As the next generation in wireless local area networking (WLAN) development, 802.11n aims to open up a new world of high-speed local networking. This new technology will have the bandwidth required for high-definition video, high-quality audio and data-intensive applications such as interactive gaming. It will likely drive a change in the way networks are used throughout homes and offices. For designers of the power amplifier (PA) in the 802.11n radio, it is more challenging than ever to define and achieve the levels of radio-frequency performance required to meet the 802.11n standard in increasingly small portable and mobile applications.
Since the development of the 802.11a protocol (1999), WLANs have deployed orthogonal frequency-division multiplexing (OFDM). In this technique, the usable bandwidth is precisely divided into a large number of smaller bandwidths, or "subcarriers." The high-speed information is then divided onto these multiple lower-speed signals, which are transmitted simultaneously on different frequencies in parallel. The resulting low-data-rate carriers are more tolerant of fading because of multiple reflections.
The challenge for the PA lies in the highly variable amplitude of RF signals that is the sum of the orthogonal subcarriers. Because each subcarrier has an essentially random amplitude and phase, if the vector sum of all of the subcarriers happens to add in phase, a large peak (in the time domain) occurs. To avoid excessive distortion, the PA must accommodate these peaks, which can range more than 10 dB higher than the average. In reality, some distortion is permitted on the highest, most infrequent peaks, so typically a 7-dB peak needs to be accommodated.
CMOS process technology
Traditionally, WLAN PAs have been designed using GaAs heterojunction bipolar transistor (HBT) technology. Recently, however, equivalent performance in the 2.5 -Hz band has been achieved using SiGe bipolar transistors in a BiCMOS silicon-based process technology. The advantages of using silicon technology are numerous and include higher levels of integration (allowing designers to build in advanced control circuitry) and lower cost. There has been much discussion about manufacturing CMOS PAs for 2.5 GHz, which would allow integration with transceivers. Although this technology may have potential, the results have been disappointing: low output power and efficiency on the order of 10 percent. Nonetheless, we can expect that CMOS PAs will be regularly used in certain low-power applications.
Many 802.11n WLAN designs using 2.5 GHz are in development, but the new standard also covers 5 GHz. Currently, all of the 5-GHz designs use GaAs PAs. However, SiGe PAs are also in development for this bandwidth, and we can expect to see them in the near future. The availability of silicon-based PAs for 5 GHz might provide a boost to 5-GHz 802.11n, which is relatively expensive to deploy. As the 2.5-GHz band becomes even more saturated, there will likely be a shift to 5 GHz, and silicon-based PAs could be an important enabler.
Key performance metrics of 802.11n
As a WLAN standard, 802.11n's primary focus is on improving throughput. To achieve this, it departs from the time-, frequency- and code-division schemes that came before it. This protocol aims to send simultaneous data on the same frequency. Normally, these signals would interfere with each other. However, 802.11n can take advantage of a concept called "spatial diversity," which separates the data streams by using multiple antennas and radios. This is referred to as multiple input, multiple output (MIMO). Spatial diversity uses these multiple paths (multipath was previously an impairment) to effectively increase throughput. Thus, it is ideal for cluttered indoor and urban environments.
The SE2546A is a dual-band, dual-stream front end module (FEM) designed for MIMO.
FEM designed for MIMO can shrink the footprint of the PA and necessary passive circuitry.
Click here for larger image
Because of the MIMO configuration often used in 802.11n, the major performance metrics for the PA are output power, efficiency and linearity (or error vector magnitude, which is closely linked to linearity). The 802.11n standard requires higher performance in these areas without allowing for increase in size or component count.
Typical 2.5-GHz 802.11g PAs can deliver about +19 dBm of output power. There is a 7-dB peak-to-average power ratio for OFDM, which requires a PA capable of about +26-dBm peak power. This poses the first design challenge: +26 dBm delivered into a 50-ohm load is equivalent to 13 volts peak-to-peak. Because WLAN PAs are typically operated from a 3.3-V supply, a large voltage transformation is required. This must be done using passive components, and space requirements force integration of these passives into the PA or the front-end module.
In PAs, efficiency is the measure of the average RF power out vs. the average dc power in. Despite its need to handle large peaks, the PA operates at a lower power (about 7 dB below peak) 90 percent of the time. Standard classes and topologies of PAs generally offer their best efficiency at high output power, and efficiency decreases as power is backed off.
Although there are exotic PA designs that claim to offer high efficiency over a range of powers, PA designers typically opt for Class AB, which offers good efficiency at reasonable cost.
PA efficiency has been a major target for designers of 802.11g systems. Now, for 802.11n, this challenge becomes even more difficult because of the use of multiple radios and antennas in MIMO configurations. Today's state-of-the-art PAs for 2.5-GHz WLANs operate with 20 percent efficiency, which means that delivering 100 mW of RF power requires 500 mW of total power.
The SE2593 FEM is optimized for 802.11n use and measures 5mm x 6mm x 1.0mm. RF circuitry for 802.11n needs to occupy a comparable footprint to its 802.11g predecessors.
Click here for larger image
An ideal linear PA produces only the original frequency from the input signal. In real-world implementations, PA nonlinearities introduce new frequencies, or intermodulation distortion (IMD), which generates out-of-band signals that interfere with adjacent users (referred to as spectral regrowth or spectral leakage). Nonlinearities also generate in-band products that can cause errors in the data stream. The amount of out-of-band signals is measured as adjacent channel power (ACP) leakage. The FCC strictly regulates the amount of ACP emitted from a Wi-Fi transmitter. To adhere to these guidelines, the PA must be linear.
Unwanted in-band signals are tracked using a parameter called error vector magnitude (EVM) that quantifies the modulation accuracy of a transmitter. It is the measure of the difference between the ideal performance and the actual performance, with each point in the EVM constellation representing a particular magnitude and phase of one symbol in the OFDM packet. In 802.11n, EVM requirements are stricter because the noise from each radio will add noise to the other(s) in the system, further degrading performance and EVM.
Design trade-offs
PA designers for 802.11n face a conundrum. Nonlinearity leads to more ACP leakage and poor EVM. Greater linearity requires a PA designed to operate at average power levels of 7 dB (peak-to-average) below its maximum output power, decreasing efficiency. Finding the optimal solution requires a careful assessment of trade-offs, as well as clever engineering and circuit design.
A PA's linearity is determined by factors such as the semiconductor technology used, amplifier design, use of earlier stages, predistortion of later stages, bias circuits, matching networks, and even the impedance seen at harmonic frequencies. Designers have proprietary techniques that allow improvements in linearity beyond "textbook" designs.
For instance, most designers use standard Class AB uncorrected PAs. However, some specialized techniques traditionally used in larger systems (such as predistortion) are being explored to improve PA performance. It is theoretically possible to make the PA more efficient at the expense of linearity and afterward improve linearity by means of predistortion circuitry. However, at the relatively low RF power levels used in Wi-Fi, the extra power required to implement predistortion is often greater than the power savings achieved in the PA. As geometries used in the baseband processor move to 65 nm and below, this balance may tip, making it quite possible that future wireless PAs will use predistortion techniques.
Size/integration
PAs for 802.11n also face significant size constraints. The expectation is that four power amplifiers (two for 2.4 GHz and two for 5 GHz) will fit in the same space occupied by the two PAs used in a non-MIMO, dual-band configuration. The best approach is to design the PAs as part of a highly integrated package, fitting more dice in the package and spacing them closer together.
Some of the latest FEMs for 802.11n integrate multiple PAs, LNAs, power detectors, filters, diplexers (frequency selective splitters that divide a signal into separate 2-GHz and 5-GHz paths), and switches for switching between receive paths and transmit paths. These flexible building blocks can be designed with connections that allow multiple modules to be cascaded or stacked for use in MIMO applications of various configurations. At these levels of integration, the footprint for the PA and its associated circuitry is comparable to current 802.11g solutions.
Compliance and coexistence
Every country has its own regulatory requirements for WLAN emissions. The United States and Japan, generally speaking, have the most stringent demands, so most WLAN solutions are designed to satisfy the requirements of those countries. Because an 802.11n WLAN device is likely transmitting twice as much power as a nearby 802.11g device (because there are now two transmit channels), the amount of ACP leakage has to be well managed in order to comply. If the PA emits too much power on an adjacent channel, it could interfere with other services operating in that channel, thus failing to comply with regulatory requirements. Such risks are best minimized by using linear PAs.
Even with the best-available linearity specs, there can still be some coexistence issues with other wireless technologies. 802.11n is designed to be compatible with previous 802.11 standards, so there are no coexistence issues there. Most cellular frequencies operate far enough away from the ones used by 802.11n to avoid any problems, and their standard frequency filtering mitigates any possible interference. However, when an 802.11n device and cellular radio operate within a single device (as in a cell phone with WLAN capability), additional filtering is required. Bluetooth and WiMax, for example, operate in the same bands as 802.11n, so avoiding interference issues will be more challenging.
The final 802.11n standard is not expected to be officially released until October, yet there is already significant activity in product development based on the draft standards. The good news is that new PA designs and FEMs are already in place to address issues of performance, compliance and footprint.
 Darcy Poulin (dp@sige.com) brings RF expertise to SiGe Semiconductor (Ottawa). He has a B.Sc. in engineering physics and a PhD in applied physics.
|
 Gord Rabjohn (gr@sige.com) specializes in GaAs and
power amplifiers at SiGe Semiconductor. Rabjohn holds B.A.Sc. and M.Eng. degrees.
|
