Вы находитесь на странице: 1из 4

SURVEY OF LINEAR MIMO DECODERS AND THEIR IMPACT ON SPATIAL MULTIPLEXING PERFORMANCE IN NON-LINEAR LAYERED RECEIVERS, APRIL 20091

Survey of Linear MIMO Decoders and Their


Impact on Spatial Multiplexing Performance in
Non-Linear Layered Receivers
John W. Thomas, Student, University of Texas at Dallas

Abstract—This paper is a short survey of a few available linear Where fX,Y is the mutual probability distribution function
decoders and their impact on a Spatial Multiplexing (SM) scheme (pdf) of the input and output of the system and fX and fY
in a multiple-input, multiple-output (MIMO) environment using are the pdf of the input and output respectively. B is equal to
layered non-linear receivers. The ultimate advantage of SM is
to increase capacity (b/s/Hz), but another aspect of performance 2 when speaking about bits.
is the rate of errors over a channel. Each used receiver has a We arrive at the below equation for mutual information after
different impact on the overall performance. This paper attempts assuming zero-mean circular symmetric complex Gaussian
to summarize some of these simulated observations. (ZMCSCG) input and output and a power restraint at the
transmitter, meaning the trace of the covariance of the input is
I. I NTRODUCTION equal to ρ. I represents the identity matrix, H represents the
MIMO channel, and Rx represents the autocorrelation of the

T HE choice of decoders at the receiver have an immense


impact on performance. An even greater impact is felt
by a spatial multiplexing MIMO system due to the multitude
input. .
I(X, Y ) = log2 det[I + HRx HH ] (2)
of receivers and independent demultiplexing of bits/symbols After maximizing the mutual information with Rx , the
over multiple antennas at the transmitter. This creates a phe- generalized Shannon Capacity describes the upper bound of
nomenon where the different transmitted streams from each error-free capacity in additive white Gaussian noise is given
antenna interfere at the receiver. This is known as multi-stream below.
interference (MSI). This paper attempts to summarize how
a selective choice of a linear decoders at a receiver impacts Capacity = max log2 det[I + HRx HH ] (3)
Rx
performance. This section describes the differences in capacity for differ-
ent antenna configuration schemes. Figures 1 and 2 compare
A. Assumptions the capacity of antenna systems with respect to the number of
A few assumptions made in this paper are that the channel antennas used for transmission, reception, or both.
experiences flat Raleigh fading, channel coding is absent,
antennas are decorrelated or antennas spacing is at least half A. Single Antennas System - SISO
the wavelength of the signal, perfect channel knowledge is This antenna system has one transmit antenna and one
only available at the receiver, uniform power across transmit receive antenna.
antennas, and there is independent additive white gaussian Capacity = log2 (1 + ρχ22 ) (4)
noise (AWGN) at each receiver.
Where ρ is equal to the average received SNR and χ22
is a
chi-squared random variable with 2 degrees of freedom.
B. Organization of Paper
This paper is outlined as follows: Section 2 will describe the B. Multiple Transmit Antenna System - MISO
capacity performance metric, Section 3 will summarize both
linear and non-linear decoders used at the receiver, Section 4 ρ 2
Capacity = log2 (1 + χ ) (5)
will describe how the choice of receivers impact the error rate M 2M
performance metric, and Section 5 will conclude the paper. Where ρ is equal to the average received SNR and χ22M is
a chi-squared random variable with 2M degrees of freedom,
II. C APACITY while M is the number of transmit antennas.
Capacity describes the amount of bits that can be sent over C. Multiple Receive Antenna System - SIMO
the channel in one cycle or second per Hertz. It can also be
defined by maximizing the mutual information, defined below,
of the input and output of the system Capacity = log2 (1 + ρχ22N ) (6)
Where ρ is equal to the average received SNR and isχ22N
X fX,Y
I(X, Y ) = fX,Y logB ( ) (1) a chi-squared random variable with 2N degrees of freedom,
fX fY while N is the number of receive antennas.
X,Y
SURVEY OF LINEAR MIMO DECODERS AND THEIR IMPACT ON SPATIAL MULTIPLEXING PERFORMANCE IN NON-LINEAR LAYERED RECEIVERS, APRIL 20092

D. Multiple Transmit and Receive Antenna System - MIMO

M
X ρ 2
Capacity > log2 (1 + χ ) (7)
M 2k
k=M −N −1

Where ρ is equal to the average received SNR and χ22k is 600


Capacity at 10 dB AVG. at RX

a chi-squared random variable with 2k degrees of freedom. SISO


SIMO
MISO
One can observe from Figures 1 and 2 that MIMO provides 500
MIMO lower bound

the greatest amount of capacity out of the displayed antenna


systems, followed by SIMO. Another observation is that MISO 400

Capacity (b/s/Hz)
and SISO are very close in terms of capacity regardless of how
300
many transmit antennas are used. This shows that the amount
of receive antennas, used only in MIMO and SIMO, has the 200

greatest impact on capacity.


100

0
1 2 3 4 5 6 7 8 9 10
# of RX ants.

Capacity at 10 dB AVG. at RX
600
SISO
SIMO
MISO
MIMO lower bound
500

400 Fig. 2. Multiple Antenna System w/ avg. receive SNR of -10 dB


Capacity (b/s/Hz)

300

200
G=I (9)
100
This inhibits an exhaustive search through all available
symbols, given the used modulation technique, to find each
0
1 2 3 4 5 6
# of RX ants.
7 8 9 10
estimated symbol, ŝ. Decoding complexity in an ML receiver
is considered to be high, but there are algorithms, such as
sphere decoding, that have been proposed that combat this.

B. Zero-Forcing or Inverse Channel Receiver


Zero-Forcing (ZF) receivers look to eliminate interference
between each independent stream originating from one of the
Fig. 1. Multiple Antenna System w/ avg. receive SNR of 10 dB multiple transmit antennas. This is known as Multiple Stream
Interference (MSI). This is done through inverting the known
channel matrix, H, and setting it equal to G.
III. L INEAR R ECEIVERS
One aspect that makes linear receivers distinct is the G = H −1 (10)
methodology of using a linear filter to separate each of the
distinct transmitted streams. This filter is represented by the 1) Simple Analytical Solution:
weighted matrix G. The symbol estimate, ŝ is found by 2) Block-wise Analytical Solution:
selecting the symbol, s, multiplied by the weighted matrix, G, 3) Gauss-Jordan Method:
and the known channel matrix H, that is closest in Euclidean 4) Gaussian Elimination:
distance to the received symbol, y. This is displayed in the 5) LU decomposition:
below equation. 6) Moore-Penrose psuedoinverse: In this paper, we will
assume that H is at full rank and if H is not a square matrix
ŝ = mins kGy − GHsk2 (8) then the Moore-Penrose psuedo-inverse matrix will be applied
to find H̃ −1 . This is described in the following equations:
A. Maximum Likelihood Receiver
H̃ −1 = H̃ −1 HH̃ −1 (11)
A Maximum Likelihood (ML) Receiver provides the opti-
mal performance. In this case the linear filter matrix, G, is
equal to the identity matrix. H = HH˜−1 H (12)
SURVEY OF LINEAR MIMO DECODERS AND THEIR IMPACT ON SPATIAL MULTIPLEXING PERFORMANCE IN NON-LINEAR LAYERED RECEIVERS, APRIL 20093

2) Diagonal Bell-Laboratories Layered Space-Time: The


H̃ −1 H = (H̃ −1 H)H (13) OSUC receiver known as DBLAST has layers that can be
described as a vector of symbols on one antenna at a particular
time slot plus the next successive antenna at a future time slot.
HH̃ −1 = (HH˜−1 )H (14)
In the case of DBLAST a layer is diagonal from top left to
bottom right of the figure. One layer in the case of the above
C. Z inf ty Receiver figure, the symbol at antenna M and time slot t1 , the symbol
D. Minimum Mean-Squared Estimate Receiver at antenna M-1 and time slot t2 , and the symbol at antenna
The Minimum Mean-Squared Estimate Receiver (MMSE) M-2 and time slot t3 is stripped and decoded at one time.
also minimizes MSI, but also minimizes noise to lower errors.
C. Sphere Detection Receiver
The linear filter matrix, G, is found by choosing the G
that minimizes the expected Euclidean distance between the D. Lattice Reduction Receiver
product of the linear filter matrix and the received symbol and V. P ERFORMANCE C OMPARISON
the transmitted symbol. This is represented by the equation One can observe from figure 3, 4, and 5 that the SER/BER
below: performance is much better when using a combination of
2
an ML detector with an OSUC receiver, VBLAST. An ML
G = minG E[kGy − sk ] (15) estimator should provide you with optimal detector.

E. Joint-Channel Diagonalization 10
0
BER/SER Plots

BER
SER
F. Sphere Detector Receiver
IV. N ON -L INEAR −1
10

A. Successive Cancellation Receiver


One example of a non-linear receiver is a Successive Can- BER/SER

cellation (SUC) Receiver. The idea behind a SUC receiver 10


−2

is to treat each symbol, which can be described in terms of


space (antenna) and time, as a layer and is stripped away to
be decoded. This idea is displayed in the below figure. 10
−3

−4
10
−10 −5 0 5 10 15 20 25 30
SNR [dB]

Fig. 4. Performance of QPSK using ML/VBLAST receivers in 2x2 MIMO

BER/SER Plots
0
10
BER
SER

−1
10
BER/SER

−2
10

Fig. 3. Basic Diagram of Space-Time Layered Decoder


−3
10

B. Ordered Successive Cancellation Receiver


−4

An OSUC receiver is an SUC receiver, but first the layer 10


−10 −5 0 5 10
SNR [dB]
15 20 25 30

with the highest signal-to-interference ratio is selected to be


stripped and decoded first.
Fig. 5. Performance of QPSK using MMSE/VBLAST receivers in 2x2
1) Vertical Bell-Laboratories Layered Space-Time: The MIMO
OSUC receiver known as VBLAST has layers that can be
described as a vector of symbols in one time slot across
multiple elements in space. In the case of the above figure, VI. C ONCLUSION
all the symbols across antennas, M-2 to M+2, in one time slot R EFERENCES
t2 is considered to be a layer and is stripped and decoded at [1] On Limits of Wireless Communications in a Fading Environment when
one time. Using Multiple Antennas .
SURVEY OF LINEAR MIMO DECODERS AND THEIR IMPACT ON SPATIAL MULTIPLEXING PERFORMANCE IN NON-LINEAR LAYERED RECEIVERS, APRIL 20094

BER/SER Plots
0
10
BER
SER

−1
10
BER/SER

−2
10

−3
10

−4
10
−10 −5 0 5 10 15 20 25 30
SNR [dB]

Fig. 6. Performance of QPSK using ZF/VBLAST receivers in 2x2 MIMO

Вам также может понравиться