Moving wireless local area networks (WLANs) to the 5-GHz band is very appealing because of the leap in data rates from 11 or 22 Mb/s to 54 Mb/s. But, making the leap to 5-GHz also means the use of orthogonal frequency division multiplexing (OFDM), which comes with its share of challenges.
OFDM is frequently referred to as multi-carrier modulation because it transmits signals over multiple subcarriers simultaneously, enabling it to boost WLAN data rates up to 54 Mb/s.1 It is based on the Fast Fourier transform (FFT) concept, which allows the multiple subcarriers to overlap yet maintain their integrity. 2 Because these subcarriers are sent at lower data rates, OFDM systems perform robustly in severe multi-path environments. As a result, this modulation approach is being widely adopted in 5-GHz WLAN implementations.
High throughput data rates are achieved in OFDM due to precise carrier spacing and exact amplitude and phase settings for each individual carrier constellation. This is accomplished using computational modulation schemes rather than traditional analog modulation. OFDM's resistance to multi-path interference results from the increased symbol duration for each individual carrier (as compared to other modulation schemes with the same data throughput) and from the use of a cyclic prefix (guard interval) preceding each symbol.
When considering the overall engineering of an OFDM modem, the design of the digital baseband processor (BBP) has many benefits, and few, if any, drawbacks. However, from the perspective of the RF section, traditional OFDM modem designs face a number of key design issues that impact system cost and performance which must be overcome in order to deliver an industry-leading WLAN modem. These include power consumption, linearity, image rejection, phase distortion and phase noise. If these issues are not resolved, the resulting 5-GHz WLAN modem will deliver lower than anticipated data rates and compromised range.
Below we will explore each of these topics and talk about potential solutions to these problems. Let's start with peak-to-average power ratios.
One of the most difficult engineering concerns in the RF portion of traditional OFDM modems is handling very large peak-to-average power ratios (PAPRs). A peak in the signal power will occur when all, or most, of the sub-carriers align themselves in phase. In general, this will occur once every symbol period (see Figure 1). The value of the PAPR is directly proportional to the number of carriers, and is given by:
PAPR (dB) = 10 log (N)
where N is the number of carriers.
For example, for the 802.11a OFDM standard, a PAPR of 17 dB results if the phases of all 52 carriers line up during a symbol period. In order to accommodate this peak in power, the radio's RF power amplifier (PA) must provide gain without compression for every peak power level. In other words, the PA will provide less RF power between peaks, by an amount given by the PAPR.
Not only does this imply that the power amplifier must be oversized in terms of its average power requirement, it also means that the efficiency of the PA will suffer dramatically, since its DC power consumption is determined by the peak power level. Since the DC power consumption of the PA represents a significant portion of the total DC power for a radio, traditionally designed OFDM modems are not power efficient.
To put this in perspective, the maximum power efficiency of a Class B PA is 78.5%. However, when accounting for a signal having a PAPR of 10 dB, this efficiency drops to 7.85 %. Hence, to achieve a power level of 100mW will require a DC power consumption of 1.3 W. This very high DC power consumption limits the attractiveness of OFDM modulation for portable applications, including 5-GHz WLAN enabled devices such as PC laptops and personal digital assistants (PDAs).
More PAPR problems
In addition to creating design problems for the PA, the high peak-to-average power ratio of the OFDM modulation scheme also requires highly linear upconverters. This means that the upconverter must have a high-level compression point, which also results in high DC power consumption.
In addition to a drain on battery life, high power consumption has other effects on the cost of the design. For instance, the power supply unit becomes more expensive. And, in order to dissipate high peak transmit powers of 3.5 W or higher, heat dissipation techniques (such as a large heat sink and special board designs) will increase overall system cost and form factor. Finally, semiconductor manufacturers will have to use more expensive packaging designed to sustain high power consumption, thereby increasing the component cost.
To meet consumer expectations for user autonomy (long battery life in portable applications), OEM manufacturers are specifying low power consumption as part of their system requirements. But, merely specifying a change does not solve the problem. The technical challenge, then, is before the OFDM modem designers who must achieve acceptable levels of power consumption in their designs. Until then, the penetration of WLANs using OFDM will likely be limited to high-end, low-volume products.
OFDM modulation is also very sensitive to the inter-modulation distortion (IMD) that results from mild non-linearity in the RF. Because the subcarriers are equally spaced, the third-order intermodulation products (IP3) will appear exactly on top of another carrier Figure 2). These intermodulation products will contribute to a noise-like cloud surrounding each constellation point.
For higher-level modulations such as a 64 quadrature amplitude modulation (QAM) scheme, these constellation clouds can contribute to an increase in bit errors for each carrier. Even a modest increase in bit error rate (BER) for each carrier can result in a dramatic increase in the cumulative error rate over a packet. Thus, in an OFDM modem design, linearity must be carefully controlled.
For the RF components in the transmitter portion of a transceiver, the linearity requirement must be met at reasonably high power levels. With the associated requirement for high efficiency, this places extreme design challenges upon the architectural realization.
For the RF components in the receiver, the power levels are much lower but the linearity requirement is even more challenging. This arises from the possible presence of adjacent channel interference from other, independent but closely located, WLAN networks.
Resulting from the possibility of the interfering transmitter location being nearer than the desired transmitter, as well as the possibility of channel shadowing on the desired signal, the adjacent channel blocker is often received with a significantly stronger signal power level than the desired channel. The linearity requirement for the receiver components must include this anticipated level of blocker power above the desired channel. Two direct consequences of this are difficulty in achieving acceptable noise figure performance for the overall receiver, and difficulty in achieving sufficient dynamic range for the ADCs.
The latest trend in receiver design is to reduce or eliminate the intermediate frequency (IF) stage in order to eliminate costly additional filtering circuitry. While this streamlines the receiver, it makes it more difficult to control image rejection, because designers cannot make use of external RF and IF filtering. At best, insufficient image rejection in an OFDM modem reduces the carrier-to-noise ratio available for the demodulator. At worst, it allows another signal to interfere with the OFDM signal, resulting in a catastrophic increase in error rate.
Recent architectures are addressing the image rejection requirements using double-quadrature down-converters. These designs are based on first balancing, and then canceling the residual image-frequency errors in two traditional, quadrature down-converters. A difficulty being faced by such designs is achieving sufficient amplitude and phase matching of the mixer devices at the relatively high 5-GHz.
OFDM provides a selection of modulation constellations for each carrier. For example, the 802.11a standard provides for individual carrier modulation up to 64 quadrature amplitude modulation (QAM). Although signal impairments due to multi-path are eliminated through the use of a cyclic prefix guard interval (during which no demodulation is performed in the receiver), the closeness of the constellation points can result in significant errors due to dispersion. This dispersion can be caused by motion of the radio units or from motion of any other object in the channel. It can also be caused by phase variation with frequency in the radio antennas, filters, and other components.
If not accounted for in the design, phase distortion can prevent the use of higher-level modulation schemes (such as 64 QAM), thereby reducing the maximum achievable data throughput rate.
Channel equalization circuits following the analog-to-digital converters (ADCs) preceding the demodulation circuits can provide compensation for phase distortion. An ongoing issue is whether the equalizers must be adaptive, or whether static equalization is sufficient. Adaptive equalization will compensate for slowly varying channels, whereas static equalization can only account for phase distortion introduced by the analog components. However, static equalizers are inherently simpler and do not have stability issues. If most of the phase distortion is a result of narrow band filters and mixers, then static equalization is all that is required.
Phase Noise Issues
By definition, multi-carrier modulation requires very close proximity of the adjacent individual carriers. This is possible due to the relatively low data rate for each carrier. However, upon down-conversion in the receiver, any phase noise associated with the local oscillator (LO) synthesizer will be superimposed onto the low data rate modulation (see Figure 3).
As a result, maintaining sufficiently low phase noise levels close to the LO frequency becomes extremely important in achieving low bit rate performance in an OFDM modem. As in the case of IMD, a modest increase in the BER for each carrier can result in a dramatic increase in the cumulative error rate over a packet. Thus, phase noise must be carefully contained in the OFDM modem system.
In any LO synthesizer, the close-in phase noise is composed of the frequency-multiplied crystal reference phase noise, phase noise due to the synthesizer circuits (charge pump, phase/frequency comparator, digital dividers), and the voltage controlled oscillator (VCO). Of these, the phase noise of an on-chip VCO will dominate, due to the low Q of its spiral inductors. Techniques to improve the Q of the on-chip spirals continue to be reported and include the use of copper as the top metal layer, and trench isolation in the substrate.
Addressing the RF Issues
While typical end users of 5-GHz WLANs may never have heard of linearity, image rejection, phase distortion and phase noise, they will keenly feel the effects with lower than expected data throughputs and range of their 5-GHz WLAN devices.
Typically linearity, image rejection, phase distortion and phase noise issues for OFDM are addressed through the addition of costly and power-hungry external components including surface acoustic wave (SAW) filters and crystal oscillators that contribute to a higher system bill of materials (BOM) and increased power consumption.
Modern radio designs utilize innovative architectures and on-chip filters to reduce BOM costs and power dissipation; however, the signal corruptions previously discussed can become problematic.
Solutions for improved performance in "radios-on-a-chip" include process selection, innovative architectures, the use of on-chip passive components where size and performance constraints permit, as well as taking advantage of quirks in the air interface definition.
Although there are some advantages to using a GaAs (i.e. linearity) or a bipolar process (reduced noise) for RF signal processing, the all CMOS radio is the preferred choice for cost, especially when designing radios with significant digital subsections. Innovative architectures use complex domain RF and digital signal processing techniques to mitigate the effects of analog imperfections. Air interface quirks allow one to trade a difficult specifications with a more relaxed one by migrating problem areas associated with difficult requirements to "don't care" areas of the frequency and temporal characteristics of the air interface. Holes in the temporal characteristics of the air interface also allow one to initiate calibration sequences transparent to the end user to improve performance during transmit and receive bursts.
For designers of 5-GHz WLAN semiconductor solutions, addressing these RF issues is a major engineering challenge in the design of production-ready 5-GHz WLAN devices.
As a result, OFDM IC solution providers have to resolve a series of technical issues and trade-offs as well as some marketing issues. One consequence of these tradeoffs could be a lower bit rate than expected for a given transmit power and cost. Or, it could result in more external components, which would cause manufacturing issues and increase cost for a given data rate and range.
On the marketing side, dictated power specifications could lead to implementation issues that constrain OEM manufacturers to an expensive or unpractical solution, resulting in a worst-case scenario of no commitments to the technology or limited production.
These issues directly impact high-bandwidth WLAN designs. Engineers, product designers, and product managers need a realistic availability estimate for next generation WLAN technology (802.11a or HiperLAN2). By definition, when compared to 802.11b, next generation technology needs to offer a higher bandwidth and superior interference immunization performance at a lower cost, all at power consumption levels low enough to make the technology viable for laptops, access points, and other mobile applications.
Jim Wight is the principal architect at IceFyre Semiconductor. He has over 28 years of experience in research and development in the wireless and satellite industry. Wight received is a doctorate in electronics and has held a faculty position at Carleton University for 25 years. He can be reached at email@example.com
- Grier, Jim. "Enabling Fast Wireless Networks with OFDM" Communication Systems Design Online, 02/01/01.
- Buckley, Sean. "Is OFDM Ready for Prime Time?" Telecommunications, February 2001.