Tài liệu Adaptive thu phát không dây P7 pdf

L7/
Practical Considerations
of
Wideband
AQAM
In deriving the upper bound performance of wideband AQAM portrayed in Figure 4.21(b),
various assumptions were made and stated in Section
4.3.1.
However, in order to provide a
more accurate comparison between AQAM and its constituent fixed modulation modes, those
assumptions must be justified and their effects have to be investigated. Specifically, perfect,
i.e. error-free feedback was assumed for the DFE, while in practice erroneous decision can be
fed back, which results in error propagation. Consequently the impact of error propagation
is studied in the context of both fixed and adaptive QAM schemes. Furthermore, as stated in
Section 4.3.1, perfect modulation mode selection was assumed, whereby the output
SNR
of
the DFE was estimated perfectly prior to transmission. However, in stipulating this assump-
tion, the delay incurred between channel quality estimation and the actual utilization of the
estimate was neglected in the wideband AQAM scheme.
In this chapter the impact of co-channel interference on the wideband AQAM scheme
is
also investigated. In this respect, interference compensation techniques are invoked in order
to reduce the degradation resulting from the co-channel interference. Let us now commence
our investigations by studying the error propagation phenomenon in the DFE.
7.1
Impact
of
Error
Propagation
Error propagation is a phenomenon that occurs, whenever an erroneous decision is fed back

into the feedback filter
of
the DFE. When a wrong decision is fed back, the feedback filter
produces an output estimate which is erroneous. The incorrect estimate precipitates further
errors at the output of the equalizer. This leads to another erroneous decision being fed
back into the feedback filter. Consequently, this recursive phenomenon degrades the BER
performance of the DFE. Intuitively, the effects of this error will last throughout the memory
span of the feedback filter. This causes an error propagation throughout the feedback filter,
until the memory
of
the feedback filter is cleared of any erroneous feedback inputs.
The performance of the fixed modulation modes of our AQAM scheme in conjunction
257
Adaptive Wireless Tranceivers
L. Hanzo, C.H. Wong, M.S. Yee
Copyright © 2002 John Wiley & Sons Ltd
ISBNs: 0-470-84689-5 (Hardback); 0-470-84776-X (Electronic)
258
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
Transmission Burst type:
Non-Spread
Speech Burst
of Figure 4.13.
DFE
Parameters:
No.

of feedforward taps,
Nf
7
No.
of feedback taps,Nb
35
See Figure 4.12 and Typical Urban Rayleigh-faded Weights
Channel Parameters:
8
Number of RKCE taps
0
Initial Channel Estimate Vector
h(o)
2
300
6
(see Equation 3.47)
0.011
System error covariance Matrix,
Q(k)
0<g11
Measurement error covariance Matrix,
R(k)
=
g1
Recursive Kalman Channel Estimator Parameters:
Past Decision
Decision Feedback
Normalized Doppler Frequency: 3.25
x

10-5
Table
4.5
Table 7.1:
Generic simulation parameters that were utilized in
our
experiments.
with error propagation is depicted in Figure 7.1, where the corresponding curve of the error-
free feedback scenario is also displayed for comparison. Perfect channel compensation was
applied at the receiver and the other simulation parameters are listed in Table 7.1. There
was only a slight degradation in the BER performance of the BPSK and 4QAM modes,
as
evidenced by Figure 7.1. However, for the higher-order modulation modes of 16QAM and
64QAM, a more severe degradation
of
approximately
1.5
and 3.0dB was recorded, respec-
tively. These results were expected, since the higher-order modulation modes were more
susceptible to feedback errors due to the smaller Euclidean distance of their constellation
points.
The impact of error propagation
on
the wideband AQAM scheme over a TU Rayleigh fad-
ing channel was also investigated and the results are shown in Figure 7.2. The corresponding
curve of the wideband AQAM scheme with error-free decision feedback was also shown for
comparison and the switching thresholds of the wideband AQAM scheme were set according
to Table
4.8
for target BERs of

1%
and
0.01%.
At low to medium average channel SNRs the
BER performance of the wideband AQAM scheme exposed to error propagation was similar
to that of the AQAM scheme with error-free decision feedback. However, at higher average
channel SNRs, as a result of error propagation, a BEWSNR degradation of approximately
3dB was observed. These results were consistent with the results shown for the fixed mod-
ulation modes of Figure 7.1. At low to medium average channel SNRs, the impact of error
propagation was negligible due to two factors. Firstly, at those channel SNRs the lower-order
modulation modes, which were more robust against error propagation were utilized more
frequently. Secondly, the higher-order modulation modes were only utilized, when the chan-
nel quality was favourable, which resulted in low instantaneous BERs. Consequently, less
erroneous decisions were made, which reduced the impact of error propagation.
However, at higher average channel SNRs, the probability of modulation mode switching
7.2.
CHANNEL QUALITY ESTIMATION LATENCY
259
10.'
W
p!
m
1
o-~
1
rr-6
Channel
SNR(dB)
Figure
7.1:

Impact of error propagation
on
the modulation modes of
BPSK,
4QAM, 16QAM and
64QAM over the TU Rayleigh fading channel of Figure 4.12. Perfect channel compen-
sation was applied and the simulation parameters are listed in Table
7.1.
was low, where the 64QAM mode was frequently chosen. Consequently, the impact of error
propagation was more apparent, as it was observed the case
of
the fixed modulation mode of
64QAM in Figure
7.1.
Nevertheless, the target performance of
1%
and
0.01%
was achieved
even in the presence of erroneous decision feedback.
7.2
Channel Quality Estimation Latency
The estimation of the channel quality prior to transmission is vital in the implementation of
the wideband AQAM scheme, since it is used in the selection of the appropriate modulation
mode for the next transmission burst. In generating the upper bound performance curves
depicted in Figure 4.21(b), we assumed that the required modulation mode was selected per-
fectly prior to transmission, as stated in Section 4.3.1. However, in a realistic and practical
wideband AQAM scheme this assumption must be discarded as a result
of
the inherent chan-

nel quality estimation delay incurred by the scheme. Nevertheless, it must be stressed that
260
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
1
o-2
IO-^
S
2
5
2
d
p32
IO-^
1
o-6
W5
S
2
5
0
1
%
BER,
0
0.01
%

BER
-
Error
Prop.
.
.
. . .
.
.
Perfect
0
5
10
15 20 25
30
35
40
Channel
SNR(dB)
Figure
7.2:
Impact
of
error propagation on the wideband
AQAM
scheme over a TU Rayleigh fading
channel, where the switching thresholds were set according to Table
4.8
for target
BERs

of
1%
and
0.01%.
Perfect channel compensation was applied and the simulation parameters
are listed
in
Table
7.1.
the assumption was essential in order to record the upper bound performance of the
AQAM
scheme.
The channel quality estimation latency is defined as the delay incurred between the event
of estimating the channel quality to the actual moment of transmission using the modem mode
deemed optimum at the instant of the channel quality estimation. During this delay, the fad-
ing channel quality varies according to the Doppler frequency and consequently, the channel
quality estimates perceived prior to transmission may become obsolete. Consequently, the
chosen modulation mode is not optimum with regards to the actual channel quality and this
degrades the BER performance of the wideband
AQAM
scheme. This degradation is de-
pendent on the amount
of
delay incurred and the rate at which the fading channel quality
fluctuates, as quantified by its Doppler frequency. Before we proceed to investigate the per-
formance degradation as a result of the channel quality estimation latency, let us present two
possible time-frame structures, where wideband
AQAM
can be implemented. This will pro-
vide us with a clearer understanding concerning the amount of delay incurred by the scheme.

7.2.
CHANNEL OUALITY ESTIMATION LATENCY
261
7.2.1 Sub-frame Based Time Division Dupleflime Division Multiple
Access System
In this sub-frame based Time Division Duplexmime Division Multiple Access (TDD
/
TDMA)
system, the uplink and downlink time-slots
are
separated equally into two halves of the
TDMA frame, as shown in Figure 7.3. In this respect the time-slot is defined as the window in
time, in which the transmission burst is received or transmitted. By utilizing the time-frame
configuration shown in Figure 7.3, we will explain the operation of the wideband AQAM
scheme and the corresponding channel quality estimation latency that is incurred. In the up-
link transmission, shown in Figure 7.3, the channel quality was estimated at the Base Station
(BS) and subsequently an appropriate modulation mode was selected for its next downlink
transmission. This was achieved by exploiting the channel’s reciprocity during the uplink
and downlink transmissions, since the transmission frequencies for both links were identical
in a TDD system. Having selected the modulation mode, a delay
of
half a TDMA frame
was incurred at the BS before the downlink transmission was activated
as
shown in Figure
7.3. We refer to this regime as open-loop controlled AQAM. Let
us
now in the next section
consider closed-loop control.
7.2.2 Closed-Loop Time Division Multiple Access System

The corresponding closed-loop TDMA construction was similar to that of the sub-frame
TDDRDMA with the exception that the uplink and downlink transmission frequencies were
different. Hence this was a Frequency Division Duplex (FDD) system. Consequently, the
assumed channel reciprocity
-
which was invoked in the sub-frame based TDDmDMA sys-
tem
-
was less applicable. Hence a closed-loop signalling system was required in order to
implement the wideband AQAM scheme, which is shown in Figure 7.4. In the uplink trans-
mission, the channel quality was estimated at the BS, in order to select the next uplink modu-
lation mode. Subsequently, the selected uplink modulation mode was conveyed to the Mobile
Station (MS) with the aid of control symbols during the next downlink transmission. Conse-
quently, the selected modulation mode was utilized by the MS in its next uplink transmission.
As a result of the closed-loop signalling regime, the delay incurred by the system was equal
to the duration of one TDMA time-frame. Consequently, the open-loop system described in
Section 7.2.
l
was more applicable to AQAM transmission as a result
of
its lower delay, when
compared to the close-loop system. This latency can be substantially reduced using slot-by-
slot TDDEDMA, where the uplink and downlink slots are adjacent, which is also supported
by the third-generation Universal Mobile Telecommunication System (UMTS) [221].
7.2.3 Impact
of
Channel Quality Estimation Latency
Regardless of the type of wideband AQAM scheme that was implemented, we investigated
the maximum delay that could be tolerated by the AQAM scheme by assuming that the per-
formance degradation in the uplink and downlink transmission was identical. In our experi-

ments, the delay was measured in terms of a time-slot duration of
72ps,
as proposed in the
Pan European FRAMES framework
[l5
l]. Mid-amble associated CIR estimation based on
the Kalman algorithm
-
which was discussed in Chapter 3
-
was implemented, in order to es-
timate the channel quality. The normalized Doppler frequency was set to
3.25
x
lop5,
which
262
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AOAM
Downlink Transmission
I
TDMA Frame
I
W
MS
Uplink
-

-
Downri~k
Band
Band
'\\
Select Activate
,l'
Mode Mode
,,'
Modulation Modulation,'
I
W
,'
Delay
=
Frame
._
__ __
Half
aTDMA
Figure
7.3:
Sub-frame based
TDD/TDMA
system
for
the uplink and downlink transmission, as de-
scribed in Section
7.2.1.

Channel reciprocity was exploited in this system and the channel
quality estimation latency was equivalent to half a
TDMA
frame.
was equivalent to a TDMA system using a 1.9GHz in carrier frequency, transmission rate
of
2.6
MSymbols/s and a vehicular speed
of
13.33ds. The specific simulation parameters
used in our subsequent experiments are listed in Table 7.1. The AQAM switching thresholds
were set according to Table 4.8, which were optimised for maintaining target BERs of
1%
and 0.01%.
The results of our investigations are shown
in
Figures 7.5(a) and
73b)
for target BERs
of 1% and 0.01%, respectively. In these figures the wideband AQAM scheme was subjected
to a delay of
8,
16
and
32
time-slots and the performance was compared to that of the zero-
delay upper bound performance. For the target BER of
1%
we can observe that the BER
performance degradation increased, as delay was increased as evidenced by Figure 7.S(a). At

high average channel SNRs, the BER degradation was minimal as a result of the reduction
of modulation mode switching frequency, where the 64QAM mode was frequently selected.
The BER degradation was more evident for the AQAM scheme designed
for
a low target BER
of 0.01% as a result of its increased sensitivity to errors. By referring to Figure 7.S(b), at a
7.2.
CHANNEL QUALITY ESTIMATION LATENCY
263
Uplink
Transm:lssion ‘
_
.
Downlin~ Transmission
1
TDMA Eqme
*
D
Uplink
Dowili.qk Activate
Band
= =
Band
‘\,
*
Modulation
‘\
Mode
t
I1

II
II
l
Select Signal
Modulation Modulation.”
Mode Mode
,,‘
4
D
Delay
e~
4-
TDMA Frame
_-
Figure
7.4:
Closed-loop FDD/TDMA system for the uplink and downlink transmission, as described
in Section
7.2.2.
Channel reciprocity was not assumed in this system in favour
of
a closed-
loop signalling regime and the channel quality estimation latency was equivalent
to
the
duration of one TDMA frame.
channel SNR of 20dB and at a delay of
32
time-slots, the BER performance was degraded by
approximately two orders of magnitude

in
comparison to the upper bound performance.
In
these experiments, the modulation mode selection regime was affected by the delay incurred
by the system. The impact was especially significant, when the channel quality was low and
a less robust higher-order modulation mode was utilized erroneously. The BPS performance
in Figures 7.5(a) and 7.5(b) remained unchanged for different delays. This can be readily
explained by observing that
on
average the throughput was the same even
if
the modulation
mode selected was erroneous.
As discussed previously, the performance of the wideband AQAM scheme depended
on
the channel quality estimation delay incurred, as well as
on
the Doppler frequency of the
fading channel. In order to investigate the system’s performance dependency
on
the Doppler
frequency, a slower fading channel having a normalized Doppler frequency of
2.17
x
IOp6
was utilized. This corresponded to a carrier frequency of
l.SGHz,
transmission rate of
2.6
Msymbols/s and a pedestrian speed of

0.89ds
in the Pan European FRAMES Proposal
[
15
1
1.
264
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
IOU
6
10
'
5
IO
4
e!
210'
m
3a
m
IOJ
10
I
IOh
0
0

5
10
15
20
25
30
35
40
Channel
SNR(dB)
(a)
Performance at a target
BER
of
1%
at channel quality estimation delays
of
8,
16
and
32
time-slots, where each time-slot
is
of
72ps
duration.
6
5
4
32

m
2
I
''.'l!
10
l5
20
25
30
35
y2
Channel
SNR(dB)
(b)
Performance at a target
BER
of
0.01%
at channel quality estimation
delays
of
8,
16
and
32
time-slots, where each time-slot is
of
72ps
duration.
Figure

7.5:
Impact
of
channel quality estimation latency upon the wideband
AQAM
scheme, where
the modem mode switching thresholds were set according to Table
4.8.
The normalized
Doppler frequency was set to
3.25
x
and the other simulation parameters are listed
in Table
7.1.
7.2.
CHANNEL OUALITY ESTIMATION LATENCY
265
The other simulation parameters were set according to Table 7.1. The BER and BPS perfor-
mances of the AQAM scheme over this slower fading channel
are
shown in Figures 7.6(a) and
7.6(b) for
a
target BER of
1%
and
0.01%,
respectively. In these figures, the characteristics
observed in Figures 7.5(a) and 7.5(b) were also evident and hence the associated trends can be

explained similarly. However. in order to investigate the impact of the Doppler frequency, the
BER performance at an average channel SNR over the two fading channels exhibiting differ-
ent Doppler frequencies were recorded against different delays in Figures
7.7(a)
and 7.7(b).
For
a
target BER of
1%
a
higher BER degradation was experienced by the higher Doppler
frequency scheme, where at
a
BER of
2
x
lop2
the lower Doppler frequency scheme can
tolerate an additional delay of
7
time-slots,
as
evidenced by Figure 7.7(a). Similarly, at
a
BER of
1
x
for the scheme having
a
target BER of

0.01%,
an additional
5
time-slots
delay can be tolerated by the scheme with the lower Doppler frequency.
From the above experiments, we can conclude that
as
the channel quality estimation
de-
lay and Doppler frequency increased, the performance degradation of the wideband AQAM
scheme was higher. Furthermore, the impact of channel quality estimation latency was more
evident at low target BERs due to its increased error sensitivity.
In
order to improve the ro-
bustness of the AQAM scheme against channel quality estimation delay, in the next section
we will invoke
a
simple channel quality prediction method and experimentally optimise the
modem mode switching thresholds.
7.2.4
Linear Prediction
of
Channel Quality
In order to mitigate the effects of channel quality estimation delay on the wideband AQAM
scheme, the next channel quality estimate can be predicted using linear prediction. This sim-
ple technique utilizes the previous channel estimates for linear prediction, in order to predict
the next channel quality estimate. Subsequently, if the prediction is accurate, the modulation
mode selection errors will decrease, yielding
a
more delay-robust wideband AQAM scheme.

This linear prediction technique was applied to the wideband AQAM scheme in conjunction
with two different Doppler frequencies and various time delays for target BERs of
1%
and
0.01%.
The results are depicted in Figures 7.8(a) and 7.8(b) for an average channel SNR of
20dB,
where the performance without linear prediction is also shown for comparison. In these
figures, the linearly predictive scheme exhibited
a
higher tolerance against channel quality es-
timation delay. The maximum delays that can be tolerated for
a
target BER of
1%
and
0.01%
are tabulated in Table
7.2
for the schemes with and without linear prediction. From this table,
channel quality estimation delay gains of approximately
8
time-slots can be achieved using
the above linear predictive techniques for the lower Doppler frequency scheme. Similarly,
delay gains of
6
time-slots were recorded for the higher Doppler frequency scheme.
In
these experiments we have highlighted that
a

simple channel quality prediction tech-
nique can substantially improve the robustness of the wideband AQAM scheme against chan-
nel quality estimation delay. However, it must be stressed that the AQAM scheme performed
better in
a
slowly varying environment, which also facilitated
a
better channel prediction
performance.
266
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
1"
0
5
IO
15
20
25
30
35
40'
Channel
SNR(dB)
(a)
Performance at a target
BER

of
1%
at channel quality estimation delays
of
8,
16
and
32
time-slots, where each time-slot
is
of
72ps
duration.
IO2
a:
i&
10"
IO4
32
m
2
1
n
0
5
IO
15
20 25
30
35

40'
Channel
SNR(dB)
(b)
Performance at a target
BER
of
0.01%
at channel quality estimation
delays
of
8,
16
and
32
time-slots, where each time-slot is
of
72ps
duration.
Figure
7.6:
Impact
of
channel quality estimation latency upon the wideband
AQAM
scheme, where
the modem mode switching thresholds were set according
to
Table
4.8.

The normalized
Doppler frequency was set to
2.17
x
and the other simulation parameters are listed
in Table
7.1.
7.2. CHANNEL OUALITY ESTIMATION LATENCY 267
(a)
Performance at a target BER
01
1%
for different channel quality esti-
mation delays in terms of time-slots
(TS),
where each time-slot is of
72ps
duration.
(b)
Performance at a target BER
of
0.01%
for different channel quality es-
timation delays in terms of time-slots
(TS),
where each time-slot
is
of
72ps
duration.

Figure
7.7:
Impact
of
channel quality estimation latency upon the wideband
AQAM
scheme for two
different normalized Doppler frequencies
of
3.25
x
(at 13.3ds) and 2.17
x
1OP6(at
0.89m/s),
where the modem mode switching thresholds were set according to Table
4.8.
The average channel
SNR
was set to 20dB and the other simulation parameters are listed
in Table
7.1.
268
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
m
IO.'

v
n
5
IO
15
20
25
30
35
Time
(TS)
(a)
Performance at
a
target
BER
of
1%
for different
channel
quality
esti-
mation delays
in
terms
of
time-slots
(TS),
where each time-slot
is

of
72~s
duration.
0
0.89ds
BER
Performance
(0.01%)
0
5
IO
IS
20
25
30
35
Time
(TS)
(b)
Performance
at
a
target
BER
of
0.01%
for
different
channel quality es-
timation

delays
in
terms
of
time-slots
(TS),
where each
time-slot
is
of
72~s
duration.
Figure
7.8:
Impact
of
channel quality estimation latency upon the wideband
AQAM
scheme for two
different normalized Doppler frequencies of
3.25
x
(at
13.3ds)
and
2.17
x
lO-'(at
0.89ds),
where the switching thresholds were set according to Table

4.8.
The performance
utilizing the linear prediction technique (denoted by Linear Prediction) was compared to
the conventional non-predicted technique (denoted by Past Estimate). The average channel
SNR
was set to 20dB and the other simulation parameters are listed
in
Table
7.1.
7.2.
CHANNEL QUALITY ESTIMATION LATENCY
269
1% 0.01%
Speed(m/s)
3
8
5
11
13.33
4 12
8
16
0.89
Past(TS) Linear(TS)
Past(TS)
Linear(TS)
Table
7.2:
The channel quality estimation delays
in

an AQAM wideband scheme in order to achieve
target BERs
of
1%
and
0.01%,
which were extracted from Figure
7.8.
The delays were
measured for different normalized Doppler frequencies of 3.25
x
lop5
(at a
vehicular speed
of 13.3ds) and 2.17
x
1OP6(at a vehicular speed
of
0.89ds).
Further comparisons were
made between the performance achieved
by
utilizing the linear prediction technique of Sec-
tion
7.2.4
(denoted
by
Linear)
and
the conventional non-predicted scheme (denoted

by
Past).
The delays were measured in terms of time-slots (TS), where each time-slot was of 72ps du-
ration.
7.2.5
Sub-frame TDDRDMA Wideband AQAM Performance
Having considered the implications of channel quality estimation latency, we will now in-
vestigate the performance of wideband AQAM in a sub-frame based TDDA'DMA scheme,
which was discussed in Section 7.2.1. The channel quality estimation latency incurred was
equivalent to half of a TDMA frame, which was set to 4.615ms according
to
the Pan Euro-
pean FRAMES proposal
[
1511. Hence the channel quality estimation latency incurred was
2.3075ms or 32 time-slots, where each time-slot was of 72ps duration. The linear prediction
technique of Section 7.2.4 was invoked, in order to predict the next channel quality. The
modem mode switching thresholds
-
which are shown in Table 7.3
-
were experimentally
determined in order to achieve the target BERs of 1% and 0.01%, since the impact of delay
prohibited the utilization
of
the Powell optimization technique discussed in Section 4.3.5.
The normalized Doppler frequency was set to 2.17
x
lop6
in order to create a slowly varying

propagation environment and the other simulation parameters were set according to Table
7.1. The associated wideband AQAM performances for target BERs of
1%
and 0.01% are
shown in Figures 7.9 and 7.10, respectively. In both of these figures, the corresponding upper
bound performance was also included for benchmarking.
tl(dB)
t4(dB)
t3(W
l2(dB)
1%
25.00
18.80
13.56
11.25 0.01%
18.68 13.65 8.00
5.64
Table
7.3:
The switching thresholds that were manually optimised
in
order to achieve target BERs
of
1%
and
0.01%.
The wideband AQAM regime was implemented in a sub-frame
based TDDiTDMA system having
a
channel quality estimation latency

of
32 time-slots or
2.3075ms as shown
in
Figures
7.9
and
7.10.
Referring to Figures 7.9 and 7.10, the target BERs of 1% and 0.01% were achieved with
slight degradation in terms of its throughput performance, when compared to the upper bound
performance. Explicitly, a BPS/SNR degradation of 0.9dB and 1.8dB was observed for tar-
get BERs of 1% and 0.01%, respectively. The BPS throughput performance of the latency-
impaired wideband AQAM scheme was also compared to that of the fixed modulation modes
270
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
Channel SNR(dB)
Figure
7.9:
The performance of a sub-frame TDD/TDMA based wideband AQAM scheme having
a
channel quality estimation latency
of
32
time-slots
or
2.3075ms, as described in Section

7.2.1. The switching thresholds are set according to Table
7.3
for
a target
BER
of
1
%
and
the simulation parameters are listed in Table 7.1. The upper-bound performance without
channel quality estimation delay was
also
displayed
for
comparison.
of Figure 7.1 for target BERs of
1%
and 0.01%. The results are shown in Figure 7.1
1
exhib-
ited the same characteristics as Figure 4.25 and hence can be justified similarly. The gains
achieved by the latency-impaired wideband AQAM are tabulated in Table 7.4 for a through-
put of
1,
2 and 4 bits per symbol, corresponding to the throughput
of
BPSK, 4QAM and
16QAM modes, respectively.
1%
0.01%

BPS
5.70 25.70 32.00
1.40
22.40 21.00
4
5.40
29.00
23.60
0.70
13.70
13.00
2
6.40 26.70 20.30
1.40
11.10 9.70
1
Gain(dB) AQAM(dB) Fixed(dB)
Gain(dB) AQAM(dB)
Fixed(dB)
I
I
I
I
I I I
I
Table 7.4:
The channel
SNR
gain achieved by the sub-frame based TDDiTDMA wideband AQAM
scheme with a channel quality estimation latency of 32 time-slots

or
2.3075ms, when com-
pared to the fixed modulation modes, at throughputs
of
1,
2 and
4
bits per symbol
(BPS).
The values were extracted from Figure 7.1 1.
7.3.
EFFECT OF CO-CHANNEL INTERFERENCE ON
AOAM
271
1
o-2
a!
m
W
l
10-
61
10-4
1
1
5
10
15
20
25

30
35
40
Channel SNR(dB)
0
6
5
4
3%
2
Figure 7.10:
The performance
of a
sub-frame TDDiTDMA based wideband AQAM scheme with
a
channel quality estimation latency of 32 time-slots
or
2.3075ms,
as
described in Section
7.2.1.
The switching thresholds are set according to Table
7.3
for
a
target
BER
of
0.01%
and the simulation parameters are listed in Table

7.1.
The upper-bound performance with-
out channel quality estimation delay was also displayed
for
comparison.
In this section, we have analysed and recorded the impact of channel quality estimation
latency over channels having different Doppler frequencies upon a wideband AQAM scheme,
We invoked a simple linear prediction technique, in order to predict the next channel quality
estimate, which allowed the wideband AQAM scheme to be more robust against channel
quality estimation delay. Subsequently, the maximum channel quality estimation delay that
can be tolerated by a wideband AQAM scheme was recorded in Table 7.2 for target BERs
of
1%
and
0.01%.
Finally, we characterized a realistic and practical sub-frame TDD/TDMA
based AQAM system, which was robust up to delays of 2.3ms and still achieved substantial
BEWSNR gains over fixed modulation modes, as evidenced by Figure 7.11 and Table 7.4.
7.3
Effect of Co-channel Interference on
AQAM
In all our previous experiments our work has been restricted to a noise limited environ-
ment. However, in a cellular mobile environment the impact of interference
-
in particular
co-channel interference
-
has to be considered in a wideband AQAM scheme. In order to
increase the capacity of a cellular mobile environment, tight frequency reuse techniques
are

frequently utilized
[222].
This
is
a technique, whereby a particular radio channel of a cell can
272
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
10;
8
l
6
S
4
CA
a
E43
2
AQAM
1%
-
0.01%
.
. . . . . .
100
-
/

l
10
15
20 25 30 35
40
Channel
SNR (dB)
Figure
7.11:
The BPS throughput
of
the sub-frame TDDRDMA based wideband AQAM scheme with
a channel quality estimation latency
of
32
time-slots or
2.3075ms
and that of the individ-
ual
fixed modulation modes
of
BPSK, 4QAM, 16QAM. The BPS throughput values were
extracted from Figures 7.10 and
7.2
and the simulation parameters are listed
in
Table
7.1.
be reused in another cell, which is separated by a certain distance.
As

a
result of the utiliza-
tion of a common radio channel in these two cells, transmission in one cell can propagate to
and distort the co-channel transmissions in the other cell. Hence the interference is termed as
CO-Channel Interference (CCI).
In our subsequent experiments the interferer was assumed to be temporally synchronous,
in other words the signals transmitted by the interferer and the reference user were perfectly
synchronous at the receiver. This approach was
also
adopted by
-
amongst others
-
Torrance
and Webb [61,223]. The signal
of
the interferer was also assumed to be phase non-coherent
with the reference signal at the receiver. In this respect, we have assumed that the independent
fading nature of each user’s channel resulted in a phase non-coherent scenario. The channel
model of the interferer and desired user was assumed identical,
as
described by Table
4.5
and
Figure 4.12. The Signal to Interference Ratio (SIR) is
a
parameter which characterizes an
interference-limited environment and is defined
as
follows [222,224]

:
7.3. EFFECT OF CO-CHANNEL INTERFERENCE ON
AQAM
273
c
where
K1
is the number of interferers,
S
is the signal power
of
the reference user and
Pptf
is the power transmitted by the kth interferer.
7.3.1
Impact
of
Co-Channel Interference on
Channel Quality Estimation
In a wideband AQAM scheme the presence of CC1 can potentially degrade the accuracy of
the demodulation process and the channel quality estimation, which is needed for AQAM
mode selection. The issues associated with the impact of CC1 upon the demodulation pro-
cess is discussed in Section 7.3.2 in more depth, while here we focus our attention on the
performance degradation inflicted by the channel quality estimation in this section.
In Section 7.2 we have discussed the importance of channel quality estimation, in order to
ensure that the selected modulation mode was optimum. However, in an interference-limited
environment the performance of a wideband AQAM scheme is degraded due to two factors:
0
The presence of CC1 degrades the ability of the receiver to accurately estimate the
channel quality on a burst-by-burst basis. Consequently the modulation mode selection

errors increase, yielding a degraded BER performance.
0
In a TDD/TDMA AQAM scheme, the channel’s reciprocity is exploited in the uplink
and downlink transmission, in order to estimate the channel quality, as highlighted in
Section 7.2. However, this reciprocity is not applicable in estimating the CCI, since the
uplink and downlink CC1 possess different propagation paths and different transmitted
powers. Consequently the modulation mode selection regime of the receiver
-
which is
subjected to uncorrelated uplink and downlink CC1
-
may not be optimum.
By assuming that statistically speaking the impact of CC1 on the receiver is identical in
the uplink and downlink transmission, we can focus our investigations on the downlink per-
formance for simplicity. In order to isolate and study the impact of CC1 on the modem mode
switching regime, we assumed that the CC1 was only present during the uplink transmission,
but not during the downlink transmission. Again, this was a hypothetical situation, but it was
necessary to stipulate these conditions, in order to analyse the impact of CC1 on the switching
regime. With this assumption in place, the reception at the BS was contaminated with CC1
while the MS experienced a interference-free demodulation conditions.
In our subsequent discussion, we considered only a single-interferer scenario and the
modulation mode of the interferer in the wideband AQAM scheme was chosen randomly
from the set of permissible modes of the AQAM regime. The uplink channel quality was
estimated using the CIR estimator based on the Kalman algorithm of Section 3.2.1 with the
aid of the mid-amble sequence of the FRAMES non-spread speech burst of Figure
4.13.
The
switching thresholds were set according to Table
4.8
for target BERs of

1%
and
0.01%.
The
other simulation parameters are listed in Table 7.
l
An informative insight into the impact of CC1 on the AQAM switching regime can be
obtained by observing the average channel quality estimation errors for different average
274
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
02
'
I
-
0
5
10
15 20 25 30 35
40
Average SIR (dB)
Figure
7.12:
The downlink average channel quality error performance defined by Equation
7.2
for
a

wideband
AQAM
scheme at an average
SNR
of 20dB. An interference-free scenario was
assumed at the
MS
and the switching thresholds were set according to Table
4.8
for
a
target BER
of
0.01%.
The performance was averaged over 10000 transmission bursts and
the simulation parameters were set according to Table
7.1.
SIRs, which is shown in Figure 7.12. The average channel quality estimation error is defined
as
follows
:
c"
(
acc
-
Average Channel Quality Estimation Error
=
%=l
Ydfe
N

(7.2)
where
$F:
is the accurate SNR output
of
the DFE without CC1 and with perfect channel
estimation, while
-yz$
is the estimated
SNR
output of the DFE in the presence of uplink CCI.
The average value was obtained over
N
=
10000
transmission bursts.
In Figure 7.12 the average difference between the actual and estimated SNR output
of
the DFE was recorded for different SIRs over the COST207 CIR
of
Table 4.5 and Figure
4.12, measured at an average channel SNR of 20dB. As expected, at low average SIRs the
average channel quality estimation error was high
as
a result of inaccurate channel estimation
of the reference user. Conversely, at higher average SIRs, the magnitude
of
the CC1 was
lower, resulting in better channel estimation of the reference user. Consequently the average
channel quality estimation error converged to an average minimum of 0.25dB, as evidenced

by Figure 7.12.
The BER and BPS performances for target BERs
of
1%
and
0.01%
for average uplink
SIRs of
0,
10,
20
and 30dB are shown in Figures 7.13 and 7.14, respectively. In terms of
BER performance, as the average SIR increased, the CC1 induced BER and
BPS
degrada-
7.3.
EFFECT
OF
CO-CHANNEL INTERFERENCE ON AQAM
275
2
1
o-*
e!
W2
5
a
1
o-3
5

2
IO-^
1
o-~
5
7
4
38
2
1
0
0
5
LO
15 20 25 30 35
40
Channel
SNR(dB)
Figure
7.13:
The
downlink
performance
of a wideband AQAM scheme contaminated with co-channel
interference at the
BS
and that
of
an interference-free scenario at the MS. This hypothet-
ical situation was necessary in order to study the impact

of
CC1
on the channel quality
estimation process, as explained
in
Section
7.3.1.
The modem mode switching thresholds
were set according to Table
4.8
for
a target
BER
of
1%
and the simulation parameters are
listed
in
Table
7.1.
Midamble channel estimation was applied at both the
BS
and MS.
tion decreased, as evidenced by Figures 7.13 and
7.14.
The degradation was more evident in
the context of the wideband AQAM scheme, which was optimised for a low BER of
0.01%
due to its increased sensitivity to modulation mode selection errors. At a low average SIR
of OdB, the estimation of the reference user's channel quality degraded and hence the effect

of modulation selection errors increased. Consequently, both schemes encountered severe
degradation in terms of their BER performance. However, at higher average SIRs
-
above
lOdB
-
the BER performance approached the target BER, for which it was optimised. This
was achieved as a result of sufficiently accurate channel quality estimates. The BPS perfor-
mance did
not
change significantly for different average SIRs. However, at an average SIR of
OdB, the BPS throughput increased, which was consistent with the corresponding BER degra-
dation. In this respect, as a result of channel quality estimation errors, less robust modulation
modes were erroneously selected, yielding a degraded BER performance and an increased
BPS throughput.
By referring to Figures 7.13 and 7.14, the wideband AQAM scheme was sufficiently
robust against CC1 in terms of its modem mode switching regime performance for average
276
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
6
5
4
3g
2
1
0

5
10
15
20
25
30
35
40
Channel
SNR(dB)
Figure
7.14:
The
downlink performance of
a wideband AQAM scheme contaminated with co-channel
interference at the
BS
and an interference-free scenario at the
MS.
This hypothetical situ-
ation was necessary in order to study the impact
of
CC1
on the channel quality estimation
process,
as
explained in Section
7.3.1.
The modem mode switching thresholds were set
according to Table

4.8
for a target
BER
of
0.01%
and the simulation parameters are listed
in Table
7.1.
Midamble channel estimation was applied
at
both the
BS
and
MS.
SIRs
above 10dB. Hence, the subsequent experiments incorporating CC1 will only consider
average
SIRs
equal to
or
in excess of 10dB.
7.3.2
Impact
of
Co-Channel Interference on the Demodulation Process
In the last section we have quantified and investigated the impact
of
CC1
on
the modem mode

switching regime
of
the AQAM scheme. Consequently, in this section we will investigate
the impact of CC1 on the AQAM demodulation process in the presence of CCI. In order to
isolate and study the impact
of
CC1 on the demodulation process, we have considered an
interference-free scenario at the BS and
a
CCI-impaired receiver at the MS in a downlink
transmission scenario. This is justified upon assuming that the average
SIRs
considered here
are in excess
of
lOdB, which was shown to be sufficient for an AQAM scheme in terms of
reliable channel quality estimation.
In
order to mitigate the impact
of
CC1
on
the demodulation process, two different ap-
proaches are presented here. In the first approach we will utilize Joint Detection (JD) tech-
7.3.
EFFECT
OF
CO-CHANNEL INTERFERENCE
ON
AOAM

277
niques [208], in order
to
jointly mitigate the impact of CCI, IS1 and noise. Alternatively,
in the second approach, we will exploit the modem mode switching regime of the AQAM
scheme in reducing the impact of CC1 on the demodulation process. Before we invoke these
two approaches, let us first quantify the impact of CC1
on
both fixed and adaptive modulation
modes without the aid of CC1 compensation techniques.
In
the following fixed modulation mode based experiments, the modulation mode of the
interferer and reference user was identical. We
also
assumed perfect CIR quality estimation
for the reference user, although no information regarding the interferer was utilized in the
demodulation process at the reference receiver. The other simulation parameters are listed in
Table 7.1.
The associated results for the BPSK, 4QAM, 16QAM and 64QAM modulation modes
are shown in Figures 7.1.5(a), 7.15(b), 7.15(c) and 7.1S(d), respectively with their individual
single-user performances depicted for comparison purposes. The single-user performance
was quantified, when there was
no
CC1 in the system. In each of these figures we can observe
different characteristics at low and high average channel SNRs. At low average channel SIRs,
where noise was dominant, the performance of all modulation modes approached that
of
the
single-user scenario. However, at high average channel SNRs, the impact of CC1 became
dominant, resulting in

an
error floor for all modulation modes. The value of the error floors
decreased with increasing average SIRs for the simple reason that the CC1 became less dom-
inant. Consequently,
as
the average SIR increased, the BER performance for all modulation
modes approached that of the single-user scenario. The other notable characteristic was that
the impact of CC1 was more severe in the higher-order modulation modes of 16QAM and
64QAM, which resulted in higher error floors. This was consistent with our expectations,
since the higher-order modulation modes, with their relatively small constellation Euclidean
distances, were more susceptible to CCI.
Having studied the CCI-impairment on the performance of the individual modem modes
in Figure 7.1
S,
the impact of CC1
on
the wideband AQAM scheme is quantified in Figures
7.16 and 7.17 for target BERs of
1%
and
0.01%,
respectively for average SIRs of
10,
20
and 30dB. The switching thresholds were set according to Table 4.8 for the respective target
BERs and the modulation mode of the interferer was selected randomly. As evidenced by
Figures 7.16 and 7.17, the BER degradation increased as the average SIR decreased. The
observed BER degradation was several orders of magnitude for the wideband AQAM scheme
designed for
a

target BER
of
0.01%.
This was significantly more severe, when compared to
the wideband AQAM scheme with
a
target BER
of
1%.
This highlighted again the increased
error sensitivity of the wideband AQAM scheme designed for a lower target BER. The BER
deterioration for a certain average SIR was higher
as
the average channel SNR increased.
Again, this was consistent with our expectations, since the higher-order modulation modes
which were less robust to CC1
-
as evidenced by Figure 7.15
-
were utilized more frequently,
as
the average channel SNR increased. The BPS performance was unchanged, since a CCI-
free scenario was imposed at the BS, which resulted in optimum modulation mode selection,
with regards to the channel quality estimation. In order to mitigate the impact of CCI, joint
detection techniques [225] are invoked in the next section, which are well known in the field
of CDMA-based systems.
278
CHAPTER
7.
PRACTICAL CONSIDERATIONS

OF
WIDEBAND AOAM
(a)
BPSK
performance with a
BPSK
interferer
\
',
(b)
4QAM
performance with a
4QAM
interferer
1;
IO
'
W
(c)
16QAM
performance with a
I6QAM
interferer
(d)
64QAM
performance with a
64QAM
interferer
Figure
7.15:

BER
performance
of
the fixed modulation modes in the presence
of
CC1
without the
utilization
of
CC1
compensation schemes.
Perfect
CIR
and channel quality estimation
was used
for
the reference user and the other simulation parameters
of
the reference user
are listed in Table
7.1.
7.3.
EFFECT OF CO-CHANNEL INTERFERENCE ON AOAM
279
IO0
6
10.'
5
1
o-2

4
e?:
W
m
10~5
n
10~6;.'
5
10
15
20
25
Channel SNR(dB)
0
10
dB
SIR
V
30dB
SIR
30
35
40
Figure
7.16:
Downlink BER and BPS performance
of
the wideband
AQAM
scheme in the presence

of
a co-channel interferer with associated random modulation modes at the
MS,
without the
utilization
of
CC1
compensation schemes.
Perfect CIR and channel quality estimation
of
the reference user was implemented and the other simulation parameters
of
the
refer-
ence user are listed
in
Table 7.1. The performance was targeted at a BER
of
1%
using the
switching thresholds
of
Table
4.8
and an CCI-free scenario was assumed at the BS, which
was justified in Section 7.3.2.
7.3.3
Joint Detection
Based
CC1

Compensation Scheme
The Joint Detection (JD) receivers are derivatives of the single-user equalizers described in
Chapter
2,
which are used to equalize signals that have been distorted by inter-symbol in-
terference
(ISI)
due to multi-path channels. The problem of CC1 is very similar to that of
multi-path propagation-induced
ISI.
Each user in a K-user system suffers from CC1 due to
the other
(K
-
1)
users. This CC1 can also be viewed as a single-user signal perturbed
by
IS1
from
(K
-
1)
paths in a multi-path channel. Therefore, classic equalization tech-
niques
[87,118]
used to mitigate the effects of
IS1
were modified for multiuser detection and
these types of multiuser detectors were classified as joint detection receivers. The concept
of joint detection for the uplink was proposed by Klein and Baier

[226]
for synchronous
burst transmission, where the performance of a zero-forcing block linear equalizer (ZF-BLE)
was investigated for frequency-selective channels. Other joint detection schemes for up-
link situations were also proposed by Jung, Blanz, Nasshan, Steil, Baier and Klein, such as
280
CHAPTER
7.
PRACTICAL CONSIDERATIONS
OF
WIDEBAND AQAM
0
10dB SIR
0
20
dB SIR
V
30 dB SIR
oodB
SIR

,
U
,"
;;;;
l,,n'
h
1
o~s
1

O-b
.5
~
BER
BPS

0
5
10
Channel SNR(dB)
15
20
25
30
35
40
6
5
4
3g
2
1
0
Figure
7.17:
Downlink BER and BPS performance of the wideband
AQAM
scheme in the presence
of
a

co-channel interferer associated with random modulation modes
at
the
MS,
without the
utilization
of
CC1
compensation schemes.
Perfect CIR and channel quality estimation
of
the
reference user was implemented and
the
other simulation parameters
of
the refer-
ence user are listed
in
Table 7.
I.
The performance was targeted at
a
BER
of
0.01% using
the switching thresholds
of
Table
4.8

and
a
CCI-free scenario was assumed at the BS,
which
was
justified
in
Section 7.3.2.
the minimum mean-square error block linear equalizer (MMSE-BLE) [208,219,227,228],
the zero-forcing block decision feedback equalizer (ZF-BDFE) [219,228] and the minimum
mean-square error block decision feedback equalizer ("SE-BDFE) [219,228]. The uti-
lization
of
the multiuser detection concept
in
a
TDMA environment for
IS1
and co-channel
interference cancellation was implemented by amongst others Yoshino
et
d.
[229], Valenti
et
al.
[230] and Joung
et
al.
[23
l].

The main motivation in these
JD
techniques was to jointly mitigate the effects of
ISI,
noise
and Multiple Access Interference (MAI), which was dominant in a CDMA multiuser system.
This was somewhat analogous to the conventional MMSE-type equalization schemes, where
noise and
IS1
were jointly optimised in order to reduce the effective mean square error at
the input of the detector. In a TDMA environment, we can exploit the JD techniques by
considering the co-channel interference as multiple-access interference and assuming that
the spreading sequence length was restricted to a single-chip, which conformed
to
symbol-
based TDMA transmission. Since the DFE with its MMSE criterion was mainly featured
7.3.
EFFECT
OF
CO-CHANNEL INTERFERENCE ON AQAM
281
in our work, we have opted for invoking the so-called Minimum Mean Square Error Block
Decision Feedback Equalizer (JD-"SE-BDFE) [208] in order to mitigate the impact of
CCI.
We will now present
a
rudimentary introduction to the operation of a JD-"SE-BDFE
receiver, when applied in a TDMA system. For
a
more detailed exposure, the interested

reader is referred to [232].
whitening
whitening
scaling
matched
filter
filter
J
feedback
operator
Figure 7.18:
The schematic of the
JD-"SE-BDFE
receiver structure featuring the different filter
functionalities and
their
corresponding coefficients as explained
in
Section
7.3.3.1.
7.3.3.1
Theory
of
the
JD-"SE-BDFE
The basic structure of the JD-"SE-BDFE is shown in Figure
7.18,
where the feedforward
filter of the structure consists of
a

whitening matched filter, a whitening filter and
a
scaling
filter, while the feedback filter is similar to that of the DFE. Let us now analyse in detail the
structure shown in Figure
7.18.
The received signal vector
y
is represented
as
follows:
y=Ad+n,
(7.3)
where
d
is
the concatenated data vector of the reference user and the interferers, while
n
is
the corresponding noise sample vector. The vector
d
represents the data frame defined by the
data symbols of the reference and interfering users
as
follows:
d
=
di.
di,
dsi

d;,
di, di

dk,
(7.4)
where
dk
represents the 7~th symbol
of
the kth user for
n
=
1,2

N
and k
=
1,2

K.
The system matrix,
A,
which contains the
CIR
estimates of the reference user and inter-
ferers, is
of
size
KN

columns and
(N
+
W
-
1)
rows, where
W
is the length of the
CIR.
Physically, this matrix describes the
IS1
and co-channel interference inflicted on the reference
user. The matrix is constructed from
a
set of column vectors
aJ
for
j
=
1,2,

KN,
which
can be written as
: