Design of Broadband RF PAs Using an Improved Bayesian Optimization Algorithm

The present work proposes an improved Bayesian optimization (BO) method for the design of broadband radio frequency (RF) power amplifiers (PAs). A new acquisition function (AF) is incorporated into this proposed BO method, which is called the probability improvement with maximum upper confidence function (PIMUCB). With the consideration of both the magnitude of the maximum expected improvement (EI) at the evaluation point and the probability that the evaluation point will exceed the current optimal value, the optimization efficiency can be greatly increased. On the basis of this method, a broadband single-ended PA and a broadband sequential load-modulated balanced amplifier (SLMBA) are designed, fabricated, and tested. Results from measurements indicate that the single-ended PA is capable of delivering an output power exceeding 40.4 dBm, a drain efficiency (DE) exceeding 65.9%, and a gain of over 10.4 dB between 2.5 and 3.5 GHz. As for the broadband SLMBA, it shows a saturated output power of 42.2–43.3 dBm, a saturation DE of 58.8%–79.8%, and a 10-dB output power back-off (OBO) DE of 53%–72.5% between 1.6 and 2.4 GHz. By comparing the proposed method to the existing BO methods, the proposed method not only achieves better performance for both PAs but also shows a faster convergence rate. Additionally, both PAs undergo a modulated signal test and a digital predistortion (DPD) test using a 20-MHz fifth generation (5G) new radio signal with 10-dB peak-to-average power ratios (PAPRs), and excellent linearity is achieved for both PAs.