Each signal path from the transmitter to the receiver has a unique time delay and phase shift associated with it. For this reason, the received signal can be severely distorted. Some frequencies within the signal bandwidth combine constructively, increasing signal strength. Others combine destructively, thereby reducing signal strengths at that particular frequency. In the extreme, deep nulls (or fades) are formed as shown in Figure 2.

In the indoor environment, energy reaching the receiver antenna via delayed paths can spill from one symbol into subsequent symbols. In fact, some secondary paths can have delays equivalent to several symbol times. Consequently, energy transmitted in one symbol period can distort several subsequent symbols. Thus, multipath can cause inter-symbol interference (ISI) resulting in severe signal distortion. Further, this signal distortion varies even within a single symbol period.

Dealing with Multipath in .11b
As most designers know, IEEE 802.11b devices employ a waveform known as complementary code keying (CCK). The underlying modulation is single-carrier quadrature phase-shift keying (QPSK).

At 11 Mbit/s, groups of 8 QPSK symbol are used to form code words. Each code word represents 8 bits of information. The QPSK symbol rate is 11 million symbols per second (Msymbols/s). Thus a symbol period is about 91 ns. However, some secondary paths have delays of 400 to 500 ns. In these situations, ISI can result in distortion of as many as five or six subsequent symbols (Figure 3). Recall that multipath can result in the presence of many delayed paths, each with a different time delay.

Fortunately, in single carrier systems designers can compensate for the effects induced by multipath by employing time domain-based channel equalization techniques. Channel equalization can overcome the effects of multipath-induced ISI, but there are some caveats. As data rates are increased, either by increasing the symbol rate or by using more complex signal constellations, equalizer complexity must increase to maintain the same level of system performance.

Eliminating ISI Altogether
In contrast to single carrier systems, OFDM systems do not compensate for multipath-induced ISI—they eliminate the problem altogether. OFDM-based radios distribute the data payload among many "subcarriers", each closely spaced as shown in Figure 4. For 802.11 a/g applications, there are 52 subcarriers, 48 of which are used for data transmission. The remaining four subcarriers are used for pilot tones. Thus, the data rate on each individual OFDM subcarrier is much lower than on a comparable single carrier system. This enables a significantly longer symbol period. For 802.11 a/g, the symbol duration is 4 μec, or about 44 times longer than a CCK symbol.

As Figure 5 points out, the OFDM pulse contains a guard interval. It can be thought of as a prefix to the symbol containing redundant information that can be discarded at the receiver without affecting the ability to correctly decode the symbol. At first glance, the underlying reason for carrying this "excess baggage" might not be apparent. In fact, the guard interval is a crucial feature of OFDM systems. The length of the guard interval is selected specifically to be longer than the delayed paths encountered.

For 802.11 applications, the OFDM guard interval is 800 ns. When the signal is digitally processed at the receiver, the signal content in the guard interval is effectively rejected. The remaining rectangular pulse is 3200 ns in length. Significantly, this rectangular pulse is completely free of multipath induced ISI, since the guard interval is longer than any anticipated delayed path. Distortion due to multipath is still quite possible, but only from within the same symbol. The effects of multipath can be effectively compensated for on a subcarrier-by-subcarrier basis by means of an amplitude shift and phase correction, both of which are constant for each subcarrier over the entire 3200 nsec symbol period. Since multipath distortion is time-invariant within a symbol period, the OFDM signal can be efficiently analyzed in the frequency domain. Once the guard interval is eliminated, the remaining portion of the OFDM symbol is a 3200 ns rectangular pulse.

In the frequency-domain, the rectangular pulse is represented by a sync function with zero-crossings at intervals corresponding to the inverse of the pulse period—312.5 kHz (1/3200 ns) in this case. This is the exact frequency spacing for the subcarriers. At the zero crossings, there is no energy from adjacent subcarriers. Subcarriers are therefore said to be "orthogonal". Even though the OFDM subcarriers are very closely spaced, they do not interfere with each other.

There is another major benefit of using frequency-domain methods. There are a number of efficient Fast Fourier transform (FFT) algorithms that can be used to perform the channel compensation. As data rates and multipath delays increase (up to 800 ns), the complexity of the channel compensation circuitry in an OFDM baseband processor does not increase. This is dramatically different from the single carrier case, which explains why OFDM systems are thought to be much more practical for WLAN applications at higher data rates.

Working with Legacy Equipment
Discussions about waveforms are great, but unless it all works in practice, it doesn't amount to much. OFDM was attractive to the members of the IEEE 802.11g Task Group due in part to the reasons described above. Another benefit was that the technology had already been developed for use with 802.11a systems already operating in the 5 GHz bands. The main hurdle in bringing OFDM to the 2.4 GHz band was making it work well with legacy 802.11b equipment.

The main channel sharing mechanism for 802.11 WLAN systems is carrier sense multiple access/collision avoidance (CSMA/CA). In essence, CSMA/CA is a "listen-before-talk" scheme. Every node in a basic service set (BSS), which consists of an AP and all associated clients, listens for transmissions. If the channel is idle, each client begins decrementing an internal back off timer. Once the back off timer reaches zero, the node may begin transmission. There are additional features that ensure the back-off timer settings are randomized to reduce (but not eliminate) the possibility of a collision.

... and there in lies the problem. In order for this happy little scheme to work out, each radio must be able to hear all of the other radios associated with the same AP (including of course, the AP itself). Legacy 802.11b radios would be effectively unable to hear newer 802.11g radios using OFDM.

The 802.11g Task Group solved this problem by use of a request-to-send/clear-to-send (RTS-CTS) feature that is already supported by every 802.11 radio. The original intent of the RTS-CTS feature was to help minimize the impact of collisions if some radios are out of range of other radios associated with the same AP. Although this feature is rarely used in practice, it's a mandatory feature that is tested in every Wi-Fi device.

Due to the topology of an 802.11 network, all nodes must be within range of the AP, even though they might not be within range of each other. This is often referred to as the "hidden node" problem. When this condition is detected, the RTS-CTS mechanism can be invoked to reduce the probability of a collision. In 802.11g, the RTS-CTS feature can be used to facilitate network operation when a mixture of 802.11g and legacy 802.11b clients are operating within the same BSS. Referring to Figure 6, a CCK RTS-CTS exchange precedes each OFDM high rate packet and the subsequent OFDM ACK.

When RTS/CTS is in use, most stations will hear the RTS, and all stations will hear the CTS. In either case, each node receives information indicating the length of the subsequent OFDM packet and ACK transmission.

Every station has an internal timer referred to as the network allocation vector (NAV). The NAV is set to have the same duration as the OFDM packet exchange. The NAV acts as in parallel with conventional carrier sensing, and is referred to as a virtual carrier sense mechanism. The channel is not considered idle unless 1) no active signal is detected, and 2) the NAV timer has expired. Once both criteria are met, stations can once again begin to contend for channel access. In this manner, 802.11b and 802.11g radios can operate in a mixed environment with an 802.11g AP.

It should also be noted that every 802.11g client and AP must be capable of falling back and operating exactly like a legacy 802.11b device. Therefore, migration to 802.11g technology can be smooth and painless. As new 802.11g AP's are brought online, legacy 802.11b AP's can remain in service and will be fully interoperable with newer 802.11g clients.

Answering RTS-CTS Concerns
Much has been made of the fact that use of RTS/CTS as a means of facilitating backward compatibility results in extra network overhead. While this argument may have merit in some regards, results of throughput calculations presented in Figure 7 demonstrate that this mechanism results in higher throughput than any other backward-compatible alternative.

There is an added bonus. Through RTS-CTS, every 802.11g AP can monitor for the presence of legacy equipment. In those instances when legacy 802.11b equipment is not present, use of RTS/CTS can be dropped, and throughput improves correspondingly.

Results in Figure 8 show the dramatic improvement in throughput under this condition. All throughput results assume three active clients and include a statistical estimate for the probability of collisions.

It's All About Range
As mentioned above, OFDM is an excellent waveform for indoor WLAN applications. In addition, simple physics indicates that range is proportional to wavelength. Therefore, use of the OFDM waveform at 2.4 GHz should result in greater range than similar systems operating at 5 GHz. This is certainly true in free-space, but there has been a lot of controversy about how this would apply indoors in a home or office. As it turns out, there's ample evidence to suggest that the range advantage at lower frequencies should hold true indoors as well.

From an RF propagation perspective, objects are measured in terms of wavelengths. At 5.3 GHz, the signal wavelength is less than half that of a 2.4 GHz signal and, as a result, everything in the environment is larger. In other words, walls are "thicker", and people and furniture are "bigger". Therefore, objects scatter and attenuate the RF energy more effectively.

For a line-of-sight (LOS) scenario, propagation differences shouldn't prove to be much of a problem. However, most WLAN systems operate indoors where the line-of-sight scenarios are limited and most radio signals must penetrate through walls and around furniture and other obstructions.

All else being equal, a system operating at 2.4 GHz should have a significant range advantage over a similar system operating at 5 GHz, indoors or outdoors. Until recently, this has been difficult to demonstrate. Prior to the emergence of 802.11g, equipment operating at 2.4 GHz has used a much different waveform and supported lower data rates than 802.11a equipment operating at 5 GHz. Therefore, any differences in performance could be attributed to a number of factors, operating frequency being just one.

The arrival of 802.11g equipment has therefore been met with some anticipation. Because 802.11g and 802.11a devices can use essentially the same waveform, meaningful direct comparisons can now be made.

To compare the range capabilities provided by 802.11a and g, a series of tests were conducted at a facility in Palm Bay, Florida. Figure 9 shows the floor plan and test point designations. This indoor office environment is also fairly representative of radio propagation conditions in a "typical" home. Ceilings are 9 feet high and internal wall construction is drywall over studs. Importantly, there are no cubicles. Offices have floor to ceiling walls and are equipped with doors.

During the testing, three types of equipment were used. These include:

1. 802.11a equipment using OFDM at 5 GHz: Equipment tested was a commercially available SMC EZ Connect client (Model # SMC2735W) and AP (Model# SMC2755W).
2. 802.11g equipment using OFDM in the 2.4 GHz band. Equipment tested was an Intersil 802.11g reference design. Transmit power was 30 mW @ 6 Mbps, with power reduced at higher data rates to ensure compliance with the IEEE 802.11g requirements for error vector magnitude (EVM). The optional mode which does not use RTS-CTS was tested in order to provide a direct comparison with 802.11a equipment.
3. PBCC-based equipment operating at 2.4 GHz was also tested. PBCC-22 is also an optional mode of operation for 802.11g devices. This equipment was included in order to demonstrate the performance of OFDM vs. a single carrier high rate system operating in the same 2.4 GHz band. Equipment tested was a commercially available D-Link AirPlus client (Model# DWL-650+) and access point (Model # DWL-900AP+)

Under practical operating conditions, actual network throughput is nearly always lower than theoretical limits. Measured throughput results presented in Figure 10 through 12 are consistent with this trend. With that said, measured end-to-end throughput numbers are nevertheless still quite impressive and will enable high rate applications to be distributed over these wireless LANs.

Referring to Figure 10, it is clear that 802.11a equipment achieves high data rates at short range, or down the hallway where unobstructed line-of-sight propagation is possible. However, data rate falls off rapidly when the signal must penetrate walls and furniture. This effect is so pronounced that there are significant areas that have no service at all. This is entirely consistent with expectations for 802.11a equipment.

In comparison, Figure 11 shows coverage for 802.11g equipment operating in the 2.4 GHz band. RTS-CTS was not used for this round of testing in order to test equipment that was as close as possible to 802.11a equipment. Any differences in performance can therefore be attributed to the operating frequency. And, as is clearly seen, there are significant differences. The region of peak throughput (greater than 15 Mbit/s) is much larger. In fact, range at all data rates was much further than for the 802.11a equipment tested.

Note that connectivity was preserved in all but the extreme edge of the floor plan, at ranges up to 150 feet through several interior walls. Based on the results shown, it is reasonable to assume that a centrally located 802.11g AP could provide complete coverage of a large home with a throughput of at least 10 Mbit/s.

Figure 12 shows the coverage map for equipment using PBCC-22. Peak throughput for this equipment was approximately 7 Mbit/s. Relative to the 802.11g equipment, coverage is not as complete. The areas of the floor plan for which there was no coverage are somewhat larger for the PBCC-22 equipment. Further, peak throughput rates for PBCC-22 are less than half of those achieved by either 802.11a or 802.11g equipment.

Measured throughput numbers at the test locations shown in Figure 10 through 12 are shown in the bar chart of Figure13. Note that 802.11g equipment offers superior data rates to PBCC-based equipment at all test points. It is also clearly superior in terms of measured throughput to the 802.11a equipment at all data points except in those situations for which line-of-sight propagation is available (i.e. the hallway). In these cases, 802.11g and 802.11a demonstrate roughly similar throughput numbers.

In enterprise and public access applications, 802.11g and 802.11a will compliment each other. 802.11a has more channels available in the 5 GHz bands, and therefore offers better scalability. Range limitations can be compensated for in the enterprise and public access use via installation of more AP's (not a cost effective option in the home space). At the same time, there are many situations in the enterprise for which range is still a critical consideration. In those circumstances, 802.11g infrastructure will be the superior choice. The advent of low cost dual band clients will mean that enterprise users will be able to roam seamlessly between 802.11a and 802.11g APs.

About the Authors
Jim Zyren is the director of strategic marketing for Intersil's Wireless Networking Product Group. He holds an MBA from the University of Central Florida, a Master of Science in electrical engineering, and a Bachelor of Science degree in electrical engineering from the University of Michigan. Jim can be reached at jzyren@intersil.com.

Eddie Enders is a systems engineer with Intersil's Wireless Networking Product Group and holds a BS degree in Electrical Engineering from Florida Institute of Technology. Eddie can be reached at eenders@intersil.com.