Abstract
This paper presents the design and simulation of a 2.44 GHz downconversion mixer implemented in a 0.5 μm CMOS process. By adapting the BJT-based "micromixer" topology to CMOS, the design overcomes the traditional trade-offs between power consumption, gain, and linearity. Utilizing a Class-AB translinear loop input stage, the mixer achieves a conversion gain of 7.5 dB and an IIP3 of 4.6 dBm while dissipating less than 3 mW. These results demonstrate the micromixer's potential for high-performance, low-power RF front-ends in wireless applications.
Introduction
In the RF front-end of wireless systems, the downconversion mixer is a critical component responsible for translating high-frequency signals (RF) to lower intermediate frequencies (IF). However, designing these components in CMOS technology is challenging due to the low transconductance-to-current ($g_m/I$) ratio of MOS transistors. Engineers typically face a "pick two" dilemma: high gain, high linearity, or low power consumption.
Standard Gilbert cell mixers rely on common-source differential pairs, where transconductance drops as the input signal increases, leading to significant non-linear distortion. This paper explores an architectural shift toward a micromixer topology to break this trade-off for WCDMA systems operating in the 2.4 GHz range.
Methodology
The core innovation of this work is the implementation of a CMOS Micromixer, which replaces the standard differential input stage with a Class-AB translinear loop. This loop, consisting of transistors $M_{b1}, M_{b2}, M_1,$ and $M_2$, is mathematically derived to show that while individual branch currents are non-linear, their difference remains a linear function of the input voltage over a much wider range than a standard differential pair.
To maintain performance at GHz frequencies, the authors addressed parasitic-induced phase shifts that can degrade the 180° symmetry required for a double-balanced mixer. By introducing a 2nH spiral inductor ($L_1$) and tuning the size of transistor $M_1$, the design restores perfect phase opposition and maximizes gain. The circuit was simulated using HPADS (High-Frequency Structure Simulator) based on a 0.5 μm CMOS process model.
Results
The simulation results for the 2.44 GHz mixer are highly competitive for the technology era. The Class-AB stage provides a theoretical 9 dB gain advantage over a standard differential pair with the same biasing.
As shown in Table 1, the design achieves an IIP3 of 4.6 dBm, which is well above the typical -5 dBm requirement for such systems. The conversion gain is 7.5 dB, and the power dissipation is kept below 3 mW at a 3V supply. Input matching ($S_{11}$) was recorded at better than -20 dB, achieved via a simple series capacitor.
Discussion and Conclusion
The CMOS micromixer demonstrates that a Class-AB input stage can maintain nearly constant large-signal transconductance until the input signal becomes very large, offering a significant linearity advantage over traditional Gilbert cells. While there is a slight "physics tax" in the form of a higher theoretical noise figure (13 dB) compared to differential pairs, this is considered a worthy trade-off for the superior linearity required in stages following the Low Noise Amplifier (LNA).
Limitations of this work include the reliance on simulation data rather than physical measurements and the use of a 0.5 μm process. Future work should involve scaling this topology to deep-submicron nodes (e.g., 28nm), where lower voltage headroom may complicate the translinear loop biasing. Nevertheless, the micromixer remains a robust architectural choice for low-power, high-linearity wireless applications.
Summary updated