Professional Documents
Culture Documents
x(t)
2
_
dB (3.1)
where x(t)
dB (3.3)
1
In the present work we will always refer to Release 99 (R99) of the standard, unless dierently
stated.
40
Figure 3.1: Rapps model AM/AM characteristic as obtained implementing equa-
tion 3.2 with a Matlab c function.
OBO = 10log
10
A
2
0
P
OUT
dB (3.4)
In this work we will always use the denition in 3.3.
WCDMA standard imposes tight limits on both OOB emissions and signal qual-
ity [33]; therefore solutions have to be found to improve amplier eciency within
the limits of the regulation. This is the reason why many eorts have been done
in studying PAPR reduction techniques [30], which should guarantee PA linearity
without resorting to large BO values. In this work, we focus our attention on a par-
ticular category of PAPR reduction strategies: signal distortion techniques, which
reduce the peak amplitudes by distorting the signal at or around peaks. Amongst
the available options for such category, we applied three dierent techniques to a
WCDMA transmitter and checked their eectiveness in terms of the parameters im-
posed by the standard. Simulation results obtained in each case have been compared
in order to nd out which strategy is more suitable for implementation.
41
3.2 WCDMA uplink air interface: Physical chan-
nels, Spreading and Modulation
The 3GPP uplink modulation scheme [4] allows up to six Dedicated Physical Data
Channels (DPDCHs) and one Dedicated Physical Control Channel (DPCCH). The
uplink DPDCH transports dedicated data generated at OSI Layer 2 and above,
i.e. the dedicated transport channel (DCH)
2
, whilst the uplink DPCCH is used to
carry control information, such as pilot bits to support channel estimation, trans-
mit power-control (TPC) commands, feedback information (FBI) and an optional
transport-format combination indicator (TFCI) [3]. Physical channels are struc-
tured in frames of 10 ms duration, each frame is split into 15 slots of xed length
(equal to 2560 chips). A super-frame corresponds to 72 consecutive frames, i.e. its
length is 720 ms. The frame structure for the uplink dedicated physical channel is
shown in Figure 3.2, where the parameter k indicates the number of bits per slot
and it is directly related to the SF of the channel as SF = 256/2
k
.
The channels conguration is shown in Figure 3.3. WCDMA uses a two level code
Figure 3.2: Frame structure for uplink DPDCH/DPCCH [3].
system: orthogonal spreading codes (c
d,n
) and pseudo random scrambling codes
(S
dpch,n
). In order to support variable data rates, the air interface allows to pick
2
There might be more than one DPDCH per active connection but always only one DPCCH.
42
Figure 3.3: DPDCHs and DPCH uplink conguration [4].
up spreading codes with dierent Spreading Factor (SF) and this family of codes
is called Orthogonal Variable Spreading Factor (OVSF). In the uplink OVSF codes
are used to separate data and control channels from a specic user. After spread-
ing control and data channels are I/Q multiplexed. Scrambling codes identify a
specic mobile terminal and it is applied on top of spreading. A detailed descrip-
tion of WCDMA uplink physical layer is beyond the scope of the this work, further
details can be found in [26] [34] other than in the related 3GPP specications [3] [4].
For uplink data rate below 450 kbps, a single DPDCH is always used, in which
case the in-phase component is made of the data channel and the in-quadrature
component is obtained from the control channel (DPCCH). The data rate of the
DPDCH can be variable and dierent values of SF can be used during the trans-
mission, whereas the control channel rate is xed and it uses a spreading code with
SF= 256. Table 3.4 contains the bit and symbol rates available (before spreading)
for a data channel according to the SF used. When employing multiple data chan-
nels the SF is set to 4 for each DPDCH, which corresponds to highest available bit
rate (around 2 Mbps), whilst the SF for the control channel remains unchanged at
43
Figure 3.4: Uplink DPDCH available bit/symbol rates along with correspondent
SF [3]. The number of bits per DPDCH is determined by the parameter k (slot
format) according to the following equation: N
bits
= 10 2
k
, k = 0, . . . , 6.
256. It is worth to point out that only three codes are available at the SF of four,
consequently the codes are reused on the in-phase and quadrature branches.
c
and
d
are adjustable weighting parameters which help to maintain constant the
signal-to-noise ratio of both data and control channels
3
.
The combined data and control channels are multiplied by a complex scrambling
code (S
dpch,n
), which has been designed to avoid 180 degrees transitions from one
chip to another, hence leading to a reduction in PAPR. This modulation scheme
takes the name of Hybrid Phase Shift Keying (HPSK) and it relies on a proper com-
bination of spreading and scrambling codes, a comprehensive explanation of HPSK
can be found in [35]. The PAPR for a single DPDCH is lower than for standard
QPSK thanks to HPSK modulation scheme (see section 3.3).
The bandwidth of the modulated signal is dened by ltering the I and Q com-
ponents with Root Raised Cosine (RRC) lter with a roll-o factor of 0.22 and a
bandwidth of 1.92 MHz (half the chip rate). The ltered signal trajectory for a
single DPDCH with
c
=
d
= 1 is depicted in Figure 3.5 (a): it is worth to notice
the low number of 180 degrees phase shifts hence proving the eectiveness of HPSK
modulation. As a comparison, in Figure 3.5 (b) the vector diagram obtained from a
wrong set of spreading codes is shown: the number of zero crossings has noticeably
increased.
3
Throughout this work both these values are always set to 1.
44
(a) (b)
Figure 3.5: Signal trajectory for an RRC ltered single data channel when using
correct HPSK (a) and employing a wrong choice of spreading codes (b).
3.3 Distribution of PAPR and its eect on PA
In Figure 3.6 the CDFs for the PAPR of single data channel and multiple data
channels WCDMA signals are shown. With regard to the 1 DPDCH case, it is easy
to notice that HPSK modulation eectively keeps the PAPR within 3.5 dB, which is
a remarkably low value. On the other hand, when employing six data channels, the
PAPR rises up to a maximum value of around 8 dB. Due to ampliers nonlinear-
ities, the eects of a higher PAPR are the increased level of out-of-band emissions
and decreased performance because of bad modulation accuracy (or in-band dis-
tortion) [36]. As mentioned in section 3.1, power ampliers have to operate near
their saturation point to reach high power eciency, unfortunately this region of
the AM/AM characteristic is non-linear, therefore -if the signals PAPR increases-
the PA is forced to work in its saturation region and this causes the negative eects
described above. As an example, Figure 3.7 illustrates the spectrum of WCDMA
signal at the output of the PA when 1 and 6 data channels are employed: the
level of OOB radiation is clearly higher in the latter case than in the former. A
solution might be setting the amplier operation point back to the linear region
by introducing a back-o to the input, but this would result in a lower eciency.
Alternatively, PA with a wider linear characteristic could be used to overcome the
unwanted eects, unfortunately very linear ampliers are both very expensive and
scarcely ecient. Furthermore, WCDMA standard sets strict limits on signals qual-
ity (i.e. allowed level of in-band distortion) and adjacent channel power leakage (i.e.
OOB emissions), in the next section ( 3.4) such requirements will be discussed in
45
3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 9
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PAPR [dB]
PAPR CDFs
Figure 3.6: Cumulative distribution functions of PAPR for a ltered WCDMA signal
with a single data channel (blue line) and 6 data channels (red line).
(a) Spectrum at the output of the PA. 1
DPDCH.
(b) Spectrum at the output of the PA. 6
DPDCH.
Figure 3.7: Eects of an increased PAPR due to PA nonlinearities.
detail.
In this scenario clearly appears the necessity of nding a trade-o between all these
clashing requisites. Summarising, we can conclude that the challenging goal ahead
consists in maximising the PA eciency by reducing signals PAPR without unduly
compromising the performance and complying with the restrictions imposed by the
46
standard [37].
It is now evident that PAPR reduction techniques represent essential features to
achieve the objective because they allow to reduce the PAP ratio and consequently
to reach a higher eciency. The main drawback of this approach is given by the
distortions introduced on the signal, which is going to inevitably have a lower qual-
ity, hence worse performances (e.g. BER). However had such techniques not been
used a large back-o value would be required thus drastically aecting the system
performance.
In section 3.5 we will go through the details of clipping schemes we applied to limit
the PAPR.
3.4 Radio Specics for WCDMA transmitter
3GPP standard for WCDMA species the following signal-quality requirements for
the waveform at the output of a 3G transmitter [33]:
Adjacent Channel Leakage Ratio (ACLR) which determines how much of
transmitted power can leak into the rst (ACLR1) or second (ACLR2) neigh-
bouring carrier. ACLR is dened as the ratio of the RRC ltered mean power
centered on the assigned channel frequency to the RRC ltered mean power
centered on the adjacent channel frequency. ACLR1 should be larger than
33 dB and ACLR2 than 43 dB
4
.
Error Vector Magnitude (EVM) which is a measure of the dierence between
the reference waveform and the measured waveform. This dierence is called
the error vector. The EVM result is dened as the root of the ratio of the
mean vector power to the mean reference power expressed in %. EVM value
should be below 17.5 %.
In fact, EVM is a measure of the signals modulation quality, it helps to imme-
diately highlight possible causes of degradation, such as signal compression,
local oscillators timing errors, I/Q impairment, etc. In digital modulation
the signals amplitude and phase can be measured at any time. Such values
dene the actual or measured phasor. Similarly, a corresponding ideal or
reference phasor can be calculated, given the knowledge of data transmit-
ted, clock timing and so on. The dierences between these phasors form the
basis for EVM measurement [5]. A graphical explanation of this concept is
in Figure 3.12, actually EVM is the scalar distance between the two phasors
(i.e. the magnitude of the dierence vector) and, by convention, is usually re-
ported as percentage of the peak signal level. In other words, EVM measures
4
The nominal channel spacing for WCDMA is 5 MHz, therefore the rst adjacent channel is at
5 MHz and the second at 10 MHz.
47
the dierence between the transmitted signal and the ideally-modulated one.
Figure 3.8: Error Vector Magnitude (EVM) and related quantities [5].
Peak Code Domain Error (PCDE) which is computed by projecting the power
of the error vector (as dened for EVM) onto the code domain at a specic
spreading factor. The Code Domain Error for every code in the domain is
dened as the ratio of the mean power of the projection onto that code, to
the mean power of the composite reference waveform. This ratio is expressed
in dB. The Peak Code Domain Error is dened as the maximum value for the
Code Domain Error for all codes. PCDE value shall not exceed 15 dB for
spreading factor 4.
This gure of merit of the signals quality has been specically introduced
in WCDMA standard to address the possibility of uneven error distribution.
This test is required only for multicode transmission.
In the present work we considered ACLR1, EVM and PCDE as limiting factors
that considerably aect the design of power ampliers for UMTS terminals [38] [39].
ACLR2 has not been included in the analysis because the allowed level of emissions
is, in any case, below the threshold. In Figure 3.9 (a) the ACLR1 value for an
unclipped 6 data channels signal is plotted as a function of the IBO of the amplier,
the ACLR1 limit is met for back-o values greater than 5 dB only, whereas the
ACLR2 value is above the threshold of 43 dB regardless of the IBO applied to the
PA (Figure 3.9 (b)). In the following of this work we will benchmark the performance
of dierent PAPR reduction techniques which guarantee a level of spectral splatter
above the minimum limit imposed by ACLR1.
48
0 1 2 3 4 5 6 7
20
22
24
26
28
30
32
34
36
38
IBO [dB]
A
C
L
R
1
[
d
B
]
ACLR1 6 DPDCHs
ACLR1 6 DPDCHs
ACLR1 limit 33 dB
(a)
0 1 2 3 4 5 6 7
42
44
46
48
50
52
54
56
58
60
62
IBO [dB]
A
C
L
R
2
[
d
B
]
ACLR2 6 DPDCHs
ACLR2 6 DPDCHs
ACLR2 limit 43 dB
(b)
Figure 3.9: Power leakage eect on adjacent channels of a 6 DPDCHs signal through
a nonlinear amplier: ACLR1 vs IBO (a). ACLR2 vs IBO (b).
3.5 PAPR reduction techniques
Several approaches have been proposed to overcome the problems posed by signals
with high PAPR (a comprehensive summary can be found in [40]). Among the
solutions devised, we decided to chose three techniques that belong to the category
named signal distortion techniques. As the name suggests, these schemes aim at
reducing PAPR by directly modifying the modulated signal waveform. The reason
for this choice essentially lays in the simplicity of the approach proposed, which ts
well with the standard boundaries and the requirements of a future implementation.
The following sections include a detailed description of the techniques we applied
to the model of a WCDMA mobile terminal..
Clipping and Filtering
The simplest way to reduce the PAPR is clipping the signal, such that peak ampli-
tude becomes limited to a desired level, A. We dene the Clipping Ratio (CR) as
the clip level A over the rms of the signal:
CR = 20 log
10
_
A
_
dB (3.5)
The main drawback related to this approach is the introduction of in-band and OOB
interference, as clipping is a non-linear operation it causes distortion by giving birth
49
to spurious harmonics (exactly the same eect as real amplier). Self interference
(or in-band distortion) degrades the signal quality which has to stay within the
allowed values of EVM and PCDE; on the other hand, OOB emissions must be kept
within the limits imposed by the ACLR. Filtering after the clipping stage reduces
the spectral splatter, but causes peak regrowth [41]. Despite these side-eects,
this simple approach proved to be eective in reducing the amplitude signal range
and suppressing unwanted OOB emissions. Obviously, the performance will mainly
depend on the characteristics of the lter used. In our case, the lter employed in
this work is a simple equiripple FIR lter with stopband attenuation around 40 dB.
Peak Cancellation
As pointed out in the previous section the main disadvantage of clipping is repre-
sented by OOB radiations due to the sharp corners present in the clipped version
of the signal. This undesirable eect can be avoided by doing a linear peak cancel-
lation: a time-shifted and scaled reference function is subtracted from the original
signal in order to reduce the peak power. This technique could not cause any
spectral splatter if a proper function (with the same bandwidth as the transmitted
signal) is selected. One example of suitable function is the sinc function, unfortu-
nately it is of no practical use as it has an innite support. Hence, for practical use,
time-limited functions with adequate spectral properties are created.
Figure 3.10 (a) shows part of the truncated sinc function which we used as correction
function (other types of reference functions can be found in [42] [43]). The spectral
properties of such function assure low OOB emissions as its attenuation outside the
signals bandwidth is around 70 dB (Figure 3.10 (b)), conversely no attempts are
made to reduce the in-band distortion.
The peak cancellation scheme suggested in [43] employs a correction function which
suppress both in-band and out-of-band clipping noise. Due to the strict require-
ments posed by the lters mask specics in that particular case, the outcome is a
correction function with not enough
5
OOB suppression capability, that is the main
reason why we preferred to let down in-band attenuation and focus on out-of-band
one.
Peak Windowing
Like clipping and ltering and peak cancellation, peak windowing is a signal distor-
tion technique (sec. 3.1). We have already pointed out that conventional clipping
causes sharps corners in a clipped signal. This leads to an increased spectral splatter
and to a higher ACLR. Peak windowing is meant to reduce the level of unwanted
emissions by smoothing the sharp corners related to the hard clipping process. The
5
According to the 3GPP specics [33].
50
500 550 600 650 700 750
0.2
0
0.2
0.4
0.6
0.8
1
samples
a
m
p
l
i
t
u
d
e
Reference Sinc function for Peak Cancellation
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
70
60
50
40
30
20
10
0
10
20
Normalized Frequency ( rad/sample)
M
a
g
n
itu
d
e
(
d
B
)
Magnitude Response (dB)
(b)
Figure 3.10: Truncated sinc function used in the peak cancellation algorithm (a)
and its spectral properties (b).
goal is achieved multiplying the signal with a window function. The spectral prop-
erties of this function determine the amount of out-band emissions. Conventional
clipping can be expressed as a multiplication [37]:
x
clip
(n) = c(n)x(n) (3.6)
where
c(n) =
_
1, |x(n)|A
A
|x(n)|
, |x(n)|>A
(3.7)
where A is the clipping level. The idea of this method is to replace the function
c(n) with the following function:
b(n) = 1
+
k=
a
k
w(n k) (3.8)
w(n) is the window function and a
k
is a weighting coecient. To achieve the
desired clipping level the funciton b(n) must satisfy the inequality
1
+
k=
a
k
w(n k) c(n) (3.9)
for all n. The dierence between c(n) and b(n) depends on the window length and
the coecients a
k
. Once the window is chosen, the weighting coecients a
k
must
51
be optimised. Usually it is assumed that the window length is so small (according
to the clipping probability) that the windows do not overlap and the easiest way
to calculate the optimal a
k
is convolving 1 c(n) with the window function w(n),
yielding to the following expression for b(n):
b(n) = 1
+
k=
[1 c(k)]w(n k) (3.10)
Unfortunately in a real case windows overlap and, as a result, the signal is clipped
more than needed (overclipping). Many solutions have been proposed to avoid this
problem [44] [37].
In this work we used an approach similar to [44] to overcome overclipping. The
window employed is a Hann with 49 samples. This number of samples sets a good
compromise between the overclipping problem and spectral performance. In Fig-
ure 3.11 the window as used in our simulations is depicted.
0 10 20 30 40 50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
samples
a
m
p
l
i
t
u
d
e
Hann Window for Peak Windowing
Figure 3.11: Hann window with 49 samples. This is the function applied to our
peak windowing scheme.
3.6 WCDMA Uplink Transmitter Simulator
The performances of the clipping techniques described in the above sections have
been evaluated through a Simulink c model of a standard-compliant WCDMA
52
Figure 3.12: WCDMA-compliant transmitter Simulink c model developed at PRL.
transmitter.
The block scheme of the implemented simulator is shown in Figure 3.12. The model
comprises a standard transmitter and a simplied WCDMA receiver. The trans-
mitters modulation and spreading strategy completely fulls the specics in [3].
3.6.1 WCDMA uplink transmitter
Figure 3.13 shows the constituent elements of the models Data generation and
spreading block (see Figure 3.12). Data are generated according to the frame and
slot structure specied by the standard and summarised in section 3.2. Each frame
has a length of 10 ms and it is divided into 15 time slots. The number of chips
per slot is set to 2560 regardless of the spreading factor, whilst the number of bits
per slot depends on the SF employed. The model we devised can support each of
the slot format allowed by the standard specics (see Figure 3.4), although when
using multicode transmission the only SF value available is 4, which yields to 640
bits per slot of a data channel (DPDCH). As already mentioned in section 3.2, the
uplink control channel (DPCCH) has a xed rate and uses a constant spreading
factor set to 256, which corresponds to 10 bits per slot. On each channel, bits are
generated independently according to uniform random distributions with dierent
initial statistical seeds.
53
After this stage, data are passed on to the spreading block in which data and
control bits are multiplied by the OVSF codes, that are allocated according to the
rule dened in [3]. At the output of the spreading block, each time slot is a sequence
of 2560 chips. The model then provides I/Q multiplexing of the physical channels
following the scheme in Figure 3.3. At this stage the signal must be multiplied by
Figure 3.13: Constituent elements of the WCDMA models block that provide data
generation and spreading operation as specied in [3].
the long scrambling code: the result of such operation is a signal with a noise-like
spectrum. It is also worth to remind that scrambling codes identify specic users
in a network, hence they have to be uniquely assigned to each mobile terminal that
access the network. This means that the uplink scrambling sequence has to be
long enough to provide a huge number of dierent codes, in WCDMA standard a
length of 38400 chip has been chosen (corresponding to one frame, i.e. 10 ms). The
generation of such code is quite complicated and a comprehensive analysis would go
beyond the scope of the work, nonetheless the standard indicates how generating
the long scrambling code and the relative details are available in 3.2. Figure 3.15
shows the Simulink c implementation of the scrambling sequence generator used in
our model. After the scrambling stage, the signal is fed into the chip shaping block.
The Chip Shaping block provides in-phase and in-quadrature modulation with an
interpolation factor I = 6, clipping is performed after this stage for a more eective
peak capturing [30]. The RRC lter roll-o factor has been set to 0.22, which
corresponds to a signal bandwidth of 4.68 MHz, given the chiprate= 3.84 Mcps.
The simulated signals spectrum at the output of the chip shaping lter block is
depicted in Figure 3.14.
Once the root raise cosine ltering has been applied to both in-phase and in-
54
Figure 3.14:
quadrature components of the signal, the uplink transmitter model that we con-
ceived allows either processing the modulated signal with clipping techniques or
feeding it directly into the PA without any further manipulation (see Figure 3.12).
Furthermore the block named Parameters settings gives the user the possibility of
Figure 3.15: Implementation through Simulink c blocks of the long scrambling code
generator as used in the WCDMA transmitter model we developed.
55
easily varying some key-parameters of the transmitter,i.e. choosing between the
three clipping strategies described in section 3.5, setting the IBO of the amplier
as well as the clipping ratio level (CR).
PA model is the one described in section 3.1. The amplied WCDMA signal is sent
to the receiver through a gaussian channel; the noise power introduced by the chan-
nel can be adjusted simply typing the desired value into a tab of the Parameters
Settings block.
3.6.2 WCDMA downlink receiver
The receiving end comprises RRC receiving lters and de-scrambling/de-spreading
blocks (see Figure 3.12).
De-scrambling operation consists in multiplying the received signal by a complex
conjugated copy of the scrambling code used at the transmitter. Similarly, informa-
tion bits are demodulated by multiplying each DPPCH/DPCCH by a synchronised
copy of the OVSF codes followed by an integration over the symbol period, which
completes de-spreading delivering a voltage value per symbol to the hard-decision
threshold block. Figure 3.16 illustrates the blocks employed to build up the sim-
ulators de-spreading stage. Overall the receiver structure is quite basic because
of the absence of any kind of channel impairment, i.e. multipath propagation, or
synchronisation issues. In fact, PAPR analysis does not need that level of detail
therefore such additional features have not been included in the simulator design,
although - thanks to its exility - they may be added without signicant eort.
Figure 3.16: WCDMA simulator: despreading stage and bit demodulation.
56
3.6.3 ACLR, EVM and PCDE blocks
ACLR is calculated by means of a bank of RRC lters with the same chiprate
as the shaping lters and centered on the two adjacent channels at 5 MHz and
10 MHz [33]. Figure 3.17 shows the Simulink c implementation of the block used
in our simulation that measures ACLR values.
Figure 3.17: Details of the block that calculates the out-of-band power leakage
(ACLR) of the WCDMA transmitter.
EVM and PCDE have been evaluated
6
following the approach suggested in [45]
and [39]. A brief, although exhaustive, description of the methods used to calculate
these parameters is presented in this section.
EVM and PCDE are two quality metrics that estimate the amount of distortion
on the transmitted signal introduced by non-linear devices, such as real power am-
pliers, mixers, possible clipping stages, etc. According to the denitions given in
section 3.2, EVM represents the dierence -in the time domain- between the ide-
ally modulated signal and the one actually transmitted. In real testing equipment,
EVM can be calculated by substituting to the ideally modulated signal a reference
signal derived from the real waveform at the transmitters output. Such signal is
constructed by the measuring equipment demodulating the signal under test (i.e.
at the output of the transmitter) and remodulating it perfectly according to the
standard specications. This error free version of the signal is then compared the
6
WCDMA standard does not indicate a specic algorithm to calculate these quality metrics.
Only their denition as reported in 3.4 is given.
57
actual output of the transmitter under test. Dening the reference signal as a vec-
tor, R, of N = nm complex samples, where n is the number the of symbols in the
measurement interval and m is the number of chips per symbol, and Z the vector
containing N samples collected from the signal under test, then the error vector E
in the time domain is:
E = Z R. (3.11)
Given E, EVM is calculated according to the following equation:
EV M =
RMS(E)
RMS(R)
100%, (3.12)
where the RMS (Root Mean Square) is calculated over the entire measurement in-
terval, which is usually a time slot.
PCDE is calculated by projecting the time domain error signal onto the channeliza-
tion codes domain and selecting, amongst the obtained values, the maximum error.
In order to achieve correct PCDE results, the error vector E must be divided into n
time-sequential vectors, e, with m complex samples comprising one symbol interval.
Given the matrix C, containing the vectors of the orthogonal channelization codes
belonging to one spreading factor
7
, the inner product between C and e is calculate
for all the symbols in the measurement interval, the result is a kn matrix (k is the
number of codes associated with the spreading factor) and its values represent an
error voltage corresponding to a specic symbol and code. The root mean square
value is then performed over each row of the matrix, thus giving a vector, e
k
, of k
elements. Amongst these k values the peak, e
k
, is selected, nally PCDE can be
estimated through the following equation:
PCDE = 10 log
10
_
( e
k
)
2
RMS(R)
2
_
[dB]. (3.13)
Figures 3.18, 3.19 illustrate the solution devised in our model to implement the
EVM and PCDE calculation algorithms described above.
3.7 Simulations Results: a comparison
The goal of the simulation work done was establishing which is the best clipping
scheme for a WCDMA transmitter. Other than compliancy with the standards
requirements in terms of signals quality (EVM, PCDE) and power leakage in neigh-
bouring channels, two additional parameters steered our analysis: BER performance
and simplicity, i.e. suitability for implementation.
7
In uplink multicode transmission SF= 4 is the only allowed, see section 3.4.
58
Figure 3.18: Constituent blocks of the simulators module that estimates EVM.
Figure 3.19: Constituent blocks of the simulators module that estimates PCDE.
The rst step in evaluating the BER performance in each single case was taking
into account the sole ACLR1 as limiting factor. The back-o of the PA has been set
to the minimum value that gives an ACLR1 with a margin of at least 1 dB over the
limit (33 dB). Given this value, the performance in terms of BER has been assessed.
Figure 3.20 shows the BER curve for clipping and ltering with CR= 3 dB com-
pared to the performances of the unprocessed signal (no IBO applied, hence the
requirements on adjacent channel power leakage are not met) and the unclipped
case, where a back-o of 5.5 dB is used to comply with the ACLR restrictions. It
is worth to notice that clipping is clearly eective in increasing PA eciency as the
unclipped signal has worse performance, the amount of distortion introduced by the
clipping process is small enough to achieve a gain around 1 dB in signal-to-noise
ratio. Now the back-o needed is decreased to 4.1 dB, which conrms that the PA
could be used more eciently.
In Figure 3.21 and Figure 3.22 the results obtained for peak cancellation and peak
windowing are presented. Peak cancellation outperforms both clipping and ltering
59
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
10
6
10
5
10
4
10
3
10
2
10
1
10
0
AWGN power (dBm)
B
E
R
BER performance in AWGN channel
BO=0 dB unprocessed signal
BO=5.5 dB unprocessed signal
Clip&Filt BO=4.1 dB CR= 3 dB
Figure 3.20: BER performance over AWGN channel for Clipping and Filter-
ing (green dashed-dotted line), Unprocessed signal (blue dashed line) and Unclipped
signal (red line).
and peak windowing, which has the worst BER performance. IBO values required
to work above the ACLR1 limit were respectively decreased to 3.2 and 3.1 dB,
suggesting that these techniques reduce the signal peaks sharply around the clip-
ping level, whilst clipping and ltering suers from peak regrowth which forces the
designer to resort to a larger back-o. However, the improvement given by Peak
Cancellation is just slightly better than the one reached by clipping and ltering
and we may claim that they have same roughly performances.
In order to clarify and complete the analysis simulation results, including EVM
and PCDE, have been summarised in Table 3.1. For the sake of simplicity, we pre-
sented BER at xed noise power level (30 dBm). The column indicated as Back-o
contains the results of the unclipped case, i.e. where the only back-o has been ap-
plied. Table 3.1 eectively helps to understand which is the PAPR reduction scheme
that sets the best compromise between BER performance and fullment of the stan-
dards requisites. We have already pointed out that BER performances are quite
close for all the three techniques examined, none of them has a stark advantage over
the others. So it now interesting to note that clipping and ltering introduces less
signal distortion having lower EVM and PCDE values. At this stage, we can state
that, among the techniques tested, there is not a clear winner, all of them seem
60
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
10
6
10
5
10
4
10
3
10
2
10
1
10
0
AWGN power (dBm)
B
E
R
BER performance in AWGN channel
BO=0 dB unprocessed signal
BO=5.5 dB unprocessed signal
BO=3.2 dB; CR=3 dB Peak Cancellation
Figure 3.21: BER performance over AWGN channel for Peak Cancellation (green
dashed-dotted line), Unprocessed signal (blue dashed line) and Unclipped signal (red
line).
to be suitable for implementation in a real system. Nonetheless clipping and lter-
ing could represent the most appealing solution due to its straightforward simplicity.
In general, simulation results have clearly shown that employing PAPR schemes
is not only essential but also advantageous for WCDMA terminals. Comparing the
BER values of the unprocessed signal case to the ones obtained by applying PAPR
reduction strategies, we can observe a notable gain in performance: passing from
6.0 10
2
to a minum of 3.86 10 2 for peak cancellation (best values which duly re-
spects the imposed limitations). Yet, a signicant reduction in back-o requirement
has to be added to the aforementioned improvement. Decreasing the back-o values
allows a more eective use of the PA amplication range, which is a key-aspect in the
design of nal stage ampliers. In order to further improve the PA eciency (thus
improving BER performance), we set the CR to 1 dB. In this way, the signal suf-
fers from more distortion and its quality will deteriorate, on the other hand PAPR
is now greatly reduced, consequently PAs eciency should be enhanced leading
to better BER performance. Table 3.2 illustrates the simulation results obtained.
The IBO values which guarantee the required ACLR1 margin are now respectively:
2.6 (clipping an ltering), 1.3 (peak cancellation) and 0.8 dB (peak windowing).
It is worth to stress out that clipping and ltering and peak cancellation improve
61
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
10
6
10
5
10
4
10
3
10
2
10
1
10
0
AWGN power (dBm)
B
E
R
BER performance in AWGN channel
BO=0 dB unprocessed signal
BO=5.5 dB unprocessed signal
BO=3.1 dB; CR=3 dB Peak Windowing
Figure 3.22: BER performance over AWGN channel for Peak Windowing (green
dashed-dotted line), Unprocessed signal (blue dashed line) and Unclipped signal (red
line).
their BER performance compared to the previous case, whilst peak windowing is
less eective and has a higher error rate value. This is probably due to excessive
level of distortion introduced by the clipping stage, which outbalance the benets
of a lower PAPR. As a conrm, the quality parameters assume values that do not
comply with the standard. Again peak cancellation takes the most advantage out
of the lower clipping ratio in terms of BER and ACLR1, unfortunately the quality
metrics requirements are not met as EVM is greater than 17%.
Actually, the only clipping scheme that would allow to operate within the regulation
is clipping and ltering, even though its modulation accuracy is dangerously close
to admittable limits (EVM= 14.3%).
3.8 Discussion and Conclusions
The multicode transmission option in uplink WCDMA standard is a crucial fea-
ture to support those services that require higher bit rate, e.g streaming applica-
tions. The superimposition of numerous channels (up to six) inevitably increases
the PAPR, i.e. large amplitude peaks are present in the modulated signal. High
PAPR puts strain on the mobiles amplier forcing it to work beyond its saturation
point. The consequences stemming out from this unwanted eect are essentially
62
Table 3.1: Simulated values of ACLR1, BER, EVM and PCDE of the clipping
strategies tested. CR= 3 dB.
CR=3 (dB) Back-O Clip & Filt. Peak Canc. Peak Wind.
ACLR1 (dB) 34.9 34.5 34.3 34.2
BER @ 30 dBm Noise Pwr 6.0 10
2
4.44 10
2
3.86 10
2
4.33 10
2
EV M (%) 3 8.5 12.7 12.5
PCDE (dB) -38.9 -29.2 -25.7 -25.9
Table 3.2: Simulated values of ACLR1, BER, EVM and PCDE of the clipping
strategies tested. CR= 1 dB.
CR=1 (dB) Back-O Clip & Filt. Peak Canc. Peak Wind.
ACLR1 dB 34.9 34.4 34.5 34.2
BER @ 30 dBm Noise Pwr 6.0 10
2
3.73 10
2
3.35 10
2
4.61 10
2
EV M (%) 3 14.3 20.5 20.5
PCDE (dB) -38.9 -24.6 -21.5 -21.5
63
two:
in-band and out-of-band interference/distortion
decreased amplier eciency
WCDMA sets the limits of tolerable modulation distortion (EVM and PCDE) and
PA eciency cannot be too low otherwise terminals would need to recharge their
batteries frequently. The only viable solution to overcome this such detrimental
eects consists in reducing the PAPR and many approached have been devised to
accomplish the task. In this work we developed a WCDMA-compliant model of a
transmitter and simulated its performance applying a particular category of PAPR
reduction techniques (see section 3.5). Among the possible option available in such
category, we picked up three schemes, namely: clipping and ltering, peak cancella-
tion and peak windowing. System performance in terms of BER, spectral splatter
(ACLR1) and modulation quality (EVM, PCDE) have been evaluated in each case
examined. It is worth reminding that our goal was nding a suitable clipping scheme
which could reach the best trade-o between performance e implementation exi-
bility.
From the results we have shown in the previous section we are able to draw some
conclusions about the eectiveness of the clipping scheme considered:
Clipping the WCDMA multiple data channels signal is useful to reduce its
PAPR and restore part of the PA eciency which would otherwise lost in
large back-o values needed to meet the requirements of power leakage.
Among the technique we took into account peak cancellation has the best
performance in terms of BER. At CR= 3 dB both peak cancellation and
clipping after ltering outperform peak windowing, although they show similar
results. Clipping and ltering adds less distortion with respect to the other
techniques showing better EVM and PCDE results. As expected IBO can
now assume lower values thus improving the PA eciency.
When the clipping ratio is decreased to 1 dB peak cancellation again sets
itself as the more eective solution in terms of BER performance, but it does
not meet the modulation accuracy requisites being EVM greater than the
17% threshold. Peak windowing presents BER values even worse than the
previous case (with CR set to 3 dB), EVM is also out of range as for peak
cancellation. Clipping and ltering is the only scheme that allows the system
to operate within the standard limitations, even if an EVM of 14.3% is very
close to the permitted level. It seems evident that decreasing the clipping
ratio introduced additional distortion that outbalanced the benets deriving
from a reduced signals PAPR.
64
Concluding we can claim that simple signal distortion techniques like the one
we analysed are eective in limiting the signals amplitude range according
to the limits imposed both by the standard and the PA nonlinearities. In
particular, we noticed that applying PAPR strategies the system is able to
achieve better BER performance with respect to the unprocessed signal case.
This due to a more ecient use of the PA. On the other hand, the signal
quality is deteriorated and the CR cannot be too low otherwise the quality
requirements are not met (see Table 3.2).
With regard to suitability for possible implementation, we must conclude that
none of the three schemes has shown performances remarkably higher than the
others, therefore we would indicate clipping and ltering as the most appealing
because it merely based on a digital FIR lter.
65
66
Chapter 4
WiMAX modelling
4.1 Introduction
In chapter 3 we highlighted the issues related to signals with large values of Peak-
to-Average-Power Ratio (PAPR) in WCDMA systems. Basically, a large dynamic
range makes the signal very sensitive to nonlinearities due to the transmitter Power
Amplier (PA) or other nonlinear devices. We pointed out that this result in un-
wanted out-of-band and in-band distortion, which degrades the signal quality and
aect the performance of the whole system. Unfortunately, the same problems are
encountered when employing another modulation scheme, namely Orthogonal Di-
vision Frequency Multiplexing (OFDM)
1
. An OFDM signal consists of a number
of independently modulated subcarriers, which can lead to high PAPR when added
up coherently. PAPR values increase with the number of subcarriers employed, a
statistical analysis has been carried out in [40] for QPSK symbols and generalised
to QAM constellations in [46]. Simulation results show that an OFDM signal can
reach a PAPR greater than 12 dB for 1024 QPSK modulated subcarriers, even if
the probability of such event is extremely low (0.001%).
A physical layer (PHY) based on OFDM technology has been adopted in many mod-
ern digital communications systems, i.e. the IEEE 802.11 Wireless LAN (WLAN),
Digital Video Broadcast (DVB), as well as the IEEE 802.16 Broadband Wireless
Access (BWA). In particular, a broad industry consortium, the Worldwide Interop-
erability for Microwave Access (WiMAX) Forum has started certifying products
that comply with the standard (and its subsequent amendments) developed by the
802.16 group. In the following sections of this chapter we will focus our attention on
this recent technology providing a brief overview of WiMAX and its salient features,
many of which are included in the model we built up in order to assess the eects of
1
In this chapter we will not go through the basics of OFDM. Detailed information can be found
in many published books and websites, e.g. [40] [46], http://en.wikipedia.org/wiki/OFDM
67
nonlinearities on the system. The remaining part of this chapter will be dedicated
to a detailed description of the simulator, after which we will address the issues
arising from large peak values against the strict quality requirements imposed by
the standard.
4.2 WiMAX: an overview
Originally the IEEE 802.16 group focused on developing air interface standard for
LOS point-to-multipoint wireless broadband applications in the 1066 GHz band.
The group subsequently produced a new standard, 802.16a, targeting NLOS com-
munications in the 2 11 GHz spectrum, using an OFDM-based physical inter-
face. Further revisions in the following years resulted in a new standard in 2004,
802.16 2004 [47], which forms the basis for WiMAX. Being a collection of stan-
dards rather than a single interoperable one, 802.16 2004 allows several design
options, such as a variety of dierent choices in PHY and MAC layers features [46].
For obvious practical reasons of interoperability, the scope of the standard had to
be restricted, hence the WiMAX Forum was created to dene a limited set of sys-
tem proles, i.e. mandatory and optional features selected from the wide range of
options given by the standard.
The rst WiMAX solution based on 802.16 2004, known as xed WiMAX, ad-
dressed xed applications that could represent a cost-eective alternative to Digital
Subscriber Lines (DSL) and cables. In December 2005, the IEEE group approved
an amendment to 802.16d, namely 802.16e 2005 (or simply 802.16e) [48], which
added mobility support. 802.16e standard paved the way for mobile broadband
services and it is usually referred as mobile WiMAX. The air interface for mobile
WiMAX is based on Orthogonal Frequency Division Multiple Access (OFDMA),
which introduces the possibility of dividing the available OFDM subcarriers into
subsets (or subchannels) to be allocated to dierent users. OFDMA will be pre-
sented in greater detail in section 4.2.1. Similarly to the previous xed version, the
WiMAX Forum agreed a number of certied proles that could ensure backward
compatibility with xed WiMAX and interoperability between the new mobile ver-
sions. Currently the WiMAX Forum specied two dierent system proles: one
based on IEEE 802.16 2004, OFDM PHY, called xed system prole; the other
one based on 802.16e2005 scalable OFDMA PHY, called the mobility system pro-
le [46]. Besides, the WiMAX Forum, in its role as guarantor for interoperability,
also dened certication proles, which specify the operating frequency, the chan-
nel bandwidth and duplexing mode. WiMAX compliant equipment are certied for
interoperability against a particular certication prole.
Today there are ve certication proles for xed WiMAX and fourteen for the
mobile version (see Table 4.1). All the initial proles for mobile WiMAX use Time
68
Division Duplexing (TDD), although Frequency Division Duplexing (FDD) might be
considered to address specic market opportunities where the spectrum regulatory
requirements either prohibit TDD or are more suitable for FDD deployment [6]. It
Table 4.1: Fixed and Mobile WiMAX Initial Certication Proles.
is also worth to point out that that the standard only covers PHY and MAC layers,
this forced WiMAX Forum to provide guidelines for an end-to-end implementation
of the system [6].
In the reminder of this introduction we will focus on WiMAX salient features, es-
pecially those introduced at physical layer level.
4.2.1 WiMAX salient features
WiMAX is a wireless broadband solution that supports a wide range of features with
a lot of exibility in terms of deployment options and potential service oerings.
The main features oered by WiMAX can be summarised as follows [6] [46]:
OFDM-based physical layer: As already mentioned, the WiMAX physical
layer is based on OFDM, which is a scheme that oers robustness towards
multipath and allows WiMAX operating in NLOS conditions. Furthermore
OFDM is easily adaptable to dierent channelisation bandwidths.
High data rates: The inclusion of MIMO antenna techniques coupled with
exible sub-channelisation schemes and Adaptive Modulation and Coding
(AMC) enable Mobile WiMAX to achieve DL peak data rates per sector up
to 46 Mbps, assuming a TDD system with 10 MHz channel and DL/UL ratio
3:1, and UL rates up to 14 Mbps, given a DL/UL ratio 1:1.
69
Scalability: In order to comply with dierent spectrum requirements, Mobile
(and Fixed) WiMAX is able to scale to work in various channelisations from
1.25 to 20 MHz. This scalability is supported in the OFDMA mode, where the
Fast Fourier Transform (FFT) size may be varied according to the available
channel bandwidth.
Support for TDD and FDD: A strength of WiMAX is that it can operate
either in TDD and FDD modes. TDD is the preferred option in most part
of the system implementations
2
because it oers the possibility to adjust the
downlink/uplink ratio, thus asymmetric trac can be eciently supported.
Other advantages consist in a less complex transceiver design and the ability
to exploit channel reciprocity. The main drawback related to TDD access is
the need for a synchronised network to counter interference.
Quality of service: 802.16e MAC layer ha been designed to oer an end-
to-end IP based QoS for various classes of services, such as VoIP (Voice over
Internet Protocol), streaming audio/video, etc. The goal is reached through
a exible mechanism of resources allocation (scheduling) over the air interface
on a frame-by-frame basis.
Mobility: Mobile WiMAx allows optimised handover strategies with latency
less to 50 ms to ensure real-time applications such as VoIP to perform without
service degradation.
IP-based architecture: The WiMAX forum has dened an all-IP net-
work architecture. This facilitates integration with other networks and oers
the adavantage of a reduced total cost of ownership during the lifecycle of a
WiMAX network deployment [6].
The features listed above make WiMAX a suitable solution for two deployment
scenarios:
WiMAX as an access solution: as DSL complement to support Internet
access and VoIP telephony where the cabled option is unfeasible or too ex-
pensive, e.g. rural and remote subscribers. WiMAX can also be deployed to
deliver wide area BWA, this is an attractive solution for customer who desire
a high-speed internet access but do not wish to be tied down with the PSTN
line. A WiMAX network might even complement other wide area networks,
such as 3G, or be used in conjunction with other wireless technologies, like
WiFi. To date, South Korea is rst country that developed and rolled out
2
All the initial WiMAX proles are based on TDD, except for two xed WiMAX proles in
3.5 GHz
70
a version of mobile WiMAX, called WiBro (www.http://www.wibro.or.kr).
WiBro delivers broadband internet service with a peak throughput (per user)
of 3 Mbps in DL and supports mobility up to 60 Km/h.
WiMAX as a transport/backhaul solution: can be used as base for de-
velopment of point-to-point or point-to-multipoint wireless IP/Ethernet trans-
port connections, e.g. for mobile networks or WiFi hotspots.
In the following section we will focus on the key-features of the WiMAX physical
layer.
PHY layer main characteristics
WiMAX radio interface uses OFDM-based technology, like WiFi and many others
transmission systems. OFDM is an ecient scheme for high data rate transmission
in NLOS or multipath radio environment. OFDM modulation relies on the idea of
dividing a given high rate data stream into several parallel lower rate streams and
nodulating each stream on separate carriers, called subcarriers. OFDM provides
protection against Inter Symbol Interference (ISI) by making the symbol time large
so that channel-induced delay are only a small fraction of the total duration. This is
usually achieved by repeating part of the last OFDM data symbols at the beginning
of the symbol, the technique is called cyclic prex and the time duration of the
copied symbols is referred as guard time. Yet, OFDM is spectrally ecient as the
subcarriers are orthogonal one to each other over the symbol time. Another ad-
vantage of this modulation scheme resides in its easiness of implementation in both
transmitter and receiver: OFDM signals can be digitally generated using IFFT al-
gorithm and received performing FFT.
On the other hand, OFDM is very sensitive to frequency oset and synchronisation,
this aspect limits the support of mobility to speeds up to 120 Km/h [6]. Another
disadvantage is associated with the signals presenting high PAPR, which causes
degradation in signal quality and forces designers to use very linear devices (e.g.
power ampliers, digital-to-analog converters) to avoid unwanted distortion eects.
Mobile WiMAX multiple access is based on OFDMA (Mobile WiMAX OFDMA-
PHY), which is multiplexing scheme relying on subchannelisation
3
of both UL and
DL to accommodate usersdata streams. OFDMA symbol structure consists of three
types of sub-carriers:
data used for carrying data symbols.
3
Subchannelisation, as well as frame and slot structure of WiMAX, are quite complex topics
and a comprehensive discussion would be beyond the scope of this work. A detail explanation can
be found in [46], besides the published standards [48] [47].
71
pilot used for carrying pilot symbols, which are known a priori and are funda-
mental for channel estiamtion/tracking. Power on pilot subcarriers is boosted
by 2.5 dB to increase channel estimation reliability even at low signal-to-noise
ratios.
null have no power allocated to them. They are located at DC frequency and
towards the edges of the spectrum, guard subcarriers.
Figure 4.1 shows a frequency domain representation of a mobile WiMAX OFDM
symbol. Active sub-carriers (data and pilot) are grouped in sub-sets of sub-carriers
Figure 4.1: OFDMA subcarriers structure [6].
called sub-channel (48 data sub-carriers +6 pilot tones). In principle, the network
assigns a sub-channel for each active user, thus avoiding intra-cell interference. In
xed WiMAX (xed WiMAX OFDM-PHY), subchannelisation is limited to
the UL only.
The FFT size of mobile WiMAX PHY is scalable from 128 to 2048. When the
available bandwidth increases, the FFT size is increased accordingly so that the
subcarrier spacing remains xed at 10.94 kHz. As a consequence, the OFDM
symbol duration is constant regardless of the number of subcarriers, hence the
impact on higher levels in minimal. On the contrary, in xed WiMAX the number
of subcarriers (the FFT size) is xed at 256, therefore the subcarrier spacing varies
with the channel bandwidth.
Other relevant characteristics common to both xed and mobile WiMAX PHY are
listed below:
MIMO: use of multiple antenna techniques are integral part of the standard
802.16e. Beamforming, Spatial Multiplexing (SM) and Space-Time Codes
(STC) are supported. MIMO schemes are essential in order to increase through-
put and coverage.
Hybrid ARQ: Chase Combining HARQ is a mandatory feature of WiMAX,
which provides fast response to packet errors and improves cell edge coverage.
72
Adaptive Modulation and Coding (AMC): AMC is an ecient mecha-
nism to maximise throughput in time-varying channels: the highest modula-
tion and coding scheme is selected according to the signal-to-noise/interference
ratio measured on the radio link. Mandatory modulations in DL are QPSK
and 16 QAM, whilst in 64 QAM is optional both in UL and DL.
CQICH (fast channel feedback): this channel is used by the MS to send
relevant channel-state information to the BS. It is essential to support MIMO
mode.
Forward Error Correction (FEC): Convolutional Codes (CC), Convolu-
tional Turbo Codes (CTC) and Repetition Codes are supported.
So far, we outlined the main features of WiMAX with particular attention to the
physical layer, highlighting the the slight dierences between xed and mobile ver-
sions. From now on, we will take refer to mobile WiMAX as it is subject of our work,
all the attributes include in the simulator we developed comply with the 802.16e
amendment. Figure 4.2 shows the OFDM parameters values as used in the model
of mobile WiMAX PHY that will be described in the following sections. The pro-
les in Table 4.2 are only a set of the possible implementations and they have been
chosen because they are likely to be deployed in the initial roll-outs. In particular,
our simulations are based on a channel bandwidth of 10 MHz, the FFT size has
been set to 1024 and the frame duration is 5 ms.
Table 4.2: OFDM parameters used in mobile WiMAX. Boldfaced values correspond
to those used in our simulator.
73
4.3 Mobile WiMAX system model
In this section we provide an in-depth description of the Simulink c model for
mobile WiMAX physical layer that was developed in few months in summer 2007.
The simulator in based on IEEE 802.16e specications and includes only part of the
wide set of functionalities oered by the standard. In fact, due to time constraints,
many key-features have not been implemented, such as subchannelisation or channel
estimation. Nonetheless the options available are sucient to highlight the critical
issues arising from high PAPR, which was the principal driver of our research work.
In our analysis we restricted the scope to DL transmission assuming that the base
station (BS) is transmitting continuously and using all the data and pilot subcarriers
available (full load), this assumption represent the so-called worst case scenario but
does not aect the PAPR analysis in any way. However, the system model we
conceived can be easily upgraded adding those attributes not included: so far, it
can be regarded as a starting point for further development of mobile WiMAX
physical layer.
4.3.1 Transmitter
Figure 4.2 depicts the functional blocks
4
that constitutes the transmitter. The ba-
sic time-unit of the simulator is an OFDM symbol; a stream of randomised bits
is generated by a source block (which is not shown in the gure), then bits are
encoded and mapped into QAM symbols. Each OFDM symbol comprises 720 data
subcarriers, 120 pilots and 184 null/guardband subcarriers. Tones are assembled in
the assembler block according to the rules in [48] and its structure in the frequency
domain is illustrated in 4.1. Once symbols have been properly ordered, IFFT oper-
ation transforms the array of symbols into a time domain signal. Cyclic Prex (CP)
is then added and the signal is fed into the block that performs digital interpolation.
Similarly to WCDMA (see chapter 3), the upsampling stage has been included for a
more eective signals envelope reconstruction, which leads to a more realistic PAPR
estimation. After this process, the signal is passed through the power amplier (RF
HPA).
Source
Information bits are generated in random sequences of sequences of length:
N
bits
= N
data
R log
2
(M), (4.1)
where N
data
is the number of data subcarriers, R represents the overall coding rate
and M denes the modulation order. As stated in the 802.16e specication [48], the
4
A complete Simulink c block diagram of the simulator is given in Appendix A.
74
Figure 4.2: Functional blocks diagram of the mobile WiMAX transmitter model.
informations bits must be randomised before encoding and the initialisation vector
for a shift register randomiser is specied. In our simulator, we preferred using the
in-built option of the binary source block given by Simulink c instead of performing
the standard-compliant bits randomisation, but this does not substantially aect the
results obtained.
Encoder
Mobile WiMAX standard supports both convolutional and turbo coding (abbrevi-
ated as CC and CTC respectively), although the mandatory option is CC only.
In our case, the sole CC has been implemented according to the parameters given
by [48]. The encoder has native code rate (R) of 1/2 and a constraint length of 7.
The generator polynomials, expressed in octal alphabet, are:
G
1
= 171
OCT
for X
G
2
= 133
OCT
for Y (4.2)
where X and Y represent the encoder outputs.
In order to achieve code rates greater than 1/2, the encoder output is punctured
following the puncturing vectors in [47]. In the DL of xed WiMAX, where subchan-
nelisation is not used, an outer Reed Solomon code is added before the convolutional
encoder. The permitted code rates are: 1/2, 2/3, 3/4, 5/6
5
. In Simulink c convolu-
tional coding and puncturing are supported in a single block, which was used for
implementation.
Interleaving
After the coding stage, bits are interleaved using a two steps mechanism. The rst
step ensures that adjacent coded bits are mapped onto non-adjacent subcarriers,
5
This value is used in conjuction with Reed Solomon code only.
75
whilst in the second step adjacent bits are alternately mapped to less and more
signicant bits of the modulation constellation. This strategy has been designed to
exploit frequency diversity, thus improving the performance of the decoder.
The interleaver implemented in our simulator comprises two blocks that perform
the aforementioned operations:
Block 1 provides block interleaving by lling the matrix row by row, then values
are read by columns. The matrix has sixteen rows (N
rows
= 16) and a num-
ber of columns which varies according to size of the encoded bits array, i.e.
N
columns
= N
tcb
/N
rows
, where N
tcb
is the total number of coded bits;
Block 2 rearranges the elements of the input vector according to an index vector
given by the following equation:
m
k
= s
_
j
k
s
_
+mod
_
j
k
+N
tcb
_
j
k
N
rows
N
tcb
_
, s
_
+ 1, (4.3)
where m
k
and j
k
represent the indices of the bits after and before the inter-
leaving operation, whereas s is dened as the half of log
2
(M) (number of bits
per subcarrier, N
cpc
in the standard), i.e. s = (log
2
(M)/2).
Modulation Mapper
The modulation mapper (or symbol mapper) converts a sequence of binary data
into a sequence of complex-valued symbols. In mobile WiMAX the mandatory
constellations are QPSK and 16 QAM, with an optional 64 QAM also dened in
the standard which is likely to be implemented - at least for the downlink - in most
of the systems [46].
The allocation of bits to a symbol follows the Gray coding method, which ensure
that adjacent constellation points only dier by a single bit. In order to have
equal average symbol power, each constellation is normalised to unit average power
applying an appropriate scaling factor. In our model, we used the symbol mapper
available in the Simulink c library.
Assembler
The mobile WiMAX prole we chose to implement (see Figure 4.2) supports 720
data subcarriers, 120 pilots tones and the remaining are null subcarriers. The
Assembler block rearranges data and pilots symbols according to the OFDM sym-
bol structure of Figure 4.1. The exact distribution of subcarriers depends on the
permutation mode selected. The PHY model we developed does not include sub-
channelisation, therefore pilots tones have been evenly spread across the spectrum,
resulting in a pattern of one pilot tone every six data subcarriers. Pilots subcarriers
76
are modulated by pseudo-random sequences of BPSK symbols and their power level
is boosted by 2.5 dB with respect to data subcarriers. Furthermore, the absence of a
channel estimation functionality makes unnecessary the implementation of training
sequences.
The null subcarriers are inserted at the edges of the OFDM symbol (guard bands)
as this leads to decreasing the emissions in the neighbouring bands; the DC carrier
is also suppressed.
IFFT and Cyclic Prex
As already mentioned in section 4.2.1, OFDM modulation strategy splits a high
data rate sequence of symbols into multiple parallel low data rate streams, each of
which separately modulates an orthogonal subcarrier. The output of the assembler
block is vector of symbols and each element of this array can be regarded as a sample
of one of these multiple parallel symbol sequences. The IFFT operation converts
the array of symbols into a discrete-time version of the OFDM signal; the output of
the IFFT block is a vector containing time samples whose values are given by the
following equation:
x[i] =
1
N
c
Nc1
k=0
s[k]e
2j(kB
c
)i
T
u
N
c
, i = 0, . . . , N
c
1, (4.4)
where N
c
stands for the total number of tones (which, in our case, is 1024), s[k] is
the symbol modulating the kth subcarrier, B
c
represents the frequency separation
between two adjacent subcarriers and T
u
is the useful OFDM symbol time (i.e.
excluding the cyclic prex). Actually, the IFFT algorithm returns time-samples of
the ideal continuous OFDM signal taken at instants multiple of T
u
/N
c
, which yields
to a sampling frequency of 11.2 MHz given the set of parameters chosen.
After IFFT, the cyclic prex is added to the signal. The guard interval is created
by copying part of the last samples and appending them to the beginning of the
OFDM symbol. The resulting time duration of this portion of samples is named
guard interval (T
g
). As a consequence, the total OFDM symbol length (T
sym
) is
now increased (see Figure 4.3):
T
sym
= T
u
+Tg. (4.5)
Guard time T
g
values are usually indicated as fraction of the useful symbol time,G
G =
T
g
T
u
. (4.6)
In our case, we kept the cyclic prex length xed at 1/8 of the useful symbol du-
ration, i.e. G = 1/8 or 12.5%. Yet, in mobile WiMAX OFDMA PHY the length
77
of the useful symbol time is constant, hence the guard time has xed duration too
(see Figure 4.2).
Figure 4.3: Insertion of cyclic prex in OFDM symbols.
The simulator uses the standard IFFT Simulink c block and CP is appended
using a simple matrix concatenation block available in the library.
Interpolation
During this stage the signal is upsampled and ltered by a direct-form FIR lter.
Upsampling operation inserts L 1 zeros between signals samples, L is called the
upsampling factor. The subsequent ltering stage reconstructs (interpolates) the
signal envelope between the original input signals samples. This block has been
introduced in the attempt of simulating the digital-to-analogue conversion that takes
place in any digital communications device.
The signals shape at the output of the interpolator presents a greater number of
peaks with respect to its original version. This is due to the fact that the sampling
time has been articially increased (the new sampling time is L times smaller) by
adding non-informative samples and the signals shape has been predicted through
linear ltering. Equation 4.4 indicates that the IFFT algorithm outputs a discrete-
time version of the ideally modulated OFDM signal (i.e. the one we would obtain
from modulating continuous-time tones) and its sampling time is equal to T
u
/N
c
.
As a result, it may happen that possible high signal peaks are not present because
of a slow sampling rate [30].
Real digital-to-analogue converter transforms numerical values into a voltage signal
through interpolation as described here.
The interpolation factor, L, used throughout our simulation work has been set to
3.
78
Power Amplier Model
The AM/AM characteristic
6
of the amplier has been modeled with Rapps curve
as described in section 3.1. The smoothness factor value, p, has been set to 3 as in
WCDMA model (section 3.1), whilst the output saturation power has been raised
to 43 dBm. The linear gain, G, is 30 dB, which is a typical value for DL power
ampliers for mobile WiMAX applications.
The PA model now contain a new feature as a source of white noise has been added
and its power level can be tuned modifying the Noise Figure (NF) parameter. NF
is dened as:
NF = log
10
_
1 +
Noise Temperature
290
_
, (4.7)
the value 290 corresponds to the standard thermal noise temperature in kelvin
degrees and Noise Temperature indicates the equivalent thermal noise temperature
of the PA. The NF range of values can be quite wide depending on the type of PA
employed, in our simulation such value has been set to 3.5 dB, which is usual in
good quality ampliers for base stations.
4.3.2 Receiver
In Figure 4.4 the functional block scheme of the mobile WiMAX receiver is illus-
trated. Basically, the receiver performs the inverse operations as the transmitter.
At rst, the received OFDM symbols are downsampled to rule out the redundant
samples introduced by interpolation, then the cyclic prex is removed and the signal
is sent to the FFT block, which demodulates the symbols on each subacarrier. The
Disassembler block is in charge of separating the dierent types of data, i.e. pilot,
data or null tones. Actually, this block removes guardband and pilot tones and
groups together data symbols, which are successively sent to the demapper. Once
data symbols have been revealed, they enter the decoding stage.
Figure 4.4: Functional blocks diagram of the mobile WiMAX receiver model.
6
Rapps model for PAs does not introduce phase noise, so the only input-output amplitude
function is available.
79
FFT
In section 4.3.1 we pointed out that the IFFT algorithm converts a set of QAM
symbols into a discrete-time version of an OFDM signal, in which each of the input
symbols modulates an orthogonal subcarrier. The FFT alogrithm performs the
inverse operation of demodulating each tone that compose the OFDM symbols,
thus the output of the FFT block is an array of QAM symbols whose size equals the
total number of subcarriers. Resorting to Equation 4.4, we can derive the following
relation between the received samples, x[i], and demodulated symbols outputted by
FFT algorithm, s[k]:
s[k] =
Nc1
i=0
x[i]e
2j(kB
c
)i
T
u
N
c
, k = 0, . . . , N
c
1, (4.8)
Disassembler
This block is needed to separate data, pilot and null subcarriers. In our simulator,
null subcarriers, i.e. guardabands and DC tones, are discarded along with pilots. As
channel estimation has not been implemented pilot symbols are useless, otherwise
they would sent to the channel estimator block.
Demapper
The stream of data symbols at the output of the Disassembler is fed into the Demap-
per, which converts complex constellation points into a bit sequence. In our model,
demapping is based on hard decision metric, which means that the Euclidean dis-
tance between the received symbol and the whole set of allowed points of the con-
stellation is calculated. The received symbol is then associated to the constellation
point with the smallest distance. The simulator uses a standard hard decision
demapper available in Simulink c communications library.
Decoder
The stream of bits at the output of the demapping stage is passed on to the decoder
that comprises two steps: deinterleaving and Viterbi decoding.
Deinterleaving rearranges the bits in the correct order as before the interleaving
stage. Similarly to the interleaver block in the transmitter, deinterleaving is made
up of two blocks, which perform exactly the same operations as described in sec-
tion 4.3.1.
The decoding strategy is based on the well-known Viterbi algorithm, which oer
a low-complexity solution leveraging an approximated maximum likelihood metric.
80
An exhaustive description of Viterbi decoding is beyond the scope of this disser-
tation, more details can be found in [49]. Simulink c provides a specic Viterbi
decoding block with a set of tunable parameters, which conveys the exibility to
adapt to many cases other than WiMAX.
4.3.3 Addressing high PAPR in WiMAX systems
As we anticipated in the introduction to this chapter, OFDM signals present large
PAPR values, which leads to deleterious eects in presence of nonlinear devices, such
power ampliers or digital-to-analogue converters (DAC). Being based on an OFDM
PHY, WiMAX waveforms are characterised by high peaks well above its average
level, as a consequence very linear PA and DAC must be employed, otherwise signals
would be unduly distorted aecting the system performance.
In WiMAX, problems related to high PAPR are particularly severe resulting in
dicult challenges to face when designing the radio. In section 4.1, we pointed
out that transmitting high PAPR signals through a nonlinear devices causes in-
band and out-of-band distortion. In order to overcome these unwanted eects, two
options are available:
1. choosing devices that can accommodate signals with wide dynamic range in
their linear region,
2. reducing the PAPR by means of appropriate techniques.
In chapter 3 we discussed about the drawbacks stemming from the rst approach:
high-linearity PA are expensive and characterised by low eciency, yet the alter-
native solution of applying a back-o to the input signal, IBO (see section 3.1),
equally leads to a decreased eciency that limits battery life and PA gain is pe-
nalised by IBO (which means a restriction of coverage range in mobile applications).
Conversely, PAPR reduction techniques try lowering the peaks either modifying the
waveform or coding the source data in a way that avoids high PAPR combinations.
As shown in section 3.7, approaches based on digital waveform manipulation eec-
tively reduce the PAPR, although they introduce a certain amount of distortion.
The amount of signals degradation allowed is a key aspect associated to radio design
as its limit is usually imposed by the standard, e.g. in WCDMA systems, ACLR,
EVM and PCDE are quality metrics that specify the permitted distortion (see sec-
tion 3.6.3). Similarly to WCDMA, mobile WiMAX specication [48] imposes tight
limits on each QAM constellation quality, again the parameter used is EVM, even
if its denition has been adapted to OFDM signals.
Evaluating PAPR The mobile WiMAX PHY model has been provide with a
block that calculates PAPR for each OFDM symbol transmitted. PAPR denition
81
is given in Equation 3.1. In order to describe the statistical distribution of PAPR,
the SImulink c automatically creates an array containing several PAPR values cor-
responding to 1 second record of the interpolated
7
WiMAX signal (roughly 10.000
OFDM symbols). This set of data is then used to build the Complementary Cumu-
lative Distribution Function (CCDF) of the PAPR. The OFDM parameters used
throughout our simulations are shown in Figure 4.2, the constellation is 16 QAM
and the overall code rate, R, is 3/4.
The CCDF curve is depicted in Figure 4.5, where it appears evident that OFDM
symbols aected by extremely high peaks seldom occur, nonetheless they must be
taken into account in the radio design.
Theoretically, an IBO value greater or equal to the PAPR would prevent the input
7 8 9 10 11 12 13
10
4
10
3
10
2
10
1
10
0
CCDF of PAPR
l
o
g
1
0
(
1
C
D
F
)
PAPR(dB)
Figure 4.5: CCDF of PAPR for mobile WiMAX system. 1024 subcarriers, 16 QAM,
R = 3/4.
signal driving the the PA into its nonlinear region. In our case, Figure 4.5 shows
that an IBO value of -at least- 12 dB would be necessary, although such a conser-
vative back-o does not guarantee that a given OFDM symbol will have a PAPR
exceeding the IBO. In practice, such large back-o values would decrease the PA
performance to intolerable level.
Due to its nonlinear AM/AM characteristic, the PA will clip the high peaks intro-
7
It is worth to notice that PAPR calculation is performed after the interpolation stage as this
operation approximates a DAC of a real transmitter.
82
ducing distortion. As an example, Figure 4.6 shows the signal at the PA input along
with its amplied
8
version at the output, clearly peaks with amplitudes beyond the
saturation level have been cut. As a result, the signal quality is now deteriorated.
0 50 100 150 200 250
0
5
10
15
20
25
30
35
40
45
50
samples
A
m
p
l
i
t
u
d
e
[
V
o
l
t
]
Amplifiers clipping effect on high PAPR input signal
Figure 4.6: PA clipping eect: input signal (blue line) and amplied output signal
(red line). Amplitude peaks have been constrained to saturation level.
We have already mentioned that the distortion caused by a nonlinear operation
like amplifying can be divided into two categories: in-band and out-of-band dis-
tortion. In-band interference can be identied using a I/Q constellation diagram
(symbols have been acquired from the demapper input, see Appendix A) as in Fig-
ure 4.7, where the scattered distribution of points around the theoretical 16 QAM
constellation symbols reveals the presence of noise induced by the amplier clip-
ping eect. With regard to out-of-band distortion, it can be easily observed in
the frequency domain through a spectrum analyser (which, in our case, is an FFT
operation) as done in section 3.3. Figure 4.8 depicts the signals spectrum both
at the PA input and output. As we noticed for WCDMA systems, the growth of
spurious spectral components outside the ideal signal band is a direct consequence
of PA nonlinearity and the eect increases as the signals input power approaches
the ampliers saturation point.
8
For the sake of a neat visualisation the PA gain has been set to 0 dB.
83
Figure 4.7: 16 QAM constellation diagram immediately pinpoints the presence of
noise due to PA nonlinearity (all the other possible source of interference have been
excluded). Input power 10 dBm, output saturation power 43 dBm, linear gain
30 dB.
The amount of degradation introduced on the signal by the various system com-
ponents, such as PA, mixers, local oscillators, etc. must be kept within specied
limits, otherwise it would negatively aect the overall performance. In the 802.16e
specication, EVM is used quality metric to estimate the modulation accuracy.
EVM has been comprehensively discussed in section 3.4 and, being a generic pa-
rameter commonly used for measuring digital modulations errors, the same general
concepts can be applied to both WCDMA and WiMAX. Obviously, the denition
had to be slightly adapted to OFDM signals, yielding to the following equation [48]:
EV M =
kS
(I
2
k
+ Q
2
k
)
kS
_
I
2
0,k
+Q
2
0,k
_ , (4.9)
where I
0
and Q
0
represent the ideal symbol point on the k-th subcarrier, whilst
I
k
, Q
k
denote the dierence between the actual observed point and the refer-
ence one. S is the group of modulated data subcarriers where the measurement is
performed, which, in our case, corresponds to 720 tones. Maximum allowed limit of
EVM are given as function of the data rate, i.e. constellation order and code rate,
and they are expressed in decibels
9
. For 16 QAM-R = 3/4, EVM must not exceed
9
In WCDMA standard EVM is given as percentage, see section 3.4.
84
Figure 4.8: Signal spectrum distorted by PA nonlinearity (black line) compared
to the spectrum calculated from the input signal. Input power 10 dBm, output
saturation power 43 dBm, linear gain 30 dB.
24 dB [48], which is a notably strict limit.
In order to obtain a more detailed estimate of possible EVM values in a real WiMAX
transmitter, we introduced a baseband model of a mixer both in uplink and down-
link. In UL, the mixer has been placed immediately after the interpolation stage
and before the radio amplier model, whereas in DL it is positioned at the top of
the receiving chain (see gure in Appendix A). In Figure 4.9 simulated values of
EVM have been plotted against the power emitted by the PA; results show that the
EVM rapidly increases as the output power level approaches the saturation limit.
At 40 dBm the EVM is scarcely below the imposed limit of 24 dB, at maximum
output power EVM has a value greater than 10 dB, which denotes an unacceptable
amount of distortion. As a consequence, the limitation imposed by EVM heavily
penalises the PA power range, which would be limited to 40 dBm instead of the
nominal 43 dBm. Resorting to a conservative IBO would ulteriorly reduce the PA
range and eciency, thus the only viable solution seems to be represented by PAPR
reduction techniques.
In section 3.5 we pointed out that clipping a signal reduces the PAPR at the ex-
pense of adding distortion, in particular, the level of induced noise increases as the
clipping amplitude is lowered. As explained in section 3.8, designers are forced to
seek for a compromise between clashing requirements: improving the PA eciency
-hence lowering the PAPR- and respecting the quality requisites imposed by the
85
10 15 20 25 30 35 40 43 45
40
35
30
25
20
15
10
5
Pout [dBm]
E
V
M
[
d
B
]
EVM vs PA output power
16 QAM R=3/4
Figure 4.9: Simulated EVM values as a function of the PA output power. Input
back-o not applied.
standard. Furthermore, WiMAX designers have to face an even harder task be-
cause OFDM signals are characterised by very high values of PAPR and the margin
for error is quite tight.
Unfortunately, we could not go further with the analysis due to time constraints.
Nonetheless, the main diculties that may be encountered in addressing PAPR for
WiMAX systems have been emphasized. The Simulink c model has been included
in the CD-rom attached to this dissertation. So far, the simulator can be regarded
as a good starting point for further development.
4.4 Conclusions
In this chapter we provided a brief overview of WiMAX focusing on those innova-
tive features that make it a promising wireless broadband system in the next future.
Both xed and mobile WiMAX are able to deliver high data rates coupled with an
unprecedented exibility. There are no doubts that the success of WiMAX will be
mainly determined by the local spectrum regulatory policies, nonetheless from a
technical viewpoint WiMAX must be recognised as a milestone towards next gen-
eration wireless networks and services. As an example, the WiBro network recently
rolled out in South Korea is the rst large-scale system that employs multiple an-
86
tennas schemes.
The physical layer of mobile WiMAX, based on the IEEE 802.16e 2005 has been
described in more detail as it represents the core of the work brought forward in
this chapter. In fact, sections 4.3.1 and 4.3.2 contain an exhaustive description of
the Simulink c model for mobile WiMAX PHY that we developed in summer 2007
at Electric & Electronic Engineering department of UCL. Although many essen-
tial features are missing, the simulator allowed to carry out an accurate analysis
about the potential issues related to high PAPR in WiMAX systems. Being based
on OFDM technology, WiMAX signals waveform present particularly large peaks
compared to their average amplitude level. We have seen in the previous chapter
dedicated to WCDMA that high PAPR leads to detrimental eects in terms of con-
stellation accuracy and out-of-band spectral splatter. In WiMAX this problem is
also more severe because of the strict limits by PHY specics, i.e. allowed values
of constellation error (EVM) are meant to ensure a high signal quality, thus ruling
out every possibility of employing aggressive PAPR reduction techniques. In simple
terms, the modulation accuracy required by [48] only allows very little distortion,
whilst PAPR scheme may introduce signicant distortion. As a result, the margin
for operation is quite tight.
The work-plan we had in mind could not be brought to the desired end; an in-depth
analysis of possible solution for decreasing the PAPR (as done for WCDMA trans-
mitter) has not been performed, nonetheless the work done so far forms a solid base
for further development.
87
88
Chapter 5
Conclusions
In the vast panorama of wireless communications, Ultrawide Band systems have
gained a crucial position in the area of short-range applications. In chapter 2 we
noted that UWB systems that employ radio pulses with extremely short time dura-
tion as physical interface are able to perform very accurate ranging, which is required
in several applications requiring such high precision, e.g. human/asset tracking and
positioning. Nonetheless the UWB specied bandwidth can be exploited using dif-
ferent radio interfaces, such as OFDM. In this case the priority is given to delivering
high data rates rather than positioning, as an example, the so-called wireless USB
applications follow this approach as they are meant to replace standard USB cables
delivering up to 480 Mbps over 3 meters range.
The LDR-LT prototype developed by the PULSERS consortium, and tested at PRL,
transmits pulses of few nanosecond duration and was designed to provide a resolu-
tion of 30 cm in distance measurement coupled with a useful data rate of 12.5 Mbps.
PULSERS put particular emphasis on the low-complexity architecture of the plat-
form for two main reasons. First the LDR-LT demonstrator should evolve into a
commercially viable product, hence the overall cost of the system has been kept
intentionally low. The second point is that PULSERS wanted to show that accu-
rate ranging is feasible even with devices based on low-complexity architecture and
algorithm, conversely systems already available on the market are based on more
sophisticated (and more expensive) architectures. In fact, the results obtained by
the extensive eld trials campaign conceived by the author and held at PRL clearly
show that the prototype is able to perform ranging with an accuracy far better than
any narrow-band system, which is promising outcome considering that the margin
for improving the platform with new features is still broad. It suce to remind
that the system had no data protection codes or advanced data post-processing,
whilst the whole range of commercial positioning devices today available have. In
harsh radio conditions, such Non-Line-of-Sight environments, the LDR-LT platform
has an average distance estimation error that does not exceed 45 cm, which is an
89
excellent result given the simplicity of the system.
The analysis of collected data indicates that many errors in ranging estimation are
imputable to synchronisation failures or misdetections due to multipath. Such errors
might be avoided by introducing error correction codes and xing the synchronisa-
tion mechanism. For such reasons, we decided to remove -as much as possible- those
spurious measurements through a simplied Kalman ltering process. As a result,
most part of the unwanted errors have been ruled out and distance estimations can
rely on a more accurate set of data. Most importantly this allowed us to get closer
to the actual systems performance, hence we have been able to draw more reliable
and realistic conclusions on its eective capabilities.
In conclusion, we can state that the measurement campaign carried out at PRL con-
tributed to demonstrate that UWB-technology is capable of high precision ranging,
even with a low-complexity platform like LDR-LT. The results drawn from eld
trials form the base for the future development of the prototype into completely
functional system for asset tracking and intelligent sensor network applications.
The PULSERS project has now entered into its second phase and the partners are
working on those essential improvements needed to make LDR-LT a fully opera-
tional system.
In chapters 3 and 4 we addressed the issues related to high PAPR signals for
two dierent modulation schemes, namely WCDMA and OFDM.
WCDMA has been chosen as the physical layer interface for UMTS, which is a cel-
lular system adopted in many countries worldwide. In recent years dedicated stan-
dardisation groups have worked to adapt WCDMA to the increasingly demanding
requirements for higher data rates. Recently a revised version of WCDMA has been
released and UMTS can now support higher bit rates both in downlink and uplink
(respectively HSDPA and HSUPA). Chapter 3 includes a detailed description of the
Simulink c model for WCDMA physical layer (based on the original version com-
monly knows as Release 99) along with a comprehensive study on possible PAPR
reduction techniques that may be applied to mobile devices. The model developed
is fully compliant with the standard, this confers a solid reliability to simulation
results. As widely discussed in the introduction to the chapter, PAPR has seri-
ous negative eects on modulation quality and systems performance in presence of
nonlinear devices, such as power ampliers or digital-to-analog converters. Signals
with high PAPR drive ampliers (that are the most powering consuming devices
in mobile equipment) into saturation. As a consequence, if not properly addressed,
high PAPR can shorten battery life, which is a particularly signicant problem in
mobile devices. Up to date, several approaches have been devised to tame the detri-
mental eects of PAPR and some of them are especially attractive because of their
suitability to digital implementation. Such techniques are usually know as signal
distortion techniques, as they directly modify the signals waveform in order to lower
the highest peaks. Being in the form of digital FIR lters, their implementation
90
impact on WCDMA systems is minimal. That is the reason why we decided to
analyse the performance of three dierent strategies belonging to this category.
Simulation results show that all the techniques employed are eective in reducing
the signals PAPR, although each of them introduces a dierent amount of dis-
tortion. In fact, limiting the signals amplitude range to a desired level causes a
degradation in modulation quality. The WCDMA standard imposes tight limits
on the tolerable amount of signal distortion, this compels the designers to strike a
compromise between PAPR reduction and modulation deterioration. Amongst the
clipping strategies considered, the one that oers better performance against the
standards requirements is clipping and ltering. This technique merely consists in
hard clipping the signal and removing the out-of-band distortion due to nonlinear-
ities by digital ltering. With the regard to the ease of implementation, clipping
and ltering is surely the most suitable. The others strategies show performance
close to clipping and ltering, though they rely on slightly more complex algorithms.
Like in WCDMA, OFDM-based systems generate signals with large PAPR val-
ues. In these systems the problem is even more signicant than in WCDMA because
the PAPR value increases with the number of modulated subcarriers. Amongst the
wireless systems employing an air-interface based on OFDM modulation scheme,
WiMAX is -at the moment- the most promising technology for delivering broadband
services. In chapter 4 we highlighted the principal features of WiMAX, especially
those characteristics that make WiMAX unique in the present scenario of wireless
communications.
Similarly to WCDMA, we developed a Simulink c model for mobile WiMAX phys-
ical layer. A detailed description of the model is given in chapter 4. Although the
model lacks some key-functionalities, like subchannelisation or channel estimation,
the options available are enough to pinpoint those critical points arising from high
PAPR. In the case of WiMAX, the standard specications on modulation accuracy
are more stringent than in WCDMA, this narrow the margin left for possible use
of PAPR reduction techniques. Our preliminary study on the eect of PAPR in
the presence of a nonlinear amplier clearly indicates that the power emitted by a
WiMAX base station should be unduly reduced to satisfy the signals quality req-
uisites imposed by the standard. In real scenarios, this would tremendously shrink
the coverage area, therefore solutions must be found that ensure an ecient use
of ampliers. Undoubtedly, WiMAX designers have to face a hard task given the
extremely high values of PAPR and stringent limitations on modulation quality.
All the cited Simulink c models and relative Matlab c codes have been include
in the CD-rom attached to this dissertation.
91
92
Appendix A
IEEE 802.16e PHY layer model
Figure A.1 shows the Simulink c block model for the IEEE 802.16e physical layer
as developed at the Electrical and Electronic Engineering department of UCL. The
functionalities of each constituent block have been explained in detail in chapter 4.
Main simulation paramenters, such as IBO, total emitted power, modulation or-
der, etc. can be changed simply typing the desired values into the dialog box
named paramenters settings. As mentioned in chapter 4, the model does not
include transmit diversity, hence it is assumed that there is a single transmitting
antenna. The block named EVM calculates the modulation distortion according to
the equation 4.9 specied by the standard and reported in chapter 4, section 4.3.3.
PAPR is evaluated for each transmitted WiMAX frame and values are displayed in
a Simulink c display box placed at bottom right side of the screen.
93
Figure A.1: Mobile WiMAX PHY model
94
Bibliography
[1] A.F. Molisch I. Oppermann C. Politano D. Porcino M.G. Di Benedetto,
T. Kaiser. UWB communications systems: A Comprehensive Overview. Hin-
dawi Publishing Corporation, 2006.
[2] W.Hirt D.Porcino. Ultra-wideband radio technology: Potential and Challanges
ahead. In IEEE Communications Magazine, volume 41, pages 6674, July 2003.
[3] ETSI. Physical channels and mapping of transport channels onto physical
channels (FDD). Technical Report TS 25.211 v2.5.0 (1999-03), 3GPP, 1999.
[4] ETSI. Universal Mobile Transmission System (UMTS); Spreading and Modu-
lation. Technical Report TS 125 213 v6.5.0 (2006-03), 3GPP, 2006.
[5] Agilent. Agilent PN 89400-14; Using Error Vector Magnitude Measurements
to Analyze and Troubleshoot Vector-Modulated Signals. Product note.
[6] http://www.wimaxforum.org/technology/downloads/. WiMAX White Paper
Part1: A Technical Overview and Performance Evaluation. Technical report,
WiMAX Forum, August 2006.
[7] The Commission of European Communities. on allowing the use of radio spec-
trum for equipment using ultra-wideband technology in a harmonised manner
in the Community. 2007/131/EC, February 2007.
[8] R.Khono M.Ghavami, L.B.Michael. ultra wideband signals and systems in com-
munication engineering. John Wiley & Sons, Ltd, 2004.
[9] R.J.Fontana. Recent System Applications of Short-Pulse Ultra-Wideband
(UWB) Technology. In IEEE Transactions on Microwave Theory and Tech-
niques, volume 52, Sept. 2004.
[10] Federal Communications Commisssion (FCC). Revision of Part 15 of the Com-
missions Rules Regarding Ultra-Wideband Transmission Systems. First Report
and Order. ET Docket 98 153, FCC 02 48, April 2002.
95
[11] R.A.Scholtz M.Z.Win. Ultra-Wide Bandwidth Signal Propagation for Indoor
Wireless Communications. In Towards the knowledge Millenium International
Conference - ICC 97, volume 1, pages 5660, June 1997.
[12] M.Quinlan S.J.Ingram, D.Harmer. UltraWideBand Indoor Positioning Systems
and their Use in Emergencies. In Position Location and Navigation Symposium
2004 - PLANS 2004, pages 706715, April 2004.
[13] IEEE 802.15 TG 4a. http://www.ieee802.org/15/pub/tg4a.html.
[14] W.Hirt D.Porcino. Pervasive Ultra-wideband Low Spectral Energy Radio Sys-
tems (PULSERS). In Contribution to the 7
th
WWRF Meeting, Dec. 2002.
[15] S.Zeisberg. Publishable Final Activity Report. PULSERS, Feb. 2006.
[16] 1
st
PULSERS Workshop (hosted at the IST Summit 2005). Dresden (Ger-
many), June 2005.
[17] S.Zeisberg V.Bramkovic. Public project presentation. Integrated PULSERS
Project Phase 2 - Deliverable number: D1.3, Feb. 2006.
[18] J.Barney R.J.Fontana, E.Richley. Commercialization of an Ultra Wideband
Precision Asset Location System. In IEEE Ultra Wideband Systems and Tech-
nologies Conference, Nov. 2003.
[19] G.J.M.Janssen H.Nikookar C.Tiberius Z.Irahhauten, G.Bellusci. Investigation
of UWB Ranging in Dense Indoor Multipath Environments. In 10
th
IEEE
Singapore International Conference on Communication Systems ICCS 2006,
pages 15, Oct. 2006.
[20] C.J.Law W.T.Ng Y.J.Lee Z.N.Low, J.H.Cheong. Pulse detection algorithm
for line-of-sight (LOS) UWB ranging applications. In Antennas and Wireless
Propagation Letters, volume 4, pages 6367. IEEE, 2005.
[21] S.Yamaguchi T.Tanaka. GPS Standard Positioning using Kalman lter. In
SICE-ICASE - International Joint Conference, pages 13511354, Oct. 2006.
[22] L.N.de Rivera J.C.S.Garcia V.I.Ponomaryov, O.B.Pogrebnyak. Increasing the
accuracy of dierential global positioning system by means of use the Kalman
ltering technique. In IEEE International Symposium on Industrial Electronics
- ISIE, volume 2, pages 637642. IEEE, Dec. 2000.
[23] M.Najar E.Medina. High Resolution Location in Ultra Wideband Communi-
cations Systems. In Networking with Ultra Wide Band and Workshop on Ultra
Wide Band for Sensor Networks, pages 4044. IEEE, July 2005.
96
[24] Ubisense (www.ubisense.net). The Ubisense Precise Real-time Location Sys-
tem, 2007.
[25] D.Porcino O.Gremigni. Field Trials of an Accurate UWB Ranging System.
In Seminar on Ultra Wideband Systems, Technology and Applications, pages
169175. IET, April 2006.
[26] H. Holma and A. Toskala. WCDMA for UMTS. John Wiley & Sons, 2000.
[27] N.Whinnett R.Ratasuk X.Weimin R.Kuchibhotla A.Ghosh, R.Love. Overview
of enhanced uplink for 3GPP W-CDMA. In IEEE 59th Vehicular Technology
Conference. VTC 2004-Spring, volume 4, pages 22612265, May 2004.
[28] N.Whinnett R.Ratasuk X.Weimin R.Ratasuk A.Ghosh, R.Love. Overview of
enhanced uplink for 3GPP W-CDMA. In IEEE 60th Vehicular Technology
Conference. VTC 2004-Fall, volume 5, pages 33593363, Sept. 2004.
[29] E.Malkam aki M.Bertinelli. WCDMA ENHANCED UPLINK: HSUPA LINK
LEVEL PERFORMANCE STUDY. In IEEE 16
th
International Symposium on
Personal, Indoor and Mobile Radio Communciations. PIMRC-2005, volume 2,
pages 895899, Sept. 2005.
[30] R. ONeill and L.B. Lopes. Envelope Variations and Spectral Splatter in
Clipped Multicarrier Signals. In Proc. of IEEE PIMRC95, pages 7175, 1995.
[31] M. Honkanen and S.G. Hggman. New Aspects on Nonlinear Power Amplier
Modeling in Radio Communication System Simulations. In Proc. of PIMRC
97, volume 3, pages 844848, 1997.
[32] E. Costa, M. Midrio, and S. Pupolin. Impact of Amplier Nonlinearities on
OFDM Transmission System Performance. IEEE Communications Letters,
3(2), February 1999.
[33] ETSI. Universal Mobile Transmission System (UMTS); UE Radio Transmission
and Reception (FDD). Technical Report TS 25.101 v4.5.0 (2002-06), 3GPP,
2006.
[34] R. Tanner and J. Woodard. WCDMA Requirements and Practical Design. John
Wiley & Sons, 2004.
[35] Agilent. Agilent AN 1335; HPSK Spreading for 3G. Application note.
[36] T.Palenius T.Ottosson. THE IMPACT OF USING MULTICODE TRANS-
MISSION IN THE WCDMA SYSTEM. In 49
th
Vehicular Technolgy Confer-
ence. VTC 1999., volume 2, pages 15501554. IEEE, May 1999.
97
[37] O. Vaan anen, J. Vankka, and K. Halonen. Simple algorithm for peak windowing
and its application in GSM, EDGE and WCDMA systems. IEE Proceedings
Communications, 152(3):357362, June 2005.
[38] S. Marsili. Algorithm for Peak to Average Power Ratio Reduction operating
at Symbol Rate. In Proc. of IEEE Int. Symp. ISCAS 2005, volume 6, pages
54975500, May 2005.
[39] A. Springer, T. Frauscher, B. Adler, D. Pimingsdorfer, and R. Weigel. Impact
of Nonlinear Amplers on the UMTS System. In 6th Int. Symp. on Spread-
Spectrum Tech & Appli., 2000.
[40] R. Van Nee and R. Prasad. OFDM for Wireless Multimedia Communications.
Artech House Publishers, 2000.
[41] X. Li and L.J. Cimini. Eects of Clipping and Filtering on the Performance of
OFDM. IEEE Communications Letters, 2, May 1998.
[42] T. May and H. Rohling. Reducing the Peak-to-Average Power Ratio in OFDM
Radio Transmission Systems. In Proc. of Vehicular Technology Conference
(VTC98), volume 3, pages 24742478, May 1998.
[43] J. Akhtman, B.Z. Bobrovsky, and L. Hanzo. Peak-to-Average Power Ratio
Reduction for OFDM Modems. In Proc. of IEEE Vehicular Technology Con-
ference (VTC 2003), volume 2, pages 11881192, April 2003.
[44] H. Kuo, S.W. Cheung, and Simon S.F. Hau. A Novel Peak-Windowing Tech-
nique for WCDMA Systems. In Vehicular Technology Conference VTC 2006
Spring, volume 4, pages 313316, 2006.
[45] K. Halonen O.Vaan anen, J. Vankka. Eect of Baseband Clipping in Wide-
band CDMA System. In 7
th
International Symposium on Spread-Spectrum
Tech.&Appl. IEEE, Sept. 2002.
[46] R. Muhamed J.G. Andrews, A. Ghosh. Fundamentals of WiMAX: understand-
ing broadband wireless networking. Prentice Hall, 2007.
[47] IEEE. Standard 802.16-2004. part16: Air interface for xed broadband wireless
access systems. Technical report, IEEE, October 2004.
[48] IEEE. Standard 802.16e-2004. part16: Air interface for xed and mobile broad-
band wireless access systems - amendment for physical and medium access con-
trol layers for combined xed and mobile operation in licensed band. Technical
report, IEEE, December 2005.
98
[49] J.G. Proakis. Digital Communications. McGraw-Hill, 4th edition, 2000.
[50] R. van Nee and Arnout de Wild. Reducing the Peak-to-Average Power Ra-
tio of OFDM. In Proc. of IEEE Vehicular Technology Conference (VTC98),
volume 3, pages 20722076, May 1998.
[51] N.J. Fliege. Multirate Digital Signal Processing. John Wiley & Sons, 1994.
[52] Agilent Application Note 1536. Designing and Testing 3GPP W-CDMA User
Equipment.
99