Joint Design and Operation of Shared Spectrum

Access for Radar and Communications

J. R. Guerci and R. M. Guerci A. Lackpour and D. Moskowitz
Cognitive Technology Division Advanced Technology Laboratories
Guerci Consulting LLC Lockheed Martin Corporation
Arlington, VA USA Cherry Hill, NJ USA

Abstract—A new theoretical foundation for the joint design optimization for the general constrained optimization case. The
and operation (JDO) of shared spectrum access for radar and framework makes clear the need for, and dependency on,
communications (SSPARC) is presented. The JDO SSPARC multidimensional channel information. To that end, in Section
framework entails advanced radar-comms channel estimation, IV, a comprehensive framework for advanced JDO SSPARC
along with a real-time adaptive space-time transmit and receive channel estimation is outlined that builds on advances in both
optimization procedure to maximize forward channel signal-to- knowledge-aided (KA) processing, and the latest cognitive
noise while simultaneously minimizing co-channel interference. radar/comms technologies [2-5]. Lastly, in Section V, a high
Additionally, a new expression for radar capacity is introduced fidelity and site-specific electromagnetic (EM) propagation
that when combined with traditional communication capacity
simulation and analysis is conducted to gauge the potential
provides a unified measure of the total capacity of the combined
performance improvements achievable in a realistic setting as
radar-comms network. High fidelity site-specific electromagnetic
radiation propagation simulations are conducted to provide a compared to conventional (non-optimized) approaches.
sense of the real-world potential gains achievable with JDO
II. A NEW MEASURE OF RADAR CAPACITY
SSPARC as compared with non-optimized approaches.
Radar is most assuredly a special case of a communication
Keywords—DARPA, SSPARC, combined radar and system. So it stands to reason that a measure of capacity
communications, radar capacity, adaptive waveforms, channel directly analogous to that developed by Hartley and Shannon
estimation, space-time optimization, knowledge-aided (KA) should not be too much of a stretch [6]. Though Woodward’s
processing, cognitive radar, cognitive radio landmark book examined radar from an information theoretic
framework (and introduced the ambiguity function along the
I. INTRODUCTION way!), it did not develop a measure of capacity directly
The increasing global demand for useful radio frequency analogous to that of comms [7]. Rather, a general measure of
(RF) spectrum by wireless communications has led to a the information gain from a measurement was developed. In
conflict with traditional primary users of legacy spectrum such this section an intuitively appealing and relatively simple
as radar. This has prompted new research into new approaches measure of radar capacity is introduced.
to potentially sharing spectrum between radar and
communications. To that end, the Defense Advanced Research For an MTI radar, one can consider each independent
Projects Agency (DARPA) has created a major new project resolution cell (e.g., range-angle-Doppler) as a binary
titled Shared Spectrum Access for Radar and Communications information storage unit, i.e., “0” = target absent, “1” = target
(SSPARC) [1]. The overall scope of the SSPARC program present. Viewed in this manner, the maximum capacity of an
includes techniques that may be incorporated by existing MTI radar performing a periodic search is given by the Hartley
systems with minimal changes, to the most advanced spectrum capacity measure [6]:
sharing approaches that entail a “back to the drawing board” CR = log N (1)
approach. The JDO SSPARC approach described herein is an
example of the latter. An additional goal of the DARPA where N is the total number of independent MTI resolution
SSPARC program is to not just share spectrum, but use cells with a minimum prescribed signal-to-noise-ratio (SNR)
spectrum in a much more efficient way that results in improved for an assumed target RCS, denoted as SNR min , and is given
performance for both radar and communications. by
To that end, this paper introduces a new and fundamental
approach to both the joint design and operation of a radar-  2π   PRF 
N ∝ ( BRmax )    (2)
comms (RC) SSPARC network. In Section II, a new  ∆θ   ∆f D 
expression for the radar “capacity” is derived that allows for a
joint radar-comms capacity measure for the entire network. In where where ∆θ and ∆f D denote the angle (e.g., azimuth) and
Section III, a mathematical framework for the joint Doppler resolution respectively, B is the operating bandwidth,
optimization of an N-node RC network is developed and is
shown to result in generalized eigenvalue problem for the and Rmax is the maximum range with SNR equal to SNR min ,
unconstrained case, or a constrained Rayleigh quotient and can be derived from the so-called radar range equation [8]

This research was developed with funding from the Defense Advanced
Research Projects Agency (DARPA). The views, opinions, and/or findings
contained in this article/presentation are those of the author(s)/presenter(s)
and should not be interpreted as representing the official views or policies of
978-1-4799-8232-5/151$31.00@2015IEEE 0761
the Department of Defense or the U.S. Government." This is in accordance
with DODI 5230.29, current edition, contract # HR0011-14-C-0142.
(consistent units assumed). Note that in (2) we have assumed and communications performance. If either or both the radar
the radar scans 360-deg azimuthal ( 2π radians). This can and communication systems are capable of performing both
obviously be modified to suit the specific application. Rmax radar and communications functions then the joint capacity
expression in (5) contains and even greater degree of coupling
can in turn be expressed in terms of SNR min and the total
than just the co-channel noise.
noise σ n2 , namely, In the next section we introduce a theoretical framework for
optimizing the transmit-receive functions of an N-node RF
κ SSPARC network. The resulting design equations provide a
Rmax = (3)
σn ⋅4 SNR min tight bound on theoretical performance (as opposed to
generally approximate methods such as Cramer-Rao [9]).
where κ is a fixed constant of proportionality that subsumes
all of the radar range equation parameters (antenna gain, III. THEORY OF JDO SSPARC
losses, etc.). Substituting (3) into (2) then into (1) yields the Consider a notional geographically distributed radar-
final basic MTI radar capacity expression comms (RC) scenario in which there is significant potential for
co-channel interference (see Fig. 1). If the available and limited
1 κB  2π   PRF  
CR =     (4) RF spectrum is to be shared by all parties in a dynamic fashion,
Ts  σ n ⋅ 4 SNR min  ∆θ   ∆f D   knowledge and adaptive functionality will be required to avoid
highly sub-optimal performance. For any RC node (radar or
where Ts denotes the scan or update time so that the capacity wireless communication device), at any given moment, there is
is in the conventional “bit rate” form [6]. Note that for a given a set of desired channels (target channel for radar, forward link
application, SNR min is specified ahead of time, and does not for comms), and other channels that are not of interest which
might include radar clutter, and other co-channel links.
usually vary during operation. Typical values for an MTI
radarfor SNR min would be between 10 and 15 dB depending Radar Comms Network
on acceptable false alarm levels, for a given assumed target
RCS.
Note that an analogous expression for radar capacity can be
readily formulated for synthetic aperture radar (SAR) by
simply recognizing that the equivalent of a “bit” is a SAR •••
“pixel”—again the minimum independent resolution cell with
a prescribed minimum SNR (sometimes referred to as terrain-
to-noise ratio (TNR)). Additionally, the most general radar
“bit” for either MTI or SAR is inherently an “M-ary” versus Comms Network Radar
binary piece of information since the amplitude (and possibly
polarization, etc.) contains additional information that can be
Fig. 1. Notional combined multi-radar and multi-comms network (RC) system
very useful such as target ID in MTI, or contrast in SAR.
Obviously for the M-ary case, there is an increase in capacity
We begin with the following set of definitions:
relative to the binary case considered in (4).
Finally for the case where both radar and communications H ij = transfer function from the j -th to the i -th nodes.
are operating simultaneously the combined expression for the s j = signal transmitted by j -th node
total capacity is given by (6)
y i = recieved signal at i-th node
1  κ Br  2π   PRF   ni = additive receive noise at i-th node
Ctot = α     +
Ts  σ ⋅ 4 SNR ∆θ   ∆f D   (5)
 n min  The dimensionality of the above variables depends on the
β ⋅ Bc log (1 + SNR c ) choice of degrees-of-freedom (DoFs) to be included in the
optimization. For example, if each node had say L spatial DoFs
where Br and Bc denote the radar and communication on both transmit and receive, the vectors in (6) would be L
dimensional and complex valued, and the constituent transfer
bandwidths respectively, and SNR c denotes the additive matrices would be L × L (and complex valued). We first
Gaussian white noise (AGWN) communications channel consider the above model over a short period of time for which
SNR. For the assumed co-channel interference model, the the channels can assumed to be non-varying, though they can
coupling between radar and communications occurs in the σ n still be stochastic. In this way we can arrive at a tight
theoretical performance bound as discussed below.
term for radar, and the SNR c term for communications. Thus,
in a co-design environment where the cross coupling terms Interpretation of the constituent block matrices H ij { } is
would depend on operating powers, bandwidths, and relative based on whether the nodes are radar or comms. For example,
locations and terrain, one would seek to jointly maximize it is convenient to define the submatrices with equal indices as
capacity while maintain a minimally acceptable level of radar follows:

978-1-4799-8232-5/151$31.00@2015IEEE 0762
 H ii = ∅, for comms nodes channels (at most N-1) for which it is desired to minimize
(7) interference. The situation is depicted in Fig. 2.
H ii ≠ ∅, for monostatic radar nodes
where ∅ denotes the null (zero) matrix or vector. In other H FCi ,1
words, a comms node does not radiate to itself. However, for a
monostatic MTI radar {H ii } represents the total signal- Forward Channel

dependent radar channel that in general consists of targets and

 Responses

clutter. More specifically, H FCi , N FC

i
N TOT
H ii = ∑H
n =1
Tn ,i + H Ci (8) si

where {H }
Tn ,i and {H C } i
denote the target and clutter H CCi ,1
th
channels for the i radar.
With the above definitions, the entire composite RC
 Co-Channel
Responses

channel matrix is given by H CCi , NCC

i

 H11 H12  H1 N 
H H 22 
Fig. 2. Illustration of forward and co-channel signal paths.
H =  21  (9)
    
  Assuming for the moment that all channels have equal
 H N1  H NN  “importance” (relative weighting will be introduced later), one
common design goal would be to maximize the signal power
If channel reciprocity is assumed, then H ij = H ji (symmetric through the forward channels while simultaneously minimizing
but not Hermitian). Moreover, some number of diagonal and the response in the co-channels. For the unconstrained
off diagonal submatrices will be null. As mentioned optimization case (other than finite norm) this leads to
previously, for comms nodes, the corresponding diagonal Rayleigh quotient maximization whose solution is a
submatrices will be null. Also, for monostatic radar, the generalized eigenvector problem, i.e.,
“reverse” channel is also null. As with co-channel comms
2
nodes, it is assumed that radar nodes can interfere with each E  H FCi si 
 (
si′ E H FC )
′ i H FCi si
other and other comms nodes.   = max
max (11)
si′ E ( H CC
′ H CC ) si
 2
The total composite transmit signal defined in (6) has the si E  H CCi si  si i i
structure  

 s1  where H FCi , H CCi denote the composite forward and co-

s 
channel transfer matrices respectively, denotes the L2 -
s= 2 (10)
   norm, and E ( ) denotes the probabilistic expectation operator.
 
s N  For the N-node case, the composite forward and co-channel
transfer functions H FCi , H CCi are given by (see Fig. 2)
where s i denotes the spatio-temporal transmit signal from the
ith RF node. Note some nodes may be receive only, or simply N
not transmitting at any given moment, in which case the
corresponding entries are the null vector.
H FCi = ∑w
n =1
FCi , n H FCi , n

(12)
N
For both radar and comms there are generally both
“forward,” or desired channels, and “co-channels” or H CCi = ∑w CCi , n H CCi , n
undesired propagation channels. A fundamental goal of JDO n =1
SSPARC is to maximize transmission through forward where H FCi ,n , HCCi , n denote the forward and co-channel
channels, while simultaneously minimizing co-channel
interference. Thus for the ith transmit node, there can be as transfer matrices from the ith transmit node to the nth receive
many as N forward channels if the ith node is simultaneously node, and wFCi ,n , wCCi , n ∈ [ 0,1] are relative weighting
acting as a monostatic radar and/or communications transmitter coefficients.
and/or multistatic radar (for which the other nodes are passive
receivers). In general, the number of forward channels for the The solution to (11) is based on solving the following
ith transmit node will be less than N, say N FCi . Additionally, generalized eigenvalue problem
for the ith transmit channel, there will be some number of co-

978-1-4799-8232-5/151$31.00@2015IEEE 0763
 ( ) (
′ i H FCi s = λ E H CC
E H FC ′ i H CCi s ) (13) accounted for in the generally colored noise matched filters [5,
9].
The eigenfunction solution with associated maximum 10

eigenvalue yields the optimum space-time transmit signal. Forward Channel

Note that if either (or both) of the matrices in (13) are
0

nonsingular, the solution is that of an ordinary eigenvalue -10

problem. It should also be noted that the kernel matrices in (13) -20
are positive semi-definite or definite, and thus all eigenvalues
are non-negative. -30

Magnitude (dB)
In general, additional constraints (beyond finite norm and -40

length) on s may be imposed such as constant modulus, etc., -50

that result in generally nonlinear programming problem of the -60

form
-70
*
 2  Co-Channels
E  H FCi si  -80 *
 WFCi 
max -90

si  2  -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5

E  H CCi si 
Normalized Angle

 WCCi  Fig. 3. Illustration of constrained optimum beamforming with co-channel

subject to: (14) “keep out” directions.

f (s) ≥ ∅ Although the basic theory for optimal reception is well

One important special case is when the constraints are in established, JDO SSPARC will leverage the recent advances in
the form of angular “keep-out” zones. This can arise for knowledge-aided (KA), physics-based/model-based, and
example if the locations of co-channel RF nodes are known or cognitive signal processing [4, 5, 12] to provide enhanced
where there is a desire to achieve a low probability of intercept knowledge of the colored noise channel, represented by the
(LPI) capability. If we let v1 ,, v J , where J < N , denote the space-time covariance matrix Ri for the ith receive node. The
spatial steering vectors (narrowband) associated with the keep- associated optimal receiver weight vector wi is given by (see
out directions, (14) takes on the form [5, 9] for further details)

E  H FCi si
2  w i = R −1s (16)
WFCi 
max  
where s is a desired steering vector of interest.
si  2 
E  H CCi si 
 WCCi 
IV. JDO SSPARC CHANNEL ESTIMATION
subject to: (15)
The previous section makes it clear that “as goes channel
V 's = ∅ knowledge, so goes performance.” Unlike the succinct
where the columns of V are the keep-out steering vectors. Fig. governing design equations derived in the previous section,
3 shows an example for the case of one forward channel and high-fidelity channel estimation has many facets and potential
two keep-out directions. The antenna consisted of a 16 element solutions, most of which are not mutually exclusive and can
uniform linear array (ULA), with half-wavelength element thus be combined into an overall effective approach. Thus we
spacing, operating from +/- 90-deg (+/- 0.5 normalized angle)) will first provide an overview summary of the different
[5]. In practice, the beam-patterns would be far more complex approaches, before delving into the mathematical and
due to non-line-of-sight (NLOS) propagation. implementation details later in this section.
In general the constraints in (14) will be nonlinear and not A. Overview of JDO SSPARC Channel Estimation
amenable to analytical solution. One such example is a The following is a fairly comprehensive listing and brief
constant modulus constraint on the fast-time waveforms that description of adaptive channel estimation techniques that can
allows for Class C (saturation) operation of the transmit be combined into the JDO SSPARC framework:
amplifiers, and thus the most effective radiated power (ERP)
[10]. Besides strictly numerical methods, one approximate • Training Sequences: This is the classical approach used in
solution to this that is asymptotically optimal is based on the wireless RF communication systems. A known signal is
method of stationary phase and nonlinear FM (NLFM) [11]. transmitted from one node to another. At the receive node
and adaptive estimation problem is solved to derive the
A. Optimal Receiver Structure channel transfer function. Performance of this approach
Once all RC transmit space-time waveforms have been depends on a number of factors including the complexity of
designed, the optimal receiver functions can be designed. The the training sequence, available computational receiver
reason the transmit waveforms are designed first is because resources, and the degree of channel stationarity. The price
they establish the colored noise component that must be paid is a reduction in overall capacity as compared to the

978-1-4799-8232-5/151$31.00@2015IEEE 0764
 case where the channel was known. In JDO SSPARC, this inherent in KA methods. KA/CRF HPEC architectures
concept can easily be extended to allow for channel are well documented in the literature [2-5, 10, 12, 14].
estimation between both comms, radar and comms-radar.
This can also be combined with MIMO techniques and o Fully Adaptive RF: Lastly, it assumed that adaptivity
channel reciprocity (see below). exists both in the receiver and the transmitter. The full
power of KA/CRF in general, and JDO SSPARC in
• MIMO: Spatial MIMO techniques can enhance the amount particular, is only realized when both the Tx/Rx
of information available by a channel estimator for a given functions are adaptive [4, 5].
training period. For example, instead of a single scalar
Although all of the above could be incorporated in a JDO
transmit training waveform, several simultaneous training
SSPARC system, the specific instantiations will vary
sequences can be transmitted from different antennas. A
significantly based on scenario specifics, and available
receiver would then use a bank of matched filters to
resources.
reconstruct each transmit degree-of-freedom. An adaptive
channel estimate that includes both spatial and temporal V. HIGH-FIDELITY JDO SSPARC DESIGN EXAMPLE
information can then be derived. This in turn can provide
In this section, high-fidelity ray tracing software was
significant performance improvement over the
utilized to better gauge the theoretical performance bounds
aforementioned SISO approach.
developed in Section III. While there is little opportunity for
• Reciprocity: Except for very pathological cases (e.g., significant performance gains in benign direct line-of-sight
“ducting”), Maxwell’s equations dictate the law of (DLOS) propagation environments, non LOS (NLOS)
reciprocity for electromagnetic propagation. That is the scenarios provide a rich opportunity for space-time adaptive
forward channel is equal to the time reversed reverse waveforms and receiver processing.
channel for the same transmit waveform. In theory this A five (5) node JDO SSPARC scenario was modeled using
means that one forward channel training sequence can be Wireless InSite™ [16], a high-fidelity RF ray tracing software
used to estimate the reverse channel. In practice the main package. The nodes were distributed in a 3 km x 2.5 km set of
impediment is the generally non-reciprocal nature of the terrain and building structures from an existing model of the
active electronics in the transmit-receiver chains. However, city of Philadelphia, PA, USA. The laydown is shown in Fig.
with the advent of “smart RF” electronics (see for example 4, along with a sample of the propagation paths from Node 1 to
[13]), it is now possible to have well calibrated electronics 4.
that can support the exploitation of reciprocity.
• Knowledge-Aided (KA) and Cognitive RF (CRF): KA and
CRF channel estimation methods are far and away the most
sophisticated and difficult to implement, but can in turn
yield the biggest benefits that are well documented in the
literature [2-5, 10, 12, 14]. Though there are a number of
KA/CRF variants, in JDO SSPARC we will assume the
most general instantiation that includes:
o High-Fidelity Environmental Dynamic Database
(EDDB): Environmental databases are increasingly
becoming the norm in normal operations. The
information they contain can include digital terrain
maps, land-cover land-use (LCLU), meteorological,
background traffic tracks, and RF “terrain” databases
such as the DARPA Radiomap [15]. It is further
assumed that mechanisms exist for the continual
updating of the database. This is a special case of the
assumed supervised learning function of the KA/CRF
system.
o KA/CRF Real-Time HPEC Architecture: We assume
the existence of a real-time high-performance Fig. 4. View of the simulated urban scenario showing the laydown of the 5 RF
embedded computing (HPEC) architecture that is nodes and propagation paths when Node 1 transmits and Node 4
capable of exploiting the aforementioned database in receives.
real-time and perform all necessary online calculations.
The DARPA KASSPER project successfully Table I below contains a description of the settings used to
developed and demonstrated several such HPEC predict the interference CIRs in the modeled urban scenario.
architectures. A key enabler is the existence of a “look These are the standard model parameters for the full 3D solver
ahead” scheduler and operating system that uses basic that we previously verified by reviewing the results against our
kinematic assumptions and RF scheduler information expectations of RF propagation. Our team has previously
to overcome the inevitable memory access latencies validated WI’s received power predictions against calibrated

978-1-4799-8232-5/151$31.00@2015IEEE 0765
measurements for a suburban propagation environment in New Fig. 6 shows a typical performance result for the case when
Jersey. Node 1 is “broadcasting” to Nodes 2 and 3, while
Table I. Specific simulation parameters
simultaneously attempting to minimize its co-channel
interference to Nodes 4 and 5. The 17 dB eigenvalue spread
Wireless InSite™ Solver Settings Tx/Rx Node Configuration again indicates there is significant opportunity for space-time
optimization.
• Solver = 3D Shooting and • Isotropic antenna (Max Gain = 0
Bouncing Rays dBi) Eigenvalue Distribution f or Complex Urban Propagation Example
20
• Ray Angle Spacing = 0.18° • Antenna Height = 2 m above
• 6 reflections/path (max) ground level 15
• 1 diffraction/path (max) • Waveform = Continuous Wave
• 0 Transmissions (buildings (CW) 10

absorb rays) • Carrier Frequency = 3000 MHz ~ 17 dB

5

Eigenvalue (dB)
Fig. 5 contains the full set of unity power normalized pair- 0

wise CIRs between the five modeled nodes. This result shows -5

that within this relatively small scale urban scenario that most
of the multipath components occur within 2 micro-seconds of -10

the arrival of the first ray. It should be noted that the first -15
arriving component is not necessarily the direct line-of-sight
path because we have disabled the “transmission” through -20
5 10 15 20 25

buildings to account for the relatively high absorption by Eigenvalue Index

building materials at 3 GHz. Fig. 6. Eigenvalue distribution for a particular wireless network topology. The
significant spread indicates there is significant potential for
performance enhancements using adaptive waveforms.
Tx from N1 RxN1
Norm Power. (dB)

0
RxN2
-20 RxN3 REFERENCES
-40 RxN4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 RxN5
Time w.r.t First Arrival (u-sec) [1] J. Chapin, "Shared Spectrum Access for Radar and Communications
Tx from N2 RxN1
(SSPARC)," DARPA BAA-13-24 www.darpa.mil, 2012.
Norm Power. (dB)

0
RxN2
-50 RxN3
[2] J. R. Guerci and E. J. Baranoski, "Knowledge-aided adaptive radar at
RxN4 DARPA: an overview," Signal Processing Magazine, IEEE, vol. 23, pp.
-100
0 1 2 3 4 5 6 RxN5 41-50, 2006.
Time w.r.t First Arrival (u-sec) [3] J. R. Guerci, "Cognitive Radar: The Next Radar Wave?," Microwave
Tx from N3 RxN1
Journal, vol. 54, pp. 22-36, January 2011.
Norm Power. (dB)

0
RxN2
-50 RxN3
[4] J. R. Guerci, et al., "CoFAR: Cognitive Fully Adaptive Radar,"
RxN4 presented at the IEEE Radar Conference, 2014.
-100
0 1 2 3 4 5 6 7 8 9 10 RxN5 [5] J. R. Guerci, Space-Time Adaptive Processing for Radar, 2nd Edition.
Time w.r.t First Arrival (u-sec) Norwood, MA: Artech House, 2014.
Tx from N4 RxN1
C. E. Shannon, et al., The mathematical theory of communication vol.
Norm Power. (dB)

0
RxN2
[6]
-50 RxN3
117: University of Illinois press Urbana, 1949.
RxN4 [7] P. M. Woodward, Probability and information theory, with applications
-100
0 1 2 3 4 5 6 7 RxN5 to radar vol. 3: Pergamon Press London, 1953.
Time w.r.t First Arrival (u-sec) [8] D. K. Barton, Modern radar system analysis. Norwood, MA: Artech
Tx from N5 RxN1
Norm Power. (dB)

0
RxN2
House, 1988.
-50 RxN3
[9] H. L. V. Trees, Detection, Estimation and Modulation Theory. Part I.
RxN4 New York: Wiley, 1968.
-100
0 1 2 3 4 5 6 7 8 9 10 RxN5 [10] J. R. Guerci, Cognitive Radar: The Knowledge-Aided Fully Adaptive
Time w.r.t First Arrival (u-sec) Approach. Norwood, MA: Artech House, 2010.
[11] C. E. Cook and M. Bernfeld, Radar signals: Academic Press New York,
1967.
[12] G. R. Legters and J. R. Guerci, "Physics-based airborne GMTI radar
Fig. 5. Unity power normalized pair wise impulse responses between signal processing," in Radar Conference, 2004. Proceedings of the
the five nodes. IEEE, 2004, pp. 283-288.
[13] N. Kingsley and J. R. Guerci, "Adaptive Amplifier Module Technique to
To gauge the potential performance improvements possible Support Cognitive RF Architectures," presented at the IEEE Radar
Conference, Cincinnati, OH, 2014.
using adaptive JDO SSPARC space-time waveforms, equation
[14] J. Mitola, et al., "Achieving 5G QoE through Spectrum Sharing and
(13) was solved for a variety of transmit-receive scenarios. In Inhomogeneous Heterogenous Networks (accepted for publication),"
each case significant eigenvalue spread was observed, often Communications Magazine, IEEE, 2014.
greater than 10 to 15 dB. This in turn implies there is [15] J. M. Chapin. DARPA Radiomap Program. Available:
significant opportunity for performance gains if space-time http://www.darpa.mil/OurWork/STO/Programs/Advanced-
transmit-receive adaptation is performed via the JDO SSPARC Mapping(RadioMap).aspx
[16] REMCON. Wireless InSite. Available:
design equations (or constrained variants thereof). http://www.remcom.com/wireless-insite

978-1-4799-8232-5/151$31.00@2015IEEE 0766