Digital Predistortion Linearizes Wireless Power Amplifiers

Reliable cellular service depends upon clean, consistent transmission from base stations under widely and rapidly changing conditions. The base station’s RF power amplifiers (PAs) are key to guaranteeing this reliability. Spectral efficiency has always been important in mobile communications, but modern second- and third-generation digital systems now demand that linearity and efficiency of PAs are crucial performance requirements. Such amplifiers are found in cellular base stations that support any of the code division, multiple access (CDMA) family of wireless standards, such as CDMA2000, the Third Generation Partnership Project (3GPP), wideband CDMA(W-CDMA), etc., as well as improvements to existing standards, such as Enhanced Data Rates for GSM Evolution (EDGE). In many of these applications, due to the use of quadrature modulation and multiple carriers, the signal’s power varies or fluctuates significantly over time, as compared with analog frequency modulation (FM) or Gaussian minimum shift keying (GMSK) modulation as used in global system for mobile communications (GSM). Although all of the above-mentioned systems have good spectral efficiency, the signal’s varying envelope generates intermodulation distortion (IMD) when amplified. Most of the IMD power appears as interference between adjacent channels, which requires the use of highly linear amplifiers. However, PAs typically operate at their best linearity and efficiency over only a narrow range of power. It is, therefore, difficult to find a given operating point or “back-off” for a given amplifier and modulation scheme that is both efficient and September 2005 55 free of distortion products, e.g., linear operation. Also, backing-off the power into an amplifier lowers power efficiency and increases heat dissipation, both of which are highly undesirable effects in the confined space of a base-station shelter. To further complicate things, note that a CDMA-type base station’s signal will change depending upon the instantaneous number of users in the cell. In practice, linearity can be improved at the expense of efficiency or through the use of linearization techniques. In this article, we will present a detailed look at one such approach known as digital predistortion (DP). This will be implemented, for our purposes, using electronic design automation (EDA) sofware connected to standard laboratory test equipment. Let us first review some of the available linearization techniques and compare them.  

Comparing Linearization Techniques 

The feedforward linearization technique is well known for providing good linearity but with poor efficiency and an undesirable reliance on complex, expensive, and potentially difficult-to-maintain analog hardware. Feedforward works by sampling the amplifier’s output and reducing it to the same level as the input signal, then subtracting it from the input, leaving only the distortion generated by the amplifier. The distortion signal is increased with a separate amplifier so that it has the same level as the main output and is then subtracted from the original amplifier output signal. The result is a linearly amplified version of the input signal. Adding adaptation can further enhance the effectiveness of feedforward. Adaptation may use pilot tones to control the signal cancellation circuit so as to minimize the power of the error signal in the frequency band occupied by the distortion. Other, more complex types of adaptation are based on continuously computing the gradient of the threedimensional power “surface” for both the signal and error cancellation signals, or by measuring the adjacent- channel power ratio (ACPR) and attempting to minimize it [1]. All of these techniques are complex and depend on hardware that is prone to changing characteristics over time and temperature, making practical implementation and calibration difficult and expensive. In addition to feedforward, with or without adaptation, several other linearization methods have been used. These include analog predistortion (PD), linear amplification using nonlinear components (LINC), and Cartesian feedback. As with feedforward and its variants, these techniques involve a considerable amount of added analog hardware or require the use of nonlinear components whose characteristics are difficult to control to the degree of precision necessary to achieve the desired improvements.  

Implementing the Adaptive DP System 

The intent of DP is to linearize the nonlinear response of a PA over an operating region. DP employs DSP techniques to predistort a baseband signal prior to modulation, up-conversion, and amplification by the PA. As a result, the cascade of the DP response and the PAresponse produces the desired linear response. Figure 2 shows the simplified block diagram. The gain, G, of the PA is modeled as a function of the magnitude of the PA input signal, Vp. The function G is, in this case, memoryless and nonlinear in amplitude and phase. The use of a memoryless model that is only dependent upon input signal magnitude is a simplification of the actual response of a typical PA. Other variables will impact the PA response, including, most notably, frequency and instantaneous operating temperature.   

A simple adaptive algorithm may be employed to update the LUT coefficients until an optimum setting is achieved and used to predistort the input to the amplifier.  

Conclusion  

When compared to other linearization methods, adaptive DP provides sufficient linearization with less complex RF hardware by depending primarily upon DSP rather than analog manipulation. The adaptive DP system may be implemented in EDA software along with interconnected test equipment (arbitrary RF signal source and vector signal analyzer). Through the use of a training signal, a simple adaptive algorithm may be employed to update the LUT coefficients until an optimum setting is achieved and used to predistort the input to the amplifier. This “connected solution” approach can provide key information to design engineers for optimizing the DSP architecture of a PA. For future work, other issues will be taken into account, such as temperature and electrical memory effects