CMOS power amplifiers (PAs) are prevalent in the Wi-Fi and Bluetooth markets, yet are only just emerging today for 2G and 3G mobile handset radios. In the 2G entry segments of the handset market, the price advantage of CMOS is especially important. However, the stringent performance requirements of 3G and 4G handsets allow no compromise, and simply matching the performance of the incumbent GaAs PAs isn’t likely to be enough to create a widespread shift to CMOS PAs. For increased market success, it is necessary for the CMOS PAs not only to deliver lower system cost, but also to outperform the GaAs PAs on important system metrics.
A close examination of the 3G mobile handset indicates that significant system improvement may be achieved by using a PA that simplifies the transmitter (TX) filtering requirements. A CMOS 3G PA with integrated TX filtering can outperform a GaAs PA, and it eliminates the TX SAW filter and other components from the handset bill of materials (BOM), reducing board space requirements and overall costs.
3G Mobile Handset TX Filtering
Figure 1 shows a block diagram of a single W-CDMA band of a 3G mobile handset. Signal is transmitted and received by the transceiver and baseband (XCVR / BB) and separated by the duplexer. Since the 3G transmitter operates at the same time as the 3G receiver, the PA amplifies unwanted transceiver noise that degrades the received signal. This effect, known as “receiver desensitization” or RX desense, is a well-known challenge for the mobile handset RF designer. The RX desense occurs not only for the 3G radio, but also for the GPS and Bluetooth/Wi-Fi/ISM receivers when they are operated simultaneously with the 3G transmitter.
To help reduce noise, handset designers typically use a TX SAW to filter the transmitted signal before it is amplified by the 3G PA. Though the TX SAW filter may not be recommended by the chipset vendor or used on the chipset reference design, many handsets require use of the TX SAW filter to ensure requirements are met with margin for high volume production. The TX SAW typically adds 2 dB of insertion loss which causes the XCVR / BB to operate at a higher output power, adding a few mA of extra current to the system budget and can degrade the system linearity margin. Furthermore, the cost and footprint of the filter can be a significant portion of the RF BOM, especially when replicated over multiple 3G handset frequency bands.
One approach to save handset BOM and improve system performance is to choose a device that integrates TX filtering into the 3G PA, such as the JAV5101 CMOS 3G PA from Javelin Semiconductor. Based on a Bandpass PA™ architecture, the JAV5101 integrates the PA, control, input and output matching, bias, and regulator functions directly with TX filtering into a monolithic CMOS 3G PA die. Other approaches simply add the TX filter into a PA module rather than directly integrating the function into the CMOS PA die.
Figure 1. Block diagram of 3G mobile handset showing integrated TX filtering in the CMOS 3G PA.
Figure 2 shows the effects of the TX SAW removal on a 3G mobile handset PCB. In this example, the CMOS 3G PA is pin-compatible with the previously used GaAs PA which simplifies the implementation. By moving to the CMOS 3G PA, designers could eliminate a TX SAW filter plus four additional components that were needed for matching around the TX SAW. In this sample implementation, only the TX SAW filter, 3G PA, and matching components were changed.
Figure 2 TX SAW filter removal on a 3G mobile handset using a CMOS 3G PA.
It is good to see a potentially market viable linear CMOS PA come to market. Blacksand announced sampling their 3G CMOS PAs a couple of months ago, so more than player is ready to enter the sand box.
The Javelin proposal is an interesting approach to differentiate from the current GaAs PA solutions, but it is a circuit solution and not a CMOS versus GaAs technology solution. Integrating the bandpass response into the PA design is what saves the BOM cost. The GaAs PA folks could probably find an equivalent solution if the market pushes them into doing it.
However, there are questions about the approach that need answering.
1. How well controlled is the bandpass response from part-to-part? Is the 15dB Rx rejection a guaranteed minimum rejection.
2. What is the worst case gain variation across the band. Small shifts in the response could make a big change in gain across the band. Perhaps this is ok to some extent since 2dB of filter loss has been eliminated.
There are market questions too...
Some GaAs PAs are multi-band such that the same PA is used across several near bands (800MHz-900MHz and 1.8GHz-2GHz). Such PAs offer cost savings by using one PA where there would normally be separate PAs for each band. So, how does the Javelin bandpass approach address this?
There is also a trend in RFIC research towards designing transmitters with very low Rxband noise (below -160dBm/Hz) to eliminate the Tx SAW filter.
Despite these questions, I hope linear CMOS PAs do make their way into the market. Once they get beyond having their foot in the door then the game will be on between CMOS and GaAs PAs.
We've been hearing about CMOS PA's for years and though there has been impressive improvements and design wins, still no real threat to GaAs PA's. GaAs ASP's are below 50 cents/module and the major players all own their own fabs. CMOS has its advantage with very low cost handsets, mainly 2G/2.5G, in developing countries and with second and third-tier OEM's in places like India and China.
I remember the hype about SiGe PA's a decade ago and the linearity, PAE and reliability still could not match GaAs. The only major SiGe player just got bought by Skyworks and maybe had major design win slots with WLAN and BT.
BTW, RFAxis is mostly hype and has a bulk of their design wins with ZigBee module companies and small ODM's.
Join our online Radio Show on Friday 11th July starting at 2:00pm Eastern, when EETimes editor of all things fun and interesting, Max Maxfield, and embedded systems expert, Jack Ganssle, will debate as to just what is, and is not, and embedded system.