [10] H.L. Van Trees. Detection, Estimation and Modulation Theory, Part III.
perfectly orthogonal. Although this perfect orthogonality is John Wiley & Sons, 1971.

162