WSA 2016 • March 9-11, 2016, Munich, Germany

On EIRP Control in Downlink Precoding for

Massive MIMO Arrays
Niklas Doose and Peter A. Hoeher
Information Theory and Coding
Faculty of Engineering, University of Kiel, 24143 Kiel, Germany
Email: {nd,ph}@tf.uni-kiel.de

Abstract—The usual performance comparisons of wireless to be transmitted. However, with the target of an ultra-high
systems are based on a constrained sum-power at the transmitter. speed link to a single user it is crucial to split the signal into
However, many wireless systems are actually constrained in their multiple streams to relax the requirements for the decoding in
equivalent isotropic radiated power (EIRP). So far, research
effort in the area of beamforming under constrained EIRP seems the receiver, which becomes a bottleneck in such systems [11].
to be done solely for single-user systems. As an original contri- Most other research in the area of beamforming and/or precod-
bution, this contribution develops a theoretical description of an ing does not even mention the EIRP or deals with the EIRP as
EIRP-limited multi-user system with joint consideration of the a static property of the system. To our best knowledge multi-
EIRP and the capacity. Accordingly, the EIRP is interpreted as a layer beamforming with explicit consideration of the variable
function of the precoding matrix instead of a static measure of the
antenna. The theoretical formulations is applicable to arbitrary EIRP has not been investigated in the research literature so
antenna designs, including multi-mode antennas. Down-scaling of far. This contribution introduces a joint consideration of the
any coventional linear precoding solution is proposed as a simple capacity of a multi-layer beamforming system together with
strategy to comply with the EIRP limit. Numerical results are an EIRP limitation. Therefore, a theoretic problem formulation
provided in the context of massive MIMO with simple linear is derived to give insight to the radiation characteristics of
precoding techniques.
precoded systems. In this context the resulting EIRP is a
I. I NTRODUCTION function of the precoding and not a static measure of the
The received signal power to noise power ratio (SNR) is antenna configuration and the power. The performance gain
a crucial parameter with respect to the capacity of wire- of the consideration of the variable EIRP compared to a static
less communication systems. Therefore, it is necessary and EIRP is investigated in terms of numerical simulations.
suitable to compare different systems in terms of some sort The employment of so-called multi-mode antennas as an-
of power constraint to prevent a capacity increasement by ticipated in [12] will be included in the investigations of
simple up-scaling of the transmit power. The most common this contribution. The multi-mode antenna approach aims at
approach is the assumption of a constrained sum-power at very compact multiple-element antennas. The design process
the transmitter side of the system. However, in the beginning makes use of the theory of characteristic modes that enables
of the successful multiple-input multiple-output (MIMO) era multiple—theoretically orthogonal—radiation patterns on sin-
an equivalent isotropic radiated power (EIRP) constraint has gle physical elements as shown in [13]. Just the same as for
been investigated. The radiation characteristic has been taken spatially separated, discrete single antenna elements, multiple
into account in terms of the array factor. In [1] and [2] the multi-mode antennas can be used to form an array as intro-
EIRP constraint is mentioned to be a regulatory condition for duced in [14], where a 484 port array has been realized with
WLAN-type systems. Again, in [3] the EIRP is studied as 11×11 physical elements. Note that the utilization of multi-
one possible constraint in the context of WLAN. Surprisingly, mode antennas results in a generalization of the considerations
since then joint precoding and power control has rarely been rather than being a special case.
taken into account in research papers on WLAN systems, Section II introduces the system model under consideration
although these radio systems are restricted with respect to their and gives insight to the configuration and flexibility of the
EIRP. system design. In Section III the problem formulation is
With the upcoming interest in ultra-wideband (UWB) sys- derived and solved by a simple down-scaling scheme. In
tems, EIRP constraints were reconsidered for example in [4]– Section IV precoding methods for massive MIMO systems
[8]. UWB systems have been standardized by the FCC, ETSI are investigated in the context of a constrained EIRP.
and several other institutions for license-free communication
purposes [9], [10]. In order to limit interference, a regulation of II. S YSTEM M ODEL
the EIRP is particularly important for UWB systems, because
they operate in the same frequency bands as primary, licensed Most physical wireless channels exhibit multipath propaga-
systems. tion, which leads to dispersion and frequency-selectivity. By
The publications mentioned so far considering EIRP-limited means of an orthogonalization in frequency like OFDM, the
beamforming assume only one user and a single data stream system equation of each orthogonal sub-channel can be written

ISBN 978-3-8007-4177-9 184 © VDE VERLAG GMBH · Berlin · Offenbach, Germany

Authorized licensed use limited to: SOCIETY FOR APPLIED MICROWAVE ELEC ENG AND RES. Downloaded on June 03,2020 at 12:03:21 UTC from IEEE Xplore. Restrictions apply.
 WSA 2016 • March 9-11, 2016, Munich, Germany

in multiplicative vector-matrix form as has to be maximized. The capacity is given as a function of the
precoding matrix and the channel matrix, which is assumed
y = Hx + n. (1)
to be perfectly known at the transmitter side. The nominal
The received vector y for one time instant and sub-channel signal-to-noise ratio SNR0 is defined as the maximal SNR
(time index and frequency index are omitted for simplicity) that can be transmitted and is therefore given by the ratio of
is given by the multiplication of the transmit vector x with the EIRP limit EIRP0 and the noise spectral density. This SNR
the channel matrix H and the additive noise vector n. The is reached for isotropic radiation, where the EIRP constraint
transmit vector x is constructed by the precoding matrix W is fulfilled by equality in each direction.
and the symbol vector s as In contrast to former publications elaborating on the topic
x = W s. (2) of joint precoding and power control for EIRP-limited MIMO
systems, in this contribution the three dimensional space is
The vector s has the dimension NS × 1, with NS being the considered rather than a plane. Additionally, non-isotropic
number of streams (or layers) in the precoder. The channel antenna characteristics generalize the formulations from earlier
matrix is assumed to be known at the transmitter side from publications, which is necessary for the usage of physically
uplink pilots and successive channel estimation. This method realizable antennas. We include the characteristics F nT (θ, φ)
is commonly assumed in time-division duplex systems like of each antenna element nT for both polarization directions.
massive MIMO with reciprocal uplink and downlink channel Please note that the characteristic F nT (θ, φ) is a angle-
matrices [15]–[17]. The elements of n are chosen from a dependent measure at a constant distance of 1 m from the
circular symmetric complex zero-mean Gaussian distribution. physical radiator. The unit of F nT (θ, φ) is volt [V]. In the
The system setup is sketched in Fig. 1. Please note that the information theory community it is commonly assumed that
sketched multi-mode antennas (MMA) offer three inputs that all antenna elements are positioned in a (typically linear)
can be used by the digital baseband processing. Therefore, it array and have an identical characteristic, which simply leads
is meaningful to distinguish between the number of physical to a multiplication of the element characteristic with the
antenna elements and the number of effective antenna ports. array factor. Please note that for multi-mode antennas this
The system offers flexibility in terms of trade-offs between simplification is not possible, because for each port the antenna
multiplexing and diversity. The extreme cases are (i) full characteristic is different.
multiplexing, where the number of streams NS is equal to the The EIRP is a measure for the maximum directivity of the
number of effective receive ports NR,eff and (ii) full diversity, antenna plus array factor (AAF ):
where just one stream is used. In the case of full multiplexing
 
the matrix W acts as a precoder, for the case with one input EIRP =
4π
max |AAF (θ, φ, W )| ,
2
(4)
stream it breaks down to a beamforming vector. The receiver 2ZF0 θ,φ
has to be configured by control bits in the MAC layer to
where ZF0 is the free space impedance, which is approximated
adapt the coding rates and modulation formats. If any diversity
by ZF0 = 120π. The unit of ZF0 is ohm [Ω]. The absolute
is used, the parallel to serial conversion in the receiver unit
squared AAF is given by
needs a scheme to combine the diversely transceived streams.
Similarly, a control unit in the transmitter takes care of channel |AAF (θ, φ, W )| =
2
estimation from pilots, calculation of EIRP-limited precoding,  2
 
adaptive modulation formats and coding rates and the loading  nT 
 W[nT ,nS ] · Fϑ (θ, φ) · ej2πk(θ,φ)·r(nT )  +
of bits onto the adaptive number of streams. The control unit  
nT nS
has to consider quality of service (QoS) requirements and data  2
 
rate demand to adapt the parameters of the system blocks. In  
 W[nT ,nS ] · FϕnT (θ, φ) · ej2πk(θ,φ)·r(nT )  , (5)
the context of ultra-high data rates in the region of 100 Gbps  
nT nS
[11], the case of full multiplexing is the suitable scenario.
Therefore, in this contribution we will neglect the case of where ϑ and ϕ are horizontal and vertical polarization, respec-
diversity and narrow the scope to full multiplexing. tively. It is obvious that the EIRP is a function of the current
III. P ROBLEM F ORMULATION precoding and is not a static measure. The max-operator in (4)
acts on the whole sphere, so any incident angle represented
In multi-user systems the optimization criterion is usually by the wave vector k(θ, φ) is considered. The sub-scripted
the maximization of the signal-to-interference-plus-noise ratio brackets (·)[n,m] denote the element in the nth row and the
(SINR). In the scenario proposed in [11], [12], the multiplexed mth column of a matrix. In this general form arbitrary antenna
data streams are actually used by one mobile terminal. There- positions can be specified via the positioning vectors r(nT ) for
fore all SINRs have to be maximized simultaneously, which is each element.
novel compared to previous EIRP-limited beamforming con-
The constraint on the EIRP at the transmitter can be written
siderations. In a different view, the overall capacity including
as
a specific precoding solution,  
   4π 2
C(W ) = log2 det INR + SNR0 HW W H H H , (3) · max |AAF (θ, φ, W )| ≤ EIRP0 (6)
2ZF0 θ,φ

ISBN 978-3-8007-4177-9 185 © VDE VERLAG GMBH · Berlin · Offenbach, Germany

Authorized licensed use limited to: SOCIETY FOR APPLIED MICROWAVE ELEC ENG AND RES. Downloaded on June 03,2020 at 12:03:21 UTC from IEEE Xplore. Restrictions apply.
 WSA 2016 • March 9-11, 2016, Munich, Germany

IDFT P/S CP RF

IDFT P/S CP RF MMA

IDFT P/S CP RF
ENC MOD RF CP S/P DFT DEM DEC

S/P ENC MOD W MMA RF CP S/P DFT DEM DEC P/S

ENC MOD RF CP S/P DFT DEM DEC

IDFT P/S CP RF

IDFT P/S CP RF MMA

IDFT P/S CP RF

Figure 1. Block diagram of the downlink employing three streams that are encoded and modulated independently. An OFDM-like structure is used to create
flat-fading channels, where each channel is beamformed separately, i.e. each frequency sub-channel has one matrix W . Additional frequency bins are indicated
in gray.

and the problem formulation for the maximization of the sum that the array could produce, due to a different main radiating
capacity given the constrained EIRP yields a conventional direction than broadside. Therefore, the investigation in this
minimization problem with one inequality constraint: contribution compares the capacities of the downlink for the
   case where the actual gain denoted by gPRE for a specific
minimize − log2 det INR + SNR0 HW W H H H
W ∈CNT ×NR precoding is used and the case where the maximum gain gANT
4π   of the antenna is used for the down-scaling of the precoding
2
subject to EIRP0 − max |AAF (θ, φ, W )| ≥ 0.
2ZF0 θ,φ matrix.
(7)
To our best knowledge, there is neither a known optimal value IV. N UMERICAL R ESULTS FOR M ATCHED F ILTER
for the capacity, nor a known analytic way of solving the P RECODING
problem stated above to reach that optimal value. Besides Massive MIMO research prefers linear precoding techniques
the conventional formulation, the optimization problem bears like matched filter (MF) precoding, because their performance
severe problems for numerical optimization techniques. gets close to optimal precoding under certain propagation
The transmit power PS is obtained by an integration over conditions [16]. Therefore, this section investigates the ca-
the whole sphere of the absolute squared AAF . Similarly, pacity for MF precoding under a constrained EIRP. Formally,
the EIRP in (4) is the integration over the maximum of the unscaled MF precoding corresponds to
absolute squared AAF . Because the maximum of the absolute 
WMF = H H. (9)
squared AAF is constant, the integration yields 4π. For both
computations the sphere is assumed to have a radius of 1 m As a channel model, the geometric, ray-based WINNER II
and the result is a measure of power with the unit watts [W]. channel model has been used [18]. It computes the effective
The fraction g = EIRP/PS is the antenna gain. In a practical channel coefficients between antenna pairs by a summation
system, the simplest approach to comply with the EIRP limit over clustered rays. The statistic distributions of the rays’
and provide a sub-optimal solution to (7) is scaling down any angles, delays and powers are derived from excessive measure-
precoding matrix W  by the square root of the gain factor, ment campaigns. Despite the fact that the specified bandwidth
i.e. of the chosen model slightly higher than the anticipated
W
W = √ . (8) band considered in [11], it is assumed that the results will
g qualitatively hold.
It is assumed that the theoretic maximum gain that an array Figure 2 shows the capacity for EIRP limited beamforming
can produce is in the ”broadside” direction. The broadside is for a system with one mobile terminal. The terminal is
commonly agreed to be perpendicular to the two-dimensional equipped with a single multi-mode antenna with NR,eff = 4
plane that is spanned by the array. The assumption is justified ports and the base station (BS) is equipped a variable number
because of the maximized virtual aperture that is seen from this of multi-mode antennas in a two-dimensional square array.
direction. The precoding solution for this case is assumed to be The precoding matrix is scaled down according to (8) with
a feeding with equal weights and phases to the antenna ports. gANT to yield the capacity CANT and with gPRE to yield the
The value of the gain for this case can be calculated offline capacity CPRE , respectively.
and will be denoted as gANT . However, for a specific precoding For an increasing number of effective antenna ports in
solution the actual gain typically is than this maximum gain the BS the variable down-scaling scheme outperforms the

ISBN 978-3-8007-4177-9 186 © VDE VERLAG GMBH · Berlin · Offenbach, Germany

Authorized licensed use limited to: SOCIETY FOR APPLIED MICROWAVE ELEC ENG AND RES. Downloaded on June 03,2020 at 12:03:21 UTC from IEEE Xplore. Restrictions apply.
 WSA 2016 • March 9-11, 2016, Munich, Germany

30 of the equal powers of the rays. In this example, however,

in Fig. 3(a) the MF solution creates only one of the lobes.
In other words, for the given example the MF beamforming
technique tries to focus all available power in one direction.
25 Under a constrained sum-power this might lead to the same
Capacity C [bit/s/Hz]

performance as forming a beam in the direction of the other

ray, or to distribute the available power among both ray
directions. For EIRP-constrained systems, however, a single
20
narrow lobe turns out to be sub-optimal, because each lobe
has to fulfill the constraint. An additional lobe in the other
ray direction will therefore increase the sum-power budget
15 of the transmitter, but does not alter the fact that the other
lobe still complies with the EIRP limit. At the same time the
CPRE instantaneous sum-capacity increases because of the additional
CANT lobe and the system benefits from the transmission of a second
10 stream over the other lobe. In Fig. 3(b) classical phased array
0 100 200 300 400 500
beamforming has been applied, where a fixed phase difference
Effective antenna ports in BS between the signals of the elements leads to a specific main
radiation direction at a given frequency.
Figure 2. Ergodic capacity of channel realizations over number of effective V. C ONCLUSION
base station ports for multi-mode system with NR,eff = 4. CANT corresponds
to the capacity, where the theoretical maximum gain of the antenna gANT is
In the evolution of wireless communication standards,
used for down-scaling, and CPRE corresponds to the capacity, where the actual multi-stream processing is becoming more important in multi-
gain of the respective precoding solution gPRE is used for the down-scaling user scenarios as well as in point-to-point links. A crucial
as explained in the text. The SNR0 is 30 dB.
performance parameter is the EIRP limitation. However, most
researchers use a normalized sum-power constraint to optimize
static scheme. In fact the performance gap increases with an and compare different systems. In contrast, in this contri-
increasing number of ports. Besides the relative performance bution the problem formulation for a multi-stream point-to-
gap, the capacities reach a maximum for NT,eff = 324 and point downlink in conjunction with an EIRP constraint is
decrease for a larger array. This effect is due to the inability developed. Down-scaling of the precoding matrix is proposed
of the matched filter precoding to use the spatial channel effec- as a sub-optimal solution to the stated optimization problem.
tively. As an illustrative, purely academic example, in Fig. 3 The numerical results take a massive MIMO scenario under
one can see the radiated pattern for a BS with NT,eff = 24 consideration. It is shown that it is crucial to perform the
antenna ports, while a MS with three ports is considered. down-scaling with the knowledge of the actual precoding
The channel model is a reduced complexity, ray-based channel matrix used instead of an offline computed value. Additionally,
under an EIRP constraint simple algorithms like MF precoding
appear to be too simple to benefit from massive MIMO arrays
90◦ 90◦
135◦ 1 45◦ 135◦ 1 45◦ with vast degrees of freedom. A growth of the number of
elements in the array even leads to a slight decrease of the
0.5 0.5
channel capacity after reaching a maximum.
180◦ 0◦ 180◦ 0◦ ACKNOWLEDGEMENT
This work is supported by the Deutsche Forschungsgemein-
schaft (DFG) within the priority program SPP 1655 under
225◦ 315◦ 225◦ 315◦ grant HO 2226/14-1.
270◦ 270◦
R EFERENCES
(a) MF (b) Phased Array
[1] P. Zetterberg, M. Bengtsson, D. McNamara, P. Karlsson, and M. A.
Figure 3. Radiated antenna plus array pattern for under EIRP constraint, Beach, “Downlink beamforming with delayed channel estimates under
given a two-ray channel model and a BS with NT,eff = 24 ports for a total power, element power and equivalent isotropic radiated power
linear antenna array with omni-directional antennas. Case (a) refers to the (EIRP) constraints,” in Proc. IEEE Vehicular Technology Conf. (VTC
MF solution (WMF ) and case (b) refers to phased array beamforming, when Fall), vol. 1, 2001, pp. 516–520.
the two rays are perfectly known and each ray is used by one stream (i. e. each [2] ——, “Performance of multiple-receive multiple-transmit beamforming
stream forms one lobe). The capacities at 30 dB is given by CMF = 19.07 in WLAN-type systems under power or EIRP constraints with delayed
bps/Hz and Cphasedarray = 20.97 bps/Hz. channel estimates,” in Proc. IEEE Vehicular Technology Conf. (VTC
Spring), vol. 4, 2002, pp. 1906–1910.
with two rays with equal power going out of the BS in the [3] A. M. Kuzminskiy, “EIRP-restricted downlink beamforming in WLAN
OFDM systems,” in Proc. IEEE Signal Processing Workshop on Signal
directions Φ1 = 0◦ and Φ2 = 45◦ . Intuitively, one would Processing Advances in Wireless Communications (SPAWC), Jul. 2006,
expect lobes in the ray directions with equal power, because pp. 1–5.

ISBN 978-3-8007-4177-9 187 © VDE VERLAG GMBH · Berlin · Offenbach, Germany

Authorized licensed use limited to: SOCIETY FOR APPLIED MICROWAVE ELEC ENG AND RES. Downloaded on June 03,2020 at 12:03:21 UTC from IEEE Xplore. Restrictions apply.
 WSA 2016 • March 9-11, 2016, Munich, Germany

[4] C. M. Vithanage, Y. Wang, and J. P. Coon, “Spatial PAPR reduction Communications and Coding (SCC), Feb. 2015, pp. 1–6.
based beamforming scheme for EIRP constrained systems,” in Proc. [12] P. A. Hoeher and N. Doose, “A massive MIMO terminal
IEEE Global Communications Conf. (GLOBECOM), Nov. 2008, pp. 1– concept based on small-size multi-mode antennas,” Trans. Emerging
5. Telecommunications Technologies, Mar. 2015. [Online]. Available:
[5] C. M. Vithanage, J. P. Coon, and S. C. J. Parker, “On capacity-optimal http://dx.doi.org/10.1002/ett.2934
precoding for multiple antenna systems subject to EIRP restrictions,” [13] R. Martens and D. Manteuffel, “Systematic design method of a mobile
IEEE Trans. Wireless Commun., vol. 7, no. 12, pp. 5182–5187, Dec. multiple antenna system using the theory of characteristic modes,” IET
2008. Microwaves, Antennas & Propagation, vol. 8, pp. 887–893, Sep. 2014.
[6] C. M. Vithanage, Y. Wang, and J. P. Coon, “Transmit beamform- [14] D. Manteuffel and R. Martens, “Compact multi mode multi element
ing methods for improved received signal-to-noise ratio in equivalent antenna for indoor UWB massive MIMO,” submitted to IEEE Trans.
isotropic radiated power-constrained systems,” IET Communications, Antennas and Propagation, 2015.
vol. 3, no. 1, pp. 38–47, Jan. 2009. [15] T. L. Marzetta, “Noncooperative cellular wireless with unlimited num-
[7] C. M. Vithanage, M. Sandell, J. P. Coon, and Y. Wang, “Precoding in bers of base station antennas,” IEEE Trans. Wireless Commun., vol. 9,
OFDM-based multi-antenna ultra-wideband systems,” IEEE Commun. no. 11, pp. 3590–3600, Nov. 2010.
Mag., vol. 47, no. 1, pp. 41–47, Jan. 2009. [16] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,
[8] T. M. Kim, A. Ghaderipoor, and A. Paulraj, “Transmit beamforming for O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities and
EIRP-limited MIMO systems based on Golay sequence,” in Proc. IEEE challenges with very large arrays,” IEEE Signal Processing Mag.,
Global Communications Conf. (GLOBECOM), Dec. 2012, pp. 4798– vol. 30, no. 1, pp. 40–60, Jan. 2013.
4803. [17] E. G. Larsson, F. Tufvesson, O. Edfors, and T. L. Marzetta, “Massive
[9] FCC, Code of Federal Regulations, Title 47, Part 15, Subpart F, Federal MIMO for next generation wireless systems,” IEEE Commun. Mag.,
Communications Commission, Feb. 2002. vol. 52, no. 2, pp. 186–195, Feb. 2014.
[10] ETSI, European Standard EN 302 065 V1.3.1, European Telecommuni- [18] P. Kyösti, J. Meinilä, L. Hentilä, X. Zhao et al., “IST-WINNER D1.1.2:
cations Standards Institute, Apr. 2014. WINNER II channel models,” Technical Report (ver. 1.1), Sep. 2007.
[11] N. Doose and P. A. Hoeher, “Massive MIMO ultra-wideband communi- [Online]. Available: http://www.ist-winner.org/WINNER2-Deliverables/
cations using multi-mode antennas,” in Proc. Int. ITG Conf. on Systems, D1.1.2v1.1.pdf

ISBN 978-3-8007-4177-9 188 © VDE VERLAG GMBH · Berlin · Offenbach, Germany

Authorized licensed use limited to: SOCIETY FOR APPLIED MICROWAVE ELEC ENG AND RES. Downloaded on June 03,2020 at 12:03:21 UTC from IEEE Xplore. Restrictions apply.