The Journal of China

Universities of Posts and

Telecommunications
February 2010, 17(1): 51–55
www.sciencedirect.com/science/journal/10058885 www.buptjournal.cn/xben

SER bound for ordered ZF-SIC receiver in M-QAM MIMO system

HAN Juan1,2 ( ), CUI Qi-mei1,2, TAO Xiao-feng1,2, ZHANG Ping1,2

1. Wireless Technology Innovation Institute, Beijing University of Posts and Telecommunications, Beijing 100876, China
2. Key Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunications, Beijing 100876, China

Abstract

With the advances of the 4th generation (4G) mobile communications targeting multi-user QoS-guaranteed multimedia services,
multiple input multiple output (MIMO) combined with multi-level quadrature amplitude modulation (M-QAM) technique gains
great popularity for its capability of supporting high data rate. In this article, the statistical distribution of signal to noise plus
interference ratio (SINR) in each recursive step of optimal ordered zero-forcing perfect successive interference cancellation
(ZF-OOPSIC) detection is deduced based on order statistics theory. The symbol error rate (SER) of M-QAM MIMO system with
ZF-OOPSIC detector is analyzed under flat Rayleigh fading channel. This precise analytical SER result could be taken as the
lower bound of various zero-forcing ordered successive interference cancellation (ZF-OSIC) receivers. Combining the analytical
SER expression for linear zero-forcing (ZF) receiver as upper bound, the rough range of SER for various ZF-OSIC detectors
could be obtained. Monte Carlo (MC) simulations validate the analytical results and prove the conclusions.

Keywords SER, ZF, ordered successive interference cancel (SIC), M-QAM, MIMO

1 Introduction Comprehensive evaluation of system performance is

required because wireless propagation channel may severely
In recent years, as 4G wireless systems are designed to degrade the performance of M-QAM MIMO [3–5]. In Ref. [6],
offer high-quality multimedia services such as voice, data and the closed-form analytical expression of averaged bit error
video, the required data rate is increasing compared with rate (BER) is given for MIMO systems with two transmitting
other existing services. Thus the conflict between high data antennas and multiple receiving antennas. For the case of
rate transmission and limited spectrum resource is becoming arbitrary number of transmitting antennas, the results are not
more serious. To solve this problem, research attention has directly available. In Ref. [7], for ZF successive interference
been paid to improving the spectral efficiency in wireless cancellation (ZF-SIC) receiver, the analytical BER performance
environment. MIMO techniques have drawn much attention for binary phase shift keying (BPSK) mode is deduced for the
due to its superiority in exploring spatial resource, enhancing systems with arbitrary number of transmitting and receiving
system capacity and increasing the spectral efficiency [1]. antennas, but the high order M-QAM mode is not discussed.
Theoretical analysis has proved that the channel capacity of Among the most recent research on this topic [8], the SER of
MIMO system is directly proportional to the minor number of vertical-bell laboratory layered space-time (V-BLAST) with
transmitting antennas and receiving antennas in rich scattering M-QAM is analyzed under rich scattering Rayleigh fading
propagation environment. However, in practice, due to the channel for ZF receiver and ZF perfect successive interference
limited size and cost of equipment, it is difficult to improve cancellation (ZF-PSIC) receiver, which are used respectively
data transmission rate only through increasing the number of as the upper and lower bound of ZF-SIC receiver without
antennas. Therefore, there has been a trend toward the use of ordering. Since ordering is pervasively used in ZF-SIC receiver,
M-QAM MIMO to increase transmission rate [2]. and can dramatically enhance the system performance [9–11],
analytical results for SER of M-QAM MIMO with ZF-OSIC
Received date: 19-11-2008 receiver will be discussed in this article.
Corresponding author: HAN Juan, E-mail: [email protected]
DOI: 10.1016/S1005-8885(09)60423-4 Among all the ordering criterions, SINR based ordering
52 The Journal of China Universities of Posts and Telecommunications 2010

method applied in Ref. [9] is optimal. Hence, with perfect antenna k. Then the system model could be expressed as:
Nt
interference cancellation, that is, without error propagation,
r = Hs + n = ∑ hk sk + n (1)
the ZF-OOPSIC receiver obviously achieves the lower bound k =1

of various ZF-OSIC algorithms. In this article, by deducing

the probability density function (PDF) of SINR in each 3 Analytical SER bound of M-QAM MIMO system
recursive step of ZF-OOPSIC detection, analytical SER
performance expression for ZF-OOPSIC receiver in M-QAM For M-QAM, it is not convenient to obtain closed-form
MIMO system is derived. Taking this result as lower bound BER result because of different amplitudes of symbols in
and the analytical SER performance for ZF receiver obtained constellation. Thereby, SER is utilized for system performance
in Ref. [8] as the upper bound, the above analysis makes it evaluation here.
possible to efficiently estimate the SER range for a ZF-OSIC
receiver. Hence, time-consuming Monte Carlo simulation is 3.1 SER lower bound for ZF-OSIC algorithm in M-QAM
MIMO
not required, which provides an efficient method for radio
link optimization and system resource configuration.
ZF-OSIC algorithm is commonly used for detection of
The article is organized as follows. In Sect. 2, the basic
spatial multiplexing based MIMO, which enhances the system
system model is introduced. Sect. 3 presents the SER
performance dramatically. Among all ordering criterions,
performance bound of M-QAM MIMO system with ZF-OSIC
SINR based ordering method mentioned in Ref. [9] is optimal.
receiver. MC simulation results and performance comparisons
Only the layer with the highest SINR is effective in each
are presented in Sect. 4. Finally, conclusions are drawn in
recursive step, which globally guarantees the lowest SER of
Sect. 5.
each layer. Furthermore, ZF-OSIC suffers performance
2 System model degradation when the estimated symbols are not precise.
Besides, the imperfect interference cancellation causes error
In this article, a single-user, point to point MIMO system propagation and results in certain variable distributions
with N t transmitting and N r receiving antennas ( N t ≤N r ) changing, which makes the analytical performance highly
is considered, as shown in Fig. 1. complicated to deduce. To solve these problems and facilitate
analysis process, perfect SIC (PSIC) is constructed, which
assumes that the cancelled interference is accurate. Thus, in
the following part, the SER performance of ordered ZF-PSIC
with SINR based ordering (i.e. ZF-OOPSIC detector) is
analyzed and taken as the lower bound of various ZF-OSIC
algorithms.
Fig. 1 Model of MIMO systems
For ZF-OOPSIC, all the following discussion is developed
Here, V-BLAST scheme [12] is taken as an example to in the jth recursive step. Assume that in the jth recursive step,
carry out the performance analysis. The transmitted signal is the transmitted symbol from the kjth antenna holds the highest
denoted as s = [ s1 , s2 ,..., sN t ]T and the received signal is SINR among all the Nt-j+1 undetected symbols. Furthermore,
r = [r1 , r2 ,..., rN r ]T . Channel matrix H = ( hik ) N × N follows r ( j ) denotes the interference-canceled receive signal obtained
r t
in the jth recursive step. Considering the perfect interference
independent identically distributed (i.i.d.) complex Gaussian
cancellation assumption, r ( j ) can be written as:
distribution with zero mean and unit variance, in which fading Nt j −1 Nt

coefficient hik (i = 1,2,..., N r ; k = 1,2,..., N t ) indicates the

a= j a =1
( )
r ( j ) = ∑ hka ska + ∑ hka ska − Qy ( sˆka ) + n = ∑ hka ska + n
a= j
(2)
complex channel gain from transmit antenna k to receive
where Qy ( sˆka ) is the hard decision of estimated value of ska .
antenna i. n = [n1 , n2 ,..., nNr ]T represents the zero-mean
If gl( j ) is defined as the mulling vector for the lth undetected
additive white Gaussian noise (AWGN) with variance σ 2 .
symbol in the jth recursive step where l = 1,2,..., N t − j + 1 ,
Further denote H as H = [h1 , h2 ,..., hN t ], in which
gl( j ) is the lth row vector of Moore-Penrose pseudo-inverse
hk = [h1k , h2 k ,..., hNr k ]T (k = 1,2,..., N t ) is the column vector
of H ( j ) = [h1( j ) , h2( j ) ,..., hN( j−) j+1 ] , which is obtained by extracting
consisting of all the fading coefficients from the transmitting t
Issue 1 HAN Juan, et al. / SER bound for ordered ZF-SIC receiver in M-QAM MIMO system 53

the columns indexed by k1 , k2 ,..., k j −1 from channel matrix x((lj)) is:

⎧ 1 (l = p ) n!
H. Having gl( j ) hp( j ) = ⎨ , the N t − j + 1 estimated pl ( x ( j ) ) = F ( x ( j ) )l −1[1 − F ( x ( j ) )]n −l p ( x ( j ) ) (7)
⎩0 (l ≠ p ) (l − 1)! ( n − l )!
symbols derived in the jth recursive step could be expressed where n = N t − j + 1 . According to Eqs. (5)–(7), in the jth
( j)
by s as follows: recursive step, the PDF for the maximum x ( j ) (i.e. x((Nj )t − j +1) )
⎛ Nt ⎞ is derived as:
sˆl( j ) = gl( j ) r ( j ) = gl( j ) ⎜ ∑ hka ska + n ⎟ = gl( j ) hl( j ) sl + Nt − j
⎝ a= j ⎠ 2( N t − j + 1) ⎡ x ⎤
N r − N t + j −1 2 k

Nt
p((Nj )t − j +1) ( x)= ⎢1 − e
( N r − N t + j − 1)! ⎣
− x2
∑ k! ⎦
⎥ ⋅
∑
k =0
g h s + g n = sl + g n
( j)
l ka ka
( j)
l
( j)
l
(3)
x 2( Nr − N t + j ) −1e − x
2
a = j , ka ≠ l (8)
Particularly, among all the N t − j + 1 estimated symbols, Under AWGN channel, the SER of optimized receiver
the one related to the kjth antenna, which obtains the highest utilizing M-QAM with rectangular signal constellations could
SINR in the current recursive step, is effective. Based on be expressed as [15]:
2
Eq. (3), if total power P is divided equally among all the ⎛ 1 ⎞ ⎛ 3RSINR ⎞ ⎡⎛ 1 ⎞ ⎛ 3RSINR ⎞ ⎤
PM = 4 ⎜ 1 − ⎟ Q ⎜⎜ M − 1 ⎟⎟ − 4 ⎢⎜1 − ⎟Q⎜ ⎟⎥ =
M ⎠ ⎜⎝ M − 1 ⎟⎠ ⎦⎥
transmitting antennas, the SINR of the lth undetected layer in
⎝ M ⎠ ⎝ ⎠ ⎣⎢⎝
the jth recursive step of ZF-OOPSIC algorithm is
( ) ⎛ 3RSINR ⎞ ⎛ 2
2 ⎛ 1 ⎞ 1 ⎞
E sl λ0 2 ⎜1 − ⎟ erfc ⎜⎜ 2( M − 1) ⎟⎟ − ⎜ 1 − ⎟ ⋅
ZF-OOPSIC( j )
= = ⎝ M ⎠ ⎠ ⎝ M ⎠
( )
RSINR ,l 2 2
(4) ⎝
E gl( j ) n gl( j ) 2
⎛ 3RSINR ⎞
where E(⋅) denotes the mathematical expectation, ⋅ is the erfc ⎜⎜ ⎟⎟ (9)
⎝ 2( M − 1) ⎠
2-norm of vector, and λ0 = P σ 2 represents the averaged where Q (i ) is the Q function. By defining
transmitted signal to noise ratio (SNR). In Eq. (4), 1 g ( j) 2
l A = ⎡1 − 1
⎣ ( )
M ⎤, ξ =
⎦ [3λ0 (2( M − 1)) ] , and using Eqs. (8)
follows independent χ 2 distribution with degree of freedom and (9), the averaged SER performance of ZF-OOPSIC
2 ( N r − N t + j ) and variance 1/2 [13–14]. If one defines a algorithm under flat Rayleigh fading channel can be derived
2 as:
variable xl( j ) = 1 gl( j ) , (l = 1,2,..., N t − j + 1) , for any l,

xl( j ) is amenable to generalized Rayleigh distribution with

PMZF-OOPSIC =
1 N t ZF-OOPSIC( j ) 1 N t ∞
∑ PM
N t j =1
= ∑ PM ( x) ⋅
N t j =1 ∫ 0 (
degree of freedom 2 ( N r − N t + j ) and variance 1/2 [15],
and its PDF meets:
p((Nj )t − j +1) ( x)dx = ) 1 N t ⎧⎪ 4 A( N t − j + 1) ∞
∑⎨
N t j =1 ⎪⎩ ( N r − N t + j − 1)! ∫ 0
erfc (ξ x ) ⋅
2
p( x( j ) ) = ( x ( j ) ) 2( N r − N t + j ) −1 e − ( x ) Nt − j
( j) 2
(5) ⎡ x ⎤
N r − N t + j −1 2 k
( N r − N t + j − 1)! ⎢1 − e
− x2
∑ ⎥ x 2( Nr − Nt + j ) −1e− x dx −
2

Since the degree of freedom 2 ( N r − N t + j ) is even, the ⎣ k =0 k! ⎦

Nt − j
2 A2 ( N t − j + 1) ∞ ⎡ x ⎤
N r − N t + j −1 2 k
cumulative density function (CDF) of xl( j ) has the closed-
∫ erfc 2 (ξ x ) ⎢1 − e − x ∑ ⋅
2

⎥
form expression as [15]:
( N r − N t + j − 1)! 0 ⎣ k =0 k! ⎦
N r − N t + j −1
( x( j ) )2 k ⎫⎪
∑ x 2( Nr − N t + j ) −1e − x dx ⎬
2
− ( x( j ) )2
F ( x( j ) ) = 1 − e (6) (10)
k =0 k! ⎪⎭
For optimal ordering operation, the maximum SINR in
each recursive step should be selected, thus the largest one is 3.2 SER upper bound for ZF-OSIC algorithms in M-QAM
identified among the N t − j + 1 i.i.d. random variables MIMO
xl( j ) (l = 1,2,..., N t − j + 1) . This issue could be modelled as an
( j) ( j)
The linear ZF detector only contains the operation of
order statistics problem [16]. Let x(1) , x(2) ,..., x((Nj )t − j +1) denote
zero-forcing without interference cancellation, which makes
the order statistics of the N t − j + 1 random samples xl( j ) its performance inferior to that of ZF-OSIC. Therefore, the
from the continuous population with CDF F ( x ( j ) ) and PDF analytical SER results of ZF detector in Eq. (11) obtained in
p( x ( j ) ) , where x(1)
( j)
< x(2)
( j)
< ... < x((Nj )t − j +1) . Then, the PDF of Ref. [8] is introduced here as the upper bound of ZF-OSIC
54 The Journal of China Universities of Posts and Telecommunications 2010

algorithms. propagation. Meanwhile, SER curves of ZF-OOSIC detector

1 N t ZF
( ) = (P ) 4A ∞
∑ PM
have a relatively significant gap with that of ZF receiver,
erfc (ξ x ) ⋅
( N r − N t )! ∫ 0
PMZF = ZF
=
N t j =1 j M j especially when modulation order is low or N r − N t is large,
2 A2 ∞ which exactly indicates that the proper ordering and SIC
x 2( N r − N t ) +1e− x dx − ∫ erfc2 (ξ x ) ⋅
2

( N r − N t )! 0 elevate system performance dramatically.

It is also validated in Fig. 2 that the analytical results of ZF
x 2( N r − N t ) +1e− x dx
2

(11) detector in Eq. (11) match well with the MC simulations,

Based on the above analysis, although it is difficult to which has been depicted in Ref. [8]. Since curves for
obtain an accurate expression of M-QAM MIMO system SER ZF-OOSIC detector is above the analytical results of Eq. (10),
performance with ZF-OSIC algorithm, it can be considered in it can be concluded that Eq. (10) could be used as the lower
the range of ZF algorithm and ZF-OOPSIC algorithm as bound for all kinds of ZF-OSIC detectors. Therefore, it is
expressed in Eq. (12). reasonable and convenient to take the two analytical results in
PMZF-OOPSIC≤PMZF-OSIC≤PMZF (12) Eqs. (10) and (11) as the lower and upper bound of various
ZF-OSIC receivers, respectively.
4 Simulation results and analysis Figs. 3 and 4 illustrate the performance of MIMO receiver
utilizing the deduced analytical SER bounds herein and in
In this section, extensive simulations are carried out to
Ref. [8]. In Fig. 3, results of numerical simulations for 2T4R
support the derived analytical results. SER performance of
and 4T6R MIMO systems with 16QAM are demonstrated.
M-QAM MIMO systems with different antenna configurations
For each system configuration, three curves of analytical SER
and modulation modes are investigated.
for ZF, ZF-PSIC and ZF-OOPSIC detectors are plotted for
In Fig. 2, the results of analytical and MC simulations for
comparison, respectively. As the SER performance of ZF
2T4R (2 transmitting antennas, 4 receiving antennas) with
receiver is only related to the difference between the
16QAM, 2T4R with 64QAM and 4T8R with 16QAM MIMO
transmitting and the receiving antenna numbers, the ZF
systems are demonstrated. As shown in Fig. 2, the analytical
curves of 2T4R and 4T6R MIMO systems are identical.
results achieved by Eq. (10) have no significant difference
However, compared with the results in the 2T4R MIMO
compared with the MC simulation results of corresponding
system, the gap between ZF-OOPSIC and ZF-PSIC detectors
ZF-OOPSIC receivers. Thus conclusions can be drawn that
is even larger, especially in high-SNR region when antenna
the analytical SER for ZF-OOPSIC receiver in Eq. (10) is configuration is 4T6R. Since N r − N t are the same for the
reasonable.
two groups of curves, ordering operation benefits from a
bigger Nt. The main reason is that recursion time is decided by
Nt. More recursion processes hold more potential in elevating
detection performance, which could be realized by proper
ordering operations.

Fig. 2 SER performance of M-QAM MIMO system with

ZF-OSIC detector by analytical and MC simulations

Additionally, all the MC simulation results for optimal

ordered ZF-SIC (ZF-OOSIC) detector are tightly approaching
the deduced lower bound in Eq. (10), which illustrates the
Fig. 3 Analytical SER performance of 16QAM MIMO
slight but confirmed performance degradation caused by error systems with different receivers, same Nr − Nt
Issue 1 HAN Juan, et al. / SER bound for ordered ZF-SIC receiver in M-QAM MIMO system 55

Fig. 4 shows the case of 4T6R and 4T8R MIMO with Technology Project of China (2008ZX03003－004, 2009ZX03003－
16QAM. Obviously, on the premise of identical number of 009), the National Natural Science Foundation of China (60872048),
transmitting antenna, larger number of receiving antenna Sino-Sweden Cooperative Project (2008DFA12110) and
induces performance elevation for all kinds of detectors. Sino-Australia Collaborations Project.
Especially, with the same SNR, 4T8R MIMO sees more
performance promotion caused by ordering operation than References
that of 4T6R MIMO.
1. Foschini G J, Gans M J. On limits of wireless communications in a fading
environment when using multiple antennas. Wireless Personal Com-
munications, 1998, 6(3): 311−335
2. Zhang P, Tao X F, Zhang J H, et al. A visions from the future: beyond 3G
TDD. IEEE Communications Magazine, 2005, 43(1): 38−44
3. Loyka S, Mosig J. Channel capacity of n-antenna BLAST architecture.
Electronics Letters, 2000, 36(7): 660−661
4. Loyka S L. Channel capacity of MIMO architecture using the exponential
correlation matrix. IEEE Communications Letters, 2001, 5(9): 369−371
5. Loyka S, Tsoulos G. Estimating MIMO system performance using the
correlation matrix approach. IEEE Communications Letters, 2002, 6(1):
19−21
6. Loyka S, Gagnon F. Performance analysis of the V-BLAST algorithm: an
analytical approach. IEEE Transactions on Wireless Communications,
2004, 3(4): 1326−1337
7. Shen C, Zhu Y, Zhou S D, et al. On the performance of V-BLAST with
zero-forcing successive interference cancellation receiver. Proceedings of
Fig. 4 Analytical SER performance of 16QAM MIMO IEEE 2004 Global Communications Conference (GLOBECOM’04): Vol
systems with different receivers, same Nt
5, Nov 29−Dec 3, 2004, Dallas, TX, USA. New York, NY, USA: IEEE,
2004: 2818−2822
5 Conclusions 8. Xu J, Tao X F, Zhang P. Analytical SER performance bound of M-QAM
MIMO system with ZF-SIC receiver. Proceedings of IEEE 2008
International Conference on Communications (ICC’08), May 19−23,
MIMO systems draw much attention due to its high
2008, Beijing, China. Piscataway, NJ, USA: IEEE, 2008: 5103−5107
spectrum efficiency and capability of exploring spatial 9. Wolniansky P W, Foschini G J, Golden G D, et al. V-BLAST: an
resource. With high order M-QAM, MIMO system owns even architecture for realizing very high data rates over the rich-scattering
wireless channel. Proceedings of 1998 URSI International Symposium
more superiority in supporting high data rate transmission.
on Signals, Systems, and Electronics (ISSSE’98), Sep 29−Oct 2, 1998,
Therefore, solid contribution towards theoretical analysis of Pisa, Italy. Piscataway, NJ, USA: IEEE, 1998: 295−300
M-QAM MIMO system is needed. In this article, SER 10. Tao X F, Yu Z Z, Qin H Y, et al. New sub-optimal detection algorithm of
layered space-time code. Proceedings of the 55th Vehicular Technology
performance of M-QAM MIMO systems with ZF-OSIC
Conference (VTC-Spring’02): Vol 4, May 6−9, 2002, Birmingham, UK.
detector is studied under Rayleigh fading channel. The Piscataway, NJ, USA: IEEE, 2002: 1791−1794
statistical distribution of SINR in each recursive step of 11. Guo Z H, Li L H, Tao X F. Fast ordering detection algorithm for
V-BLAST system. Journal of Beijing University of Posts and
ZF-OOPSIC detection is derived here, which is afterward
Telecommunications, 2007, 30(4): 83−87 (in Chinese)
employed in deducing the analytical SER for ZF-OOPSIC 12. Foschini G J. Layered space-time architecture for wireless com-
receiver. Taking the analytical SER of ZF-OOPSIC detector munication in a fading environment when using multi-element antennas.
as the lower bound and the analytical SER of ZF detector in Bell Labs Technical Journal, 1996, 1(2): 41−59
13. Liu J S. The Introduction of Wishart Distribution. Beijing: Science Press,
Ref. [8] as the upper bound, it can roughly predicate the 2005: 70−97 (in Chinese)
performance of various ZF-OSIC receivers for link optimization 14. Li X, Nie Z P. Performance losses in V-BLAST due to correlation. IEEE
and system configuration. Extensive MC simulations are Antennas and Wireless Propagation Letters, 2004, 3(1): 29−294
15. Proakis J G. Digital Communications. 4th ed. New York, NY, USA:
carried out to verify the validity of the analytical results. McGraw-Hill, 2001: 45−46, 278
16. Casella G, Berger R L. Statistical inference. 2nd ed. Pacific Grove, CA,
Acknowledgements USA : Thomson Learning , 2002

This work was supported by the Specialized Major Science and

(Editor: WANG Xu-ying)