Duality of antennas and subcarriers in On the basis of the big number law and large-scale random matrix

massive MIMO-OFDM downlink system theory, one can develop

HH H HH H
Ahmed Bannour and Yichuang Sun lim = lim
nt +1 nc nt nt nc +1 nc nt

N
Massive multiple-inputmultiple-output (MIMO) can signicantly
outperform conventional MIMO in terms of spectrum efciency and HH H
= lim
link reliability. For massive MIMO, there are still theoretical and prac- N +1 N
tical issues that have to be addressed. The capacity of the massive


tr(H2 )
MIMO-orthogonal frequency division multiplexing (OFDM) downlink = IN
N
system is derived and analysed and the duality of antennas and subcar- 


riers in such system is demonstrated analytically and by simulation. A p
detailed comparison between massive MIMO, massive MIMO-OFDM
= sh IN (3)
2
and MIMO-OFDM with large subcarriers is presented.
Hence, combining (3) and (2) one can obtain
 


Introduction: To satisfy the demands for high throughput commun- nc nt p
ications a massive number of antennas [massive multiple-input lim C = log2 det Inc nt (1 + 2 sh )
nt +1 sw 2
multiple-output (MIMO)] [1, 2] can be adopted for 5G systems, since  


massive MIMO helps to reduce transmit power, noise and fast fading. sh p
= nc nt log2 2 nc nt (4)
The reliability and spectral efciency are much more improved com- sw 2
pared with the conventional MIMO techniques. Large-scale MIMO
If we set h = 1 then
can also provide a real capacity enhancement. However, there are a
number of problems that need to be solved, such as the estimation of 





p/2
a large number of channels between nt transmit antennas and nr lim C = nc nt log2 n c n t (5)
nt +1 s2w
receive antennas, the use of tremendous pilot overhead which results
in pilot contamination and the increase of computational power. The capacity performed over ncnt narrowband channels can be written as
Although the theoretical aspects of massive MIMO systems have 





gained signicant attention in the research community [3, 4], much p/2
lim CAV = log2 nc nt (6)
less is known about the practical transmission schemes. As pointed nt +1 sw
2

out in [5], practical realisations of large-scale MIMO systems will

require the use of low-cost and low-power radio-frequency (RF) com-
ponents. Practical wireless channels typically exhibit frequency selective
fading, which requires high-cost RF components. The combination of
massive MIMO with orthogonal frequency division multiplexing
(OFDM) can overcome the frequency selective problems and reduce
the burden on RF components.
This Letter discusses massive MIMO-OFDM based on the capacity
and achievable rate. The contribution of the Letter is:
The analytic capacity of the massive MIMO-OFDM system is derived.
We prove theoretically and by simulation, the duality between the
number of antennas in MIMO and the number of subcarriers in OFDM.
We show the benet of MIMO-OFDM with high subcarriers com-
pared with massive MIMO and massive MIMO-OFDM.
System model: The MIMO-OFDM system in a baseband model, equipped
with nt transmit antennas and nr user antennas is considered (nr = nt). To sim-
plify the discussion, we give the overall mathematical model as presented in
[6]. MIMO-OFDM symbols are supposed to be transmitted over a frequency
and time selective Rayleigh channel. The channel taps are assumed constant
when transmission occurs. The channel impulse response distribution
follows the Rayleigh law between the qth transmitting




antenna and pth
receiving antenna and hkp, q
Rayleigh(sh p/2, ((4 p)/2)s2h ).
Note that hkp, q (l) is the lth path from the qth transmitting antenna to the
pth receiving antenna at time k and L is the largest order among all
From (6), one can conclude that a small antenna number nt can achieve
similar performance as a large antenna number when an OFDM modu-
lator is included due to the factor of ncnt. Rather than focusing on the
antenna number, one can target the number of subcarriers nc, in other
words we speak about (nc, nt) duality. Without loss of generality, we
start with the couple (nt , nc )nt 1 at this stage and we can inverse the
situation to (nc , nt )nc b , where is a big subcarrier number. In
general, a very large number of antennas will result in much more RF
modules, including a low-noise amplier, a frequency down-converter
and an analogue-to-digital converter. To reduce the cost associated
with RF modules, the OFDM technique can be used to achieve the
needed capacity with a smaller number of antennas.
Simulations: The cumulative complementary density function (CCDF)
is used as a tool to give the probability that the capacity C is larger than
the capacity abscissa Cx.
Fig. 1 depicts the CCDF of the capacity developed in (6), for massive
MIMO, MIMO-OFDM and massive MIMO-OFDM congurations.
1.0
nc = 48, nt = 1000
0.9 nc = 1024, nt = 1000
0.8 nt = 1000, no OFDM
massive MIMO
nc = 512, nt = 500
0.7
gap = 10 bits/s/Hz
0.6

impulse responses. The mathematic model of the kth MIMO-OFDM

CCDF

0.5
symbol at the pth receiving antenna is
0.4 gap = 4 bit/s/Hz

nT 
L1
0.3
ypk = hkp, q (l)uqk (k l) + wpk (1)
0.2
q=1 l=0
0.1
where uqk is the symbol vector transmitted by the qth antenna and wpk is zero MIMO-OFDM
0
mean white Gaussian complex noise of variance N0/2. 16 18 20 22 24 26 28 30 32 34
Cx
Capacity of massive MIMO-OFDM system: The mathematical
expression of the MIMO-OFDM system using the maximisation of
mutual information is presented in detail in a previous work by Fig. 1 CCDF of massive MIMO, MIMO-OFDM and massive MIMO-OFDM,
SNR = 10 dB
Bannour et al. [68]. To achieve the maximum capacity for a large
number of antennas, one can derive the theoretical limit of the capacity
In Fig. 1, the CCDF windows is spread for up to 10 dB. The capacity
C as when nt goes to innity as
is larger than 18, 24 and 32 bits/s/Hz, respectively, for massive MIMO,
 
1 MIMO-OFDM and massive MIMO-OFDM for 90% of probability. The
lim C = lim log2 det I nc nt + 2 HH H capacity increases when the couple (nc, nt) increases. We note the gap of
nt +1 nt +1 sw
  10 bits/s/Hz at 60% of probability between massive MIMO and massive
nc nt H H H
= lim log2 det I nc nt + 2 (2) MIMO-OFDM and a negligible gap between massive MIMO-OFDM
nt +1 sw nc nt and conventional MIMO-OFDM with a high number of subcarriers

ELECTRONICS LETTERS 9th July 2015 Vol. 51 No. 14 pp. 11151117

nc = 1024. It can be noted that for a high number of subcarriers nc a high 24
number of antennas is no longer needed, because we can achieve the
22
theoretical limit of the downlink capacity with a reduced antenna
number due to the ncnt effect. 20
Fig. 2 shows clearly the impact of subcarrier number on the capacity.
We note a 10 bits/s/Hz gap at 600 antennas between massive MIMO and 18 MIMO-OFDM nt = 2

ergodic capacity
MIMO-OFDM with high nc values and a negligible gap between MIMO-OFDM nt = 1000
16
massive MIMO-OFDM and MIMO-OFDM with high nc.
14

12

24 10

nc = 1024
22 8
0 200 400 600 800 1000 1200
subcarrier number
20
MIMO-OFDM

Fig. 4 Capacity for MIMO-OFDM and massive MIMO-OFDM with respect

ergodic capacity

18 nc = 512
to subcarrier number, SNR = 10 dB
16
nc = 48
Conclusion: We have studied a MIMO-OFDM conguration in the
14
gap = 10 bits/s/Hz massive MIMO context. It has been shown analytically that in rich
12 scattered channels, MIMO-OFDM can achieve the same capacity ef-
MIMO ciencies with a high subcarrier number nc, as those using the recently
10
proposed massive MIMO scheme. Simulations showed the effectiveness
8 of the proposed method for the case where nc is high and nt still a con-
0 200 400 600 800 1000 ventional value. The proposed scheme can use fewer transmit antennas
antenna number
combined with a high number of OFDM subcarriers to reduce the high
cost of RF components in practical wireless systems.
Fig. 2 Asymptotic behaviour of capacity for massive MIMO, MIMO-OFDM
and massive MIMO-OFDM, SNR = 10 dB, with respect to antenna number The Institution of Engineering and Technology 2015
Submitted: 28 February 2015 E-rst: 18 June 2015
doi: 10.1049/el.2015.0355
In Fig 3, for both the approximate capacity expression against One or more of the Figures in this Letter are available in colour online.
signal-to-noise ratio (SNR) per receive antenna, derived in (6), and
Ahmed Bannour (Higher School of Data Processing of Mahdia,
the theoretical result using (2), it is shown that the match between the
University of Monastir, Avenue Habib Bourguiba, Sidi Massoud BP
two gets closer as SNR increases.
49, Mahdia 5111, Tunisia)
E-mail: bannour.ahmed@gmail.com
Yichuang Sun (School of Engineering and Technology, University of
nt = 1000, nc = 48
Hertfordshire, Hateld, Herts, AL10 9AB, United Kingdom)
700
nt = 2, nc = 1024

600 nt = 1000, no OFDM

References
nt = 1000, nc = 1000 1 Marzetta, T.: Noncooperative cellular wireless with unlimited numbers
500
theoretical nt = 1000, nc = 1000 of base station antennas, IEEE Trans. Wirel. Commun., 2010, 9,
pp. 35903600
ergodic capacity

400
300
approximate
200
100
0
0 5 10 15
Eb/N0, dB
Fig. 3 Capacity for massive MIMO, MIMO-OFDM and massive
MIMO-OFDM with respect to SNR
Fig. 4 illustrates the impact of the subcarrier number on capacity for
both a small and a very large number of antennas. We note that the
capacity increases asymptotically as the number of subcarriers increases.
It can also be observed that the gap between the two curves get wide as
the number of subcarriers increases and increasing the number of sub-
carriers is more effective for a large number of antennas than a small
number of antennas because the duality of subcarriers and antennas
only exists when the number of antennas is very large.
2 Kuo, P.-H., Kung, H.T., and Ting, P.: Compressive sensing based
channel feedback protocols for spatially-correlated massive antenna
arrays. IEEE Wireless Communications and Networking Conf., Paris,
France, April 2012, pp. 492497
3 Rusek, F., and Persson, D.: Scaling up MIMO: opportunities and chal-
theoretical
lenges with very large arrays, IEEE Signal Process. Mag., 2013, 30, (1),
pp. 4060
4 Ngo, H.Q., Larsson, D., and Marzetta, T.: Energy and spectral efciency
20 25 30 35 40 of very large multiuser MIMO systems, IEEE Trans. Commun., 2013,
61, (4), pp. 14361449
5 Mohammed, S., and Larsson, D.: Per-antenna constant envelope precod-
ing for large multi-user MIMO systems, IEEE Trans. Commun., 2013,
61, (3), pp. 10591071
6 Bannour, A., Ammari, M.L., Sun, Y., and Bouallegue, R.: The capacity
performance of ASTC-MIMO-OFDM system in a correlated Rayleigh
frequency-selective channel, Wirel. Pers. Commun., 2013, 68, (4),
pp. 13651376
7 Bannour, A., Sun, Y., Ammari, M.L., Delestre, F., and Bouallegue, R.:
A novel algebraic carrier frequency offset estimator for
ASTC-MIMO-OFDM systems over a correlated frequency-selective
channel, IEEE Trans. Veh. Technol., 2012, 61, (6), pp. 24682475
8 Bannour, A., and Sun, Y.: Frequency synchronization performance of
algebraic space time coded OFDM system within correlated channel.
19th European Wireless Conf., 2013, University of Surrey, Guildford,
UK, April 2013

ELECTRONICS LETTERS 9th July 2015 Vol. 51 No. 14 pp. 11151117