The AT&T Labs Broadband Fixed Wireless Field Experiment Byoung-Jo Kim, N. K. Shankaranarayanan, Paul S. Henry, Kevin Schlosser, and Thomas K. Fong, AT&T Labs-Research

ABSTRACT

In this article, we describe an ongoing broadband fixed two-way wireless field experiment conducted by AT&T Laboratories — Research in Monmouth County, New Jersey. Our experiment, which is one of the first twoway broadband fixed wireless systems, offers an end-to-end broadband packet access service, with telecommuting as the primary application for our employee users. It operates in the 2.6 GHz MMDS spectrum, and is based on cable modem technology. We have developed a Web-based network monitoring/management tool that greatly enhances the ability to manage, diagnose, and optimize the system. The lengthy period of operation has allowed us to make observations about user behavior, weather-related channel impairments, and equipment performance. We have identified several design issues related to the application of cable modem technology to the fixed wireless environment. Also, we have measured a significant path loss effect arising from a combination of rain and foliage.

I

n this article, we describe an ongoing broadband fixed two-way wireless field experiment conducted by AT&T Laboratories — Research in Monmouth County, New Jersey, United States. The experimental system offers a broadband packet service to residential users. The experiment started in late 1997, and currently there are 12 sites which are the houses of volunteer AT&T employees. This system is one of the first (if not the first) two-way broadband fixed wireless systems. It operates in the 2.6 GHz multichannel multipoint distribution service (MMDS) spectrum, and is based on cable modem technology. The objectives of the experiment were to: • Install a field-grade broadband two-way fixed wireless network • Operate a broadband packet access network offering high-speed access to corporate intranets • Study various issues related to fixed wireless and broadband packet access networks The primary service which we were interested in providing was a seamless extension of the corporate TCP/IP intranet out to user homes. The approach taken in our experiment was to set up a network to provide a previously unavailable two-way, always-on, high-speed packet access service that users would use in their daily routine. The system was thus treated as a whole as we investigated in depth the various components, such as radio receiver technology and medium access control (MAC) protocol. The long-term nature of the project provided a unique opportunity to study user behavior patterns as well as radio performance issues related to weather. The objective of this article is to provide a broad overview of the system. We briefly discuss broadband fixed wireless access in the following section, then describe our experimental system, and finally highlight our observations and lessons.

BROADBAND FIXED WIRELESS PACKET ACCESS USING MMDS A fixed wireless network is an attractive option for broadband packet access networks [1-3]. Such a network typically has base stations serving three to six (or more) sectors in a cell with 2–15 km radius. User sites have directional terminal

56

0163-6804/99/$10.00 © 1999 IEEE

antennas oriented toward a suitable base. Broadband fixed wireless networks are similar to hybrid fiber/coax (HFC) networks in that the access medium is a shared network carrying radio frequency (RF)-modulated data signals. Fixed wireless offers rapid and flexible deployment and a low initial cost; this makes it suitable for low take-rate scenarios. Our experiment uses the 2.6 GHz MMDS spectrum. Some of the first broadband fixed wireless data systems used this band. MMDS or “wireless cable” systems were originally designed for broadcast video services covering a single cell with 10–40 km radius. The service as well as the technology was similar to cable television. The first MMDS data systems, as well as HFC cable systems, had a high-speed packet channel on the downlink, with the uplink provided by a telephone modem link. This constrained the uplink bit rate and also required a telephone line for the data service. Two-way wireless overcomes these disadvantages, and our experiment is one of the first two-way MMDS systems. With the growing popularity of the Internet, MMDS systems have now evolved to two-way access networks [4]. In the initial single-cell systems there is no co-channel interference, and the system could support the use of 64-quadrature amplitude modulation (QAM) on the downlink. In order to cover large metropolitan areas with sufficient capacity, the new MMDS systems have to use a cellular architecture which reuses radio spectrum. The preferred cellular designs involve moderate signal-to-interference ratios of 12–15 dB rather than the 25 dB required for a 64-QAM downlink [5]. While (multicell) fixed wireless deployments are expected to use modulation schemes such as quadrature phase shift keying (QPSK) and 16-QAM, our single-cell experimental system uses 64-QAM modems, since those were the only modems available to us. Since 64QAM modems are sensitive to channel impairments, they gave us an opportunity to study channel effects closely.

THE EXPERIMENTAL SYSTEM ARCHITECTURE OVERALL Figure 1 depicts the major components of the experimental system, which has been continuously operating since late 1997. There are 12 user sites scattered across the Red Bank, New Jersey area within an 8 mi radius of the base station, which houses radio systems and IP data networking equipment. The user sites are houses of AT&T employee volunteers. An AT&T Labs Research location is also one of the user sites. The base T. K. Fong is now with @Home Corporation.

IEEE Communications Magazine • October 1999

Rooftop antenna

Downstream at 2.6 GHz 27 Mb/s Upstream at 2.1 GHz (3 x 650 kb/s) 3 x T1 capacity

Transverter

Base station antenna

Internet

Cable modem

Single tower shared RF system

AT&T intranet

Alcatel point-to-point radio

Cable headend

Alcatel

Rooftop antenna

landline T1

Cisco4500 router

Cisco4500 router

Transverter

100BaseT Ethernet switch

VoIP gateway

Application servers

User computers

Simple user site

Backup

FDDI ring

Video

10BaseT Ethernet

Cable modem

DHCP/Web/FTP/file/ monitoring servers Base station

10BaseT Ethernet bridge

PBX

User computers

PSTN VoIP client

WLAN access point

Telephone Telephone

Fully functional user site

■ Figure 1. The AT&T Labs — Research broadband fixed wireless packet access field experiment.

station is connected to the AT&T Labs IP intranet (5.3 mi away) by a point-to-point radio link with 3 x T1 (4.5 Mb/s) capacity and one backup landline T1. The system is based on two-way cable modem technology and uses IP as the network layer protocol. In this section we describe the radio network, the IP network, the services, and the network management.

received are downconverted to 5–40 MHz. The cable headend extracts user IP packets from the uplink signal at a postFEC 650 kb/s effective bit rate. Our system supports three uplink channels that are quasi-statically assigned to individual user modems. While the downlink is continuously on, the uplink is time-shared using polling-based multiple access.

RADIO AND MAC SUBSYSTEMS

User Site Equipment — Figure 2 shows the user site equipment, which is based on off-the-shelf products. The user sites have 24-in roof-mounted antennas with 7.5˚ beamwidth and 24 dBi gain, and effective heights of 12–30 ft above ground. The antenna is directly connected to transverters that contain uplink/downlink amplifiers and mixers for up/downconversion. The downlink is converted to 285 MHz to match the cable modem specifications, while the upconversion matches the downconversion at the base station. (There is a small amount of additional loss and distortion due to the concatenation of transverter and modem RF stages.) The transverters are typically mounted close to the antennas to minimize signal loss, and connected via 75-Ω coaxial cables to cable modems, which are placed indoors. If necessary, attenuators are used to match the signal power levels to the input dynamic range of the modems. The transverters are powered from indoors via the RF coaxial cable using a DC coupler. Cable modems communicate with users’ IP device(s) using a 10BT Ethernet interface.

The experimental system is based on a commercial two-way HFC cable modem system. The base station houses a cable headend that operates at 44 MHz downstream (downlink) and 5–40 MHz upstream (uplink). Compatible cable modems are used at user sites. At both ends, the cable frequency signals are translated to the wireless frequencies. Base Station Equipment — For downlink signals (from the base station to user sites), the 44 MHz signal from the cable head-end is upconverted to 2.6 GHz and transmitted at 48 dBm equivalent isotropic radiated power (EIRP) to user sites. The base antenna has a 15 dB gain, 180˚ beamwidth (cardioid) and an effective height of 400 ft above grade level. Although the cable head-end was developed before the current widely adopted Data Over Cable Service Interface Specifications standard (DOCSIS) [6], its specifications are similar to DOCSIS. The continuous 6-MHz-wide downlink uses MPEG-2 transport. It carries IP packets at 27 Mb/s using 64QAM at a 5 MHz symbol rate, and forward error correction (FEC) overhead of 3 Mb/s. The uplink signals are burst-mode QPSK signals within 600 kHz channels. The uplink signals

IEEE Communications Magazine • October 1999

IP NETWORK AND APPLICATIONS The IP Network — The IP network for the experiment essentially consists of three connected networks: the user sub-

57

net, the base station subnet, and the AT&T Labs intranet. All user IP devices are on the user subnet, and the head-end routes IP traffic between this subnet and the base station subnet. (The traffic between any two user sites is treated as layer 2 traffic, and does not involve IP layer routing.) The base station subnet uses switched 100BT Ethernet and contains a dynamic host configuration protocol (DHCP) server and other application servers. A mid-range IP router connects the base station subnet to the AT&T Labs intranet via the backhaul T1 links. Another IP router terminates these T1 links inside the AT&T Labs intranet, and connects to the Internet as well as various internal application servers. Applications — The base station subnet has an FTP server and a Web server, which are used for testing, and to provide network monitoring and management functions as described in the next section. This exploits the full capacity of the wireless channel on the downlink, which exceeds the capacity of the backhaul T1 links. The user applications supported by the system include: • Telecommuting: The principal use of the fixed wireless packet data system is always-on telecommuting. Users at home can access file and print servers at their respective offices controlled by different Windows NT domains, as well as internal Web sites in the AT&T Labs intranet. Internet access is through the same firewalls that protect the AT&T Labs intranet. The families of the users also have access to the Internet, while AT&T intranet resources are protected by passwords. One user site is an AT&T Labs office building where some users use the experimental system routinely for their day-to-day office work. A modem at this location is connected to a wireless LAN (WLAN) that supports several laptop computers. This system also offers seamless portability of laptops

Antenna

Transverter

Modem

Indoor

Outdoor

■ Figure 2. Fixed wireless access user site equipment.

58

between work and home; users can use their laptops both at work and at home without any change in setup, and get the same environment in both places. • Voice over IP (VoIP): The fixed wireless system supports VoIP for a few user sites. The client devices at the user sites have two RJ-11 jacks to accommodate two analog telephones with separate numbers. Voice conversations are encoded using International Telecommunication Union (ITU) Recommendation G.711. The packetized voice is sent as IP packets to another client device or to a VoIP gateway in the intranet. The gateway interfaces to the corporate private branch exchange (PBX), and the client devices work like regular PBX extensions. Calls can be made using the PBX without additional dial tone. Weighted Fair Queuing (WFQ) is used to provide basic QoS for VoIP, and more sophisticated QoS approaches are being implemented and studied over the experimental system. • Video on demand and video multicast: The users also access a video-on-demand service and live broadcasts of AT&T internal presentations. This service is enabled through IP multicasting from a server in the intranet over the fixed wireless system, and requires 50–500 kb/s downlink bandwidth. • Thin-client computing: A thin-client computing service using Windows NT Terminal Server/Citrix MetaFrame is supported. Enabled by simple client software, users can run various Windows applications remotely on a highperformance server in the intranet, with local display and input devices. This includes several wireless thin-client devices which are supported by the WLAN mentioned above. These lightweight “wireless pads” with pen-based input allow access to standard Windows applications on the remote server while roaming in the building.

WEB-BASED NETWORK MANAGEMENT SYSTEM Configuring, monitoring, and debugging various network and computer devices initially required multiple separate interfaces (e.g., telnet, tftp, Web), each with a separate login. Also, since the experimental system covers a substantial geographic area, frequent visits to user sites and the base station were required. This was very tedious, especially for a continuous long-term experiment. Therefore, a Web-based network management system was implemented by integrating several commercially or publicly available software systems around a commercial network management development platform. The system resides on a server inside the intranet and supports the following: • Monitoring and configuration of any Simple Network Management Protocol (SNMP)-enabled device in the experimental system through a Web browser interface anywhere in the AT&T intranet. • SNMP management information base (MIB) manipulation for customized interface. • Real-time statistics collection, presentation, and manipulation on a Web browser using Java applets. • Alarms for device failure, signal quality degradation, and other customizable fault conditions via e-mail, paging, fax, or Web browsers. • Remote control of various servers and user computers inside a browser window using Virtual Network Computing (VNC) Java servers developed by AT&T Labs [7]. Using VNC, computers at user sites can be remotely configured to participate in experiments, performance measurements, and technical support. • Access to, and control of, packet capture probes deployed at various locations in the system. These remote-controlled software probes were placed in the base subnet, a user site, and the intranet, since current SNMP implementations do

IEEE Communications Magazine • October 1999

■ Figure 3. The Web-based network management system. Top: real-time statistics graphs; bottom left: overview of major equipment with clickable image to access/configure details; bottom right: top-level interface to the network management system.

OBSERVATIONS AND LESSONS In this section we highlight some of our observations related to weather-related radio channel impairments, modem performance in wireless systems, and user application performance.

COMBINED RAIN/FOLIAGE EFFECT ON PROPAGATION Figure 4 shows the link path loss for the various user site locations, along with the free space loss and the predicted median loss based on the AT&T Labs-Research path loss model for fixed wireless systems [8]. The link losses are 10–30 dB greater than the free space loss due to scattering and diffrac-

IEEE Communications Magazine • October 1999

-100 -110 Radio path loss (dB)

not provide enough detail about network traffic. The probes capture and analyze Ethernet packets to assist in debugging and measurement, including long-term measurements. The above capabilities significantly reduce the time and effort required for experiments, measurements, configuration, and maintenance on the system. This is significant since, compared to telephone networks, these functions are likely to be more complex in packet access networks. Figure 3 shows a screen shot of a typical view of the management system in several browser windows. They show real-time statistics graphs, live information on individual sites, and a clickable image map for accessing/configuring most devices used in the experiment.

Free space loss

-120 -130

Median loss from AT&T path loss model for fixed wireless

-140 -150 -160 100

Distance in km

101

■ Figure 4. Radiolink path loss for user sites with 2.56 GHz free space loss and median path loss predicted by the AT&T path loss model for fixed wireless systems (tx height = 130 m, rx height = 10 m, light foliage).

59

Rainfall and downlink attenuation

Relative attenuation: dB

tion from foliage, terrain, and buildings. The losses are con16 Rainfall Site 1 sistent with the AT&T path loss model. As part of normal Site 1 attenuation Site 2 attenuation 14 operation, the network management system collects various statistics over time, including hourly downlink and uplink 12 0.6 signal strengths for each site based on modem measurements. Although not as accurate as calibrated signal 0.5 10 strength measurements reported in the literature, these 8 hourly records are collected over a long period of time 0.4 (approximately a year) compared to previously reported 0.3 6 studies. One interesting finding from these records is the com4 0.2 Site 2 bined effect of rain and foliage on propagation path loss. Rainfall 0.1 2 Figure 5 shows the relative downlink power level variation observed at two sites along with the hourly rainfall record 0.0 0 obtained from a nearby weather station. The wireless link between the base station and site 1 (approximately 5 mi 6/10 6/15 6/20 6/25 6/30 long) has several large trees in the path near the house. With Date this relatively heavy foliage, the path attenuation increases by ■ Figure 5. Fixed wireless rain/foliage-related signal attenuation. more than 12 dB under heavy rain. The limited dynamic range of the cable modem system is indicated by the clipping of the attenuation curve at the top. Thus, the rain/foliage attenuation could be more than 12 dB in some cases. Site 2, symbol timing errors [9], and longer equalizers or fracwhich is approximately 3 mi from the base station, has a path tionally spaced equalizers improve this problem. partially obscured by a single tree branch. The relative path loss • Equalizer adaptation: The cable modem equalizers used for this site compared to its nominal value is also highly correin our system do not continuously adapt after initial lated with rainfall, but the attenuation under heavy rain is acquisition. Although this may be acceptable for the observed to be less than 3 dB. HFC environment, fixed wireless channels are time-varyThe comparison between the two sites suggests that the ing, albeit at slow speed. Over a period of several days, combination of heavy foliage and rain can cause a significant the downlink error rate increased at all user sites. To increase of path loss, while rain alone has a modest effect at overcome this problem, we force equalizer tap updates 2.6 GHz. This is due to the increased scattering caused by wet by remotely resetting all modems twice a day when user leaves. As can be seen in the graph, the path loss of site 1 activity is low. However, one of the user sites exhibits continues to stay high for several hours after a heavy rain, especially fast channel variation (on the order of tens of until most of the water has dried/shed from the leaves. seconds). Since the channel time variation is not tracked Records collected for other sites also support the above conby the equalizer, this site has the highest downlink error clusion. When planning for the coverage of a fixed wireless rate. These problems can easily be solved by slow equalsystem, this rain/foliage-related attenuation must be taken izer channel tracking. into account in addition to all the conventional margins for • Carrier recovery and stability: In most cable modem seasonal, log-normal, and Rayleigh fading. This result is sigdesigns, carrier recovery uses a second-order loop with nificant since the 2.6 GHz frequency is often assumed to be a decision-directed (DD) phase discriminator, and can unaffected by rain (as opposed to 28 GHz and 38 GHz sysusually correct carrier frequency errors up to 10 pertems), and we see that this is not so in some special foliagecent of the symbol rate under ideal conditions [9]. (The related cases. Another important implication of this performance is worse with low signal-to-noise ratio, rain/foliage attenuation is that slow transmit power control, SNR, and/or delay spread.) This tolerance range is receiver automatic gain control (AGC), and continuous equalabout 50 kHz for our downlink and 5 kHz for the izer adaptation are necessary for reliable service. Only a uplink. The carrier stability requirement is more minor impact has been observed due to snow accumulation on demanding for the uplink, especially considering its tree branches. bursty transmission. At 2.6 GHz carrier frequencies, an uplink transmit carrier stability of 2 ppm is required. APPLYING CABLE MODEM TECHNOLOGY TO FIXED WIRELESS (For an HFC system with uplink carrier frequencies under 100 MHz, the carrier stability requirement is As described earlier, the experimental system is an adaptation about 100 ppm.) In our system, uplink carrier stability of an early two-way cable modem system to fixed wireless has been the most common cause of service disruption, applications. Since cable modem technology continues to be a primarily due to the effect of temperature on outdoorstrong option, it is important to understand the issues involved mounted transverters. User sites with high SNR and litin such adaptation. These issues are mainly due to differences tle multipath delay spread operated reliably with up to in the HFC and wireless physical media. 7 kHz carrier offsets, while user sites with low SNR • Symbol timing and downlink equalization: According to and/or larger multipath delay spread failed with only 3 simulations using channel models from [8] and actual kHz drifts. Achieving satisfactory performance at these system tests, the downlink equalizer lengths of most sites required more stable (and more expensive) oscilcable modems (as short as eight feed-forward and eight lators. Improving cable modem carrier recovery perforfeedback taps, and longer for more recent modems) are mance would be more reliable and cost-efficient than adequate for mitigating the delay spread of fixed wireless improving transverter oscillator stability. There are channels, provided that the symbol timing error is kept many simple synchronization methods to ease the carrismall. However, most current implementations of cable er stability requirement in the literature [9]. modem equalizers (including the one used in the experi• RF stage control: The transverters used in the system are ment) show inadequate symbol timing recovery perforbasic analog devices that cannot be remotely controlled. mance. There are many effective methods to achieve low

60

Rainfall inch/hour

18

IEEE Communications Magazine • October 1999

The above transverter-related problems can be eliminated if they are monitored and controlled by cable modems, for example, by a control signal path between transverters and cable modems, or integration of modem functions and RF functions. • Dynamic range: The wireless environment requires a much larger receiver dynamic range than does the HFC environment. The signal level variation with weather/foliage needs to be addressed with suitable AGC and/or larger dynamic range. The static part of this variation can be compensated for by attenuators, and using programmable attenuators would avoid a complex installation procedure. • Modulation: As mentioned earlier, support for modulations such as QPSK or 16-QAM for downlink is preferable for robust operation in a wireless environment with co-channel interference and/or to achieve better coverage. • Link-layer error recovery: Cable modem systems usually have no mechanism for packet error recovery in the MAC layer and leave this function to higher-layer protocols such as TCP. This is a reasonable design in cable environments, where the packet error rate is low. However, for fixed wireless systems this is likely to be inadequate due to interference and fading, and link-level retransmission techniques are preferable [10].

SERVICES AND APPLICATION PERFORMANCE System Reliability — After initial equipment validation and improvement, most user sites have been continuously operational for about a year, although a few of them have experienced short periods of outage while transverters were being replaced. These were due to the limited carrier recovery and equalization performance of the cable modem system combined with the carrier drift (up to 8 kHz) of transverters over time and temperature. Best Effort IP Data — The IP data performance of the system is determined by the cable modem system. Round-trip delays between user computers and the AT&T intranet measured by the ping program range from 10 ms when lightly loaded to 70 ms when many modems are active (especially on the uplink). The peak FTP throughputs are 7 Mb/s for download (close to ideal over the 10BaseT Ethernet interface to the modems) and 650 kb/s for upload. Stress-testing was conducted on the base station Web server and one user modem by simulating the workload of 200 Web users on one modem using the SURGE Web workload model [11]. There were no impairments under load, and the performance was as if the computer were connected directly to the server over a regular 10BT connection. VoIP — In our experiment, we focused on the routers at the ends of the backhaul link between the user subnet and the AT&T intranet. WFQ was implemented on the routers to preferentially treat the VoIP traffic between the user subnet and the AT&T intranet. Consequently, even with heavy data traffic between the user subnet and the AT&T intranet, the voice quality of the VoIP calls had little degradation, and the packet round-trip delay between the VoIP gateway and client devices was kept under 20 ms. Telecommuter Applications and Traffic — The measurement system records 1 min averages of various IP traffic characteristics. Although this may not capture the true dynamics of IP packet data traffic, we found some interesting aspects of telecommuters’ applications and traffic: • The main applications were Web browsing, e-mail, network storage/printer access, database access, streaming

IEEE Communications Magazine • October 1999

and push applications (Real Audio, Pointcast, etc.), and IP multicasting. The use of streaming and push applications is increased due to the always-on nature of the service. Except for IP multicasting video service and Web browsing, most applications are symmetric (within a factor of 2) in terms of number of bytes as well as number of packets. Web browsing is symmetric in packets (due to acknowledgments) but fairly asymmetric in terms of bytes. • We observed that many applications are very chatty. While most Web traffic is highly correlated to the input of users, other applications such as desktop productivity applications and server-based e-mail applications generate traffic that is not well correlated with user activity. This is especially true for applications configured to use network storage. Such applications sometimes use network storage for temporary files, creating unnecessary traffic on the network. • Many desktop applications do not react well to, or recover well from, packet losses due to short-term fading of wireless channels lasting several seconds or minutes. These include server-based e-mail applications such as MS Exchange, domain login services, and many desktop productivity applications that use network storage. Most of them eventually time out and alert users with an unclear error message. However, the timeout periods of these wired-LAN-oriented applications are sometimes excessively long when waiting for server responses, and users often manually terminate them too early in the absence of status information. This is in contrast to Web browsers, where users may not even notice temporary channel degradations. This may be due to the fact that Web browsers running over TCP are designed to operate with various bandwidths ranging from dialup modem speeds to LAN speeds, while desktop productivity applications usually assume reliable high-bandwidth LAN connectivity. User Experience — Several users report that there has been a fundamental change in their computing and work behavior since they joined the experiment. The always-on characteristic of the system encourages more efficient online time at home between other activities, and enables quick lookup of e-mail and information on the Internet for immediate use for professional and personal activities. Broadband access enables them to participate in videoconferencing, provides access to multimedia streaming and multicast, allows access to/sharing of large files at work for collaboration, and lets them spend more time at home while maintaining higher productivity. Users also enjoy quick technical support via high-speed software download, remote configuration through SNMP andVNC, and early detection of potential problems by the management system. Also, they greatly appreciate the simplicity of having virtually identical computing environments at home and at work.

CONCLUSIONS AND FUTURE WORK We have been operating a two-way broadband fixed wireless field experiment in the 2.6 GHz MMDS band for more than a year. It provides high-speed intranet and Internet access for telecommuting employees. The network has been operating continuously, except for short, isolated periods of outage related to failure of RF subsystems that were still early in their product cycle. Our experiment, which is one of the first two-way broadband fixed wireless field experiments, offers an end-to-end solution which our satisfied users use in their daily

61

routine. We have developed a Web-based network monitoring/management tool that greatly enhances the ability to manage, diagnose, and optimize the system. The lengthy period of operation has allowed us to make observations about user behavior, weather-related channel impairments, and equipment performance. We have identified several issues in the design of cable-modem-based systems which need to be tailored to the wireless environment. Also, we have measured a significant path loss effect arising from a combination of rain and foliage. Future directions for the experiment include: • Enhancements to IP network features such as QoS provisioning and network management • Use of DOCSIS 1.1 modems that may address some of the performance issues raised in the article and provide voice/data integration features • Provision of in-home wireless LAN extensions at more user sites • Continued measurements and analysis of user traffic and weather-related impairments • Use of lower-order modulation schemes to improve robustness and coverage.

ACKNOWLEDGMENTS We wish to acknowledge a partner in this effort, CAI Wireless, who has provided the use of the spectrum as well as equipment and manpower. We also acknowledge Zhimei Jiang for the load-testing experiments, and our computer administration team for their assistance. We would like to thank all the volunteer users, especially Larry Murphy, Carl Lundgren, and Norm Dannen, for their support in testing various applications and equipment.

REFERENCES [1] T. K. Fong et al., “Radio resource allocation in fixed broadband wireless networks,” IEEE Trans. Commun., vol. 46, no. 6, Jun. 1998, pp. 806–18. [2] M. Luise and S. Pupolin, Eds., Broadband Wireless Communications, London: Springer, 1998 [3] W. Honcharenko et al., “Broadband wireless access,” IEEE Commun. Mag., vol. 35, no 1, Jan. 1997, pp. 20–26. [4] FCC 97-360, “NPRM on two-way transmission in the MDS/MMDS/ITFS spectrum bands,” http://www.fcc.gov/Bureaus/Mass_Media/Notices/ 1997/fcc97360.txt [5] N. K. Shankaranarayanan and A. Srivastava, “On the design of broadband fixed wireless networks,” Proc. IEEE GLOBECOM, vol. 4, article 68.4, Nov. 1998. [6] MCNS/DOCSIS, DOCSIS 1.1 RF Specification, http://www.cablemodem.com [7] T. Richardson et al., “Virtual network computing,” IEEE Internet Comp., vol. 2, no.1, Jan./Feb. 1998, pp. 33–38. [8] V. Erceg et al., “A model for the delay spread profile for fixed wireless channels,” IEEE JSAC, vol. 17, no. 3, Mar. 1999, pp. 399–410. [9] H. Meyr, M. Moeneclaey, and S. A. Fechtel, Digital Communication Receivers: Synchronization, Channel Estimation, and Signal Processing, Wiley, 1997. [10] H. Balakrishnan et al., “A comparison of mechanisms for improving TCP performance over wireless links,” IEEE/ACM Trans. Net., vol. 5, no. 6, Dec. 1997, pp. 756–69.

62

[11] P. Barford and M. Crovella, “Generating representative Web workloads for network and server performance evaluation,” Proc. 1998 ACM SIGMETRICS Int’l. Conf. Measurement and Modeling of Comp. Sys., Jul. 1998, pp. 151–60.

BIOGRAPHIES BYOUNG-JO KIM (S’93–M’98) ([email protected]) received his B.S. from Seoul National University, Korea, in 1993, and his M.S. in 1995 and Ph.D. from Stanford University in 1997, all in electrical engineering. He consulted for several wireless startup companies in Silicon Valley before he joined the Broadband Wireless Systems Research Department of AT&T Laboratories — Research in 1998. He has worked on signal processing techniques for wireless communications, including radio propagation, equalization, diversity techniques, and sequence estimation. His current research interests are mobile communications, networks, and computing, especially wireless/wireline QoS, thin-client mobile computing, and “open” approaches for cross-layer optimization of mobile networks from applications to physical layers. N. K. SHANKARANARAYANAN (S’83-M’92-SM’98) ([email protected]) received a B.Tech degree from the Indian Institute of Technology, Mumbai, in 1985, an M.S. degree from Virginia Polytechnic Institute and State University, Blacksburg, in 1987, and a Ph.D. degree from Columbia University, New York, in 1992, all in electrical engineering. During 1991 he was a visiting researcher at the University of California, Berkeley. He joined AT&T Bell Laboratories in 1992, and is currently a principal technical staff member with the Broadband Wireless Systems Research Department of AT&T Laboratories, Red Bank, New Jersey. His current research interests cover various aspects of wireless and broadband networks, including: performance of shared packet access networks, broadband user experience, broadband fixed wireless networks, resource allocation, microcellular PCS networks, and radio propagation. His earlier work was in the areas of WDM optical network architecture and technology, optical beat interference, subcarrier optical networks, and optical fiber sensors. PAUL S. HENRY [F’87] ([email protected]) is manager of the Broadband Wireless Systems Research Department at AT&T Labs. He received A.B. (1965) and Ph.D. (1971) degrees in physics from Harvard and Princeton University, respectively. Since joining AT&T (Bell) Labs in 1970 he has published papers or patented inventions in several fields, including millimeterwave radio techniques, cosmology, wireless systems, cryptography, and lightwave communications. KEVIN P. SCHLOSSER received an Electronics Technician diploma from DeVry Technical Institute of Technology, Woodbridge, New Jersey, in 1990. He has worked for several IT organizations until late 1992, when he joined Andersen Consulting, LLP, Florham Park, New Jersey. In 1996 he joined AT&T Labs — Advanced Network Solutions. He was one of the designers of a Data Quality Platform for optimizing voice transcription data. Later he was assigned to assist in the deployment of the nation’s first completely electronic bank, Atlanta Internet Bank. In 1998 he joined AT&T Research to assist in the research of broadband fixed wireless systems. He assisted in network design and testing as well as the development of a Web-based network management system for the user trial. THOMAS K. FONG (S’89–M’96) received B.E. and M.Eng.Sc. degrees from the University of New South Wales, Australia, in 1989 and 1991, respectively, and a Ph.D. degree from Stanford University, Stanford, CA, 1995, all in electrical engineering. Upon graduation he joined the Broadband Access Research Department of AT&T Bell Laboratories, Crawford Hill, New Jersey, where he worked on broadband cellular architectures and wireless cable systems. In 1996 he moved to the Broadband Wireless Research Department, AT&T Labs, Red Bank, New Jersey, and studied service and deployment issues of residential broadband access networks. Since May 1998 he has been with @Home Corporation in California working on IP over cable performance and network management for HFC systems.

IEEE Communications Magazine • October 1999