Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
A D VA N C E S I N S M A RT A N T E N N A S
6 6
64-QAM
64-QAM
5 5
ayleigh)
Unattainable
it (R
region
N)
capacity lim
WG
N)
16-QAM 16-QAM
WG
it (A
4 4
Channel capacity C (bits)
it (A
y li m
lim
acit
MIMO 4x4
y
acit
Cap
Cap
64-QAM
3 3 turbo
16-QAM
QPSK limit turbo QPSK
2 2
Uncoded Uncoded
R = 1/2
convolutional
Repetition code code Repetition code
R = 1/2 R = 1/2
1 1
bol on the channel corresponds to one coded bit Multilevel Coding — In MLC the QAM symbol
The choice of the (per real dimension). The detection at the receiv- channel is partitioned into M parallel bit chan-
er is trivial, using a matched filter that directly nels, implicitly defined by the bit labeling of the
right channel feeds its soft information output to the outer 2M–QAM constellation. For each bit channel, a
channel decoder. The ultimate Shannon limit is at maximal information rate can be derived; more-
interface, i.e., the E b/N 0 = –1.59 dB, and no matter how low of a over, it can be shown that the sum of the indi-
coding rate we use, we will not be able to achieve vidual information rates is equal to the maximal
mapping of binary reliable communication below that Eb/N0. For ref- information rate of the corresponding QAM
codes to the erence, the performance of several coding
schemes is plotted at a BER of 10–5. Binary con-
channel. Thus, channel coding is reduced to cod-
ing over M binary channels with appropriately
communication volutional codes are characterized by the memory chosen code rates. This strategy can approach
of the code; increasing the memory from 2 to 8, capacity; however, the handling of different cod-
channel, and the as shown in the figure, improves the coding gain, ing rates appears to be inconvenient in practice.
but the decoder complexity grows exponentially.
corresponding With the discovery of turbo codes [3] in 1993 it Bit-Interleaved Coded Modulation — Figure 2 depicts
became possible to approach the AWGN capacity a communication link using BICM. The output
demapping strategy very closely by concatenating simple component of a single channel encoder is bit-interleaved and
at the receiver, codes and using an iterative decoding strategy. mapped onto a QAM constellation. The bit inter-
leaving can be thought of as performing an aver-
is a central BANDWIDTH-LIMITED REGIME aging over different bit channels defined by the
At high SNR or high spectral efficiency, the constellation labeling, resulting in a single effec-
question in digital bandwidth is the precious resource, as usually is tive information rate. Each codeword experiences
the case in wireless cellular systems. For the a maximal number of possible states of the chan-
communications, power-limited case, we were able to simply map nel, approaching the desirable case of an ergodic
the binary coded sequence to antipodal signals. channel, which ensures robust performance. It
even more so when Now, to achieve higher spectral efficiency per can be shown that BICM can operate close to
communicating using transmitted symbol we need to use a larger signal
alphabet, based on either nonbinary codes or
the capacity limit provided that Gray labeling is
used. With Gray labeling, neighboring constella-
multiple antennas. grouping several coded bits of a binary code into tion points differ by only one binary digit. Thanks
one symbol on the channel. Typically, using bina- to the bit interleaver, BICM performs quite well
ry codes is the preferred way, as several efficient over fading channels, which makes it a popular
hardware architectures for encoding and decod- choice for wireless communications. In fact, most
ing are readily available. With multi-amplitude cellular systems as well as satellite and wireless
modulation like pulse amplitude modulation networking systems employ BICM.
(PAM) for real and quadrature amplitude modu- All of the above modulation strategies can be
lation (QAM) for complex (bandpass) signals, we combined with MIMO techniques. As we will
can communicate M b/symbol using 2M/2 ampli- see, BICM in particular is very easy to extend to
tude levels per real dimension (“spectrally effi- the MIMO channel. A first hint of the potential
cient modulation”). The mapping of the binary capacity increase of communicating using multi-
code to the multi-signal-level channel input is not ple antennas (four transmit and receive anten-
straightforward, and several approaches have nas) is given in Fig. 1.
been devised, such as trellis-coded modulation
(TCM), multilevel coding (MLC), and bit-inter- ADAPTIVE MODULATION AND RATE TABLE
leaved coded modulation (BICM) [4, references
therein]. While demapping was trivial for the With impairments such as fading, multiuser
BPSK/QPSK case, there might be several ways of interference, and bursts of noise, the wireless
demapping the signal at the receiver when using communication channel is a much more chal-
multiple amplitude levels, with different complex- lenging transmission medium than cable or the
ity/performance trade-offs. The choice of the idealized AWGN channel. To account for the
right channel interface (i.e., the mapping of binary time- and location-varying behavior of the wire-
codes to the communication channel) and the less channel, many wireless systems support rate
corresponding demapping strategy (also referred control by adaptive modulation — the most sim-
to as detection) at the receiver is a central ques- ple form of feedback. A modulation/coding
tion in digital communications, even more so scheme table (MCS, also referred to as rate
when communicating using multiple antennas. table) defines different combinations of code
rate and modulation rate, with different SNR
Trellis-Coded Modulation — In TCM, channel coding requirements. Depending on the current channel
and modulation are viewed as a single entity, with condition, as reported back from the receiver,
dedicated bit labelings and encoders that generate the transmitter chooses the appropriate MCS
multilevel signals, separating codewords in entry. From Fig. 1 we can infer a few guidelines
Euclidean space rather than in Hamming space as for the design of an MCS table:
is most relevant for binary antipodal (BPSK) trans- • For coding rates of less than about 3/4, the
mission. The added redundancy simply expands QAM input-constrained mutual information
the constellation size to achieve coding gain; thus, limit incurs hardly any loss compared to the
spectral efficiency is increased at no bandwidth capacity limit.
expansion. TCM has turned out to be a highly suc- • A high modulation rate (e.g., 64-QAM) should
cessful technique, and is widely used in voiceband not be used in combination with a low coding
modem standards ITU-T V.32–V.34. It is consid- rate (e.g., rate 1/3), as this would unnecessari-
ered to be the first successful application of chan- ly increase the complexity of the detection, as
nel coding to spectrally efficient modulation. well as put more demanding constraints on
Discovered by
Source Sink P. Elias in 1954,
convolutional codes
Video decoder
Convolutional code
or turbo code Channel Channel or iterative turbo became the
encoder decoder decoder or iterative
or LDPC code Channel as seen
by channel code
LDPC decoder workhorses of
channel coding
Iterative detection
Π–1
Bit interleaver Π Π and decoding for wireless
QAM or vector-QAM
and/or
QAM communications:
Channel Channel with ZF, MMSE,
space/time mapper
and/or mapper demapper APP SMX and/or
space/time-combiner
they operate from
Channel as seen
subset selection by channel interface and/or subset
selection and/or deep space, increase
MRC
the reliability of
Matrix cellular radio,
channel +
H and allow us to
Additive connect to the
NT NR noise
Internet through
■ Figure 2. Communication link with channel coding and different channel interfaces.
WLAN/WPAN.
RF linearity than actually required for com- most prominent results of this effort are the
munication at the respective information rate. codes discovered by I. S. Reed and G. Solomon,
In WLAN IEEE 802.11a, which is a single-anten- with the corresponding algebraic decoding algo-
na system, the MCS table has only eight entries rithm by E. R. Berlekamp and J. L. Massey.
[5]. Obviously, with the inclusion of MIMO tech- Most properties of linear block codes were well
niques, the MCS table has to grow with the num- understood at the beginning of the 1970s.
ber of different antenna modes possible. And
indeed, the high-rate successor system 802.11n [2], CONVOLUTIONAL CODES
which uses MIMO techniques, has more than 100 Discovered by P. Elias in 1954, convolutional
entries, varying from simple non-MIMO/single- codes became the workhorses of channel coding
antenna modes to four spatial streams (transmit for wireless communications: they operate from
antennas), even supporting a different modulation deep space, increase the reliability of cellular
rate per antenna — a complexity increase that has radio — Global System for Mobile Communica-
to be taken into account in system design. tions (GSM), code-division multiple access
(CDMA), Universal Mobile Telecommunicaitons
System (UMTS) — and allow us to connect to
CHANNEL CODES AND THEIR DECODERS the internet through WLAN/WPAN. The encoder
LINEAR BLOCK CODES is a simple shift register with ν memory elements.
The optimal maximum likelihood decoding algo-
Binary linear block codes were the first error rithm was introduced by A. Viterbi in 1967, using
correcting codes, introduced by R. W. Hamming 2ν states per information bit, efficiently organized
in 1950 to improve the reliability of electrome- in a trellis structure. In view of the fact that
chanical computing machines [6]. They can be decoding of linear block codes was dominated by
described by a generator matrix G of dimension hard decision decoding at that time, the availabili-
K × N having binary entries, characterizing the ty of low-complexity soft-input decoding using the
encoding process, or a parity check matrix H of Viterbi algorithm did mean a big step forward.
dimension (K – N) × N, more naturally repre- With the new soft decoding paradigm, no hard
senting the decoding process. A codeword vector decisions need to be made after the demapper,
c is generated by c = uG, and for all codewords but soft values are passed on to the decoder,
c we have HcT = 0, and thus GHT = 0. A linear improving performance by 2–3 dB. The Viterbi
block code is characterized by the minimum algorithm offers an excellent performance/com-
Hamming distance d min, which corresponds to plexity trade-off when no close-to-capacity perfor-
the minimum number of differing digits between mance is required. The performance can be
any two codewords. Large d min means good improved by increasing the memory of the code;
asymptotic error correcting performance, and however, the decoding complexity grows exponen-
the early coding research of the 1950s and ’60s tially in the memory ν, making codes with big
almost exclusively focused on finding codes with memory unattractive for implementation. Table 1
large dmin. Linear block codes are based on sim- provides some further properties of convolutional
ple algebraic descriptions, lending themselves to codes and compares them to iteratively decodable
algebraic hard decision decoding. One of the codes, discussed in the following.
Compact, by generator Compact, by generator polynomials, Sparse parity-check matrix (compact nota-
Code definition
polynomials interleaving rule tion possible through a “protograph”)
Decoder
Linear (BCJR algorithm), iterative
complexity over Linear (Viterbi algorithm) Linear (message passing), iterative decoding
decoding
message length
Simple puncturing of Optimized patterns, or simple pseudo- Simple puncturing results in performance
Different code
single mother code using random puncturing achieves good degradation; best complete parity-check
rates
optimized patterns performance matrix redesign
AWGN 3 dB away from capacity < 1 dB from capacity for big block < 1 dB from capacity for big block lengths
performance for reasonable complexity lengths (104 or higher) (104 or higher)
Convergence Not applicable Typically fast, eight iterations Typically more than 12 iterations
+ +
Serially Concatenated Codes — The classic serially con-
catenated code is composed of an inner convolu- +
tional code and an outer Reed-Solomon code, From +
connected by an interleaver. The interleaver channel +
breaks up bursts of errors at the Viterbi decoder (or demapper)
output, and the Reed-Solomon decoder tries to +
+
clean up any remaining errors. With the success of
+
iterative decoding of parallel concatenated codes,
similar schemes using inner and outer convolu- + +
tional codes were investigated. It quickly became
clear that many components in a digital communi- + dc = 4
cation system can be thought of as serial concate- dv = 1 Edge interleaving
nations, including processing elements such as
equalizers for intersymbol interference channels, N = 8 variable nodes N – K = 4 check nodes
or multiuser detectors. Using an appropriate itera- (repetition codes) (parity check codes)
tive decoding algorithm at the receiver (e.g., iter-
ating over an equalizer and a channel decoder) Parity check matrix
allows further performance improvement. 1 0 1 0 1 0 0 0
0 1 0 1 1 1 0 0
H(N–K)×N =
LOW-DENSITY PARITY-CHECK CODES 1 1 0 0 0 1 1 0
A low-density parity-check (LDPC) code [8, ref-
erences therein] is a linear block code described 0 0 1 1 0 0 1 1
by a parity-check matrix that is sparse (i.e., has
only a few entries set to one). Apart from special Concatenated encoder
subclasses, the code itself does not quite qualify K
as a concatenated code. However, due to the
sparseness of the parity-check matrix, LDPC + +
codes can be decoded efficiently using iterative
decoding, with a structure reminiscent of an iter- K + + N–K
ative decoder for serially concatenated codes. A + +
toy example for K = 4 is shown in Fig. 3. An
inner variable node decoder comprising N vari- + +
able nodes (which can be thought of decoders
for repetition codes) and an outer single parity- ■ Figure 3. Decoder of low-density parity-check (LDPC) code, parity-check
check node decoder performing N – K parity matrix, and encoder for repeat-accumulate (RA) codes, a subclass of LDPC
checks are connected through an edge inter- codes with linear encoding complexity.
leaver, exchanging soft reliability (extrinsic)
information to recover the transmitted message.
Code design is done by choosing variable node eral that it poses a serious design burden for
and check node degree distributions (i.e., a mix- implementation with respect to interconnection
ture of repetition code rates and single parity- network routing and memory size. Luckily, there
check code rates) governing the convergence of are plenty of parity-check matrices to choose
the iterative decoder. This can be accomplished from that describe codes with similar properties
using density evolution [8] or extrinsic informa- — so we can be picky, and select structures that
tion transfer (EXIT) charts [12]. For a regular simplify implementation while hardly degrading
LDPC code with variable node degree dν, corre- performance. For a protograph notation, the par-
sponding to a repetition code of rate 1/d ν, and ity-check matrix H is composed of square matri-
check node degree dc, corresponding to a single ces of dimension P × P. A typical value for P is
parity-check code of rate (dc – 1)/dc, the parity- 50. Each square matrix is either a “permutation
check matrix H has d ν ones per column and d c matrix” Ps, which is obtained from the identity
ones in each row. For an irregular LDPC code, matrix by s consecutive cyclic shifts, or a null
the values for dν and dc follow a certain distribu- matrix. Obviously, the corresponding protograph
tion obtained from code design to optimize the matrix H p provides a P-times more compact
convergence behavior of the iterative decoder. notation: For each permutation matrix only the
cyclic shift index s, with s = 0, …, P – 1, needs
Protograph Notation — The many degrees of free- to be stored. Dedicated hardware architectures
dom in designing LDPC codes, as well as a flexi- have been developed that can efficiently exploit
ble silicon-area vs. clock-frequency trade-off are the parallelism inherent to the protograph struc-
amongst the main reasons why they have attract- ture. An iterative decoding schedule that walks
ed a lot of interest in standardization groups along the rows or columns of Hp, while decoding
over the past few years. An LDPC decoder can P single parity check codes or repetition codes in
be fully parallelized in hardware. However, the parallel, was found to be particularly convenient.
description by a parity-check matrix H is so gen- This is sometimes referred to as “layered decod-
1 1
Bit error rate
below 0.1%
Inner
Extrinsic output from MIMO demapper
0 0
0 1 0 1
A priori feedback from outer channel decoder A priori feedback from outer channel decoder
■ Figure 4. Spatial multiplexing and bit-interleaved coded modulation with iterative detection and decoding for NT ≈ NR (left) and
NT >> NR (right).
When there are many more transmit than receive transmitted from antenna 1 at time instance 2,
antennas (NT >> NR), however, the demapper and from antenna 2 at time instance 3; and so
output improves significantly with increasing a on. This can be extended to any number of trans-
priori knowledge (steep slope), as depicted in mit antennas, corresponding to a rate 1/NT repe-
the right chart; many iterations over a MIMO tition code. The downside is that detection at
detector and an outer decoder are required to the receiver is still rather complex, applying con-
converge toward low BER. Furthermore, the ventional methods known from equalization of
behavior of the channel decoder needs to be ISI channels, such as optimal Viterbi equaliza-
matched to that of the MIMO detector. LDPC tion. Of course, any of the SMX detectors dis-
codes can be specifically designed for this by cussed in the previous section will do the job as
appropriately adjusting the degree distributions, well. So we have suffered a rate loss, giving up
that is, choosing the right mixture of repetition some spectral efficiency to collect some more
code rates and parity-check code rates; this, spatial diversity, but have not gotten simpler
however, complicates system design. detection in return. Does this pay off? Can we
As N T > N R is a quite common scenario in, not get both, the benefit of transmit diversity
say, the downlink from a base station (with many and simple detection at the receiver? Another
antennas) to a mobile terminal (with only a few space-time technique comes to the rescue, com-
antennas), there needs to be a better way of monly referred to as space-time block coding,
benefiting from excess transmit antennas. Simply which is discussed next.
repeating symbols over different transmit anten-
nas sounds, intuitively, like a good idea. Let us SPACE-TIME BLOCK CODING
take a closer look. As with CDD, space-time block codes [13, 14]
can be viewed as repetition codes over space and
CYCLIC DELAY DIVERSITY time, simultaneously transmitting the same data
In cyclic delay diversity (CDD), delayed replicas over different antennas. Similar to MRC, a fad-
of the message are transmitted over different ing channel can be made more AWGN-like
antennas. In a way, it is like deliberately creating using this technique, providing increased robust-
a single antenna intersymbol interference (ISI) ness and range extension. A quite attractive vari-
channel from an otherwise perfectly frequency- ant of space-time codes is orthogonal space-time
flat multiple antenna channel. The upside is that block codes, which can be detected optimally at
all transmitted symbols benefit from the full spa- the receiver with very simple linear operations.
tial diversity, as can be seen from the transmit They enjoy popularity in both wireless cellular
matrix given for NT = 2, (e.g., UMTS) and wireless networking standards
(i.e., 802.11n, 802.16e). The Alamouti orthogo-
s s2 s3 … nal space-time block code for the 2 × 1 channel
S = 1 .
0 s1 s2 … [13] is given by
s − s2*
Constellation symbol s 1 is transmitted from
S = 1 .
antenna 1 at time instance 1, and from antenna s2 s1*
2 at time instance 2; constellation symbol s 2 is
1 a b
AL 1 a,b 2
2 -b* a* or or
1 a b a b 0 0
AL
+ 1 a,b 3 2 -b* a* 0 0 a b
zero
3 0 0 -b* a* -b* a*
1 a b
AL
+ 1 a,b 3 2 -b* a*
rep.
3 a b
1 a b
a,b
AL
+ 2 3 2 -b* a*
SMX
c,d
3 c d or or or
1 a b a b 0 0 0 0
AL 2 0 0 0 0 a b a b
+ 1 a,b 4
zero 3 0 0 -b* a* 0 0 -b* a*
4 -b* a* 0 0 -b* a* 0 0
1 a b
AL 2 -b* a*
+
AL 1 a,b 4
(rep) 3 a b
4 -b* a*
1 a b
a,b
AL 2 -b* a*
+
2 4
AL
(SMX) 3 c d
c, d
4 -d* c*
a,b 1 a b
AL 2 -b* a*
+ 3 c, d 4
SMX 3 c d
e, f 4 e f
■ Figure 6. Extensions of the basic 2 × 1 orthogonal space-time block code to more than two transmit antennas by zero insertion, repeti-
tion, and combinations with spatial multiplexing, as done in WLAN 802.11n.
ing, and we can expect advances in this field over cial systems) has now turned into something
the next few years. called cognitive radio — and this time it seems it
is becoming real. In cognitive radio [17] it is all
Cognitive Radio — What used to be software radio about multidimensional spectrum reuse: in
(which has never has really taken off in commer- space, in time, in frequency, breaking free of
With the ever- most of the constraints that hinder further opti-
mization of today’s wireless systems. A cognitive
single antenna channels. With the ever increasing
demand for faster wireless connectivity, even the
increasing demand radio can sense its environment and location, seemingly vast bandwidth available for UWB com-
learn from its measurements, and react by adapt- munication will become a bottleneck — thus, it is
for faster wireless ing power, frequency, bandwidth, MIMO mode, likely that a further evolution of current single
modulation, coding, protocol stack, and several antenna systems toward including MIMO tech-
connectivity, it is other parameters of its transmission. The auto- niques is just a matter of time.
matic channel switching of WLAN 802.11 can
likely that a further already be regarded as a first humble step toward REFERENCES
evolution of current an environment-aware radio. But there is more
on the horizon: In 2002 FCC opened up the
[1] C. E. Shannon, “A Mathematical Theory of Communica-
tion,” Bell Sys. Tech. J., vol. 27, July/Oct. 1948, pp.
379–423, 623–56.
single antenna spectrum from 3.1 to 10.6 GHz for unlicensed [2] “Joint Proposal: High Throughput Extension to the
ultrawideband (UWB) communications — across 802.11 Standard: PHY Draft 1.0,” Task Group n, doc.
systems toward several already existing licensed and unlicensed IEEE 802.11-05/1102r4.
bands. Although the allowed transmit levels are [3] C. Berrou, A. Glavieux, and P. Thitimajshima, “Near
including MIMO extremely low (–41 dBm/MHz), other regulatory Shannon Limit Error-Correcting Coding and Decoding:
Turbo-Codes,” Proc. IEEE Conf. Commun., May 1993,
domains, like Japan and Europe, demand special
techniques is just a detect-and-avoid (DAA) modes, the details of
pp. 1064–70.
[4] G. D. Forney, Jr. and G. Ungerboeck, “Modulation and
Coding for Linear Gaussian Channels,” IEEE Trans. Info.
matter of time. which (e.g., threshold levels and response times)
are still to be defined. The basic idea, however, Theory, IT-44, Oct. 1998, pp. 2384–2415.
[5] “Part 11: Wireless LAN Medium Access Control (MAC)
is simple: The device has to detect victim sys- and Physical Layer (PHY) Specifications, High-Speed
tems, like WiMAX or WLAN services, using Physical Layer in the 5 GHz Band,” IEEE Std 802.11a-
spectral measurements, and avoid interference 1999.
[6] D. J. Costello et al., “Applications of Error-Control Cod-
by notching out the respective frequency compo- ing,” IEEE Trans. Info. Theory, vol. 44, no. 6, Oct. 1998,
nents from its transmit spectral mask. This allows pp. 2531–60.
to satisfy different regulatory domains, and to [7] J. Hagenauer, E. Offer, and L. Papke, “Iterative Decoding of
coexist with the primary users of the respective Binary Block and Convolutional Codes,” IEEE Trans. Info.
Theory, vol. 42, no. 2, Mar. 1996, pp. 429–45.
spectrum. Depending on the flavor of UWB, [8] T. J. Richardson and R. L. Urbanke, “The Capacity of
DAA might be more or less difficult to realize. Low-Density Parity-Check Codes under Message-Passing
The UWB system according to the WiMedia Decoding,” IEEE Trans. Info. Theory, vol. 47, no. 2, Feb.
standard [18] has a built-in OFDM receiver that 2001, pp. 599–618.
[9] J. H. Winters, J. Salz and R. D. Gitlin, “The Impact of
is already capable of performing Fourier trans- Antenna Diversity on the Capacity of Wireless Commu-
forms to decode the OFDM symbols. The nication Systems,” IEEE Trans. Commun., vol. 42, no.
respective frequency domain processor just 2/3/4, Feb./Mar./Apr. 1994, pp. 1740–51.
needs some further tuning to provide accurate [10] G. J. Foschini, “Layered Space-Time Architecture for
Wireless Communication in A Fading Environment
spectral energy measurements. What cognitive When Using Multi-Element Antennas,” Bell Labs. Tech.
radio means for channel coding is less obvious, J., vol. 1, no. 2, 1996, pp. 41–59.
but it is something like “one-code-fits-all”: The [11] G. D. Golden et al., “Detection Algorithm and Initial
code should be flexible with regard to block Laboratory Results using V-BLAST Spacetime Communi-
cation Architecture,” IEE Elect. Lett., vol. 35, Jan. 1999,
length and code rate, and should be robust pp. 14–16.
against transmit notches, or support simple [12] S. ten Brink, G. Kramer, and A. Ashikhmin, “Design of
means to “code around” those nulled subcarriers Low-Density Parity Check Codes for Modulation and
in combination with a good interleaving strategy. Detection,” IEEE Trans. Commun., vol. 52, no. 4, Apr.
2004, pp. 670–78.
It should be simple to adapt to different channel [13] S. M. Alamouti, “A Simple Transmit Diversity Tech-
interfaces, like different numbers of transmit nique for Wireless Communications,” IEEE JSAC, vol.
antennas and modulation formats, and provide 16, no. 8, Oct. 1998, pp. 1451–58.
excellent performance in conjunction with itera- [14] V. Tarokh, H. Jafarkani and A. R. Calderbank, “Space-
Time Block Codes from Orthogonal Designs,” IEEE
tive detection and decoding. A lot of creativity Trans. Info. Theory, vol. 45, July 1999, pp. 1456–67.
has already gone into this, and more is to come. [15] A. Goldsmith et al., “Capacity Limits of MIMO Channels,”
IEEE JSAC, vol. 21, no. 5, June 2003, pp. 684–702.
SUMMARY [16] W. Yu, J. M. Cioffi, “Sum Capacity of Gaussian Vector
Broadcast Channels,” IEEE Trans. Info. Theory, vol. 50,
no. 9, Sept. 2004, pp. 1875–92.
This article is an attempt to shed some light on [17] S. Ashley, “Cognitive Radio,” Sci. Amer., Mar. 2006,
wireless communication using multiple antennas, as pp. 66–73.
seen from the perspective of the channel code. [18] “MultiBand OFDM Physical Layer Specification,” WiMe-
dia Alliance specification document, ver. 1.1, May 26,
Extending simple bit-interleaved coded modulation 2005, Also, “High Rate Ultra Wideband PHY and MAC
to the matrix channel, with a single outer encoder Standard,” ECMA-368.
and an inner vector-QAM mapping at the transmit-
ter, appears to be the most pragmatic choice for BIOGRAPHY
materializing a significant portion of the promised S TEPHAN TEN B RINK (stenbrink@realtek-us.com) received a
high spectral efficiencies. Alamouti’s unique orthog- Dipl.-Ing. degree in electrical engineering and a Dr.-Ing.
onal space-time block code in combination with degree from the University of Stuttgart, Germany, in 1997
repetitions and zero insertions as extensions to and 2000, respectively. From 2000 to June 2003 he was
with the Wireless Research Laboratory, Bell Laboratories,
more than two transmit antennas makes the chan- Lucent Technologies, Holmdel, New Jersey, conducting
nel more robust against fading and interference. research on channel coding for multiple-antenna systems.
The commercial application of multiple antenna Since July 2003 he has been with Realtek Semiconductor
techniques has just begun. While WLAN 802.11n Corp., Irvine, California, where he is involved in the devel-
opment and standardization of high-throughput wireless
has moved to MIMO just recently, many other sys- systems. His research interests include concatenated coding
tems, like ultrawideband WPAN systems, are still and iterative decoding, multiple-antenna communications,
based on spectrally inefficient communication over cognitive radios and watermarking schemes.