|Wireless Ethernet - The Next Generation|
|Originally published March, 2001|
|¿ 2001, 2005 Carlo Kopp|
No sooner than the ink dried on the baseline IEEE 802.11 Wireless Ethernet standard, we began to see a plethora of proprietary bit rate extending enhancements to the basic standard, pushing its achievable bit rate out further and further than the modest 1-2 Megabits/s it started at. In this month's feature Carlo Kopp explores one of the latest developments in the IEEE 802.11 saga - the very high speed variants being developed around the COFDM modulation scheme.
At this time wireless Ethernet is a rapidly growing market, as it has proven to be an excellent connectivity tool for laptops, notebooks and various other bits of portable/wearable technogadgetry. Given the boom in mobile phones, it is inevitable that wireless connectivity will become an increasingly popular part of the portable and mobile computing game.
After an initial period during which proprietary products dominated the market, the 802.11 standard was adopted and has become the baseline for this ilk of mobile connectivity (an excellent web primer by Jean Tourrilhes of HP is posted at http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/). While the 900 MHz band achieved some success, most current products are firmly centred in the 2.45 GHz ISM (Industrial-Scientific-Medical) band, for which there is no licensing requirement if the transmitter is appropriately limited in power output (1 Watt, although the definition is really related to the antenna Equivalent Isotropic Radiated Power or EIRP) and band coverage.
The baseline 802.11 offering comes in two forms. The first is the Direct Sequence or Direct Spreading (DS) scheme at 1 or 2 Mbits/s, in which each bit in the binary stream to be transmitted in encoded into an 11-bit Barker code, which is then phase modulated upon a carrier wave. With four mutually orthogonal forms of the 11-bit code, 802.11-DS can have four channels sharing the same bandwidth, in a manner not unlike CDMA telephony.
The second form is the Frequency Hopping (FH) scheme at 1 Mbits/s, in which the band is chopped up into channels, and the carrier is pseudo-randomly hopped between channels, thus spreading the energy of the signal evenly across the band.
The aim of using spread spectrum techniques in 802.11 was twofold, both to allow low power operation with a useful bit rate, and to allow operation in a free-for-all unlicensed portion of the spectrum. In general, wideband spread spectrum systems have significantly better resilience to narrowband interfering signals than conventional narrowband modulations.
Inevitably, users wanted more channel throughput, and thus the race was on to find ways of extending the standard to get higher bit rates, or put simply, squeeze more blood out of the stone. After much haggling between vendors on the committee, an agreement was reached which led to the adoption of the IEEE 802.11-b or 802.11 HR standard, which is an extension of the direct spreading modulation scheme to provide either 5.5 Mbits/s or 11 Mbits/s by imposing an additional modulation on the signal, with some inevitable loss in signal robustness as Shannon predicts. There can be no doubt that 11 Mbits/s in a wireless Ethernet is a useful throughput, and the standard has proven to be very popular in the market, especially for environments needing only small footprints.
Not surprising navigating in the 802.11 marketplace is not a game for the faint-hearted or technologically challenged, since we now have three different derivatives bundled in the same standard, of which only two have a one-way backward compatibility relationship.
The use of spread spectrum techniques for 802.11 was a radical departure from the established wireless communications game, but given the very low spreading ratios resulting from short pseudo-random coding, it has not delivered the level of robustness many users may have expected, after hearing tales of near jam-proof military spread spectrum communications. This was an inevitability, insofar as the robustness of any spread spectrum transmission scheme depends critically upon the spreading ratio between the baseband signal and the modulation on the carrier. In practice, the 10 dB or so offered by 802.11 is not much of a defence against the neighbour's leaky microwave, or ISM-barely-compliant spread spectrum telephone handset.
Interference rejection proved to be one weakness of the established 802.11 protocols, but it is not the only one, as was soon discovered once the product arrived in the market en-masse. Multipath, the curse of analogue television, FM radio in cars and mobile telephony, proved to be just as much of a curse for wireless LANs.
To those who have a strong background in radio-frequency communications, multipath is an inevitable and well understood fact of life in terrestrial transmission. Radio signals propagating along the surface will bounce off obstructions such a buildings or hills, and what a receiver sees at its antenna is a jumble of variously time delayed and variously weakened copies of the signal which left the transmitting antenna. In engineering speak, a vector sum of signals.
What this looks like at the receiver end depends very much on the geometry and relative strengths of the signal received directly, and its delayed copies. The worst case situation is where the delayed signal is just as strong, but delayed by half a carrier cycle, or technically out-of-phase. When this happens, the two signals cancel each other out and the signal fades to nothing, causing a dropout. The term fading, used to describe the deleterious effects of multipath, is aptly chosen.
How well a receiver copes with fading depends on the technology in use, and the severity of the multipath effects. Direct spreading spread spectrum receivers have traditionally performed quite well under modestly severe fading conditions, by using rake receivers, in effect a battery of parallel receivers each of which locks on to one of the multipath delayed copies of the carrier. But even this technique fails if the multipath fading is serious enough, since mutually cancelling or almost cancelling carriers leave very little indeed for a rake receiver to lock on to.
Other alternatives do exist, some of which are very effective. Military GPS receivers use smart adaptive beamforming and nulling antennas, which cleverly point beams at satellites, and antenna nulls at interfering sources, or jammers. Needless to say it doesn't take a genius engineer to figure out that a very nice way to jam GPS is to simply retransmit a time delayed copy of the GPS signal to artificially introduce a serious fading problem. Adaptive beamforming and nulling can pretty much nullify such mischief. The snag? The antenna and its supporting beamforming electronics can cost tens of thousands of dollars, weight several kilograms, and are definitely not a near term candidate for an IEEE 802.11 laptop user, no matter how profligate he or she may be with the departmental IT budget!
Fading has been a thorn in the side of every radio engineer throughout the history of broadcast communications, and inevitably became a subject for much practical and academic research. Some urgency arose with the growth in mobile communications, since there was no simple way of dealing with the problem when using a cheap whip antenna, or other simple omni-directional antenna type.
One of the interesting facts which follows from some thought into the subject is that fading is frequency dependent, as well as spatially dependent. If the propagation path of the interfering copy of the radio wave has some fixed length, retuning the carrier wave will see the fading periodically increase and decrease, as the two carrier waves move in and out of phase with one another. So just as one can drive in and out of an area of severe fading when talking on one's mobile, one could achieve a similar effect by having the transmitters and receivers retune themselves to suppress fading.
Attempting such a scheme is however not a very practical idea, for good engineering but also regulatory reasons. So the big question does remain of how to exploit this aspect of multipath propagation physics to advantage. A very elegant solution does indeed exist, and we can expect to see much more of it in the foreseeable future - COFDM.
Coherent Orthogonal Frequency Division Multiplexing (COFDM)
COFDM is the basis of the new European (and soon Australian) High Definition TV broadcast standard, the European Digital Audio Broadcasting (DAB) standard, and the new 802.11-a 5 GHz band wireless LAN standard. The in the direction of COFDM, particularly its early adoption in Europe, has much to do with Europe's endemic multipath problems resulting from very high population density and a lot of very hilly terrain. What is an annoyance in urban Australia and the US, and largely a non-issue in most rural areas, is a do-or-die problem for the Europeans.
To best appreciate how COFDM achieves good rejection of multipath fading, it is useful to do a little experiment. Let us contrive a radio link which allows us to retune both ends concurrently, and then let us find a place in the neighbourhood known for horrendous fading problems. What next? We pick a slice of the radio spectrum we intend to work in, and tune the carrier across our frequency range of interest. What we will find is that for a reasonably wide slice of the spectrum, typical for a wideband data signal, we are likely to get some particular patches where multipath bites a lot, and many others where the phase differences cause no pain at all.
Can we think about this effect differently? Let's assume our carrier is modulated in amplitude with a simple high speed binary data stream. It will produce a spectrum with sidebands, and as a result of the multipath fading being frequency dependent, parts of these sidebands will be chopped out or damaged. What happens when we view the recovered data stream? We will see nasty shape distortion, causing some bits to run into their neighbours, not unlike the ugliness we encounter on a cable which distorts signals by suppressing higher frequency components. Different physics, but related consequences.
How can we exploit this behaviour to advantage? If we transmit the data modulation at a much slower bit rate, the distortion will become increasingly less significant. Why? The sidebands contract and less of them fall into the part of the spectrum where the multipath causes them to fade out.
This inevitably leads to the basic idea behind COFDM. Rather than transmitting a very fast digital modulation with redundant data bits on a single carrier, which is vulnerable to fading because the modulation sidebands become damaged, we transmit a very large number of redundant subcarriers each with a very slow modulation. As a result, if fading knocks out one or more subcarriers, we can still recover the data safely, which is not necessarily true of the single carrier/fast modulation scheme. In concept, this is not unlike the comparison between a redundant parallel bus with many wires, against a serial bus with one wire. Parallel busses have always been easier to build for a given throughput, because they can be clocked with data N-times slower for any given bus throughput.
This analogy isn't quite as silly as it may seem, on closer examination, because the problems which kill bus performance arise also from pulse distortion in the transmission medium, in this case a cable. The physics via which the distortion arises may be quite different, but they produce much the same effect at a system level.
A trivial multiple subcarrier FDM system could be simply built by stacking a large number of low speed data modems in parallel, each tuned to its subcarrier frequency, and then suitably multiplexing and demultiplexing the digital data stream going into the system. However, if we want several hundred subcarriers to properly exploit the available benefit, we end up with something which is prohibitively expensive to build and much too bulky for ordinary users.
Modern COFDM systems are affordable and compact, as they exploit some clever idiosyncrasies in the mathematics of the problem.
The origins of COFDM go back to 1971, when a pair of very clever research engineers, Paul Ebert and S. Weinstein, both working for Bell Labs in New Jersey, discovered a curious relationship between the Fourier transform, beloved by engineers, and the behaviour of coherent FDM systems using large numbers of subcarriers. The FDM signal, made up of a large number of coherent (ie having a fixed frequency relationship) subcarriers, could be shown to be the Fourier transform of the digital data stream, and that the behaviour of the stack of coherent demodulators could be described by the inverse Fourier transform. Since Fourier transforms can be crunched on computers, Ebert and Weinstein suggested a new modulator design for this purpose, a completely digital modem built around a special purpose computer performing the fast Fourier transform (FFT) algorithm (the author is indebted to Dr Chintha Tellambura of Monash Uni for providing a copy of this 1971 paper).
Given the speed and cost of computers during that period, the Ebert-Weinstein modem had to wait three decades before it could be produced economically.
Before practical COFDM systems could be produced, other theoretical refinements had to be developed. In the Ebert-Weinstein model, the data could not be transmitted continuously, since the sidebands of the subcarriers interfered with one another. They dealt with this problem by leaving gaps in the transmission, which was inefficient. The solution to the problem was found in 1980 by NEC research scientist Botaro Hirosaki, who discovered that making the subcarriers orthogonal mathematically, allowed transmission without interference between the sidebands of the subcarriers. The condition for orthogonality was found to be very simple - the spacing of the subcarrier frequencies had to be the inverse of the data symbol period (ie bit cell duration on each subcarrier). If this condition was satisfied, then the system could transmit data on every subcarrier with no intervening gaps.
Thus was born COFDM. Before practical commercial systems could be built, advances had to come in FFT processing. By the 1990s, FFT processor chips reached the cost and performance level where mass production COFDM modems became feasible.
Modern COFDM transceivers rely fundamentally on the availability of high speed signal processing chips, FFT processor chips and analogue/digital and digital/analogue converter chips.
A typical design will see the digital data stream converted into a parallel set of N bits for N subcarriers. It is then used to produce, in software, the phase keyed modulation values for each subcarrier (for readers with an engineering background, this amounts to calculating the respective amplitudes of the real and complex components of each subcarrier, to get the phase angle required to encode the bit value). These are then fed into an FFT chip which performs an inverse FFT into the time domain. These samples are then fed into a digital/analogue converter to produce the modulation envelope for the signal. At the receiver end, the signal is digitised, and the samples fed into a FFT chip to perform the forward FFT transform into the frequency domain, to recover the subcarriers and their respective phase shifts. Once the phase values are established, the bits are recovered and the data stream can be produced.
Since fading may knock out some of the subcarriers, a block coding scheme with some redundancy will be used to recover the data free of errors. This is usually backed up with further data redundancy and error control measures in the data stream.
There can be little doubt that COFDM is the most complex modulation scheme yet to penetrate into the mass production techno-commodity market.
The 802.11-a COFDM Wireless Ethernet
The 5 GHz band 802.11-a standard will provide unprecedented speed for a wireless application - no less than 54 Mbits/s. This is almost five times the throughput of the current 802.11 HR standard, which tops out at 11 Mbits/s under optimal transmission conditions. Since COFDM is used, it is expected that the 802.11-a standard will deliver significantly better robustness in fading environments.
An 802.11-a link requires a modest 20 MHz bandwidth, and encodes 64-subcarriers with the Quadrature Amplitude Modulation envelope carrying the data, using a pair of 64-point FFT chips and a Viterbi encoding scheme. A typical 802.11-a COFDM modem chipset such as the Radiata R-M11a uses 10-bit resolution analogue/digital and digital/analogue conversion at 80 Megasamples/s.
Not to break the well established trend in this market, some manufacturers are already selling designs with proprietary enhancements. The Atheros AR5110 radio-on-a-chip is designed to operate in 802.11-a compliant 54 Mbits/s mode, but also in a proprietary turbo mode of up to 72 Mbits/s, out to 100 feet of distance.
What conclusions can we draw at this stage? The market is still in its infancy, but the throughput advantages of 802.11-a COFDM are so dramatic against the basic 802.11 and enhanced 802.11-a (HR) standards, that we can expect to see a mad scramble by OEMs, WLAN board manufacturers and mainstream computer manufacturers to incorporate the COFDM product at the earliest possible date. Since 802.11-a operates outside the established 2.45 GHz band, virtually all in service WLAN hardware will be effectively obsoleted. Very few 2.45 GHz antennas and cables will perform well at 5 GHz, so we are likely to see a lot of surplus 2.45 GHz equipment appearing on the market over the next 2-3 years.
To yet again paraphrase Larry Niven, it is another case of evolution in action.
|$Revision: 1.1 $|
|Last Updated: Sun Apr 24 11:22:45 GMT 2005|
|Artwork and text ¿ 2005 Carlo Kopp|