You are on page 1of 5

Performance Analysis of Massive MIMO

with/without Hardware Impairments


AbstractMassive MIMO (multiple-input multiple-output) is a In this model, y is a K 1 vector, representing signals
fast-growing and is foreseen as one of the main technology received by users. P is the transmit power from
components in next generation cellular communication (5G).
Massive MIMO technology can offer some big advantages in the transmitter. H = [h1, h2, hk]T is a K M channel
wireless communication, such as high capacity and energy matrix with hk being a 1 M row vector. The precoding
efficiency and low cost of building. matrix W = [w1, w2, wk] is a M K matrix. Wk is a M
1 column vector. x denotes the transmitted symbols.
While Massive MIMO provides some benefits to the wireless
communication, it uncovers some challenging problems.
Its a K 1 column vector. And zero-mean complex
Hardware impairment is one of the challenging problems. Its Gaussian noise vector is modeled as e, a K 1 vector.
caused by low-cost components used to build Massive MIMO. In To compensate the increase in the number of antennas
this paper, the fundamental limit of massive MIMO downlink resulting the increase in array gain, the transmitted power
system has been derived for Matched filter and Zero-forcing
is scaled by 1/M. Therefore, a constant signal-to-noise
precoders. The performance of system with imperfect channel
information is also investigated. ratio (SNR) is obtained for a given number of antennas
elements. In other words, SNRt for given M antennas
elements in an interference free environment is given by
I. INTRODUCTION
Massive MIMO has been seen as a key technology in P = N0 SNRt / M (2)
the future wireless communication. It employs large- No is the power of the noise.
scale antenna systems in base station to communicate
with many single-antenna terminals at the same time. III. SYSTEM WITH PERFECT CHANNEL STATE
Massive MIMO can handle large orders of magnitude INFORMATION
data traffic and achieve high capacity and radiated power In this section, we assume that the BS has perfect
efficiency. However, due to its very large dimensions, channel state information (CSI). We derive the formula
massive MIMO must be built on inexpensive and low- for signal-interference-noise ratio (SINR) and sum rate.
power consumption components in practical. Using Then the SINR and sum rate are derived for matched
cheap components inevitable causes side effects on the filter (MF) and zero-forcing (ZF) precoder under IID
performance of massive MIMO, such as power distortion Rayleigh channel model. Some analytical and simulation
and phase noise, as known as Hardware impairments. results are generated and compared to see the agreement.
We start by introducing the SINR for user k for a
In this paper, the performance analysis of downlink given channel matrix H. As we later show, it is given by
massive MIMO system with/without hardware
impairments is presented. The signal-to-interference-
noise ratio and sum rate for matched filter and zero-
forcing precoder are derived for assuming transmitting (3)
signal in IID Rayleigh channel model. The comparisons From (1), we can obtain that
between analytical and simulation results for systems
without hardware impairments are presents. Then the
impact of amplitude and phase noise on the performance (4)
of the system is also investigated. It shows the received signal of the user k. For user k, the
only useful part is and the rest of the rest of
II. SYSTEM MODEL the received signal are interference and noise. Thus,
The system model proposed below models a base SINR for user k is calculated by the power of
station (BS) that has M antennas serves K co-scheduled divided by the power of the rest of the received signal,
users with no external interference. which is the exactly same as (3).
Using (3), the maximum achievable data rate for user k is
(1) obtained by
(5) The ZF precoding vector for user k is given by
Therefore, the maximum sum rate is obtained by adding
up the maximum achievable data rate of all served users,
given by
(15)
The H+ is the pseudoinverse of channel matrix H and
. The ||.|F means the Frobenius norm.
(6) The precoding matrix can be approximated by
In IID Rayleigh channel model, we assume that the
amplitude of a signal that will vary or fade according to (16)
Rayleigh distribution. Due to the difficulty of calculating In order to solve for c, first we need to solve the
the exception in (5), a simple accurate approximation expectation of the square of the Frobenius norm of
that is proven by simulation results later is made below pseudoinverse of H. This can be solved by

(7)
where SINRk is given by
(17)
Hence, the normalization constant c is obtained by
(8)
(18)
Substituting (16) into (1), the received signal is the given
Then in next two sections, we derive SINRk and Rk for
by
MF and ZF precoder, respectively.

1) MF (19)
The MF precoding vector for user k is given by Thus, the SINRk is obtained by

(20)
(9)
And the expectations in (8) is given by
3) Analysis of Analytical and Simulation results
(10) In this part, we present some figures that compare the
analytical approximations and simulation results. The
analytical approximations use (12), (20) to calculate the
(11) SINRk for MF and ZF, respectively. The exact data rates
Using (10) and (11), the SINRk is given by are obtained from (3) and (4). The plot shows the average
sum rate vs. number of antennas for different values of K
and SNRt. All figures clearly show an excellent
agreement between analytical and simulation results.
(12)
We can express N0 in terms of SNRt and P by rewriting
(2)

(13)
By substituting (13) into (12), the SINR for user k is
shown as

(14)
2) ZF
Figure 1. Average sum rate vs. number of antennas using MF and
ZF precoders. K = 10 and SNRt = 10 dB.

Figure 4. Average sum rate vs. number of antennas using MF and ZF


precoders. K = 10 and SNRt = 1 dB
Figure 2. Average sum rate vs. number of antennas using MF and
ZF precoders. K = 30 and SNRt = 10 dB.

Figure 3. Average sum rate vs. number of antennas using MF and ZF Figure 5. Average sum rate vs. number of antennas using MF and ZF
precoders. K = 50 and SNRt = 10 dB precoders. K = 10 and SNRt = 15 dB

From Fig. 1 to Fig. 3, these three plots show averages Figure 4, 1 and 5 show average sum rate vs. number of
sum rate vs. number of antenna elements when the SNRt antenna elements when SNRt changes from 1 dB (1.259)
is fixed at 10 dB and K increases from 10 to 30 to 50 for to 10 dB (10) to 15 dB (31.623) and K is fixed at 10 for
MF and ZF precoder, respectively. The results show that MF and ZF precoder, respectively. The results clearly
average sum rate increases as number of user increasing, show that the capacity limit increases as the SNRt
when SNRt is constant. It also can be seen that ZF increasing for same number of served uers. Also, the
precoding technology usually has a larger average sum performance of MF precoder and ZF precoder become
rate than MF precoding method does for the same K and almost the same when the signal-to-noise ratio is
SNRt. In additional, as the number of antenna elements relatively small (1dB).
increases, the average sum rate of ZF precoder and MF
precoder seems converge to the same capacity limit. IV. SYSTEM WITH IMPERFECT CHANNEL STATE
INFORMATION
In this section, we analyze the impact of hardware
impairments by introducing a general model of channel
matrix H with imperfect CSI. This model includes
amplitude and phase error in the channel matrix. It is
given by
(21)
The hkm denotes the true channel matrix between user k
and antenna k. and are amplitude error and phase
error in the channel matrix, respectively. These two
errors represent two of the largest hardware impairments:
the power amplitude distortion and phase noise. They are
assumed to be independent, zero-mean Gaussian noise. a
and are standard deviation of amplitude and phase
error. Using the (21) and defined variables before, we can
derive the SINRk for MF with imperfect CSI.
1) MF
The MF precoding vector with error is given by

Figure 6. Relative average sum rate vs. phase error for MF precoder.
K = 10, SNRt = 10 dB.
(22)
The norm in the denominator can be approximated by its
expected value in order to simplify calculation. This
expectation is obtained by

(23)
and

(24)
The proof of (23) and (24) are shown in the [2, P. 2].
Hence, the SINRk in (8) is given by

(25)
Figure 7. Relative average sum rate vs. phase error for MF precoder. K
Here, / is the factor that reduce the error- = 50, SNRt = 10 dB
free SINRk. If the error is small, the approximation of
error factor is given by

(26)
The is the total error variance which is equal to
. We can see that the error factor is independent
of the number of antenna elements. Its only affected by
the standard deviation of amplitude and phase error.

2) Analysis of Simulation results


In this part, we analyze the impact of amplitude and
phase error independently for MF precoder with different
K and SNRt by comparing simulation results using (25)
and (26). Figure 8. Relative average sum rate vs. phase error for MF precoder. K
= 50, SNRt = 1 dB
These three plots show the relative average sum rate vs.
phase error with 50, 100, 250 and 500 antenna elements.
To determine the impact of phase error to the average
sum rate, the amplitude error is set to 0. From Fig. 6 and
Fig. 7, as number of served users K increases from 10 to
50 while SNRt hold the same, the relative average sum
rate decreases, implying that the impact of phase error
become bigger when the number of users in the system
become more and more. From Fig. 6 and Fig. 7, when
hold K constant and decrease SNRt, the relative average
sum rate decreases. It shows that the phase error will
cause more problems when SNRt is small. In addition, we
can clearly see that the more antenna elements in the
system, the less impact of the phase error on the average
sum rate. Figure 11. Relative average sum rate vs. amplitude error for MF
precoder. K = 50, SNRt = 1 dB

Figure 9 11 show the relative average sum rate vs.


amplitude error with 50,100, 250 and 500 antenna
elements. We set the phase error to 0 to investigate the
impact of amplitude error on the system performance.
From Fig .9 and Fig. 10, Its clear that the amplitude
error has a bigger impact on the average sum rate when
increasing the number of users. From Fig .10 and Fig .11,
we can also see that reducing SNRt will cause the impact
of amplitude error bigger. Also, the system is more
resistible to the amplitude error if the system has more
antenna elements.
V. CONCLUSION
In this paper, a performance analysis of downlink
Figure 9. Relative average sum rate vs. amplitude error for MF massive MIMO system is presented. The SINRk and Rk
precoder. K = 10, SNRt = 10 dB without hardware impairments for MF and ZF precoders
are first derived. Then using SINRk and Rk to
approximate the average sum rate for MF and ZF
precoders. The comparison between analytical and
simulation results has an excellent agreement. The results
show that MF and ZF precoders approach to the same
capacity limit as the number of antenna elements is large.
Furthermore, the impact of hardware impairments on
system performance is investigated for MF precoder. The
general model of imperfect channel matrix is introduced.
The SINRk with incorrect CSI is then derived. In the
performance analysis, the impact of amplitude noise and
phase noise are solely investigated. The results show that
the system containing more antenna elements is more
resistible to the hardware impairments.

Figure 10. Relative average sum rate vs. amplitude error for MF
precoder. K = 50, SNRt = 10 dB

You might also like