IC Substrate - Optimize the long-established Doherty power amplifier

The Doherty Power Amplifier (PA), invented nearly 100 years ago, is used to improve energy efficiency in a large number of radio transmitters, and there are many ways to make such a power amplifier. This article first outlines linearization and efficiency enhancement, and highlights related challenges and a few of the many solutions based on the background. Finally, a case study is used to illustrate an alternative design process, and in-depth discussion of the design and how to achieve the best compromise between performance and cost.

Linearization technology
The four main technical performance parameters in the transmit (Tx) radio frequency front end (RFFE) are efficiency, output power, linearity and bandwidth. The last three parameters usually depend on system requirements, such as communication standards. The first parameter (ie energy efficiency) is the distinguishing factor. If all other performance parameters are the same, the higher front-end efficiency is better.

The devices used in RFFE have non-linear characteristics and cannot be used directly as ideal modules. Through linearization technology, the linearity of Tx RFFE can be improved. This usually increases the original cost of the Tx RFFE, and what you get is improvements in efficiency, linearity, and output power. Many linearity improvement methods have been published, at least back to the feedforward 1 and feedback 2 patents. It can be considered that the date of application of nonlinear predistortion is similar to the date of invention of compression and expansion technology3. These programs can be classified according to how they work (see Figure 1 and Table 1)4. One of the distinguishing criteria of linearization technology is: whether the scheme predicts or extracts useless signals, and whether to perform correction before or after output. Classification is useful for understanding general characteristics and identifying the best application method.

Feedforward is an example of a post-measurement correction scheme, feedback is a measurement pre-correction scheme, and predistortion is a predictive pre-correction scheme. Predictive solutions rely on the generation of unwanted signals, which can be very troublesome for digital predistortion (DPD) in systems with wider frequency bands and lower power. On the other hand, predictive solutions do not require distortion, and may completely eliminate distortion.

What is missing from these examples is the entire category of linearization techniques that employ predictive post-correction. In the past 100 years, people have conducted in-depth research and records on this technology series. Outphasing 5, Envelope 6, and Doherty 7 transmitters and hybrid transmitters introduced by Choi 8, Andersson 9 and Chung 10 are examples of these technologies, but these technologies are mainly used to improve efficiency rather than as linearization technologies. Market Development. The purest form of the envelope and out-of-phase schemes uses amplification and path summation, respectively, to construct their signals from nonlinear components that are efficiently generated. The Doherty amplifier contains a reference path called the "main path" or "carrier" and an efficiency path called the "peak path" or "secondary path". A more comprehensive mathematical analysis of the Doherty design is beyond the scope of this article and is available in many documents. For detailed information, readers can refer specifically to Cripps article 11.

DOHERTY implementation
It can be considered that the most common and usually the fastest starting point for Doherty amplifier design is the "zeroth embodiment" (see Figure 2), including:

* Fixed RF input to the final power splitter.

* The main amplifier and auxiliary amplifier are biased differently (for example, using class AB and class C).

* The Doherty synthesizer is formed by a quarter-wavelength transmission line.

* In most applications, this architecture will not provide sufficient power gain (at least not from a single final stage), and additional gain stages are cascaded in front of the power splitter. The disadvantages of this most common implementation include:

* After the design is frozen, there is no way to compensate the gain and phase in any domain.

* Due to the bias stage, there is a trade-off between efficiency and output power. It is actually C-level bias (an open-loop analog circuit) to accomplish this task.

* The efficiency improvement is limited to a single level. The situation of multi-stage cascade will limit the performance improvement, especially the gain will be reduced at higher frequencies.
From another perspective, the Doherty engine is an open-loop solution with several important functional mechanisms derived from the bias point of the transistor. Once other variables (such as phase offset, splitter design, etc.) are defined, only one or two operating points on which a variety of key adjustments are dependent are provided.

challenge

One of the ways Doherty improves efficiency is load modulation. The driving engine behind this modulation is the difference between the output current from two or more amplifiers into the synthesizer. Since the engine can only approximate Doherty operation, the challenge for designers is to make the engine approach this operation in the best possible way, but still have an appropriate price/performance ratio. Potential obstacles to Doherty performance include: 1) The amplitude and phase matching of the signal entering the merge node, especially the overfrequency (see Figure 3a). Deviation from the ideal value will reduce efficiency and output power. The latter may be more destructive, because the devices are intentionally not isolated, and the increase in efficiency depends on the interaction achieved through the synthesizer. 2) Ideally, the auxiliary path of the Doherty engine exhibits a polyline or hockey stick characteristic (see Figure 3b). Failure to reach the ideal value is often the main reason for not achieving the well-known efficiency saddle point. Since this characteristic tends to change from an ideal value to a linear response, the behavior of the Doherty amplifier will gradually become similar to that of a quadrature balanced amplifier (although with an unisolated synthesizer), especially its efficiency performance. 3) The commonly used "differential bias" of the main amplifier and auxiliary amplifier in Class AB and Class C will force the output power and efficiency of the two amplifiers to decrease (see Figure 3c). As explained by Cripps 11, the continuity of class A to class C quasi-linear amplifiers (theoretically these two stages will work through the sinusoidal voltage across their sources) will change the corresponding maximum output power and efficiency characteristics. At the same time, if bias is used to generate a differential engine (as in the traditional Doherty implementation), there is a trade-off between output power and efficiency. At the same time, the differential bias will increase the Doherty effect, but will reduce the achievable performance.

Challenges of the Doherty amplifier: synthesizer amplitude and phase matching (a), auxiliary amplifier current response (b), and power-efficiency tradeoff (c).

Variations and improvements
The following variations of the basic design may be more suitable for certain applications. In the traditional implementation, it provides designers with performance and flexibility options.

* There are multiple gain stages in the Doherty splitter and synthesizer

* N Road Doherty

* Intentionally dispersed separator

* Programmable separator

* Bias modulation

* Power modulation, that is, adding a third frequency enhancement technique to the two frequency enhancement techniques used by Doherty

* Envelope shaping

* Digital Doherty

In addition to the different architectures available to designers, adjustments can also be made at three points in the product life cycle. In the design phase, design parameters can be modified and passed to the production process as fixed values (for example, input separator design parameters). In the production process, you can usually modify or adjust the parameters based on the measured data, and then freeze or fix the parameters through programming. An example is the nominal bias voltage used to generate the target bias current in the device. After the equipment is deployed in the field, the parameters can be updated continuously or at a specific time in an open-loop or closed-loop manner. Open-loop solutions rely on fully predictable characteristics, while closed-loop solutions may require built-in measurement and control. An example is a temperature compensation circuit. These product life cycle options provide multiple solutions that are not "best". Designers know that the production and supply capabilities that follow the design are as important as the design challenges and trade-offs encountered during the design phase.

The opposite of zero-level implementation is digital Doherty (see Figure 4). The feature of this architecture is to perform input separation in the digital domain before digital-to-analog conversion. With the ability to apply digital signal processing to the signals applied to the two amplifier paths, unsurpassable performance can be obtained from a set of RF hardware. Compared with the standard Doherty implementation, the output power of the digital implementation can be increased by 60%, the efficiency by 20%, and the bandwidth by 50%, without reducing the predictive pre-correction linearity12.

Measurement Aided Design Process
In order to optimize the Doherty design, it is recommended to build a simulation environment that is well related to the design to understand trends and sensitivity. With this kind of simulation, a large part of the development process can be quickly covered. The input of the first step may include the load pulling data or model of the device, the theoretical study of the combined circuit and the response of the matching network, and the evaluation board containing measurement data or other empirical data. Based on this starting point, the design process can be supplemented with measurement aided design (see Figure 5).
For digital Doherty, the starting point of this approach is a Doherty amplifier that contains two input ports, input and output matching networks, active devices, bias networks, and combiners (see Figure 6). By measuring the prototype Doherty of the dual-input device, it is possible to gain a deeper understanding of the performance limitations, tradeoffs, and repeatability expected in the production environment. Crucial to the test configuration are the two signal paths, the signals of which may change with each other. In addition to applying precise, stable and repeatable amplitude and phase offsets to these signals, it is also very beneficial to be able to apply nonlinear shaping to at least one of the signal paths.