Table of Contents

Introduction

This report presents the results of several preliminary tests of high speed digital data transfers between different computer systems, or networks, with these transfers supported by wireless (Wi-Fi) connections. In particular, the ability to support such transfers was evaluated in different test environments at distances ranging from several meters to several kilometers. These tests started in November 2001 and were completed in January 2003, with testing performed as the wireless equipment, supporting software, human resources and test environments became available.

As will be shown in this report, data transfer rates of 1 Mbps or greater are possible over long distances with the use of relatively inexpensive, commercially available wireless equipment. It should be noted as well that the maximum data transfer rates observed for some of the tests were limited not by environmental conditions, but rather by the wireless hardware itself and by the parameter values available for configuring this hardware.

The Wi-Fi equipment used in these tests was easily configured, monitored and maintained through the use of well supported utilities, such as ap-config. These utilities are available as GPL'ed software from the usual Internet software repositories, such as sourceforge.net, for use on Linux and other open source platforms. Proprietary utilities, as well as a limited selection of GPL'ed software, are also available to manage commercially available Wi-Fi equipment on "closed" source platforms, such as Windows..

Several issues related to the use of wireless connections were investigated during the calibration and data transfer tests performed during the course of this project. A number of factors were examined which effect the use of wireless connections and the data transfer rates that can be attained with these connections. In particular, a number of sources of potential interference with these transfers, such as the presence of natural (trees, hills, rain...) or man-made (electrical power lines, microwave repeater antennas, metal obstructions...) barriers, were included within some of the tests.

It is clear from the results of these tests that it is critical to know which sources of interference are present in the path of a particular wireless connection. In some cases, by removing or reducing the impact of these sources on the signal transmitted between wireless end-points, a significant improvement in transfer rates can be attained. As will be shown in this report for tests performed at a distance of 3150 meters between wireless end-points, reducing the amount of interference from selected sources improved the data transfer rates from 0Mbps to sustained transfer rates of 1Mbps.

Based on the positive results presented in this report for these preliminary tests, it is recommended that more formal testing of "wireless connections at a distance" be performed during 2003 by My_Sponsor personnel. Their primary objective would be to determine the wireless solutions which can satisfy the technical, legal and business requirements that must be addressed to support the high speed data transfer needs of existing or potential customers of My_Sponsor.

All of the figures in this report have higher resolution versions of the images they present available from the author. Also, all text that is blue and underlined represents either an active hyperlink to another document or to another location within this report. A link can be activated by simply mouse-clicking on a figure or on a blue, underlined section of text.

Additional information added on March 2004

This report can be freely redistributed as long as appropriate credit is provided to the author, Johnny M. Brown. The author does not intend to recommend or disapprove of any particular vendor of software or hardware identified in this report, nor is he acting as the agent for a particular vendor.

To cover any loose ends, consider the contents of this report as being covered by the GNU Free Documentation License (GFDL).

The name of My_Sponsor will be provided on request if approved by My_Sponsor.

This report was based on work done in France before February 2003. Hardware and software technologies, prices, references, etc. have certainly been updated, replaced or modified since this work was performed. Some of the Internet references in the text have already disappeared since this report was written, so overstrikes have been applied to the related links (rather than removed, in case the information mysteriously reappears). The author is currently using 802.11b/g+ APs and modems in his own LAN (near Renton, Washington), while examining 802.16 technology for use by customers and his company, ITGS.

Some of the internal links in this report may not respond, depending on which application or browser you are using to view it with. None of the links are active within the pdf version of this report.

Acknowledgments

I wish to thank the owner of My_Sponsor for the loan of the wireless equipment and server LAN from his company. Without the availability of this material, his direct assistance during the project and the interest he showed during the progress of these wireless connection tests, it would not have been possible to complete the project.

I would also like to express my appreciation to one of the senior engineers in My_Sponsor for his assistance in the tests performed in and around the My_Sponsor offices at Pau in November 2002. The success of these tests was a critical step in determining whether it would be useful to attempt additional tests at greater distances between separate networks linked by wireless connections. In addition, he was generous with the interest he showed in the use of Wi-Fi from the beginning of this project, including a visit he made to Pau in early 2002 to discuss how Wi-Fi networks could be used to extend or combine existing wired networks for My_Sponsor customers.

Patrica and Francois Dabat are certainly to be thanked for providing me with unlimited access to their home near Bosdarros. Without this access, it would not have been possible to conduct the medium distance tests, which produced some of the most important results of this report.

Finally, my appreciation is given to my wife, Mary Shaver. She was gracious enough to accept without complaint my very unusual work schedule during these tests, as well as the unpredictable times at which which I had to leave from, or return to, our home in order to conduct these tests.

Test Sites

Pau

Indoor Tests (< 15 meters)

I had almost no hands-on experience with the use of wireless equipment when this project began in mid-November, 2001. As such, one of the objectives of the work I performed during these initial tests was to improve my understanding of the hardware and software which could be candidates for use in later tests. Also, the equipment for the later tests was not available during the first 9 months of this project. Therefore, I hoped that by working with temporary substitutes in the interim, I would acquire at least some of the skills needed to use the equipment effectively when it did become available.

Test Environments

There were three different test environments used in my "basic training" introduction to wireless technology. These environments all made use of the servers within the computer center LAN (Local Area Network) located on the ground floor of the My_Sponsor building in Pau. In addition, they all made use of a PC client supporting several Linux systems (RedHat, Mandrake, SuSE, Debian) which was installed in an office on the floor above the computer center in the My_Sponsor building . Each Linux system could be booted independently through the use of the grub boot utility. The antennas for the wireless devices connected to the client PC were located less than 15 meters from the antennas connected to the server APs (Access Points). However, several thick walls and floors of varying material (stone, cement, wood, metal) were between the client and server antennas.

All of these initial tests were performed within the My_Sponsor building at Pau. This limitation was imposed in part due to lack of the wireless equipment needed to perform tests at longer ranges and in part due to ART regulations. Until 2003, it was officially not legal to conduct tests of wireless network connections within France outside of a building (or single campus of buildings) using the unlicensed 2.4GHz and 5.0GHz bands. These regulations were relaxed in late 2002 and general testing, with permission from ART, of longer range connections in these bandwidths can now be conducted. The tests described throughout this report are only preliminary tests, performed with minimal equipment and human resources to determine if more formal testing would be desirable in 2003.

Hardware Installation

The wireless devices used in these tests were selected because

• they operated in the unlicensed 2.4GHz band
  [FCC channels are 12MHz wide and they are separated by 5 MHz between their centers, with the 1^st channel centered at 2.412GHz and the last, 11^th channel centered at 2.462GHz. ERTI regulations support equivalent channel configurations with up to 2 more channels, while regulations in Japan permit 3 more radio channels. Wireless devices built to support the FCC channels will often also provide transparent support to the full range of ERTI and Japanese channels. In fact, many wireless devices support radio bands above the official ones. For example, the D-Link DWL-900AP supports radio bands up to 2.497GHz.]

• Wi-Fi hardware is widely distributed
  [3 to 4 million APs were sold world-wide in 2002, according to Analysys Consulting in Cambridge, England.]

• they were available at My_Sponsor or could be easily purchased for use at My_Sponsor

• they were relatively inexpensive.
  [The 12 dBi Hyperlink Yagi antennas used in these tests were the most expensive 2.4GHz components at $135 each, which was about 140 euros when they were purchased in September 2002. The D-Link APs cost slightly less than $110 each, with the internal wireless modems costing well under $100 each. Note that 24 dBi Yagi antennas were available for $165 from Hyperlink. Their use would have made it much easier to succeed with connections for the long range tests, but for various reasons (size, availability, too much gain relative to ART regulations...) they were not acquired for these tests.]

• each device was capable of supporting an external antenna.
  [The need to support external antennas by the wireless devices will become clear in later tests described in this report. In brief, to test wireless connections at distances of more than about100 meters requires the use of antennas or the transmitted and received signals will not be strong enough to be detected by the devices.]

The computer center LAN mentioned earlier and the client PC located on the floor above it were used with all three tests. The only difference in the three test environments were the types of wireless devices which were used. These devices were

1. A Proxim Symphony 4110 wireless PCI modem installed in the client PC. This HomeRF (Home Radio Frequency) modem communicated with a Symphony Cordless Ethernet Bridge, which in turn was connected to the computer center LAN using Ethernet. This modem has a maximum connection speed rating of 1.6 Mbps and is capable of falling back to 0.8 Mbps.

2. A D-Link DWL-520 internal PCI modem supplied by the author of this report which was installed in the client PC. This Wi-Fi modem communicated with an Airport or a D-Link DWL-900 AP, which in turn was connected to the computer center LAN using Ethernet. Either a 2.5 dBi "pig-tail" external antenna or a 8 dBi "flat patch" antenna was connected to the internal modem during the tests. This modem has a maximum connection speed rating of 11 Mbps. It is capable of automatically dropping down through several speeds (5.5, 2, 1 Mbps) to support connections when the radio signal is weak. Both 64 bit and 128 bit WEP (Wired Equivalent Privacy) encryption is supported by this modem.

3. A D-Link DWL-900 AP connected to the client PC using Ethernet. This Wi-Fi AP communicated with a D-Link DWL-900 AP, which in turn was connected to the computer center LAN using Ethernet. Various external antennas, ranging in estimated gain levels from 2.5 dBi to 12 dBi, were attached to either or both APs during the tests. The DWL-900 AP has a maximum connection speed rating of 11 Mbps and is capable of dropping down through several speeds (5.5, 2, 1 Mbps) to connect to weak signals. Both 64 bit and 128 bit WEP (Wired Equivalent Privacy) encryption is supported by this AP. Finally, this AP supports SNMP management, which made configuration and monitoring of the hardware very easy.

The PC used as the client for these tests made use of MSI motherboard, a Duron 1.2GHz processor, 256 MBytes of SDRAM memory (100 MHz), PCI 2.0 slots and nearly 30 GBytes of storage and swap space on the main IDE (ATA-66) hard disk.

Software Configuration

Most of these tests were performed using RedHat 7.3, however a few comparison tests were run using the Mandrake 8.1 operating systems. Mandrake was installed in three partitions on a removable IDE (ATA-66) hard disk supplied by the author of this report, with over 20 GBytes of storage and swap supported for this installation. The latest operating system and application updates for each distribution were applied as they became available.

Depending on the wireless equipment used, different device drivers had to be installed to support the Symphony and D-Link internal modems. The firmware on the D-Link internal modem and the D-Link APs was flashed to the latest stable version as it became available. Since the chip sets and firmware on the APs and each of the internal modems were different, different combinations of configuration and monitoring packages were used in each of the three test environments. The wireless specific software consisted of

• The Symphony internal PCI modem: The rl2 package, v1.7.2beta2, developed by Dave Koberstein (davek@komacke.com) and locally patched by the author of this report to compile cleanly under RedHat 7.2, was installed to provide the modem device driver (rlmod.o – proprietary code) and the configuration utility used to set the modem flag values (device ID, host name, card type...). The ifconfig and iwconfig utilities were used to monitor the performance of the modem, modify flag values and detect problems with the wireless communications. Note that the rl2 package could be modified to compile correctly under RedHat 8.0, when this distro was later installed in November 2002 to use for longer range wireless tests, but the driver would not load successfully.

• The D-Link DWL-520 internal PCI modem: Several versions of the linux-wlan package, up through 1.1.16pre7, were used during the testing of this modem. This package provided the modem device driver and configuration utility. As with the Symphony wireless modem, the ifconfig and iwconfig utilities were used to monitor the D-Link modem.

• The D-Link DWL-900AP: Version 1.1.1 of the ap-util package was used to provide the configuration utility for this Access Point. The welcome page for a recent version of the ap-config utility contained in this package is presented in the figure below.
  
  This same package was used to monitor the performance of the APs, examine and modify parameter settings for the APs, start or reset the APs and identify problems with the wireless communications. The author of this package, Roman Festchuck, maintains a mailing list at ap-utils@kiev.iorta.com. Roman was very responsive to questions regarding his software, which I posted to him during my tests, as well as to minor proposals I made to him for improvements in this software.

It is important to note that none of the packages mentioned above, except for the ifconfig utility, are provided as part of a Linux distribution. Each of these packages had to be downloaded from the Internet, compiled against the current version of the kernel being used and updated manually as needed. However, the linux-wlan and ap-config packages were available as rpm files, while ap-config was also available as a deb file, so it is possible that these packages will become part of one or more Linux distributions in the near future.

Test Results

To ensure that the maximum transfer rates observed in the tests reported below were not limited by the bandwidth available in the wired My_Sponsor LAN, the client PC used in the wireless tests was temporarily connected directly to this LAN using a 10/100 Ethernet link. Transfer rates within the PC itself were also tested between the hard disks.

Using the Ethernet connection, maximum transfer rates of 96Mbps were observed, while transfer rates of up to 400Mbps (clearly local caching was involved) between local hard disks were measured. Therefore, the connectivity available within the LAN and within the client PC would certainly not limit the expected transfer rates through a wireless connection between the client PC and the LAN. As a reminder, the maximum transfer rates which can be achieved in theory with these wireless connections are

• 1.6 Mbps for HomeRF

• 11 Mbps for 802.11b / Wi-Fi (IEEE)¹ or HiperLAN (ERTI) standards at 2.4GHz

• 22 Mbps for 802.11b+
  "Wi-Fi+" is a non-standard vendor extension to 802.11b

• 54 Mbps for 802.11g (IEEE)
  802.11g devices started to appear on the market from a few vendors early in 2003. However, 802.11g will not be an accepted IEEE standard, with validation tests available, until the summer or autumn of 2003

• 54 Mbps for 802.11a (IEEE) or HiperLAN2 (ERTI)standards at 5.0GHz
  80-100 Mbps are supported for "turbo-charged" devices available from several vendors as proprietary extensions to this standard.

Symphony 4110 internal PCI modem

I began this test by making some minor modifications to the rl2 package code during November 2001 so that it would compile cleanly under RedHat 7.3 . With some assistance from a My_Sponsor system administrator for selecting modem flag values, the Symphony HomeRF modem wireless connection to the computer center LAN was established successfully in early December 2001. This connection was found to be very reliable over the months that followed. The maximum data transfer rates observed between the client PC and servers in the computer center LAN were approximately 59 KBytes/sec, equivalent to 0.485 Mbps. This rate is very respectable when compared to the maximum data transfer rates available with ADSL wired systems, but significantly less than the theoretical maximum transfer rate of 1.6 Mbps for HomeRF wireless systems.²

Moving the small external antenna to different locations, all within 1.5 meters of the client PC, did not make any difference in the maximum data transfer rates observed. However, the signal strengths measured by the iwconfig tool did vary slightly with the location of the antenna.

D-Link DWL-520 internal PCI modem

I began testing the D-Link DWL-520 modem early in the summer of 2002. The installation instructions for this modem indicate that it should be installed in the PCI 2.1 slot, however there are many messages on the Internet that indicate it will work with a PCI 2.0 slot. In fact, if the external LED on this modem lights up, it is supposed to perform at maximum transfer rates.

Two different types of MSI motherboards were used during these tests, both with PCI 2.0 slots, to see if there was any difference in how well they supported this modem. Unfortunately, both motherboards had difficulties with the internal modem, even though the modem LED would light, the modem would be recognized and the proper drivers installed during a Linux boot, and the modem could be accessed by the linux-wlan and iwconfig utilities after the system was booted.

The iwconfig utility indicated that a packet loss rate of more than 10% was occurring during data transfers between the Airport AP and this modem. If the Mozilla browser was started up, pages would hang during download and sometimes the browser itself would hang. The gFTP utility was used to attempt the transfer of various files, however it was not possible to transfer files of greater than a few dozen KBytes without the transfers halting or the utility hanging.

In summary, the tests using this modem were largely unsuccessful, possibly due to defects in the modem hardware or to the need to install it in PCI 2.0 slots.

D-Link DWL-900AP Access Points

I began tests with the D-Link access points in early October of 2002. Several initial attempts to communicate between the Airport access point in the My_Sponsor computing center and a D-Link DWL-900 access point attached to the client PC were unsuccessful, so I only tried to establish connections thereafter between D-Link devices. It is still very possible that an Airport and a DWL-900 AP can communicate successfully if the D-Link device is properly configured. However, establishing communications between APs provided by different vendors was not as important to me as successfully establishing any wireless communications, a task I believed would be more easily achieved with material from a single vendor.

The biggest problem I had in finally getting a successful link between the two DWL-900 APs was in determining the correct way of configuring the operational mode for each AP. In particular, the choices offered for configuring each AP

• as an Access Point

• as a wireless bridge in Point-to-Point mode

• as a wireless bridge in Point-to-Multipoint mode

• or as an Access Point Client

appeared to give me several possibilities. However, the only configurations that I could get the two APs to work together in was with one AP configured as an Access Point and the other AP configured as an Access Point Client.

By using the point-to-point wireless connections supported by the bridge modes for the APs, I had hoped that I could establish a more secure connection and have a stronger signal between the two APs³ than I would have achieved by configuring one of the APs as a "true" AP and the other AP as an AP client. However, I could not get the bridge modes to work, possibly because

1. I was using versions of the ap-util package in October which had some bugs in them or

2. perhaps because the firmware installed in the APs at the time was not working correctly.

Although I updated the firmware for the DWL-900 APs in November 2002 and used a more recent version of the ap-config package as well, I did not retest the bridge modes for these APs.

After I configured the AP connected to the client PC (using an Ethernet switch) as an Access Point Client and the AP connected to the computer center LAN as an Access Point, I was able to successfully transfer information between the two networks as of October 21, 2002. As I was moving the client PC distro installations from RedHat 7.3 to RedHat 8.0 and Mandrake 8.1 to Mandrake 9.0 during this same week, I was able to test the DWL-900 AP connections successfully using all four distro versions.

Various parameter settings were tested during the weeks that followed. In particular, the settings were evaluated for parameter values which could optimize the security of a wireless connection before tests began outside of the My_Sponsor building. For maximum security, settings were selected and set for

• Admin Password
  As a D-Link DWL-900 AP, like most commercially available APs, is shipped with a default password ("public") preset for the administrator of this device, changing this default value should be the first security action taken by a network administrator! Changing this password is particularly important for the DWL-900 AP, as password management can be controlled by remote access anywhere in the network through SNMP. For the DWL-900 APs used in these tests, the Admin password of one AP was set to "My_SponsorAP1" and, for the other AP, the Admin password was set to "My_SponsorAP2".

• ESSID (Extended Service Set IDentifier)
  The value of "My_Sponsor2002test2003" was set for this parameter in both APs. Only those APs which have registered the same ESSID should be able to communicate with each other, which means that only the two DWL-900 APs used in our tests were supported within our wireless LAN(s). If the ESSID is not set, then the default action of a DWL-900 AP is to attempt a connection with any other wireless device that is within range.

• WEP 128 encryption
  The DWL-900 APs support both 64 bit and 128 bit WEP encryption, with 128 bit encryption clearly preferred for security reasons. The encryption key C7D82F33619966338855221144 was used for all tests conducted with these APs.
  
  WEP encryption is not very secure, even with the 256 bit encryption supported by newer wireless devices (for example, the D-Link 520+ modem). Therefore, it is recommended that a better, external method be used (SSL/TSL encryption provided by most browsers, VPN...) if very secure data transfers are required for wireless transmissions. In addition, the use of WEP encryption has been reported as slowing down data transmissions by nearly 20%. If performance is a concern, which it is very likely to be, then the use of WEP encryption is certainly not the best choice for more secure data transfers.

• MAC address
  It is possible to specify the MAC (Media Access Controller) address for the APs allowed in a wireless network within an ACL (Access Control List) maintained in each AP. As such, the MAC address of the Access Point Client was set in the Access Point ACL and the MAC address of the Access Point was set in the Access Point Client ACL. Setting the allowed MAC addresses meant that in principle, these two devices could only communicate with each other at the hardware level.

Performance of the wireless connection was tested using data transfers managed by the gFTP utility or the Mozilla browser. The maximum data file transfer rate observed with gFTP or Mozilla when using this wireless connection was approximately 2.9 Mbps, which clearly is much less than the theoretical optimum rate of 11 Mbps. The iptraf utility identified maximum transfer rates of approximately 3.2 Mbps for equivalent tests when it was added as a transfer rate reporting utility on November 6, 2002. This 10% apparent increase in the data transfer rate is probably caused by iptraf including the data packet headers and trailers, acknowledgment packets and similar contributors to overhead in the Ethernet transfer rate statistics.

Note that the PC Magazine tests of 802.11b products had consistently shown transfer rates comparable to those found in these tests. For example, in the December 24, 2002 issue of this journal, an article ("Dual Mode Routers Do More than an AP, page 54) identified a transfer time of 177 seconds for a single 50MByte file, which is equivalent to 2.4Mbps. This transfer rate was observed for both Netgear and D-Link wireless dual band routers operating in Wi-Fi mode.

Various antennas were connected to the APs during these tests, with a gain of 2.5, 8 or 12 dBi provided by each antenna. However, the distance between the antennas was less than 15 meters, so even though the radio signal strength was attenuated by walls, floors and other material within the My_Sponsor building, no reduction was seen on the stability or speed of the data transfers when lower gain antennas were used.

It would have been useful for each test to record the signal strength received at the "server" Access Point in dBm units. This server was the DWL-900 AP connected to the computer center by Ethernet. However, with the version of the ap-utils package used at this time (1.1.1), a measurement for received signal strength in dBm units was not available. This capability was added to version 1.3 of the package, which was released in December 2002.

Local Tests (< 300 meters)

Test Environments

There were two test environments used for local tests at the Pau offices of My_Sponsor. The first environment, LocalPau1, included the My_Sponsor building at the server end of the wireless LAN and the home of the parents of the coordinating senior engineer at My_Sponsor, at the client end. The distance between the antennas at each end of the wireless LAN was approximately 300 meters.

The second test environment, LocalPau2, also included the My_Sponsor building at the server end, but the home of a friend of the owner of My_Sponsor was used to located the client end of the wireless LAN.

The figure on the right presents a view of the My_Sponsor building as seen from the location of the client LAN used for the LocalPau2 test. The chimney at the front right corner of the My_Sponsor building can barely be seen between the tall tree in the middle of the figure and the much smaller, bushy tree immediately to the left of this tree.

In this same figure, the ground floor of the My_Sponsor building is blocked from our view by a hedge planted on the left wall. This hedge, as well as other obstacles, could also interfere with the Wi-Fi signal sent between the antenna connected to the AP on the client LAN and the antenna connected to the AP on the computer center LAN.

The distance between the two antennas in this test was approximately 180 meters.

The figure on the right presents a view out of the front door of the My_Sponsor building. We are looking in the direction of the client LAN used for the LocalPau2 test. The hedge on the wall in the previous figure can be seen just on the right of the large tree in the middle of this picture.

As with the preceding figure, it is evident that there is no clear line-of-sight between the antenna connected to the AP, which was located just inside the front door, on the computer center LAN and the antenna connected to the AP on the client LAN. Various trees, walls, cars, bushes and other obstructions shown in the figure could easily interfere with the signal exchanged between the two antennas.

Hardware Installation

The client PC used in the preceding tests was replaced by the notebook computer that was provided by the My_Sponsor offices in Toulouse. This notebook computer and supporting electronic equipment were carried to the two houses used for the test environments described above and the various connections required to assemble the client LAN and power it up were made. The antenna was then located in various positions at each house to determine where the best signal, if any, could be received.

The Access Point Client device was connected by an Ethernet cable to a 5 port, 10Mbps switch, as was the notebook computer, to create a small LAN. A 12 dBi Yagi antenna, with a horizontal beam width of 45 degrees, was connected to the Access Point Client by means of a 6.1 meter length of N type cable. Reverse SMA adapters were used between this cable and the antenna cable as well as between this cable and the Access Point Client device.

Of note is that the signal lose through these cables, adapters and a lightening protector would reduce the true gain of the antenna from 12 dBi to approximately 10 dBi. This estimate consists of about 1.35 dB of attenuation from the 6.1 meters of N type cable and about 0.65 dB of attenuation from the adapters, the short lengths of reverse SMA-compatible cables between the adapters and the antenna (or the AP), and the lightening protector.

The Access Point which was connected by Ethernet to the computer center in the preceding series of tests was used in the same configuration for these tests. This AP was also connected to an 8 dBi (for a horizontal beam of 75 degrees) patch panel antenna by a 2 meter length of reverse SMA-compatible cable. The attenuation from this connection was estimated to be about 0.5 dB, so the actual gain from the antenna was about 7.5 dBi.

Note that the AP power rating of 12.5 dBm combined with this antenna gain of 7.5 dBi results in an EIRP⁴ of 20 dBm, which is equivalent to 100mW. This happens to be the maximum power rating allowed by ART regulations for France up through the year 2002, for a Wi-Fi channel transmission above 2.4465GHz (channels 9 through 13) that is made either indoors or within private property boundaries.

The 2.5 dBi pigtail antenna that comes with the D-Link DWL-900AP was not used for tests over distances of more than 100 meters since

1. it did not provide enough gain to support such long distance wireless connections affected by "real-world" signal degradation sources and

2. it connected directly to the base of the AP hardware, so it was difficult to orient this antenna in an optimal direction.

Software Configuration

The software configuration used in these two tests was the same as that used for the local tests involving the two DWL-900 APs, which was reported above. The same configuration utilities were used as well. However, at the recommendation of the coordinating senior engineer, the iptraf utility was added to provide measurements of the transfer rates which were independent from the Mozilla and gFTP results.

As the notebook computer provided by the coordinating senior engineer for these tests had already been configured at Toulouse for use in the My_Sponsor Ethernet network, we left all of this information (IP address, gateway IP...) unchanged.

Test Results

Before moving the client LAN to one of the two houses used in the LocalPau1 or LocalPau2 test environments, the coordinating senior engineer and I first tested it within the My_Sponsor building. Once the client LAN was assembled and the wireless utilities were installed on the notebook computer (running the Debian distro), we were able to immediately connect to the DWL-900 AP on the computer center LAN. The senior coordinating engineer was able to verify that all of the nodes on the subnet were accessible from the notebook by doing a ping -b. He then launched Netscape and we were quickly able to browse the My_Sponsor home page and download files. The senior coordinating engineer also demonstrated the capabilities of the iptraf utility to me at this time.

After we verified that the client LAN was working properly, the senior coordinating engineer and I moved its components to the house used in the LocalPau1 environment. We were able to set up the client LAN fairly quickly, with the antenna connected to the DWL-900 AP located in various places both in front of and behind the house itself. As the My_Sponsor building could not be seen from this house, the best direction for the antenna was determined by guesswork.

The patch panel antenna connected to the DWL-900 AP on the computer center LAN was moved outside of the My_Sponsor building and placed at a height of about 2 meters. The direction of the antenna was changed several times in our attempts to establish a wireless connection with the client LAN.

Unfortunately, in the LocalPau1 environment, we were not able to establish a wireless connection at any time between the client LAN and the DWL-900 AP connected to the computer center LAN.

Again, there was no line-of-sight between the AP antenna used with the client LAN and the AP antenna used with the computer center LAN. Power lines, trees, buildings, fences, bushes and other materials effectively blocked the Wi-Fi signal between the two antennas. While this result was somewhat disappointing, it did make it very clear to us that wireless network installers must carefully consider all of the possible items in their environment which can interfere with the signal strength of a wireless connection.

Rather than concede defeat in this first attempt to establish a wireless connection outside of the My_Sponsor building, the owner of My_Sponsor contacted a friend who lived in the house we would use for the LocalPau2 tests. This friend gave us permission to set up our client LAN in front of her house (and under protection from the rain, which was increasing in volume). We first moved the 8 dBi patch panel antenna connected to the AP on the computer center LAN to be just inside the front door to the My_Sponsor building and left the front door open. We then set up the client LAN in front of the home of a friend of the owner of My_Sponsor, with the 12 dBi Yagi antenna attached to the AP directed at the front door of the My_Sponsor building.

Success!!!

Although, as the figures above showed, there were a number of trees, masonry walls and metal material in the path of the signal between the two AP antennas, we were able to successfully establish a strong wireless connection between the two LANs. As before, a ping -b command was issued on the notebook used in the client LAN to verify that we could access the My_Sponsor subnet nodes available through the computer center LAN. The Mozilla browser was also launched. With it we were able to browse the Internet, the My_Sponsor home page and ftp servers within the My_Sponsor wired network.

The measurement of the data transfer rates that could be achieved over a wireless connection at a distance was not an important concern for the LocalPau2 tests. The two primary goals of these tests were to

1. verify that a reliable wireless Wi-Fi connection could be made at a distance of a few hundred meters in a real-world environment and

2. that access to a wired network would be transparent, from a customer's point of view, from a wireless network.

However, we did notice that while using Mozilla to download files from the Internet onto the notebook computer in the client LAN, average data transfer rates of over 0.400Kbps were reliably maintained. The computer center LAN used an ADSL line for Internet access, which in principle supports a maximum data transfer rate of 1Mbps. As our downloads using Mozilla had to compete with normal My_Sponsor "wired" Internet traffic for, at most, 1Mbps of bandwidth, the data transfer rates observed over this wireless connection were very acceptable.

Long Distance Tests (~ 15500 meters)

Test Environment

The 10,000 meter view of the environment used in these long distance tests is presented on the right. The My_Sponsor building in Pau contained the north "server" LAN (NSLAN) for this long distance wireless connection test, while the author's home near Haut de Gan contained the south "client" LAN (SCLAN). The distance between the NSLAN and the SCLAN was approximately 15500 meters. The area along the direction of the connection for about 5 kilometers south of the NSLAN is highly urbanized, while the next 5 kilometers are moderately urbanized. The final 5500 meters along the direction of the wireless connection is a mixture of farms and low density urban areas.

A high ridge that often reaches an altitude of over 310 meters is present along the south side of the Pau river, at a distance of between 6 and 7 kilometers from the NSLAN. Numerous trees of over 20 meters in height are on this ridge as well. The antenna for the NSLAN was installed at an altitude of approximately 210 meters and the SCLAN AP antenna was installed an altitude of approximately 450 meters, with the trees on the ridge line between the antennas at an average altitude of approximately 330 meters. As such, the trees on this ridge acted as an effective physical barrier "wall" in the line-of-sight between the two endpoints of the wireless connection.

The figure on the right presents the view in the direction of the NSLAN as seen from the 12 dBi directional Yagi antenna, which was connected to the AP "client" in the SCLAN. As is evident in the figure, there are no natural obstructions to the Wi-Fi signal between the antenna and the distant ridge along the Pau river valley,. The poles, or pylons, for a small power line can be seen just above the roof of the neighboring house. However, any interference from this power line would be minimal since the line is well below the direction of the wireless signal.

Of greater concern are a number of commercial microwave repeater towers on the ridge itself, which cannot be seen in this figure. While these towers are not in the line-of-sight for this test, their antennas could generate interference within the standard Wi-Fi frequency bands.

The figure on the right shows the view in the direction of the SCLAN from the location of the 12 dBi directional Yagi antenna, which was connected to the AP "server" in the NSLAN. The plastic rings attached to the wall on the left side of the figure were used to support the antenna, which was connected by a 6.1 meter long N-type cable to the DWL-900 AP on the server LAN.

At the bottom of the figure, one can see a number of large apartment buildings and a variety of antennas installed on these buildings. As the direction to the SCLAN in this test is between these apartment buildings, clearly the building themselves or the antennas they support could interfere with the Wi-Fi signals that were transmitted during these tests.

The figure on the right contains an enlarged view taken in the direction of the SCLAN from the location of the NSLAN. The highlighted window in this figure indicates the specific direction in which the Yagi antenna was aimed for these tests. Clearly the Wi-Fi signals transmitted through the antenna are likely to be absorbed by the nearby buildings or interfered with by the antennas. Since the antenna radiates its signal over a horizontal beam width of 45 degrees, much of the Wi-Fi signal strength would never get beyond these man-made barriers.

Hardware Installation

The NSLAN was composed of

• the client PC used in the tests performed earlier within the My_Sponsor building

• a 10baseT Ethernet switch (10/100 Mbps)

• a D-Link DWL-900AP which retained its configuration as as an Access Point from earlier tests

• one of the 12 dBi directional Yagi antennas used in the previous tests

• Ethernet cabling to connect the PC, AP and Ethernet switch

• various cables needed to connect the antenna to the AP

• plastic plumbing pipe and various plastic pipe accessories used to securely attach the antenna to the My_Sponsor building (with a minimum of modifications needed to the building), while still permitting someone to easily change the direction of the antenna.

The SCLAN was composed of

• A firewall/web server running SuSE 8.0 that was built from spare parts (M-549 motherboard from M Technology, 350 MHz AMD K6 processor, 64MBytes of DRAM and 128MBytes of PC100 SDRAM, 20GByte boot disk...)

• An application/database server (ABIT KT7-RAID motherboard, 950MHz AMD Athlon processor overclocked to 1.1GHz, 1GByte of PC133 SDRAM, 3 bootable disks containing 140 GBytes of operating systems with applications and documents, 240GBytes of RAID 0 data storage...) Any of the RedHat 8, SuSE 8, Mandrake 9, Debian 3 or Windows 2000 sp3 operating systems could be booted up on this server. One could use VMware to access multiple operating systems, with RedHat acting as the host and one of the other systems acting as a guest, or one could use grub to boot a single operating system.

• A Windows ME system (hardware details of no importance to these tests)

• a 10Mbps 5 port Ethernet switch

• a D-Link DWL-900AP which retained its configuration as an Access Point Client from earlier tests

• one of the 12 dBi directional Yagi antennas used in the previous tests

• mounting equipment purchased at Castorama for the antenna

• Ethernet cabling to connect the PCs, AP and Ethernet switch

• various cables needed to connect the antenna to the Access Point Client

• plastic plumbing pipe and various plastic pipe accessories used to securely attach the antenna to the exterior of the house, while still permitting someone to easily change the direction of the antenna.

Software Configuration

The same software configuration and monitoring utilities (ifconfig, iwconfig, ap-utils, iptraf) that were used in the earlier tests involving the two DWL-900 APs were used for these tests. These utilities were installed on the RedHat and Mandrake systems on the application/data server in the SCLAN. These utilities were also installed on the SuSE system on the firewall/web server in the SCLAN. Version 1.2 of the ap-util package became available with some useful bug fixes in mid-November 2002. Therefore, this package was updated in late November on the PC in the NSLAN and on the two servers in the SCLAN.

All configuration settings for the DWL-900 APs used in the previous tests (ESSID, MAC addresses, WEP encryption key...) were left the same for these tests. As they were working properly in the preceding tests, it seemed less risky to change them than to try to optimize any of them for the longer distances involved. The Wi-Fi channel used for the first set of tests at 15500 meters was channel 5 (2432MHz), although channel 11 (2462MHz) was used for the final set of tests. Channel 11 would be more likely to be used in commercial Wi-Fi applications (ART regulations, less interference...) than channel 5.

Test Results

In a word, unsuccessful. However, several lessons were learned from these tests that would prove useful for later tests.

Before the tests were started, an estimate was made of the reduction in signal strength that might be expected for a Wi-Fi transmission over 15500 meters. Using the power output of the two APs (12.5dBi each)⁵, the gain of the two directional Yagi antennas 12.0dBi each), and the free space signal loss due to the distance separating the two antennas, the reduced power level was estimated as follows:

free space signal loss

= dBm

= dBm

= dBm

= 123.98591 dBm

total power output (AP power + antenna gain)

=

= 45 dBm

estimated reduced power level

= total power output - free space signal loss

=

= -78.9 dBm

A DWL-900AP is rated for detection of Wi-Fi signals at a reduced power level, or signal strength, of -89 dBm. Therefore, the above estimate would seem to indicate that a wireless connection over the distance of 15500 meters should be possible, even with additional factors present which might degrade the strength of the Wi-Fi signal.

The long distance tests were started on November 25^th, 2002. After resolving some configuration difficulties that were triggered by some bugs in the earlier version (1.1.1) of the ap-config utility, the tests were restarted on November 27. These tests continued through the first week of December 2002, at which time they were terminated without successfully making a wireless connection between the NSLAN and the SCLAN.

Some of the lessons learned from these tests were the following

• Total power output was overestimated
  As the SCLAN AP had to be configured as an access point client so the access points as configured by the ap-config utility would communicate, this means that the SCLAN AP operated in a passive mode. In other words, the AP power did not contribute to the total power available for this long distance wireless connection.
  
  In addition, the cables connecting each antenna to its respective AP attenuate the signal sent to to each. The loss of signal from 30.5 meters of N type cable alone is
  cable loss
  = dB
  = dB
  = 6.663 dB
  for 6.1 meters of cable, this is a loss of approximately 1.33 dB. If the loss of the other cables and connectors between each antenna and AP is estimated to be about 0.7 dB, then the total cable loss is about 2.0 dB for each antenna. Using these corrections, an approximation to the corrected total power output is given by
  corrected total power output
  ~
  ~ 28.5 dBm

• Environmental effects were underestimated
  As a clear line-of-sight between the NSLAN and the SCLAN was not possible, a correction to the free space signal loss for "real space" effects is necessary. An estimate of least 3.0 dB was made for the signal attenuation caused by the interference of the buildings just south of the NSLAN. An estimate was also made of an additional signal attenuation of at least 3.0 dB due to the Pau river valley ridge preventing a direct connection between the two LANs.⁶ With these corrections, an approximation to the corrected free space signal loss is given by
  real space signal loss
  ~
  ~ 129.98591 dBm

• The reduced power level was overestimated
  With the corrections made for total power output and space signal loss presented above, the
  estimated reduced power level
  ~ corrected total power output - real space signal loss
  ~
  ~ -101 dBm
  According to D-Link technical specifications, the minimum detection level for a DWL-900AP is -89 dBm. As the estimated reduced power level of -101 dBm is much weaker than this detection level, it is extremely unlikely that a usable Wi-Fi connection could be established between the two LANs with the equipment used for these tests.

• Installing a wireless connection can be difficult for a network administrator
  Although the reader of this report may already be aware of this fact, it is important to note that installing a wireless connection can be more complicated than installing a standard wired connection. Even if the configurations of the APs or modems used for the wireless connection have been preset and tested before placing them in the target environment, as was done for this long distance test, just establishing the connection can be a very challenging task.
  
  As an example, it is necessary to adjust the horizontal and vertical direction of each antenna so that it is aligned as exactly as possible in the direction of the antenna on the other LAN. For this test, the use of highly detailed maps, the clear identification of landmarks and the availability of good quality binoculars was very necessary. Since the Pau river valley ridge made it impossible to see one LAN from the location of the other LAN, aligning the two antennas was very challenging. Of course, since it was not possible to detect a minimum signal, the initial estimates of antenna directions could not be improved by observing changes in signal strengths as the antenna directions were modified.
  
  Of course, other variables impact the success of setting up a wireless connection (the Wi-Fi channel used, lower AP power levels may produce stronger signals than higher AP power levels, antenna movement at one LAN due to wind and rain after the network installer has returned to the location of the other LAN...), with none of these variables of relevance when a wired connection is being installed.
  
  In brief, a network administrator should expect to take more time installing a wireless connection within a network than a wired connection. The greater the separation is between the two endpoints of a wireless connection, either due to the actual distance or to other factors (walls, trees, dust, other signals...), the more difficult it can be to initially install this connection.

• Antennas can be easily installed using low cost, interference-free materials
  With the simple installation kits that came with the antennas used in these tests, supplemented by some inexpensive plumbing supplies purchased at the local Castorama, it was easy to build a variety of mounting structures for these antennas.
  
  Several figures in this report show different ways in which the antennas were attached to buildings or to plastic pipe "masts". Besides being transparent to Wi-Fi radiation bands (tested by putting the plastic pieces in a microwave oven for a couple of minutes), these structures were very light, easy to transport, weather resistant and could be set up either temporarily or for permanent use. Of some interest is that this plumbing material could be used itself to construct inexpensive Wi-Fi antennas, such as the helical antennas described in the Miscellaneous Wireless Web Sites reference below.

• The ap-config utility has some limitations
  Although the ap-config utility is a very good tool and it works very well on many Linux distributions, it does have some limitations that the user needs to be aware of. These include

• An Upload must be performed to save each modified parameter
  It is necessary to upload the value of a parameter to the access point firmware after each parameter is modified. For version 1.3.1 or earlier of this utility, if you changed the values of two or more parameters before performing an upload, only the new value for the last parameter changed would be saved.

• Uploads will not accidentally force firmware resets to default values.
  The warning message concerning the possible resetting of parameter values other than the ones being intentionally changed can be ignored. After dozens of parameter uploads, I have never seen this happen for a DWL-900AP.

• The WEP key values cannot be seen after they have been uploaded
  According to the ap-config author, the chip set used for the DWL-900AP does not allow retrieval of the WEP key values, even by an administrator. This "feature" can be irritating, since the only way an administrator can determine if the correct values are entered is to test the AP with a connection to another AP using the same WEP key values.

• A different password for the administrator and the user is desirable
  There is a reason other than the obvious one of security for setting different passwords for the administrator and user. At times it was noticed that if these two passwords are the same, the ap-config utility would assume a user was connected instead of an administrator. As a result, if a parameter value was changed by the administrator and uploaded to the AP firmware, the value was not really modified.

• The AP is sensitive to voltage changes
  While this is not really a problem for the ap-config utility, it can easily be detected by this utility. Twice the AP values were reset due to accidentally using a voltage transformer delivering 5.5 volts instead of 6.0 volts to the AP. A few times the connections between the AP and the voltage transformer were loose, causing the values to be modified. In one case the AP had to be reset to its defaults and all the parameter values re-entered. In other cases the ap-config showed that some values had been changed (for example,., the type of AP changed from access point client to access point) while the rest of the values were the same.
  
  An update to the DWL-900AP seemed to eliminate the problems caused by loose connections or by connecting and disconnecting the APs from electrical power sources. However, a network administrator should always check the AP parameter values if any problems are noticed with the use of an AP.

Haut de Gan

Indoor Tests (< 10 meters)

All of these tests were done at my home in Haut de Gan using the PC I had available in my own LAN to evaluate Wi-Fi modems and access points. Some of these tests were performed to either confirm results found during the indoor tests at Pau while others were performed as continuations of the tests at Pau. All of these tests were performed between mid-November and mid-December of 2002.

Test Environments

In addition to the reasons given earlier for performing the indoor tests at Pau, the reasons for performing similar tests at Haut de Gan included

• easy access to the equipment at all hours for the tests

• more isolation from man-made sources of signal interference than the My_Sponsor building in Pau

• preparation of my systems for the configuration and middle distance tests which followed these tests and which are described later in this section

As with the indoor tests performed at Pau, the thick masonry walls, doors, floors and other materials used in the construction of my home were used to test their effects on the attenuation of Wi-Fi signal strengths.

Hardware Installation

The SCLAN described in the long distance tests above was split into two smaller LANs for these tests. The DWL-900AP configured as an access point client in the Pau tests was connected to the firewall/web server LAN (FWLAN), while the Wi-Fi modems or the DWL-900AP configured as an access point were connected when needed to the application/data server LAN (ADLAN). The 10Mbps switch used in the tests at Pau handled the Ethernet traffic for the ADLAN. The servers for these two LANs were both installed in my home office for easy access. The antennas connected to the wireless modems, or the antennas and the APs they were connected to, were moved up to 10 meters apart for these tests. This amount of separation required the use of Ethernet and wireless N-type cables which were each several meters long.

As with the indoor tests at Pau, the wireless devices used in these tests fell into three different categories, with these devices sharing the same LAN infrastructures. These categories consisted of

• a D-Link DWL-520 PCI modem and a D-Link DWL-900 AP
  Both of these devices have already been described in previous tests.

• a D-Link DWL-520+ PCI modem and a D-Link DWL-900 AP
  The DWL-520+ modem uses a Texas Instruments ACX100 chip set to support data transfer rates of, in principle, up to 22Mbps. This Wi-Fi+ device does not conform to existing Wi-Fi standards, but it is supposed to interact transparently with existing Wi-Fi devices. As the DWL-520+ modem was not made available for general use until late in the second half of the year 2002, configuration utilities and device drivers have not yet been developed for Linux. Note that D-Link has indicated that their device driver and configuration utilities for this modem will not be made available as open source code.

• a D-Link DWL-900 AP and a D-Link DWL-900 AP
  Both of these devices have already been described in previous tests.

Software Configuration

The usual software configuration and monitoring utilities (ifconfig, iwconfig, ap-utils, iptraf, linux-wlan) available on Linux were used for these tests. However, as the DWL-520+ modem was not supported on Linux at this time, only the Windows utilities available for this modem were used to install and configure the modem. As usual, the latest version of the Windows utilities and drivers were downloaded from the D-Link home page for the modems, while the firmware used for the APs was the latest version available for them.

Test Results

D-Link DWL-520 PCI modem and a D-Link DWL-900 AP

The D-Link DWL-520 modem was taken to my home in mid-November 2002, where it was installed on two different PCs. The first PC, which booted up into Window ME, was one that normally acted as a node in the FWLAN, but was disconnected from the Ethernet for these tests. The second PC used was the application/data server in the ADLAN, with it booted into either the Windows 2000 or the RedHat operating system. The modem was easily configured on both the Windows and Linux systems, with no problems reported by the configuration utilities.

The D-Link DWL-900 AP which had been configured in Pau as an access point was connected through an Ethernet cable to the server in the FWLAN. After the AP configuration was verified, communications between the AP and the modem were successfully established for each installation in turn. However, as was the case with the earlier tests performed at the My_Sponsor offices in Pau with this modem using Linux operating systems, the modem did not work very well. Although a connection to the web server home page could be established from a browser on both PCs, the connections were extremely slow and very unreliable.

As these tests were so unsuccessful, the only conclusions that could be reached were that either

• the modem would not work properly with PCI 2.0 slots on the motherboards tested,

• the modem contained recent hardware updates (many users reported these caused lockups when using WEP encryption),

• the modem hardware was defective, or

• the firmware loaded into the modem was not working correctly.

The latest official firmware was reloaded from the D-Link downloads page and reinstalled, then all the parameter values were entered again. WEP encryption was turned off as well for the modem and set up externally with the linux-wlan utility. However, no improvements in transfer rates resulted from these changes.

Although the test described in the next section showed that a DWL-520+ modem will work correctly when installed at my home on the PCs with PCI 2.0 slots, the DWL-520+ is built on a completely different chip set than the DWL-520 modem. Therefore, it could only be determined that either the modem is defective or this particular modem will not work correctly when installed in a PCI 2.0 slot, although it could be configured without problems on every PC it was installed in. The modem is available for further tests if anyone at My_Sponsor would have a need for it.

D-Link DWL-520+ PCI modem and a D-Link DWL-900 AP

A D-Link DWL-520+ modem supports, in principle, data transfers at twice the speed of a 802.11b device when connected to other 802.11b+ devices. Independent laboratory tests have also shown a 50% increase in throughput when this modem communicates with standard 802.11b devices. Tests to determine if this modem could improve observed Wi-Fi data transfer rates when used with a DWL-900AP were conducted in late November 2002.

The DWL-900AP which had already been included in the FWLAN was used to set up a wireless connection with the DWL-520+ modem. This connection was successfully established with the DWL-520+ modem .

Both the pig-tail (2.5 dBi) antenna that came with the modem and a 12 dBi Yagi antenna were connected to the modem for these tests, with equivalent antennas connected to the AP. Separations from 0.5 meters up to 4 meters were tried in these tests. With the 4 meter separation, there were also two 1 meter thick masonry walls between the antennas. However, changing the antennas or their separations did not effect to any noticeable degree the data transfer rates over the wireless connections established between the two LANs.

The fastest transfer rates recorded during these tests, which were recorded during Mozilla or gFTP file downloads, were approximately 2.2Mbps, which is about 25% less than the maximum rates found in the Pau tests for transfers between two DWL-900AP devices. This transfer rate is much lower than the expected rate of approximately 4.4Mbps, with no apparent reason for this slow down. No information was found on the Internet which might explain the slow down either.

The only reason I could think of for this slowdown was, again, the presence of PCI 2.0 slots in the PCs the modem was installed in. The DWL-520+, like the DWL-520, is designed to use a PCI 2.1 slot, although users of this modem have reported that it works properly if the operating system can detect the modem hardware. Perhaps the PCI 2.1 slots support functionality related to transfer rates that is not available with PCI 2.0 slots. Unfortunately I did not have access to PCs with PCI 2.1 or later slots to determine if their use would correct this problem.

It was useful to verify that the DWL-520+ modem would work reasonably well with the DWL-900AP, as two different Wi-Fi protocols were being used. Also, it was good to see that successful wireless connections could be made with the available Wi-Fi devices between Windows and Linux systems, with no apparent effects noted on the data transfer rates because of the use of different operating systems.

D-Link DWL-900 AP and a D-Link DWL-900 AP

The DWL-900 AP which had been configured as an access point client in the Pau tests was connected to the ADLAN. As with the indoor Pau tests, maximum data transfer rates of approximately 2.9Mbps were observed, with these transfer rates apparently independent from the locations used for the antennas or the antennas used (2.5 dBi pig-tails, 8 dBi patch panel, 12 dBi Yagi). These tests were conducted from late November through mid-December 2002.

The antennas were placed up to 10 meters apart, with up to three masonry walls also between then. The increased distance and the presence of signal attenuating material clearly impacted the reduced power levels, dropping these levels from about -35 dBm for antennas that were 0.5 meters apart to about -75 dBm for the 10 meter separation. However, the data transfer rates were still maintained at nearly 2.9Mbps.

Besides verifying that the two DWL-900AP devices would work properly when connected to either the FWLAN or the ADLAN, these tests also provided the opportunity to try many different parameter settings for the APs. Different cable combinations between the antennas and the APs were also tested, verifying the estimates made for the signal attenuation caused by these cables. Also, as with the DWL-520+ tests, the success of data transfers and the data transfer rates did not appear to depend on which PC operating systems were in use during these tests.

One interesting observation made during these tests was that often the signal strength observed when using the 2.5 dBi antennas was stronger than that observed when the 12 dBi antennas were used. This would often occur when the antennas were very close (less than 3 meters apart, but also sometimes when the antennas were up to 10 meters apart.

An explanation for this non-intuitive result for nearby antennas could be that the antenna fields for the Yagi antennas were not well focused at short distances, so that their effective EIRPs were much less than 12 dBi. For longer distances the reasons are unclear, with further tests needed to determine what the possible explanations might be.

Outside Configuration Tests (~18 meters)

As the initial results from the local and long distance tests conducted at Pau were reviewed, as well as the indoor tests at Haut de Gan, it became evident that results from more controlled tests would be desirable. More pressure for conducting these controlled configuration tests also mounted when early attempts to establish wireless connections over a medium distance, which are discussed in the next section, were not successful.

These tests needed to be conducted in the open air, with full control over any elements in the test environment which might interfere with a Wi-Fi signal and with the same equipment that was being used in the medium distance tests described below. As the earlier indoor tests with the high-gain Yagi antennas at close distances indicated that they were weaker than the low-gain pigtail antennas, a separation of nearly 20 meters between them was desired to ensure that the electrical fields radiated by these antennas were fully formed. Finally, reasonably good weather was needed so that rain, wind or similar factors would not interfere with measurements gathered during the configuration tests.

Test Environment

The figure on the right shows the back side of the west server LAN (WSLAN) antenna used in these configuration tests, with a view of the east client LAN (ECLAN) antenna. The two antennas are separated by a distance of 18 meters. Although there are stone and metal elements evident in the figure, these elements do not interfere with the line-of-sight between the two antennas.

The figure on the right shows the ECLAN antenna looking directly west in the direction of the WSLAN antenna. The solid stone wall of the barn seen behind the WSLAN antenna will probably absorb and reflect some of the Wi-Fi signals transmitted between the two antennas. However, this wall should not interfere with the Wi-Fi signals which are directly transmitted between the two antennas.

The same conditions are true for the house seen in the right side of the figure. The nearest house wall is located about 3 meters to the north of the direct path between the two antennas.

The narrowest gap the Wi-Fi signal must traverse between the two antennas is about 4.5 meters, which is between the well house in the left side of the figure above and the house on the right side. The antennas themselves are between 1 and 1.2 meters above the surface of the ground. The weather during these configuration tests was cool (~10° C), partially overcast, with occasional light rain and very little wind.

The next figure illustrates a series of configuration tests performed to estimate the effects of a metallic object blocking (or absorbing) part of the Wi-Fi signal exchanged between the two antennas. The metal sheet shown in the figure, which is actually a side panel from a PC tower, was placed at a distance of 9, 3 and 1 meters from the WSLAN antenna in the direction of the ECLAN antenna. The signal strengths, or reduced power levels, reported by the ap-config utility on the ECLAN were recorded 16 times at each distance with the block in place. The results of these tests are presented under the Performance Issues section of this report.

A configuration test environment for measuring blocked signal strengths

Hardware Installation

The figure on the right shows the 12 dBi Yagi antenna that was connected to the WSLAN AP. The AP itself is in the box on the right of the figure. The the light gray cable connecting the AP to the WSLAN Ethernet switch can be seen descending to the bottom right corner of the figure. The black electrical power cord that is connected to the back side of the AP is seen descending to the transformer that is out of view below the bottom of the figure. The AP was placed in the box to protect it from the weather, while the concrete tile was placed on the box to prevent gusts of wind from blowing the box over.

To minimize signal loss from the cabling used between the antenna and the AP in the WSLAN, the N-type cable was not used. Instead, only the reverse SMA cables and adapters were used.

The figure on the right shows the 12 dBi Yagi antenna that was connected to the ECLAN. One can see a large black cable that is descending towards the ground in the bottom right side of the figure. This is a 6.1 meter N-type cable used to connect the antenna to the AP for the ECLAN that is inside the house (not visible in this figure).

The vertical adjustment for the antenna can be seen where the antenna is connected to the vertical piece of plastic pipe. This pipe is used to adjust the horizontal direction of the antenna by simply rotating it on the shaft it is mounted on. This shaft and the four legged base it is mounted on were "borrowed" from a large umbrella of the type used on a patio or terrace. The antenna used for the WSLAN was mounted on a similar shaft with a heavier base.

The figure on the right shows the server used for the WSLAN. This server is the same PC that was used as the client PC for the indoor tests at Pau. The 10baseT Ethernet switch used to connect this server to the AP in this LAN is the small rectangular box just on the right of the PC monitor.

On the screen of the monitor is presented one of the displays supported by the ap-config utility. To the left of the screen one can see part of an oil painting created by my wife. This painting has absolutely nothing to do with the configuration tests, but it is for sale.

The figure on the right shows the client PC used in the ECLAN. This client is the same firewall/web server that was used in the SCLAN and the FWLAN for some of the earlier wireless tests. Next to the PC monitor is the 10Mbps switch used to connect the PC with the AP.

This AP, configured as an access point client, can be seen on the window ledge in the upper right portion of the figure. The adapter between the reverse SMA coming from the AP and the N-type cable that is connected to the external antenna (by an adapter to another reverse SMA cable) can be seen just inside the middle of the window. As in the previous figure, on of the displays supported by the ap-config utility is shown on the monitor.

A discerning reader can see that both the monitor and the tower it is sitting on in this figure are Gateway 2000 products. The monitor contains all original components from 1995. By contrast, none of the original electronic components remain in the tower.

Software Configuration

The usual software configuration and monitoring utilities (ifconfig, iwconfig, ap-utils, iptraf) available on Linux were used for these tests. These utilities have been described several times already in this report, so I will not duplicate these descriptions here.

The power levels set for the APs for different Wi-Fi channels and the selection of which channel or channels were made available to the access point client in the ECLAN from the access point in the WSLAN were modified for each configuration test. All remaining parameter values for the APs were unchanged from previous tests.

Each configuration test performed for a specific combination of Wi-Fi channels (1, 3, 5, 6 or 11) and AP power levels was repeated 16 times to get a minimal set of statistics for the test. Repeating these tests many more times could have been very useful. However, the time available for these tests, a rapid degeneration in weather conditions during the test period and my limited patience combined to impose this limit of 16 repetitions.

Test Results

The main results of these configuration tests are summarized in the following list. All of these tests were conducted in late January of 2003 (22/01 to 24/01).

1. The best reduced power level was observed with Wi-Fi channel 11 (2462MHz)
   The result was rather surprising, since the lower the frequency is of a radio transmission, the better should be its reduced power levels. Whether this result is due to the characteristics of the DWL-900AP hardware or some other factor is not clear. However, the difference between the reduced power levels observed for channel 11 and the other Wi-Fi channels, while measurable, was not very large. At most a 10% difference in maximum reduced power levels was measured.

2. The reduced power levels for the channels occurred in at least 3 modes.
   The maximum reduced power levels for the channels ranged from -39 dBm down to about -44 dBm. Intermediate reduced power levels for these channels occurred from -59 dBm down to about -65 dBm. Weak reduced power levels were more channel dependent, occurring around -70 dBm, -80 dBm and -90 dBm. More measurements could have produced better statistics, especially for the weak reduced power levels. However, the presence of these power level modes is clear in the available data.
   
   The power level difference between the maximum and intermediate modes for all the channels appears to be about 20 dBm. It is very tempting to attribute this difference to the combined gain contributed by the Yagi antennas. That is, the intermediate reduced power level readings could result from a good wireless connection being established without the need for the gain provided by the antennas. The maximum power gain would then result from the antennas contributing their combined maximum gain to the wireless connection.
   
   The weak power levels could be due to harmonics between the two APs, such as signals coming from the server antenna reflecting from the barn to the client antenna, signals from the server antenna passing through the house wall back to the client antenna, by modes in the radio frequency field generated by the server antenna, etc. As I am clearly just guessing here, this could be a fruitful area of further investigation.

3. Changing power levels to the WCLAN antenna from the client AP had no effects, but these same changes made to the ESLAN antenna from the server AP did make a noticeable difference.
   As the AP for the WCLAN was operating as an access point client, these results confirmed the hypothesis presented in the long range test at Pau that the client AP does not contribute to the power levels for the Wi-Fi signal. For the server AP, the difference between sending no power to the antenna and sending the maximum power possible was from 5 to 6 dBm. The weaker the maximum power level was for the signal (for example, if the signal was partially blocked), the greater was the effect of changing the power level.

4. Materials placed in the line-of-sight between the antennas will reduce the maximum reduced power levels observed
   Perhaps this result may appear to be self-evident. However, as the configuration test environment was already in place, performing tests with blocking material seemed like a good idea. With the blocking material positioned 1 meter in front of the WSLAN antenna, the attenuation of the reduced power level for channel 5 was a consistent 15 dBm. As this value is comparable to the gain provided to the wireless connections in the configuration tests by both Yagi antennas, clearly this test demonstrates the impact that interference by man-made or natural elements in the wireless environment can have on the signal strength for a wireless connection.

From these configuration test results, I decided to use only Wi-Fi channel 11 in my attempts to establish a wireless connection for the medium distance tests conducted in late January of 2003. The power level settings between the APs and their antennas was set to their maximum levels, although this setting probably had no effect on the LAN with the access point client. Finally, as much material as was possible was removed from the line-of-sight between the two antennas for the medium distance tests.

Some calculations made after the medium distance tests were concluded are presented below. If these calculations had been made before these tests, then perhaps I would have been more optimistic about the outcome of these tests. Then again, this optimism might have been misplaced, since the most positive results of these tests were not produced until the very last day of the test period.

Estimated maximum distance for a wireless connection at 2.462GHz

These calculations make use of the information gathered during the configuration tests to estimate the maximum distance at which a wireless connection could be established using the available equipment. We begin with the free space signal loss that is applicable for the distance between the two antennas in all the configuration tests (18 meters) and the radio frequency at which Wi-Fi channel 11 is centered, 2462MHz. This loss is calculated using the relation

free space signal loss (2462MHz)

= dBm

= dBm

= dBm

= 65.3312111 dBm

Next we take the results measured for the reduced power level output of channel 11

measured reduced power level (2462MHz)

= total power output - free space signal loss

~ -40 dBm

and rearrange the equation used to calculate it so we can back out an estimate of the real total power output produced by the APs and the antennas in our two LANs.

real total power output (AP power + antenna gain - cabling attenuation)

= Measured reduced power level + free space signal loss

~ -40dBm + 65.33

~ 25.33 dBm

If we assume that the antenna gains are contributing 19 to 20 dBi to the real total power output, this means the AP itself is only contributing 5 to 6 dBm. So either the antennas are not providing nearly as much gain as expected, or the AP power levels are much lower than advertised (see the related results for varying power levels from APs to Antennas which are presented in the the Observations section below), or something completely unaccounted for is reducing the power output. Although this is an interesting subject for possible future investigation, for now we can take this result and try to extract a little more information with it.

If we go back to the free space signal loss equation, we can invert it to determine the maximum distance at which a connection can be made with our equipment, assuming the published detection sensitivity of the DWL-900AP, -89 dBm, is correct. That is, we can derive the relationship

free space signal loss at maximum distance (2462MHz)

= real total power output + minimum reduced power level

= 25.33 dBm + 89 dBm

= 114.33 dBm

= dBm

By rearranging the terms in the previous two lines, we find the maximum distance that our equipment might be able to establish a working wireless connection over is

maximum distance for free space signal loss (2462MHz)

~

~ 5.07 kilometers

As is evident from the above result, presuming it is roughly correct, it is very unlikely that the long distance test in Pau could possibly have succeeded. However, as these configuration tests were conducted while I was trying to get the medium distance tests discussed below to succeed, the above result did give some hope that these tests could succeed. The major unknown in these final series of wireless connection tests, as will be seen, was how much signal attenuation was produced by the material in the line-of-sight between the two LANs.

Medium Distance Tests (~3150 meters)

This is the final series of tests to be described in this report. These tests were started in the second week of December 2002 and continued until January 31 2003. No testing was attempted between December 19 2002 and January 6 2003 as the author of this report was otherwise engaged at the time.

Test Environment

The topography for these tests in shown in the following figure. The line indicates the direction between the two LANs used in these tests, with the arrows indicating the locations of the two LANs. The L'Oustalet house is on a hill above the town of Haut de Gan, while the Barbe house is just south of the town of Bosdarros. The departmental road (D-934) shown near the middle of the figure arrives at the town of Gan after continuing a few more kilometers due north of the figure. Gan is approximately 8 kilometers directly south of Pau, as can be seen in the map showing the topography for the long distance tests presented earlier in this report.

A clear line-of-sight is potentially possible between the antenna attached to the Loustalet client LAN (LCLAN) and the antenna attached to the Barbe server LAN (BSLAN). However, a number of trees near the LCLAN house in the west and on a ridge about 1 kilometer east of this house could have caused considerable signal attenuation. On a positive note, since these tests were performed in mid-winter, there were no leaves on these trees to make this possible attenuation problem even worse.

In addition to the trees on the ridge mentioned above, there was an electrical power line installed just below the top of the ridge that ran parallel to the ridge. Finally, at about a kilometer north and a kilometer south of this ridge relative to the line shown in the figure, there were two large microwave repeater towers installed. However as the electrical power line and the microwave towers were well outside of the line-of-sight between the two Wi-Fi antennas, I hoped that interference from them would be negligible.

Below is a figure that displays the view from the LCLAN antenna location looking in the direction of the BSLAN. As there are over 3 kilometers between these trees and the BSLAN itself, it is evident that any Wi-Fi signals coming from the BSLAN will have difficulties penetrating this natural barrier. The curious reader might like to know that I had already removed a number of branches from the tree in the immediate foreground, as well as several branches from a tree about 10 meters directly behind this tree.

However, as the weather during these tests was often very bad, initial attempts to make a wireless connection between the LCLAN and the BSLAN were conducted with the trees in place as shown.

As initial attempts to establish a wireless connection between the two LANs were very unsuccessful, the rather low-tech task of clearing out possible sources of signal interference was initiated. As this task involved the removal of several trees and/or branches near the LCLAN house, I first acquired the approval of my landlord to do so. The reasons for clearing these trees that were given to my landlord are not necessary to repeat here. Suffice it to say that these reasons were acceptable to him.

By late January, I had improved my health considerably due to my involvement in much more strenuous physical exercise than usual and I had recovered some of my lost skills in tree maintenance. To keep the costs associated with these tests at a minimum, I used only the tools I already had available to cut down trees and remove branches. As a result of my labors, by late January the view from the LCLAN house in the direction of the BSLAN was as it appears in the next figure.

As the reader probably does not have a pair of binoculars available so that the BSLAN house can be seen in the center of the previous figure, an enlargement of the "target" is provided in the next figure. The light colored region in the highlighted area at the middle of the figure is the side of the BSLAN house which faces the LCLAN location. From this enlargement, it would appear that there is now a clear line-of-sight between the two LANs, but this is not completely correct.

The next figure presents a view of the LCLAN house that was photographed through an open skylight from which the antenna for the BSLAN protruded. The LCLAN house is represented by the blurry rectangle that can barely be seen, just below the sky line and behind some trees in the upper middle portion of the figure.

Using skills at digital photography of which I was not previously aware that I possessed, I took a digital photograph of the LCLAN house from the BSLAN house skylight through one lens of a 7x35 pair of binoculars. After making some digital enhancements to this photograph with the GIMP tool, the figure below was produced. The antenna for the LCLAN is located a few meters from the right end of the light-colored wall in the figure. As is evident from this figure (well, sort of), there are still a number of trees remaining in the line-of-sight which could attenuate the Wi-Fi signal transmitted between the two LANs.

I attempted to minimize the possible effects of the remaining trees interfering with the Wi-FI signal at the LCLAN end of the wireless connection by placing the antenna as high as I could. By connecting a couple of plastic pipes together and mounting the antenna on the top of these pipes, I was able to position the antenna as is shown in the figure below.

The LCLAN which is connected to this antenna is inside the windows directly behind the antenna. Note the snow falling in the air and on the roofs of the buildings in this figure. The presence of this snow visually demonstrates that a wireless connection outside of a building must take into account many conditions which can degrade the Wi-Fi signal strength which have no impact on an indoor connection.

Hardware Installation

The figure on the right shows the antenna as it was installed for the BSLAN. The N-type cable used for the LCLAN was not used with this antenna so that attenuation of the Wi-Fi signal could be minimized. However, this meant that the AP must be rather close to the antenna. As the antenna was about 3 meters above the roof of the house, the AP was forced to dangle below it, supported only by the reverse SMA cable that connected it to the antenna and a very unstable pile of boxes.

This was clearly a temporary arrangement, since the AP was not protected from the weather (it did rain during these tests and the temperature was just above 0° C). Also, as the owner of the BSLAN home was not in France during these tests, the skylight through which the antenna mast was extended could not be left open for security reasons. However, this very non-professional installation worked very well during the tests.

In the figure above it is possible to see a thick cable going across the sky behind the antenna. This is an electrical power line, which could case interference with the Wi-Fi signal. However, this house was the last one on this power line and only two circuits were turned on in the house. Therefore, the current flow through the line should be minimal, so that an induced interference that was generated at GHz frequencies would be nearly unmeasurable.

The figure on the right shows the antenna installation for the LCLAN. As the mast of the antenna is over 4 meters long, it was necessary to use the N-type cable to connect the antenna to the AP. However, the extra height provided by this mast was expected to minimize the signal attenuation caused by the trees shown in earlier figures that were between the two LANs. Therefore, the loss of about 1.35 dB due to the use of the N-type cable was considered to be an acceptable exchange for the possible very large signal attenuation caused by the trees. Some of the N-type cable can be seen looped near the bottom of the figure.

An electrical power line which terminates at this house is just to the right of the chimney in this figure. It was hoped that this power line was far enough away from the antenna that it would not cause any interference with the Wi-Fi signal, but this was only an assumption.

The BSLAN is shown in the figure to the left. This is the same LAN as the WSLAN used in the configuration tests described earlier. The 10baseT Ethernet switch used to connect the PC to the AP in this LAN can be seen sitting on the the top of the PC monitor.

There was no electrical power in the room in which the BSLAN was installed. However, there was one working electrical outlet that worked elsewhere in this house, so the roll of extension cord in the foreground of the figure was used to bring power into this room for the LAN. Did I mention that this room was very cold as well since there was no heat available in the house?

The LCLAN is shown in the figure to the left, with the shutter on which the antenna connected to the AP was installed visible through the window in the background. The DWL-900AP configured as an Access Point Client can be seen on the window ledge. This LAN is the same one as was described in the long distance tests as the SCLAN. The application/data server is on the left in this figure, with the firewall/web server in the middle of the figure. The PC supporting Windows ME is installed in a separate room about 5 meters away.

Sitting on top of the application/data server PC is a pair of binoculars. Another pair was kept at the LCLAN. These binoculars, as well as detailed maps and a compass, were found to be important tools for use in the correct alignment of the Yagi antennas. These antennas have a fairly wide horizontal beam width of 45° across which their maximum gain is projected. However, the narrow, clear field of view between the two LANs required that these antennas be aligned rather carefully or the their EIRPs were drastically reduced.

Unfortunately, two days after these tests were completed, the power supply in the application/data server shorted out, causing a catastrophic failure for most of the electrical components contained in this PC (graphic card, SCSI card, disk drives...). A partial replacement of both servers has been installed since then, with some components from the firewall/web server used in building the replacement servers. However, it will be impossible to repeat any of the wireless connection tests made with these servers exactly as they were described in this report, since these servers no longer exist.

Software Configuration

The usual software configuration and monitoring utilities (ifconfig, iwconfig, ap-utils, iptraf) available on Linux were used for these tests.

The next figure presents a screen shot of the BSLAN PC monitor. On the screen are a couple of windows which are provided by the ap-config utility and a small window that monitors file download statistics for Mozilla. The lower ap-config utility indicates that the wireless connection made between the two LANs has a reduced power level of -88 dBm (RSSI=relative signal strength indicator = reduced power level) and a Link Quality of 15 (1 is very good, 20 is very bad), while the file download monitor is indicating an average data transfer rate of approximately 1Mbps over a time period of nearly 10 minutes.

The next figure also presents windows on the BSLAN server screen which were produced by the ap-config utility and the Mozilla download monitor. In addition, it shows a window containing real-time information that was updated by the iptraf utility. In this figure, the reduced power level for this wireless connection is only -90dBm although the Link Quality has improved slightly to 12. The download tool indicates an average data transfer rate of only 450Kbps. However, these statistics are somewhat misleading since they were gathered while the ap-config utility was being used to reset the wireless connection to measure several reduced power level and Link Quality values. The iptraf tool indicates that the current instantaneous data transfer rates over the Ethernet were around 500Kbps, which are consistent with the Mozilla download monitor results.

Test Results

!!!! Success !!!!!

Well, to be more precise, eventually the wireless connections were successful. During the initial attempts to establish a wireless connection between the LCLAN and the BSLAN, which started in early December 2002, the only results were frustration on my part. This state of affairs continued throughout January. Several possible factors were considered in trying to determine the causes of these continued failures, such as

• the foggy weather made it difficult to establish a good alignment between antennas

• the rain, snow and occasional high winds interfering with the Wi-Fi signal

• interference caused by materials in the windows and the window frames when attempts were made to use the antennas from inside either the BSLAN or LCLAN houses when the weather was very bad

• wrong configurations for the DWL-900APs

• all of the above plus signal attenuation caused by the trees between the LANs

Continued attempts to account for all of these possible factors were made throughout January. The most satisfying efforts I made to reduce or eliminate these factors came from removing branches or cutting down trees near the LCLAN house. Some of the results of these efforts can be seen in the figures describing the test environment for these tests. While this work was definitely low-tech, apparently it contributed strongly to what I considered to be a high-tech success.

After I installed longer masts to support both antennas, I made one final attempt to establish a wireless connection between the two LANs on the very rainy, windy day of January 31, 2003 (31/01/2003). As I had indicated to the owner of My_Sponsor that I would return the wireless equipment and the BSLAN PC to My_Sponsor during the first week of February, the tension was starting to mount.

Almost to my disbelief, I succeeded! Although the wind kept changing the alignment of the antenna on the BSLAN, so that I had to keep repositioning it to detect a Wi-Fi signal, I was able to perform several successful connection tests during the day as well as make a few observations. Such as

• I was able to log onto the servers on the LCLAN from the PC on the BSLAN using ssh to bring up the iptraf and ap-config utility windows from the LCLAN servers o on the screen of the LCLAN PC.

• I could reset the wireless connection from the BSLAN PC , to collect reduced power level and Link Quality statistics, and the connection would successfully restart each time.

• become aware of the following issue: if the wireless signal was lost due to a system problem (for example, the antennas were out of alignment for too long), the wireless signal would not automatically reconnect when the problem was resolved. The connection had to be reset with the ap-config utility operating on the BSLAN AP to restart the connection.

• File transfers interrupted by resetting the wireless connection or by a short time period when the wireless connection was lost would continue when the connection restarted.

• The relative percentage of transmission failures (ACK failures, failed packets, FCS failures...) observed during these tests were much higher than seen during the configuration or indoor tests performed earlier. Whether these higher failure rates were caused by the lower Link Quality and lower reduced power levels seen in these tests or whether the reverse is true could be an interesting area for further investigation.

• The lower reduced power levels clearly reduced the data transfer rates for this Wi-Fi connection. This was an expected result, based on the information available for the DWL-900AP devices. However, these reduced data transfer rates for a wireless Wi-Fi connection compare very favorably with the maximum rates supported by a wired ADSL connection

The following figure provides some statistics regarding the wireless connections that were made during the middle distance tests at Haut de Gan. As can be seen, the signal strengths, or reduced power levels observed during these tests , average about -90 dBm. The Link Quality averaged about 16, which is very low as well. Clearly these wireless connections are operating at the very edge of supportability by the equipment used in these tests.

Wi-Fi signal attenuation factors

The free space signal loss for the middle distance tests is given by the relationship

free space signal loss (2462MHz)

= dBm

= dBm

= dBm

~ 110.2 dBm

the actual signal loss observed, using the real total power output estimated during the configuration tests, was

Actual signal loss

= real total power output + observed reduced power level at 3150 meters

= 25.33 dBm + 90 dBm

= 115.33 dBm

The summation of the attenuation factors that reduced the strength of the Wi-Fi signal is simply the difference between the results of the above two relationships, which is approximately 5 dBm. While it is not possible to determine the magnitude of the contributions of each of the possible attenuation factors to this result, it is very likely that most of the attenuation was caused by the tree branches that still remained in the line-of-sight between the two LANs. The second most important contributor was probably the humidity (rain, snow, mist...) in the air between the two LANs. All the remaining possible factors (dust, wind, electrical interference...) probably contributed very little to this result.

Observations

AP and Modem Selection

Vendor

The D-Link access point and modem equipment used during these tests was selected simply because it was of low cost, supported external antennas and was available. It is highly recommended that other vendors, such as SMC and Linksys, be considered in acquiring additional wireless equipment for commercial use. Of course, there are more expensive alternatives to the AP equipment used in these tests, which are available from several vendors , that support additional functionality (Ethernet routers, firewalls, more power). Hopefully the results of the tests in this report will provide some guidance in selecting which APs and modems satisfy the requirements for specific wireless installations.

Channel and Bandwidth

Some of the results from the configuration tests indicate that the use of some Wi-Fi channels may be preferred to using others. I suspect which channels are best will depend not only on the hardware and firmware used for the APs or modems selected, but also on the environment these modems are placed in. For example, in an urban environment, selecting an optimum channel could depend strongly on what other wireless networks are in the area and the channels they make use of. On the other hand, for ease of maintenance it may be preferred to simply select certain channels to use in any environment. Finally, ART regulations could be a determining factor in selecting which channels are to be used for wireless connections that are needed beyond the boundaries of a single location.

Wireless Technology

Moore's Law is clearly controlling the future of wireless technology for at least the next few years. However, which type of technology will be most successful in France will depend at least as much on regulations and licensing issues as on advances in wireless technology. Some of the competing technologies are indicated in the following list, these being

• 802.11a (HiperLAN2)

• 802.11b (HiperLAN)

• 802.11b+
  As 802.11g devices are arriving soon, the use of 802.11b+ devices, with a maximum published data transfer rate of 22Mbps, is likely to be of no interest for business use. The 802.11g devices will be IEEE and ESRI compliant, unlike the 802.11b+ devices, and they will support faster data transfer rates for a similar cost.

• 802.11g
  I believe that 802.11g will largely replace 802.11b APs and modems during the next couple of years. However, as the 802.11g standards will not be finalized until the middle of the year 2003, use of this equipment to support customers is not recommended until late 2003 at the earliest. Of course, testing of 802.11g devices is highly recommended during the year. Some questions which need to be answered are how well 802.11g devices work with existing 802.11b devices and how many channels will 802.11g devices support.

• Dual Band (802.11a and 802.11b, 802.11g and 802.11a) devices
  Use of dual band APs and modems could be important for the next few years as existing wireless networks are upgraded and users of mobile devices (PIM, PDAs, tablets, notebooks) try to connect to different wireless providers. It may also be desirable to support both 2.4GHz and 5.0GHz technologies for technical reasons.
  
  For example, customer requirements may be better satisfied by using 802.11b for longer range networks outside of buildings and 802.11a for internal networks. The 802.11a networks may be usable only over shorter distances than 802.11g. However, they are likely to maintain an advantage other 802.11g devices with regards to data transfer rates, the number of channels they support, and lower interference from other electrical devices (cell phones, microwave ovens....).

• non-802.11a/b/g wireless solutions [3G, 802.16, satellite, laser...]
  Yes, there are many other wireless technologies out there. However, they are either much more expensive than Wi-Fi solutions, less well distributed, less mature, or suffer from other difficulties. If 802.16 devices are ever built for unlicensed frequencies, they could be of considerable interest to customers demanding high data transfer rates (100Mbps or better) for fixed wireless connectivity. There are no line-of-sight restrictions for 802.16 devices as they operate at 266GHz and 802.16e devices are supposed to provide connectivity like 802.11 devices for mobile users.

Antenna Selection

Purchase or Build Options

The tests described in this report definitely required the use of external antennas to support Wi-Fi connections at distances of a few hundred meters or greater. There are a small number of vendors which provide these antennas at reasonable prices. These antennas are well built, reliable, and of course will work with any wireless technology that used the 2.4GHz bands.

It is also possible to build antennas for use in Wi-Fi networks. The cost of the materials used in constructing antennas with 8 dBi to 24 dBi of gain can be less than 10% of the cost of purchasing equivalent antennas from vendors. Each antenna can be easily designed to provide a specified gain using GPLed software. Also, most of the materials needed to build these antennas can be found at a local home supply store, such as Castorama, with the few remaining materials available from local electronics supply stores.

While building their own antennas may not be a task a wireless network provider wants to be involved in, there could be a market segment of customers who would be very interested in doing so. In particular, small businesses, community organizations or academic centers might welcome the opportunity to build these antennas if it helped keep their wireless network costs down to a minimum. Therefore, some experience in building these antennas by the wireless provider could be very useful, as would guidance for some customers in how to build their own antennas.

Regulatory Issues

The most important constraint on the deployment and use of wireless networks in France are the ART regulations and the ESRI policies these regulations are based on. By far the biggest difficulty with the ART regulations is their very strong restrictions on the EIRP that is permitted for a wireless transmission outside of a single location (campus or building). As these regulations were written at the beginning of the year 2003, they effectively prevent the creation of Wireless LANs (WLANs) such as Metropolitan LANs (MLANs), Community LANs (CLANs) or WLANs in sparsely populated areas (SLANs).

As ART has become somewhat more flexible with during the last 18 months, at least in granting permission for testing Wi-Fi network components by potential providers, perhaps they will also support the installation of WLANs on a case-by-case basis. If this is not the case, then the high bandwidth wireless communications market in France will continue to be restricted to deployment within a building or on a single campus.

Business Opportunities

The distribution of wireless services and products to customers is only briefly touched upon in this report, as the main purpose of this report it to describe the tests and their results conducted over the last few months with wireless devices. However, if the results of this report are to be of any practical benefit to its readers, some attention to the business opportunities presented by the support of wireless networks is probably warranted.

Types of Wireless Installations

There are several possible types of wireless installations that could be deployed by, or services supported for, a wireless network provider. These include

• WIPOP (Wireless Internet Point of Presence)
  Streaming media or Intranet/Internet services

• Hot spots or Internet cafes

• WAN (Wide Area Network) extensions with WLANs
  This market segment should grow rapidly over the next few years

• Backbones for SOHO (Small Office Home Office) LANs

• campus communications (academic, government, industry)

• Alternative/supplemental/backup/replacement WLANs for wired networks
  This should be a very important market segment in France during the next few years, with a wide variety of customers potentially interested in integrating WLANs with existing or planned wired networks.

Types of Wireless LANs (WLANs)

There are many forms in which a WLAN can appear. A few of them are listed below, but there are several others which could easily be of interest to a wireless service provider.

• BLAN (Business LAN)
  I invented this term to account for a WLAN that is deployed to support a single business customer. This business may be located at one or more sites, with the WLAN used to support communications between one or more of these sites. BLANs may be completely internal to the customers intranet, which both increases and decreases some of the security issues associated with wireless communications. A BLAN will usually be integrated with the customers existing LAN or WLAN, either to extend these wired connections or provide an alternative to them (load balancing, emergency communications...).

• CLAN (Community LAN)
  CLANs are very popular in North America and England, with their presence rapidly expanding into other European countries (Sweden, Norway, Estonia...). Some of these CLANs are very large, supporting their own Internet Web sites and providing extensive help information on these Web sites about CLAN technologies and services. CLANs tend to operate as self-help organizations, without any or very little assistance requested from wireless service providers. However, new CLANs or CLANs that start to encounter administrative overhead costs could be markets for wireless service providers in France.

• SLAN (Sparse LAN)
  I invented this term to describe the wireless LANs which are starting to appear in rural or transitional (between a city and its surrounding countryside) regions in many countries. SLANs are often started because the user communities have no other alternative to gaining high speed access to the Internet. These potential customers may or may not even have dial-up access, depending on the presence of and cost of their wired telephone connections. Within France there are many small communities which would benefit from an SLAN. In addition, gaining permission from ART to service an SLAN should not be too difficult, as the benefits of SLAN use to these communities is clear. Also, as these communities are not a very attractive market for large telecommunications companies (did someone mention France Telecom?), there should be little resistance from such companies to attempts by small wireless providers to support SLANs.

• MLAN (Metropolitan LAN)
  As MLANs are deployed in high density population areas and can be used by a wide variety of customers, the service and support for MLANs can generate considerable revenues. However, the competition for MLANs will be very strong and, particularly in France, the regulatory issues can be difficult to resolve.

• ...

Services and Products

There are a wide variety of services a wireless provider could offer to customers, many of which have wired equivalents. The important point to consider with these services is that they are available to customers without the need to build a wired network infrastructure to support them. In addition, the wireless support for these services can be upgraded as the customers needs expand with little or no impact on the customer.

Some of the services and products that could be offered to users of wireless networks include

• VOIP

• Audio or video (A/V) streaming
  Existing 802.11b devices already support some configuration options which improve data transmissions for audio and video streaming. For example, it is possible to select short packet headers instead of the standard long packet headers used for file transmissions to reduce overhead.

• standard home user services
  These services would include e-mail, Internet chat, basic desktop publishing support and moderate amounts of data storage

• thin client or diskless client support
  I do not know if Microsoft is aware of this, but reliable high bandwidth wireless connections could damage their business model considerably. If sustained data transfer rates of 20Mbps or better are provided by wireless networks, this would mean that many customers would longer need to purchase fat clients and all of the expensive Windows software that usually comes on them.

• SOHO services
  These services would include the standard home user services, with additional business applications, more desktop publishing support and larger amounts of data storage. Some use of remote computing services could also be supported (for example, basic ASP services)

• medium to large customer services
  The full range of xSP services could be offered to large an medium sized customers over wireless networks, as well as the wireless services (failover for wired networks, load balancing, mobile user support...) themselves

• turn-key installations

• corporate wireless network administration

• training

• ...

Performance Issues

Alternatives to Wi-Fi

Wi-Fi is certainly not the only wireless technology available in the marketplace. The various levels of support for cellular telephones (2G, 2.5G, 3G) are alternatives to Wi-FI, although only 3G can offer close to the same data transfer rates as are supported by Wi-Fi devices. Newer versions of 2.4GHz technologies, such as 802.11b+ and 802.11g are natural evolutions of the 802.11b technology in the direction of faster, more secure data transfers.

Of course, high speed wireless communications are supported at many other bandwidths, ranging from 1 GHz to 30 GHz, for a variety of wireless architectures. These include satellite broadband transmissions of up to several Mbps, point-to-point laser transmissions in the 1Gbps to 10Gbps range, and proprietary multi-GHz transmissions which use licensed transmission bands which support up to 1Gbps.

In general, however, these technologies are much more expensive than 802.11 technologies, have very limited markets, require special training to support and are more difficult to maintain. In addition, licensing fees and regulations put many of these technologies out of the reach of smaller wireless service providers.

Environmental

Examples of the effects of a number of factors in the environment on Wi-Fi communications have been presented in the test results for this report. Some of these factors are listed below.

• Distance

• Weather (rain, wind, fog...)

• Visibility (trees, dust, smoke, hills...)

• Man-made (buildings, electric power lines, microwave interference, wireless hot spots...)

A test which involved applying different AP power levels, combined with partially blocking the Wi-Fi signal, was performed during the configuration tests at Haut de Gan. Some of the results from this test are presented in the next figure. This test emulates the effects of man-made interference with the signal, such as might be caused by a building, combined with a weakened signal, such as could be caused by a long distance between the endpoints of a wireless connection. The effects of both types of environmental effects are quite evident.

The figure below shows the effects of blocking the Wi-Fi signal at various distances from the antenna connected to the WSLAN used in the configuration tests. The effects on the maximum reduced power levels measured for the Wi-Fi signal are evident, with the block causing up to a 15 dBm reduction in signal strength. This blockage emulates the effects of man-made sources with different attenuation characteristics interfering with a Wi-Fi signal.

Technical

Antennas

Although only a limited number of antennas were used in this report, they provided many useful results. For example, the indoor test results at Haut de Gan indicated that sometimes a low gain antenna delivers a Wi-Fi signal that is less attenuated than does a high gain antenna. Some of the of issues related to antenna use and selection are listed below.

• Maximum Estimated Gain
  Naturally the gain that is provided by an antenna can have a large influence on a long distance wireless connection. However, it is also important that as much of this gain be usable as is possible.

• Directional and Omni-Directional Types ("Pig-tail", Yagi, Helical...)
  If an antenna is mostly intended for support of point-to-point connections, then a highly directional antenna is preferred to make maximum use of the gain it supplies. If a wireless connection is multipoint or required in arbitrary directions between an AP and its clients, then increasingly wider horizontal or vertical beams will be useful. The directionality of an antenna depends on the type of radio field it produces. For example, Yagi antennas produce radio beams that are directional, but still rather broad (40 to 80 degrees). A helical antenna will produce a very tightly focused radio beam, which is approximately 5 degrees across in both the vertical and horizontal directions.

• Cabling and Related Connections
  A Wi-Fi antenna must be attached in some manner to an access point or modem to be of any use to a wireless connection. However, the cables, adapters, lightening protectors and any other materials used in establishing this connection can significantly reduce the gain of an antenna. For example, the combination of cables and adapters used to connect a Yagi antenna to a DWL-900AP in the tests described in this report reduced the maximum 12 dBi gain for this antenna to an actual gain of between 8 and 10 dBi. As a reduction of 6 dBi is equivalent to doubling the distance between two Wi-Fi antennas, the impact of these cables and adapters on a wireless connection is definitely of great importance.

• Effective Isotropic Radiated Power (EIRP)
  As this term is used by the ART to identify how much power can be produced at a wireless endpoint, determination of its value for a "French connection" is required.
  

APs and Wireless Modems

Many technical factors can impact the rate of data transfers for a wireless connection and the Quality of Service that can be maintained for a wireless connection. The list below presents some of these factors.

• Bus Adaptation (ISA to PCI, PCI 2.0 versus PCI 2.1...)
  For older 802.11b modems, it has been reported that their internal use of ISA protocols caused considerable overhead with the PCI bus. This overhead effectively reduces their maximum possible data transfer rate to less than 7.5Mbps. Newer 802.11b modems often use PCI 2.1 or PCI 2.2 functions to reduce overhead on the PCI bus, but these modems often do not work well on PCI 2.0 compliant buses.

• Public Layer Compatibility Protocols (PLCP)
  All Wi-FI APs and modems must convert the signals transmitted at the public layer level into internal formats their hardware can process. This conversion can use up to 15% of the bandwidth supported by these devices.

• Packet Header Length (long for normal file transfer, short for video streaming)

• Encoding (FHSS, DSSS,CDMA, W-CDMA...)

• Encryption (external, WEP 64 bit, 128 bit, 256 bit...)

• Vendor (chip sets used, firmware, dBm power of hardware)
  Even if the same chip sets are used, modems and APs produced by vendors often are supported by different, incompatible firmware from the viewpoint of device drivers. This is even true of firmware installed by some vendors for different updates of the same model of a device!
  The dBm power level produced by an AP or modem can vary considerably between vendors. At times this power difference is used to justify large price increases for the more powerful devices, since it increases their effective ranges relative to the less powerful devices. However, if the less powerful devices support an external antenna, it is often a less expensive and more flexible solution to purchase the less powerful device and connect it to a higher gain antenna than to buy the more powerful device.

• Channel Selection and Channel Power

The importance of selecting the Wi-Fi channel that is the best match for the environment the wireless connection is to be used in was discussed in the test results for the configuration tests. The figure below shows the reduced power levels that were measured in one of the configuration tests when the only change made in the wireless connection configuration was which channel was being used.

While the differences in the maximum reduced power levels measured for each channel were not very large, they are noticeable. In the case of the medium distance test, if another channel other than channel 11 had been used, the small reduction in the maximum reduced power level detected for any other channel in these tests could have prevented the middle distance test from succeeding.

The figure below shows the results of changing the power levels from the WSLAN (S) and ECLAN ( C ) APs to their respective antennas during the configuration tests conducted at Haut de Gan. While the impact on the maximum reduced power level measurements is clear for the WSLAN changes, there is only a minor change, if any, caused by the ECLAN power level changes.

Although the WSLAN is supposedly changing the power levels by 100mW, the reduced power level only changes by 5 to 6 dBm, which is far less than 100mW. Also, as a maximum change of 6 dBm is much less than the rated power output of the DWL-900AP, while100mW is much greater than the maximum rated power output, it appears that some thing is going on here that is not very well understood. Of interest is that an AP power output of 5 to 6 dBm is consistent with the real total power output results determined in the test results for the configuration tests. The readers of this report are welcome to contribute their own opinions.

Access Point Connection Configurations

The performance for data transfers in a wireless network will depend somewhat on the different ways in which the access point connections are configured. A few of the possible connection possibilities are listed belows.

• AP to Ethernet
  The data transfer rates can be degraded in this configuration by the need to convert packets of information built using wireless data protocols to packets that use Ethernet protocols. Adding to this problem is the need to add acknowledgment packets and packet delivery failure notifications into the data stream for both protocols.

• AP to selected AP clients

• AP multicaste
  Multicaste wireless communications add considerable overhead in the form of addition packet filtering and automated channel selections to maintain optimum data transfer performance levels across several wireless connections. As channels become saturated, an AP such as the DWL-900AP is supposed to provide access to other available channels for its clients to connect to.
  
  Neither the impact of having more than one AP client on a single channel, nor the effects of an AP changing channels to load balance its clients, on the overall performance of wireless connections made with an AP (or group of APs) have been examined by the tests presented in this report. However, this is definitely a topic that should be investigated.

• ...

Conclusions

This report has presented the results of a wide variety of wireless connection tests performed both within a single building and across various distances between LAN sites. The ability to successfully establish wireless connections using Wi-Fi devices in a number of environments was demonstrated. In particular, a wireless connection between LANs separated by more than 3 kilometers was successfully tested, with a sustained data transfer rate of 1Mbps recorded over this connection.

While it was unfortunate that a wireless connection over a distance of nearly 16 kilometers could not be created, the calculations presented in this report indicate quite clearly that such a connection exceeded the capabilities of the available equipment. That is, with somewhat more powerful antennas or a more connection friendly environment, a connection over such a distance is certainly feasible.

Wi-Fi devices currently available in the market could support the deployment of WLANs in France with acceptable costs to their users and reasonable profits to their providers. With faster data transfers already possible with 802.11b+ and 802.11g devices at 2.4GHz, WLANs could easily compete with, or complement, the use of wired networks in a number of market segments. The main constraint that I can identify on the deployment of WLANs in France, and in the southwest of France in particular, are the current ART regulations for the use of unlicensed transmissions at 2.4GHz.

References

There are thousands of references on wireless networking subjects which can be found on the Internet, in journals and in books. Some of the Web sites which present lists of bookmarks on various wireless networking topics are provided in the Wireless Bookmark Lists below. The Miscellaneous Wireless Web Sites list which follows presents selected web sites containing more specific information about wireless networking subjects.

Wireless Bookmark Lists

1. Linux Wireless LAN Project HowTo
   http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Wireless.html

2. Robert's Linux Wireless Bookmarks
   http://www.csoft.net/~dummy/robert/bookmarks/technology/os/linux/wireless/index.shtml

3. Wireless Bookmarks At Magnolia Road
   http://www.magnoliaroad.net/wireless_bookmarks.html

4. Wireless Bookmarks From Turnpoint
   http://www.turnpoint.net/wireless/bookmarks.html

5. Wes's Bookmarks: Linux Wireless
   http://www.atlantek.com.au/~wes/bookmarks/Linux/Wireless/

Miscellaneous Wireless Web Sites

If the description of a web site is in italics, then the version of that reference available in 2003 is available from the author of this report. If you want to be certain that you have access to the latest copy of a reference, assuming it is still available on the remote Web site, please click on the URL shown for that reference.

1. Helical Antennas – Design and Assembly Instructions for 2.4GHz Wireless Systems
   http://www.wireless.org.au/~jhecker/

2. Linux Wireless LAN Project FAQs
   http://www.linux-wlan.com/linux-wlan/index.html#FAQ

3. Michael Young - Understanding Decibels and Their Use in Radio Systems
   http://www.ydi.com/

4. Richard Miller-Smith – Limits on Bandwidth for Prism 2.5 PCI Modems
   http://lists.linux-wlan.com/pipermail/linux-wlan-devel/2002-April/000987.html

5. Rob Flickenger - Tips and Tricks For Setting Up Wireless Networks
   http://oreillynet.com/wireless

6. Wireless Definitions and Radio Transmission Equations
   http://www.qsl.net/n9zia/wireless/index.html

References Specific To This Report

Several references that are specific to the content and format of this report are provided below to assist two types of readers of this report. The first type of reader is one with interactive access to this report who would like to look at additional information that was gathered to produce this report. The second type of reader is one who does not have interactive access to this report. The full Internet address of the links for some of the entries in this report are presented in a list below in case such a reader would like to view these links with their own Internet access applications.

Additional Information Gathered For This Report

[The additional information described in this section is available from the author on request. It is not included with this document primarily due to the size of the folders containing this information. This information consists of 40 MBytes for various references and 41 MBytes for all of the unedited photographs taken during the creation of this report]

Several references used during this project have been copied into a local folder and linked to this document. If hyperlinks are activated for the format of this document that is available to you (HTML, pdf...)and this local folder has been placed in the correct location relative to this document, you can view these references by clicking on this underlined text.

If you would like access to all of the figures presented in this report, you can view them in a single folder which can be accessed by clicking on this underlined text. The unedited versions of these figures, along with additional digital photographs taken during the tests which are not included in this report, can be accessed by clicking on this underlined text.

Finally, the statistics gathered during the configuration tests conducted at Haut de Gan can be accessed by clicking on this underlined text.

Explicit Internet Addresses

1. Johnny Brown – http://johnny_m_brown.home.comcast.net

2. sourceforge.net - http://sourceforge.net

3. ap-config - http://ap-utils.polesye.net/

4. linux-wan - http://www.linux-wlan.com/

5. D-Link - http://www.dlink.com

6. Hyperlink Technologies - http://www.hyperlinktech.com/

7. 802.11g devices - search "802.11g" - http://www.pcmag.com/
   
   
   

1While the IEEE (Institute of Electronic and Electrical Engineers) determines the specifications for these standards, the regulation of their use and the certification of devices which comply with these standards is performed by the FCC (Federal Communications Commission) for all of the states in the USA. The ERTI (European Telecommunications Standards Institute) performs a function similar to that of the IEEE, but regulation of the standards proposed by the ERTI is done at the national level within European member states. For example, ART (Autorité de Régulation des Télécommunications ) regulates the licensed (2.5G, 3G, 30GHz...) and unlicensed (2.4GHz, 5.0GHz...) radio bands in France.

2The Symphony Bridge speeds may have been restricted by the transfer rates for the ADSL line used in the computer center LAN for Internet connections. If so, this restriction would explain in part why the maximum transfer rates for this wireless connection were less than expected.

3I had hoped that if both APs were configured as bridges, this would result in the transmitters and receivers of both APs contributing to the signal power level . This could be an incorrect assumption on my part, but it is worth investigating further.

4Effective Isotropic Radiated Power, actual power transmitted in the main lobe after taking in account all cable losses and antenna gain. Based on an isotropic antenna.

5The ap-config documentation indicates that it is possible to set power levels of up to 100mW for each Wi-Fi channel when configuring an access point. As 100mW is equivalent to 20dBm and D-Link rates the maximum power of this AP model at 12.5dBm, perhaps this is not true for all APs. The author presumes that it is really possible to set power levels up to 100% for each Wi-Fi channel according to the AP being configured, which for the DWL-900AP would be about18mW.

6Each 3.0 dB reduction in signal strength reduces the distance at which a Wi-Fi signal can be detected by 50%.

Johnny Brown                      France -3- 2003                      Wi-Fi Connection Tests