464 IEEE TRANSACTIONS ON SIGNAL PROCESSING

Inner Bounds on Performance of Radar and

Communications Co-Existence
Alex R. Chiriyath, Bryan Paul, Garry M. Jacyna, and Daniel W. Bliss, Fellow, IEEE

Abstract—We investigate methods of co-existence between A. Background

radar and communications systems. Each system typically con-
siders the other system a source of interference. Consequently, The increasing demand for spectrum and the need for efficient
the traditional solution is to isolate the two systems spectrally or spectrum sharing techniques is highlighted in [3]. It is worth
spatially. By considering a cooperative radar and communications noting that in our efforts presented here, we focus on radar es-
signaling scheme, we derive achievable bounds on performance timation performance as opposed to radar detection which was
for a receiver that observes communications and radar return in considered in [4]–[11]. To be more specific, in our work, we
the same frequency allocation. We assume the radar and com- have focused on the estimation of a dynamic target parameter,
munications operations to be a single joint system. Bounds on
performance of the joint system are measured in terms of data time delay or target range, from the received target return and
information rate for communications and a novel radar estimation the performance of the radar system is measured in terms of the
information rate for the radar. estimation rate.
Index Terms—Performance bounds, radar-communications The work presented in [4]–[7] investigated the application
co-existence. of information theory to improve radar system performance. In
those works, the idea that signal-to-noise ratio (SNR) does not
measure information is introduced. Maximizing SNR does not,
I. INTRODUCTION in general, maximize information. By using information theory,

T HERE is an ever increasing demand for spectrum and a new type of receiver, the a posteriori radar receiver, is devel-
given the limit on resources, communications and radar oped that does not try and maximize output SNR but attempts to
systems are increasingly encouraged to share bandwidth. This maximize the quantity of information, given by the a posteriori
can cause inter-system interference that degrades the perfor- distribution of a target parameter.
mance of both systems. The standard solution is to separate In [8], waveform optimization for detection and target infor-
(temporally, spatially or spectrally) the radar and communica- mation extraction are considered. The radar waveform is de-
tions systems. In this paper, we do not require this separation, signed so as to maximize the mutual information between the
and we explore the fundamental radar and communications target parameter of interest and the measurements obtained from
co-existence performance bounds. An important contribution the receiver. It is shown that the maximization of mutual in-
that enables this exploration is the novel parameterization of formation improves the radar system performance measured in
estimation information rate. The estimation information rate in- terms of target classification ability or average measurement
corporates the insights of rate distortion theory but emphasizes error. However, performance of the optimized waveforms in
the symmetry with the communications bound. In this paper, terms of target parameter estimation is not explicitly discussed.
we refine and extend the performance bounds introduced in In [12], information theory is utilized to develop a mutual in-
[1], as well as the additional bounds discussed in [2]. We also formation measure used for waveform and power spectrum de-
expand the results in [1] in greater detail. The two new inner sign to jointly optimize the performance of radar and commu-
bounds on performance discussed in this paper are the isolated nications systems that overlap in frequency. Similarly, the work
sub-band inner bound and the optimal Fisher information inner presented in [11] also uses information theory to develop an ex-
bound. pression for radar capacity (for radar systems performing target
detection only) which, in combination with traditional commu-
nications capacity, can be used to measure the total capacity of
Manuscript received October 25, 2014; revised June 22, 2015; accepted
September 08, 2015. Date of publication September 28, 2015. The associate
a joint radar-communications network.
editor coordinating the review of this manuscript and approving it for publi- Current research has investigated the benefits of using
cation was Dr. Fauzia Ahmad. This work was sponsored in part by DARPA methods similar to cooperative sensing to solve the problem
under the SSPARC program . The views expressed are those of the author and of radar and communications co-existence [13]–[18]. Radar
do not reflect the official policy or position of the Department of Defense or
the U.S. Government. nodes that employ some form of cooperative sensing have an
A. R. Chiriyath and D. W. Bliss are with the Bliss Laboratory of Information, improvement when compared to traditional nodes. In [13], it is
Signals, and Systems, Arizona State University, Tempe, AZ 85281 USA (e-mail: shown that co-existence between radar and communications is
[email protected]).
B. Paul is with the Bliss Laboratory of Information, Signals, and Systems, feasible for radar nodes that utilize cooperative sensing with very
Arizona State University, Tempe, AZ 85281 USA, and also with the General loose constraints on interference restrictions, such as low radar
Dynamics Mission Systems, Scottsdale, AZ. transmit power. Another approach is employed in [14] wherein
G. M. Jacyna is with The MITRE Corporation, McLean, VA 22102 USA.
Color versions of one or more of the figures in this paper are available online
the surveillance space of the radar system is divided into sectors
at http://ieeexplore.ieee.org. and priorities are assigned (using fuzzy logic) to all radar and
Digital Object Identifier 10.1109/TSP.2015.2483485 communications systems that want to transmit in each sector.

1053-587X © 2015 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/
redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
CHIRIYATH et al.: INNER BOUNDS ON PERFORMANCE OF RADAR AND COMMUNICATIONS CO-EXISTENCE 465

This paper is organized as follows. In Section II, we present

the channel and signal model for a joint radar-communications
system that is employed in this paper. We also introduce the
idea of an observed signal with a predicted radar return signal
suppressed. In Section III, we review the Cramér-Rao lower
bound on time-delay estimation. In Section IV, we develop the
radar estimation information rate. The estimation information
rate is the metric which we use to measure performance of the
radar system. In Section V, we present the multiple access com-
munications performance bound as an analogy to the bounds
on performance for a joint radar-communications system. In
Fig. 1. Joint radar-communications system. Section VI, we develop several inner bounds on the perfor-
mance of the joint radar-communications system by considering
various scenarios and developing estimation and data rates for
Techniques such as interference mitigation [19], precoding the radar and communications systems respectively. Finally, in
or spatial separation [20], waveform design [21]–[23] or wave- Section VII, we draw conclusions from the results obtained in
form shaping [24]–[27] allow both radar and communications this paper and discuss avenues for future research.
to share the spectrum and co-exist. In [24], waveform shaping
is done by projecting radar waveforms onto the null space of II. JOINT RADAR-COMMUNICATIONS CHANNEL MODEL
the interference channel matrix in an attempt to spatially mini- In this section, we consider the joint radar-communications
mize interference to the communications system from the radar system complex baseband received signal, , for a multiple-
system. access communications and radar return channel. We make the
Radar systems based on communications systems, where following assumptions:
radar systems use OFDM or DSSS signals as radar illumina- • Target cross-section is a nuisance parameter.
tion signals, have also been considered [28]–[32]. Similarly, • Target cross-section is estimated well.
communications systems using radar illumination signals • Targets are well separated.
like linear frequency modulation (LFM) chirp waveforms as • Residual of unpredicted radar return is modeled well by a
modulation signals to transmit data have been developed [33]. Gaussian distribution before pulse compression.
Finally, signal sharing, a method in which both radar and • Target range known, due to target tracking, up to some
communications systems utilize the same waveform has also Gaussian random process variation which is within one
been applied to the radar and communications co-existence over the bandwidth.
problem [34]–[36]. • Portion of time during which radar return and communica-
tions signal overlap is only considered.
B. Contributions For the sake of simplicity in this introductory paper, we focus
In this paper, the joint radar-communications system consists only on range estimation, though many extensions could be
of an active, mono-static, pulsed radar and a single-user com- made to other estimation parameters. We present a table of sig-
munications system. We assume that the radar system operates nificant notation that is employed in this paper in Table I.
without any maximum unambiguous range. We consider the
joint radar-communications receiver to be a radar transmitter/re- A. Radar Return Signal Model
ceiver that can act as a communications receiver. The joint re- Unless stated otherwise, we always assume that all signals are
ceiver can simultaneously estimate the radar target parameters in complex-baseband. While this may not seem pertinent now,
from the radar return and decode a received communications this becomes convenient later, when computing the information
signal. While the node architecture can easily be generalized to rates (estimation and data) of both radar and communications
function as a communications relay by including a communi- systems [37]. It is assumed that the received signal has been
cations transmitter, this is not explicitly discussed in this paper. passed through a brick-wall filter matched to the bandwidth of
We present a diagram of a jointly operating radar and commu- interest, .
nications system in Fig. 1. For targets, the observed complex baseband [37] radar re-
In this paper, we develop a new approach for producing joint turn in the presence of a communications signal and noise
radar-communications performance bounds. The principle con- is given by
tributions of this paper are as follows:
• Develop novel joint receiver formulation analogous to
(1)
communications multiple access channel.
• Develop radar estimation rate, a metric analogous to data
information rate. where complex combined antenna, cross-section, and propaga-
• Derive the isolated sub-band inner bound. tion gain for target, , is a parameter such that the radar
• Derive the successive interference cancellation inner range equation for received power for the target can be
bound. written as .
• Derive the communications water-filling inner bound. The zero-mean thermal noise is drawn from a complex
• Derive the optimal Fisher information inner bound. Gaussian distribution with variance . A
466 IEEE TRANSACTIONS ON SIGNAL PROCESSING

TABLE I the presence of the radar signal by using the predicted target
SURVEY OF NOTATION range to generate a predicted radar return and subtract it from
the received signal at the receiver.
For targets, the received signal at the communications re-
ceiver with the predicted radar return suppressed is given by

(4)

Note that we have assumed here that the estimated amplitude

is equal to the actual amplitude. This approach is only useful
if the error in delay is smaller that . For small fluctuations
in delay, we can replace the difference between the actual and
predicted radar return waveforms with a derivative,

(5)

The signal observed by the communications receiver is then

given by

(6)

The interference plus noise from the communications system’s

point of view is given by

reasonable time-delay estimator (particularly if targets are well

separated), assuming there is no communications interference,
is the correlation or matched filter estimator. As stated in
Section II, because we assume we are tracking the target, we (7)
have some knowledge of the target’s range (based upon prior
observations), up to some range fluctuation in the return due (8)
to an underlying target random process. This range fluctuation
is interpreted as a fluctuation in delay which is modeled by a
where is frequency, represents the frequency spectrum
Gaussian distribution . During the observation, the
of the unit variance radar illumination signal , and
delay for the target is given by,
comes from employing Parseval’s theorem to convert
into the frequency domain and then using the differen-
(2) tiation property of the Fourier transform [37]. is extracted
from bandwidth as follows
The function is a prediction function which de-
pends on , the pulse repetition interval, and a set of non- (9)
specific system and target parameters, . The variance of the
range fluctuation process is given by where the value is the scaling constant between and
times that is dependent upon the shape of the radar wave-
(3) form’s power spectral density. For a flat spectral shape,
.
Once the receiver has decoded the communications signal, it
B. Communications Signal With Predicted Radar Return can be removed from the observed waveform to obtain the orig-
Suppressed inal radar return signal free of any communications interference.
In order to improve the performance of the communications This technique is known as Successive Interference Cancella-
system, we try to mitigate unnecessary interference caused by tion (SIC) [1]. We assume that SIC is employed by the joint
CHIRIYATH et al.: INNER BOUNDS ON PERFORMANCE OF RADAR AND COMMUNICATIONS CO-EXISTENCE 467

radar-communications receiver whenever there is any overlap the radar and bits per pulse repetition interval , the
between the radar and communications signals. radar estimation information rate is bounded by

III. CRAMÉR-RAO LOWER BOUND FOR (13)

TIME-DELAY ESTIMATION
In this section, we develop the Cramér-Rao lower bound on where is the received signal entropy and is the
time-delay estimation on a SISO (single-input single-output) estimation entropy.
channel with circularly symmetric Gaussian noise [38]. This is The received signal entropy of the radar or the entropy of the
a standard result that is presented here as exposition to connect process uncertainty plus estimation uncertainty, assuming that
to the rest of the paper. The Cramér-Rao bound gives the best both are Gaussian, is given by [37], [39]
performance (in terms of variance of estimation error) of an un-
biased estimator. (14)
For a single target with delay and combined radar cross-
section, antenna and propagation gain , we assume that the To find the estimation entropy, we find the delay estimation un-
received signal of the time-delay estimator is given by certainty for each target. Under the assumption of Gaussian es-
timation error, the resulting entropy of the error is given by
(10)

where is the transmitted signal with power whose fre-

quency representation, has full bandwidth , (15)
is the delayed version of the transmitted signal and, is cir-
cularly symmetric Gaussian noise with zero mean and variance where the variance of delay estimation for the target is
. given by (12).
Let be the parameter to be estimated. From (10), we Finally, after putting it all together, we see that the radar esti-
see that and has the following mation information rate is given by
probability density function,

(11)

The Cramér-Rao lower bound for time delay estimation [38],

, is given by

(12) (16)

where ISNR stands for integrated SNR and is given by It is worth noting, that by employing this estimation entropy
. By centering the spectrum at an appropriate point (by in the rate bound, it is assumed that the estimator achieves the
choosing the origin of the spectrum), we get the RMS bandwidth Cramér-Rao performance. If the error variance is larger, then
, given by (8). the rate bound is lowered.

IV. RADAR ESTIMATION INFORMATION RATE V. MULTIPLE-ACCESS COMMUNICATIONS

PERFORMANCE BOUND
Here we develop a novel parameterization of the radar in
terms of information rate, the ‘radar estimation rate.’ The metric We present the multiple-access communications system
is analogous to data information rate for the communications performance bound [37], [39] as motivation to develop inner
system. If radar illumination on a target can be viewed as the bounds on the performance of a joint radar-communications
target unwillingly communicating information on its parame- system [1]. In this scenario, the channel propagation gain for
ters (range, cross-section etc.), the radar channel can be char- the first communications system is given by and channel
acterized by a uncooperative communications channel and the propagation gain for the second communications system is
estimation rate can be viewed as a mutual information between given by . The power of the first communications transmitter
the radar and the target. We construct this information rate by is denoted by and the power of the second communications
considering the entropy of a random parameter being estimated transmitter is given by . Their corresponding rates are de-
and the entropy of the estimation uncertainty of that parameter noted and . Assuming that the noise variance is given by
[1]. As mentioned earlier, this estimation information rate em- , the fundamental limits on rate are given by
ploys the insights of rate distortion theory [39] and highlights
the symmetry with the communications information rate bound.
As an observation, if the targets are well separated, then each
target estimation can be considered an independent information
channel.
Motivated by the mutual information rate (or radar estimation
(17)
rate) in terms of estimation entropy, random process entropy of
468 IEEE TRANSACTIONS ON SIGNAL PROCESSING

Fig. 3. Joint radar-communications system block diagram for SIC scenario.

radar pulse duration would cause the radar pulse duration to

exceed the pulse repetition interval of the radar system, i.e.,
, which would render the radar system unable to func-
tion correctly.
Fig. 2. Pentagon containing Communications Multiple-Access Achievable
Rate Region.
A. Isolated Sub-Band Inner Bound
Vertices are found by jointly solving two bounds to get, In this section, we derive an inner bound by considering a
scenario in which we partition the total bandwidth into two
sub-bands, one for radar only and the other for communica-
tions. Each system functions without any interference in their
respective sub-band. This is the traditional solution to the joint
(18)
radar-communications co-existence problem.
The total bandwidth is split between the two sub-bands ac-
The other vertex can be found by switching the subscripts 1 and cording to some such that,
2 in (18). The region that satisfies these theoretical bounds is
depicted in Fig. 2.
The achievable rate region is obtained by taking the convex The corresponding communications rate (for the communi-
hull [40] of the vertices 1–4. Because a radar signal is not de- cations only sub-band) is given by
rived from a countable dictionary, the fundamental assumption
of a communications signal is violated, and the bounds pre-
sented here can not be achieved by a joint radar-communica-
tions system. The result presented in this section can be ex- (19)
tended for more than two communications systems. For dif-
ferent communications systems, the resultant achievable rate re-
and the corresponding radar estimation rate is given by
gion is a -dimensional polytope [39].

VI. PERFORMANCE BOUNDS OF A JOINT

RADAR-COMMUNICATIONS SYSTEM
In this section, we derive inner bounds on the performance (20)
of the joint radar-communications system. As mentioned ear-
lier, performance is measured in data information rate for the
communications system and estimation information rate for the B. Successive Interference Cancellation (SIC) Inner Bound
radar system. To find these inner bounds, we hypothesize an ide- We consider a scenario in which the joint radar-communi-
alized receiver and determine the bounding rates. To simplify cations system first suppresses the predicted radar return and
the discussion, we consider only a single radar target with delay then attempts to decode the communications signal. After the
and gain-propagation-cross-section product . receiver has decoded the communications signal, it can remove
Additionally, for the sake of simplicity, we assume that the communications signal from the observed waveform. We
there is only a single communications link present. The results can then obtain the original radar return signal free of any com-
derived in this section can be extended for more than the single munications interference. This sequential interference mitiga-
communications channel link. Similar to the multi-access tion technique is called SIC. An achievable inner bound on joint
channel case described in Section V, this extension results in an radar-communications system performance can be derived by
-dimensional polytope, for communications users taking the convex hull of all achievable communications and es-
and a single radar case. timation rate pairs, the SIC inner bound. The block diagram of
Furthermore, for the derivations of the bounds presented in the joint radar-communications system considered in this sce-
this section, the radar pulse duration is held constant. In nario is shown in Fig. 3.
some scenarios, this implies that the time-bandwidth product of If , it is as if the radar interference is not present
the radar system is not constant. In [1], a case was made for and the communications system can operate at a data rate deter-
ensuring that the time-bandwidth product of the radar system mined by the isolated communications bound,
would be fixed as constant which meant that the radar pulse du-
ration would not be constant. This would cause the duty-factor
of the radar system to vary as well, which is not a desirable
feature for radar systems. Furthermore, in some cases, a varying (21)
CHIRIYATH et al.: INNER BOUNDS ON PERFORMANCE OF RADAR AND COMMUNICATIONS CO-EXISTENCE 469

There are two effective channels,

(25)

for the communications only channel and

(26)

for the mixed use channel. The communications power is split

Fig. 4. Joint radar-communications system block diagram for communications
only and mixed use sub-bands. between the two channels [37], [39],

If is sufficiently low for a given transmit power, then as (27)

described above, the receiver can successfully decode the com-
munications signal and remove it from the observed waveform,
leaving just the radar return. Thus, the radar parameters, such as where if ; otherwise .
target range, can be estimated without corruption from any out- The critical point (the transition between using one or both
side interference. This implies that from the communications channels for communications) occurs when
receiver’s perspective, it observes interference plus noise as de-
scribed by (7), and the corresponding communications rate is
given by
(28)

so both channels are used if

(22) (29)

In this regime, the corresponding estimation rate bound is If the communications-only channel is used exclusively for
given by (16). communications, then . If both channels are em-
The vertices formed by (21), (22) and (16) correspond to the ployed for communications then
points 2, 3 and 4 in Fig. 2, assuming that is the estimation
rate, and is the communications rate. An achievable rate lies
within the quadrilateral constructed by constructing the convex
hull between these points. This is the SIC inner bound. (30)

C. Communications Water-Filling Inner Bound and thus when (29) is satisfied,

In this section, we consider a scenario in which the total band-
width is split into two sub-bands, one sub-band for communi-
cations only and the other sub-band for both radar and com-
munications. It is not necessary that the sub-bands be of equal (31)
bandwidth. We use a novel water-filling approach to distribute
the total communications power between the two sub-bands [1]. The value of power fraction is then given by
Water-filling optimizes the power and rate allocation between
multiple channels [37], [39]. In this scenario, the bandwidths
of the two channels need not be equal. This means that the
problem formulation in this scenario is not a standard formula-
tion. Hence, we expect that the shape of the inner-bound derived
by employing water-filling to be non-intuitive. The mixed use
channel operates at the SIC rate vertex defined by (16) and (22).
The block diagram of the joint radar-communications system
considered in this scenario is shown in Fig. 4.
(32)
Given some that defines the bandwidth separation,

(23) The resulting communications rate bound in the communica-

tions-only sub-band is given by
we optimize the power utilization, , between sub-bands,
(33)
(24)
470 IEEE TRANSACTIONS ON SIGNAL PROCESSING

The mixed use communications rate inner bound is given by

(34)

The corresponding radar estimation rate inner bound is then

given by

(35) Fig. 5. Joint radar-communications system block diagram for radar only and
mixed use sub-bands.

D. Optimal Fisher Information Inner Bound

In this section, we construct another inner rate bound by con- where ,
sidering an approach similar to the one utilized in the previous is the communications signal that is only present in the
section. We split the total bandwidth into two sub-bands and dis- mixed use channel, is complex additive
tribute the radar power (or power spectral density) between the white Gaussian noise (AWGN) in the radar only sub-band and
two sub-bands (instead of the communications power) in a way is complex AWGN in the mixed
that minimizes the Cramér-Rao lower bound (or maximizes the use sub-band and is the variance of additive
Fisher information) on the variance of a time-delay estimator. white Gaussian noise in a channel that was the sum of both sub
Hence, we have one channel that has radar only and the other channels.
channel has both communications and radar. The block diagram Using this return signal, we derive the Cramér-Rao lower
of the joint radar-communications system considered in this sce- bound on the variance for joint time-delay estimation. Let
nario is shown in Fig. 5. The bandwidth is split between the two be the parameter to be estimated. From (40), we see that
sub-bands according to some such that, and has the following
probability density function,
(36)

and we optimize the power spectral densities, and , (41)

utilized by the radar only and mixed use sub-bands respectively,
The corresponding log-likelihood function is given by
to maximize the Fisher information, where

(42)
(37)
We have the following constraints on power and energy of the and the score function, is given by
radar system in the two sub-channels,
(38)
(39) (43)
where c.c. stands for complex conjugate term and
Now, consider a radar signal with bandwidth B, whose fre- . Now, the Fisher information for this estimation
quency spectrum is centered around . We assume that problem, , is given by
is spectrally flat. We now partition the frequency spec-
trum into two portions, and with bandwidths
and respectively, thereby creating two new sig-
nals, and which is used in transmissions in the
radar only sub-band and mixed use sub-band respectively. Be-
cause is spectrally flat, this implies that both and
are spectrally flat as well. This partitioning in the fre-
quency domain also makes the two signals orthogonal in fre-
quency.
Thus, after transmission, the radar receiver observes the fol- On simplification, we see that,
lowing return signal,
(44)

where the cross-terms in the product become 0 due to

(40) and the independence of and . The factor of two
CHIRIYATH et al.: INNER BOUNDS ON PERFORMANCE OF RADAR AND COMMUNICATIONS CO-EXISTENCE 471

comes from the complex conjugate term. Using the fact that a time-delay estimator is obtained by plugging in (46) and (48)
and simplifying, we see that, into (45) and taking its derivative with respect to , setting
the resultant equation to 0 and solving for . is obtained
in a similar way, except that (47) is used in (45) instead of (48).
and are given by the following equations,

By multiplying the terms out, converting to frequency domain

and applying Parseval’s Theorem, the time-shift and differentia- (49)
tion properties of the Fourier Transform and the orthogonality of
and , for spectrally flat and ,
we get

(50)

where

The resultant estimation rate bound for the radar system in both
sub-channels is given by

to finally get (51)

where is the variance of the time-delay estimation given by

and is given by (45). The corresponding commu-
nications rate bound in the mixed use channel is

(45)

We consider to be a free parameter. We select a value (52)

of by looking at the reduced Fisher information [37], [41]
for time-delay estimation derived from the Fisher information We expect the resulting inner bound to have end points given
matrix of joint amplitude and time-delay estimation. We set the by (16) and (22) (SIC vertex) when and by (16) when
value of such that the regular Fisher information for time- .
delay estimation, given by (45), and the reduced Fisher informa-
tion for time delay estimation are equal. In general, the reduced E. Examples
Fisher information is given by (58). As shown in Appendix A, In Fig. 6, we display an example of the inner bounds on per-
the resultant value for is given by formance. The parameters used in the example are displayed
in Table II. It is assumed that the communications signal is re-
(46) ceived through an antenna sidelobe, so that the radar and com-
munications receive gain are not identical. In general, the inner
bound is produced by the convex hull of all contributing inner
From (38), we see that,
bounds.
For the optimal Fisher information bound, while optimizing
(47) the distribution of radar power between the two sub-channels,
it was found that the power becomes complex for and
(48) . In order to get an inner-bound on rate over all values
of , the power in each sub-band has been set linearly for
The value of the power spectral density utilized by the radar and such that the total power used by both
only sub-band, , that maximizes the Fisher information for sub-channels at value is always the total radar power, .
472 IEEE TRANSACTIONS ON SIGNAL PROCESSING

multiple access channel was developed. We developed an

estimation information rate for time-delay estimation for a
radar system which is then used to evaluate performance.
Various inner bounds on performance such as the SIC inner
bound, the isolated sub-band inner bound, the communications
water-filling inner bound and the optimal Fisher information
inner bound were developed for a joint receiver. There are a
range of potentially interesting scenarios to which these bounds
may be applied. Given a set of parameters and constraints
on the radar and communications systems, we can use the
derived inner bounds to design appropriate joint systems. In the
case where we have full control over all parameters for both
systems, we can simply calculate the convex hull of all inner
Fig. 6. Data rate and estimation rate bounds for parameters in Table II.
bounds and, depending on the requirements of the system, we
can choose which region to operate in, thus also choosing the
TABLE II algorithm that will be implemented by the system. In the case
PARAMETERS FOR EXAMPLE PERFORMANCE BOUND #1 where we have no control over some of a system’s parameters,
we can design a joint system based on just the convex hull of
an appropriate subset of the derived inner bounds. This gives a
unique profile of all possible data and estimation rates and we
can then choose which region to operate in, depending on the
requirements of the system.

APPENDIX A
DERIVATION OF REDUCED FISHER INFORMATION FOR
TIME-DELAY ESTIMATION
In this section, we first derive the Fisher information cross
terms for joint amplitude and time-delay estimation and find the
value of the free parameter that sets these cross-terms to 0.
We then show that by setting the cross-terms to 0, the reduced
In Fig. 6, we indicate in green, the bound on successive in- Fisher information for time-delay estimation is the same as the
terference cancellation (SIC), presented in (22). The best case Fisher information for time-delay estimation, given by (45).
system performance given SIC is at the vertex (at the intersec- We consider the same scenario as described in Section VI-D.
tion of the green and gray lines in Fig. 6), which is determined The total bandwidth is split into two sub-bands and the radar
by the joint solution of (22) and (16). The inner bound that lin- power (or power spectral density) is distributed between the two
early interpolates between this vertex and the radar-free com- sub-bands. The bandwidth and radar power (power spectral den-
sities) are split between the two sub-bands according to some .
munications bound in (21) is indicated by the gray dashed line.
Now, consider a radar signal with bandwidth , whose
The water-filling bound is indicated by the blue line. The water-
frequency spectrum is flat and centered around .
filling bound is not guaranteed to be convex. The water-filling
and are the spectrally orthogonal sub-band
bound is not guaranteed to be greater than the linearly interpo-
signals with bandwidths and respectively.
lated bound. The isolated sub-band inner bound is indicated by
Thus, the joint receiver observes the following return signal
the brown line and the optimal Fisher information bound is in-
dicated by the black line.
In the example, we see that the water-filling bound exceeds (53)
the linearly interpolated bound and all other inner bounds. We
also see that the optimal Fisher information bound is always where ,
lower than the water-filling bound and the linearly interpolated is the communications signal that is present in the mixed use
SIC bound. The optimal Fisher information bound can either ex- channel and is circularly symmetric Gaussian noise with
ceed the isolated sub-band bound or be lower than the isolated zero mean and variance .
sub-band bound depending on the value of used. As men-
Let be the parameters to be estimated. From (53),
tioned in Section VI-C, the shape of the water-filling curve is
non-intuitive. Finally, we see that the end points of the optimal we see that and has
Fisher bound are as expected. the following score function, is given by

VII. CONCLUSION
In this paper, we provided a novel approach for producing
joint radar-communications performance bounds. A unique (54)
joint receiver signal model similar to the communications
CHIRIYATH et al.: INNER BOUNDS ON PERFORMANCE OF RADAR AND COMMUNICATIONS CO-EXISTENCE 473

where c.c. stands for the complex conjugate term and reduced Fisher Information [37], [41] for time-delay estimation
. Now, the Fisher Information Matrix for this is
estimation problem, , is given by

(58)

REFERENCES
[1] D. Bliss, “Cooperative radar and communications signaling: The esti-
mation and information theory odd couple,” in Proc. IEEE Int. Radar
Conf., May 2014, pp. 50–55.
[2] A. Chiriyath, “Joint radar-communications performance bounds: Data
versus estimation information rates,” M.S. thesis, Dept. of Electrical,
We now simplify the cross terms of the Fisher information Computer and Energy Engineering, Arizona State Univ., Tempe, AZ,
matrix. Starting with , we see that on simplification, USA, 2014.
[3] H. Hayvaci and B. Tavli, “Spectrum sharing in radar and wireless com-
munication systems: A review,” in Proc. Int. Conf. IEEE Electromagn.
Adv. Appl. (ICEAA), Aug. 2014, pp. 810–813.
[4] P. Woodward and I. Davies, “A theory of radar information,”
Philosoph. Mag. Series 7, vol. 41, no. 321, pp. 1001–1017, 1993.
By multiplying the terms out, converting to frequency domain [5] P. Woodward, “Information theory and the design of radar receivers,”
and applying Parseval’s Theorem and the time-shift and differ- Proc. IRE, vol. 39, no. 12, pp. 1521–1524, Dec. 1993.
[6] P. M. Woodward, Probability and Information Theory: With Applica-
entiation properties of the Fourier Transform, for spectrally flat tions to Radar. Norwood, MA, USA: Artech House, 1953.
(or and ), we get [7] P. Woodward, “Radar ambiguity analysis,” RRE Tech. Note, no. 731,
Feb. 1967.
[8] M. Bell, “Information theory and radar waveform design,” IEEE Trans.
Inf. Theory, vol. 39, no. 5, pp. 1578–1597, Sep. 1993.
(55) [9] S. U. Pillai, H. S. Oh, D. C. Youla, and J. R. Guerci, “Optimal transmit-
receiver design in the presence of signal-dependent interference and
channel noise,” IEEE Trans. Inf. Theory, vol. 46, no. 2, pp. 577–584,
Applying the definition of and simplifying, we get Mar. 2000.
[10] D. A. Garren, M. K. Osborn, A. C. Odom, J. S. Goldstein, S. U. Pillai,
and J. R. Guerci, “Enhanced target detection and identification via op-
timised radar transmission pulse shape,” IEEE Proc.—Radar, Sonar,
Navig., vol. 148, no. 3, pp. 130–138, Jun. 2001.
[11] J. Guerci, R. Guerci, A. Lackpour, and D. Moskowitz, “Joint design and
operation of shared spectrum access for radar and communications,” in
Proc. IEEE Radar Conf., May 2015, pp. 761–767.
[12] A. Turlapaty and Y. Jin, “A joint design of transmit waveforms for
radar and communications systems in coexistence,” in Proc. IEEE
Radar Conf., May 2014, pp. 315–319.
[13] L. S. Wang, J. P. McGeehan, C. Williams, and A. Doufexi, “Applica-
tion of cooperative sensing in radar-communications coexistence,” IET
Commun., vol. 2, no. 6, pp. 856–868, Jul. 2008.
[14] S. S. Bhat, R. M. Narayanan, and M. Rangaswamy, “Bandwidth
sharing and scheduling for multimodal radar with communications and
tracking,” in Proc. IEEE Sensor Array Multichannel Signal Process.
Workshop, Jun. 2012, pp. 233–236.
[15] M. Fitz, T. Halford, and I. H. S. Enserink, “Towards simultaneous radar
(56) and spectral sensing,” in Proc. IEEE Int. Symp. Dynam. Spectrum Ac-
cess Netw. (DYSPAN), Apr. 2014, pp. 15–19.
[16] R. Saruthirathanaworakun, J. M. Peha, and L. M. Correia, “Oppor-
Similarly, using the same properties as mentioned above, on tunistic sharing between rotating radar and cellular,” IEEE J. Sel. Areas
Commun., vol. 30, no. 10, pp. 1900–1910, 2012.
simplifying the other cross term in the Fisher information ma- [17] F. Paisana, J. Miranda, and N. M. L. Dasilva, “Database-aided sensing
trix we see that for radar bands,” in Proc. IEEE Int. Symp. Dynam. Spectrum Access
Netw. (DYSPAN), Apr. 2014, pp. 1–6.
[18] H. Wang, J. Johnson, C. Baker, L. Ye, and C. Zhang, “On spectrum
sharing between communications and air traffic control radar systems,”
in Proc. IEEE Radar Conf., May 2015, pp. 1545–1551.
In order to find the value of that sets the Fisher informa- [19] H. Deng and B. Himed, “Interference mitigation processing for spec-
tion matrix cross-terms and to 0, we set trum-sharing between radar and wireless communications systems,”
and solve for . The resultant value for is IEEE Trans. Aerosp. Electron. Syst., vol. 49, no. 3, pp. 1911–1919, Jul.
2013.
[20] A. Babaei, W. Tranter, and T. Bose, “A practical precoding approach
for radar/communications spectrum sharing,” in Proc. Int. Conf.
(57) Cognit. Radio Orient. Wireless Netw. (ICST), Jul. 2013, pp. 13–18.
[21] A. Aubry, A. D. Maio, M. Piezzo, and A. Farina, “Radar waveform
design in a spectrally crowded environment via nonconvex quadratic
This means that the Fisher information cross terms are be 0 optimization,” IEEE Trans. Aerosp. Electron. Syst., vol. 50, no. 2, pp.
whenever the value of is given by (57). In this case, the 1138–1152, Apr. 2014.
474 IEEE TRANSACTIONS ON SIGNAL PROCESSING

[22] A. Aubry, A. D. Maio, M. Piezzo, M. Naghsh, M. Soltanalian, and P. Bryan Paul received the B.S. degree (Highest
Stoica, “Cognitive radar waveform design for spectral coexistence in Honors) in electrical engineering from the University
signal-dependent interference,” in Proc. IEEE Radar Conf., May 2014, of Illinois at Urbana-Champaign in 2010, and the
pp. 474–478. M.S. degree in electrical engineering from Arizona
[23] K.-W. Huang, M. Bica, U. Mitra, and V. Koivunen, “Radar waveform State University in 2014. He is currently pursuing
design in spectrum sharing environment: Coexistence and cognition,” the Ph.D. degree in electrical engineering from
in Proc. IEEE Radar Conf., May 2015, pp. 1698–1704. Arizona State University.
[24] S. Sodagari, A. Khawar, T. Clancy, and R. McGwier, “A projection- From 2002 to 2008, he was enlisted in the Illinois
based approach for radar and telecommunication systems coexistence,” Air National guard completing multiple tours of duty
in Proc. IEEE Global Commun. Conf., Dec. 2012, pp. 5232–5236. in support of Operation Enduring Freedom and Op-
[25] S. C. Surender, R. M. Narayanan, and C. R. Das, “Performance anal- eration Iraqi Freedom. From 2010 to 2012, he was
ysis of communications and radar coexistence in a covert UWB OSA with Validus Technologies, Peoria, Illinois, working in the area of embedded
system,” in Proc. IEEE Global Telecommun. Conf., Dec. 2010, pp. 1–5. software and digital signal processing. He is currently with General Dynamics
[26] A. Khawar, A. Abdel-Hadi, and T. Clancy, “Spectrum sharing between Mission Systems (legacy General Dynamics Advanced Information Systems)
S-band radar and LTE cellular system: A spatial approach,” in Proc. in Scottsdale, Arizona, working in the area of digital signal processing and
IEEE Int. Symp. Dynam. Spectrum Access Netw. (DYSPAN), Apr. 2014, systems. His current research interests include signal processing, information
pp. 7–14. theory, radar, and communications.
[27] A. Khawar, A. Abdel-Hadi, and T. Clancy, “MIMO radar waveform Bryan has received numerous military decorations, including the Humani-
design for coexistence with cellular systems,” in Proc. IEEE Int. Symp. tarian Service Medal. He is a co-inventor of one awarded U.S. patent.
Dynam. Spectrum Access Netw. (DYSPAN), Apr. 2014, pp. 20–26.
[28] D. Garmatyuk, Y. Morton, and X. Mao, “On co-existence of in-band
UWB-OFDM and GPS signals: Tracking performance analysis,” in
Proc. IEEE/ION Position, Location, Navig. Symp., May 2008, pp.
196–202. Garry M. Jacyna received the B.S. degree in
[29] S. Blunt, P. Yatham, and J. Stiles, “Intrapulse radar-embedded com- physics in 1973 and the M.S. and Ph.D. degrees
munications,” IEEE Trans. Aerosp. Electron. Syst., vol. 46, no. 3, pp. in mathematics in 1974 and 1977, respectively, all
1185–1200, Jul. 2010. from Rensselaer Polytechnic Institute, Troy, NY.
[30] S. Gogineni, M. Rangaswamy, and A. Nehorai, “Multi-modal OFDM Prior to joining MITRE in 1987, he was employed
waveform design,” in Proc. IEEE Radar Conf., May 2013, pp. 1–5. at UNISYS, Reston, VA, performing EW system
[31] D. Garmatyuk, J. Schuerger, K. Kauffman, and S. Spalding, “Wide- studies including wideband detection, robust lo-
band OFDM system for radar and communications,” in Proc. IEEE calization, direction finding, and signal parameter
Veh. Technol. Conf., May 2009, pp. 1–6. identification and characterization. He was with
[32] S. Thompson and J. Stralka, “Constant envelope OFDM for power-ef- Planning Systems Incorporated, McLean, from 1977
ficient radar and data communications,” in Proc. Int. Waveform Divers. to 1984 as a senior systems analyst responsible
Des. Conf., Feb. 2009, pp. 291–295. for the analysis and review of advanced sonar equipment. He is currently
[33] M. Roberton and E. Brown, “Integrated radar and communications a MITRE fellow of the MITRE Corporation, McLean, VA. His specialties
based on chirped spread-spectrum techniques,” in IEEE MTT-S Int. Mi- include analytical-based performance studies for DoD and DHS programs as
crow. Symp. Dig, Jun. 2003, vol. 1, pp. 611–614. well as extensive research in the areas of sonar system performance modeling,
[34] B. J. Donnet and I. D. Longstaff, “Combining MIMO radar with OFDM artificial neural networks, wavelet detection and estimation, higher order spec-
communications,” in Proc. 3rd Eur. Radar Conf., Sep. 2006, pp. 37–40. tral analysis, adaptive beamforming, nonlinear control theory, and nonlinear
[35] C. Sturm, T. Zwick, and W. Wiesbeck, “An OFDM system concept signal and noise modeling. He has designed complexity-based analysis tools
for joint radar and communications operations,” in Proc. IEEE Veh. for Marine Corps agent-based simulation models and distributed detection,
Technol. Conf., Apr. 2009, pp. 1–5. classification, and tracking algorithms for netted sensor systems. He was also
[36] C. Sturm and W. Wiesbeck, “Waveform design and signal processing an Adjunct Assistant Professor of electrical engineering at Catholic University,
aspects for fusion of wireless communications and radar sensing,” Washington, DC, teaching in the areas of communication theory, stochastic
Proc. IEEE, vol. 99, no. 7, pp. 1236–1259, Jul. 2011. processes, sonar signal processing, detection and estimation theory, and neural
[37] D. W. Bliss and S. Govindasamy, Adaptive Wireless Communications: networks.
MIMO Channels and Networks. Cambridge, U.K.: Cambridge Univ.
Press, 2013.
[38] M. A. Richards, J. A. Scheer, and W. A. Holm, Principles of Modern
Radar: Basic Principles. Raleigh, NC, USA: SciTech Publishing, Daniel W. Bliss (F’15) received his BSEE in Elec-
2010. trical Engineering from Arizona State University
[39] T. M. Cover and J. A. Thomas, Elements of Information Theory, 2nd in 1989 and his M.S. in Physics and Ph.D. and
ed. New York, NY, USA: Wiley, 2006. from the University of California at San Diego in
[40] M. de Berg, C. Otfried, M. van Kreveld, and M. Overmars, Compu- 1995 and 1997, respectively. Employed by General
tational Geometry: Algorithms and Applications, 3rd ed. New York, Dynamics from 1989 to 1993, he designed rocket
NY, USA: Springer, 2008. avionics and performed magnetic field calculations
[41] D. F. Delong, “Multiple signal direction finding with thinned linear and optimization for high-energy particle-accel-
arrays,” MIT Lincoln Lab., Lexington, MA, USA, Tech. Rep. TST-68, erator superconducting magnets. His doctoral
1983, DTIC:ADA128924. work (1993–1997) was in the area of high-energy
particle physics. He was a senior member of the
technical staff at MIT Lincoln Laboratory from 1997 to 2012. He is currently
an Associate Professor in the School of Electrical, Computer and Energy
Alex R. Chiriyath received his BSEE degree (Cum Engineering at Arizona State University. His current research topics include
Laude) in electrical engineering from the University multiple-input multiple-output (MIMO) wireless communications, MIMO
of Michigan at Ann Arbor in 2012 and his M.S. radar, cognitive radios, radio network performance, geolocation, and statistical
degree in electrical engineering from Arizona State signal processing for anticipatory physiological analytics. Dan has been the
University in 2014. He is currently pursuing his principal investigator on numerous programs with applications to radio, radar,
Ph.D. degree in electrical engineering from Arizona and medical monitoring. He has made significant contributions to robust mul-
State University. tiple-antenna communications including theory, patents, and the development
of advanced prototypes. He is responsible for some of the foundational MIMO
radar literature.