Received 2 September 2022, accepted 21 September 2022, date of publication 30 September 2022, date of current version 7 October 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3210938

Closed-Form Bit Error Probability of ZF Detection

for OFDM-M-MIMO Systems Using Effective
Noise PDF
DITSAPON CHUMCHEWKUL 1 AND CHARALAMPOS C. TSIMENIDIS 2, (Senior Member, IEEE)
1 Schoolof Engineering, Newcastle University, NE1 7RU Newcastle upon Tyne, U.K.
2 Department of Engineering, Nottingham Trent University, NG1 4FQ Nottingham, U.K.

Corresponding author: Ditsapon Chumchewkul ([email protected])

This work was supported in part by the National Science and Technology Development Agency (NSTDA) under Grant ST G5671, and in
part by the Engineering and Physical Sciences Research Council (EPSRC) under Grant EP/R002665/1.

ABSTRACT This paper derives the closed-form bit error probability (BEP) of massive multiple-input,
multiple-output (M-MIMO) systems using orthogonal frequency-division multiplexing (OFDM) and zero-
forcing (ZF) detection. We improve the BEP accuracy by increasing the Neumann series expansion (NSE)
to second order for the system that we previously analyzed by deriving the probability distribution function
(PDF) of the effective noise. The proposed PDF is then utilized to evaluate the BEP, the PDF of output
signal-to-noise ratio (SNR), and the outage probability as a function of the output SNR of the system.
Furthermore, a simplified closed-form expression for the effective noise PDF, in terms of the Gaussian
distribution, and the noise variance are firstly derived in this paper for simplifying the performance analysis.
Monte-Carlo simulation results confirm that the outcome from the derived equation and the approximation
closely matched those obtained by simulation. In addition, we employ the proposed noise variance to estimate
the log-likelihood ratio (LLR) instead of the approximate noise variance for the low-complexity soft-output
ZF detection. The computational complexity of the proposed detection is thus significantly reduced, whereas
its bit error rate (BER) is lower than that of the classical detection. Focusing on a 10 × 200 Coded-OFDM-
M-MIMO system, 97.81% of multiplications, required for producing the LLR from the estimated symbol,
were minimized by utilizing the proposed detection. Therefore, the derived equations can be efficiently used
for analyzing the performance of OFDM-M-MIMO systems, and reducing the computational complexity of
the soft-output ZF detection.

INDEX TERMS Massive multiple-input, multiple-output (M-MIMO), Neumann series expansion (NSE),
orthogonal frequency-division multiplexing (OFDM), zero forcing (ZF) detection.

I. INTRODUCTION antennas over the same time slots, the received signal at the
Massive multiple-input, multiple-output (M-MIMO) has BS includes the effects of both co-channel interference (CCI)
become a promising technique to enhance the spectral and and thermal noise. Therefore, a M-MIMO symbol detector is
energy efficiencies for the 5th generation (5G) cellular net- required at the receiver to estimate the transmitted symbols
works [1]. Both the mobile terminals (MT) and the base from the CCI and noise contaminated signal. Although sev-
station (BS) of M-MIMO systems can use several antennas eral symbol detection techniques can be chosen for receiver,
for data communication, and the spatial multiplexing tech- the number of transmit and receive antennas in M-MIMO
nique is utilized for enhancing the data transmission rate. system is very large, and the complexity of M-MIMO detector
Since the MT sends their messages using multiple transmit greatly increases according to the number of antennas. As a
result, the computational complexity of M-MIMO detection
The associate editor coordinating the review of this manuscript and
becomes a critical factor for the implementation of M-MIMO
approving it for publication was Luca Barletta. systems in practice [2]. Focusing on an uplink M-MIMO

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
104384 VOLUME 10, 2022
D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

system, where the number of the receive antennas is cho- interference-to-noise ratio and the output SNR at high SNR
sen to be larger than that of the transmit antennas, the region were approximated. The equation for the latter was
Gram matrix of the system tends to be a diagonal matrix. then utilized to derive a tight approximation of the BEP,
As a result, linear symbol detection provides an excellent outage probabilities, diversity-multiplexing gain trade-off for
result to enhance the spectral and energy efficiencies of the the system. The authors also demonstrated a gap between the
system [1]. Maximal-ratio combining (MRC) detection has performance of MMSE and ZF detection, which cannot be
become an attractive technique to detect transmitted infor- decreased at a higher SNR region. The gap was then rewritten
mation symbols in practical M-MIMO systems, since the in terms of a Hermitian quadratic form in [18], and the authors
detector requires only a matrix multiplication to produce an employed the equation to derive the PDF of SNR, the sym-
estimated symbol. However, for a medium-size M-MIMO bol error rate and the outage probabilities of the M-MIMO
system, the number of the receive antennas is not much larger system. A BEP analysis for the MIMO system over corre-
than that of the transmit antennas and the outcome from lated Rician and Rayleigh fading was derived in [19]. The
the MRC detection still contains remaining effects of CCI, noncentral Wishart distributed matrix was utilized to approx-
thus, degrading system performance. Therefore, zero-forcing imate the distribution of the Gram matrix, and the authors
(ZF) and minimum mean square error (MMSE) detection proved that the PDF of the SNR can be approximated as an
are still required for this case. These detectors use inverse infinite linear summation of the Gamma distribution. Channel
matrix operation to recover the contaminated symbol at the estimation is generally employed to the receiver to estimate
receiver and the computational complexity of the detection the information, making a deviation between the actual and
significantly increases according to the number of antennas. the estimated information. As a result, the effects of channel
Thus, Neumann series expansion (NSE), Gauss-Seidal itera- estimation errors in ZF detection and the BEP approximation
tion, and other approximations can be utilized to reduce the were then investigated by a number of research works. In
computational complexity at the receiver [3], [4], [5], [6], [20], the authors utilized Taylor expansion to approximate the
[7], [8], [9]. If the number of the receive antennas is much pseudo-inverse of the channel matrix and employed Gaussian
larger than that of the transmit antennas, the outcome from random variable to model the channel estimation errors. The
the approximation significantly matches that of the general SNR and the BEP for the system using M -PSK and M -QAM
detection [3]. were then derived in the work. The authors also demonstrated
ZF detection is a classical linear symbol detection, which that the PDF of the SNR exhibits the Chi-square distribution.
uses Moore-Penrose inverse matrix operation to recover con- The BEP analysis was then extended to the M-MIMO system
taminated symbols at the receiver. Generally, the bit error and a lower SNR region in [21]. Improved SNR and BEP
probability (BEP) of the system utilizing ZF detection is analysis for M -QAM were introduced in the work. The BEP
higher than that of MMSE detection for low signal-to-noise for MRC, ZF, and MMSE detection in M-MIMO system
ratio (SNR). However, the computational complexity of the using pilot symbols to estimate the channel state information
detection is less than that of the MMSE detection since the was derived in [22]. The average SNR and the BEP were
SNR information is not required by the detector [10]. Fur- derived in the work.
thermore, the difference in BEP of ZF and MMSE detection According to the literature survey, the BEP analysis for
at a higher SNR region is marginal. Therefore, a number of the MIMO system using ZF detection and BPSK modulation
research works focused on BEP performance for this type of over uncorrelated frequency-flat Rayleigh fading channel at
detection. Exponential tight bounds on BEP of the data com- high SNR region was previously proposed in [17]. The BEP
munications using multiple antennas and optimal combin- for M -QAM and other type of modulation schemes was
ing were proposed for flat and frequency-selective Rayleigh neglected in the paper, however, the readers can employ the
fading in [11] and [12]. The numerical results confirm that PDF of the derive SNR to derive the BEP using the equa-
user capacity can be significantly enhanced by increasing tions in [23], [24]. Although the outcome from the derived
the number of the antennas. The BEP of the MIMO system equation significantly matched the exact BEP, the distribution
using ZF detection was then analyzed under various fading of the effective noise in the work was assumed to be the
channels in [13] and [14]. The authors employed the complex Gaussian distribution. Therefore, an accurate effective noise
Wishart distribution and a derived probability distribution PDF of ZF detection is still required to derive the BEP of the
function (PDF) of SNR in [15] to approximate the PDF of system utilizing channel coding techniques [24]. Moreover,
the Gram matrix. A closed-form expression for the BEP was most research work involving low-complexity linear symbol
also proposed in the work. In [16], the BEP of the system over detection in [3], [4], [5], [6], [7], [8] and [9] assumed that the
uncorrelated Rayleigh fading channel was approximated. The effective noise after the detection exhibits the Gaussian dis-
Gram matrix was assumed to be a diagonal matrix and the tribution. Thus, the soft-output linear symbol detection used
BEP was then derived employing the Gaussian approxima- the Gaussian assumption to produce the log-likelihood ratio
tion. The output SNR, BEP, outage probabilities, diversity- (LLR) from the estimated transmit symbol. However, as far
multiplexing gain trade-off, and SNR gain of the MIMO as the research literature goes, there is no prior research work
system using ZF and MMSE detection over uncorrelated flat that derives a closed-form expression of the noise variance.
Rayleigh fading were extensively investigated in [17]. The As a result, the soft-output linear symbol detector requires

VOLUME 10, 2022 104385

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

most of the arithmetic operations to evaluate the exact noise these derived PDFs to analyze BEP performance, the PDF of
variance for producing the LLR. the output SNR, and the outage probability of the OFDM-M-
In this paper, we introduces an accurate PDF of the MIMO system utilizing ZF detection in sections VI to VIII,
effective noise of ZF detection for orthogonal frequency- respectively. Furthermore, the operation of the proposed
division multiplexing-M-MIMO (OFDM-M-MIMO) system low-complexity soft-output ZF detection is summarized in
for Gray-coded BPSK, QPSK, and M -QAM modulation the section IX, where the derived noise variance is used
types. This work is motivated by a derived effective noise to produce the LLRs. We use Monte-Carlo simulations to
PDF for the system, proposed in [25]. The authors in the work demonstrate the validity of the proposed equations and obtain
used NSE to approximate the inverse matrix operation as a numerical results to evaluate the system performance, and the
Taylor series, and the effective noise PDF was then derived results are discussed in the section X.
from the joint probability of the involved random variables. Notation: In this paragraph, we clarify the notation of
The BEP of the system with Gray-encoded M -QAM was mathematical symbols used in this paper. We use bold charac-
analyzed utilizing the derived PDF. In addition, the pairwise ters H to denote matrix and vector variables. Their elements
error probability of the system was then determined from the λ represents
are written as Hn,m . Let λ = I , Q, the symbol Hn,m
PDF and was used to evaluate the upper-bounds of coded the in-phase and the quadrature component of Hn,m . The oper-
systems. Although the BEP analysis in [25] provided accu- ator ( .)† denotes the Hermitian transpose of a matrix or vector
rate analytical results, only the 1st order of Taylor’s series and is equivalent to complex conjugate ( .)∗ transposition
†
∗
was used by the research work, and the Gram matrix was ( .) , e.g., Hf = (Hf ) . 0(a) represents Gamma function
T T
assumed to be a diagonal matrix. As a result, there is still a of a, and the upper incomplete Gamma function is 0(a, b).
small deviation between the analytical results and the exact
BEP. More accurate analytical results could be theoretically
achieved if a higher order of Taylor’s series is chosen for the II. SYSTEM MODEL
analysis. Therefore, in this paper, we employ the NSE with Fig. 1 illustrates the block diagram of an uplink OFDM-M-
the 2nd order Taylor series to analyze the effect of noise in MIMO system considered in this paper. The M -QAM mapper
ZF detection and its PDF. The PDF is additionally utilized requires N = Nb Nt Nf bits, where Nb = log 2(M ). Nt
to determine the BEP performance, the PDF of output SNR, represents the number of transmit antennas, and Nf is the
and outage probability as a function of the output SNR of the number of sub-carriers in frequency domain. M denotes the
OFDM-M-MIMO system. QAM order and additionally corresponds to the number of
Although the derived equation provides an accurate result, points in the signal constellation. In practice, Nf is chosen
their computational complexity is still too high. Therefore, to be a power of 2 so that the inverse fast-Fourier transform
we firstly prove that the effective noise PDF of ZF detec- (IFFT) can be used to transfer the M -QAM symbols from
tion tends to become a zero-mean, complex-valued Gaussian frequency to time domain. The M -QAM mapper produces
distribution, and a simplified closed-form expression of the a symbol matrix, X ∈ CNt ×Nf , and the IFFT is applied
noise variance is then derived. In practice, the analytical row-wise to obtain a matrix 9 ∈ CNt ×Nf . Subsequently, Ncp
results from the approximate PDF are closely matching those cyclic prefix (CP) samples are appended to 9 to eliminate the
from the exact equation if the number of the receive antennas effect of inter-block interference (IBI) induced by the delay
is larger than 100, and the ratio of number of the receive spread, τ , of the M-MIMO multipath channels. To avoid IBI,
antennas to that of the transmit antennas is larger than 10. Ncp is chosen to be greater than τ . The resulting symbol is
As a result, the classical Gaussian distribution approximation transmitted through the frequency-selective Rayleigh fading
can be also utilized to simplify the performance analysis M-MIMO channel.
of OFDM-M-MIMO system for special cases. Furthermore, At the BS, the cyclic prefix samples are removed from the
we utilize the Gaussian approximation to reduce the com- received signal and a fast-Fourier transform (FFT) is applied
putational complexity of the soft-output ZF detection. The row-wise to transfer the signal in frequency domain and pro-
proposed detector uses the derived noise variance for produc- duce Y ∈ CNr ×Nf , where Nr is the number of receive anten-
ing the LLR instead of the exact noise variance. Since the nas. This matrix is used by the M-MIMO detector to obtain
effective noise tends to become a Gaussian distribution, the the estimate, X̂, of the transmitted symbol matrix, X. Finally,
BEP of the proposed detection is theoretically close to that the receiver uses an M -QAM demapper to estimate the trans-
of the classical detection, whereas its operational complexity mitted N information bits. By designing Ncp > τ , the IBI
can be significantly minimized. is assumed to be completely eliminated. As a result, the
The remainder of this article is organized as follows. received signal in frequency domain can be expressed as a
A block diagram of the OFDM-M-MIMO system and the linear matrix equation. Let f = 0, 1, . . . , Nf − 1 denotes
channel model are described in the next section. In section the sub-carrier index, and Xf ∈ CNt ×1 be the vector of
III, the operation of classical ZF detection is summarized. The the transmit symbols of the f -th sub-carrier, with elements
effective noise of ZF detection and its PDFs are then analyzed {Xm }N t
m=1 . The value of Xm is chosen from a square-shaped
in section IV. The Gaussian approximation of the effective M -QAM constellation, and the probability of the occurrence
noise of ZF detection is derived in section V. We also use of each symbol is identical, i.e., 1/M . The average transmit

104386 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

energy per symbol is denoted as Es . By assuming that the where the diagonal and off-diagonal components of Gf are
IBI is completely eliminated by utilizing a CP, the received represented as Df and Ef , respectively. Here, the elements
PNr
symbol vector of the f -th sub-carrier Yf ∈ CNr ×1 can be of Df and Ef are denoted as Dm,m = 2
n=1 |Hn,m | and
PNr
expressed as ∗
Em,n = k=1 Hk,m Hk,n , respectively. Generally, the order
Yf = Hf Xf + Wf , (1) L of Taylor series in (5) should be large enough to minimize
the differences between G−1f from (5) and their actual inverse
where Hf ∈ CNr ×Nt is channel frequency response (CFR) matrix operation. However, since Nr in M-MIMO system is
matrix of the f -th sub-carrier. If there is no line-of-sight much larger than Nt , the results from NSE with L = 2 can
link between transmitter and receiver, each component Hn,m provide an accurate result for this case. If G−1
f in (5) and
in Hf is represented by zero-mean, complex-valued, Gaus- L = 2 are substituted in (4), Zf becomes
sian random variables CN (0, 2σh2 ) [26]. Their variance per
†
f Ef )Df Hf Wf .
Zf = (I − D−1 −1
dimension σh2 is 0.5 so that the channel energy is normalized. (6)
Furthermore, Wf ∈ CNr ×1 represents the vector of addi-
† PNr
tive white Gaussian noise (AWGN). Each element, {Wn }N r Let Af = Hf Wf , with elements αm = ∗
n=1 Hn,m Wn ,
n=1 ,
in Wf exhibits a zero-mean, complex-valued, Gaussian dis- each component of Zf in (6) can be written as
tribution, i.e., CN (0, 2σw2 ), and σw2 is determined by 
Nt

1  X Em,n αn 
Nt Es Zm = αm − . (7)
σw2 = , (2) Dm,m Dn,n
2 log2 (M )γb n=1,n6=m

where γb is the signal-to-noise ratio (SNR) per bit. It is It is worth noting that a higher order L of Taylor series in
worth pointing out that data communication links using (5) is not applied for this research work since the mathemat-
OFDM waveforms are generally degraded due to the car- ical expression of Zm for this case involves with dependent
rier frequency offset (CFO) and channel estimation errors. variables. As a result, the derived PDF, obtained by joint
As a result, more accurate mathematical expressions of the probabilities of the random variables, is still inaccurate.
received symbol and the performance analysis were essen-
tially proposed by a number of research work [27], [28], IV. DERIVING EFFECTIVE NOISE PDF OF ZF DETECTION
[29], [30]. However, considering CFO and other impair- Zm , in (7), shows the effects of noise for the received symbol
ments such as IQ imbalance and frame synchronization in of ZF detection, and their distribution depends on Nt , Nr ,
the system model is beyond the scope of this work. Uti- and the operating SNR point of the OFDM-M-MIMO system.
lizing all these impairments simultaneously to analyze the We now use this equation to derive the PDF of Zm . Due to the
performance of the OFDM-M-MIMO system will become fact that Dm,m , Em,n , and αm in (7) are functions of random
too complex. Therefore, the PDF of Hn,m in (1) is assumed to variables, their joint probabilities can be used to derive their
be the Gaussian distribution for simplifying the performance PDFs. However, there are several arithmetic operation in
analysis. (7), and determining the PDF of Zm is complex. Therefore,
approximations are now utilized to simplify this operation.
III. CLASSICAL ZF DETECTION The PDF of Dm,m was previously derived from (8) in [25].
In this section, we discuss the operation of ZF detection and Since Dm,m can be rewritten as a 2Nr times summation of
the effect of noise in its performance. The detection employs squared, zero-mean, Gaussian random variables with vari-
Moore-Penrose pseudoinverse of the CFR matrix to estimate ance, σh2 , the PDF of Dm,m , pD (Dm,m ), exhibits a Chi-square
the transmitted symbol from Yf as [2] distribution and tends to be the N (2Nr σh2 , 2Nr σh2 ) if Nr is
† † large enough [31]. Focusing on an OFDM-M-MIMO system
X̂f = (Hf Hf )−1 Hf Yf . (3)
where Nr = 100, the probability of the occurrence of Dm,m ,
Substituting Yf from (1) in (3), we obtain ranging from 85 to 115, is 86.76%. Therefore, if Nr ≥ 100,
† most of the variation of Dm,m is less than 15% of its mean and
f Hf W f ,
X̂f = Xf + G−1 (4) Dm,m can be approximated as a constant 1, i.e.,
†
where the Gram matrix is denoted as Gf = Hf Hf ∈ CNt ×Nt . 1 = 2Nr σh2 . (8)
The receiver uses X̂f in (3) to determine the received message,
and this symbol is usually contaminated by the effective noise If 1 in (8) is substituted into (7), Zm becomes
†
Zf = G−1 f Hf Wf . Let Zm denotes the m-th vector element in αm 1
Nt
X
Zf ∈ CNt ×1 . In order to analyze the distribution of Zm , the Zm = − 2 Em,n αn . (9)
1 1
NSE approach is utilized to approximate the inverse matrix n=1,n6=m
operation of the Gram matrix [3], i.e., PNt
Let α̃m = αm /1, βm = n=1,n6=m Em,n αn and β̃m =
L−1
X βm /12 , (9) can be expressed as
G−1 f Ef ) Df ,
(−D−1 l −1
f ' (5)
l=0 Zm = α̃m − β̃m . (10)

VOLUME 10, 2022 104387

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

FIGURE 1. Block diagram of OFDM-M-MIMO system.

Undoubtedly, αm , βm , α̃m , β̃m and Zm are a function of CFR

Nr
X (Nr + k − 2)!|αm λ |Nr −k
× .
and noise vector element Wn . Therefore, their PDFs can be 2Nr +k−1 (σh σw )Nr −k+1 (Nr − k)!0(k)
k=1
derived utilizing joint probabilities as described in the next
(14)
sub-sections.
In addition, if Nr is large enough, pα (αm λ ) can be approxi-
A. THE APPROXIMATE PDF OF Em,n mated as a N (0, σα ), and their variance per dimension is given
2
Em,n is an off-diagonal element of Gf . The in-phase and as
quadrature components of Em,n can be expressed as
σα2 = 2σh2 σw2 Nr . (15)
Nr Nr
∗,Q Q The PDF of βm is also derived utilizing the joint proba-
X ∗,I I
X
I
Em,n = Hk,m Hk,n − Hk,m Hk,n , (11a)
k=1 k=1 bilities of random variables. βmλ , in (9), is 2(Nt − 1) times
Nr Nr the summation of the product of Em,n λ and α λ . If N is large
m r
Q ∗,Q
X ∗,I
X
Q
Em,n = Hk,m Hk,n + I
Hk,m Hk,n . (11b) enough, these variables tend to be a Gaussian distribution.
k=1 k=1 Therefore, we can use (6.9) in [32] to determine the PDF of
Due to the fact that Hn,mλ is a Gaussian random variable βmλ . As a result, pβ (βmλ ) becomes as (16), shown at the bottom
λ of the page.
with λ = {I , Q}, Em,n , in (11), is 2Nr times the summation
of the product of two N (0, σh2 ). For the system with large
λ tends to become a Gaussian distribution C. PDF OF α̃m AND β̃m
Nr , the PDF of Em,n
α̃m and β̃m in (10) are ratio of αm and βm to constants,
N (0, σE ), and the variance per dimension is given as
2
λ ), p (β λ ), and
and their PDF can be obtained utilizing pα (αm β m
σE2 = 2Nr σh4 . (12) (4.19) in [31], i.e.,
λ λ
B. DERIVING PDF OF αm AND βm
pα̃ (α̃m ) = 1pα (1α̃m ), (17a)
Af is a product of
†
Hf
and Wf . Therefore, the in-phase and pβ̃ (β̃mλ ) =1 2
pβ (12 β̃mλ ). (17b)
quadrature elements of αm , in (7), can be expressed as λ ), p (β λ ), and 1 from (14), (16), and (8)
Substituting pα (αm β m
Nr Nr λ and β̃ λ become
into (17), the PDFs of α̃m
X X m
αm
I
= ∗,I
Hn,m WnI − Hn,m Wn ,
∗,Q Q
(13a) 
λ|

Nr 2Nr σh |α̃m
n=1 n=1 − σw
λ λ Nr −k
X
Nr Nr pα̃ (α̃m )= fα̃ (k)|α̃m | e , (18a)
X X
αm
Q
= ∗,I
Hn,m WnQ + Hn,m Wn .
∗,Q I
(13b) k=1

λ|

t −1
NX 2Nr σh |β̃m
n=1 n=1 − σw
λ in (13) is 2N times the summation
pβ̃ (β̃mλ ) = fβ̃ (k)|β̃mλ |Nt −k−1 e . (18b)
Due to the fact that αm r k=1
of the product of Hn,m and Wnλ , their PDF is determined
∗,λ
The function fα̃ (k) and fβ̃ (k) in (18) are defined as
utilizing (6.9) of [32] and the results from this operation are
|α λ | (Nr σh )Nr −k+1 (Nr + k − 2)!
 
λ 1
pα (αm )= exp − m fα̃ (k) = , (19a)
0(Nr ) σh σw 22k−2 σwNr −k+1 0(Nr )0(k)(Nr − k)!

! N −1
1 |βmλ | t
(Nt + k − 3)!|βmλ |Nt −k−1
pβ (βmλ )
X
= exp − . (16)
22Nt −2 2Nr σh3 σw k=1
(Nr σh3σw )Nt −k(Nt − 2)!(Nt − k − 1)!0(k)

104388 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

(Nr σh )Nt −k (Nt + k − 3)! Substituting σE2 and σα2 from (12) and (15) into (24), this
fβ̃ (k) = . (19b)
22k−2 σwNt −k 0(k)(Nt − 2)!(Nt − k − 1)! equation becomes

D. EFFECTIVE NOISE PDF OF ZF DETECTION (Nt − 1)σw2

σβ̃2 = . (25)
According to (10), Zm is a subtraction of α̃m and β̃m . Due 2Nr2 σh2
λ ) and p (β̃ λ ) are even functions, the PDF of Z λ is
to pα̃ (α̃m β̃ m m Due to the fact that Zm , in (10), is constructed by subtracting
analyzed by utilizing convolution integral operation [31], i.e., α̃m from β̃m , and both of them are approximated by a Gaus-
pz (Zmλ ) = pα̃ (α̃m
λ
) ∗ pβ̃ (β̃mλ ). (20) sian distribution, the PDF of Zmλ becomes a N (0, σz2 ) too, i.e.,
(Z λ )2
 
1
If the PDFs of α̃m λ and β̃ λ from (18) are substituted into
m pz (Zmλ ) ' q exp − m 2 . (26)
2π σz2 2σz
(20) and the binomial expansion is employed to decompose
the polynomial in the equation, the results from the integral
operation becomes (21), as shown at the bottom of the page, σz2 can be determined from sum of σα̃2 in (23) and σ 2 in (25),
β̃
i.e., the PDF of the effective noise of ZF detection. The and the result from this operation is
function fz (k, p) in (21) is defined as σ2

1 Nt − 1

σz2 = w2 + . (27)
(Nr + k − 2)!(Nt + p − 3)! σh 2Nr 2Nr2
fz (k, p) =
22k+2p−4 (Nt − 2)!(Nr − k)!0(Nr )0(k)0(p)
VI. BEP ANALYSIS
Nr σh Nt +Nr −k−p+1
 
× . (22) In the previous section, we have derived the effective noise
σw PDF of ZF detection in (21) as well as their Gaussian approx-
λ) imation in (26). We now use these PDFs to evaluate the BEP
V. THE GAUSSIAN APPROXIMATION OF pz (Zm
performance of the proposed OFDM-M-MIMO system. If the
Although pz (Zmλ ) in (21) provides an accurate noise PDF
transmitter uses BPSK, QPSK or squared-shape M -QAM
of ZF detection, their computation is complex and using
modulation with Gray encoding, the receiver estimates the
this equation to determine performance of OFDM-M-MIMO
transmitted message from the in-phase and quadrature com-
system is still inconvenient. Therefore, we now prove that the
ponents of the received symbols independently. Let Mλ rep-
PDF of Zmλ can be alternatively approximated as a classical
resent the number of constellation points per dimension, the
Gaussian distribution. The variance of Zmλ is also analyzed in
λ and E λ can be symbol error probability (SEP) of the system is determined by
this section. If Nr is large enough, both αm m,n averaging the error probabilities of Mλ − 2 inner and 2 outer
approximated as a Gaussian random variable. If we use (17a)
λ , their distribution becomes Gaussian constellation points in each dimension. As a result, the SEP
to derive the PDF of α̃m
can be expressed as (5.105) in [26], i.e.,
too, i.e., N (0, σα̃ ), and the variance of this random variable
2

can be expressed as σα2 /12 , i.e., 2(Mλ − 1)

Ps = P(X̂mλ |Xmλ = X̄mλ , K ). (28)
Mλ
σw2
σα̃2 = . (23) P(X̂mλ |Xmλ = X̄mλ , K ) in (28) denotes the cumulative distribu-
2Nr σh2
tion function (CDF) of Zmλ . The CDF demonstrates a pairwise
In addition, βmλ is 2(Nt − 1) times the summation of the error probability that a constellation point is transmitted by
product of αmλ and E λ . Therefore, it can be approximated
m,n the transmitter and a nearest constellation point X̄mλ is detected
as a N (0, 2σβ2 ), and their variance σβ2 is 2(Nt − 1)σE2 σα2 . by the receiver. The equation is written in terms of pz (Zmλ ) as
By using (17b), β̃mλ is also a Gaussian distribution. Its variance
Z ∞
is determined by P(X̂mλ |Xmλ = X̄mλ , K ) = pz (Zmλ + K )dZmλ , (29)
0
2(Nt − 1)σE2 σα2 where the shortest distance between the constellation points
σβ̃2 = . (24)
14 in each dimension and the decision boundary is denoted as

λ|
 
2Nr σh |Zm t −1
Nr NX −p−1
NtX
− σw 1
pz (Zmλ )
X
=e fz (k, p)
(Nt − p − q − 1)!q!
k=1 p=1 q=0

(−1)q |Zmλ |Nt +Nr −k−p

Nt +Nr −k−p−q
σw

× + (Nt + Nr − k − p − q − 1)!|Zmλ |q
Nr − k + q + 1 4Nr σh

Nt +Nr −k−p−q−1
(−1)q (Nt + Nr − k − p − q − 1)!|Zmλ |Nt +Nr −k−p−r−1
r+1
σw
X 
+ . (21)
(Nt + Nr − k − p − q − r − 1)! 4Nr σh
r=0

VOLUME 10, 2022 104389

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

K , i.e., 1 for 16-QAM using the constellation points per Using P(X̂mλ |Xmλ = K ) from (35) in (32) and (33), the Pe of
dimension {−3, −1, 1, 3}. We then utilize the Ps in (28) to the systems using BPSK, QPSK and 16-QAM modulation
derive the BEP as Ps / log2 (Mλ ) [26], and results in becomes
2(Mλ − 1)
P(X̂mλ |Xmλ = X̄mλ , K ).
 
Pe = (30) 1
Mλ log2 (Mλ )
BPSK,QPSK
Pe 'Q , (36a)
σ
z 
By substituting the PDF of Zmλ from (21) in (29) and 3 1
P16−QAM ' Q . (36b)
utilizing the binomial expansion to expand the polynomial, e
4 σz
P(X̂mλ |Xmλ = X̄mλ , K ) becomes (31), as shown at the bottom of
the page. We use P(X̂mλ |Xmλ = X̄mλ , K ) in the equation to evalu- It is evident that the computation complexity of the approx-
ate the BEP of the proposed OFDM-M-MIMO system in (30). imate BEP equations in (36) is much lower complex than
For example, if BPSK or QPSK modulation is selected, the that of the accurate BEP analysis in (32) and (33). However,
constellation points per dimension, K , and Mλ are {−1, 1}, we have found experimentally that these approximate equa-
1, and 2, respectively. By substituting these variables into tions provide an accurate BEP if Nr ≥ 10Nt and Nr ≥ 100.
(30), the BEP analysis for the system with BPSK and QPSK
modulation becomes VII. DERIVING PDF OF SNR OF ZF DETECTION
In this section, the effective noise PDFs are used to derive the
PBPSK,QPSK
e = P(X̂mλ |Xmλ = X̄mλ , K = 1). (32)
PDF of the output SNR of ZF detection. If a square-shaped,
In addition, if the system utilizes 16-QAM modulation with Gray-coded, M -QAM is selected as modulation scheme, the
the constellation points {−3, −1, 1, 3}, K = 1, and Mλ = 4. detector uses the in-phase and quadrature components of the
Therefore, the BEP in (30) is written as received symbols X̂m to estimate their information messages
3 independently. In addition, the average symbol energy per
P16−QAM
e = P(X̂mλ |Xmλ = X̄mλ , K = 1). (33) dimension for Xmλ is Es /2 and the output SNR per symbol, γs ,
4
of their in-phase and quadrature components can be expressed
It is worth pointing out that the approximate PDF of Zmλ
as
in (26) can be also utilized to minimize the computational
complexity of the BEP analysis. Due to the fact that Zmλ is Es
γs = . (37)
assumed to be N (0, σz2 ) in (26), P(X̂mλ |Xmλ = X̄mλ , K ) in (29) 2(Zmλ )2
becomes
1 Due to the fact that γs in (37) is a function of Zmλ , we use (4.19)
P(X̂mλ |Xmλ = X̄mλ , K ) = q in [31] to analyze the PDF of γs in terms of pz (Zmλ ), i.e.,
2π σz2
s s !
(Zmλ + K )2
Z ∞  
Es Es
× exp − dZmλ . (34) pγ (γs ) = pz . (38)
0 2σz2 2γs3 2γs
R ∞ −t 2
Let Q(x) = √1 x e− 2 dt be the Q-function, (34) can be If the PDF of Zmλ in (21) is substituted in (38), pγ (γs ) becomes
2π
written in terms of the function as (39), as shown at the bottom of the next page. In addition,
if the approximate PDF of Zmλ in (26) is utilized instead of
 
K
P(X̂mλ |Xmλ = X̄mλ , K ) = Q . (35)
σz their accurate equation, the PDF of the output SNR of ZF


2Nr σh K
 Nr Nt −1 −p−1
NtX
1
P(X̂mλ |Xmλ = X̄mλ , K ) = e
− σw
XX
fz (k, p)
(Nt − p − q − 1)!q!
k=1 p=1 q=0

Nt +Nr −k−p Nt +Nr −k−p−s+1
(−1)q (Nt + Nr − k − p)! Ks σw
X 
×
Nr − k + q + 1 s! 2Nr σh
s=0
q Nt +Nr −k−p−s+1
Ks σw
X 
+(Nt + Nr − k − p − q − 1)!q!
22Nt +2Nr −2k−2p−q−s+1 s! Nr σh
s=0
Nt +Nr −k−p−q−1 Nt +Nr −k−p−r−1
X X (−1)q K s (Nt + Nr − k − p − q − 1)!
+
2Nt +Nr −k−p+r−s+2 (Nt + Nr − k − p − q − r − 1)!s!
r=0 s=0
!
σw Nt +Nr −k−p−s+1
 
× (Nt + Nr − k − p − r − 1)! . (31)
Nr σh

104390 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

detection can be approximated as used by (41), this equation can be expressed as

s   s Z γth  
Es Es Es 1 Es
pγ (γs ) ' exp − . (40) Pout ' exp − dγs . (44)
4π σz2 γs3 4γs σz2 4π σz2 0 γs3 4γs σz2
p

By using the substitution method in [34] to solve the integral

VIII. OUTAGE PROBABILITY
operation in (44), Pout becomes
In this section, the outage probability as a function of the  
output SNR of the proposed OFDM-M-MIMO system is Es
Pout ' 0 , 1/2 . (45)
analyzed utilizing the PDF of γs in (39) and the approximate 4σz2 γth
PDF in (40). In general, the outage probability is defined
as the probability that γs is below a threshold γth , and this IX. LOW-COMPLEXITY SOFT-OUTPUT ZF DETECTION
parameter can be evaluated from pγ (γs ) as USING THE DERIVED NOISE VARIANCE
Z γth Channel coding is an efficient technique to reduce the BEP
Pout = pγ (γs )dγs . (41) in data communications. If the technique is employed by the
0 OFDM-M-MIMO system, the receiver requires soft-output
If pγ (γs ) in (39) is substituted in (41), this equation becomes ZF detection to produce the LLRs for channel decoder. Since
(42), as shown at the bottom of the page. After using (3.351.2) the exact noise variance of Zmλ is required by the detector
in [33] and the substitution method in [34] to solve the inte- for producing the LLRs and the computational complexity of
gral operation in this equation, Pout of the OFDM-M-MIMO the operation significantly increases according to the number
system can be written as (43), as shown at the bottom of the of antennas, the detector uses an increased number of the
next page. In addition, if the approximate equation in (40) is arithmetic operation to evaluate the noise variance. Therefore,


Nr σh
q  Nr Nt −1 −p−1
NtX
− σw
2Es
γs
XX 1
pγ (γs ) = e fz (k, p)
(Nt − p − q − 1)!q!
k=1 p=1 q=0
 !1/2
N +Nr −k−p+1
(−1)q Es t
× Nt +Nr −k−p+3
Nr − k + q + 1 2Nt +Nr −k−p+1 γs
q+1
!1/2  Nt +Nr −k−p−q
Es σw
+(Nt + Nr − k − p − q − 1)!
2q+1 γs
q+3 4Nr σh
Nt +Nr −k−p−q−1 N +Nr −k−p−r
!1/2
X (−1)q (Nt + Nr − k − p − q − 1)! Es t
+ Nt +Nr −k−p−r+2
r=0
(Nt + Nr − k − p − q − r − 1)! 2Nt +Nr −k−p−r γs
r+1 !
σw

× . (39)
4Nr σh

t −1
Nr NX −p−1
NtX
X 1
Pout = fz (k, p)
(Nt − p − q − 1)!q!
k=1 p=1 q=0
Nr σh
 q 
2Es
(Nt +Nr −k−p+1)/2 Z γth − σw γs
(−1)q

Es e
× (N +N −k−p+3)/2
dγs
(Nr − k + q + 1) 2 0 γs t r
Nr σh
 q 
2Es
(q+1)/2  Nt +Nr −k−p−q Z γth −
σw σw γs

Es e
+(Nt + Nr − k − p − q − 1)! dγs
2 4Nr σh γs 0
(q+3)/2

Nt +Nr −k−p−q−1 (Nt +Nr −k−p−r)/2  r+1

(−1)q (Nt + Nr − k − p − q − 1)! σw

X Es
+
(Nt + Nr − k − p − q − r − 1)! 2 4Nr σh
r=0
Nr σh
 q 
2Es

Z γth − σw γs
e
× (Nt +Nr −k−p−r+2)/2
dγs  . (42)
0 γs

VOLUME 10, 2022 104391

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

we now utilize the derived noise variance in (27) to reduce the received symbol in the frequency domain Yf , and the
the computational complexity of the detection. The model soft-output ZF detection is then utilized to produce the LLR
of the coded-OFDM-M-MIMO system and the operation of vector, 3f .
classical soft-output ZF detection are summarized in this We now summarize the operation of the classical low-
section. We then describe the operation of soft-output ZF complexity soft-output ZF detection where NSE in [3] is
detector, where the derived noise variance is used. Although employed to minimize the computational complexity. The
several low-complexity soft-output linear symbol detections detector uses (3) to estimate X̂f from Yf . Since the compu-
have been proposed for the M-MIMO system using approx- tational complexity in the equation is significantly increase
imate inverse matrix operation and iterative methods in [3], according to the number of antennas, the L-th order NSE
[4], [5], [6], [7], [8] and [9], most of the works focus on in (5) is employed to approximate G−1 f in the equation [3].
MMSE detection. Since we utilize NSE in [3] to derive the Generally, the distribution of Zm for the m-th antenna is
effective noise PDF, we compare the bit error rate (BER) and 2 ). If BPSK, QPSK, or M -QAM
assumed to be the CN (0, 2σz,m
the computational complexity of the proposed detection with modulation is chosen by the transmitter, the q-th tuple of the
that of the classical soft-output ZF detection, where NSE in LLR for the m-th antenna is then estimated from the in-phase
the work is employed to reduce the calculation of the LLRs or the quadrature components of X̂m as [7]
and the noise variance.
min (X̂mλ − X̄kλ )2 − min (X̂mλ − X̄kλ )2
Most of the operation of the coded-OFDM-M-MIMO X̄kλ ,dm,q =0 X̄kλ ,dm,q =1
system is identical to that of the conventional uncoded 3m,q = 2
. (47)
2σz,m
OFDM-M-MIMO, described in the section II. However, the
transmitter separates the information bits into Nt sub-streams, 2 and X̄ λ in (47) represents the exact noise variance for the
σz,m k
and the code words cm ∈ N1×Ncoded for the m-th trans- m-th transmitter and the constellation points per dimension,
mitter are then generated from the information using the respectively. dm,q denotes the q-th tuple of the information
†
m-th channel encoder, where N = {0, 1} denotes the binary bits from the m-th transmitted symbol. Let UZF,f = G−1 f Hf
set. The code words are then interleaved and grouped into denotes the weight matrix of ZF detection, Zf in (4) can be
log2 (M )-tuples. The information is mapped into a M -QAM rewritten as UZF,f Wf . Therefore, the classical soft-output ZF
constellation using the M -QAM mapper, and becomes the detection estimates σz,m
2 for the m-th transmitter as
M -ary coded symbols. Finally, the IFFT operation produces
Nr
the OFDM transmit symbol 9. If a channel coding technique X
with the coding rate R is chosen by the coded-OFDM-M- σz,m
2
= σw2 |Un,m |2 , (48)
n=1
MIMO system, the size of the code words for the m-th trans-
mitter is larger than the number of the information bits, i.e., where an element in UZF,f is denoted as Un,m . (48) involves
Ncoded = Nb Nf /R. Therefore, the variance of Wn for this case an inversion operation of the Gram matrix, resulting in high-
is now redefined as complexity. Therefore, the 1-st order NSE is employed to
†
Nt Es approximate UZF,f in the equation as D−1
f Hf [3]. As a result,
σw,coded
2
= . (46) the variance in (48) becomes
2 log2 (M )Rγb
Nr
σw2 X
In addition, after removing the CP, the receiver of the coded- σz,m
2
= ∗ 2
|Hm,n | . (49)
OFDM-M-MIMO system uses the FFT operation to estimate D2m,m
n=1


Nr σh
q  Nr Nt −1 −p−1
NtX
− σw
2Es
γth
XX 1
Pout = e fz (k, p)
(Nt − p − q − 1)!q!
k=1 p=1 q=0
 !1/2 Nt +Nr −k−p
(−1) q (N + N − k − p)! E Nt +Nr −k−p+1 1

σw
Nt +Nr −k−p−r+1
t r s
X
×  √
2Nt +Nr −k−p−1 r! γthr Nr σh 2Es
p
Nr − k + q + 1
r=0
q+1 1/2  q
! Nt +Nr −k−p−q X q−r+1
σw σw

Es 1
+q!(Nt + Nr − k − p − q − 1)! √
2q−1 4Nr σh r! γthr Nr σh 2Es
p
r=0
Nt +Nr −k−p−q−1 r+1
σw (−1)q (Nt + Nr − k − p − q − 1)!(Nt + Nr − k − p − r − 1)!
X 
+
4Nr σh (Nt + Nr − k − p − q − r − 1)!
r=0
N +N −k−p−r
! 1/2 Nt +Nr −k−p−r−1 Nt +Nr −k−p−r−s !
Es t r σw

X 1
× N +N −k−p−r−2 √ (43)
s! γths Nr σh 2Es
p
2 t r
s=0

104392 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

Although (49) requires less arithmetic operations than that

of the exact equation in (48) for producing σz,m 2 , the com-

putational complexity of the equation is still high since the

size of Hf is very large. Moreover, the equation employs the
1-st order NSE to approximate the inverse matrix operation,
resulting in a deviation between the outcome from (49) and
the exact value of σz,m2 . Therefore, a higher accuracy and

lower complexity equation for estimating the noise variance is

still required for enhancing the BER performance. In section
V, we have proved that the PDF of Zmλ exhibits a normal
distribution, i.e., N (0, σz2 ), and the noise variance can be
evaluated using (27). The equation to produce the derived
noise variance uses a few arithmetic operators to evaluate FIGURE 2. The enhanced noise PDF of ZF detection.
the outcome. Therefore, we now use the derived σz2 from the
equation to produce the LLRs in (47) instead of the approxi-
mate noise variance from (49). Empirically, if Nr ≥ 100 and
Nr /Nt ≥ 10, the derived equation produces an accurate σz2 ,
compared to the exact noise variance. Therefore, the BEP
of the coded-OFDM-M-MIMO system, utilizing the derived
noise variance is significantly close to that of the general
detection for this case.

X. NUMERICAL RESULTS
This research work uses Monte-Carlo simulation to evalu-
ate the analytical results from the proposed equations. The FIGURE 3. Derived effective noise variance for the OFDM-M-MIMO
system.
derived effective noise PDFs, the PDF of output SNR, the
BEP, and the outage probabilities from the proposed equa-
tions are compared with those of the numerical simulations. its average value, and the outcome from the approximate
The OFDM-M-MIMO system and channel model utilized equation in (9) is inaccurate for this case.
are as described in section II. Gray-coded, QPSK and M - In addition, we employ the derived noise variance in (27) to
QAM are chosen as modulations schemes for the presented evaluate the effective noise variance of the OFDM-M-MIMO
numerical results. system in Fig. 3. The outcome of the approximation using
Fig. 2 compares the PDF of the in-phase component of NSE in (49) is also included in the figure. The results are
the effective noise ZmI from the simulation results with that also compared to that of the empirical. The system operated
of the accurate noise PDF in (21) and the Gaussian approxi- at Eb /N0 = −10 dB and the transmitter used the 16-QAM to
mation in (26). 16-QAM was selected for these results, with produce the transmit symbols. Nr = {200, 300, 400}, and Nt
Nt = 10 and Nr = {50, 100, 200}. A closer look at the was varied from 2 to 20. The σz2 for the OFDM-M-MIMO
graph reveals a good match between analytical and numer- system, where Nt > 20, was not included in the results
ical results. For the 10 × 200 system and ZmI = −2.086, since the outcome from (27) is inaccurate if the ratio of Nr
pz (ZmI ) from the accurate noise PDF and simulation result to Nt is less than 10. According to the results, the outcome
was 3.54 × 10−3 and 3.69 × 10−3 , respectively exhibiting from the derived equation was more accurate than that of
a difference of 1.51 × 10−4 . In addition, the accuracy of the NSE. The difference in σz2 between the simulation results
the Gaussian approximation PDF was slightly lower than the and the derived equation for the 20 × 200 OFDM-M-MIMO
accurate noise PDF, and the deviation between this equation system was 2.04×10−2 . Therefore, the σz2 from the proposed
and simulation results was 1.66 × 10−4 . We also used a 2- equation was only 1.47% different from the simulation result,
sample Kolmogorov–Smirnov test with significance level of and the number was 0.11 closer to the exact value than that of
5% to compare the analytical results with that of the empiri- the NSE. In addition to the results in Fig. 3, the distribution of
cal. The results confirmed that there is no difference between the derived noise variance dramatically increases according
the outcome from the derived PDFs and the simulation. How- to Nt . The σz2 of the 20 × 200 OFDM-M-MIMO system was
ever, there were small deviations between simulation and ana- 1.24 higher than that of the 2 × 200 system. Furthermore,
lytical results, especially for the system with Nr = 50. This is the system using a higher Nr obtained a lower σz2 , and the
due to the fact that we used 1 = 50 to approximate the value number for the 20 × 400 system was 0.71 lower than that of
of Dm,m in (7), although its true value was ranging between the 20 × 200 system. These results confirm that the OFDM-
41 and 59 for 79.93% of the time. Therefore, the maximal M-MIMO system using a higher Nr can be efficiently used to
deviation between the value of 1 and Dm,m was ±18% of reduce the distribution of the effective noise in the detection.

VOLUME 10, 2022 104393

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

FIGURE 4. BEP of OFDM-M-MIMO system utilizing QPSK modulation.

FIGURE 5. BEP of OFDM-M-MIMO system with Gray-coded, 16-QAM.

The BEP of the OFDM-M-MIMO system using

BPSK/QPSK modulation is illustrated in Fig. 4. The analyt-
ical results from (32) and those obtained using the approxi-
mate equation in (36a) are compared with simulation results.
The BEP analysis for QPSK modulation (Eq. (27) in [25])
and BPSK modulation (Eq. (39) in [17]) are also included in
the results. We considered the following M-MIMO Tx/Rx
antenna combinations, i.e., Nt = {3, 6, 9} and Nr = 100.
Focusing on the 5 × 100 system, the BEP at Eb /N0 = −7 dB
from the exact equation, the approximate equation and the
simulation result was 2.93×10−3 , 2.8×10−3 and 3.08×10−3 ,
respectively. The deviation between the simulation result
and the exact equation was only 1.53 × 10−4 . These results FIGURE 6. Deriving PDF of output SNR for OFDM-M-MIMO system.
confirm that the proposed equations produced an accurate
BEP for the OFDM-M-MIMO system. In addition, the BEP is reduced if Nr /Nt < 10. Focusing on a 10 × 50 system
analysis of [17], [25] at Eb /N0 = −7 dB was 3.09 × 10−3 operating at Eb /N0 = 6 dB, the difference in BEP between
and 3.36 × 10−3 , respectively. Therefore, the BEP analysis in the empirical results and that of (33) was 7.29 × 10−5 , while
(32) was significantly more accurate than the number in [25]. the value obtained in [25] was only 5.72 × 10−5 .
However, the deviation in BEP between the analytical result The PDF of the output SNR using the analytical results in
from the proposed equation and the exact BEP was still higher (39) and the approximate results in (40) is shown in Fig. 6.
than that of the BEP in [17] at 1.63 × 10−4 . 16-QAM was employed as a modulation scheme operating at
Fig. 5 compares the BEP analysis for the OFDM-M-MIMO an Eb /N0 = 0 dB, with Nt = 10 and Nr = {50, 100, 200}.
system using 16-QAM with Gray-encoding from (33) and According to the results, the PDFs from the derived equations
(36b) with that of simulation results and (29) in [25]. Nt was matched the simulation results. For a 10 × 200 system, the
10 and Nr = {50, 100, 200}. Since the equation in [17] was value of pγ (γs ) at γs = 16 dB from the simulation was 1.29×
only derived for BPSK modulation, the outcome from the 10−2 , and the difference from the analytical result using (39)
work was not included in the results. According to the results, was 2.98 × 10−5 . In addition, the PDF from the approximate
the proposed equations improved the BEP analysis, and the equation was, as expected, slightly less accurate than the
analytical results for the system with large Nr were closer to exact equation, i.e., the difference between the result from
the simulation results than that of [25]. In a 10 × 200 system this equation and the simulation results was 6.22 × 10−5 . The
operating at Eb /N0 = 0 dB, the BEP analysis from the exact PDFs from the derived equations were also compared using a
equation and the approximate equation were 3.83 × 10−5 2-sample Kolmogorov–Smirnov test with a significance level
and 3.42 × 10−5 , respectively, demonstrating a difference of of 5%, and the results confirm that there is no significant
2.04×10−6 and 6.12×10−6 from the simulation results. The different between simulation and analytical results. However,
number from (29) in [25] was 4.78 × 10−5 and the difference there was a small deviation between simulation and analytical
between the analytical and empirical results was 7.53×10−6 . results for the system with small Nr due to the underlying
Therefore, BEP from the proposed equations was more accu- assumptions made in the derivations.
rate than that of the equation derived in [25]. However, the The effects of Nr in the accuracy of outage probabilities
results also show that the proposed BEP analysis accuracy from the derived equations in (43) and (45) are evaluated in

104394 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

FIGURE 7. The effects of Nr in analytical results for outage probability.

FIGURE 8. Analytical results of outage probability for OFDM-M-MIMO
system.
Fig. 7. 16-QAM using Gray encoding operating at Eb /N0 =
−5 dB and γth = −5 dB. We considered different Tx/Rx 16-QAM was chosen by the system. We used an uplink 5G
antenna combinations, i.e., Nt = {6, 9, 12} and Nr varies NR LDPC with the base matrix 1 and the revision number
from 20 to 200 in steps of 20. According to the findings, 0 to encode the information bits. The channel decoder used
both equations produced accurate results. For instance, in a the Belief propagation to estimate the transmitted information
6 × 200 system, the Pout from the simulation result was from the LLR. The coding rate and the number of itera-
3.57 × 10−4 . The numerical values from the derived and tion were 1/2 and 25, respectively. Nt was 10 and Nr =
approximate equations were 3.37 × 10−4 and 3.1 × 10−4 , {50, 100, 200}. A pseudo-random interleaver was chosen for
respectively, matching very closely to the simulation results. the simulation for converting burst errors to random error
However, there was a small difference between the analytical events. Evidently, the difference in BER between the sys-
and simulation results, especially if Nr < 40, since the tem using the proposed detection and that of the classical
approximation of the value of Dm,m utilizing the constant 1 detection was small, especially for a system with a higher Nr .
was imprecise. The 10 × 50 system using the proposed detection required
In addition, Fig. 8 shows the outage probabilities of the Eb /N0 = −1.8 dB to maintain the BER of 10−3 , which was
system using 16-QAM with Gray encoding. Nr and γth were 0.47 dB higher than that of the classical detection. If Nr =
100 and 5 dB, respectively. Nt was 6, 9 and 12. The results 200, the difference in Eb /N0 between the system using the
confirm that the analytical results from the proposed equa- derived noise variance and the classical detection at BER
tions closely matched the simulation results. For the system of 10−3 was smaller at 1.35 × 10−2 dB. Furthermore, the
with Nt = 6 and operating at Eb /N0 = −1 dB, the value BER for the system using the derived noise variance was less
of Pout from the accurate equation was 8.26 × 10−5 , which than that of the approximate noise variance. Focusing on the
was only 9.12 × 10−6 away from the simulation result. The 10 × 50 system at Eb /N0 = −2 dB, the BER of the proposed
deviation between the analytical result from the approximate system was 3.15 × 10−3 , which was 2.79 × 10−3 lower than
equation and the simulation was 2.66 × 10−5 , which was that of the system with the approximate noise variance.
slightly higher than that of the accurate equation. However, In addition to the soft-output ZF detection, Table 1 demon-
there were small deviations between the analytical and the strates the number of real arithmetic operators, which are
simulation results, especially for the system with Nt = 12. required for producing the LLRs for all Nf sub-carriers
These differences are due to using the 2-nd order NSE to in (47). The proposed soft-output ZF detection utilized
approximate the inverse matrix operation and analyze the (27) to evaluate the noise variance for producing the LLR,
noise PDFs. Empirically, we found that the outcome from whereas the general detection employed the approximate
the approximate equation is inaccurate if Nr /Nt < 10. The noise variance in (49) for the operation. Total complex
deviation between the exact and the derived Pout is still high operations, which was required to determine the LLR from
since this condition is not satisfied. X̂mλ , were determined in terms of real arithmetic operations.
Fig. 9 shows the BER of coded-OFDM-M-MIMO system We assumed that the operation to evaluate the minimal value
where the low-complexity soft-output ZF detection, and the of X variables requires X −1 comparisons, and the summation
derived noise variance in (27) are utilized by the receiver of X variables uses X − 1 additions. Since the derived noise
for producing the LLR in (47). The detector employed the variance was a function of constants, the variable was evalu-
3-rd order NSE to simplify the operation in (3) for producing ated only one time for all sub-carriers. In addition, the number
X̂f from Yf . We compare the BER with that of the system of arithmetic operations, required for producing Df in (49),
using the approximate σz,m 2 in (49). In addition, the BER of was not included in the results since the diagonal components
the classical soft-output detector, using general inverse matrix of the Gram matrix were previously generated by (3) for
operation and the exact noise variance in (48), is also included producing X̂mλ . Although the number of comparisons in both
in the results. The block size utilization was 1024 bits and detections was equal, the proposed detection required less

VOLUME 10, 2022 104395

 D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

TABLE 1. Comparison of real-arithmetic operations in soft-output ZF detection.

soft-output ZF detection can be minimized by utilizing the

proposed detection.

XI. CONCLUSION
In this paper, an accurate PDF is derived using the effective
noise in a OFDM-M-MIMO system, where ZF detection is
used along with Gray-coded BPSK, QPSK, and M -QAM
modulation. We use joint probabilities of random variables
involved in the system model to derive their PDFs. Since
the computational complexity of the accurate PDFs derived
was too high, the Gaussian approximation of the effective
noise PDF was additionally derived. The obtained PDFs were
subsequently utilized to compute the BEP performance, the
PDF of the output SNR, and the outage probability of the
FIGURE 9. BER of coded-OFDM-M-MIMO system using low-complexity
proposed OFDM-M-MIMO system. The simulations con-
soft-output ZF detection. firmed that the outcomes from the derived equations were
closely matching the numerical results. Although the com-
putational complexity of the approximation is minimized,
the deviation between the analytical result from the approxi-
mate equations and the exact results was small, especially if
Nr ≥ 100. Therefore, they can be efficiently used to eval-
uate performance of OFDM-M-MIMO system. In addition,
we employed the derived noise variance to produce the LLRs
utilized in soft-output ZF detection. By utilizing the proposed
detection, the number of arithmetic operators, required for
producing the LLRs, can be significantly decreased. There
is a very small reduction in BER performance compared to
the classical detection, and the proposed detection requires
less Eb /N0 to maintain the BER than that of the soft-output
detection using NSE. Thus, the proposed soft-output detec-
tion can be efficiently utilized for reducing the computational
FIGURE 10. Computational complexity of soft-output ZF detection.
complexity in coded OFDM-M-MIMO systems.
Although the BEP from the proposed exact and the approx-
additions, subtractions, multiplications, and divisions than imate equations is more accurate than that of the BEP in [25],
that of the general detection. Fig. 10 shows the number of a number of approximations were chosen for this research
real arithmetic operators, required for the detections in the work. The diagonal components of the Gram matrix are
OFDM-M-MIMO system, using 16-QAM, Nt = 10 and assumed to be a constant and this approximation is pre-
Nf = 512 symbols. The size of Nr was varied from 20 to cise if Nr ≥ 100. This work presented closed-form BEP
200. Evidently, utilizing the derived noise variance provided for M -QAM modulations using 2-nd order NSE in (5) and
a massive reduction in real arithmetic operators to the ZF extending the BPSK work in [17] to general M -QAM formats
detector. If Nr = 200, the number of additions and subtrac- for Nr > 10Nt . To overcome the latter limitation, future
tions for the classical and the proposed detection was 2.1 × research work will consider a higher order NSE, the presence
106 and 6.14 × 104 , respectively. Therefore, 97.08% of the of diagonal components in the Gram matrix as well as other
arithmetic operations can be reduced by utilizing the derived types of inverse matrix operations. Furthermore, more accu-
noise variance. In addition, the number of multiplications rate channel model for the system using OFDM waveform
significantly reduced from 2.1 × 106 to 4.61 × 104 . The should be included in the work for improving the performance
results demonstrated that 97.81% of multiplications in the analysis.

104396 VOLUME 10, 2022

D. Chumchewkul, C. C. Tsimenidis: Closed-Form BEP of ZF Detection for OFDM-M-MIMO Systems Using Effective Noise PDF

REFERENCES [23] J. Lu, K. B. Letaief, J. C.-I. Chuang, and M. L. Liou, ‘‘M-PSK and
[1] E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, ‘‘Massive M-QAM BER computation using signal-space concepts,’’ IEEE Trans.
MIMO for next generation wireless systems,’’ IEEE Commun. Mag., Commun., vol. 47, no. 2, pp. 181–184, Feb. 1999.
vol. 52, no. 2, pp. 186–195, Feb. 2014. [24] J. G. Proakis, Digital Communications, 5th ed. New York, NY, USA:
[2] S. Yang and L. Hanzo, ‘‘Fifty years of MIMO detection: The road McGraw-Hill, 2008.
to large-scale MIMOs,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 4, [25] A. J. Al-Askery, C. C. Tsimenidis, S. Boussakta, and J. A. Chambers,
pp. 1941–1988, 4th Quart., 2015. ‘‘Performance analysis of coded massive MIMO-OFDM systems using
[3] M. Wu, B. Yin, G. Wang, C. Dick, J. R. Cavallaro, and C. Studer, ‘‘Large- effective matrix inversion,’’ IEEE Trans. Commun., vol. 65, no. 12,
scale MIMO detection for 3GPP LTE: Algorithms and FPGA implemen- pp. 5244–5256, Dec. 2017.
tations,’’ IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 916–929, [26] G. L. Stüber, Principles of Mobile Communication, 4th ed. Cham,
Oct. 2014. Switzerland: Springer, 2017.
[4] C. Xiao, X. Su, J. Zeng, L. Rong, X. Xu, and J. Wang, ‘‘Low-complexity [27] P. Dharmawansa, N. Rajatheva, and H. Minn, ‘‘An exact error probability
soft-output detection for massive MIMO using SCBiCG and Lanczos analysis of OFDM systems with frequency offset,’’ IEEE Trans. Commun.,
methods,’’ China Commun., vol. 12, pp. 9–17, Dec. 2015. vol. 57, no. 1, pp. 26–31, Jan. 2009.
[5] L. Dai, X. Gao, X. Su, S. Han, I. Chih-Lin, and Z. Wang, ‘‘Low-complexity [28] D. Singh, A. Kumar, H. D. Joshi, M. Magarini, and R. Saxena, ‘‘Symbol
soft-output signal detection based on Gauss–Seidel method for uplink error rate analysis of OFDM system with CFO over TWDP fading chan-
multiuser large-scale MIMO systems,’’ IEEE Trans. Veh. Technol., vol. 64, nel,’’ Wireless Pers. Commun., vol. 109, no. 4, pp. 2187–2198, Dec. 2019.
no. 10, pp. 4839–4845, Oct. 2015. [29] D. Singh and H. D. Joshi, ‘‘Performance analysis of SFBC-OFDM system
[6] X. Qin, Z. Yan, and G. He, ‘‘A near-optimal detection scheme based with channel estimation error over generalized fading channels,’’ Trans.
on joint steepest descent and Jacobi method for uplink massive MIMO Emerg. Telecommun. Technol., vol. 29, no. 3, p. e3293, Mar. 2018.
systems,’’ IEEE Commun. Lett., vol. 20, no. 2, pp. 276–279, Feb. 2016. [30] D. Singh and H. D. Joshi, ‘‘Error probability analysis of STBC-OFDM
[7] M. Zhang and S. Kim, ‘‘Evaluation of MMSE-based iterative soft detec- systems with CFO and imperfect CSI over generalized fading channels,’’
tion schemes for coded massive MIMO system,’’ IEEE Access, vol. 7, AEU, Int. J. Electron. Commun., vol. 98, pp. 156–163, Jan. 2018.
pp. 10166–10175, 2019. [31] A. Grami, Probability, Random Variables, Statistics, and Random Pro-
[8] S. Berra, M. A. M. Albreem, M. Malek, R. Dinis, X. Li, and K. M. Rabie, cesses: Fundamentals and Applications. Hoboken, NJ, USA: Wiley, 2020.
‘‘A low-complexity soft-output signal data detection algorithm for UL [32] M. K. Simon, Probability Distributions Involving Gaussian Random
massive MIMO systems,’’ in Proc. Int. Conf. Comput., Inf. Telecommun. Variables: A Handbook for Engineers and Scientists. New York, NY,
Syst. (CITS), Nov. 2021, pp. 1–6. USA: Springer, 2006.
[9] C. Zhang, Z. Wu, C. Studer, Z. Zhang, and X. You, ‘‘Efficient soft-output [33] A. Jeffrey and D. Zwillinger, Table of Integrals, Series, and Products,
Gauss–Seidel data detector for massive MIMO systems,’’ IEEE Trans. 6th ed. New York, NY, USA: Academic, 2000.
Circuits Syst. I, Reg. Papers, vol. 68, no. 12, pp. 5049–5060, Dec. 2021. [34] G. R. Iversen, Calculus. Newbury Park, CA, USA: Sage, 1996.
[10] M. A. Albreem, M. Juntti, and S. Shahabuddin, ‘‘Massive MIMO detection
techniques: A survey,’’ IEEE Commun. Surveys Tuts., vol. 21, no. 4,
pp. 3109–3132, 4th Quart., 2019.
DITSAPON CHUMCHEWKUL received the
[11] J. H. Winters, J. Salz, and R. D. Gitlin, ‘‘The capacity of wireless B.Eng. degree in electrical and electronics engi-
communication systems can be substantially increased by the use of neering from the King Mongkut’s Institute of
antenna diversity,’’ in Proc. 1st Int. Conf. Universal Pers. Commun., 1992, Technology North Bangkok, in 2000, and the
pp. 02.01/1–02.01/5. M.Eng. degree in information engineering from
[12] J. H. Winters, J. Salz, and R. D. Gitlin, ‘‘The impact of antenna diversity on the King Mongkut’s Institute of Technology
the capacity of wireless communication systems,’’ IEEE Trans. Commun., Ladkrabang, in 2003. He is currently pursuing the
vol. 42, no. 234, pp. 1740–1751, Feb./Mar./Apr. 1994. Ph.D. degree in electrical and electronics engineer-
[13] R. Xu and F. C. M. Lau, ‘‘Performance analysis for MIMO systems using ing with the School Engineering, Newcastle Uni-
zero forcing detector over Rice fading channel,’’ in Proc. IEEE Int. Symp. versity. He is also a Lecturer with the Department
Circuits Syst., vol. 5, May 2005, pp. 4955–4958. of Telecommunication Engineering, Faculty of Engineering, Rajamangala
[14] H. A. Saleh and W. Hamouda, ‘‘Performance of zero-forcing detectors University of Technology Rattanakosin. His education in M.Eng. and Ph.D.
over MIMO flat-correlated Ricean fading channels,’’ IET Commun., vol. 3, degrees was supported by the scholarships from Thailand’s National Science
no. 1, pp. 10–16, Jan. 2009.
and Technology Development Agency, the Thailand Graduate Institute
[15] D. Gore, R. W. Heath, Jr., and A. Paulraj, ‘‘On performance of the zero
of Science and Technology, the Office of the Civil Service Commission,
forcing receiver in presence of transmit correlation,’’ in Proc. IEEE Int.
Symp. Inf. Theory, Jun./Jul. 2002, p. 159.
the Royal Thai Embassy, The Office of Educational Affairs, and the
[16] H.-Y. Fan, ‘‘MIMO detection schemes for wireless communications,’’ Rajamangala University of Technology Rattanakosin. He has published more
MPhil. thesis, Dept. Elect. Electron. Eng., Hong Kong Univ. Sci. Technol., than 14 conference papers, a letter, and a journal articles. His research inter-
Hong Kong, 2002. ests include wireless communications, optical wireless communications, and
[17] Y. Jiang, M. K. Varanasi, and J. Li, ‘‘Performance analysis of ZF and channel coding techniques.
MMSE equalizers for MIMO systems: An in-depth study of the high SNR
regime,’’ IEEE Trans. Inf. Theory, vol. 57, no. 4, pp. 2008–2026, Apr. 2011.
[18] G. Alfano, C.-F. Chiasserini, and A. Nordio, ‘‘SINR and multiuser effi-
ciency gap between MIMO linear receivers,’’ IEEE Trans. Wireless Com- CHARALAMPOS C. TSIMENIDIS (Senior Mem-
mun., vol. 19, no. 1, pp. 106–119, Jan. 2020. ber, IEEE) received the Ph.D. degree in commu-
[19] C. Siriteanu, Y. Miyanaga, S. D. Blostein, S. Kuriki, and X. Shi, ‘‘MIMO nications and signal processing from Newcastle
zero-forcing detection analysis for correlated and estimated Rician fad- University, in 2002. He is currently a Professor of
ing,’’ IEEE Trans. Veh. Technol., vol. 61, no. 7, pp. 3087–3099, Sep. 2012. digital innovation at Nottingham Trent University.
[20] C. Wang, E. K. S. Au, R. D. Murch, W. H. Mow, R. S. Cheng, and V. Lau, He has published over 230 conference and jour-
‘‘On the performance of the MIMO zero-forcing receiver in the presence nal papers, supervised successfully three M.Phil.
of channel estimation error,’’ IEEE Trans. Wireless Commun., vol. 6, no. 3, and 50 Ph.D. students and made contributions in
pp. 805–810, Mar. 2007. the area of digital communications to several U.K.
[21] F. Jiang, C. Li, and Z. Gong, ‘‘Accurate analytical BER performance for ZF
and European funded research projects. His main
receivers under imperfect channel in low-SNR region for large receiving
research interests include digital communications and signal processing with
antennas,’’ IEEE Signal Process. Lett., vol. 25, no. 8, pp. 1246–1250,
Aug. 2018.
specialization in massive multiple-input multiple-output systems, adaptive
[22] C. D. Altamirano, J. Minango, H. C. Mora, and C. De Almeida, ‘‘BER filter, and demodulation algorithms, error control and network coding for
evaluation of linear detectors in massive MIMO systems under imperfect radio frequency, and underwater acoustic channels. He is a member of
channel estimation effects,’’ IEEE Access, vol. 7, pp. 174482–174494, the IET.
2019.

VOLUME 10, 2022 104397