A Low-Phase-Noise InGaAs/GaAs BiHEMT Balanced Colpitts VCO with Deep-Learning-Optimized On-Chip Inductor for Satellite Communication Applications

In this communication, a balanced Colpitts VCO with low phase noise and a wide tuning range is presented, which was implemented using a 2 μm GaAs BiHEMT process. To improve phase noise, a noise-shifting technique introduced by the cross-coupled pair was applied. A deep-learning-assisted genetic algorithm was employed to optimize the quality factor of an on-chip inductor, further improving the phase noise. Additionally, three-bit switches were utilized to expand the tuning range of this VCO. Finally, a VCO prototype was implemented and fabricated for verification. The measurement results demonstrate that the proposed VCO can achieve a wide tuning range of 33% (2.56–3.57 GHz) and an excellent phase noise of −136.25 dBc/Hz at a 1 MHz offset, validating the efficiency of the adopted circuit techniques and the machine learning-assisted optimization strategy for low-phase-noise design. Consequently, this VCO proves to be a highly competitive candidate for signal sources in S-band satellite communication systems.