If your ESP32 battery project dies in weeks instead of months, you're probably making at least one of these mistakes.

I've measured hundreds of ESP32 sleep-wake profiles on automated test benches. These are the five issues I see most often — and exactly how to fix them.

---

Mistake #1: Using the Wrong Voltage Regulator

The AMS1117 on most dev boards draws 5 mA quiescent current. Your ESP32 in deep sleep draws 10 µA. That means your regulator is consuming 500× more power than your microcontroller.

Fix: Use an MCP1700-3302E (1.6 µA quiescent) or go direct from a 3V coin cell to VDD — no regulator needed if your source is 2.0-3.6V.

Mistake #2: Not Saving Wi-Fi Channel in RTC Memory

Every time your ESP32 wakes up and calls WiFi.begin(), it scans all channels looking for your access point. That scan takes 3-6 seconds at 80-180 mA.

Fix: After connecting, save WiFi.channel() and WiFi.BSSID() in RTC_DATA_ATTR variables. On the next wake, pass them to WiFi.begin() to skip the scan entirely. This alone cuts Wi-Fi connection time to under 1 second.

Mistake #3: Leaving GPIOs Floating

Unconnected GPIO pins in an undefined state can create leakage paths through internal pull-up/pull-down networks. This adds 10-100 µA to your deep sleep baseline — and it's invisible unless you measure it.

Fix: Before entering deep sleep, explicitly set unused pins as inputs with pull-downs disabled:

gpio_set_direction(GPIO_NUM_XX, GPIO_MODE_INPUT);
gpio_pulldown_dis(GPIO_NUM_XX);
gpio_pullup_dis(GPIO_NUM_XX);

Or use rtc_gpio_isolate() for RTC-domain pins.

Mistake #4: Using WiFi.persistent(true)

This is the default on many ESP32 Arduino cores. It writes your Wi-Fi credentials to flash every single boot. That's a ~50 ms write operation drawing 40+ mA, and it wears out your flash over thousands of cycles.

Fix: Call WiFi.persistent(false) before WiFi.begin(). Store credentials as constants in your code.

Mistake #5: Measuring with a Multimeter

A multimeter averages current over its sampling window. An ESP32 wake cycle has current spikes from 10 µA to 350 mA in under 1 ms. Your multimeter shows you a meaningless average and misses the transients that actually drain your battery.

Fix: Use a shunt resistor + oscilloscope or DAQ system sampling at 10 kS/s minimum. Tools like the Nordic PPK2 or a LabVIEW-automated DAQ rig give you the full picture.

---

The Bottom Line

Battery life is an engineering problem, not a guessing game. Measure your baseline. Fix these five issues. Measure again. Most projects can go from "dies in 2 weeks" to "runs for 6 months" just by addressing these fundamentals.

I built an entire course around this process — automated measurement, power budgeting, and validated field deployment. If you're serious about building products that survive on batteries, check it out.