A Flexible Architecture for Future Wireless Local Area Networks

Abstract

This paper reports on the activities at IMEC that will lead to a 100Mb/s indoor communication system based on multi-carrier modulation. The goal is to build a wireless local area network, which can compete with today's wired solutions in terms of cost and performance. It is shown how a dedicated ASIC architecture implements an OFDM modem with 256 carriers that allows transmission of complex symbols (QPSK, 16 QAM) at 50Mbaud/s. It is also demonstrated how adaptive turbo-decoding allows operation close to the Shannon limit and how it can be used to provide quality of service. Adaptive loading of the carriers provides a further capacity enhancement. A gain of 6.5dB in the required signal to noise ratio can be observed compared to standard non-adaptive schemes. Finally, an SDMA scheme with full channel characterization is proposed in order to increase the channel capacity even further. It turns out that even with short training sequences a substantial capacity improvement can be observed.

Full text

Page 1

A Flexible Architecture for Future Wireless Local Area Networks
Bert Gyselinckx, Wolfgang Eberle, Marc Engels, Curt Schurgers, Steven Thoen, Patrick
Vandenameele, Liesbet Van der Perre
IMEC
Kapeldreef 75
B-3001 Leuven
Tel: +32 16 281537, email: gyselinc@imec.be
ABSTRACT
This paper reports on the activities at IMEC that will
lead to a 100Mb/s indoor communication system based
on multi-carrier modulation. The goal is to build a
wireless local area network, which can compete with
today’s wired solutions in terms of cost and performance.
It is shown how a dedicated ASIC architecture
implements an OFDM modem with 256 carriers that
allows transmission of complex symbols (QPSK, 16
QAM) at 50Mbaud/s. It is also demonstrated how
adaptive turbo-decoding allows operation close to the
Shannon limit and how it can be used to provide quality
of service. Adaptive loading of the carriers provides a
further capacity enhancement. A gain of 6.5dB in the
required signal to noise ratio can be observed compared
to standard non-adaptive schemes. Finally, an SDMA
scheme with full channel characterization is proposed in
order to increase the channel capacity even further. It
turns out that even with short training sequences a
substantial capacity improvement can be observed.
I.
INTRODUCTION

Telecommunications has mainly focussed on plain old
telephony service up to now. However, this is changing
rapidly. The telecommunications market is predicted to
grow by a factor of 20 in the next 15 years. Almost all of
this growth will come from data services.
In anticipation, standards for data services are being
defined. In the wireless society this has led to the 802.11
standard allowing spread spectrum communication at
2Mb/s in the ISM band. Also HIPERLAN seems to be
well on its way. It will enable high-speed communication
up to 23Mb/s necessary for wireless local area networks
(WLAN). In addition, the third generation of cellular
systems such as UMTS and IMT-2000 also foresee a
mode for data-communication up to 2Mb/s in small cells.
However, it is the authors’ belief that wireless broadband
communications will only massively be deployed at the
moment a performance and price can be offered that is
comparable to what a cabled network is offering.
Therefore, we are working on the next generation of
broadband WLAN, with the goal to achieve an overall
network capacity of up to 100 Mb/s comparable to fast
ethernet. Such an ambitious target asks for an accurate
design of all parts of the system, going from the high
level networking aspects down to the hardware
implementation of the functional blocks. Because price is
a key enabling factor, emphasis is put on full CMOS
integration.
In the next section, the network configuration will be
defined. Section III describes the characteristics of the
indoor channel. Based on this a transceiver architecture
is proposed in section IV.

II.
NETWORK CONFIGURATION

The envisaged network will consist of powerful
basestations connected to a high speed backbone. Each
basestation serves as a hub for all communication to and
from the different terminals within a cell. To reduce the
installation cost it is advantageous for the cells to be
quite large. This can be achieved by using the 2.4GHz or
5.7GHz ISM bands. Moreover, this also allows
integration of the RF front-end in standard CMOS or
BiCMOS technology, in contrast to pico-cellular
networks that are being investigated in the 60GHz band.
Because there are only few basestations in the network, it
is affordable to put of a lot of processing power such as
space-time processing and multi-user detection in these
stations. The asymmetric network configuration makes it
possible to reduce the complexity and the power
consumption of the terminals.

III.
THE CHANNEL
In order to design a modem it is very important to have
knowledge of the channel for which it will be used.
Therefor, the indoor channel was modeled by means of a
ray-tracing technique.
Since the envisaged application is a WLAN structured as
a star-configuration network consisting of relatively large
cells, a typical corridor in an office environment is taken
as the geometrical input for the ray-tracing algorithm.
Figure 1 shows a typical geographical output of the
implemented algorithm
propagating rays taking multi-path components into
account up to a certain threshold for the power loss
(which was 30dB for the presented example). The
basestation is located in the middle of the corridor, while
a mobile terminal is situated in one of the office rooms.
The walls are displayed by solid lines, while all dotted
lines represent propagation paths.
displaying the different

Page 2

Figure 1: Typical ray-tracing output
From the above picture, it is clear that a multitude of
reflections occur in the indoor environment, and
consequently a large number of multi-path components
arrive in the receiver.
The channel impulse response will clearly consist of a
number of discrete pulses. The power delay profile for
the above example is given in Figure 2.
Channel Time Response
00.02 0.04 0.06
Time, usec
0.080.1 0.12
−105
−100
−95
−90
−85
−80
−75
−70
−65
Power, dB
Figure 2: Power delay profile
Simulation results show that the delay spread in this
environment varies between 10 and 40ns. The delay
differences between subsequent multi-path components
are of the order of 0.1ns. From the power delay profile,
the channel transfer function in the frequency domain
was calculated by means of the FFT transform. The
result is presented in Figure 3.
This frequency response shows a very good agreement
with measurement data in a similar environment [2]. It is
clear that the indoor propagation channel is frequency
selective, with a coherence bandwidth of the order of 5 to
25 MHz (depending on the specific situation). From this
observation, we can conclude that using a flat fading
model is not reliable for the envisaged bit rate, and that it
is possible to improve the system’s performance by
exploiting frequency diversity. Dips in the spectrum of
up to 30 dB are perceived, meaning that the transmission
in these frequency bands will be dramatically degraded.
Figure 3: Frequency response
IV.
TRANSCEIVER ARCHITECTURE
Orthogonal frequency division multiplex (OFDM) is a
modulation technique that can exploit frequency
diversity. It allows to transmit data on different
orthogonal carriers. All these carriers are modulated
(PSK or QAM) by a narrow band signal and see a flat
fading channel. Because it is unlikely that all carriers are
subject to deep fades the receiver will be capable of
recovering the originally transmitted sequence. From this
reasoning one could conclude that the performance will
improve with the number of carriers that are used per
OFDM symbol. However, as the number of carriers
increases the symbol duration will also increase. From
the moment this duration starts exceeding the coherence
time of the channel the performance will degrade rapidly.
For an 100Mb/s WLAN active on an indoor channel with
a coherence bandwidth of 10MHz and a coherence time
of 50ms, 256 carriers give an optimal result.
The OFDM transceiver architecture under development
at IMEC is shown in Figure 4 [1]. It consists of a channel
(de)coding module, an adaptive (de)loader, an OFDM
modem and an RF front-end with multiple antennas. In
the following sections our research activities in these
fields will be discussed.
A. OFDM Modem
The implementation target for the OFDM modem is an
ASIC in order to achieve the very high data rates. A
prototype is under development at the moment. This will
perform all OFDM signal processing in the time and
frequency domain.
data
…
Channel
(de)coder
Adaptive
(de)loader
OFDM
Modulator/
demodulator
Spatial Division
Multiple Access
Front-end
Figure 4: OFDM transceiver architecture
Basestation
Terminal

Page 3

One of the key building blocks of an OFDM modem is
the (I)FFT-module. A dedicated pipelined architecture,
based on a recursive decomposition, allows the 256-
points complex FFT to execute in 5µs. This enables
transmission of QPSK symbols at 50Mbaud/s. Care was
taken to minimize the area and power of the FFT. This is
mainly achieved by optimizing the data flow control and
the storage of intermediate results. The layout of the
(I)FFT macrocell is shown in Figure 5. The main features
of this block are summarized in Table 1.
Figure 5: (I)FFT layout
Table 1: (I)FFT features
TechnologyMietec 0.5µ CMOS, TLM,
3V
50 MHz
195 kFFT/s
2.5 * 2.5 mm2
31000 gates
384 bytes RAM
Operating frequency
Throughput
Core size
Gate count
Memory
The ASIC consists of programmable data paths that
provide the necessary flexibility for on-site tuning
depending on the specific application and environment.
B. Channel (De)coding
Because the indoor channel suffers from severe fading,
forward error correcting codes have to be applied in
order to obtain an acceptable link quality. Turbo-codes
[3] allow operation close to the Shannon limit and are
therefore a viable option.
As described in section II, all communication passes
through the basestation. This will receive the signals
from the different mobile terminals all with their own
signal to noise ratios. Turbo-codes are parallel
concatenated convolutional codes, that are decoded using
an iterative scheme. The iterative process of the decoding
allows the receiver to tune its channel decoding for each
individual transmitter. Assume that all mobiles are
transmitting with a particular rate ½ turbo code. The goal
at the receiver is to obtain a bit error rate (BER) of 10-3.
For a transmission that is received through a good
channel the signal to noise ratio per bit (Eb/N0) will be
high, assume 8dB. For a transmission through a bad
channel Eb/N0 will be low, assume 4dB.
Figure 6 shows the simulated BER in function of Eb/N0
for successive iterations in the decoder. The more
decoding iterations are performed the lower the BER,
until after 6 to 7 iterations the solution has converged
almost to the ideal maximum likelihood decoding. From
this figure it becomes clear that for the transmitter
received at 8dB one iteration is sufficient to obtain the
required BER of 10-3. For the transmitter received at
4dB, 6 iterations will be needed. This adaptive decoding
offers two advantages:
•
The processing power can be reduced for good
channels and increased for bad channels, thus
obtaining an equal performance for all users.
Alternatively, more processing power could be
allocated to certain users in order to guarantee a
quality of service (QoS) requirement.
•
The adaptation takes place in the receiver.
Therefore, a dedicated communication link to set-up
a common encoding and decoding scheme in
transmitter and receiver is not needed.
C. Adaptive loading
Multi-carrier modulation such as OFDM makes it
possible to assign a variable number of bits to each
carrier depending on the signal to noise ratio (SNR) for
each carrier.
The SNR in our system is estimated by sending four
times a QPSK-modulated PN-sequence over the channel.
Let r(i) be the signal received on the i-th carrier.
Provided the length of the guard interval was chosen to
be longer than the delay spread of the channel, r(i)
becomes:
r iPN iH i
( )( )( )
=⋅
where H(i) stands for the complex channel attenuation
factor and n(i) the complex, additive noise on carrier i
which is considered to be white and Gaussian.
The channel attenuation ~
H is estimated from:
~
i PNir avgiH
=⋅=
(2)
Here, avg means taking the average over the 4 PN-
sequences that were sent. To diminish the second noise
term and obtain a better estimation ~
n i
( )
+
(1)
)}()({)()()}({)(
iPNin avgiH
⋅+
Hf, a lowpass filter
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
BER
23456789
Eb/N0
Figure 6: Performance of the turbo-decoder for
successive iterations
Decoding
iterations
1
2
6

Page 4

is applied to the channel estimation. The spectrum of the
channel is shown together with the unfiltered channel
estimation in Figure 7. A typical result for the filtered
channel estimation is shown in Figure 8.
Figure 7: Unfiltered channel estimation
Figure 8: Filtered channel estimation
The filtered attenuation estimation clearly forms an
accurate representation of the given fading channel.
The noise power estimation ~
manner
~( )
{| ( )
N iavg r iH i
=−
Using the noise power and attenuation estimations, the
bits are assigned in a first step with infinite granularity to
the carriers. In the next step, the assigned number of bits
is quantized to an even integer. Only even integers are
allowed in order to balance the BER on the I and Q
components. The third part of the algorithm adapts the
number of bits in order to make sure that Rtotal equals the
targeted bitrate. Finally, the power of each carrier is
adjusted so that all subchannels yield the same BER. In
the implementation of the modem this is accomplished
by a multiplication per carrier with a power coefficient.
A typical bit assignment for the channel estimation in
Figure 8 is shown in Figure 9.
N is obtained in a similar
~( )
f
( )| }
PN i
⋅
2
(3)
Figure 9: Bit assignment
This figure indicates that the dips in the spectrum are not
carrying any bits while more bits are transmitted on the
carriers with low attenuation. Furthermore, the order of
the constellation on the good carriers is low (QPSK or
16-QAM) compared to the order in ADSL, since in
general the Eb/N0 is relatively low for the indoor channel.
Figure 10 compares the achievable BER for adaptive
loading and for plain QPSK on all the carriers for the
ray-tracing channel model when the data rate is 100Mbps
and a cyclic prefix of 8 bits is used. A 6.5dB gain is
observed for BER = 10-2. From the figure, it is clear that
a substantial advantage can be achieved by adapting the
bit allocation to the spectrum of the fading channel.
Figure 10: BER vs. Eb/No for ray-tracing model
The main benefit arises from the fact that the 30dB dips
in the spectrum are now not carrying any bits while in
QPSK they are the main source of bit errors. To
accomplish this, more bits are sent on the peaks of the
spectrum.
D. Space Division Multiple Access
Channel turbo-coding allows operation close to the
Shannon limit. In order to increase the capacity further,
spatial diversity must be used. In our transceiver, we
propose to reuse bandwidth by space division multiple
access (SDMA). This technique separates multiple
simultaneously transmitting terminals based on their

Page 5

different spatial properties by processing the signals
received at an antenna array.
Because the indoor channel is characterized by heavy
multipath distortion, the signals arriving at the receiver
come from all directions. Therefore, techniques based on
beam forming are not appropriate. Instead, SDMA
algorithms based on full channel identification have to be
used. For the further analysis it is assumed that the total
network capacity of 100Mb/s is shared between 4 users,
each having a capacity of 25Mb/s. Single carrier BPSK
modulation is assumed to simplify the reasoning.
Therefore, a bandwidth of 25MHz is needed for the 4
users. A symbol will last 40ns. Assuming an indoor
channel with a delay spread of 40ns, the last arriving
signal will have a delay of approximately 120ns,
requiring a 3 tap equalizer in the receiver to remove ISI.
The channel identification is based on the transmission of
a training sequence that is known a priori by the receiver.
To limit the capacity loss resulting from this sequence, a
fast converging recursive least squares (RLS) algorithm
is used to determine the channel parameters.
Furthermore, the four users are trained simultaneously.
Because the fading rate of the indoor channel is in the
order of 10Hz, the channels estimation can run at a much
slower rate than the actual symbol detection (Figure 11).
BER versus SNR simulation results for the SDMA
receiver with RLS based channel estimation and MLSE
symbol detection are shown in Figure 12. The dashed
curve labeled perfect shows the performance of the
reception of four simultaneous users with a four antenna
receiver, in the case of perfect channel estimation. The
performance with the imperfections of the RLS channel
estimation included is given by the plain curves. The
training sequence length j is used as a parameter. It is
observed that a 30 symbol training sequence already
shows similar performance to perfect 25Mb/s single user
single-antenna Viterbi reception, which is the bold
dashed curve.
0246810 12141618
10
−7
10
−6
10
−5
10
−4
10
−3
10
−2
10
−1
10
0
BER
RLS Channel Estimation
SNR
perfect
j = 100
j = 30
j = 16
single user
Figure 12: Performance of SDMA algorithm
V.
CONCLUSIONS
We presented a transceiver architecture for the next
generation of WLAN. The prototype OFDM modem
ASIC will allow a network capacity of 100Mb/s. Further
research on turbo-coding, adaptive loading and spatial
diversity will allow to increase the capacity even further.
REFERENCES
[1]
W. Eberle, L. Van der Perre, B. Gyselinckx, M. Engels, I.
Bolsens. Design Aspects of an OFDM Based Wireless LAN
with Respect to ASIC Integration. Proc. of the 2-nd OFDM-
Fachgespräch, Braunschweig, Germany, September1997.
G.J.M. Janssen, P.A. Stigter, and R. Prasad. Wideband Indoor
Channel Measurements and BER Analysis of Frequency
Selective Multipath Channels at 2.4, 4.75, and 11.5 GHz. IEEE
Trans. on Communications, Vol. 44, No 10, Oct 1996, pp.
1272-1281.
C. Berrou, A. Glavieux, P. Thitimajshime, “Near Shannon limit
error-correcting coding and
Proceedings ICC’93, Geneva, Switzerland, May 1993, pp.
1064-1070
C. Schurgers, L. Van der Perre, M. Engels, H. De Man,
"Adaptive Turbo Decoding
Communication," (submitted), ISSSE, Pisa, Italy, September-
October 1998.
L. Van der Perre, S. Thoen, P. Vandenameele, B. Gyselinckx,
M. Engels, "Adaptive loading strategy for a high speed OFDM-
based WLAN," (submitted), Globecom 98, Sydney, Australia,
November 1998.
P. Vandenameele, L. Van der Perre, B. Gyselinckx, M. Engels,
H. De Man, "An SDMA Algorithm for High-Speed Wireless
LAN," (submitted), Globecom 98, Sydney, Australia, November
98.
[2]
[3]
decoding: Turbo-code,”
[4]
for Indoor Wireless
[5]
[6]
…
…
Channel estimation at
fading rate
Symbol detection at
symbol rate
Figure 11: Receiver with channel estimation at fading rate