To convert a 0-5V analog sensor signal into a digital value with a maximum quantization error of ±2mV, you need a minimum resolution of 12 bits, which yields a 1.22mV Least Significant Bit (LSB). For slow-moving DC signals (under 1 kSPS), the correct architecture among analog digital converter types is a Delta-Sigma chip like the Texas Instruments ADS1115 (approx. $3.50). If your signal moves faster (up to 200 kSPS), you must switch to a Successive Approximation Register (SAR) ADC like the Analog Devices AD7606 (approx. $18.50). You cannot pick an ADC architecture without first converting your physical voltage and speed requirements into hard mathematical limits.
The Core Conversion: Calculating ADC Resolution
The fundamental 'conversion' in any ADC is mapping a continuous voltage range into discrete digital steps. The size of each step is your LSB. If your application demands a specific precision, the formula dictates your minimum bit depth.
The Quantization Formula:
LSB = V_REF / 2^N
Substituted for a 5V system requiring ~1.2mV precision:
1.22 mV = 5.0V / 2^12
Here is how the resolution scales for neighboring bit depths around our 12-bit target (a ±20% range in bit depth dramatically shifts your physical precision):
| Bit Depth (N) | Total Steps (2^N) | LSB Size at 5.0V | Max Quantization Error (±0.5 LSB) |
|---|---|---|---|
| 10-bit | 1,024 | 4.88 mV | ±2.44 mV |
| 11-bit | 2,048 | 2.44 mV | ±1.22 mV |
| 12-bit | 4,096 | 1.22 mV | ±0.61 mV |
| 13-bit | 8,192 | 0.61 mV | ±0.30 mV |
| 14-bit | 16,384 | 0.305 mV | ±0.15 mV |
How Reference Voltage Shifts the Math
The assumption that fixes your bit-depth requirement is the Reference Voltage (V_REF). The LSB size is entirely dependent on this ceiling. If you change the voltage range, the 'conversion' shifts immediately:
- 3.3V Systems (e.g., ESP32/STM32 internal ADCs): A 12-bit ADC yields a 0.80 mV LSB. You get better absolute precision on a 3.3V rail than a 5V rail for the exact same silicon.
- 5V Systems (e.g., Arduino Uno ATmega328P): A 10-bit internal ADC yields a 4.88 mV LSB. This is often too coarse for precise load cell or thermocouple readings without external amplification.
- ±10V Bipolar Industrial Systems: A ±10V signal means a 20V peak-to-peak swing. A 12-bit ADC here yields a 4.88 mV LSB. To achieve 1mV precision in a 20V industrial environment, you are forced to jump to a 16-bit architecture (0.305 mV LSB).
When the Conversion Becomes Meaningless
Calculating theoretical LSB size is useless if you ignore the physical realities of your circuit. The math breaks down in two specific scenarios:
2. Violating the Nyquist-Shannon Theorem: If you are sampling a 10 kHz AC waveform, your ADC must sample at strictly greater than 20 kSPS (samples per second). If you attempt to 'convert' this signal using a slow Delta-Sigma ADC running at 10 kSPS, you will experience aliasing. The digital output will reconstruct a completely false, lower-frequency waveform. For AC and dynamic signals, sampling rate dictates the architecture just as much as voltage resolution.
Decision Tree: Picking the Right Analog Digital Converter Types
Use this decision path to terminate your component search. Match your signal bandwidth and required resolution to lock in the correct architecture and a specific, purchasable part number.
| Signal Bandwidth (Speed) | Required Resolution | Best Architecture | Concrete Part Pick (2026) |
|---|---|---|---|
| DC to 1 kSPS (Thermocouples, load cells, battery monitoring) |
16 to 24-bit | Delta-Sigma High resolution, slow speed, excellent noise rejection. |
TI ADS1115 (16-bit, I2C, ~$3.50) or ADS1256 (24-bit, SPI, ~$12.00) |
| 1 kSPS to 1 MSPS (Audio, motor control feedback, oscilloscope front-ends) |
12 to 16-bit | SAR (Successive Approximation) Balanced speed and resolution, zero latency. |
ADI AD7606 (16-bit, 200kSPS, ~$18.50) or TI ADS8688 (16-bit, 500kSPS, ~$11.00) |
| > 10 MSPS (RF sampling, radar, high-speed digital comms) |
8 to 12-bit | Pipeline / Flash Extreme speed, lower resolution, high power draw. |
TI ADC12DJ1600 (12-bit, 1.6GSPS, ~$140.00) |
FAQ: Real-World ADC Debugging
Q: Why does my 16-bit SAR ADC only give me 12 bits of clean, stable data?
A: This is almost always a reference voltage or layout issue. A 16-bit ADC requires an ultra-low-noise voltage reference (like the TI REF5050) and a dedicated ground plane. If you are powering the ADC's V_REF pin directly from a noisy 5V USB rail or a switching buck converter, the noise on the reference pin modulates the digital output. Add a 10µF X7R ceramic capacitor and a 1µF NP0/C0G capacitor directly across the VREF and GND pins, placed within 2mm of the IC.
Q: Can I just use the ESP32's internal ADC for precision 12-bit measurements?
A: No. The internal ADCs on the ESP32 (and most microcontrollers) are notoriously non-linear, especially near the 0V and 3.3V rails. The ESP32's internal ADC often exhibits an effective number of bits (ENOB) closer to 9 or 10 bits due to internal digital switching noise. For true 12-bit precision, bypass the internal ADC and wire an external I2C/SAR chip like the ADS1115 to the GPIO pins.
Q: What is the difference between unipolar and bipolar ADC inputs?
A: A unipolar ADC measures from 0V to V_REF (e.g., 0 to 5V). A bipolar ADC measures across a zero-crossing (e.g., -5V to +5V). If you feed a -2V signal into a unipolar ADC, you will clip the internal protection diodes and potentially destroy the silicon. Always verify the input topology in the datasheet before wiring up AC-coupled or motor-drive feedback signals.
