Advances in Radio Science, 3, 27–37, 2005
SRef-ID: 1684-9973/ars/2005-3-27
© Copernicus GmbH 2005
OFDM: From the Idea to Implementation
S. A. Fechtel
Infineon Technologies AG, Rosenheimer Str. 116, 81699 München, Germany
Abstract. OFDM (orthogonal frequency-division multiplex-
ing) is one of the key digital communication technologies of
the current decade. The first part of this paper presents the
fundamentals of OFDM and its benefits in the presence of
multipath propagation in a tutorial-like fashion. The second
part details on some of the most important aspects of OFDM
transceiver implementation: concept of receiver channel fil-
tering and A/D conversion, radio impairment compensation
(I/Q mismatch), and OFDM demodulator (FFT) design.
1 Introduction
The ever-increasing demand for very high-rate wireless data
transmission calls for technologies which make use of the
available electromagnetic resource in the most intelligent
way. Key objectives are spectrum efficiency (bits per second
per Hertz), robustness against multipath propagation, range,
power consumption, and implementation complexity. These
objectives are often conflicting, so techniques and implemen-
tations are sought which offer the best possible tradeoff be-
tween them.
Two of the most effective means of closing the gap be-
tween the achieved performance and channel capacity are
advanced channel coding (to combat noise) and OFDM mod-
ulation (to combat multipath). In conjunction with advanced
forward error correction (FEC) coding, e.g., turbo, spherical,
or LDPC (low-density parity check) codes, advanced OFDM
is the modulation of choice when it comes to improving ro-
bustness against multipath - possibly fading – at reasonable
cost of implementation. In order to improve spectral effi-
ciency, the best strategy is to increase the capacity of the
channel itself. This is effectuated by multiple-input multiple-
output (MIMO) smart antenna systems which make use of
the spatial dimension to essentially scale the data rate trans-
ferred over the same channel bandwidth with the number
of antennas. As MIMO systems have now become within
reach of low-cost implementation in highly-integrated de-
vices, they are currently a very active field of R&D activity
and standardization.
Correspondence to: S. A. Fechtel
(stefan.fechtel@infineon.com)
Advances in
Radio Science
Combining several advanced technologies leads to partic-
ularly attractive solutions. OFDM, which by now has ma-
tured to an established technology, also serves as a solid ba-
sis of many of the most recent R&D developments, in par-
ticular MIMO. Figure 1 gives an overview of some impor-
tant wireless communication standards and standardization
efforts in terms of range and PHY (physical layer) data rate.
OFDM and MIMO-OFDM are seen to play a key role in
high-rate data transmission over wireless channels for a wide
range of applications such as WLAN (wireless local-area net-
works), DVB-T (terrestrial digital video broadcasting), and
also UWB (ultra wideband).
Concentrating on wireless LAN in this paper, Table 1
displays basic parameters of current and forthcoming IEEE
WLAN standards. All standards but the very first (802.11b,
1...11 Mb/s) are based on OFDM. Currently, 802.11a/g-
based devices are entering the mass market, offering PHY
rates between 6 and 54 Mb/s. In the near future, 802.11n
standardization will be finalized. The two most promising
proposals are TGn Sync (task group n synchronization) and
WWiSE (world-wide spectrum efficiency). TGn Sync pro-
poses doubling the channel bandwidth to 40 MHz (rates up to
630 Mb/s), whereas WWiSE confines itself to 20 MHz chan-
nels (40 MHz optional where permitted, then with rates up
to 540 Mb/s) in order to obey current spectrum regulations.
Both contending proposals improve MAC efficiency, and,
more importantly, both are based on MIMO-OFDM tech-
nology with between 2×2 and 4×4 TX and RX antennas.
The latest development efforts for next-generation WLAN,
e.g., the WIGWAM (wireless Gigabit with advanced multi-
media support) project initiated in 2003, are targeting wide-
band MIMO-OFDM (100 MHz channel in higher bands up
to 60 GHz) to attain scalable data rates well beyond 1 Gb/s.
This paper is organized as follows: Sect. 2 features a brief
tutorial on OFDM and its benefits in the presence of multi-
path radio propagation, and Sect. 3 gives details on some of
the most important aspects of a typical OFDM-based WLAN
PHY transceiver implementation, viz., concept of receiver
channel filtering and subsequent A/D conversion, transmitter
I/Q mismatch calibration as an example of radio impairment
compensation, and the design of OFDM modulators and de-
modulators, i.e., FFT design.
28 S. A. Fechtel: OFDM: From the Idea to Implementation

data rate
1 Gbit/s
Gbit/s
WLAN OFDM

WLAN
100 Mbit/s
802.11n
WLAN
802.11a
10 Mbit/s
WLAN DVB-T
802.11b
1 Mbit/s

UMTS
100 kbit/s
Blue-
DECT
10 kbit/s tooth
GSM
range
10 m 100 m 1 km 10 km

Figure 1. Wireless
Fig. 1. Wireless Communication Communication Standards.
Standards.

1
Table 1. Basic Parameters of Wireless LAN Standards and Standard Proposals.
0.8Table 1. Basic Parameters of Wireless LAN Standards and Standard Proposals
Relative Power

0.6Standard Band, Channel, Modulation maximum PHY rates

0.4802.11b 2.4 GHz, 25 MHz channel, DSSS/CCK 11 Mb/s
0.2802.11a/g 5/2.4 GHz, 20 MHz channel, OFDM 54 Mb/s
0802.11n 5/2.4 GHz, 20/40 MHz channel, OFDM
0 50 100 150 200 250 300 350 400 450 500
TGn Sync 20 MHz (optional), 40 MHzTime [ns]
(mandatory) 140 / 315 Mb/s (2x2, 20/40 MHz)
0Proposal 2x2 (mandatory) or 4x4 MIMO OFDM 280 / 630 Mb/s (4x4, 20/40 MHz)

-10WWiSE 20 MHz (mandatory), 40 MHz (optional) 135 / 270 Mb/s (2x2, 20/40 MHz)
Pathloss [dB]

Proposal 2x2 (mandatory) or 4x4 MIMO OFDM 270 / 540 Mb/s (4x4, 20/40 MHz)
-20
Gbit/s 5/17/24/60 GHz bands, 100 MHz channel, more than 1 Gb/s
-30WLAN up to 4x4 MIMO OFDM

-40
-100 -80 -60 -40 -20 0 20 40 60 80 100
Frequency [MHz]
2 Fundamentals of OFDM both in terms of performance and complexity. An old
Table 2. WLAN Standard 802.11a/g: PHY Transmission Parameters
but effective idea to circumvent time-domain equalization
Wireless communicationFigureusually deals
2. information
Typical with
data rate multipath
Multipath Radio Channel 6, 9, consists
radioImpulse12, 18, in 48,converting
24, 36,
Response and 54 Transfer
Mbit/s aFunction
bitstream with high rate R into
channels. The time- and frequency domain representations a large number of N parallel bitstreams, each with low
channel bandwidth = sampling rate 20 MHz, occupied bandwidth 16.6 MHz
of a typical channel realization are shown in Fig. 2. The rate R/N , and transmitting the set of N3 substreams over a
baseband-equivalent channelOFDM symbolresponse
impulse length h(τ ; t) and 4.0 µscorresponding
= 80 samples set of N smallband subchannels formed by
transfer function H (f ; t) both characterize the instantaneous
useful symbol length = FFT length partitioning
3.2 µs = 64 samples the broadband channel, see Fig. 3. The receiver
channel state at time t, where the impulse response is formed essentially performs N parallel demodulations, one for each
guard interval length 0.8 µs = 16 samples
by superposition of a (possibly large) number of reflected or frequency bin, followed by parallel-to-serial conversion to
scattered multipath rays withnumber
distinct (complex-valued)
of occupied subcarriers gains 52 restore the original serial high-rate bitstream. This frequency
ci (t) and delays τi (t). Naturally,
number the gains
of data tend to become 48 partitioning approach is called frequency-division multiplex
subcarriers
smaller as the path delay rises, and the set of “relevant” paths (FDM). In order to avoid crosstalk between subchannels,
number of pilot subcarriers
with gain ci (t) above a certain significance level (depending 4 their carrier waveforms are usually chosen to be mutually
on the required signal-to-noise ratiofrequency
subcarrier particular link) 312.5orthogonal.
for thespacing kHz (=20 MHzThe / 64) modulation so formed is termed orthogonal
determines the effective spansubcarrier
τmax ofmodulation
the multipath channel. BPSK, FDM, or OFDM. Computationally efficient IFFT (inverse
QPSK,16-QAM, 64-QAM
Wideband transmission channels used in high-rate com- FFT) and FFT (fast Fourier transform) architectures are
forward error correcting code
munications are often strongly frequency-selective (lower convolutional usuallycode, K=7, rates 1/2, to
implemented 2/3, perform
3/4 OFDM modulation and
part of Fig. 2), rendering conventional equalization difficult, demodulation, respectively.

2
 10 m 100 m 1 km 10 km

FigureFrom
S. A. Fechtel: OFDM: 1. Wireless
the Idea Communication
to Implementation Standards. 29

0.8

Relative Power
0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400 450 500
Time [ns]

-10
Pathloss [dB]

-20

-30

-40
-100 -80 -60 -40 -20 0 20 40 60 80 100
Frequency [MHz]

Fig. 2. Typical Multipath Radio Channel Impulse Response and Transfer Function.
Figure 2. Typical Multipath Radio Channel Impulse Response and Transfer Function

mod 1 channel 1 demod 1 3

mod 2 channel 2 demod 2
S/P
... ... ... . . . P/S
mod N channel N demod N

Fig. 3. Parallel Transmission over Smallband Subchannels.

Figure 3. Parallel Transmission over Smallband Subchannels

guard interval. Figure 5 illustrates the insertion of a guard

Figure 4 illustrates OFDM transmission over a frequency- interval and the effect of multipath on a so-formed OFDM
selective channel. In this simple example, the four sub- symbol. The guard interval of length Tg is given by the cyclic
carrier waveforms are givenOFDM by 0,symbol
1, 2, and 3 full peri- extension at the beginning of an OFDM symbol (left-hand
ods of sinusoids within 1 length T , respectively. An OFDM side of Fig. 5), its samples are taken from the latter section
1 transmitted
of the (original) OFDM symbol. Hence, the total OFDM
symbol is formed by0 modulating each of thesef 0subcarriers
=0
0.5 symbol length increases to T +Tg . The associated loss in
by a (complex-valued) -1 PSK or QAM symbol and super-
pose the resulting modulated
0 waveforms.
0.5 The1 corresponding 0 transmission efficiency T /(T +Tg ) is greatly outweighted by

frequency-domain signals1
(right-hand side of Fig. 4) exhibit -3
the-2benefits
-1
in
0
the1 presence
2 3
of multipath,
4 5 6
as illustrated by the
0
a sin(x)/x-characteristic with center frequencies f0, =1/T right-hand side of Fig. 5: at the channel output (here two-
1 1/T , 2/T 1 channel
and 3/T , respectively.-1 At these frequencies, all but one wave- ray), delayed and weighted copies of the waveform (the one
0 0.5 1 0.5 with frequency 1/T is shown as an example) are superim-
form are seen to have1 zero crossings, so they are mutually
orthogonal. Given a 0sufficiently large number of subchan- 0 posed. Therefore, the received waveforms exhibit disconti-
f =2/T nuities at the0 (delayed) borders between consecutive OFDM
nels, thus small subcarrier bandwidth relative to2 the chan- -3 -2 -1 1 2 3 4 5 6
-1
nel coherence bandwidth, symbols. However, as long as the span of channel delays
0 the use 0.5of OFDM 1effectively con- 1 received
verts a frequency selective
1 channel into an orthogonal set of τmax is smaller than guard interval length Tg , all of these dis-
0.5 continuities fall inside the guard interval. By simply elimi-
frequency-flat subchannels.
0 f =3/T
0 nating the guard interval samples and retaining only the “use-
3
-1
Beyond orthogonality,
0
a second0.5ingredient
1
is required in ful” signal, the received waveforms turn6 out to be given by
-3 -2 -1 0 1 2 3 4 5
making OFDM symbols robust against
T multipath, viz., the 1/T
30 S. A. Fechtel: OFDM: From the Idea to Implementation

OFDM symbol
1
1 transmitted
0 f 0=0
0.5
-1
0 0.5 1 0
1
-3 -2 -1 0 1 2 3 4 5 6
0 f 1=1/T
1 channel
-1
0 0.5 1 0.5
1
0
0 f 2=2/T
-3 -2 -1 0 1 2 3 4 5 6
-1
0 0.5 1 1 received
1
0.5
0 f 3=3/T
0
-1
0 0.5 1 -3 -2 -1 0 1 2 3 4 5 6
T 1/T

Fig. 4. OFDM Symbol: Orthogonal Subcarrier Waveforms.

Figure 4. OFDM Symbol: Orthogonal Subcarrier Waveforms

guard interval insertion: cyclic extension guard interval elimination

1 1

0 f 0=0 0

-1 -1
0 0.5 1 h(τ) 0 0.5 1
1 1

0 f 1 =1/T a 0
1
-1 a -1
2
0 0.5 1 0 0.5 1
1 1

0
0 τ 0
f 2 =2/T 4
-1 -1
τ
0 0.5 1 0 0.5 1
1 1

0 f 3 =3/T 0

-1 -1
0 0.5 1 0 0.5 1
Tg T
Tg T useful signal window

Fig. 5. OFDM Symbol: Guard Interval.

Figure 5. OFDM Symbol: Guard Interval

the amplitude-weighted and phase-shifted versions of their 3 OFDM Transceiver Implementation

original counterparts. The orthogonality property of the orig-
inal OFDM symbol is thus restored, which implies both or-
thogonality in time (no intersymbol interference, ISI) and
in frequency (no intercarrier interference, ICI). As a conse-
quence, equalization in the subcarrier domain becomes par-
ticularly simple: the remaining amplitude and phase distor-
tion (or, equivalently, the complex-valued gain factor) affect-
ing each subcarrier can be easily corrected for, e.g., by com-
plex division or demapping.
In this second part of the paper, we focus our attention
to WLAN transceiver implementation. Figure 6 shows
the block diagram of a 2.4/5.2 GHz dual-band 802.11a/b/g
transceiver which has been realized as a three-chip solution.
The transceiver comprises (from left to right) the host in-
terface, MAC (medium access controller), PHY with A/D,
D/A converters and OFDM/CCK baseband processors, RF
transceiver, power amplifier, and two diversity antennas with
 Figure 5. OFDM Symbol: Guard Interval

S. A. Fechtel: OFDM: From the Idea to Implementation 31

Fig. 6. 802.11a/b/g Dual-Band WLAN Transceiver.

Figure 6. 802.11a/b/g Dual-Band WLAN Transceiver

20 MHz -> 80 MHz low-IF
center 0 -> 10 MHz signal 5
6...54 Mb/s scrambler IFFT
input DAC low-IF
convol. encoder guard low-IF TX IQ/DC IF
upcon- PA
from interleaver interval upconversion compensation DAC TX
version
MAC QAM modulator insertion 2
LO

80 MHz -> 20 MHz

low-IF center 10 MHz -> 0
signal
low-IF 6...54 Mb/s
low-IF output
PGA IF ADC filtering + guard subcarrier deinterleaver
LNA downcon- FFT
PGA filter ADC downcon- removal equalization Viterbi decoder to
version
version 3 MAC
LO 1

preamble detection channel

timing/frequency sync estimation

Fig. 7. WLAN OFDM Baseband

Figure 7.and RF PHY
WLAN Transceiver
OFDM Architecture.
Baseband and RF PHY Transceiver Architecture

Low IF Spectrum Mask

TX/RX antenna switch. -60 3.1 Receiver Channel Filtering and A/D Conversion
desired signal
interference
noise floor
level [dBm]

-70
Figure 7 displays the PHY transceiver architecture in more
-80
In our channel filtering concept, receiver selectivity is shared
detail. Transmitter baseband processing comprises channel
between analog IF filtering and digital postfiltering follow-
encoding and QAM subcarrier modulation, OFDM modula- -90

-40 -30 ing A/D conversion. The low-IF spectrum mask depicted
-20 -10 0 10 20 30 40
tion (IFFT and guard interval insertion), low-IF upconver-
0
in the upper part of Fig. 8 reflects the worst-case interfer-
Low IF Analog Filter Response

sion (center of spectrum shifted from 0 to 10 MHz), TX I/Q

-10 ence scenario (according to 802.11a) where the desired sig-
and DC compensation, and D/A conversion. Receiver base-
-20
nal (centered about 10 MHz) is surrounded by adjacent chan-
gain [dB]

band processing performs A/D conversion, channel filtering

-30

-40 nel interference (ACI) of some maximum level. The analog

and low-IF downconversion (center of spectrum shifted back
-50
polyphase bandpass IF filter (center of Fig. 8) is designed
from 10 MHz to 0), OFDM demodulation (guard interval re-
-60
-40 -30
so as to suppress only part of this interference (bottom of
-20 -10 0 10 20 30 40

moval and FFT), subcarrier equalization, channel decoding, Low IF Filtered/Aliased Spectrum
-70 Fig. 8); its main purpose is to prevent aliasing and reduce the
and all necessary ancillary functions such as preamble de-
-75
dynamic range of residual interference in subsequent A/D
sampled desired signal
sampled interference
tection, gain control, timing and frequency synchronization,
-80 aliasing
level [dBm]

-85 conversion. By virtue of this approach, analog filter com-

noise floor

channel estimation, RF and MAC interfacing, and transceiver

-90
plexity and power consumption is much reduced at the ex-
control. -95

-100
pense of only one additional A/D converter bit and the need
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
In the following, three of the most important aspects of for a digital postfilter eliminating residual interference.
frequency

OFDM transceiver implementation are discussed in some de-
tail. Section 3.1 presentsFigure 8. Low-IFof
the concept Receiver Channelfiltering
RX channel Filtering and A/D
and A/D conversion (circle 1 in Fig. 7), Sect. 3.2 deals with
an example of radio impairment compensation (circle 2), and
Sect. 3.3 details on the methodology of designing the FFT
unit for OFDM (de)modulation (circle 3).
The filtering concept just described calls for proper choice
of Conversion
the A/D conversion wordlength WA/D , which can be de-
rived by means of an A/D conversion level budget such as
the one shown in Fig. 9. Starting 6 from the A/D saturation
level determined by the particular semiconductor technol-
ogy (here 600 mVp), some headroom is added to account
 timing/frequency sync estimation

32 S. A. Fechtel: OFDM: From the Idea to Implementation

Figure 7. WLAN OFDM Baseband and RF PHY Transceiver Architecture

Low IF Spectrum Mask

-60
desired signal
interference
noise floor

level [dBm]
-70

-80

-90

-40 -30 -20 -10 0 10 20 30 40

Low IF Analog Filter Response

0

-10

-20
gain [dB]

-30

-40

-50

-60
-40 -30 -20 -10 0 10 20 30 40

Low IF Filtered/Aliased Spectrum

-70

-75 sampled desired signal

sampled interference
-80 aliasing
level [dBm]

noise floor
-85

-90

-95

-100
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
frequency

Fig. 8. Low-IF Receiver Channel Filtering and A/D Conversion.

Figure 8. Low-IF Receiver Channel Filtering and A/D Conversion

6
-4.4 dBV, 600 mVp 1.6 dB DC-Offset (100mV)
-6 dBV, 500 mVp
AGC level
-10 dBV 6 dB variation
-12 dBV, 250 mVp
PAR
(crest factor,
9 dB Vp/Vrms)
-20 dBV -21 dBV, 90 mVrms rms_total
2 dB rms_desired
-23 dBV, 70 mVrms
54 Mbps
dynamic range
~ 6 * W eff Bit

A/D wordlength
-30 dBV effective 8.6 bit
18 dB
physical 10 bit

-40 dBV 6 Mbps rms_adjacent

-41 dBV, 8.9 mVrms
4 dB rms_noise
-45 dBV, 5.6 mVrms

-50 dBV >9 dB quantization

-54 dBV, 2.0 mVrms rms_quant

Figure 9. A/D Conversion Level Budget

Fig. 9. A/D Conversion Level Budget.
 PA
TX compensation DAC TX

BB
S. A. Fechtel: OFDM: From thetest
Idea to Implementation 33
signal TX calibration
test generator
f
0 f0
OFDM TX IQ / DC DAC IF/RF
TX
TX calibration
compensation DAC TX
PA
test measurement test
RF TX DC signal
BB test signal TX TX
BB
signal DC IQ TX calibration TX IQ
test generator
f
0 f0
TX calibration
test f
measurement RF
test f
signal TX0 TX f0 2*f0 0 fc-f0 fc fc+f0
TX DC signal
BB DC IQ
(.)TX2IQ

0
RF/IF
f
f0 2*f0
ADC 0
OFDM
f
fc-f0 fc fc+f0
RX ADC (.)2 RX
RF/IF ADC OFDM
RX ADC RX
Figure 10. Calibration Scheme for Transmitter I/Q and DC Compensation
Figurefor
Fig. 10. Calibration Scheme 10.Transmitter
CalibrationI/Q
Scheme for Compensation.
and DC Transmitter I/Q and DC Compensation

Figure 11. Performance of Low-IF Transmitter I/Q Compensation

Fig. 11. Performance of Low-IF Transmitter I/Q Compensation.

Figure 11. Performance of Low-IF Transmitter I/Q Compensation
for various RF/IF receiver imperfections such as DC off-
set (1.6 dB) and gain control error (6 dB). As OFDM sig-
nals exhibit a Gaussian-like probability density, their peak-
to-average ratio (PAR, 9 dB or more) must be added. The
rms level of the composite (desired + residual interference)
signal is therefore much smaller than the A/D saturation
level (90 mVrms). Depending on the data rate (6...54 Mb/s),
the composite signal may comprise more or less amount
of interference so that the desired signal level may be fur-
ther depressed (here to 9...70 mVrms). The required SNR
(4...22 dB for 6...54 Mb/s) then determines the ambient noise
level (5.6 mVrms). In order to keep the implementation loss
below some predefined limit (here 0.2 dB), the A/D quantiza-
tion level should be lower than the noise level by a consider-
able margin (>9 dB). Summing up all contributions, the A/D
must cover a dynamic range of almost 50 dB, which corre-
sponds to an effective wordlength of 8.6 dB. Further consid-
ering possible nonidealities of the A/D itself, an A/D phys-
ical wordlength WA/D of 10 bit has been chosen for imple-
mentation.
34 Win S. A. Fechtel: OFDM: From the Idea to Implementation
Wout SA = 0 or 1
<<(WT-1-ST) >>SA
ST usually WT-1, may be WT-2
Win (WT-1-ST) usually 0, may be 1
>>ST
Wout SA SA
= 0 and
or 1 (WT-1-ST) not simultaneously 1
<<(WT-1-ST) - >>SA
>>SA
ST usually WT-1, may be WT-2
WT (WT-1-ST) usually 0, may be 1
twiddle
>>ST - >>SA SA and (WT-1-ST) not simultaneously 1

WT
twiddle (twiddle W32) (twiddle W64)
(twiddle W8) (twiddle W16)
stage 1 stage 2 stage 3 stage 4 stage 5 stage 6
(twiddle W8) (twiddle W16) (twiddle W32) (twiddle W64)
(input) (output)
stage
>> 1 stage
>> 2 stage 3 >>
>> stage
>> 4 >> stage
>> 5 >> stage
>> 6 >>
(input) SR(2) SR(3) SR(4) SR(5) SR(6) SR(7) SR(8) SR(9) SR(10) SR(11)
(output)
>> >> >> >> >> >> >> >> >> >>
(1) SR(2)
(2) SR(3)(3) SR(4)(4) SR(5) (5) SR(6) (6) SR(7) (7)SR(8) (8)SR(9) SR(10) (10)
(9) SR(11) (11)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
Figure 12. FFT Architecture
Figure 12. FFT Architecture
Fig. 12. FFT Architecture.

Wordlengths, Average/Maximum Signal Levels, Bittrue Range

14 Wordlengths, Average/Maximum Signal Levels, Bittrue Range
14
12
12
10
10
bits

8
bits

8
6
6
4
4
1 2 3 4 5 6 7 8 9 10 11
1 2 3 4 5 6 7 8 9 10 11
AverageSNR
Average SNR
5555
5050
4545
dB
dB

4040
3535
3030
2525
11 22 33 44 55 66 77 8 8 9 9 10 10 11 11
position index
position index

Fig. 13. FFT Wordlength Optimization and Quantization SNR.

Figure 13. FFT Wordlength Optimization and Quantization SNR
Figure 13. FFT Wordlength Optimization and Quantization SNR
3.2 Radio Impairment Compensation: Transmitter I/Q suppressed in order to ensure standard compliance. To this
9
Mismatch end, the I/Q mismatch and DC calibration scheme
9 shown in
Fig. 10 has been implemented.

The cost-sensitive mass market calls for direct conversion
transceiver architectures. Unfortunately, direct conversion
suffers from nonidealities such as DC offsets caused by lo-
cal oscillator (LO) leakage or I/Q mismatch due to ampli-
tude and/or phase imbalance between the two I (in-phase)
and Q (quadrature-phase) rails carrying the complex-valued
baseband signal. In the transmitter, such nonidealities lead
to the emission of unwanted spurious signals which must be
In calibration mode, a test signal is generated and applied
to TX IQ/DC compensation, D/A conversion, and the IF/RF
transmitter. The upconverted RF signal is then squared and
looped back to the RX A/D converter and from there to the
TX calibration measurement unit, see Fig. 10. It can be
shown that any I/Q mismatch manifests itself as crosstalk
into the opposite sideband, thus compromizing SSB (single
sideband) suppression. Upconversion (to carrier frequency
S. A. Fechtel: OFDM: From the Idea to Implementation
Required SNR [dB]: AWGN (blue), ETSI-A/C (red)
25
20
15
10
0
5 10
Fig. 14. Performance of Implemented WLAN PHY Transceiver.
Figure 14. Performance of Implemented WLAN PHY Transceiver
fc ) of a complex sinusoidal test signal with (programmable)
frequency +f0 therefore yields RF signal components at
fc +f0 (desired), fc (DC offset), and fc −f0 (unwanted spur
from I/Q mismatch). Squaring the RF signal then yields
(real-valued) signal components at f0 (DC) and 2f0 (I/Q)
(left-hand side of Fig. 10) which can be observed by tuning
the measurement unit to these frequencies. The IQ/DC com-
pensator settings can now be adapted such that, after sev-
eral measurement and control cycles, the unwanted signals
are suppressed to the extent that their energy is below some
predefined threshold. The optimal compensator settings thus
found are stored and used during subsequent regular WLAN
operation.
I/Q calibration performance thus achieved is illustrated by
Fig. 11. The IQ/DC-compensated low-IF transmitter (cen-
ter frequency 10 MHz upconverted to 5270 MHz) is seen to
improve SSB suppression from 25...30 dB without compen-
sation to more than 45 dB, thus complying with the 802.11a
TX spectrum mask.
3.3 FFT Design
OFDM modulation and demodulation is most efficiently per-
formed by IFFT and FFT processors, respectively. In our
implementation, both IFFT and FFT units are based on the
same radix-2 pipelined architecture shown in Fig. 12, how-
ever with different wordlengths. The 64-point (I)FFT con-
tains six stages, and each stage consists of a butterfly proces-
sor element (upper part of Fig. 12) with two input and output
ports of wordlength Win and Wout , respectively. Internal but-
terfly operations comprise twiddle multiplication (complex
35
WLAN 802.11a OFDM RF+BB PHY Performance
15 20 25 30 35 40 45 50 55
Data Rate [Mbit/s]
exponential with wordlength WT , complex multiplication,
right shift ST ), a left shift (WT −1−ST ), two complex
additions, and two right shifts SA at the output. In the first
two stages, twiddle multiplications simplify to additions.
As the 64 input signal samples ripple through the six
stages, they are subject to bit-true arithmetic operations
which introduce some amount of quantization noise. So
from one observation point to the next (1 to 11 in Fig. 12),
the quantization noise level continually rises. Focusing
on wordlength design for the OFDM demodulator, the
art of FFT wordlength design consists in minimizing the
wordlengths yet keeping the quantization noise below a cer-
tain target level, or, equivalently, keeping the loss in quanti-
zation SNR below some limit. As the received and filtered
singal may vary in strength, this must hold for a range of
input signal levels. In this design, the average FFT input
bit-true signal level may deviate considerably from its target
value 72 (∼6 bit, medium level) and range between about 12
(<4 bit, low AGC gain, strong ACI, 6 Mb/s mode) and 250
(∼8 bit, high AGC gain, no ACI, 54 Mb/s mode).
In the first step of wordlength optimization, a quantiza-
tion noise model based on Fig. 12 was developed
10 for fast
interactive exploration of the wordlength space. In a sec-
ond step, the most promising configurations thus identified
were verified by simulation. Figure 13 displays the result of
wordlength optimization. In the upper drawing, the progres-
sion of signal levels across the 11 observation points (Fig. 12)
are shown for strong, medium, and weak FFT input signals
(three lower curves). From the wordlengths at positions 1
to 11 (upper curve) it is seen that the optimized FFT ex-
pands the wordlength by only one bit from 11 bit (input)
 Proposal 2x2 (mandatory) or 4x4 MIMO OFDM 270 / 540 Mb/s (4x4, 20/40 MHz)

Gbit/s 5/17/24/60 GHz bands, 100 MHz channel, more than 1 Gb/s
WLAN up to 4x4 MIMO OFDM
36 S. A. Fechtel: OFDM: From the Idea to Implementation

Table 2. WLAN Standard 802.11a/g: PHY Transmission Parameters.

Table 2. WLAN Standard 802.11a/g: PHY Transmission Parameters

information data rate 6, 9, 12, 18, 24, 36, 48, 54 Mbit/s

channel bandwidth = sampling rate 20 MHz, occupied bandwidth 16.6 MHz

OFDM symbol length 4.0 µs = 80 samples

useful symbol length = FFT length 3.2 µs = 64 samples

guard interval length 0.8 µs = 16 samples

number of occupied subcarriers 52

number of data subcarriers 48

number of pilot subcarriers 4

subcarrier frequency spacing 312.5 kHz (=20 MHz / 64)

subcarrier modulation BPSK, QPSK,16-QAM, 64-QAM

forward error correcting code convolutional code, K=7, rates 1/2, 2/3, 3/4

BB PHY IC
RF Transceiver
Diplexer

Power Supply Power Amplifier Switch

Figure 15. First-Generation WLAN PHY Transceiver Board

Fig. 15. First-Generation WLAN PHY Transceiver Board.

to 12 bit at its output, which compares favorably with the 4 WLAN Transceiver Performance
6 extra bits (= number of stages) normally required. Also,
the bit-true saturation level (= wordlength-1) is more than
The performance of the implemented WLAN 802.11a/g
2 bits above the maximum average signal level, thus grant-
transceiver is shown in Fig. 14 in terms of required SNR
ing more than 12 dB peak-to-average ratio (PAR) headroom.
versus PHY data rate 6...54 Mb/s. The lower four pairs
From the lower drawing of Fig. 13, the loss in quantization
of curves refer to subcarrier modulation BPSK, QPSK, 16-
SNR across positions 1 to 11 (again for weak, medium, and
QAM, and 64-QAM, respectively (code rates 1/2 and 2/3 or
strong signal levels) is seen to remain small; the quantization
3/4, Table 2), and the nonselective AWGN channel, whereas
noise introduced by the optimized FFT architecture is about
the upper four pairs of curves refer to the same modu-
20 dB below the ambient noise level and therefore negligible
lation/coding parameters but frequency-selective multipath
at all relevant signal levels.
channels (ETSI channel models A and C valid for typi-
cal office scenarios). Depending on the data rate, an SNR
S. A. Fechtel: OFDM: From the Idea to Implementation
between 8 and 26 dB is required for these channels. The ref-
erence results for ideal floating-point transceiver processing
also shown in Fig. 14 (dashed lines) reveal that the imple-
mentation loss of the realized radio and baseband PHY de-
vice is less than 2 dB, which compares very favorably with
other solutions on the market. Figure 15 displays the cur-
rent WLAN PHY transceiver board housing baseband and
radio transceiver chips, power amplifier, and circuitry for an-
tenna switching and power supply. Next-generation imple-
mentation are targeting one-chip solutions for the radio and
all PHY and MAC processing.
5 Conclusions
Despite the presence of multipath, OFDM modulation – fea-
turing a set of smallband subcarriers and a guard interval –
preserves orthogonality both in time and frequency and thus
greatly facilitates channel equalization. Important aspects of
OFDM radio and baseband transceiver implementation have
been discussed. The concept of sharing receiver channel se-
lection filtering between the analog and digital domains has
a number of advantages but requires harmonizing the de-
signs of analog filter, A/D converter, and subsequent digi-
tal processing. Radio impairments such as transmitter DC
and IQ mismatch are best compensated for by means of dig-
ital predistortion with parameters determined by calibration.
A methodology for optimizing the performance-complexity
tradeoff in OFDM demodulation has been demonstrated on
the pipelined FFT architecture and its wordlengths. These
and other measures were shown to result in less than 2 dB
implementation loss, which is a very competitive figure of
merit for the entire radio and baseband transceiver.
Acknowledgements. The author would like to thank U.R.S.I. and
especially K.-J. Langenberg for the invitation to present this paper.
References
Bingham, J. A. C.: Multicarrier modulation for data transmission:
an idea whose time has come, IEEE Commun. Mag., 37, 5–14,
May 1990.
Coffey, S., Jones, V. K., Hamady, N., et al.: WWiSE IEEE 802.11n
proposal, doc.: IEEE 802.11-04/0935r3, September 2004.
ETSI EN 300 744: Digital Video Broadcasting (DVB); framing
structure, channel coding and modulation for digital terrestrial
television, V1.4.1, January 2001.
ETSI DVB Document A081: Transmission system for handheld ter-
minals (DVB-H), June 2004.
Fechtel, S. and Blaickner, A.: Efficient FFT and equalizer imple-
mentation for OFDM receivers, IEEE Trans. Consumer Electron-
ics, 45, November 1999.
37
Fechtel, S., Schöllhorn, P., Speth, M., Fock, G., and Schotten, C.:
Advanced receiver chip for terrestrial digital video broadcasting:
architecture and performance, IEEE Trans. Consumer Electron-
ics, 44, August 1998.
Foschini, G. J.: Layered space-time architecture for wireless com-
munication in a fading environment when using multi-element
antennas, Bell Labs Technical Journal, 1, No. 2, 41–59, Autumn
1996.
Foschini, G. J. and Gans, M. J.: On limits of wireless communi-
cations in a fading environment when using multiple antennas,
Wireless Personal Communications, 6, 311–335, 1998.
Foschini, G. J., Golden, G. D., Valenzuela, R. A., and Wolnianski,
P. W.: Simplified processing for high spectral efficiency wireless
communication employing multi-element arrays, IEEE Journal
on Selected Areas of Communications, 17, No. 11, 1841–1852,
November 1999.
IEEE 802.11a: Wireless LAN medium access control (MAC) and
physical layer (PHY) specifications: high speed physical layer in
the 5 GHz band, July 1999.
IEEE 802.11b: Standard for wireless LAN medium access control
(MAC) and physical layer (PHY) specifications, 2001.
IEEE 802.11g: Wireless LAN medium access control (MAC) and
physical layer (PHY) specifications further higher-speed phys-
ical layer extension in the 2.4 GHz band, Supplement to IEEE
Standard, 2003.
Jones, V. K., Raleigh, G., and van Nee, R.: MIMO answers high-
rate WLAN call, EE Times, December 2003.
Kammeyer, K. D., Schmidt, H., Rückriem, R., and Fechtel, S.:
OFDM: An old idea solves new problems, in: Proc. Int. Symp.
Theoretical Electrical Engineering (ISTET), 2, Linz, Austria,
August 2001.
Meyr, H., Moeneclaey, M., and Fechtel, S.: Digital communication
receivers, Wiley Interscience, 1997.
Mujtaba, A., Stephens, A. P., Purkovic, A., et al.: TGn Sync com-
plete proposal, IEEE presentation, September 2004.
Pawlowski, S.: Motivation for a reconfigurable wireless architec-
ture, Intel Labs, January 2003.
Pirsch, P.: Architectures for digital signal processing, Wiley Inter-
science, 1998.
Speth, M., Fechtel, S., Fock, G., and Meyr, H.: Optimum receiver
design for wireless broadband systems using OFDM – Part I,
IEEE Trans. Commun., 47, November 1999.
Speth, M., Fechtel, S., Fock, G., and Meyr, H.: Optimum receiver
design for OFDM-based broadband transmission – Part II: A case
study, IEEE Trans. Commun., 49, April 2001.
Vermeer, V. and Cramer, M.: Wireless eyes throughput challenge,
EE Times, December 2003.
Wireless Gigabit with advanced multimedia support (WIGWAM),
Project Proposal, June 2003.
Wolnianski, P. W., Foschini, G. J., Golden, G. D., and Valenzuela,
R. A.: V-BLAST: an architecture for realizing very high data
rates over the rich-scattering wireless channel, 1998 URSI Inter-
national Symposium on Signals, Systems, and Electronics, 295–
300, October 1998.