Is Cellular RF PA Integration Finally Here? Part 1

A couple of weeks ago, in RF Integration – The Final Frontier, I commented on the large number of separate devices needed to implement a multiband, multimode, multiprotocol RF section for a smartphone or tablet, and contrasted it to the fact that the entire digital baseband processing and apps processing sections had been reduced to one chip.

The biggest issue with the RF section now is the power amplifier (PA) and filtering. Looking back, the technologists who cooked up the digital cellular systems figured that the challenges of getting all the digital stuff done far outweighed the problems of putting a little 2-watt UHF power amplifier on a board. Remember: At the time the GSM standard was first created, a DSP capable of just the voice compression algorithm cost close to $100.

The creators of the standard were correct in predicting that the digital section would shrink in size, power, and cost while gaining speed and capabilities. But as new bands and modes crept in, the balance of complexity shifted to the lowly, not-real-exciting PA and associated filtering.

Analog FM cellular and even mostly-constant-envelope GSM are forgiving of PA issues. Linearity doesn’t matter much, and the big push for the first 10 years was to figure out how to get GaAs (gallium arsenide) into high-volume commercial production. It was a better technology choice because it offered greater efficiency, which delivered longer battery life.

In the meantime, CMOS PAs have emerged, trying to take advantage of the lower cost and higher availability of CMOS wafers, but their efficiency has not approached that of GaAs amplifiers, and they have so far mostly been used only in low-end phones (e.g., disposable phones). Several companies have done well making relatively simple PAs, and adding things like boost-buck regulators and filters to them in an effort to reduce package count.

Keep in mind what you learned a long time ago about power amplifiers: They can be efficient or linear, but not both. Amplifiers get very non-linear when they are driven to the maximum available output level, as the amplifier saturates and can’t faithfully reproduce the output waveform. This creates amplitude and phase distortion, which appear as newly created signals on nearby (or not-so-nearby) frequencies.

The consequences of a non-linear PA in a cellular handset are worse than in other services. Most cellular systems rely on each handset transmitter being well behaved and not interfering with other handset signals heading to the base station on the same channel (in the case of CDMA systems), not producing energy in the handset receive band (for all systems), and not creating spurious energy in other carriers (in the case of OFDM).

Newer cellular standards use more complex waveforms to convey more channels and/or more bits. These waveforms, examined in the time-domain, have occasional high peak power. If the amplifier is designed to deliver full output power on peaks without distortion, it is running very inefficiently at lower output power levels. Thus a more-linear amplifier wastes power most of the time.

Lots of techniques have emerged for linearizing PAs — taking a fundamentally not-very-linear PA and making it behave more like a linear gain block. Pre-distortion, polar modulators (open and closed-loop), envelope-tracking, Doherty amps, and Cartesian loops all have been tried with varying degrees of success, mostly in base stations, where more power and space are available. Moving these techniques to handset PAs has taken time.

Next, a deeper dive into some possibilities for integrated analog at RF: Is Cellular RF PA Integration Finally Here? Part 2.

Related posts:

5 comments on “Is Cellular RF PA Integration Finally Here? Part 1

  1. Netcrawl
    March 30, 2013

    @Scott very interesting thoughts! PAs are one of the most difficult RF components to be integrated on-chip for RF-SoC products, @Scott I agree with lots of players in that space but for me Qualcomm is the first one to take the deep and still getting better.  @Doug interesting topic! Thanks for bringing us a great article.  

  2. eafpres
    March 30, 2013

    Hi Doug–it seems there is still so much going on in spectrum allocation and use, as well as modulation and spectrum reuse schemes, that it makes it risky to invest a lot integrating a PA, as a new requirement may pop up and then the whole chipset is not applicable to the new requirement.  If a vendor has a chipset that works over a range of spectrum, but only parts of that are used in a given country, would they typically optimize the front end including the PAs for the spectrum they will actually use?

  3. Doug Grant
    April 1, 2013

    Yep, spectrum allocations, etc. make things complicated. As long as PA designers are stuck with fixed passives to do filtering and matching, it's hard. Wait for my next post to see some possible ways around that problem.

  4. Doug Grant
    April 1, 2013

    I recall seeing an article recently somewhere by Earl McCune (I think) that defined the requirements for PAs in terms of how close to saturation they can run for various combinations of signal envelope and desired output power level. I can't put my hands on it at the moment, but IIRC, it used a limited number of levels as you describe. It looked interesting, but I'm not sure it scales for a seemingly never-ending parade of signal waveforms while always meeting the spectral mask rqeuirements.

  5. Brad Albing
    April 2, 2013

Leave a Reply