Federico - Master Thesis
Federico - Master Thesis
Federico - Master Thesis
Supervisor
Prof. Marina Mondin
Ing. Bruno Melis
Candidate
Federico Pacifici
March 2014
INDEX
INTRODUCTION
10
13
14
17
18
19
20
22
23
25
27
2.1 feICIC
28
2.2 CRS-IM
29
2.3 NAICS
30
2.3.1 MMSE
31
2.3.2 MMSE-IRC
31
2.3.3 E-MMSE-IRC
33
2.3.4
34
2.3.5
35
2.3.6
36
2.3.7
ML receiver
36
2.3.8
R-ML
38
40
41
43
45
46
47
48
50
51
51
58
61
61
63
67
39
69
69
71
76
79
CONCLUSION
85
BIBLIOGRAPHY
86
INTRODUCTION
The Master Thesis was written after an internship period at Telecom Italia S.p.A
(Wireless Access Innovation group). The objective of this Master Thesis is analyzing
and simulating advanced receiver schemes with interference rejection capabilities that
represent one of the next innovative step in the physical layer of LTE/LTE-A systems
providing higher throughput especially at the cell edge. Several receiver schemes are
analyzed and some of them are simulated to obtain performance results in terms of
Throughput and Raw BER. The selected receivers are chosen considering the trade-off
between complexity and expected gains.
A low complexity version of the MMSE-IRC receiver has been implemented in a link
level simulator specific for the LTE system, so performance results have been obtained
showing interesting features and using a low complexity technique for the estimation
of the interference covariance matrix. MMSE-IRC has been implemented as an
independent block to simplify the development of innovative receivers that use it as
elementary building block. The MMSE-IRC receiver outperforms the classical detection
schemes that treat the inter-cell interference as Gaussian noise, especially in case of
no colliding pilots between the serving and interfering cells. MMSE-IRC is also a
fundamental block of successive interference cancellation receivers operating at
symbol level (SLIC, Symbol Level Interference Cancellation) and bit level (BLIC, Bit Level
Interference Cancellation). In a second step of the analysis SLIC and BLIC receivers have
been implemented in a simplified link level simulator based on MATLAB and the
simulated performances are compared with the other considered receivers.
The
analysis showed that, even if the SLIC receiver complexity is higher than MMSE-IRC
one, it provides some gain especially in the low SINR region, while for higher SINR, the
successive interference cancellation functionality must be switched off to avoid the
error propagation effect. BLIC is more powerful, but its complexity is very high because
it performs the channel decoding also for the interfering signals. The Master Thesis is
structured into four chapters. In the first chapter, a Physical Layer overview of LTE and
LTE-Advanced systems is provided, focusing on the aspects that have been considered
for the receiver implementations. Chapter two gives an overview of the activity carried
out by 3GPP on advanced receivers. The third chapter shows the algorithm and the
implementation of the MMSE-IRC receiver in the LTE link level simulator, discussing
also the performance results. The last chapter describes the SLIC and BLIC receivers,
showing simulation performance in terms of throughput and Raw BER.
standards. Another important target is to provide higher data rates over the entire cell
area, including users at the cell edge. Theoretically, the maximum rate is limited by the
channel capacity that depends on the channel bandwidth and on the signal to noise
ratio, in presence of AWGN noise. This is a noise limited scenario, in which, the data
rates are always limited by the available received power or by the received signal
power to noise power ratio. When the bandwidth utilization is low, so the data rate is
lower than the available bandwidth, increasing the data rate requires a higher received
power, so an increase in the available bandwidth does not substantially impact what
received signal power is required for a certain data rate. On the other hand, in the case
of high bandwidth utilization, when the data rates is equal or higher than the available
bandwidth, an increase of data rate requires a much larger increase in the received
signal power, so an increase in the bandwidth will reduce the received signal power
required for a certain data rate. In conclusion, the transmission bandwidth should at
least be of the same order as the data rates to be provided.
Fixing a transmit power, to increase the received one, it is possible reduce the
attenuations, decreasing the distance, planning small cells and increasing the number
of cells. At the receiver side, another useful technique to provide high data rates is
using additional antennas, known as receive antenna diversity. Even at transmit side it
is possible to use multiple antennas, so combining signals received at the different
antennas the signal to noise ratio can be increased in proportion to the number of
antennas, allowing higher data rates. Multiple transmit or receive antennas techniques
are efficient up to a certain level beyond which there is only a marginal increase in the
data rates. This limit can be avoided using multiple antennas at both the transmitter
and the receiver side, using the spatial multiplexing or MIMO. There are also other
techniques, for examples focusing the total transmit power in the direction of the
receiver or reducing the noise power density improving the receiver design.
In the previous cases, the AWGN noise is the main negative contribution, but in real
scenarios, especially in mobile communication fields, the interference from
transmissions in neighboring cells, called inter cell interference, is the dominant source
of radio link impairment that usually occurs with a high traffic load. In addition to inter
cell interference, there could be another kind of interference, called intra cell
interference in which the useful signal is interfered by other signals within the current
cell. In this case, the maximum data rate that can be achieved in a given bandwidth is
limited by the SINR (Signal power to Interference and Noise Ratio).
One important difference between interference and noise is that interference, in
contrast to noise, typically has a certain structure which makes it, at least to some
extent, predictable and thus possible to further suppress or even remove completely.
More advanced topics about interference cancellation will be addressed carefully in
the next chapters, emphasizing some aspects that are the main job of this Master
Thesis, focusing on the implementations and performances of advanced receivers able
to cancel interferences in various scenarios.
From the operator point of view, bandwidth is a scarce and expensive resource, so
telecom operator would like to provide very high data rates within a limited
bandwidth. One way to increase the data rate is to use higher order modulations. In
3G systems (i.e. WCDMA) is used the QPSK modulation, nowadays high order
modulations such as 16QAM or 64QAM are used in HSPA to improve the bandwidth
utilization, providing higher data rates within a given bandwidth at the cost of reduced
robustness to noise and interference. Higher order modulation are normally combined
with channel coding giving more efficiency, paying attention that an additional channel
coding applied by using a higher order modulation scheme such as 16QAM may lead to
an overall gain in power efficiency compared to the use of QPSK. Setting a SINR there is
an optimal choice of modulation and channel coding to obtain the highest bandwidth
utilization.
Wider band transmissions are subjected to frequency channel selectivity that corrupt
the frequency domain structure of the signal, leading to higher error rates for a given
SINR. It is necessary to design a transmission scheme that avoids frequency channel
selectivity with low complexity. This goal can be reached by OFDM. This scheme
provides a lot of other benefits such as robustness against Intersymbol Interference
(ISI) through cyclic prefix insertion, IFFT/FFT digital processing, user multiplexing, multi
access etc.
transmission, in particular for cell edge users that experience low Signal to
Interference plus Noise Ratio (SINR) values.
In case of spatial multiplexing, up to four antennas at both the transmitter (base
station) and the receiver (terminal) side are used to provide simultaneous transmission
of multiple parallel data streams, also known as layers, over a single radio link, thereby
significantly increasing the peak data rates that can be provided over the radio link. As
an example, with four base-station transmit antennas, and a corresponding set of (at
least) four receive antennas at the terminal side, up to four layers can be transmitted
in parallel over the same radio link, effectively quadrupling the peak data rate with
respect to a single antenna system (i.e. SISO).
10
Downlink
Fallback
Transmission
Transmission
Mode
Scheme
Mode 1
Mode 2
Notes
LTE Rel.8
Transmit diversity
Single antenna
port
Transmit diversity
Mode 3
Transmit diversity
LTE Rel.8
Mode 4
Transmit diversity
LTE Rel.8
Mode 5
Multi-user MIMO
Transmit diversity
LTE Rel.8
Mode 6
Transmit diversity
LTE Rel.8
Mode 7
LTE Rel.8
Mode 8
Transmit diversity
or single-antenna
port
Transmit diversity
Mode 9
Up to 8 layer transmission
Transmit diversity
LTE Rel.10
LTE Rel.8
LTE Rel.9
Among the transmission modes defined in the 3GPP standard, Transmit Diversity
(Mode 2) and Open Loop Spatial Multiplexing (Mode 3) are supported in the first
equipment and terminal implementations and thus are of importance for the initial
roll-out of the LTE network. Switching between these two modes is decided by the
network as a function of the channel conditions, which is known to the eNode B
through the channel state information reported by the UE (CQI and RI). The accuracy
of RI reporting, which indicates the estimated number of simultaneous layers that can
be received by the UE, is a critical information for the optimal usage of TxD and SM in a
real LTE network.
11
This figure shows how is convenient to switch in a transmit diversity mode when SINR
is low.
The LTE physical layer offers data transport services to higher layers. The access to
these services is through the use of a transport channel via the MAC sub-layer. The
physical layer is designed to perform the following functions:
Rate matching
RF processing
12
An important characteristic of the LTE radio interface is that the frame duration and
Transmission Time Interval (TTI) are harmonized with those of UMTS/HSDPA system. In
particular the frame duration is equal to 10 ms while the subframe period, which
corresponds to the Transmission Time Interval (TTI), is equal to 1 ms (compared to the
2 ms of HSPA). Each subframe is divided in two slots, where each slot has a duration of
0.5 ms. Also the sampling frequency of the baseband (BB) signals are harmonized: for
UMTS/HSPA the baseband signal is sampled at 3.84 MHz, while for LTE the baseband
sampling frequency is equal to m/n3.84 MHz, where m and n are integer factors that
13
depend on the LTE channel bandwidth. These features reduce the complexity and the
cost of dual mode terminals that will support both radio interfaces.
There is one reference signal transmitted per downlink antenna port. The number of
downlink antenna ports P equals 1, 2, or 4. The RS of different antenna ports are
orthogonal among each other because resource elements used for RS transmission of
one antenna port are not used for any transmission by the other antennas (i.e. are set
to zero power for the other antennas).
14
Figure 3, Figure 4 and Figure 5, respectively show, the pilot pattern for a SISO case
when a normal or an Extended prefix cyclic is used, the CRS signals for two transmit
antennas (MIMO 2x2) and finally CRS signals for four transmit antenna (MIMO 4x4).
The cell specific RS sequence is a PN (pseudo random) sequence defined by a length-31
Gold sequence. The pseudo-random sequence generator is initialised with a value that
depends on the cell identity (cell-ID) so that different PN sequences are associated to
different cells. In this way the RSs of different cells have low values of cross-correlation
and thus the interference from neighboring cells can be reduced by proper averaging
on frequency adjacent reference symbols received at the UE.
Frequency hopping (FH) can be applied to the cell-specific reference signals. The
frequency hopping pattern has a period of one frame (10 ms). Each frequency hopping
pattern corresponds to one cell identity group.
The LTE standard foresees also UE-specific reference signals, also denoted in the
technical documents as DeModulation Reference Signals (DM-RS). The DM-RSs are
introduced for the support of beamforming techniques. The eNode B can semistatically configure a UE to use the dedicated reference signal as the phase reference
for data demodulation of a single codeword.
DL control signalling is located in the first n OFDM symbols (n 3) of a subframe and
consists of:
15
Note that there is not mixing of control signaling and shared data in an OFDM symbol.
Figure 6 shows the mapping between Control and Data symbols.
Control channels are formed by aggregation of control channel elements (CCE), each
control channel element consisting of a set of resource elements. The modulation used
for all control channels is QPSK.
16
Multiple physical downlink control channels are supported and a UE monitors a set of
control channels.
17
The Figure 8 shows the layer mapping operation for the case of four transmit
antennas.
18
An important feature of the Alamouti code is that only simple linear operations are
needed at the receiver for decoding.
In case of four transmit antennas the LTE standard adopts a combination of the
Alamouti code and the Frequency Switching Transmit Diversity (FSTD) technique. The
Figure 10 shows the principle of SFBC+FSTD encoding where S1,,S4 are the
modulated symbols coming from the layer mapping block. Notice that also in this case
only one codeword is transmitted.
Basically the SFBC + FSTD technique consists in the application of the Alamouti code
over pair of antennas.
19
The Alamouti code is applied over the antennas 1 and 3 for symbols S1 and S2, while
for symbols S3 and S4 the code is applied over the antennas 2 and 4. The antenna
pairing (1,3) and (2,4) is done in order to balance the different pilot density that is
lower for antenna 3 and antenna 4 compared to antenna 1 and antenna 2.
20
In Table 3 are showed the eight categories from 1-5 (LTE Release 8/9/10) to 6-8 (LTEAdvanced Release 10).
Table 3: UE Category
21
22
Ior1
I oc
Ior1
N BS
I
j 2
orj
where Iorj is the average received power from the j-th strongest base station implies (
Ior1 is the serving cell average received power), 2 is the thermal noise power over the
23
received bandwidth, and NBS is the total number of base stations considered including
the serving cell.
In addition to geometry, another measure, referred to as the Dominant Interferer
Proportion (DIP) ratio, was agreed as a key parameter for defining the interference
profiles. DIP was defined as the ratio of the power of a given interfering cell over the
total other cell interference power.
DIP of synchronized, and asynchronized interference, DIPi s , DIPi a is expressed as
follows:
DIPi s
Iors (i1)
DIPi a
I oc
Iora i
I oc
Ns
Na
j 2
j 1
a
I oc Iors j Ior
j N
and NBS = NS + Na is the total number of eNodeBs considered including the serving cell.
DIP ratio statistics have been derived obtaining both unconditional DIP CDFs and
conditional median DIP values, the latter conditioned on various geometry values. An
interference profile was defined on the basis of averaging unconditional median DIP
values submitted by the different companies. DIP values conditioned to the geometry
values have also been submitted by the different companies. Starting from these
values, the interference profiles that have been defined as part of the 3GPP feasibility
study to assess link level performance of MMSE-IRC receivers, are showed in Table 5.
24
Profile
Based on conditional
median values
Geometry
0 dB geometry
-3 dB geometry
Synchronized NW
DIP1
DIP2
-3.1
-5.4
-2.8
-7.3
Asynchronized NW
DIP
-3.1
-2.8
-2.5 dB geometry
The link level simulations have been done for the Extended Pedestrian A (EPA) channel
profile defined by the 3GPP.
25
The channel is assumed constant in each TTI (Transmission Time Interval of 1 ms).
The correlation of the fading processes is set according to the three cases defined in
3GPP for the conformance tests (TS 36.101 and TS 36.104) of the equipments (low,
medium and high correlation):
The and correlation values are, respectively, the correlation coefficient at the
transmitter and the correlation coefficient at the receiver.
The fading correlation is mainly determined by two factors: antenna characteristics
(e.g. distance, polarization) and propagation environment (e.g. number and position of
the scatterers, presence of LoS, angle of arrival and angle spread of the
electromagnetic waves).
The transmission schemes of LTE are differently affected by the correlation. In
particular transmit diversity (TxD) appears rather robust whilst Spatial Multiplexing
(SM) suffers a severe performance degradation as the correlation increases.
26
Study Item
IR (Rel.11)
Description
Interference Rejection
Focus
Focus on receiver structures targeting spatial
domain
interference
mitigation.
IRC
(Rel. 11)
CRS-IM
(Rel. 12)
(RP-130393)
NAICS
Network Assisted
(Rel. 12)
Interference Cancellation
Suppression
(RP-130404)
deriving
from
network
assistance
27
2.1 feICIC
In the feICIC case the focus is on the heterogeneous network scenarios where the
interference is mainly caused by the CRS and Control Channels of the macro cell on the
UEs connected to the small cells.
The main IC candidate techniques for the implementation at the UE side include:
Puncturing; receiver that punctures REs of the wanted signal of the serving cell
that are interfered by CRS REs received from one or more dominant interfering
cells.
In the case of CRS interference cancellation, the procedure requires: the channel
estimation from the interfering cells, regeneration of all the interfering cells CRS
signals and subtraction.
Puncturing is not applicable in several scenarios, e.g. with colliding CRS in non-MBSFN
ABS because CRE REs of the serving cell cannot be punctured. For SFBC and SFBC-FSTD,
two REs used should be punctured simultaneously when one of them was
contaminated. In the other cases, it sets the LLR of bits of REs undergoing strong
interference as zero.
The results show the better and robust performance and versatility of the CRS
cancelling receiver over the CRS puncturing receivers. Also for PDSCH demodulation
the CRS cancelling receiver outperforms the CRS puncturing receivers. CRS puncturing
receiver performs reasonably for single non colliding interferer, but for the other
scenarios it does not perform well. The relatively poor performance of the CRS RE
puncturing receiver for transmission mode 2 is because strong interference on one RE
affects demodulation of the two symbols that are transmitted through the affected RE
via SFBC encoding.
28
2.2 CRS-IM
Interference Mitigation (IM) of Cell-Specific Reference Signals (CRS) has been studied
in the Rel-11 Work Item on feICIC, where interference from CRS is dominant assuming
data RE muting in ABS subframes. A new study item has been started in 3GPP on CRS
interference mitigation (IM) in homogeneous network deployments.
The main objectives of this Study Item are:
identify the partial traffic loading levels, other realistic system level parameters
(e.g. traffic and interference models, time and frequency offset between cells)
and performance metrics for studying the feasibility of CRS-IM in a
synchronized homogenous network;
identify the baseline receiver which can be used for evaluating the gain of CRS IM in a synchronized homogenous network considering the reuse of CRS-IM
receiver assumed for Release 11 feICIC and the reuse of MMSE-IRC receiver as
the baseline receiver;
evaluate the system level and link level gains of CRS-IM with respect to the
baseline MMSE-IRC receiver in a synchronized homogenous network
deployment under the various loading levels identified (e.g. gains of CRS-IM
from 1 and 2 aggressor cells CRS shall be evaluated and compared).
The objectives of the study item explicitly indicate that only Release 11 CRS assistance
information should be assumed to be available. It can be seen that the CRS assistance
information consists of a list of cells which are to be considered as candidates for CRS
interference mitigation. Therefore, for each cell the information related to the CRS
transmission (i.e. the physical cell ID, antenna port count and MBSFN configuration)
are provided to the UE.
A way forward on CRS-IM performance evaluation has been agreed. The first proposal
is the reuse of CRS-IM receiver assumed for Release 11 feICIC to mitigate CRS
interference of up to two cells. The second solution is the reuse of MMSE-IRC based
receiver with interference covariance matrix estimation, here the receiver does not
differentiate CRS or data interference when suppressing them.
29
The proposed receiver scheme for the execution of the link level simulations is the
MMSE-IRC with/without CRS-IM. Concerning the CRS-IM part of the receiver, basically
it consists in the regeneration and subtraction of the CRS signal from only the 1 st or
both the 1st and 2nd strongest interfering cell. A possible receiver implementation is
depicted below. A possible work item on this activity can follow.
2.3 NAICS
NAICS is similar to the approach of CRS-IM. The major difference for NAICS is that the
interference mitigation is now targeted not only for interfering CRS but also for
interfering PDSCH.
Objectives of this Study Item for RAN4 are:
Identify reference IS/IC receivers with and without network assistance, and
evaluate
their
performance/complexity
trade-off
and
implementation
feasibility;
Evaluate the link-level gain over baseline Rel-11 linear MMSE-IRC receivers and
Rel-11 non-linear receivers required for feICIC;
30
Indicate (to RAN1) assumptions on the network assistance information for the
evaluated receivers under possible network coordination.
In the following part of this chapter, it will be shown a brief description of the main
advanced receivers with interference cancellation/mitigation capabilities.
2.3.1 MMSE
The Rel-8/Rel-9 baseline receiver, MMSE receiver, ignores the fact that interfering
signals are spatially colored signal. MMSE receivers treat interference as white noise.
Along with the channel matrix H for the desired signal, only interference-plus-noise
power I2 n needs to be estimated by the MMSE receiver. The MMSE receiver can be
expressed as:
s H H HH H I2n I x
1
The complexity of the Rel-8/Rel-9 MMSE receiver is given by: the channel estimation
and the matrix inversion.
2.3.2 MMSE-IRC
31
A second approach to realize the MMSE-IRC receiver is using the CRS or DMRS from
the serving transmitter to estimate the channel matrix H of the desired signal, and
using the differences of the received reference signal and the re-constructed reference
signal with the estimated desired channel on the CRS or DMRS REs to estimate
interference-plus-noise covariance matrix RI n :
s H H HH H RI n x
1
RI n
k ,lRS
k ,l
H xk ,l yk ,l H xk ,l
The RAN4 Rel-11 advanced receiver study shows that CRS or DMRS-based MMSE-IRC
receiver outperforms data RE-based MMSE-IRC receiver. The above MMSE-IRC
approaches can be applied to intra-cell interference suppression in MU-MIMO
scenarios as well as to inter-cell interference suppression.
For the Rel-12 NAICS SID, it would be a logical extension to study the possible
performance gain of an MMSE-IRC receiver when the system assists UEs in performing
better channel state information estimation, for both desired and interference signals.
For example:
s H HH H H i H iH I2 I x
i 1
32
2.3.3 E-MMSE-IRC
limited complexity;
In this context, the received signal is given by the superposition of one useful signal
and N-1 interferer signals with different precoding matrix and different amplitudes:
N 1
yk
H
i
i , k Pi xi , k
nk
i 0
where, i is the amplitude of the signal transmitted from i-th cell, H i ,k is the channel
matrix of the i-th cell on the k-th tone / resource element (RE), xi ,k is the symbol
transmitted by the i-th cell in the k-th tone and
Pi is
by the i-th cell and K is the total number of observed tones. The number of cells in this
case is N with one serving cell and N 1 interferers.
33
Core receiver processing includes symbol level detection of the desired cells signals
and Turbo decoding. At the detector stage, Rel-11 MMSE-IRC receivers suppress the
transmission from interfering cells before detecting the desired symbols. The nulling
operation is performed by a front end MMSE filter, W, and Wy is the linear estimate of
the transmitted symbols. For Rel-11 MMSE IRC receivers, W is constructed using: the
channel estimation of serving cell and the total interference and noise estimated using
CRS or DMRS. In contrast, even if E-MMSE-IRC receiver perform some similar
functions, there are some key differences:
the interfering signals are modelled using the estimated channels of the
interferers, using CRS-IC;
for each signal the precoded matrix is needed and it is obtained using UE-side
blind estimation or network signaling;
There are two types of successive interference cancellation (SIC) receivers: in the first
one only symbol demodulation is involved in the SIC process and in the other one the
FEC decoding is involved. It can be expected that, if the FEC decoding is involved in the
SIC process, the performance will be improved compared to the one only using symbol
demodulation. However, FEC decoding will require that all detailed coding information
34
P
1
s H H HH H n2 I y H i ~
si
i 1
where ~
si is the quantized estimation of the interference signal.
The symbol level SIC receiver needs to know the modulation order of the interference
signal, power offset and (an estimate of) the channel matrix of the interferers as well.
This requires system assistance in providing the interference modulation order and
providing means to estimate the interference channel matrix. It is a general
understanding that an SIC receiver can perform well in case that the interference
signal is much stronger than the desired signal. Therefore, SIC receivers are well suited
for some inter-cell interference scenarios (like range extension in HetNet, or intra-cell
interference in some MU-MIMO cases). However, for inter-cell interference in
homogeneous networks, the interference signal can generally be expected to be
weaker or not much stronger than the desired signal. In this case, the performance
advantage of SIC receiver over MMSE-IRC receiver may be questionable.
The receiver attempts to detect and decode one by one the interferers of interest, also
in case of MU-MIMO and/or inter-cell interference cancellation. The decoded
interferers are subtracted step-by-step to the overall signal, obtaining at the end the
decoded useful signal.
This receiver takes advantage of the CRC attached to each transport block before
channel coding: if CRC check is successful, the block has been correctly decoded and
the interfering signal can be reconstructed (minor the channel estimation errors). The
35
bit level hard SIC to be efficient needs to find at least one interferer that can be
decoded without error (in order to subtract its interference from the useful signal).
As a result, the situations where the interference power is much higher than the useful
signal power and/or when the interference has a robust MCS are favourable situations
where it brings significant gains. In case the interference and useful signal have similar
powers, the Hard SIC imposes the constraint that the MCS used by the first interferer
be more robust than the MCS used for the signal of interest, as it will need to be
decoded under the interference of the latter.
This receiver scheme performs the soft detection and the Turbo decoding of the UE
signals which are repeatedly subtracted from the received signal.
An important parameter of these receivers is the number of Turbo-code iterations for
each detection and decoding step. In the case of Turbo-SIC receivers (also in the Hard
SIC), the victim UE needs to know the following transmission parameters of the
interferers:
PRB assignments;
MCS;
RNTI;
2.3.7 ML receiver
36
i 1
37
with Rank 1 transmission, the dominant interfering signals could be jointly detected
with limited additional NA information.
2.3.8 R-ML
where,
H
H
y Hx R1 y Hx
y Hx R1 y Hx
LLR(bi ) log e
log
e
x0 ( bi )
x1 ( bi )
where k (bi ) denotes the set of transmit vectors with bi k, k 0 ,1 , and R is the
noise covariance matrix.
Using a Rel.11 MMSE-IRC receiver, the interferer term (the second one inserted in the
received signal) can be used to calculate the interferer plus noise covariance matrix R
in this way:
R H2 W2 W2H H 2H E nn H .
Finally, about the R-ML, LLRs can be also represented by max-log approximation:
LLR bi min y Hx
x0 ( bi )
38
R1 y Hx min y Hx
x1( bi )
R 1 y Hx
matrix, y is the received symbols 2x1 matrix and k (bi ) denotes the set of transmit
vectors with bi k, k 0 ,1 .
R-ML is a reduced complexity version of ML, but it is more complex than the previous
receiver schemes, even if it provide sub-optimal performance.
39
3.1 Overview
This chapter provides a detailed vision of all the aspects that led to a low complexity
implementation of the MMSE-IRC receiver in real scenarios. Moreover, performance
results are showed and explained carefully, taking care to select relevant results that
best show the behavior of the receiver.
At the starting point, a brief analysis of the simulation platform is provided, focusing
on some key blocks that are the core of a MIMO OFDMA link level simulator and that
are useful to understand the MMSE-IRC implementation inside it. It is not possible to
describe the overall architecture, since this simulator is composed by a very large
number of blocks. The link level simulator is designed for the simulation of MIMOOFDM based wireless communication systems like LTE/LTE-A and represents an
effective tool for the research and development of innovative physical layer system
components.
Simulations are obtained adding an independent block, the MMSE-IRC receiver, into
the physical layer simulator, developed using CoCentric System Studio. MMSE-IRC
block is intentionally implemented as a unique block, putting inside the corresponding
functionalities, with the objective to have an interferer cancellation receiver that can
be accessible and modifiable quickly. The designed MMSE-IRC is a unique simulation
block implemented in C language.
The main implementation constraint for our MMSE-IRC is the low complexity. Some
techniques are used to reduce the computation burden: reducing the complexity of
the matrix inversion, averaging and weighting coefficients computation.
Interfering scenarios are selected, first of all, to test the MMSE-IRC code and after to
visualize the performance in terms of Raw BER, BLER and Throughput in presence of
single or double interfering cells selecting different spatial correlations and DIPs.
Performance results are compared with the baseline receiver based on the Alamouti
detection scheme [ref. paper di Alamouti], using ideal implementations developed in
MATLAB and also with the more realistic simulator based on CoCentric System
40
Studio, showing how in the most cases MMSE-IRC provides a performance gain with
respect to Alamouti.
In the next chapter are also shown two other advanced receiver schemes that exploit
the MMSE-IRC algorithm and are based on the symbol level interferer cancellation
(SLIC) and bit level interferer cancellation (BLIC) concept, comparing them and showing
interesting features in order to develop an adaptive receiver that is able to switch or
adapt the interference cancellation algorithm as a function of the channel and
interference conditions.
41
42
The explanation will be concentrated in the mapping block. The demapping block
basically performs the inverse operations, so just the most important differences will
be pointed out.
43
example, in the LTE system, the BRU (in this case called Resource Elements) has some
positions reserved for the pilot subcarriers. For this reason, to describe the
internal structure of the BRU, it is defined a data set containing the indexes of the
subcarriers that can be used for data transmission. By means of this data set, it is also
determined the filling order of the structure, such as frequency-first, time-first or any
other order, depending on the order the subcarriers indexes are listed.
The generic resource grid (GRG) represents all the allocable resources within a
time/frequency zone, being constituted by BRUs. It is important to remark that all
BRUs within a GRG must have the same structure and filling order, as previously
explained. The GRG has rectangular dimensions defined by the parameters
GRG_freq_size and GRG_time_size, given in number of BRUs in frequency and in time,
respectively. The numbering of the BRUs inside the GRG is shown in Figure 4.
Regarding the implementation of the block, it is also useful to view the GRG in terms of
subcarriers, with the correspondent dimensions and numbering shown in Figure 5.
44
The BRUs inside the GRG are allocated by the specification of GDRs, as will be
explained in the following. BRUs not allocated have all their subcarriers filled with
zeroes.
The purpose of the mapping block is to map the symbols of different types (data,
pilots, other signals) that arrive organized in a logical manner (logically indexed), into
theirs correspondent physical resources, given a mapping rule. A physical resource is
defined as a physical subcarrier (i.e., a given position in the IFFT/FFT) at a given time (in
terms of OFDMA symbol offset). The physical resources are positioned over a grid with
45
dimensions NFFT x Nsymb. NFFT is the IFFT/FFT size and Nsymb corresponds to the
maximum between the pilots pattern repetition period and the extension in time
where the mapping rule applies (i.e., the maximum offset in time between a logical
index and its correspondent physical index). The numbering of the physical resources
in the grid is done as shown in the Figure 6.
In general, a logically indexed subcarrier at the input can be mapped into any physical
resource in the grid.
Besides NFFT and Nsymb, other additional parameters shall be provided to the model:
Ndata: total number of data subcarriers in the grid, also equivalent to the rate
of the data input port;
Npilot: total number of pilot subcarriers in the grid, also equivalent to the rate
of the pilots input port;
Nnull: total number of null subcarriers in the grid, including guard, DC, and
other null subcarriers (when using MIMO, for example);
Nother: total number of subcarriers in the grid dedicated to other signals, such
as synchronization signals or control channels in LTE.
In transmission, the Generic IFFT & Cyclic Prefix Insertion model, as its name already
states, performs the IFFT calculation of the spectrum defined by the input subcarriers.
46
The FFT size depends on the channel bandwidth being considered. In sequence, the
cyclic prefix is inserted taking a copy of a given number of samples (Cyclic Prefix
length) at the end of the useful OFDM symbol (just after the IFFT calculation) and
inserting them before it.
In reception, considering that the system is ideally synchronized and that time
windowing is not performed over the OFDMA symbol, the Generic FFT & Cyclic Prefix
Removal model performs the inverse operations done in transmission. First, it removes
the beginning of the OFDMA symbol corresponding to the cyclic prefix. Finally, it
performs the FFT calculation of the useful OFDM symbol.
The purpose of the generic pilots compensation model is to compensate the received
pilots to remove the power boost and the specific pilot sequence, based on the
knowledge of the transmitted (reference) pilot sequence. After doing that, the value of
each pilot symbol represents an estimate of the channel seen by the pilot subcarrier
itself.
47
The block operates over the same grid of the subcarriers mapping (see the Figure 3),
therefore using the same parameters and data sets (just the necessary ones) to know
the location of the pilots subcarriers.
The purpose of the generic channel estimation block is to estimate the channel
coefficients correspondent to the received data symbols. These estimated values are
used in the subsequent blocks of the chain to perform some data processing over the
data symbols. The channel estimation is based on the received pilot subcarriers that
should be already compensated prior to enter in the block to remove power boost and
the specific pilot sequence.
The estimation of the channel coefficients is performed using linear interpolation,
linear
extrapolation and the hold operation (which is indeed a particular case of linear
extrapolation).
48
First of all, it is defined an interpolation grid, with frequency length equal to the
IFFT/FFT size and time duration Nsymb, equal to the periodicity of the interpolation
rules. The parameter Nsymb is not necessarily the same defined in the generic
subcarriers mapping model. The contents of the grid are the channel estimates of the
correspondent subcarriers. An interpolation rule is a linear operation involving 3 points
in the grid, where the channel estimate of a destination subcarrier is obtained from
the known estimates of the two source subcarriers, considering the 3 points are
positioned over a straight line.
Therefore, it is possible to calculate the channel estimate related to a given subcarrier
(destination), providing the channel estimates of the two source subcarriers and the
proper weights. The weights are a function of the subcarriers indexes and can be precalculated for every defined interpolation rule. This information is then provided to the
block by the data sets shown in the Figure 9, which contain all the interpolation rules
(meaning first operand indexes and weights, second operand indexes and weights, and
destination indexes) to be performed in the grid. The channel estimation is done in
steps, starting from the step 0, where at the beginning just the received pilot symbols
are known. The pilots are assumed to be already compensated to remove
sequence and power boost. A step includes all the interpolation rules that can be
defined using all channel estimates known at the end of the previous step. New steps
should be included until all the required channel estimates are obtained.
49
The purpose of using the technique of space-time coding and decoding is to support
Multiple Input Multiple Output (MIMO) antenna systems in order to also exploit the
spatial dimension. As consequence, an improvement in the capacity (throughput) or in
the reliability (coverage range) of a wireless communication system can be obtained.
Two possible transmission modes of MIMO systems, the spatial multiplexing (SM) and
the space-frequency block coding (SFBC). The spatial multiplexing is based on the
transmission of different data streams across the different transmitting antennas with
50
the goal of increasing the overall throughput, while the space-frequency coding
techniques transmit redundant data streams over the multiple antennas for increasing
the link reliability and extending the coverage range.
MMSE-IRC receiver is based on the MMSE criteria, but the interference rejection
combining require highly accurate channel estimation and covariance matrix
estimation that includes inter cell interference. In this scheme, the covariance matrix is
used in a modified version that provides lower complexity, avoiding the 4x4 matrix
inversion (MIMO 2x2 SFBC), leading to a trivial 2x2 matrix inversion.
Lets consider a scenario where there is an UE and some interfering cells, the received
signal by UE antennas is:
where, considering a MIMO 2x2 SFBC transmit diversity, Y is the 4x1 matrix containing
the received signals by UE,
the total interference received 4x1 matrix, N is the 4x1 noise matrix.
In a real context, the UE receives the summation of many signals plus noise composed
by the useful signal (from the serving cell) and interferences (from interfering cells):
51
where:
UE,
is the 4x2 channel frequency response matrix between the c-th cell and the
is the 2x1 transmitted signal matrix and
is the
This matrix formulation can be extended, considering the SFBC transmit diversity in a
2x2 MIMO fashion. When SFBC is enabled, considering two antennas in transmission
and two in reception (2x2 MIMO), the transmitted symbols are Alamouti coded
exploiting two adjacent subcarriers and the two antennas, sending for each time
instant four symbols mapped in subcarrier k (even) and k+1 (odd). So, the previous
matrix equation can be expanded as:
=
[
] [
]
[
The UE, implementing the MMSE-IRC, estimates the useful signal, in particular the 2x1
matrix composed by two estimated serving cell symbols transmitted by the two
antennas:
Where,
52
is the sign
where,
code and spatial domains. This matrix is generated considering the estimated channel
matrix of the useful signal and the interferers plus noise covariance matrix.
can be calculated using one of the two following methods:
1th Method
where,
is the estimated
interferences plus noise covariance matrix. From simulation tests, using this method it
was noted that the estimate received symbol have to be normalized through the
following normalizing
symbol the matrix element
so,
and
53
2nd Method
In this case, the normalization process is not necessary as the estimated symbols are
already normalized (i.e. the amplitude is correctly scaled for the subsequent symbol to
bit demapping operation).
Both methods provide the same result; the first one is a reduced complexity method
because the matrix inversion is done considering only one operand (
with
eventually a splitting zero-adding operation. The second method can be used when
does not contain zero values.
The covariance matrix
where I is the total interference received by the UE. It can be expressed neglecting the
received useful signal (
At the UE receiver, the total interference received by the UE on the subcarriers k and
k+1 can be expressed as:
54
where
:
|
|
|
|
|
|
|
the main diagonal represents the received interferer powers at antenna port 1 and
port 2,
the other expectation terms are the correlation functions between the interfering
signals at antenna port 1 and port 2, the null terms are the auto-correlation function of
the interference calculated over two adjacent subcarriers that can be assumed equal
to zero.
Besides, also the terms
],
|
|
|
|
|
|
Obviously, the complexity is much lower in terms of matrix inversion, but the price to
pay is a very small performance degradation.
For simplicity, expectations can be expressed as:
55
Considering that the interference characteristic changes slowly in time and frequency
domains, it possible to write:
So,
[
, it is only needful
is a simpler operation,
is simply:
In the following, the relative sub-carrier indexes k and k+1 are omitted, considering:
56
Calculating the inverse matrix, it is possible to show the low complexity procedures:
and,
where
][
]
[
Imposing that
is:
57
It is important to note that, the above values are complex symbols, so they must be
divided in real parts and imaginary parts, doing complex operations. Extending above
formulas, considering complex values, there are not important simplifications, so it is
not convenient splitting real part and imaginary part, but sometimes it is the only way
to proceed. Fortunately, CoCentric is able to treat complex values and operations using
a specific complex data format. So, in the following all the implemented variables are
treated as complex values.
As already mentioned, the description of the main employed simulator blocks and the
math procedure, is fundamental to build an independent block that accurately
represents the MMSE-IRC receiver.
The math description appears very simple, because the approximations and calculus
are simple to understand and to realize on paper, but a real realization into a real
LTE/LTE-A simulator or in a real LTE/LTE-A chipset has to be done opportunely, solving
several implementation problems, engineering some calculus to respect the LTE/LTE-A
standard and the simulator software context.
Choosing to implement all the operations inside the MMSE-IRC receiver (e.g.
covariance matrix estimation), the input and output ports are:
58
INPUT PORT
OUTPUT PORT
float symbols_in1_I;
float symbols_out_I;
float symbols_in1_Q;
float symbols_out_Q;
float symbols_in2_I;
float reliability;
float symbols_in2_Q;
float reference_pilots_in1_I;
float reference_pilots_in1_Q;
float reference_pilots_in2_I;
float reference_pilots_in2_Q;
float h11_I;
float h11_Q;
float h12_I;
float h12_Q;
float h21_I;
float h21_Q;
float h22_I;
float h22_Q;
Table 10: Input/Output MMSE-IRC block data
Moreover some data set files have to be loaded to know the position of useful data
(e.g. pilot subcarrier indexes). In order to estimate the covariance matrix, it is very
important know exactly the CRS positions, so mapping CRS indexes in a data file it is
possible to extract the interested data from a PRB.
So, to estimate the covariance matrix
the total received interferer is obtained subtracting the estimated received signal to
the total received signal, for each OFDM symbol and subcarrier that belong to the CRS
resource elements:
where,
59
by UE at k-th subcarrier and l-th OFDM symbol, is the estimate channel response of
the serving cell,
The average operation can be done considering a sliding windows to select the number
of PRB and so the number of CRS inside the sliding windows. The parameter K
establishes the size of the sliding window: for example, if K=1 the estimated covariance
matrix, the interferer power at the UE antenna 1 and the interferer power at the UE
antenna 2 are calculated for each PRB.
The following C-code shows the implementation of the sliding window and the
estimations of covariance matrix and interferer powers at the UE antenna 1 and 2:
for(i = 0; i < Nsymb*NFFT; i++)
{
for(i = 1; i <=Nprb-K+1; i++)
{
idx_low = G_LEFT + (i-1)*12;
/* First index within SW */
idx_high = G_LEFT + (i-1)*12 + Nrew_f-1; /* Last index */
if(idx_low >= DC_POSITION) idx_low++;
/* Skip DC */
if(idx_high >= DC_POSITION) idx_high++;
/* Vector with the pilot indexes within the sliding window */
m = 0;
x = 0;
for(n=0; n<Npilot; n++)
{
j = pilot_subcarriers_indexes_1[n] % NFFT;
f = pilot_subcarriers_indexes_2[n] % NFFT;
if(j >= idx_low && j <= idx_high)
{
p_idx_w1[m] = n;
m++;
}
if(f >= idx_low && f <= idx_high)
{
p_idx_w2[x] = n;
x++;
}
}
for(n=0;
{
l
x
j
f
n<Npw; n++)
=
=
=
=
p_idx_w1[n];
p_idx_w2[n];
pilot_subcarriers_indexes_1[l];
pilot_subcarriers_indexes_2[x];
60
IMCS 7;
QPSK modulation;
TBS=6200 bits;
);
61
ALAMOUTI
SINR [dB]
-12
MMSE-IRC
THR [Kbit/s]
3
THR [Kbit/s]
Gain [%]
5,4304 74,99919
-8
147,397
259,884 76,31566
-4
1582,582
1975,125 24,80396
0
3673,298
4321,846 17,65574
4
5765,565
5964,164 3,444571
8
6187,587
6186,036 -0,02507
12
6200
6200
0
Table 11: MMSE-IRC and Alamouti THR
6 000
5 000
4 000
3 000
ALAMOUTI
2 000
MMSE-IRC
1 000
0
ALAMOUTI
-12
3
-8
-4
12
62
After having done the preliminary simulations, the real simulator is extended adding
two transmission chains to insert real interference cells. The simulator schematic is
showed in the next figure. In the case of one interferer signal, DIP1=-0.1dB and DIP2=0.001, so the first interferer power is very high and the second one is about zero.
Moreover, several simulations are performed changing the angle of arrival (AoA) of the
interferer, from 0 degree (worst case) to 45 degree (best case). Fixing the AoA of the
useful signal at 0 degree, if the AoA of interferer is 0 degree, the interferer signal is
perfectly aligned to the useful signal so the spatial filtering of the interference is very
difficult. In contrast, if the AoA of the interferer is 45 degree and AoA of the useful
63
signal is 0 degree, the spatial filtering can provide some interference rejection. In
general, it is very important knowing performances in different network scenarios (i.e.
physical layer simulations for different IMCS), also to optimize the network planning.
Adjacent cells can use the same pilot pattern in their transmitted frames or a planning
can be done selecting different pilot positions for each cell.
64
Moreover, several simulations are performed changing the angle of arrival (AoA) of the
interferer, from 0 degree (worst case) to 45 degree (best case). Fixing the AoA of the
65
useful signal at 0 degree, if the AoA of interferer is 0 degree, the interferer signal is
perfectly aligned to the useful signal so the spatial filtering of the interference is very
difficult. In contrast, if the AoA of the interferer is 45 degree and AoA of the useful
signal is 0 degree, the spatial filtering can provide some interference rejection. In
general, it is very important knowing performances in different network scenarios (i.e.
physical layer simulations for different IMCS), also to optimize the network planning.
Adjacent cells can use the same pilot pattern in their transmitted frames or a planning
can be done selecting different pilot positions for each cell. Obviously, the UEs,
subjected to a reception of interferer signals, in the first case receives interfered pilots
(colliding pilots case), in the second case receives non interfered pilots (no colliding
pilots case). Both the cases are simulated, showing interesting results.
Lets start with the colliding single interferer case, the simulation parameters are:
IMCS 7;
QPSK modulation;
TBS=6200 bits;
);
considered AoA
66
Array elements = 2;
7 000
Throughput [kbit/s]
6 000
5 000
4 000
3 000
2 000
1 000
0
-10 -8 -6 -4 -2
10 12 14
SINR [dB]
Figure 16: THR comparison (Real Interference) - Colliding pilots
The above figure shows that the MMSE-IRC receiver provides a gain of 6-7 dB at about
3.5 Mbit/s when the interferer AoA is 45 degree. The gain decreases when the
direction of the interference is close to the useful signal one and becomes negative
when the interferer is perfectly aligned to the useful signal.
The simulation scenario is about the same, changing only the pilot pattern for the
interferer signals, modifying appropriately the CRS position inside the resource blocks
(i.e. a pilot pattern shift is applied to the interfering signal). Since pilot are not
colliding, channel estimation is more accurate than in the colliding case, so a better
estimation of interfering signal permits to estimate the interferer covariance matrix
with higher accuracy. This facts are reflected by the simulation results, in fact,
comparing the colliding case with the no colliding case, when AoA of the interferer
signal is 45 degree, the gain of no colliding case is about 6 dB at 5 Mbit/s. For different
interferer AoA the gain and throughput are always consistent.
67
In the colliding case, if the interfering AoA is between 0 and 10 degree, the Alamouti
receiver seems to provide better performances than MMSE-IRC. Conversely, about the
no colliding CRS case, the MMSE-IRC receiver always provides a gain respect to
Alamouti receiver, as shown in the figure below:
Throughput [kbit/s]
7 000
6 000
5 000
4 000
3 000
2 000
1 000
0
-10 -8 -6 -4 -2
0 2 4
SINR [dB]
10 12 14
68
69
where,
About SLIC, channel estimation, detection and interferer cancellation are performed N
times (N is the number of transmitting cells including serving plus N-1 interfering cells)
so the complexity scales linearly; the decoding procedure is perfomed only once for
the serving cell. In contrast, even the BLIC decoding is made N times, typically N turbo
decoding operations, so the BLIC complexity is much higher. Moreover, interference
parameter estimation is significantly more complex for BLIC, since the encoding
scheme and entire RB allocation information are needed for all the interferers.
Finally, SLIC receivers are a good choice for interference cancellation because their
interference cancellation capability gives a reasonable complexity that scales linearly
with the number of interferers. In contrast, the BLIC receiver complexity is higher than
SLIC one, but BLIC can give higher performance due to the turbo decoder that operates
on each interferer.
In the next paragraphs will be presented the description of SLIC and BLIC receivers,
showing implementation details and performance simulation results. The simulation
70
results are presented considering single and double interferers for different spatial
characteristics expressed in terms of Angle of Arrivals (AoAs).
The results show the SLIC gain respect to MMSE-IRC and the SINR window in which it is
possible apply the dynamic on/off switching of the interference cancellation
functionality. BLIC simulation results are compared to the others one, showing that in
terms of performance (Throughput and Raw BER), BLIC receiver is the optimal choice
to reach the highest throughput when the UE is at the cell edge.
represented in MATLAB as a function that returns the SFBC decoded symbols r and
their reliability:
[r,reliability] = slic(y,y1,y2,y3,Heff1,Heff2,Heff3,Nr,Nlay,Cmod);
y is the total received signal (useful plus interferences and AWGN noise)<
y1 is the useful signal;
y2 is the first interfering signal;
y3 in the second interfering signal;
71
72
At the output of the first step an estimation of the first interferer is provided (the
73
strongest). As shown in the Figure, for the first dominant interferer estimation, the
MMSE-IRC receiver receives as input the estimated channel matrix of the second
interfering signal Heff2, the total received signal y and the estimated covariance matrix
considering the residual interference. The covariance matrix is calculated considering
the low complexity formulation treated in the chapter 3, and it is calculated
considering the residual interferer signal viewed excluding the first dominant
interferer:
y_calc1=y-y2;
P1=1/length(y)*sum(abs(y_calc1(1,:)).^2);
P2=1/length(y)*sum(abs(y_calc1(2,:)).^2);
r12=1/length(y)*sum(y_calc1(1,:).*conj(y_calc1(2,:)));
r21=conj(r12);
Ri=[P1 r12; r21 P2];
Ri is a 2x2 matrix that is provided as input to the MMSE-IRC:
[r2,reliability] = mmse_irc(y,Heff2,Nr,1,Ri);
The MMSE-IRC receiver at the first step acts on the total received signal by UE
(denoted as y), knowing the estimated channel matrix of the first interfering signal and
the covariance matrix of the residual interference signal.
The symbols r2 represent a rough estimation of the first interferer symbols. To
improve the estimation, the proposed SLIC receiver regenerates the 1th interferer
making the following operations:
74
w_hard(idx)=0;
idx = w<0;
w_hard(idx)=1;
Modulation (x-QAM)
r2 = modulation(1,Cmod,w_hard);
Alamouti encoding
s_2 = alamouti_enc(r2);
So, knowing the regenerated Alamouti encoded symbols s2 and the estimated channel
matrix of the first interferer, it is possible to calculate the regenerated first interferer
as:
y2_est=Heff2*s_2;
This signal, subtracted to the total received signal y, permits to calculate the received
signal without the presence of the strongest interferer.
y_2=y_1-y2_est;
This is the input of the second step, in which the second interferer is estimated. At the
second step, MMSE-IRC acts on the second interfering signal knowing its channel
matrix and the covariance matrix of the residual interferer:
y_calc2=y-y3-y2_est;
P1=1/length(y)*sum(abs(y_calc2(1,:)).^2);
P2=1/length(y)*sum(abs(y_calc2(2,:)).^2);
r12=1/length(y)*sum(y_calc2(1,:).*conj(y_calc2(2,:)));
r21=conj(r12);
Ri=[P1 r12; r21 P2];
[r3,reliability] = mmse_irc(y_2,Heff3,Nr,1,Ri);
75
After MMSE-IRC, the same procedure is made: soft demodulation, hard decision,
modulation and alamouti encoding. So, the second interference can be estimated and
regenerated:
y3_est=Heff3*s_3;
Having estimated, regenerated and subtracted the first and the second interfering
signals, the remaining part is composed by the useful signal and thermal noise. So, the
useful signal plus noise, is:
y_3=y_2-y3_est;
Applying the MMSE-IRC to this signal, knowing the estimated channel matrix of the
useful signal (Heff1) and the covariance matrix calculated only for the noise, it possible
to obtain the alamouti decoded useful symbols:
y_calc3=y-y1-y2_est-y3_est;
y_3=y_2-y3_est;
P1=1/length(y)*sum(abs(y_calc3(1,:)).^2);
P2=1/length(y)*sum(abs(y_calc3(2,:)).^2);
r12=1/length(y)*sum(y_calc3(1,:).*conj(y_calc3(2,:)));
r21=conj(r12);
Ri=[P1 r12; r21 P2];
[r4,reliability] = mmse_irc(y_3,Heff1,Nr,1,Ri);
The useful symbols r4 is then soft demodulated and turbo decoded.
Finally, the performance measurements can be done, they are showed comparing
Throughput and Raw BER of BLIC, SLIC, MMSE-IRC and Alamouti receivers.
76
The method used for the SLIC receiver applied to MMSE-IRC are the same, but instead
of applying the hard decision after the soft demodulation, in the BLIC receiver the soft
demodulated bits are turbo decoded:
w = soft_demodulator(r3,Nr,Nlay,Cmod,reliability);
z = layer_demapping(w,Ncw,Nlay,M);
[soft_bits_1 soft_bits_2] =
channel_decoding(Ncw,Lp,f1,f2,n_iter,interf_power,z);
After the turbo decoder, a hard decision is done:
w_hard = zeros(size(soft_bits_2));
idx = soft_bits_2>0;
w_hard(idx)=1;
idx = soft_bits_2<0;
w_hard(idx)=0;
So, the re-generated i-th interfering transmitted symbol is turbo encoded, modulated
and SFBC encoded. Multiplying the channel matrix of the i-th interfering signal it is
possible to obtain the regenerated i-th interferer signal. For example, considering the
first step or in other word, the regeneration of the strongest interferer signal:
enc_bits_interf_2, data_bits_interf_2] =
channel_encoding_data(Ncw,Lp,rate,f1,f2,w_hard);
r3 = modulation(1,Cmod,enc_bits_interf_2);
s_3 = alamouti_enc(r3);
y3_est=Heff3*s_3;
Having regenerated the first interference signal correctly, the remaining parts are the
same, paying attention to apply the MMSE-IRC correctly as mentioned before.
In the next page is showed in detail the block diagram of the proposed BLIC receiver.
77
78
The single interferer case simulation results are showed in the figures 3 and 4. In this
case a single interferer is consider, this means that DIP1=-0.1 and DIP2=-0.001.
Considering negative SINR, BLIC and SLIC receivers shows a similar throughput, also the
Throughput of MMSE-IRC when AoA1=30 and AoA1=45 is comparable to the BLIC and
SLIC one. When SINR is about 2dB, the SLIC (AoA1=10 and AoA1=0) and BLIC AoA1=0
throughput start to decrease and starting from 3dB MMSE-IRC provides higher
throughput. From this simulations seems that SLIC receiver is the optimal choice when
the SINR is less than about 3dB, while the MMSE-IRC is optimal (in terms of throughput
and complexity) when SINR is higher than about 3dB. Regarding the Raw BER figure,
SLIC and BLIC reach a minimum that shifts varying the AoA, so increasing the SINR, SLIC
and BLIC Raw BER becomes higher than MMSE-IRC ones. The double interferer case
simulation results are shown in the figures 5 and 6, as expected when there are two
79
Figure 20: Raw BER - Single Interferer
interfering signals SLIC and BLIC provide higher gain respect to MMSE-IRC and
Alamouti receiver in the low SINR region. In this case, for negative SINR, BLIC receiver
is the optimal choice, while SLIC performances are comparable to the MMSE-IRC ones.
When SINR is higher than about 1dB, the MMSE-IRC becomes the best receiver
because, rather than SLIC and BLIC, does not suffer from the error propagation effect
in the regeneration of the interfering signals.
80
81
82
83
84
CONCLUSION
Symbol level successive interference cancellation and interference rejection combining
receivers represent a good trade-off between complexity and performance results. Bit
level successive interference cancellation is optimal in presence of two interfering
signal when SINR is low (at cell edge), but in presence of single interfering signal or for
a UE close to the base station the switching on-off procedure between SLIC and
MMSE-IRC represents the optimal choice to respect the trade-off between complexity
and performances. From the analysis and simulation tests, it follows that there is no an
optimal receiver useful for all the interferer scenarios, but surely a class of receivers
becoming to the NAICS (i.e. SLIC and BLIC that include MMSE-IRC) are very promising.
Some research activities must be done about the topic of NAICS, in particular to clarify
the type of network assistance and to find new network parameters useful to improve
the UE interference cancellation capabilities. Clearly, since UE will become more
powerful, the research must be address to receiver side, limiting the spectrum usage
foe transmitting assistance from the network. UE could be able to sense the interfering
signals to switch their IC receivers to the optimal one.
This capacity could be reached implementing in software or directly in the chipset
hardware a set of IC receivers, for example starting from: MMSE-IRC, SLIC and BLIC. As
analyzed in the chapter 4, it is possible to subdivide the SINR range in regions in which
the selection of the best receiver can be done on performance basis. Moreover,
identifying accurately (tagging) each interferer signals and tracking their received
powers, would be possible to order and filter each interferer signals in the most
suiable way.
85
BIBLIOGRAPHY
1. Complexity and link level performance analysis for feICIC CRS-IC receiver. Ericsson.
s.l. : 3GPP TSG-RAN WG4 Meeting # 64, 2012. R4-124460.
2. Discussion on the reference receiver for FeICIC. Huawei. s.l. : 3GPP TSG-RAN WG4
Meeting #63 , 2012. R4-122479.
3. FeICIC baseline receiver assumptions. Qualcomm. s.l. : RAN4 #62bis, 2012. R4122185.
4. Link level simulations for FeICIC with 9dB cell range expansion. Qualcomm. s.l. :
3GPP TSG-RAN WG4 #63, 2012. R4-123313.
5. CRS Interference Mitigation For Homogeneous Deployments. Ericsson, ST-Ericsson,
NEC, MediaTek, Sony Mobile, Verizon, Orange, Softbank, Alcatel-Lucent, LG
Electronics, Renesas. s.l. : 3GPP TSG-RAN Meeting # 59. RP-130393.
6. WF on CRS-IM performance evaluation. Ericsson. s.l. : 3GPP TSG-RAN WG4 #66bis.
R4-132020.
7. Investigation on Advanced Receiver Employing Interference Rejection Combining.
s.l. : Yusuke Ohwatari, Vol. NTT DOCOMO.
8. 4G LTE/LTE-A for Mobile broadband. s.l. : AP.
9. Analsysis of LTE physical layer and its evolution . s.l. : Telecom Italia Lab, 2013.
10. Reconfigurable OFDMA Simulation Platform - Inner Modem V 1.0. s.l. : Telecom
Italia Lab, 2007.
86