Sie sind auf Seite 1von 5

Multilevel PAM with Optimal Amplitudes for

Non-Coherent Energy Detection


Rainer Moorfeld and Adolf Finger
Communications Laboratory
Dresden University of Technology
01062 Dresden, Germany
Email: {moorfeld, nger}@ifn.et.tu-dresden.de
AbstractThis paper analyzes multilevel pulse am-
plitude modulation (M-PAM) for energy detection
receivers. Because of the asymmetric, non Gaussian
probability density function (PDF) of the energy de-
tected signal, the optimal signal constellation is not
known a priori. In order to achieve optimal symbol
error rate performance using M-PAM, the optimal
signal constellation will be calculated. Finally the SER
performance of the energy detection receiver based
multiband impulse radio UWB architecture within the
IEEE802.15.3a channel model using M-PAM will be
demonstrated.
Index TermsEnergy detection, pulse amplitude
modulation, multiband impulse radio UWB
I. Introduction
Because of the simplicity, non-coherent energy detection
receivers are of various interest. This kind of receivers
are used inter alia to demodulate optical and radio sig-
nals [1], [2]. There has also been interest in applying
energy detection to ultra wideband radios demodulating
multiband impulse radio using on/o-keying modulation
[3], [4]. The multiband impulse radio architecture has
been further investigated in [5], [6], [7]. In this scenario
(based on IEEE802.15.3a UWB channel model in [8]), the
most limiting factor of the non-coherent energy detection
receiver is the channel delay spread. In order to handle
the intersymbol interference a minimal impulse distance T
r
has to be kept. Fig. 1 shows the maximum achievable data
rate for pulse position modulation (PPM) and pulse am-
plitude modulation (PAM) considering a minimum pulse
distance T
r
(equidistant) of 25 ns in case the SNR allows
to demodulate the higher modulation scheme of the PAM.
It is not possible to go below a minimum pulse distance
(depending on the channel) without increasing the inter-
symbol interference. Note that the data rates for PPM in
Fig. 1 are pessimistic, as ISI will limit the performance of
PPM regarding the energy detection receiver.
In order to provide the full potential out of an energy
detection receiver based transmission system, the signal
constellation has to be optimized to get minimal symbol
error rate (SER) performance. Gaussian approximation
of the
2
distributed decision variables at the output of the
energy detection receiver has often been used to simplify
the error rate analysis and threshold calculation. However,
this is not suitable for a multiband impulse radio UWB
1 2 3
0
20
40
60
80
100
120
bit/symbol
d
a
t
a

r
a
t
e

[
M
b
i
t
/
s
]


PPM
PAM
Figure 1. Maximum achievable data rate for PPM and PAM with
a minimum xed pulse distance of 25 ns.
system where the degree of freedom can be relatively low
(< 5). Thus, the optimization in this paper based on the

2
distributed decision variables in order to get minimal
SER performance.
Multilevel PAM with optimized amplitudes allows to
increase the data rate in good channel conditions. This
work is not limited to multiband impulse radio UWB.
Every impulse based transmission system with an en-
ergy detection receiver, even optical transmission systems
transmitting data through air or ber and photodetecting
the laser beam have to deal with similar aspects.
The papers structure is: Section II describes the system
model and the decision statistics for the non-coherent
energy detection receiver. In Section III the symbol error
rate for multilevel PAM using an energy detector has
been calculated. In section IV the amplitudes for optimal
symbol error rate performance are calculated. In section
V the performance of the optimized PAM is demonstrated
in the IEEE802.15.3a channel model. Finally, section VI
concludes the paper.
II. System model
The basic model for this paper is the multiband impulse
radio architecture [3]. Without loss of generality only one
subband will be considered in the following sections. The
transmitter uses energy weighted pulses to transmit the
data. The modulated pulse stream is
978-1-4244-5668-0/09/$25.00 2009 IEEE
s(t) =

n=
w
n
_
E
max
p(t nT
r
), (1)
where w
n
[0, 1] and w
n
< w
n+1
is the relative
amplitude of symbol n, E
max
is the maximal symbol
energy and p(t) a cosine shaped pulse [6]. The average
symbol energy is:
E
S
=
E
max
M
M1

n=0
w
2
n
. (2)
The received waveform is
y(t) = h(t) s(t) +n(t), (3)
where h(t) is the channel impulse response and n(t) is
additive white Gaussian noise with the two-sided spectral
density of N
0
/2. The optimal amplitudes will be calculated
using the AWGN channel. But in the simulations, the
IEEE 802.15.3a channel model described in [9] reproducing
an indoor environment with a range of a few meters, will
be used.

I
T
0
( )
2
BP
Figure 2. Non-coherent energy detector.
The receiver is a simple non-coherent energy detector
(Fig. 2). The integrator output r is
r =
_
TI
0
[y(t )g()]
2
dt, (4)
where T
I
is the integration time and g(t) is the impulse
response of the receiver lter with bandwidth B. The
SNR is dened as E
S
/N
0
. The decision statistic R can
be written as [1], [10]:
R =
1

N
0
B
2BTI

i=1
s
_
i
2B
_
+n
_
i
2B
_
. (5)
If s
i
= 0, i, R is a sum of squares of 2BT
I
Gaussian
variates with zero mean and unit variance. Thus R follows
a central
2
distribution with 2BT
I
degrees of freedom.
If s
i
= 0, the test decision statistic R follows a non-
central
2
distribution with 2BT
I
degrees of freedom and
the non-centrality parameter 2E
S
/N
0
. To calculate the
symbol error rate (SER) the maximum likelihood decision
criterion and multiple hypothesis testing has been used
[11]. The decision hypothesis H
m
(m = 0, 1, ..., M 1)
that a symbol w
n
was transmitted if
p(r|H
m
) > p(r|H
k
) for all k = m, (6)
where p is the probability density function (PDF) of R:
p(r|H
m
) =
_
1
N0(L)
(
r
N0
)
L1
e

r
N
0 m = 0
1
N0
(
r
n
)
L1
2
e

r+n
N
0
I
L1
_
2

rn
N0
_
m > 0,
(7)
where
n
= w
2
n
E
max
is the symbol energy of symbol n,
is the Gamma function and I

is the -th order modied


Bessel function of the rst kind. Without loss of generality,
the time bandwidth product L = BT
I
is chosen to be an
integer.
III. Symbol error rate in the AWGN channel
A. Optimal decision criterion
The optimal decision criterion to nd the thresholds
m
is the solution of
p(r|H
m
) = p(r|H
m+1
)|
r=m
m = 0, 1, ..., M 2. (8)
Because there does not exist a closed form expression
isolating
m
in 8, having the target PDFs of 7 in mind,

m
has been calculated using the bisection method [12].
B. Symbol error rate
The symbol error rate P
e
of the M-ary PAM in the
AWGN channel is averaged over the probability of error
for all possible symbols [13]:
P
e
= 1 P
c
= 1
M1

m=0
P(r|H
m
)P(H
m
), (9)
where P(r|H
m
) is the conditional probability of a correct
decision. All symbols are equally likely to be transmitted.
Thus P(H
m
) = 1/M for all m. Depending on m, there are
three dierent decision areas. The conditional probability
of correct decision is [11]:
P(r|H
m
) =
_

_
m
_
0
p(r|H
m
) dr for m = 0,
m
_
m1
p(r|H
m
) dr for 0 < m < M 1

_
m1
p(r|H
m
) dr for m = M 1.
(10)
Using the conditional cumulative distribution function
(CDF) [13]:
F
R
(y) =
_
m
0
p(y)dy = 1 Q
D
__
2
m
N
0
,
_
2
m
N
0
_
, (11)
in (10) to evaluate (9), the SER for multilevel PAM using
an energy detector in the AWGN channel is [14]:
P
e
=
1
M
_
M 1
+
M2

m=0
Q
D
_
_

2w
2
m
E
max
N
0
,
_
2
m
N
0
_
_

M1

m=1
Q
D
_
_

2w
2
m
E
max
N
0
,
_
2
m1
N
0
_
_
_
. (12)
IV. Optimal PAM for energy detection
receivers
Due to the energy detection receiver, the pulse ampli-
tudes w
n
for optimal SER performance are not known.
The PDFs for this receiver architecture are not Gaussian
(see (7)). The presented algorithm based on a minimum
search of the constrained SER function, calculating the
optimal amplitudes.
A. Equidistant symbol amplitudes
Table I shows the equidistant amplitudes for OOK, 4-
PAM and 8-PAM.
w
0
w
1
w
2
w
3
w
4
w
5
w
6
w
7
OOK 0 1
4-PAM 0 1/3 2/3 1
8-PAM 0 1/7 2/7 3/7 4/7 5/7 6/7 1
Table I
Equidistant amplitudes for OOK, 4-PAM and 8-PAM
Fig. 3 shows the SER performance of OOK, 4-PAM and
8-PAM in the AWGN channel. The lter bandwidth is
B=250 MHz and the integration time T
I
=25 ns ( M
5). The pulse repetition period T
r
is greater than the pulse
duration, avoiding intersymbol interference. Compared to
OOK, the 4-PAM needs a 4.5 dB higher SNR and the 8-
PAM needs a 9 dB higher SNR to reach a BER of 10
3
.
Fig. 4 points out the SER P
e
against w
2
2
and w
2
3
on basis
of 4-PAM. w
2
1
is set to 0, w
2
4
is set to 1 and w
2
m
< w
2
m+1
.
A unique minimum of the SER can be found for specic
values of E
2
= w
2
2
and E
3
= w
2
3
.
0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
S
E
R


SNR [dB]
OOK
n = 4 equidistant
n = 4 optimal
n = 8 equidistant
n = 8 optimal
Figure 3. Performance of OOK, 4-PAM, and 8-PAM with equidis-
tant and optimal amplitudes (AWGN channel).
B. Optimal symbol amplitudes
To solve the optimization problem the SER has to be
minimized with respect to the chosen symbol amplitudes
w
n
:
minimize P
e
(w
n
,
n
( w
n
))
subject to 0 < w
n
< 1

n
( w
n
).
E3
E
2
0 0.2 0.4 0.6 0.8 1
10
3
10
2.5
10
2
10
1.5
10
1
10
0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Figure 4. SER versus energy values En = w
2
n
(n = 4, SNR=15 dB,
M = 5).
Constrained nonlinear optimization has been used to cal-
culate the unique minimum. The observation of Fig. 4
shows that a unique minimum exists for 0 < w
n
< 1.
The numeric optimization algorithm looks as follows:
Input: start values for w
n
Output: optimal w
n
begin
set w
n
;
while

P
e
= do
calculate
n
(w
n
);
calculate P
e
(w
n
,
n
(w
n
));
search new values for w
n
;
end
end

P
e
is the rst derivative of P
e
with respect to w
n
. For
0 a unique minimum for P
e
can be found. For the
search of new values for w
n
, the optimization has been
numerical computed, because a closed form expression for
w
n
does not exist. The optimal amplitudes depend on the
chosen SNR. The results of the optimization algorithm
for dierent SNRs for 4-PAM are displayed in Table
II. The optimal energy values for 20 dB show the best
SNR [dB] w
0
w
1
w
2
w
3
5 0 0.2753 0.6633 1
10 0 0.3831 0.6619 1
15 0 0.3805 0.6815 1
20 0 0.3615 0.6784 1
Table II
Optimal amplitudes for different SNR (4-PAM) for M=5.
overall performance. These values are used in the following
performance comparison. Fig. 3 shows the performance of
the optimal amplitudes vs. the equidistant amplitudes in
the AWGN channel. The achievable gain is about 0.5 dB.
V. BER performance in IEEE802.15.3a channel
In this section, the performance of the new mod-
ulation scheme is shown in the UWB fading chan-
nel (IEEE802.15.3a), reproducing an indoor environment
(LOS and NLOS) with a range of a few meters (0-10 m).
This channel model represents the application scenario for
the high data rate, short range point to point communi-
cation.
The SER simulations use the 90 best channel impulse
responses out of the 100 standard impulse responses pro-
vided by the IEEE task group. All symbols are Grey
coded. The integration time T
I
is chosen as a static
optimum over all 100 channel impulse responses and diers
for the channel model CM1 to CM4. Thus, the optimal
amplitudes are dierent for the channel models and have
to be adjusted once the integration time or the subband
bandwidth has changed. The subband bandwidth is xed
in the following simulations. Table III shows the optimal
integration time T
I
for the dierent channel models and
the optimal amplitudes calculated for E
S
/N
0
of 20 dB.
The optimal T
I
is identied by SER simulations. The
channel model T
I
[ns] M wn
CM1 22 5 [0, 0.3576, 0.6733, 1]
CM2/CM3 28 6 [0, 0.3618, 0.6785, 1]
CM4 84 20 [0, 0.3867, 0.6920, 1]
Table III
Optimal amplitudes (4-PAM) for different integration times
T
I
for IEEE802.5.13a channel model.
results show the single band performance of a 250 MHz
subband with the center frequency of 3.225GHz. The
transmitted pulse is a cosine shaped pulse with a -10 dB
bandwidth of 200 MHz. An elliptic bandpass lter has
been used with a stopband attenuation of 50 dB and
passband ripple of 2 dB. The lter order is 2. The whole
system uses 24 subbands within the frequency range of
3.1 GHz to 10.6 GHz. The lter and pulse conguration
guarantee a minimum intersubband interference [6]. Thus,
the multiband performance is nearly the same as the single
band performance.
A. Channel model CM1
The channel model CM1 represents a LOS scenario with
a range of 0-4 m. The average delay spread
RMS
is 5 ns.
Fig. 5 displays the simulation results for OOK and 4-PAM
in channel model CM1 compared to the AWGN channel.
T
r
is set to 75 ns in order to avoid ISI. The eect of fading
in the LOS channel is not strong, thus the performance is
nearly identical to the AWGN performance. On average
23 paths contain 85% of the signal energy.
B. Channel model CM2
The channel model CM2 describes the same range as
CM1 but represents a NLOS scenario. The average delay
spread
RMS
is 8 ns. Fig. 6 displays the simulation results
for OOK and 4-PAM in CM2. T
r
is set to 100 ns in
0 2 4 6 8 10 12 14 16 18 20 22
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR [dB]
S
E
R


OOK AWGN
OOK CM1
4PAM AWGN
4PAM CM1
Figure 5. Performance of OOK and 4-PAM in CM1 (M = 5).
0 2 4 6 8 10 12 14 16 18 20 22
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR
S
E
R


OOK AWGN
OOK CM2
4PAM AWGN
4PAM CM2
Figure 6. Performance of OOK and 4-PAM in CM2 (M = 6).
order to avoid ISI. The multipath spans over several
nanoseconds and on average 35 paths contain 85% of the
signal energy. Thus it is harder to collect a siginicant
amount of signal energy and the SER performance is worse
than the performance in the AWGN channel. Although T
r
is set to 100 ns, 4-PAM is more sensitive against ISI than
OOK.
C. Channel model CM3
The channel model CM3 represents a NLOS scenario
with a range of 4-10 m. The average delay spread
RMS
is
15 ns. Fig. 7 displays the simulation results for OOK and
4-PAM in CM4. T
r
is set to 100 ns in order to avoid ISI. On
Average 65 paths contain 85% of the signal energy. Thus
it is harder for the energy collector to get the signicant
signal energy. Compared to CM2, the performance is
slightly worse.
D. Channel model CM4
Finally CM4 describes an NLOS environment with
strong delay dispersion, resulting in a delay spread of
25 ns. This channel model represents an extreme multipath
channel. Fig. 8 displays the simulation results for OOK
and 4-PAM in CM4. T
r
is set to 150 ns in order to avoid
ISI. On average 123 paths contain 85% of the signal energy.
0 2 4 6 8 10 12 14 16 18 20 22
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR [dB]
S
E
R


OOK AWGN
OOK CM3
4PAM AWGN
4PAM CM3
Figure 7. Performance of OOK and 4-PAM in CM3 (M = 6).
0 2 4 6 8 10 12 14 16 18 20 22
10
6
10
5
10
4
10
3
10
2
10
1
10
0
SNR [dB]
S
E
R


OOK AWGN
OOK CM4
4PAM AWGN
4PAM CM4
Figure 8. Performance of OOK and 4-PAM in CM4 (M = 20).
Again the results show that 4-PAM is more sensitive
against ISI.
Table IV shows the achievable data rates in the dierent
channel models and the necessary SNR to reach a SER of
10
3
.
channel model modulation E
b
/N
0
[dB] data rate [Mbit/s]
CM1 OOK 12 26.7
4-PAM 16 53.4
CM2 OOK 12 (14) 20 (26.7)
4-PAM 16 40.0
CM3 OOK 14 13.3
4-PAM 17 20.0
CM4 OOK 14 (15) 8.0 (10.0)
4-PAM 17 (18) 13.3 (26.7)
Table IV
Achievable data rates for OOK and 4-PAM for the
IEEE802.5.13a channel model for a SER of 10
3
VI. Conclusion
The paper presents the optimal amplitudes for mul-
tilevel PAM using an energy detection receiver in the
AWGN channel. The theoretical results in chapter III
show that the gain in the symbol error rate is not great.
This small advantage shown in the AWGN channel is also
achievable in the IEEE802.15.3a channel model. Thus,
increasing the data rate using multilevel PAM (either
equidistant and optimal amplitudes) is possible under
good channel conditions. Further investigations regarding
energy detection will look into analytical calculations of
optimal amplitudes for multilevel PAM based on SER
in fading channels combined with a multistep integrator
using dierent combining schemes.
Note that the simulation results show the performance
of one single subband. A multiband impulse radio archi-
tecture with the non-coherent energy detector and 12 to 24
active subbands can achieve a data rate up to 1.2 Gbit/s
using 4-PAM.
Acknowledgment
The authors gratefully acknowledge that their work is
supported within the priority program No. 1202 (UKoLoS)
by the German Research Foundation (DFG).
References
[1] H. Urkowitz, Energy detection of unknown deterministic sig-
nals, Proceedings of the IEEE, vol. 55, no. 4, pp. 523531, April
1967.
[2] P. A. Humblet and M. Azizoglu, On the bit error rate of
lightwave systems with optical ampliers, IEEE/OSA Journal
of Lightwave Technology, vol. 9, no. 11, pp. 15761582, Nov.
1991.
[3] S. Paquelet, L. M. Aubert, and B. Uguen, An Impulse Radio
Asynchronious Transceiver for High Data Rates, in Proc. of
Joint UWBST & IWUWBS, Kyoto, May 2004.
[4] S. Paquelet and L. Aubert, An Energy Adaptive Demodulation
for High Data Rates with Impulse Radio, in Proc. of RAW-
COM, Atlanta, Sep. 2004.
[5] L.-M. Aubert, Mise en place dune couche physique pour les
futurs systemes de radiocommunications hauts debits UWB.
These docteur, 2005.
[6] M. Mittelbach, R. Moorfeld, and A. Finger, Performance of a
multiband impulse radio uwb arcchitecture, in Proc of 3rd IEE
Mobility Conference, Bangkok, Thailand, October 2006.
[7] R. Moorfeld, H. Dehner, H. Jkel, F. Jondral, and A. Fin-
ger, A simple and fast detect and avoid algorithm for non-
coherent multiband impulse radio UWB, in Proc. of 10th IEEE
International Symposium on Spread Spectrum Techniques and
Applications (ISSSTA), Bologna, Italy, August 2008.
[8] A. F. Molisch, J. R. Foerster, and M. Pendergrass, Channel
models for ultrawideband personal area networks, IEEE Wire-
less Communications Magazine, vol. 10, no. 6, pp. 1421, Dec.
2003.
[9] J. R. Foerster, M. Pendergrass, and A. Molisch, A channel
model for ultrawideband indoor communication, November
2003, mitsubishi Eletric Research Laboratory, Inc., TR-2003-73.
[10] F. F. Digham, M. S. Alouini, and M. K. Simon, On the energy
detection of unknown signals over fading channels, in Proc.
IEEE International Conference on Communications ICC 03,
vol. 5, 1115 May 2003, pp. 35753579.
[11] S. M. Kay, Fundamentals of Statistical Signal Processing: De-
tection Theory. Prentice Hall, 1998, vol. 2.
[12] P. Theis, Untersuchungen zur Modulation fr nicht kohrente
Energiedetektoren, Masters thesis, University of Technology,
Dresden, 2008.
[13] J. G. Proakis, Digital Communications, Fourth Edition.
McGraw-Hill Series in Electrical and Computer Engineering,
2001.
[14] A. Anttonen, A. Mammela, and A. Kotelba, Sensitivity of
energy detected multilevel pam systems to threshold mismatch,
in Proc. IEEE International Conference on Ultra-Wideband
ICUWB 2008, vol. 1, 1012 Sept. 2008, pp. 165168.

Das könnte Ihnen auch gefallen