Scalable Power Converter Architectures with Quantized Output and Envelope Prediction for Wireless Communication

Power supply modulation techniques such as envelope tracking (ET) have achieved wide output bandwidths and high efficiencies suitable for wireless systems in 3G and 4G communication. However, extending the bandwidth further for high power 4G LTE and 5G applications has been very challenging with the traditional multi-phase buck converter-based ET approach. This paper introduces two scalable converter architectures with quantized output and machine learning-based envelope prediction for higher bandwidth and efficiencies in future wireless base stations. The first architecture constitutes multi-level converters that are series stacked to achieve the quantized levels (or bins) depending on the required voltage resolution. The second architecture employs interleaved parallel converters with constant current control to achieve quantized current levels. Further, ‘Extreme Gradient Boosting’ algorithm is used to predict the baseband signal a few tens of nanoseconds in advance, which facilitates improved ET with minimal control latency that is vital for high linearity of the radio frequency (RF) power amplifier (PA). Detailed analysis and simulation results on PSIM using real parameters of GaN FETs validate the concepts.