Kiến trúc phần mềm Radio P14 ppsx

Softwar e Radio Arc hitecture: Object-Oriented Approac hes to Wireless Systems Engineering
Joseph Mitola III
Copyright
c
!2000 John Wiley & Sons, Inc.
ISBNs: 0-471-38492-5 (Hardback); 0-471-21664-X (Electronic)
14
Smart Antennas
Smart antennas are an important application of SDR technology [381]. An
in-depth treatment is beyond the scope of this chapter. The objective is to
introduce the topic to identify the implications of smart antennas for software-
radio architecture.
The smart antenna is a logical extension of antenna diversity described
above [382]. Smart antenna arrays integrate the contributions of spatially dis-
tributed antenna elements to provide wireless communication systems with
larger capacity and higher link quality through frequency reuse and cochannel
interference suppression [383, 384]. Since smart antennas require an order
of magnitude more IF and baseband digital processing capacity than a con-
ventional receiver, the smart antenna base station is “90% antenna.” Contrast
this to a conventional base station, which is only “10% antenna,” including
diversity processing. The rate of proliferation of the smart antenna technology
in the commercial sector has been slow because of the cost of this increase in
capability.
I. SMAR T ANTENNA DOMAINS
Four applications domains attract investments in smart antenna technology
as illustrated in Figure 14-1. Historically, military radar and communications
jamming laid the foundations of smart antenna technology. Investment lead-
ership has shifted to commercial terrestrial networks, however. For example,
a smart antenna with per-subscriber AOA estimation, interference differen-
tiation, and coherent multipath combining was demonstrated for AMPS in
1994 [385, 386]. In add ition, GSM infrastructure is amenable to smart an-

tenna applications [387]. In the future, 3G base stations with W-CDMA 1 : 1
frequency reuse also should benefit from this technology [388].
Military applications remain substantial. Technology for beamforming
on transmit for communications, for example, was sponsored by DARPA’s
GloMo program [389]. Academic interest is growing in the area of joint trans-
mission and reception diversity via smart antennas [390, 391]. Since the
smart antenna places a null on interference [385], the military could use
this COTS technology to reduce jamming effects. In addition, both military
[392] and commercial satellite communications terminals benefit from rapid
electronic beam steering [393] and o ther features of smart antennas, such as
overcoming light and heavy shadowing [394]. This chapter provides an over-
467
468
SMART ANTENNAS
Figure 14-1
Smart antenna domains.
view of the relationship between smart antenna technology and software-radio
architecture.
Levels of smart antenna technology, in order of increasing cost and com-
plexity, include:
1. Multibeam antennas to enhance SNR [395]
2. Null-forming to reduce interference in high traffic density [385]
3. Space-time adaptive processing to jointly equalize the spatially enhanced
signals [396]
4. SDMA via joint beamforming, null pointing, and equalization [397]
II. MULTIBEAM ARRAYS
The concept of operations of a multibeam antenna is illustrated in Figure
14-2. Conventional sectorized antennas cover the bulk of this notional subur-
ban area that includes an interstate highway system. Each conventional antenna
has three 120-degree sectors with frequencies assigned according to the air

interface standard’s frequency reuse plan (e.g., 1/7 for AMPS, 1/3 for GSM,
1/1 for CDMA, etc.). An area between the highways includes a high-density
commercial zone that generates high-intensity traffic.
MULTIBEAM ARRAYS
469
Figure 14-2
Multibeam array concepts.
The service-provider has only a few alternatives. If the intensity level is
several times the design capacity of the conventional sector, additional capacity
must be provided. Several additional smaller cells could be provided in the
high-intensity area. This requires the acquisition of the sites and establishing
connectivity between the new sites and the provider’s existing infrastructure.
In some areas, the opportunities to establish sites are limited and/or the cost
of backhaul from the sites is high. The multibeam antenna alternative creates
additional smaller sectors, each of which has a conventional fixed-frequency
assignment. The physical layout of the multibeam alternative is as illustrated
in Figure 14-2. In the notional highway scenario, the subscriber signal is
switched to the beam with the best CIR via high-speed analog or digital beam
switching [398]. Such a fixed multibeam antenna may use sector beamforming
technology, such as a Butler matrix [399, 400]. Figure 14-3 illustrates the
contemporary Butler matrix technology.
In spite of the level of maturity of multibeam array technology, research
challenges remain. For example, the complexity of the multibeam array tech-
nology is high, keeping costs high. The ADAMO (ADaptive Antennas for
MObiles) project, for example, addressed this challenge with a circular array
of patch antennas [401] and low-complexity analog processing. The bench-
mark set for this project is to suffer only small performance degradations
compared to (macroscale) digital techniques.
In addition, Thomson-CSF has developed prototype antennas for the eval-
uation and qualification of the SDMA concept in the field of UMTS radio

communications under contract to CNET/France TELECOM [402]. Figure
14-4 shows prototype SDMA hardware. In general, SDMA may employ multi-
beam arrays, digital beamforming, joint beamforming-equalization, and other
smart antenna techniques. As a practical matter, however, the costs of SDMA
products must be kept low in order to be affordable to infrastructure opera-
tors.
470
SMART ANTENNAS
Figure 14-3
Illustrative multibeam technology.
Figure 14-4
SDMA antenna prototypes.
III. ADAPTIVE SP ATIAL NULLING
If the multibeam array has a dozen beams, it may not be feasible to assign a
complete frequency-reuse plan to each beam. This is because of interference
with adjoining sectorized antennas. In such situations, it may be useful to
ADAPTIVE SPATIAL NULLING
471
Figure 14-5
Smart antennas complement con ventional sectors.
cancel interference by creating spatial nulls in the direction of nonsubscriber
signal components.
Figure 14-5 illustrates the deployment concept for a smart antenna with
spatial nulling. As subscribers that are on the same frequency (cf. cochan-
nel subscribers) move through the h igh-traffic-intensity area, nulls track their
movement and cancel their path components. The architecture of such a spa-
tial nulling subsystem (e.g., [385]) is illustrated in Figure 14-6. This smart
antenna replaces a conventional sectorized array, interfacing to the cell site
via the existing RF distribution system. The three 3-element sectors of a con-
ventional sectorized base station have been replaced with eight circularly dis-

posed antenna elements. The signal is preamplified and converted to digital
form by a bank of eight wideband ADCs. The angle of arrival of all incom-
ing signal components is estimated by a super-resolution DF algorithm [403,
404]. Since the DF algorithm requires a few milliseconds to compute its esti-
mates, the eight raw ADC streams are delayed so that the digital beamformer
weights correspond exactly to the received signal. Subscriber channels are
then isolated (e.g., using a bank of digital filter ASICs). The measurement
of the
supervisory audio tones
(SAT) is one of the AMPS-specific baseband
algorithms implemented in a pool of DSPs. The out-of-band SAT generated
by the basestation is transponded by the mobile. The basestation can there-
fore differentiate its subscribers from cochannel interference based on SAT.
The cross-correlation process determines the delay-azimuth parameters needed
for the final beamforming-equalization stage. The resulting 100 signals from
472
SMART ANTENNAS
Figure 14-6
Spatial nulling architecture.
the base station’s subscribers exhibit enhanced CIR. These are digitally mul-
tiplexed by adding the signals in a high-dynamic-range numerical process.
Finally, they are converted to analog and sent to the base station.
A. Algorithm Operation
This section illustrates the operation of such spatial-nulling antenna systems.
The exposition is similar to that of Kennedy and Sullivan [385]. The spatial
distribution of a wavefront arriving at a smart antenna is illustrated in Fig-
ure 14-7. The power-delay profile (a) shows the autocorrelation of a single,
direct-path wavefront arriving from a single direction (b). Multipath reflec-
tions will generally exhibit some time-delay with respect to this principal
component. The azimuth display helps visualize the distribution of energy in

space.
When multipath components are present, they are delayed with respect
to the principal component as illustrated in Figure 14-8a. In addition, the
multipath components are not collinear with the direct path and they usually
have less signal strength than the direct path as seen in Figure 14-8b.
When interference is present, it is mixed with the multipath components
as illustrated in Figure 14-9. In this case, the interference is not on the same
azimuth as the direct-path, so it may be suppressed by pointing a null in the
appropriate direction.
ADAPTIVE SPATIAL NULLING
473
Figure 14-7
Principal component distributions.
Figure 14-8
Two multipath components.
Adjustments to the weights of the beamforming matrix yield the kind of
response illustrated in Figure 14-10. Although the depth of the null exceeds
30 dB, a residual remains. The CIR, however, has been improved by 3 to 6 dB.
The simple beamformer does not equalize the received multipath compo-
nents. Such a process would delay the signal components with respect to each
474
SMART ANTENNAS
Figure 14-9
Multipath and interference.
Figure 14-10
Illustrative array manifold response.
other so that they may be combined, further enhancing the SNR. The smart
antenna described by Kennedy includes baseband equalization in each sub-
scriber channel. This is an example of a spatial beamformer followed by a
temporal equalizer in which each stage operates independently. In space-time

adaptiv e processing (STAP) the beamforming and equalization parameters are
calculated jointly.
SPACE-TIME ADAPTIVE PROCESSING
475
TABLE 14-1 Beamforming Algorithm Complexity
Algorithm Multiplications Divisions Additions
LMS 2
Q
+1 0 2
Q
RLS 2
Q
2
+7
Q
+5
Q
2
+4
Q
+3 2
Q
2
+6
Q
+4
FTF 7
Q
+12 4 6
Q

+3
LSL 10
Q
+3 6
Q
+2 8
Q
+2
Adapted from [405]
c
!IEEE 1999, with permission.
B. Beamforming Algorithm Complexity
Cellular systems structure signals such that base stations can differentiate sub-
scribers from cochannel interference. In the case of AMPS, the interference
would have a different SAT frequency. In the case of CDMA, the interference
has a different placement on the long-code. In the case of GSM, the burst has
different header bits. In both of these latter cases, the individual path com-
ponents could be demodulated in order for the system to differentiate signal
from interference. This would be computationally expensive, but might be un-
avoidable. Researchers have therefore sought less computationally intensive
algorithms.
In particular, Razavilar et al. [405] analyzed the computational aspects of
beamforming algorithms that use training sequences.
Direct matrix inversion
(DMI) is the simplest method for calculating beamforming weights based on
a known training sequence of length
Q
. Its complexity is on the order of
Q
3

,
where
Q
is the length of the training sequence. Adaptive algorithms iterate
the weights as the training sequence is received, yielding an estimate at the
conclusion of the training sequence. Razavilar characterized the complexity
of the following algorithms:
least mean square
(LMS),
recursive least square
(RLS),
fast transversal filter
(FTF), and
least squa res lattice
(LSL). Complexity
in terms of
Q
is given in Table 14-1.
IV. SPACE-TIME ADAPTIVE PROCESSING
At times, a cochannel interferer will also be collinear with the subscriber
and the base station. This situation cannot be corrected spatially: deep nulls
cancel both the interference and the desired signal. These two signals are
not likely to be mutually coherent in the time domain, however. Joint space-
time adaptive processing (STAP) uses this lack of coherence to separate the
signals in parameter-space. This allows one to cancel such collinear interfer-
ence.
A STAP array includes a tapped delay line in each antenna element’s
processing channel [406, 407], as illustrated in Figure 14-11. The matrix
W
ij

transforms the signal from multiple antenna elements into a space-time-
equalized signal. Those w eights in part reflect array normalization, weights
476
SMART ANTENNAS
Figure 14-11
Conceptual structure of STA P.
that correct differences in the magnitude and phase transfer functions be-
tween antenna elements. Such differences arise because the corresponding
analog signal processing paths are not perfectly matched. Those weights
also reflect the placement of spatial nulls. Moreover, they reflect the inver-
sion of matrix equations that compensate for relative time delays (or equi-
valently for relative phase differences) of the multipath components. STAP
therefore generally requires computationally intensive matrix factorization
[406].
Matrix in version substantially increases the processing requirements, but
yields improved performance. Consequently, many techniques have b een in-
vestigated either to reduce the computational burden of optimal STAP al-
gorithms, or to enhance the cancellation capability of simpler algorithms. A
taxonomy of smart antenna techniques is provided in Figure 14-12. In the fig-
ure, array algorithms require more than one statistically independent antenna
element. Highlights of the techniques are as follows.
In
sequential interference cancellation
(SIC), the highest-power signal is de-
modulated to estimate its bitstream. The bitstream is then remodulated and
filtered to form an idealized replica of the analog signal in the channel.
The subtraction of this replica from the composite input stream yields a
residue that includes the remaining users. The cochannel interference from
the strongest interferer has been reduced substantially. This recovery process
continues, recovering multiple users in turn. In CDMA applications, most of

the signals thus recovered are likely to be from viable users, because of soft-
handoff.
Minimum mean squared error
(MMSE) processes estimate signal parame-
ters using a Gaussian noise error model [407].
Maximum likelihood
(ML) tech-
niques formulate the likelihood ratio (for a sequence of channel symbols), the
maximum of which determines the signal parameter estimate. One variation
of this is also called
maximum likelihood sequence estimation
(MLSE) [408].
ARCHITECTURE IMPLICATIONS
477
Figure 14-12
Scope of smart antenna algorithms.
SIC, ML, and MMSE may be employed on a single channel, on a multichannel
array antenna, and in conjunction with other techniques.
The simplest STAP algorithm is the joint equalizer-beamformer. Joint beam-
forming and equalization maximizes received signal strength by coherently
combining the multipath components while placing spatial nulls on sources
of interference [409–411]. Advanced STAP techniques include multichannel
SIC [412] and other multichannel adaptive techniques [413, 414]. The perfor-
mance of smart antennas degrades as a function of the structure of the mul-
tipath environments [415].
Parallel interference cancellation
(PIC) techniques
structure the computations so that multiple subscriber signals are processed
at the same time. The parallelism uses an amplitude estimate for each user as
a soft-decision metric. The decision-biases thus introduced can be canceled

using
partial interference cancellation
(also PIC) [416]. Parallel algorithms
distribute well onto parallel DSP hardware [418].
In general, multichannel SIC and MMSE outperform the other approaches,
but they are one to four orders of magnitude more computationally inten-
sive. Additional treatment of the relationship between concrete computational
complexity and interference cancellation effectiveness in real environments
is needed. The further quantification of resource demands of candidate algo-
rithms is essential to insertion in software-radio architectures.
V. ARCHITECTURE IMPLICATIONS
The STAP algorithms are the most computationally intensive algorithms in-
vestigated to date for canceling cochannel interference. These algorithms re-
quire several orders of magnitude more processing capacity than digital beam-
formers, which require an order of m agnitude more processing capacity than
single-channel architectures. Thus, digital circuits of envisioned receivers for
478
SMART ANTENNAS
Figure 14-13
Smart antenna architecture.
some STAP algorithms may require a minimum of 50 GOPS of processing
capacity, with ASIC gate-counts of 100,000 or more [417]. Murotake’s anal-
ysis determined that 12.3 GFLOPS is required for each channel of a 5 MHz
W-CDMA modem [418]. The conclusion was that a 60 GFLOP configuration
would support a W-CDMA smart antenna using PIC algorithms. Because of
such high computational burdens, both the reduction of computational com-
plexity and the enhancement of processing platforms have received attention.
Low-cost DSP architectures are suitable for laboratory investigations of such
smart antenna algorithms [419].
A. Smart Antenna Components

The introduction of smart antennas into SDR implementations is accommo-
dated by software radio architecture. Figure 14-13 illustrates the functional
organization of a smart antenna. Delay and estimation processes vary from
algorithm to algorithm, so these blocks would be connected into the signal
flow paths as a function of algorithm.
The organization of DSP components for smart antennas may be based
on the diversity platform developed in prior chapters. The block diagram of
the (physical components) reference platform for this class of smart antennas
is given in Figure 14-14. Many possible signal flows may be implemented
on such a reference platform. In an
N
-element array, the channel isolation
filters extract channels for each of
K
subscribers on each of
N
elements.
These are processed to form beams and to extract first-stage soft-decision
ARCHITECTURE IMPLICATIONS
479
Figure 14-14
Smart antenna reference platform.
parameters. The channels with low CIR are thus identified. Their bulk-delayed
signals may be isolated for the second and higher stages of interference can-
cellation, performed in a logically separate segment of the DSP pool. This
pool also provides the processors for modulation and predistortion, which can
include beamforming on transmission. The node’s switching functions may be
implemented by addressing on the low-speed bus, sized to interconnect the
K
users locally and/or to the PSTN via other elements of the bitstream segment.

B. Design Rules
1. Digital Base Station Interface
The analog RF interface to the cell site is
one of the problematic aspects of the introduction of smart antenna technol-
ogy. The D AC, filtering, block up-conversion, and the RF distribution steps of
Figure 14-6 all add noise to the enhanced signals from the smart antenna sub-
system, countering some of the gains of the smart antenna. To help overcome
this problem, the SDR Forum has defined a digital interface between a smart
antenna and the core base station. This interface is defined between the Mo-
dem functional block and the INFOSEC functional block of the architecture.
More precisely, this Decoded Channel Bits Interface (DCBI) defines a point in
an SDR that divides a future base station into a smart antenna subsystem and a
core base station. The smart antenna includes the primary RF of the base sta-
tion, the wideband ADC and DACs, IF processing, smart antenna processing
480
SMART ANTENNAS
(e.g., beamforming, interference cancellation, and equalization), and demod-
ulation. Any soft-decision decoding, Trellis coding/decoding, etc. required to
decode the channel bits is in the domain of the smart antenna. T his leaves the
majority of the bitstream processing (bit interleaving, FEC, turbocoding, etc.),
speech processing (e.g., GSM TRAU processing), and data processing (e.g.,
billing-related, logging, operations support, maintenance diagnostics, etc.) in
the domain of the core base station.
2. Business-oriented Design Rules
The identification and adoption of such
interface standards promotes open architecture. During the deliberations of the
SDR Forum on this interface, base station manufacturers became resistant to
the publication of the DCBI standard. This reflects the business reality that
the introduction of smart antenna technology could restructure the basestation
marketplace. The smart antenna is currently positioned in a conflict between

business-oriented design rules. One rule that base station suppliers have to fol-
low is to protect and enhance their business base. A rule that service providers
like to follow is to sustain competition. The DCBI sustains competition, but at
the possible expense of established base station incumbent suppliers. It seems
likely that established suppliers will incorporate smart antenna technology into
next-generation base stations, introducing the “smart base station.” Such prod-
ucts reduce the need for a digital interface between smart antennas and the
core base station.
At some point, these issues will be resolved. Adaptive nulling provides
some enhancement to CIR, while STAP techniques improve CIR further. The
better interference suppression may be necessary to achieve the information
densities planned for 3G. It therefore seems likely that smart CDMA base
stations will proliferate with 3G deployments.
VI. EXERCISES
1.
Define smart antenna. Differentiate STA P from other smart-antenna tech-
niques.
2.
How can SIC be used in a single-antenna configuration? What additional
complexities arise from multiple-antenna applications of SIC?
3.
Express the smart antenna reference platform in a table (i.e., do not use a
block diagram). Be sure to differentiate classes o f DSP pool.
4.
How can smart antennas enhance quality, quantity, or timeliness of com-
munications?
5.
Can smart base stations introduce asymmetries into the link to the mobile
subscriber? What are the alternatives for closing the link in spite of the
asymmetries? What are the adv antages and disadvantages of each alterna-

ti ve?
EXERCISES
481
6.
What are the alternatives to smart antennas for enhancing CIR?
7.
Describe a smart antenna architecture that overcomes the business issues
of technology insertion.
8.
Apply a smart antenna to the Disaster Relief case study. In which scenarios
is the added cost worth it?