This action might not be possible to undo. Are you sure you want to continue?
Abstract— This paper explains a new technique for a single
user data transmission called interleavedivision multiplexing
(IDM). The information of the user is distributed into N
parallel data streams, which are multiplexed at the transmitter
and demultiplexed at the receiver. This technique is closely
related to interleavedivision multiple access (IDMA). In IDM
the interleavers are the only mean of layer separation. IDM
benefits from robustness in transmission and efficient iterative
detection. At the transmitter unequal power allocation can be
used for each layer. The complexity per layer at the receiver
remains constant with increasing the number of layers.
I. INTRODUCTION
CDMA is a popular multiple access technique which
provides the users a shared medium for transmission. It is
applied in the second and third generation mobile standards
with a huge success. Some of its benefits are mitigation of
cross cell interference and robustness against fading. In this
technique the only mean to distinguish users is using unique
spreading codes for each one of them. However this
redundancy leads to spreading the spectrum without coding
gain. Another disadvantage of CDMA is that with increasing
the number of users, the multiuser detection (MUD) at the
receiver side will be much more complex [1]. The increasing
number of users also leads to degradation of performance
(high BER) due to multiple access interference (MAI) [2].
In order to avoid the disadvantages mentioned in the
previous subpart, another technique, called IDMA, was
proposed. IDMA is a special case of CDMA where the only
mean of separation between users is the unique interleaver
assigned to each one. Since the coded bits are only
commutated there is no irrelevant spreading of the spectrum.
Moreover the complexity of the receiver is linear with
respect to the number of users, i.e. it is constant per user [1].
IDM is a special case of IDMA. In the system described
in the last subpart, we can use multiple number of data
streams which do not represent different users but different
applications, related to the same user. The structure of IDM
is the same as those of IDMA and it benefits most of its
advantages. Each single stream can be used for different
application like e.g. speech, video or transferring data files.
This article consists of six sections. The next section
consists of the transmitter structure and the third one
explains the iterative receiver structure and the working
principle of the receiver on chipbychip (CBC) detection
used in IDM to overcome MAI. The fourth section describes
implementation of the IDM receiver in JAVA following the
advantages and disadvantages of this scheme. Section five
includes the most important advantages and drawbacks of
IDM. The last
Fig. 1 Transmitter structure
section briefly concludes the whole article by giving some of
the potential functionalities of IDM.
II. TRANSMITTER
The structure of the transmitter containing N layers is
shown in Fig.1. The input data stream of the user is
distributed into N parallel data streams denoted by
i
d ,
where ( ) N i , 1 ∈ [3]. The input data sequence
i
d is encoded
by the encoder, which can be unique depending on the
requirements of the application, using a lowrate forward
error correction code (FEC) [4]. For instance a repetition
code or the more sophisticated turbo code can be used. The
encoded symbols are called chips [3].
The encoded data sequences are then scrambled in order
to avoid long runs of zeros and ones.
The scrambled data in each layer is then permuted frame
by frame using layerspecific interleaver
i
π [3]. The
resulting output from the interleaver is the
vector ( ) ( ) [ ]
T
i i i
J x x x ,..., 1 = , where J is the length of the
frame and
i
x is the chip. After interleaving, the successive
chips of each layer are uncorrelated which is required for the
single CBC detection at the receiver side [5].
Then each chip is multiplied by a layer specific
coefficient
i
j
i i
e a h
φ
= , where
i
a is an amplitude and
i
φ is a
phase. After summation of the layers the transmitted signal is
expressed by equation [6]:
( ) ( ) ( ) k x e a k x h k x
i
j
N
i
i i i
i
φ
∑
=
= =
1
(1)
In case of equal
i
a coefficients, an equal error protection
is provided to all layers. Depending on the application and
the importance of the layer, unequal power allocation could
be used hence providing unequal error protection. Usually
InterleaveDivision Multiplexing
Najeeb ul Hassan, Setareh Maghsudi, Shashishekar Ramakrishna, Tihomir Tsanev
naj.hassan@gmail.com, settareh.maghsudi@gmail.com, shashi792@gmail.com, tihomir_tsanev@abv.bg
2
the phases
i
φ are uniformly distributed giving us a
symmetrical constellation. The idea behind this modulation
is that using more layers, higher modulation schemes are
achievable, therefore higher data transmission rates are
achievable.
III. RECEIVER
A. Structure of the iterative receiver
Fig. 2 shows the blocks of the iterative receiver structure.
The formulations in this chapter are made with the
assumption of additive white Gaussian noise (AWGN)
channel. Thus, the received signal samples from N layers are
expressed as [6]:
( ) ) ( ) ( ) ( ) (
1
k n k x h k n k x k r
N
i
i i
+ = + =
∑
=
, (2)
where ) (k n is a noise sample of the AWGN process with
variance 2 /
0
2
N
n
= σ .
Fig. 2 Iterative receiver structure
The turbotype iterative process starts in the multilayer
detector (MLD) which computes the extrinsic¹ log
likelihood ratios (LLR) )) ( ( k x e
i MLD
for each chip. These
soft values are then given to the respective decoder as a
priori information after deinterleaving. The updated and
deinterleaved information from the decoders )) ( ( k x e
i DEC
is
returned to the MLD through a feedback. Using this a priori
information, the detector can improve its estimates for the
next iteration.
B. Loglikelihood ratio
The formula for the LLR in case of BPSK (for simplicity)
is given by [4]:
( ) ( )
( ) ( )
( ) ( )


¹

\

− =
+ =
≡
1
1
log
k x y p
k x y p
k x e
i
i
i
k i, ∀ (3)
For the case of )) ( ( k x e
i MLD
y is represented by the
received samples ) (k r and for )) ( ( k x e
i DEC
the deinterleaved
information from the decoders.
After a preset number of iterations the decoders make
hard decision for ) (k x
i
, based on the value of { } )) ( ( sgn k x e
i
.
¹The extrinsic LLR value for each chip is computed on the basis of the
LLR values of the chips from the other layers at the same time k.
The reliability of the decision is given by the
magnitude )) ( ( k x e
i
,
¹
´
¦
→
→ ∞ +
=
unreliable
certain
k x e
i
0
)) ( ( (4)
C. Detection
On account of using randomly generated interleavers
i
π ,
the detection can be done chip by chip. From (2) it is easy to
express the received noise sample for layer i at time k [5]:
∑
≠ =
+ = − =
N
i j j
j j i i i
k n k x h k x h k r k
, 1
) ( ) ( ) ( ) ( ) ( ζ (5)
From (5) it is obvious that for layer i the overall noise
sample contains the AWGN sample from the channel plus
the samples from all other layers at time k . Therefore (2)
can be rewritten as:
) ( ) ( ) ( k k x h k r
i i i
ζ + = (6)
From the central limit theorem it can be concluded that
) (k
i
ζ can be approximated as a Gaussian variable in
presence of a large number of layers. From that point the
received sample ) (k r is described by a Gaussian conditional
probability density function [5]:
( ) ( )


¹

\

+ ± −
= ± =
2
2
2
2
) (
exp
2
1
) 1 ) ( ) ( (
i
i
i
i
i
h k r
k x k r p
ζ
ζ
ζ
σ
µ
πσ
(7)
In (7)
2
i
ζ
σ and
i
ζ
µ denote the variance and the mean of
the noise samples. From here on we will omit the time index
k for simplicity.
From (5) the mean and variance of the interference are
expressed as follows:
∑
≠ =
=
N
i j j
x j
j i
h
, 1
µ µ
ζ
(8)
keeping in mind that the AWGN is zero mean.
2 2
, 1
2
2
n x
N
i j j
j
j i
h σ σ σ
ζ
+ =
∑
≠ =
(9)
On the basis of (3) and (7) the LLR of MLD is expressed
by the formula:
{ }
2
2
2
2
2
/ ) ( Re 4
) (
i i
i
i
i
i
y h
h r h r
x e
i
i i
i MLD
ζ ζ
ζ
ζ
ζ
ζ
σ µ
σ
µ
σ
µ
− =
− +
+
− −
− =
∗
(10)
In (10) the amplitude and phase coefficients
i
h are set at
the transmitter and the received samples are measurable.
3
(a) Constellation diagram at transmitter
(c) Constellation diagram at receiver after 2 iterations (1111)
(b) Constellation diagram at receiver after 1 iteration (1111)
(d) Constellation diagram at receiver after 3 iterations (1111)
Fig.3 Constellation diagrams at transmitter and receiver
In order to find the unknown terms
i
ζ
µ and
2
i
ζ
σ , the
following formulas should be considered [7]:
1
1
) (
) (
+
−
=
j DEC
j DEC
j x e
x e
x
e
e
µ (11)
2 2
1
j j
x x
µ σ − = (12)
The intermediate values of (8), (9), (11) and (12) are
calculated after each iteration in order to update the LLR
value of the detector.
An important note here is that before the first iteration the
mean values and the variances of all transmitted chips are
initialized to zero and one respectively, if there is no a priori
information [1].
The next point is to calculate the ) (
j DEC
x e values. For a
repetition decoder it can be calculated as [7]:
∑
− +
≠ =
=
1
,
) ( ) (
SF l
i j l j
j MLD i DEC
x e x e , (13)
where l is the position of the first LLR related to the
original bit and
¸ ¸
SF i SF l / = (14)
SF is the spreading factor [3].
The explanation of (13) is that at the decoder all LLR
values corresponding to one bit are gathered. Then at its
output the LLR corresponding to the given chip is the
summation of all LLRMLD except the one being
calculated. In such a way the next correction is not based on
the same estimate that was calculated before.
IV. IMPLEMENTATION
The transmitter and receiver of IDM are implemented in
JAVA IDE 6.7. The constellation diagram at the output of
the transmitter can be plotted for different possible number
of layers. Fig. 3a shows a constellation diagram for four
layers with equal power allocation and uniform phase
distribution N i
i
π φ × = where i is the layer number and N
is the total number of layers. The transmitted signal is
distorted with AWGN noise and then fed to the receiver in
Fig. 2. The MLD calculates the mean and variance of the
interference signal for each chip using (8) and (9)
4
respectively, as well as the LLR values using (10) and passes
this information to the respective decoders of each layer.
Each decoder then calculates the LLR values using (13) and
feeds this information back to the MLD.
The experiment is performed for four layers with a
spreading factor of 4 and interleaver length of 20. In (11)
j
x
µ gives the mean value for the particular chip after each
iteration. The Fig. 3b3d shows the expected value of the
symbol after each iteration which is computed by adding the
mean interference to i
th
chip calculated from (8) and the
expected value of the i
th
chip (11),
∑
=
= + =
N
j
x j j x s
j i i
h h
1
µ µ µ µ
ζ
The results show that the expected value becomes closer
to the actual symbol transmitted after each iteration. Every
second layer’s data is represented as a dashed blue line.
V. FEATURES OF IDM
IDM possesses a number of advantages that make it an
attractive modulation technique. As it was stated earlier, the
power applied in each layer can be adjusted, depending on
the application. For instance a layer dedicated to voice
transmission can be given less power compared to a layer
that serves file transmission. The reason is that no errors in
transmitting a file are wanted, whereas in speech
transmission this condition is not so crucial.
Another important application of unequal power
allocation is in case of extreme environment, e. g. an isolated
room. In such occasion the number of layers is reduced and
whole amount of power can be concentrated into a single
layer, e.g. those devoted to speech, so that at least this
service would be provided.
Moreover in the fast fading environment, one layer can be
assigned for training. Thus the channel estimation would be
more reliable even in ultra fast fading channels. This also
does not decrease the bandwidth efficiency of the system.
Another key advantage of IDM is its robustness against
deep fades. The spreading factor and the used interleaver
spread one bit into many chips at different locations of the
transmitted signal. Thus it is very unlikely that all the chips
corresponding to one bit will simultaneously be affected by a
deep fade. Furthermore the spreading factor used in encoder
does not affect the bandwidth efficiency as the number of
layers can be increased to achieve the same bandwidth and
power efficiency.
On the other side, due to the superposition of the signals
from each layer at the output of the transmitter, high peak to
average power ratio (PARP) is a major drawback in IDM
systems. Furthermore the output can only be obtained after
certain iterations which make the processing slow at the
receiver.
VI. CONCLUSION
The article outlines briefly the principles of IDM with the
focus on chipbychip decoding algorithm. IDM provides
many desirable features such as a low cost chip by chip
iterative detection algorithm. The data rate can be adapted
according to the channel conditions without using the higher
modulation formats and just by increasing the number of
layers. It also inherits most of the advantages from CDMA
and IDMA. The efficiency of IDM systems can be further
improved by using powerful low rate FEC codes. The
theoretical basis of IDM shows good possibilities for its
future applications in real communication systems.
ACKNOWLEDGMENT
We would like to thank our supervisor Meelis Nõmm for
his assistance in explaining the subject matter and giving us
comments during this project.
REFERENCES
[1] Li Ping “Interleavedivision multiple access and chipbychip
iterative multiuser detection,” IEEE Commun. Magazine, Jun. 2005,
vol. 43, no.6, pp S19S23.
[2] A. DuelHallen, J. Holtzman, Z. Zvonar, ’Multiuser detection for
CDMA systems,’ in Personal Communications, IEEE [see also IEEE
Wireless Communications] vol.2, Issue 2, 1995, pp 46 – 58
[3] D. Hao, P. Yao and P. A. Hoeher, “Analysis and design of interleaver
sets for interleavedivision multiplexing and related techniques,” in
Proc. 5
th
Int. Symp. on Turbo Codes and Related Topics, Lausanne,
Switzerland, Sep. 2008.
[4] Li Ping, Lihai Liu, K. Y. Wu, and W. K. Leung “On interleave
division MultipleAccess,” in Proc. IEEE ICC, Paris, Jun. 2004, vol.
5, pp. 286973.
[5] Li Ping, L. Liu, K. Wu, W.K Leung, “ Interleave division multiple
access,“ in Wireless Communications, IEEE Transactions on, vol.
5, Issue 4, 2006 pp 938 – 947
[6] P. Wang, L. Ping and L. Liu, “Power allocation for multiple access
systems with practical coding and iterative multiuser detection,”
IEEE Trans. Commun., vol. 11, pp.49714976, Jun. 2006.
[7] T. Wo and P. A. Hoeher, “A simple iterative Gaussian detector for
severely delayspread MIMO channels,” in Proc. IEEE Int. Conf.
Commun. (ICC ‘07), Glasgow, Scotland, Jun. 2007.
j ≠ i 2 N xj (8) keeping in mind that the AWGN is zero mean. the received signal samples from N layers are expressed as [6]: The reliability of the decision is given by the magnitude e( xi (k )) . RECEIVER A. { ∗ } 2 . From that point the received sample r (k ) is described by a Gaussian conditional probability density function [5]: r (k ) − ± hi + µζ 2 1 i (7) p(r (k ) xi (k ) = ±1) = exp 2 2 2σ ζ i 2πσ ζ i ( ( )) 2 In (7) σ ζ i and µζ i denote the variance and the mean of Fig. The formulations in this chapter are made with the assumption of additive white Gaussian noise (AWGN) channel. From (5) it is obvious that for layer i the overall noise sample contains the AWGN sample from the channel plus the samples from all other layers at time k . From (5) the mean and variance of the interference are expressed as follows: µζ i = ∑h µ j j =1. k (3) eMLD( xi ) = − r − hi − µζ i 2 σζ2 + r + hi − µζ i 2 i σζ2 i (10) For the case of eMLD ( xi (k )) y is represented by the received samples r (k ) and for eDEC ( xi (k )) the deinterleaved information from the decoders. From (2) it is easy to express the received noise sample for layer i at time k [5]: ζ i (k ) = r (k ) − hi xi (k ) = ∑h x j j =1. the detector can improve its estimates for the next iteration.the phases φi are uniformly distributed giving us a symmetrical constellation. Loglikelihood ratio The formula for the LLR in case of BPSK (for simplicity) is given by [4]: the noise samples. These soft values are then given to the respective decoder as a priori information after deinterleaving. j ≠i N j (k ) + n( k ) (5) r (k ) = x(k ) + n(k ) = ∑ h x ( k ) + n( k ) . After a preset number of iterations the decoders make hard decision for xi (k ) . + ∞ → certain e( xi (k )) = 0 → unreliable C. Structure of the iterative receiver Fig. B. Therefore (2) can be rewritten as: r (k ) = hi xi (k ) + ζ i (k ) (6) From the central limit theorem it can be concluded that ζ i (k ) can be approximated as a Gaussian variable in presence of a large number of layers. ¹The extrinsic LLR value for each chip is computed on the basis of the LLR values of the chips from the other layers at the same time k. higher modulation schemes are achievable. i i i =1 N (2) where n(k ) is a noise sample of the AWGN process with 2 variance σ n = N 0 / 2 . the detection can be done chip by chip. therefore higher data transmission rates are achievable. based on the value of sgn{e( xi (k ))} . 2 Iterative receiver structure The turbotype iterative process starts in the multilayer detector (MLD) which computes the extrinsic¹ loglikelihood ratios (LLR) eMLD ( xi (k )) for each chip. 2 σ ζi = ∑h j =1. III. Detection (4) On account of using randomly generated interleavers π i . From here on we will omit the time index k for simplicity. The updated and deinterleaved information from the decoders eDEC ( xi (k )) is returned to the MLD through a feedback. j ≠ i N j 2 2 σ xj + σ n (9) On the basis of (3) and (7) the LLR of MLD is expressed by the formula: p( y xi (k ) = +1) e(xi (k )) ≡ log p( y x (k ) = −1) i ∀i. The idea behind this modulation is that using more layers. 2 shows the blocks of the iterative receiver structure. Thus. Using this a priori information. = 4 Re hi ( y − µζ i ) / σζ2i In (10) the amplitude and phase coefficients hi are set at the transmitter and the received samples are measurable.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.