How Sigma-Delta ADCs Use Oversampling and Noise Shaping

A sigma-delta (\(\Sigma\)-\(\Delta\)) analog-to-digital converter (ADC) converts continuous analog signals into high-resolution binary representations by combining oversampling, noise shaping, and digital filtering. Rather than relying on high-precision multi-bit hardware at the Nyquist frequency, a sigma-delta ADC samples the input at a rate significantly higher than the signal bandwidth and utilizes an internal feedback loop to push quantization noise out of the band of interest. A subsequent digital decimation filter eliminates the high-frequency noise and downsamples the high-speed, low-bit stream into a precise, high-resolution multi-bit binary output.

1. The Principle of Oversampling

In conventional Nyquist-rate ADCs, the sampling frequency (\(f_s\)) is set just above twice the highest input signal frequency (\(2f_0\)). In this scenario, the total quantization noise power—determined by the step size of the quantizer—is distributed evenly across the Nyquist bandwidth from \(0\) to \(f_s/2\).

A sigma-delta ADC employs oversampling, sampling the input signal at a frequency \(f_s\) that is many times greater than the Nyquist rate (defined by the Oversampling Ratio, \(\text{OSR} = f_s / (2f_0)\)). By spreading the same fixed amount of quantization noise power across a much wider frequency spectrum (\(0\) to \(f_s/2\)), the noise density within the narrow target bandwidth (\(0\) to \(f_0\)) drops dramatically. Each doubling of the oversampling ratio reduces the in-band noise floor by approximately \(3\text{ dB}\), yielding an effective resolution increase of \(0.5\text{ bits}\).

2. Noise Shaping via the Delta-Sigma Modulator

Oversampling alone is not sufficient to achieve high resolutions (such as 16 to 24 bits) efficiently. Sigma-delta ADCs introduce noise shaping through a feedback loop containing an integrator, a quantizer (often a simple 1-bit comparator), and a digital-to-analog converter (DAC).

The modulator processes the signal and the quantization error differently: * For the Input Signal: The integrator acts as a low-pass filter within the loop, allowing the low-frequency analog input signal to pass through to the output largely unchanged. * For the Quantization Noise: The feedback loop creates a high-pass filter response for the quantization error introduced by the comparator.

Consequently, noise shaping actively pushes (or “shapes”) the majority of the quantization noise out of the low-frequency signal band and concentrates it into the high-frequency spectrum where no signal information exists.

3. Digital Filtering and Decimation

The output of the modulator is a high-speed, low-bit (often 1-bit) binary pulse-density modulated stream. To convert this into a standard high-resolution binary number, the signal passes through two digital stages:

1. Digital Low-Pass Filtering: The digital filter strictly cuts off all the high-frequency quantization noise that was shifted outside the signal band by the modulator.
2. Decimation: The filter averages the high-speed 1-bit data points and downsamples the stream to a lower sampling rate matching the Nyquist rate of the original signal.

By averaging thousands of 1-bit samples over time, the decimation process reconstructs intermediate voltage levels with extreme precision, outputting standard binary words (e.g., 16-bit, 20-bit, or 24-bit fixed-point integers) with high signal-to-noise ratios (SNR).