Esp32 RF Front End
The RF front end is the single hardest sub-block in an ESP32-class design, because it has to pull in a microvolt-level 2.4 GHz signal, amplify it cleanly, and transmit a spec-compliant signal back out — all on the same piece of silicon as a digital CPU switching millions of gates a nanometer away. A full receive/transmit chain is built from five distinct stages (LNA, downconversion mixer, PLL/VCO synthesizer, upconversion mixer, power amplifier) plus the passive matching networks between them, and every one of those stages has its own noise, linearity, or power-efficiency requirement that the next stage in the chain depends on — get one stage wrong and the whole radio's range or power consumption suffers, even if every digital block elsewhere on the die is perfect.
The LNA's job is won or lost in its first stage, because every dB of noise it adds is noise the rest of the chain can never remove. Noise figure compounds multiplicatively down the chain according to the Friis equation, which means the LNA's own noise contribution dominates the entire receiver's sensitivity almost regardless of how clean every later stage is — this is why LNA design gets disproportionate attention relative to its tiny die area, and why a cheap, noisy LNA is one of the fastest ways to quietly cut an ESP32-class part's real-world Wi-Fi range without it ever showing up as a hard functional failure in test.
The power amplifier, not the digital logic, is why battery life on a transmitting ESP32-class device is dominated by radio duty cycle rather than CPU workload. A Wi-Fi PA driving a compliant output power routinely draws tens to a few hundred milliamps for the duration of a transmission — vastly more than the digital core pulls running application code — which is exactly why real-world power budgeting for these parts focuses on minimizing transmit time and radio-on duration rather than optimizing CPU clock speed, and why the PMU's deep-sleep mode exists specifically to shut the entire RF chain down between transmissions rather than just idling the CPU.
The PLL and VCO are shared infrastructure, not a per-chain component, which is exactly where a subtle design bug hides. Both the receive mixer and the transmit mixer pull their local-oscillator signal from the same synthesizer, so any noise or instability on that shared LO — phase noise, reference spurs leaking in from the 40 MHz crystal path — degrades both receive sensitivity and transmit spectral purity simultaneously, making PLL phase-noise performance one of the few single points of failure that can silently violate both the receiver's sensitivity spec and the transmitter's spectral-mask compliance at the same time.
A real RF front-end schematic is this five-stage chain plus a shared synthesizer plus the matching-network passives tying them together — and the engineering discipline that makes an ESP32-class radio actually work in the field comes down to minimizing noise figure early in the receive chain, managing PA power draw aggressively, and keeping the shared PLL clean enough that neither direction of the radio quietly degrades the other.