|Academic Supervisor||Rob Maunder, Michael Ng, Tim Norman|
|Pure Link||Active Project|
With the growing use of mobile communications, there is a greater requirement for efficiency to control the operating costs of networks. One area which can benefit from an increase in efficiency is the power amplifier. This component of a transmitter increases the power level of the signal to increase the distance that the signal can travel. However, there is an issue with the performance of current amplifier designs, they are a compromise between three design criteria: Efficiency, linearity, and cost. This creates a situation where, in order to design amplifier that is very linear, it is generally costly to produce and inefficient to run. Digital predistortion (DPD) provides a solution to this problem by allowing the amplifier to have nonlinearity in its response, and instead modifying the input signal in a way that accounts for the flaws in the amplifier’s behaviour, so that the predistorter combined with the amplifier appears to be linear.
The aim of my project is to use artificial intelligence to provide a solution for DPD in wireless communication systems, as this offers the potential for higher levels of performance while also reducing the complexity compared to traditional methods. These traditional methods relied upon either fitting a function to the properties of the amplifier, which can quickly become mathematically complex, or using a look up table of values to correct for amplifier distortion, which can require a large amount of memory to perform effectively. Machine learning, and specifically recurrent neural networks, can provide a new way of addressing this problem with amplifiers, as they can not only learn a nonlinear function, but also learn long term patterns in data, which can be used to address issues with hysteresis in amplifiers.
In the later stages of the project, the goal is to develop a bespoke hardware-based solution for DPD, firstly using programmable logic (FPGA), and potentially an ASIC solution. This will give the best possible performance of the predistorter in terms of latency and efficiency.