(5) Advanced Topics in

MIMO-OFDM Systems

gation paths will contribute to the received signal. In such a

severe multipath environments, better performances are
obtained with Orthogonal Frequency Division Multiplexing
(OFDM) than with single carrier modulation used in 3G systems. Furthermore, Multiple Input Multiple Output (MIMO)

Naoto Matoba, Gunther Auer, Gerhard Bauch,

technology which involves the use of multiple antennas can be

Andreas Saul, Katsutoshi Kusume and Satoshi Denno

used to increase the capacity (i.e., speed) of data transmission in

a given frequency band. Therefore, with a view to the future

At DoCoMo Euro-Labs, we are concentrating on studies of

OFDM and MIMO with a view to realizing FourthGeneration mobile communication systems with high-speed
and high-capacity wireless access. When MIMO-OFDM systems are applied to mobile communications, there are numerous issues that have to be addressed, including improvement
of the required signal-to-noise ratio to achieve adequate coverage outdoors, as well as reduction of the signal processing
load, taking into account the processing capabilities of future
mobile devices. In our studies, we are bringing these issues to
light and striving for technologies that can be used to solve
them, such as OFDM channel estimation, multi-user detection, peak power reduction, channel coding, space-time signal processing, and adaptive beamforming.

realization of 4G systems, we are conducting a variety of studies to improve transmission speeds and increase the efficiency
of bandwidth use by focusing on technology relating to MIMO
and multi-carrier schemes such as OFDM.
Specifically, we have proposed a solution for MIMOOFDM channel estimation, which supports high velocities of
the mobile terminal, even at severely distorted channels. For
OFDM with spreading, we have made new proposals for detection schemes aimed at improving the Bit Error Rate (BER).
Furthermore, since multi-carrier systems cause peak power
problems, we have proposed a scheme for analyzing peak
reduction techniques, and a channel coding technique suitable
for OFDM systems.
For MIMO systems, we have proposed a transmit diversity
scheme in which different data sequences are transmitted from
transmitting antennas with temporal, spatial and frequency cor-

1. Introduction

relations. To perform coherent detection in MIMO systems, a

Rapid progress in the development of device technology has

large number of pilot signals are needed for channel estimation.

made it possible for digital equipment to handle diverse multi-

This has an adverse effect on efficiency. We have therefore pro-

media content such as speech, music and video. We envision

posed a new scheme for differential space-time modulation that

that future mobile communications will be able to provide this

makes these pilot signals unnecessary and allows the available

type of content, which can be accessed seamlessly, at higher

bandwidth to be used very efficiently. Also, in the field of direc-

speed and with higher quality. The requirements of Fourth-

tional transmission, we have proposed a beamforming technique

Generation (4G) mobile communications are substantially high-

that uses long term channel knowledge and a space-time signal

er than those of Third-Generation (3G) mobile communications

processing method that improves performence by signal pro-

(IMT-2000: International Mobile Telecommunications-2000).

cessing in time and space directions in transmitter.

In order to meet these requirements it will be necessary to

has higher capacity without sacrificing the features of existing

2. Multi-carrier Systems for Broadband

Wireless Access

2nd and 3rd Generation systems in terms of their service areas,

With the development of Digital Signal Processing (DSP)

implement wireless access technology that is higher speed and

mobile terminal size, ease of use, or cost.

34

technology and the desire for broadband connectivity even over

If the target transmission speed of 4G systems is set to 100

severely dispersive channels, multi-carrier modulation appears

Mbit/s, it will become necessary to use frequency bands of

to be a technology whose time has come. OFDM offers efficient

50100 MHz, which exceeds the bandwidth of current systems [1].

implementation through the application of the Fast Fourier

When transmitting broadband signals, a large number of propa-

Transform (FFT). Moreover, by inserting a cyclic prefix, an

NTT DoCoMo Technical Journal Vol. 6 No.1

equalizer can be omitted. How multi-carrier systems can be

Figure 1 a), for a Binary Phase Shift keying (BPSK) modulat-

implemented in future mobile communication systems is the

ed signal with unrotated spreading, a Spreading Factor (SF) of

subject of current research.

8, and eight users. Here, the area size of each distribution point
corresponds to degree of signal overlapping. Note that, about

2.1 Performance Enhancement of MC-CDMA

20% of all subcarriers are equal to zero, i.e. no signal is trans-

For Multi Carrier-Code Division Multiple Access (MC-

mitted on these subcarriers. On the other hand, a more favorable

CDMA), OFDM is combined with CDMA by spreading the

distribution of the signal space is obtained if rotated transforms

data across subcarriers in the frequency direction. If these

are used, shown in Fig.1 b). The spread signal is distributed on

spread subcarriers are subject to independent fading, diversity

256 points in the complex plane, so each signal point is unique-

can be utilized. That is, if one subcarrier is in a deep fade, its

ly defined. Hence, the signal energy is distributed more equally

information is lost. However, other subcarriers which are

across subcarriers, which allows to exploit diversity more effi-

received with better quality may compensate for that. Since the

ciently. By using rotated transforms, up to 3dB in Eb/N0 for an

information is spread over many subcarriers, diversity can be

uncoded system can be gained. For a coded system with rate of

utilized. On the other hand, orthogonality of the spreading codes

2/3, about 1dB can still be gained, while with a coding rate of

will be lost which causes multiple access interference. By using

1/2 no difference between rotated and unrotated spreading is

a Multi-User Detector (MUD) the performance degradation due

observed [3]. So, it can be concluded that MC-CDMA with

to multiple access interference can be reduced.

rotated spreading sequences is of advantage for high code rates.

We are working on advanced detection schemes to push

detection for MC-CDMA towards the theoretical limit, which is

In [4] we have verified that these gains can also be achieved if

realistic channel estimation is taken into account.

the single user bound. One approach is based on iterative processing, i.e. applying the turbo principle. In particular, we are

2.2 MIMO-OFDM Channel Estimation

investigating a soft interference canceller [2], which cancels the

A received OFDM frame may be viewed as a two

multiple access interference weighted by reliability information

Dimensional (2D) grid, where each field specifies a certain sub-

provided by the channel decoder from the previous iteration.

carrier i and OFDM symbol l. The received signal in each field

The performance of MC-CDMA with MUD can be

is a complex value Yl,i = Hl,i Xl,i + Nl,i.. In order to detect Xl,i, the

improved by using rotated spreading sequences instead of con-

Channel Transfer Function (CTF) Hl,i must be estimated in the

ventional Walsh-Hadamard sequences. If the spreading codes

entire frame. For multi-carrier systems the channel response is

are Walsh-Hadamard sequences, the distribution of the signal

typically correlated in two dimensions, in time and frequency.

space is limited to nine distinct constellation points, as shown in

By invoking the sampling theorem, the 2D channel response

im

im

re

a) Non rotated and spread BPSK symbols of 8 users

re

b) Rotated and spread BPSK symbols of 8 users

Figure 1 Distribution of the spread MC-CDMA signal in the complex plain

35

can be reconstructed by sampling the CTF in the 2D plane.

scheme suitable for MIMO-OFDM systems [6], [7].

This sampling operation is realized by periodically inserting

Furthermore, we have developed an algorithm which can effi-

pilots in the time-frequency grid, termed Pilot-symbol Aided

ciently separate Nt superimposed signals, corresponding to Nt

Channel Estimation (PACE).

transmit antennas, by exploiting the properties of the Discrete

The application of the sampling theorem is intuitively

Fourier Transform (DFT) [8].

appealing, since it yields the minimum possible 2D pilot spacing given the characteristics of a certain channel and the OFDM

2.3 Peak Power Reduction for OFDM

system parameters. In order to implement PACE, 2D filtering

OFDM suffers from high signal peaks, which originate from

algorithms have been proposed based on Wiener filtering.

the superposition of subcarriers. The high power amplifier heavi-

Unfortunately, such a 2D estimator structure may be too com-

ly distorts all signal parts that come close to or exceed saturation.

plex for practical implementation. To reduce the complexity,

The distortion causes Inter-Carrier Interference (ICI) and out-of-

two cascaded 1D estimators in time and frequency may be used

band radiation. While ICI disturbs the transmitted signal and

instead as indicated in Figure 2, termed two times one dimen-

degrades the BER, out-of-band radiation disturbs signals on adja-

sional (2x1D) PACE [5]. In the 1st step, the channel is estimat-

cent frequency bands and should also be avoided.

ed at pilot positions in the first dimension, which in Fig.2 is the

The peaks can be reduced by digital signal processing at the

frequency direction, to yield tentative estimates of all subcarri-

transmitter. Distortion techniques usually clip the peak amplitudes

ers. In the 2nd step, the channel is estimated in the time direc-

of the signal. Furthermore, it is necessary to limit the out-of-band

tion using the tentative estimates of the 1st stage, to yield the

radiation without significant increase of required Eb/N0 by filtering

final estimate.

the clipped signal before amplification and transmission.

For MIMO systems, Nt channels, corresponding to Nt trans-

Figure 3 compares the performance of different clipping

mit antennas, need to be estimated. If PACE is to be used for

techniques for QPSK modulation, namely non-recursive clip-

MIMO-OFDM channel estimation, it requires about Nt more

ping and recursive clipping [9], [10], and Active Constellation

pilot symbols than a comparable single transmit antenna system.

Extension (ACE) [11]. The y-axis shows the out-of-band radia-

Thus, the efficient use of pilot symbols becomes even more

tion. The x-axis informs about the ICI, which corresponds to a

important. We believe that the realization of 2x1D PACE makes

loss in signal to noise ratio Eb/N0, referring to the undistorted

it possible to keep the pilot symbol overhead at an acceptable

signals Eb/N0. Whereas the out-of-band radiation can be easily

level, while maintaining a practicable estimator complexity,

reduced by several decibels, it is quite difficult to achieve a very low

even at high velocities of the mobile terminal. So, 2x1D PACE

out-of-band radiation for any of these clipping techniques. This is

provides an efficient means of putting MIMO-OFDM systems

even more difficult for higher order modulation schemes [12].

into practice. To this end, we have proposed a 2x1D PACE

Subcarrier number
Pilot symbol
Tentative estimate symbol
Final estimate symbol

Estimation in
time direction

Time
Estimation in frequency direction

Figure 2 Principle of 2x1D PACE

36

NTT DoCoMo Technical Journal Vol. 6 No.1

nas are used at the transmitter and

Power Spectral Density (dBc) at f/fs=0.6

30
35
lower clipping ratio
40

Non-recursive clipping
Recursive clipping
ACE, 1 iteration
ACE, 3 iteration

the receiver MIMO [15]. Therefore,

future wireless systems will be
equipped with multiple antennas at
least at the base station. Several

45

strategies exist in order to exploit

50
55

the capacity increase offered by a

better performance

MIMO channel.
If the channel is unknown to the

60

transmitter, the theoretically opti-

65

0
0.2
0.4
0.6
0.8
1
3
Degradation increase of required Eb/N0 (dB) caused by nonlinear distortion at BER=10

mum transmission technique for

high data rate transmission in a

Figure 3 Performance comparison of some clipping techniques for uncoded QPSK

modulation and transmission over an AWGN channel

MIMO channelunder the assump-

tion that the antennas are separated far enough in order to guar-

3. The Turbo Principle for Iterative

Detection of Coded Data

antee independently fading channels from each transmit to each

receive antennais spatial multiplexing. Here, independent

Since their invention ten years ago, turbo codes have revolu-

data streams are transmitted simultaneously from multiple trans-

tionized the field of error control coding [13]. The idea of this

mit antennas. Spatial multiplexing requires at least as many

turbo principle can be applied to a large variety of detection

receive antennas as transmit antennas. Receivers are relatively

problems. We focus on ways to obtain a very flexible (adaptive

complex. Furthermore, the performance of spatial multiplexing

coding rate), powerful but still relatively simple Forward Error

degrades in the case of channel imperfections such as spatial

Control (FEC) coding scheme which can be detected using the

correlations or keyhole effects. A more robust multiple transmit

turbo principle. Particularly, we look at iterative demapping for

antenna technique is spatial diversity which can be obtained

bit-interleaved coded modulation where the Quadrature

using space-time or space-frequency codes. The idea is to trans-

Amplitude Modulation (QAM) demapper is regarded as the

mit the same information via different antennas in such a way

inner constituent code [14] (see Figure 4). However, turbo

that the receiver is able to combine the contributions from all

gains in this scheme are only possible for special mappings of

transmit antennas constructively. The reliability of the wireless

bits to QAM symbols which are different from Gray mapping.

link is improved, which enables higher data rates by using high-

We have developed an algorithm to find such optimized map-

er order modulation or lower coding rates. Orthogonal designs,

pings. With these mappings we obtained similar performance as

Transmitter

with conventional turbo codes at a significantly lower detection

complexity. Moreover, we have considered rate 1 recursive pre-

Interleaving
code

Ck

Ck

QAM/PSK
mapper

coders for further performance improvement without rate loss.

Performance results are shown in Figure 5. Our new mappings
outperform Gray mapping and previously proposed mappings

Turbo receiver

such as set partitioning. The turbo code as used in Universal

Mobile Telecommunications System (UMTS) shows earlier
convergence but a higher error floor. An error floor can be

De-interleaving
c

Le (c
k)

Soft output
demapper

L (ck )

avoided by using a rate 1 precoder.

Le (ck)

4. Multiple Antenna Systems

4.1 Transmit Diversity for MIMO-OFDM
Results from information theory show that the capacity of a
wireless system can be significantly increased if multiple anten-

decoder

Le (c
k )

Interleaving

Figure 4 Configuration of the proposed transmitter for

bit-interleaved coded modulation and the turbo
receiver for iterative demapping and decoding

37

10

1st iteration

BER

10

10th iteration

10

Gray mapping
Set partitioning
Proposed mapping

10

Proposed mapping + recursive precoder

UMTS turbo code
8

10

3
Eb/N0 (dB)

Figure 5 Performance of bit-interleaved coded modulation with iterative demapping

i.e. transmit matrices with orthogonal columns, can be used as a

uted over several antennas. At least, more pilot symbols are

transmit diversity technique where essentially a simple combin-

required. Therefore, differential transmit techniques are attrac-

er at the receiver transforms the fading MIMO channel into an

tive alternatives since they can be detected non-coherently with-

equivalent Single Input Single Output (SISO) channel with a

out any channel estimation. Differential unitary space-time

lower variance of the Signal to Noise Ratio (SNR).

modulation has been proposed in [20]. Here, bits are mapped on

We have proposed methods for the application of orthogo-

a unitary matrix Ck. The transmit matrix is obtained by differen-

nal designs in OFDM or MC-CDMA [16]-[19]. In order to

tial encoding of Ck with the previously transmitted matrix Ck1.

allow a simple receiver which does not produce intersymbol

Unitary matrices are usually constructed based on PSK constel-

interference, the channel needs to meet certain requirements

lation elements. However, the distance properties of M-ary PSK

relating to coherence time and coherence bandwidth. In our pro-

are only advantageous up to M=8. For higher order modulation

posal, the elements of orthogonal designs are mapped to both

it is better to code the information in both phase and amplitude.

different OFDM symbols and neighboring subcarriers taking

Therefore, we have developed a differential multiple transmit

into account coherence time and coherence bandwidth of the

antenna scheme where unitary space-time modulation is extend-

channel. This results in a simple and powerful orthogonal

ed by a differential cyclic amplitude modulation on matrix basis

space-time-frequency coded system.

[21] (see Figure 6). We have derived a completely non-coher-

Furthermore, we have proposed a turbo detector for orthog-

ent soft-output detector for the scheme which in contrast to

onal transmit diversity [18]. Previous publications had shown

most soft-output detectors for differential schemes does not

that due to the orthogonal structure of orthogonal designs no

require any knowledge about the channel state or its statistics.

extrinsic information can be obtained on the symbol level and

The proposed scheme significantly outperforms unitary space-

that, therefore, turbo iterations cannot be effective. Similar to

time modulation at high bandwidth-efficiencies particularly in

the iterative demapping technique described above, we have

time-varying channels or frequency-selective multicarrier chan-

derived optimized mappings of bits to transmit symbols which

nels. At the same time it has lower detection complexity.

enable significant turbo gains even with orthogonal designs at

relatively low complexity.

38

4.2 Adaptive Beamforming

The multiple antenna transmission techniques discussed so

If the transmitter has knowledge of the long-term channel

far require channel estimation at the receiver. Channel estima-

properties, we can apply beamforming techniques in order to

tion is more difficult in multiple transmit antenna systems than

concentrate the transmit energy in dimensions where the infor-

in single antenna systems since more coefficients have to be

mation can be transmitted most efficiently. A powerful beam-

estimated and the transmit power of the pilots has to be distrib-

forming technique is Eigenbeamforming where the antenna

log2 (M1 )
bits uk (1)

Mapping

Ck

Xk

Diversity combiner

Xk

Soft output

NTT DoCoMo Technical Journal Vol. 6 No.1

yt

Yk
CSI

(1)

(2)

L (u
k )

qk

Synthesis
log2 (M2 )
(2)

Mapping

bits uk

Vk

Yk1

Soft output

a Xk1
L (u
k )

Figure 6 Bandwidth-efficient differential space-time modulation

weights are given by the Eigenvectors of the spatial covariance

Detection and Channel Estimation with Rotated Transforms for MC-

matrix. We proposed transmit processing techniques which

CDMA, in Proc. World Wireless Research Forum No. 9 (WWRF #9),

combine Eigenbeamforming and MIMO concepts such as spa-

Zurich, Switzerland, Jul. 2003.

[5] P. Hoeher, S. Kaiser and P. Robertson: Pilot-Symbol-Aided Channel

tial multiplexing and transmit diversity [22]. We derived criteria

Estimation in Time and Frequency, in Proc. Communication Theory

and algorithms for adaptive transmission with the optimum

Mini-Conference (CTMC) within IEEE Global Telecommunications

number of Eigenbeams and optimum power and data rate allocation to the Eigenbeams.
As another possibility to make use of channel knowledge at
the transmitter, we have developed a space-time transmit pro-

Conference (GLOBECOM 97), Phoenix, USA, pp. 9096, 1997.

[6] G. Auer: Channel Estimation for OFDM Systems with Multiple Transmit
Antennas by Filtering in Time and Frequency, in Proc. IEEE Vehicular
Technology Conference 2003-fall (VTC F03), Orlando, FL, USA. Oct.
2003.

cessing which avoids or reduces the effect of interference due to

[7] G. Auer: Channel Estimation in Two Dimensions for OFDM Systems

a delay spread of the channel which exceeds the guard interval

with Multiple Transmit Antennas, in Proc. IEEE Global

[23]. By applying the proposed scheme to the downlink, the

receivers (i.e. the mobile terminals) can be kept simple without
the need for complex operations like equalization.

Telecommunications Conference (GLOBECOM 2003), San Francisco,

CA, USA, Dec. 2003.
[8] G. Auer, A. Dammann and S. Sand: Channel Estimation for OFDM
Systems with Multiple Transmit Antennas by Exploiting the Properties of
the Discrete Fourier Transform, in Proc. IEEE Int. Symposium on

5. Conclusion

Personal, Indoor and Mobile Radio Communications (PIMRC 2003),

Beijing, China, Sep. 2003.

This article summarized the status and achievements of

[9] J. Armstrong: Peak-to-average power reduction for OFDM by repeated

research in wireless communications at DoCoMo Euro-Labs. We

clipping and frequency domain filtering, in Electronics Letters, Vol. 38,

will continue to carry out research on 4G, and we will collaborate

with excellent European research institutes in those fields. We
hope to achieve a global impact and to contribute to the realization of 4G systems.

No.5, pp. 246247, Feb. 2003.

[10] A. Saul: Analysis of Peak Reduction in OFDM Systems Based on
Recursive Clipping, in Proc. Int. OFDM-Workshop, Vol. 1, pp. 103107,
Sep. 2003.
[11] B.S. Krongold and D.L. Jones: PAR Reduction in OFDM via Active
Constellation Extension, in Int. Conf. on Acoustics, Speech and Signal

References
[1] Sawahashi, et al: Broadband Packet Wireless Access, NTT DoCoMo
Technical Journal, Vol. 5, No. 2, pp. 1123, Sep. 2003.
[2] S. Kaiser: OFDM Code-Division Multiplexing in Fading Channels, in
IEEE Transactions on Communications, Vol. 50, pp. 12661273, Aug.
2002.
[3] R. Raulefs, A. Dammann, S. Kaiser and G. Auer: Rotated spreading
sequences for broadband Multicarrier-CDMA, in Proc. IEEE Vehicular
Technology Conference 2003-fall (VTC F03), Orlando, FL, USA. Oct
2003.
[4] R. Raulefs, A. Dammann, S. Kaiser, S. Sand and G. Auer: Multi User

Processing, Vol. 4, pp. 525528, Apr. 2003.

[12] A. Saul: Comparison between Recursive Clipping and Active
Constellation Extension for Peak Reduction in OFDM Systems, in Proc.
Int. Symp. on Wireless Personal Multimedia Communications, Vol. 1,
pp. 3741, Oct. 2003.
[13] C. Berrou, A. Glavieux and P. Thitimajshima: Near Shannon limit errorcorrecting and decoding: Turbo-codes (1), in International Conference
on Communications (ICC), Genf pp. 10641070, May 1993.
[14] F. Schreckenbach, N. Grtz, J. Hagenauer and G. Bauch: Optimized
symbol mappings for bit-interleaved coded modulation with iterative
decoding. IEEE Globecom Conference, San Francisco, USA, Dec. 15,

39

2003.

radio air interface, International Symposium on Wireless Personal

[15] G.J. Foschini and M.J. Gans: On limits of wireless communications in a

fading environment when using multiple antennas, Wireless Personal

Multimedia Communications (WPMC), Yokosuka, Japan, Oct. 1922,

2003.
[20] B. Hochwald and W. Swelden: Differential unitary space-time modula-

Communications, 1998, Vol. 6, pp. 311335.

[16] G. Bauch: Space-time block codes versus space-frequency block
codes, IEEE Vehicular Technology Conference (VTC), Jeju, Korea, Apr.
2225, 2003.

tion, IEEE Transactions on Communications, Vol. 48, No. 12, pp.

20412052, Dec. 2000.
[21] G. Bauch: A bandwidth-efficient scheme for non-coherent transmit

[17] G. Bauch: Space-time-frequency transmit diversity in broadband wireless OFDM systems, 8th International OFDM Workshop, Hamburg,
Germany, Oct. 2325, 2003.

diversity, IEEE Globecom Conference, San Francisco, USA, Dec. 15,

2003.
[22] P. Tejera, W. Utschick and G. Bauch: Pairwise error probability for

[18] G. Bauch and F. Schreckenbach: How to obtain turbo gains in coherent and non-coherent orthogonal transmit diversity, IEEE International
Symposium on Personal Indoor and Mobile Radio Communications
(PIMRC), Beijing, China, Sep. 710, 2003.

MIMO OFDM with transmit partial CSI, ITG Workshop on Applied

Information Theory, Dresden, Germany, Jun. 20, 2003.
[23] K. Kusume: Linear Space-Time Precoding for OFDM Systems based on
Channel State Information, IEEE International Symposium on Personal

[19] A. Dammann, R. Raulefs, G. Auer and G. Bauch: Comparison of spacetime block coding and cyclic delay diversity for a broadband mobile

Indoor and Mobile Radio Communications (PIMRC), Beijing, China, Sep.

710, 2003.

Abbreviations
2D: 2 Dimensiona1
ACE: Active Constellation Extension
AWGN: Additive White Gaussian Noise
BER: Bit Error Rate
BPSK: Binary Phase Shift Keying
CDMA: Code Division Multiple Access
CTF: Channel Transfer Function
CSI: Channel State Information
DFT: Discrete Fourier Transform
DSP: Digital Signal Processing
Eb/N0: Signal energy per bit to background noise power spectrum density ratio
FFT: Fast Fourier Transform
ICI: Inter Carrier Interference

40

IMT-2000: International Mobile Telecommunications-2000

MC-CDMA: Multi Carrier-Code Division Multiple Access
MIMO: Multiple Input Multiple Output
MUD: Multi User Detection
OFDM: Orthogonal Frequency Division Multiplexing
PACE: Pilot symbol Aided Channel Estimation
PSK: Phase Shift Keying
QAM: Quadrature Amplitude Modulation
QPSK: Quadrature Phase Shift Keying
SF: Spreading Factor
SISO: Single Input Single Output
UMTS: Universal Mobile Telecommunications System