2026-05-18
If you've ever wondered why a $50 handheld multimeter can deliver 4½ digits of accuracy while a $500 oscilloscope ADC drifts noticeably between calibrations, the answer is the dual-slope integrating ADC. It's slow, elegant, and brilliantly immune to the noise sources that plague faster converters.
The core trick: instead of comparing the input to a reference directly, you let the input charge a capacitor through a resistor for a fixed time, then discharge that capacitor with a known reference current and measure how long the discharge takes. The ratio of times equals the ratio of voltages — and crucially, the integrating capacitor and resistor values cancel out of the equation.
The three phases:
Why it's so accurate: Notice R, C, and clock frequency all cancel. Only VREF matters for absolute accuracy — and you can buy a 2 ppm/°C reference for under $5. Better yet, by choosing T₁ to be an integer multiple of the AC line period, line-frequency noise integrates to zero. This is called normal-mode rejection, and it's why DMMs feel rock-steady when probing noisy industrial gear.
The 50/60 Hz rule of thumb: Set T₁ = 100 ms (or any integer multiple of 1/60 s AND 1/50 s — 100 ms is the smallest that works for both). This gives infinite theoretical rejection of mains hum and its harmonics. Bench DMMs labeled "5½ digit, 10 readings/sec" are running exactly this trick — that's why you can't get faster readings without losing digits.
Real-world example: The classic ICL7106 (and its successors in Fluke 17x-series meters) is a dual-slope ADC with built-in LCD driver. Pair it with an LM4040 2.5 V reference, a 0.1 µF polypropylene integrating cap (low dielectric absorption is critical here!), and a 100 kΩ integrating resistor. You get 3½ digits, 3 readings/sec, and 50/60 Hz rejection — all from one chip plus a handful of passives.
Watch out for: dielectric absorption in the integrating cap. Ceramic caps "remember" previous charge and cause nonlinearity. Use polypropylene or polystyrene for ≥4-digit designs.
