TOPICS IN WIRELESS COMMUNICATIONS
`
`A Fourth-Generation MIMO-OFDM
`Broadband Wireless System: Design,
`Performance, and Field Trial Results
`
`Hemanth Sampath, Shilpa Talwar, Jose Tellado, and Vinko Erceg, Iospan Wireless Inc.
`Arogyaswami Paulraj, Iospan Wireless Inc. and Stanford University
`
`ABSTRACT
`Increasing demand for high-performance 4G
`broadband wireless is enabled by the use of multi-
`ple antennas at both base station and subscriber
`ends. Multiple antenna technologies enable high
`capacities suited for Internet and multimedia ser-
`vices, and also dramatically increase range and
`reliability. In this article we describe a multiple-
`input multiple-output OFDM wireless communi-
`cation system, lab test results, and recent field test
`results obtained in San Jose, California. These are
`the first MIMO system field tests to establish the
`performance of MIMO communication systems.
`Increased capacity, coverage, and reliability are
`clearly evident from the test results presented in
`this article.
`
`INTRODUCTION
`This design is motivated by the growing demand
`for broadband Internet access. The challenge for
`wireless broadband access lies in providing a com-
`parable quality of service (QoS) for similar cost as
`competing wireline technologies. The target fre-
`quency band for this system is 2–5 GHz due to
`favorable propagation characteristics and low
`radio frequency (RF) equipment cost. The broad-
`band channel is typically non-LOS channel and
`includes impairments such as time-selective fad-
`ing and frequency-selective fading. This article
`describes the physical layer design of a fourth-
`generation (4G) wireless broadband system that
`is, motivated from technical requirements of the
`broadband cellular channel, and from practical
`requirements of hardware and RF. The key objec-
`tives of the system are to provide good coverage
`in a non-line-of-sight (LOS) environment (>90
`percent of the users within a cell), reliable trans-
`mission (>99.9 percent reliability), high peak data
`rates (>1 Mb/s), and high spectrum efficiency
`(>4 b/s/Hz/sector). These system requirements
`can be met by the combination of two powerful
`technologies in the physical layer design: multi-
`
`input and multi-output (MIMO) antennas and
`orthogonal frequency division multiplexing
`(OFDM) modulation. Henceforth, the system is
`referred to as Airburst.
`Multiple antennas at the transmitter and
`receiver provide diversity in a fading environ-
`ment. By employing multiple antennas, multiple
`spatial channels are created, and it is unlikely all
`the channels will fade simultaneously. The Air-
`burst system employs two transmit antennas and
`three receive antennas at the base station (2 × 3
`downlink), and one transmit antenna and three
`receive antennas at the customer premises equip-
`ment (CPE) (1 × 3 uplink). Only one transmit
`antenna is used at the transmitter due to cost
`considerations. It is seen that spatial diversity in
`Airburst yields link budget improvements of
`10–20 dB compared to a single-input single-out-
`put (SISO) system by reducing the fade margins.
`In addition, the two base transceiver station
`(BTS) antennas are used to double the data rate
`for users with certain channel characteristics by
`transmitting independent data streams from the
`two antennas. This technique, known as spatial
`multiplexing, can significantly increase system
`capacity [1, 2]. At the receiver, multiple anten-
`nas are used to separate spatial multiplexing
`streams and for interference mitigation, which
`makes aggressive frequency reuse a reality.
`OFDM is chosen over a single-carrier solu-
`tion due to lower complexity of equalizers for
`high delay spread channels or high data rates. A
`broadband signal is broken down into multiple
`narrowband carriers (tones), where each carrier
`is more robust to multipath. In order to main-
`tain orthogonality among tones, a cyclic prefix is
`added that has length greater than the expected
`delay spread. With proper coding and interleav-
`ing across frequencies, multipath turns into an
`OFDM system advantage by yielding frequency
`diversity. OFDM can be implemented efficiently
`by using fast Fourier transforms (FFTs) at the
`transmitter and receiver. At the receiver, FFT
`reduces the channel response into a multiplica-
`
`IEEE Communications Magazine • September 2002
`
`0163-6804/02/$17.00 © 2002 IEEE
`
`143
`
`IPR2018-01476
`Apple Inc. EX1011 Page 1
`
`

`

`initially when few base stations are installed, while
`spectrum efficiency defines how many users can
`be supported per unit of spectrum over the long-
`term. Reliability determines the quality of service
`a customer receives, and correspondingly long-
`term customer satisfaction. The system is current-
`ly targeted for business, home-office, residential
`and mobile users requiring high-rate data services.
`
`MIMO-OFDM
`DESIGN CONSTRAINTS
`NON-LOS CHANNEL MODELS
`In this section we briefly describe the key chan-
`nel characteristics that influence the broadband
`wireless system design such as channel disper-
`sion, Ricean K-factor, Doppler, cross-polariza-
`tion discrimination, antenna correlation, and
`condition number. Figure 1 shows a typical non-
`LOS propagation scenario.
`
`Channel Dispersion — An important channel
`characteristic that influences a system perfor-
`mance is channel dispersion due to reflections
`from close in and far away objects. The dispersion
`is often quantified by the rms delay spread, which
`increases with distance, and changes with environ-
`ment, antenna beamwidth, and antenna height [6].
`Typical values are in the 0.1–5 µs range.
`
`K-Factor — The fading signal magnitude follows a
`Rice distribution, which can be characterized by
`two parameters: the power Pc of constant channel
`components and the power Ps from scatter channel
`components. The ratio of these two (Pc/Ps) is called
`the Ricean K-factor. The worst case fading occurs
`when Pc = 0 and the distribution is regarded as
`Rayleigh distribution (K = 0). The K-factor is an
`important parameter in system design since it
`relates to the probability of a fade of certain depth.
`Both fixed and mobile communications systems
`have to be designed for the most severe fading con-
`ditions for reliable operation (i.e., Rayleigh fading).
`
`Doppler — The fixed wireless channel Doppler
`spectrum differs from the mobile channel
`Doppler spectrum [6]. For fixed wireless chan-
`nels, it was found that the Doppler is in the
`0.1–2 Hz frequency range and has close to expo-
`nential or rounded spectrum shape. For mobile
`wireless channels, the Doppler can be on the
`order of 100 Hz and has the Jake’s spectrum.
`
`Cross-Polarization Discrimination — The
`cross-polarization discrimination (XPD) is
`defined as the ratio of the co-polarized average
`received power Pll to the cross-polarized average
`received power, P⊥. XPD quantifies the separa-
`tion between two transmission channels that use
`different polarization orientations. The larger
`the XPD, the less energy is coupled between the
`cross-polarized channels. The XPD values were
`found to decrease with increasing distance [6].
`
`Antenna Correlation — Antenna correlation
`plays a very important role in single-input multi-
`output (SIMO), multi-input single-output
`(MISO), and MIMO systems. If the complex cor-
`relation coefficient is high (e.g., greater than 0.7),
`
`Receiver
`
`Base
`station
`
`■ Figure 1. A wireless propagation scenario.
`
`tive constant on a tone-by-tone basis. With
`MIMO, the channel response becomes a matrix.
`Since each tone can be equalized independently,
`the complexity of space-time equalizers is avoid-
`ed. Multipath remains an advantage for a
`MIMO-OFDM system since frequency selectivity
`caused by multipath improves the rank distribu-
`tion of the channel matrices across frequency
`tones, thereby increasing capacity [3].
`Another key feature of the physical layer
`design is adaptive modulation and coding that
`allows different data rates to be assigned to dif-
`ferent users depending on their channel condi-
`tions. Since the channel conditions vary over
`time, the receiver collects a set of channel statis-
`tics which are used both by the transmitter and
`receiver to optimize system parameters such as
`modulation and coding, signal bandwidth, signal
`power, training period, channel estimation fil-
`ters, automatic gain control, and so on. The Air-
`burst system has a proprietary link adaptation
`algorithm (LA) that tracks channel variations
`and adapts transmission parameters to perform
`optimally under prevailing conditions [4].
`Of course, a successful broadband wireless
`access system must have an efficient co-designed
`medium access control (MAC) layer for reliable
`link performance over the lossy wireless channel.
`The corresponding MAC is designed so that the
`TCP/IP layers see a high-quality link that it
`expects. This is achieved by an automatic retrans-
`mission and fragmentation mechanism (automatic
`repeat request, ARQ), wherein the transmitter
`breaks up packets received from higher layers
`into smaller subpackets, which are transmitted
`sequentially. If a subpacket is receiver incorrectly,
`the transmitter is requested to retransmit it. ARQ
`can be seen as a mechanism for introducing time
`diversity into the system due to its capability to
`recover from noise, interference, and fades. More
`details on ARQ design can be found in [5].
`The performance of the Airburst system is
`demonstrated by lab and field trial results. The
`performance can be measured by three key met-
`rics: coverage, spectrum efficiency, and reliability.
`The first two metrics are key in determining the
`cost of the system. Good coverage is important
`
`144
`
`IEEE Communications Magazine • September 2002
`
`IPR2018-01476
`Apple Inc. EX1011 Page 2
`
`

`

`The combined
`application of
`multi-antenna
`technology and
`OFDM
`modulation
`(MIMO-OFDM)
`yields a unique
`physical layer
`capable of
`meeting the
`requirements of a
`second-
`generation
`non-LOS system.
`
`diversity and multiplexing gains can be significant-
`ly reduced (or completely diminished in the case
`of correlation of 1). Generally, it was found that
`the complex correlation coefficients are low, in
`the 0.1–0.5 range for properly selected base sta-
`tion and receiver antenna configurations.
`
`Condition Number — The condition number is
`defined as a ratio of the maximum and minimum
`eigenvalues of the MIMO channel matrix. Large
`capacity gains from spatial multiplexing opera-
`tion in MIMO wireless systems is possible when
`the statistical distributions of condition numbers
`have mostly low values. LOS conditions often
`create undesirable MIMO matrix conditions
`(i.e., high condition numbers) that can be miti-
`gated using dual-polarized antennas. For low
`BTS antennas most propagation conditions are
`non-LOS with a considerable amount of scatter-
`ing, in which case the multiplexing gains of
`MIMO systems are very significant.
`RF AND HARDWARE CONSIDERATIONS
`In addition to the wireless channel characteristics,
`we need to consider the practical hardware (HW)
`limitations of low-cost RF and mixed signal
`devices when designing a broadband wireless data
`system. Moreover, since wireless systems must
`coexist with other co-channel and adjacent-chan-
`nel services, the system must meet emission speci-
`fications at the transmitter (masks, max EIRP,
`etc.) and must be able to tolerate specified levels
`of undesired interfering signals at the receiver.
`The distortion effects from the HW will add to
`the degradation effects from the channel to yield
`the overall link performance. Moreover, under
`good channel conditions the HW distortion will
`ultimately determine the maximum performance of
`the link. For a MIMO system operating in spatial
`multiplexing mode, the HW SNDR requirement is
`only a few dB higher since the channel matrix con-
`dition number can increase the effective receive
`distortion. Measured field trial results of our sys-
`tem confirm that the HW SNDR requirement is
`only a few dB higher for a MIMO receiver operat-
`ing in spatial multiplexing mode relative to a SISO
`counterpart that operates at a fraction of the data
`rate. On the other hand, since the effective data
`rate grows logarithmically with increasing SNDR, a
`SISO system with equal data rate would require
`HW specifications to get exponentially better.
`Moreover, for a MIMO system operating in diver-
`sity mode the HW requirements are lower than its
`SISO counterpart due to HW impairment diversity
`since the distortion is typically uncorrelated across
`multiple HW chains. In the 2–5 GHz frequency
`bands, it is possible to design low-cost wireless HW
`using IC components. After aggregating all the dis-
`tortion effects, including both the transmitter and
`receiver ends, a good design will yield up to 30 dB
`of SNDR. With this SNDR it is possible to suc-
`cessfully transmit MIMO with up to 64-quadrature
`amplitude modulation (QAM) with light coding.
`There exist a large number of sources of distortion
`at both the transmit and receive ends of a broad-
`band wireless system, but the most significant are:
`DAC/ADC: Digital–analog and analog–digital
`converters, mixed signal devices, generate distor-
`tion through saturation, quantization noise, and
`spurs. For high-performance broadband wireless
`
`applications with adequate level control, 10
`effective bits with minimal oversampling are typ-
`ically enough not to degrade the overall SDR.
`DAC/ADC clocks: The sampling instants at
`both transmitter and receiver will not be uniform
`spaced and will have slightly different rates.
`Even with timing tracking loops at the receiver
`to account for clock drifts, the residual timing
`phase noise or jitter will cause residual SDR.
`The timing jitter rms must be less than 1 percent
`of the data sampling rate for SDR > 30 dB.
`Up/downconverter oscillators: The frequency
`converters will introduce frequency drift and add
`phase noise. Even with phase tracking loops, the
`integral of the phase noise beyond 1 percent of
`the OFDM tone width must be less than –30
`dBc to get SDR > 30 dB.
`Linearity and dynamic range: All HW com-
`ponents introduce noise and have a limited
`range over which the signal can be processed
`without significantly distorting it. Thus, the sig-
`nal levels must be carefully controlled with a
`combination of power control and automatic
`gain control (AGC) to maximize the signal level
`relative to the HW noise without saturating the
`device. OFDM signals have slightly higher peak
`to average ratios (PARs) than other high-perfor-
`mance modulations, and extra care is required.
`The dynamic range and linearity requirements of
`OFDM can be made comparable to single-carri-
`er modulation with PAR reduction algorithms.
`
`MIMO-OFDM ARCHITECTURE
`The combined application of multi-antenna tech-
`nology and OFDM modulation (MIMO-OFDM)
`yields a unique physical layer capable of meeting
`the requirements of a second-generation non-
`LOS system. Herein, we discuss some key design
`and algorithmic choices made in implementation
`of the Airburst modem.
`
`Transmit Diversity — Many transmit diversity
`schemes have been proposed in the literature
`offering different complexity vs. performance
`trade-offs. We chose delay diversity for downlink
`transmission due to its simple implementation,
`good performance, and no feedback require-
`ment. In this scheme, the signal sent from the
`second antenna is a delayed copy of the signal at
`the first antenna. The delay introduced at the
`transmitter results in frequency selectivity in the
`received channel response. With proper coding
`and interleaving, space-frequency diversity gain
`is achieved without requiring any channel knowl-
`edge at the transmitter. Next-generation system
`system design incorporates improved transmit
`diversity schemes. Of particular interest are
`space-time codes that require no feedback [7],
`and linear precoding based on channel statistics
`that requires minimal feedback [8]. In space-
`time coding, the same signal is encoded differ-
`ently into different streams to be transmitted
`across multiple antennas. Block codes are attrac-
`tive since they allow linear decoding at the
`receiver (e.g., the Alamouti scheme) [9]. In lin-
`ear precoding, the transmitted signals are linear-
`ly mapped onto multiple transmit antennas,
`depending on the slowly varying channel statis-
`tics such as transmit antenna correlation. Linear
`
`IEEE Communications Magazine • September 2002
`
`145
`
`IPR2018-01476
`Apple Inc. EX1011 Page 3
`
`

`

`The link
`adaptation layer
`monitors the
`channel
`conditions on a
`per user basis
`and determines
`the optimal
`transmission
`scheme:
`multiplexing
`or diversity.
`
`■ Figure 2. From left to right: a BTS chassis, a CPE board with Airburst ASIC, and CPE.
`
`precoding can be used in conjunction with space-
`time codes to provide performance gains. Pre-
`liminary field trials have shown 2–6 dB gain over
`delay diversity and up to 3 dB gain over the
`Alamouti scheme across the cell.
`
`Spatial Multiplexing — It is possible to trans-
`mit two separately encoded data streams from
`the two base station antennas. A high-rate signal
`is multiplexed into a set of lower-rate streams,
`each of which is encoded, modulated, and trans-
`mitted at a different antenna, while using the
`same time and frequency slot. Each of the three
`receiving antennas receives a linear combination
`of the two transmitted messages that have been
`filtered by different channel impulse responses.
`The receiver separates the two signals using a
`spatial equalizer, and demodulates, decodes, and
`demultiplexes them to yield the original signal.
`Separation is possible as long as each stream
`induces a different spatial signature at the receiv-
`er (i.e., the channel matrix has full rank). Since
`most spatial equalizers use some form of chan-
`nel matrix inversion, a unique solution is only
`possible if the number of receive antennas is
`greater than or equal to number of independent
`transmit signals. In practice, it is preferred that
`the number of receive antennas be greater since
`this leads to a better-conditioned channel matrix,
`and therefore more accurate channel inversion
`in fixed precision arithmetic. The link adaptation
`layer monitors the channel conditions on a per
`user basis and determines the optimal transmis-
`sion scheme: multiplexing or diversity.
`
`Receive Diversity and Interference Cancella-
`tion — Receive diversity is available at both the
`BTS and CPE due to three receive antennas. The
`diversity is leveraged by a maximal-ratio-combin-
`ing (MRC) algorithm that coherently combines
`signals at multiple receivers to maximize signal-
`to-noise ratio (SNR). There is some natural inter-
`ference suppression with MRC since it matches
`the spatial signature of the desired signal, not that
`of the interferer, which therefore gets attenuated.
`MRC cannot, however, suppress strong interfer-
`ence such as that arising from spatial multiplexing
`where the two streams interfere with each other,
`or from strong co-channel users in other cells due
`to aggressive frequency reuse. In this case, it is
`
`desirable to use the minimum mean square error
`(MMSE) algorithm that minimizes the mean
`square error between each desired signal and its
`estimate, thereby maximizing signal-to-interfer-
`ence-plus-noise ratio (SINR). The MMSE weights
`require knowledge of noise-plus-interference
`statistics. Therefore, it is important to detect
`whether the ambient environment is noise-limited
`or interference-limited, and to collect accurate
`statistics by averaging over appropriate time and
`frequency intervals. If the interferer is friendly
`(e.g., a spatial multiplexing user), the interferer’s
`spatial signature is available to be used in the
`MMSE weights. For unfriendly interference from
`neighboring cells, second-order statistics (covari-
`ance matrices) are used to capture the spatial
`structure of interference.
`
`Soft Decoding — Both MRC and MMSE algo-
`rithms yield soft signal estimates that are input
`to a soft decoder. The soft decisions are weight-
`ed by the estimated SINR on a tone-by-tone
`basis to give more weight to good tones and less
`weight to bad tones. Soft-decision decoding com-
`bined with SINR weighting provides significant
`performance gains (3–4 dB) in frequency selec-
`tive channels.
`
`Channel Estimation — The purpose of channel
`estimation is to identify the channel between each
`pair of transmit and receive antennas. The train-
`ing tones transmitted from each antenna are
`orthogonal with respect to each other so that the
`channel from each transmit antenna can be iden-
`tified uniquely. The training tones are spaced in
`frequency, with spacing less than the channel’s
`frequency coherence so that the channel can be
`interpolated between training tones. The channel
`interpolation is optimized depending on channel
`delay spread, and can be further improved by
`time-domain filtering. In the downlink, a dedicat-
`ed channel identification slot is broadcast to all
`users on a frame-by-frame basis. In the uplink,
`each slot includes both training and data tones
`since the traffic from the CPE can be bursty, and
`the channel may change between bursts.
`
`Synchronization — Both uplink and downlink
`transmission are preceded by a synchronization
`(sync) slot for timing phase, timing frequency,
`
`146
`
`IEEE Communications Magazine • September 2002
`
`IPR2018-01476
`Apple Inc. EX1011 Page 4
`
`

`

`Physim SUI-3
`Lab SUI-3
`Physim SUI-4
`Lab SUI-4
`Physim SUI-6
`Lab SUI-6
`
`100
`
`10–1
`
`10–2
`
`10–3
`
`10–4
`
`Bit error rate
`
`0
`
`2
`
`4
`
`6
`
`8
`
`12
`10
`SNR (dB)
`
`14
`
`16
`
`18
`
`20
`
`22
`
`■ Figure 3. Sample BER vs. SNR for several SUI models and link modes.
`
`It is seen that there is very good agreement
`between laboratory HW measurements and the
`Physim model.
`FIELD TRIAL RESULTS
`TEST SETUP
`The performance of the Airburst system is cur-
`rently being verified with outdoor field trials.
`The BTS antennas are located on the rooftop
`(approximately 49 ft high) of the Iospan Wire-
`less building, and the CPEs are placed at differ-
`ent locations within a 3.5-mi cell radius and a
`120° sector facing west of Iospan Wireless. The
`environment can be characterized as suburban
`with residential blocks (1–3 floors), commercial
`buildings (2–5 floors), moderate trees (30–90 ft
`high), and slightly hilly terrain. The total BTS
`transmit power and EIRP are 35.5 dBm and 51
`dBm, respectively, while the CPE maximum
`transmit power and EIRP are 30 and 42 dBm,
`respectively. The downlink was operated at 2.683
`GHz and uplink at 2.545 GHz center frequen-
`cies (multimegabit data service, MMDS, band),
`with a channel bandwidth of 2 MHz. At the
`BTS, the transmit and receive antennas are
`spaced 16 and 8 wavelengths, respectively, with a
`gain of 16 dBi and 3 dB beamwidth (azimuth) of
`100°. At the NAU, the receive antennas are
`spaced up to 1.5 wavelengths, with a gain of 12
`dBi and 3 dB bandwidth of 90°.
`Field trials include:
`• Modem performance evaluation — deter-
`mining the data rates at different locations
`using a fully functional CPE under normal
`operating conditions
`• Channel measurements — characterizing
`the wireless channel across time, space and
`frequency, by using the Airburst system in a
`test-mode.
`In test mode, the BTS transmits known training
`signals, and the CPE samples the received sig-
`nals every 50 ms in time and 72 kHz in frequen-
`
`and frequency offset estimation. The slot is
`structured such that data and training are trans-
`mitted over even numbered tones, and odd tones
`are set to zero. This introduces a repetitive pat-
`tern in the time-domain signal, which allows esti-
`mation of the above parameters. Once
`synchronization is obtained, fine timing esti-
`mates can be computed from the training tones.
`
`Adaptive Modulation and Coding — The
`Airburst system maximizes system capacity by
`optimizing the link parameters available to each
`user. There are multiple levels of coding and
`modulation that can be optimized on a per user
`data flow basis depending on the user’s SINR
`statistics at a particular location and time, and
`the user’s QoS requirements. The QAM levels
`vary from 4 to 64, and the coding consists of a
`punctured convolutional code combined with a
`Reed-Solomon code. There are six combinations
`of QAM and coding levels, referred to as coding
`modes. The coding modes 1–6 correspond to
`data rates of 1.1–6.8 Mb/s obtained over a 2
`MHz channel. For the downlink, the above rates
`are doubled when spatial multiplexing is used.
`Each coding mode has a different setpoint,
`which is defined to be the average SINR
`required to obtain a specified pre-ARQ packet
`error rate, typically in the 0.1–5 percent range.
`The setpoint is a function of channel characteris-
`tics: delay spread, K-factor, antenna correlation,
`Doppler, and so on. The link adaptation algo-
`rithm chooses the best coding mode per user
`based on SINR statistics provided by the modem.
`
`LABORATORY
`PERFORMANCE RESULTS
`For software system simulations and laboratory
`performance validation purposes, a set of non-
`LOS channel models were developed by the
`IEEE 802.16 working group [6]. These models
`were implemented in both software and hard-
`ware to test and define the system performance.
`For each transceiver six multipath channels (2 ×
`3) had to be implemented, each with three tap
`delay lines. We used three TAS-Flex 500 simula-
`tors with two channels each to emulate downlink
`fading channels. The Airburst modem was mod-
`eled in software using Matlab/C code (Physim).
`For all channel models, bit accurate simulations
`of the modem were run to predict bit error rate
`(BER) performance vs. SNR and SIR. These
`results were later compared with results obtained
`in the laboratory using HW channel emulators
`and the current ASIC based product. Figure 2
`shows a picture of some main elements of the
`product, namely a BTS chassis, a CPE board
`with an Airburst application-specific integrated
`circuit (ASIC), and CPE.
`
`Results — Figure 3 shows a BER comparison of
`Physim and laboratory results for a subset of
`channel models and transmission modes:
`• SUI-3 channel: uplink coding mode 2
`• SUI-4 channel: downlink transmit diversity
`coding mode 3
`• SUI-6 channel: downlink spatial multiplex-
`ing coding mode 3
`
`IEEE Communications Magazine • September 2002
`
`147
`
`IPR2018-01476
`Apple Inc. EX1011 Page 5
`
`

`

`the CPE antennas pointed in the general direc-
`tion of the BTS.
`
`RESULTS
`In this section we report on the 2 × 3 Airburst
`system field test results. The results include fading
`margins for 99.9 percent link reliability, cell size,
`and data rates. The results are also compared to 1
`× 2 and 1 × 1 antenna configurations (systems) by
`equivalently disconnecting the corresponding RF
`chains from the 2 × 3 Airburst system.
`
`Fading Margins — The fading margins are a
`function of the Ricean K-factor, delay spread,
`and antenna correlation. Higher delay spread
`results in frequency selectivity that can be
`exploited by OFDM systems to lower the
`required fading margins. For Rayleigh fading
`channels (K = 0) with no delay spread and zero
`antenna correlation, the fading margins (includ-
`ing array gain) for the 99.9 percent link reliabili-
`ty are 10 dB, 23 dB and 35 dB for 2 × 3, 1 × 2
`and 1 × 1 antenna configurations, respectively
`(Physim results).
`Figure 4 shows a snapshot of the received
`SNR averaged across the receive antennas for a
`1 × 1, 1 × 2 and 2 × 3 antenna configurations
`obtained from a moving measurement. The vehi-
`cle was moving at low speeds at a distance of 1.1
`mi from the BTS. The fading margins for 2 × 3,
`1 × 2, and 1 × 1 systems for 99.9 percent link
`reliability were computed to be 7.5 dB, 14 dB,
`and 21 dB, respectively. The lower fade margins
`for the 2 × 3 system are attributed to a higher
`array gain and diversity gain resulting from mul-
`tiple antennas. These margins are lower than the
`margins reported in the previous paragraph
`since the measured channel has nonzero delay
`spread and yields an additional diversity gain.
`Similarly, Fig. 5 shows the received SNR
`across the receive antennas for a 2 × 3, 1 × 2,
`and 1 × 1 system for a fixed measurement made
`at an apartment 0.8 mi away from the BTS. The
`SNR values were reported every 1 min across a
`24-h period. The weather conditions were dry
`with wind speeds ranging from 0 to 15 mi/h. We
`observe that the mean received SNR and fading
`characteristics for the three systems are compa-
`rable to that seen in a moving measurement,
`albeit across a larger timescale, due to wind
`blown foliage. Hence, for reliable operation, a
`fixed wireless system needs to be designed with
`similar fade margins as a mobile system, even
`though the former exhibits orders of magnitude
`lower Doppler relative to mobile links.
`
`Cell Size — For the same transmit power and
`99.9 percent link reliability, when compared to
`the 2 × 3 system, the higher fade margins
`required for the 1 × 2 and 1 × 1 systems result in
`a smaller cell size. For example, assuming the
`path loss exponent of 4, 12 dB higher fade mar-
`gin reduces the cell radius by half and therefore
`the cell area by one-fourth. Based on our test
`results for the given cell site and 90 percent cov-
`erage, the cell radii turned out to be 4.0 mi, 2.7
`mi, and 1.6 mi for the 2 × 3, 1 × 2, and 1 × 1 sys-
`tems, respectively.
`
`Measured Data Rates — Figure 6 shows the
`
`0
`
`1
`
`2
`
`3
`
`4
`
`5
`Time (s)
`
`6
`
`7
`
`8
`
`9
`
`10
`
`1 × 1
`1 × 2
`2 × 3
`
`25
`
`20
`
`15
`
`10
`
`5 0
`
`SNR (dB)
`
`–5
`
`–10
`
`–15
`
`■ Figure 4. Received SNR for a moving measurement.
`
`0
`
`5
`
`10
`
`15
`
`20
`
`25
`
`Time (h)
`
`1 × 1
`1 × 2
`2 × 3
`
`20
`
`15
`
`10
`
`5 0
`
`SNR (dB)
`
`–5
`
`–10
`
`–15
`
`■ Figure 5. Received SNR for a fixed long-term measurement at an apartment
`0.8 mi from the BTS.
`
`cy (spanning a total of 2 MHz in bandwidth).
`The outdoor tests and measurements are being
`carried out for both fixed (short- and long-term)
`and moving scenarios. For the short-term mea-
`surements, the CPE is placed on a mast mount-
`ed on a mobile at approximately 9 ft above the
`local ground level. Fixed measurements lasting
`approximately 5 min each are being carried out
`at different distances from the BTS. In addition,
`long-term fixed measurements lasting from a day
`to several months are being carried out by moni-
`toring fully functional CPEs in several apart-
`ments and houses at ground level. Moving
`measurements at low vehicular speeds are taken
`to cover at least 50 percent of the cell area with
`
`148
`
`IEEE Communications Magazine • September 2002
`
`IPR2018-01476
`Apple Inc. EX1011 Page 6
`
`

`

`Combining
`OFDM and CDMA
`technologies is
`attractive for
`future wireless
`broadband
`communications
`and software
`radio realization.
`
`average recorded data rates as a function of dis-
`tance from the BTS for the 2 × 3, 1 × 2, and 1 ×
`1 systems. Data rates are higher for users closer
`to the BTS due to the lower path loss and there-
`fore improved SNR. The peak data rates for the
`1 × 1 and 1 × 2 systems is 6.8 Mb/s, while that
`for the 2 × 3 system is 6.8 * 2 = 13.6 Mb/s (two
`data streams with spatial multiplexing). For good
`link reliability, spatial multiplexing requires high
`receiver SNR as well as good channel matrix
`condition numbers. Such conditions are satisfied
`in non-LOS environments for distances less than
`2 mi from the BTS. Approximately 80 percent of
`the users have data rates greater than 6.8 Mb/s
`and operate in the spatial multiplexing mode. At
`distances greater than 2 mi from the BTS, the 2
`× 3 system operates mostly in a diversity mode,
`where the signals from all antennas are com-
`bined to lower the fading margins and therefore
`improve coverage (i.e., increase cell radius). The
`effect of multiple antennas on the measured
`data rate is dramatic. The mean data rate at the
`1.1 mi distance from the BTS is approximately
`quadrupled for the 2 × 3 system when compared
`to the 1 × 1 system, and doubled when compared
`to the 1 × 2 system. As mentioned earlier, the
`gains are related to the spatial multiplexing
`capability and low fading margin of the 2 × 3 sys-
`tem.
`
`CONCLUSIONS
`In this article we describe the Airburst 4G broad-
`band wireless system, laboratory test results, and
`recent field trial results. The laboratory test and
`field trial results show the encouraging perfor-
`mance of the MIMO-OFDM system. The results
`show dramatic increase in capacity, coverage,
`and reliability over SISO, MISO, or SIMO com-
`munication systems.
`ACKNOWLEDGMENT
`Thanks are given to Prof. Willie W. Lu for the
`technical review of this article and continued
`encouragement. Without his help, this article
`would never have been possible.
`REFERENCES
`[1] A. J. Paulraj and T. Kailath, “Increasing Capacity in Wireless
`Broadcast Systems Using Distributed Transmission/Direc-
`tional Reception,” U.S. Patent, no. 5,345,599, 1994.
`[2] G. J. Foschini, “Layered Space-time Architecture for
`Wireless Communication,” Bell Labs Tech. J., vol. 1,
`Autumn 1996, pp. 41–59.
`[3] H. Bolcskei, D. Gesbert, A. J. Paulraj, “On the Capacity
`of OFDM-Based Spatial Multiplexing Systems,” IEEE
`Trans. Commun., vol. 50, no. 2, Feb. 2002, pp. 225–34.
`[4] S. Catreux et al., “Adaptive Modulation and MIMO Cod-
`ing for Broadband Wireless Data Networks, IEEE Com-
`mun. Mag., June 2002.
`[5] I. Stojanovic et al.,