01 ADC Quantization Error

1. Introduction

AD\Test\2024\April\TestATMEGA16.SchDoc

Recently, the sampling characteristics of a 10-bit ADC were measured. When the input signal changes very slightly, the quantization error of the ADC greatly affects the sampling results. Even averaging 512 data points cannot eliminate the impact of quantization error. Next, tests were conducted by introducing noise interference into the signal to mitigate the errors caused by quantization. Oversampling the data and averaging it can improve the ADC’s resolution.

2. Measurement Results

Modifying the parameters of the low-pass filter RC for the analog input signal can change the proportion of input noise in the collected signal. Reducing the input resistance from 10k ohms to 1.5k ohms decreases the filtering effect. At this time, with 512 continuous data points, the data fluctuation range is seen to be 3, while previously, the fluctuation range was only 2, indicating that the 50Hz interference signal in the current signal has indeed increased. Next, we will re-measure the characteristics of the ADC average data at different voltages, comparing the influence of noise on the ADC’s quantization error.

Measuring 150 data points, the QR10 voltage divider resistance increased from 10k ohms to 10.15k ohms. The blue data line represents the average result of 512 data points, while the orange represents a single ADC sampling result. The voltage gradually increases in the front, and then suddenly drops at the back, the specific reason for which is still unknown. Next, we will compare the average results measured with low noise and high noise.

Below are the results of two ADC measurements. The cyan data represents the results measured under low 50Hz noise conditions, while the orange data represents the results under high 50Hz noise conditions. We will not discuss the sudden drop in the orange data for now, as I cannot explain the specific reason at this time. When the input and output noise is relatively low, we can see that the quantization error of the ADC significantly affects the data. However, when the noise is relatively high, the measured data exhibits better measurement linearity, and the corresponding quantization error is significantly reduced. Theoretically, each time the data is averaged four times, one bit of sampling quantization resolution can be improved. Averaging 512 data points approximately increases the ADC resolution by 4.5 bits. Therefore, when the ADC bit count is relatively low, appropriately increasing the amplitude of the input signal noise is beneficial for improving the signal sampling resolution.

※ Conclusion ※

This article compares the effect of noise in the data on quantization error during the averaging operation of a 10-bit ADC. If the input signal is thoroughly cleaned of noise, the ADC’s quantization error cannot be eliminated through data averaging. Therefore, ‘water that is too clear has no fish’. The greater the wind and waves, the more valuable the fish.