More OFDM and CDMA
Orthogonal Frequency Division Multiplexing seems to present a refinement to CDMA on at least the reverse link. Eventually the Q will need to incorporate it? It does not seem to be a sea change improvement or ERICY, who I understand finances some of the OFDM research at the EE department at Berkeley, would never have capitulated to CDMA.
Check out this article and the companies.
Semiconductors
Systems & Software
Design
Technology
People
Broadband developers apply advanced coding to
wireless
By Loring Wirbel
EE Times
(08/05/99, 6:20 p.m. EDT)
DENVER ? Broadband radio developers are taking more risks at using advanced
modulation and coding methods to achieve more efficient use of bandwidth, as
evidenced by several sessions at the IEEE Radio and Wireless Conference
(Rawcon) this week.
Orthogonal frequency-division multiplexing, a coding scheme at the center of a
recent wireless local loop announcement from Canadian radio specialist Wi-LAN
Inc., was demonstrated in several Rawcon papers. Turbo codes, an academic
curiosity for the last six years, also are finding new homes in applications such as
Local Multipoint Distribution Service (LMDS).
And a new coding scheme from a scientist formerly with Scientific-Atlanta Inc.
and CableLabs is poised to take on some of the equalization duties of orthogonal
multiplexing (OFDM). Thomas Williams, founder of startup Holtzman Inc.
(Longmont, Colo.), used both technical-paper presentations and booth floor
space to promote frequency-domain reciprocal modulation as an alternative to
OFDM codes. Williams made some initial presentations at the National
Association of Broadcasters show for using the technique in digital broadcast,
but said the interest at Rawcon came from a much broader development
community.
The buzz over new coding concepts is tied to a larger realization in the wireless
community: that more creative use of DSP power can serve voiceband interests,
in moving to software-defined radio for handling multiple air interfaces in one
platform; as well as broadband radio, allowing faster error correction, better
channelization and greater bandwidth for emerging access systems like wireless
local loop and LMDS.
Turbo codes, a method developed by France Telecom in 1993 for performing
parallel concatenation of multiple recursive convolutional codes, can take error
correction near Shannon limits in many systems, thus holding the potential of
replacing many Viterbi and Reed-Solomon codes in error correction. Several
companies and universities have experimented with turbo-convolutional codes
over the past five years with fruitful results, but the level of excitement rose in
late 1998, as Advanced Hardware Architectures Inc. (Pullman, Wash.)
announced turbo-product-code chip sets developed in conjunction with Efficient
Channel Coding Inc.
At Rawcon, a group from Honeywell Technology Center (Minneapolis) and a
nearby spin-off, Minnetronix Inc., announced the use of turbo-convolutional
codes to implement adaptive-filter receivers that can greatly improve error
correction in CDMA phone systems. Honeywell uses traditional convolutional
methods for decoding the CDMA turbo codes, using a soft-output Viterbi
algorithm for iterative decoding at the receiver.
In optical CDMA systems, a team at Princeton University's EE department, under
Jin Young Kim, used turbo coding in conjunction with binary pulse position
modulation to significantly improve coding gains. The Princeton work has gone
only through software simulation, but Kim hopes to work on implementations of
turbo-coded optical CDMA in the future.
Meanwhile, Advanced Hardware Architectures staff scientist Eric Hewitt
described the use of turbo-product codes for LMDS applications. Pairing the
company's CMOS turbo chip set with the soft-in/soft-out algorithms developed
by Efficient Channel Coding, radio developers could cut the number of
basestations necessary for LMDS Internet access systems, while potentially
reducing the rain fade common to such broadband systems.
Product codes are multidimensional array block codes, which were well-known
but difficult to use before the advent of turbo-concatenation methods. Product
manager David Williams said that executives at Advanced Hardware
Architectures were somewhat surprised that no other DSP specialists have
applied turbo concepts to product codes, since the advantages in
error-correcting chip sets seem more straightforward for turbo products than
turbo-convolutional codes.
The company's first-generation 4501 coding chip, implemented in 0.35-micron
CMOS, potentially can decode data streams at up to 1 Gbit/second. Hewitt said
one advantage in using the turbo-product-code chips in LMDS systems is that a
single decoder can support multiple code and data packet sizes, as well as
multiple packet types, including Internet Protocol and asynchronous transfer
mode. Such coding also can be used with a variety of modulation methods
including quadrature amplitude modulation (QAM) and quadrature phase-shift
keying (QPSK).
For modulation of physical channels, OFDM is finding favor among several
developers. Motorola Inc., which is playing a key role in the U.S. Army's
software-defined radio program, is turning to OFDM modulation to add
high-data-rate capability to its multifrequency, multiband system.
Scott Chuprun of Motorola's systems solution group (Scottsdale, Ariz.) pointed
to a specific commercial incentive for adding OFDM to future spin-offs of the
Model 5004 software-defined radio: Broadband wireless-system companies like
Teledesic LLC would like to use microcellular radios as a last-mile distribution
feeder for its space-based broadband system. The OFDM transceivers could be
implemented in such processor systems as the Analog Devices Sharc family, the
Motorola G4 with AltiVec extensions or the Logic Devices' Dragonfly fast Fourier
transform processor. Using OFDM in conjunction with soft-programmable air
interfaces would make it possible to mix several modulation schemes in one
hardware platform, Chuprun said.
OFDM's use in combating dynamic multipath interference could be challenged
by a new algorithm from startup Holtzman, which is touting frequency-domain
reciprocal modulation (FDRM) to replace adaptive equalizers, and even some
phase-locking, used in a variety of radio systems.
Thomas Williams, founder of Holtzman, said that FDRM uses two consecutive
blocks of data containing the same information, with the second block based on
the frequency reciprocal of the first. The two blocks are sent out in adjacent time
slots, so that echo characteristics will be similar in both.The technology could
eliminate the need to phase-lock local oscillators and do away with the need for
automatic gain control in systems with fast fades, Williams said.
"The advantage of FDRM comes at a price," he said. "I have to send the same
information twice, doubling either the bandwidth or the transmit time." The
potential of simplifying RF/IF chains in many designs, however, was enough to
send many attendees to Holtzman's booth at the conference.
One new concept could show promise for improving equalization in radio
designs. Maximum-likelihood algorithms, used in a variety of communication and
storage applications, are used as nonlinear codes in the time domain in the
maximum-likelihood sequence estimation (MLSE) family of codes. Masami
Akaike, a professor in the Science University of Tokyo's EE department,
suggested in a paper that MLSE could be used in the frequency-domain world,
for a maximum-likelihood spectrum estimation.
By applying Fourier transform techniques, highly effective nonlinear distortion
equalization could be accomplished, he said. Akaike stressed that his work has
been limited to simulations, though he would like to extend it to hardware
implementations in the future.
While most of the Rawcon presentations were good news for programmable and
hardwired DSP developers, some papers showed that integer processors may be
able to take on more duties previously assigned solely to DSPs. The
Massachusetts Institute of Technology's Lab for Computer Science has
launched a SpectrumWare program for creating virtual radios, looking for the
most cost-effective ways to implement soft-programmable handsets.
Doctoral candidate Matthew Welborn presented results on using precomputed
waveforms to handle modulation methods like QAM and QPSK. In essence,
input waveforms are stored in blocks of off-chip memory, allowing a 400-MHz
Pentium II to perform QAM, PAM and other modulation methods at greater
speeds than hardware multipliers. Welborn said that only standard DRAM
access had been studied, though the lookup table methodology employed in the
waveform storage would appear to map well into specialty memories such as
SRAMs and content-addressable memories.
Welborn said that the wave-storage concepts bear some resemblance to direct
digital synthesis, in that both can replace real-time computation in DSP for some
software radios.
Can anyone address what plans QCOM has with OFDM and these attendant technologies. I asked Viterbi about OFDM at the last shareholders' meeting, he shrugged his shoulders, looked surprised, and said, "We must have some patents in the area". |
