Dithering, Layout, and High-Quality Components: Tools to Decrease the Noise Floor


Scientists and engineers taking analog measurements often use the term noise floor, but it is often misunderstood and, as a result, not dealt with correctly. To reduce noise that occurs on your measurements, you need to have a firm understanding of the noise floor, its components, and what you can do to decrease it in your measurement system.


Introducing Noise Floor

The noise floor of a measurement device is the measured noise level with its inputs grounded.

You will usually see it expressed either as a noise density function with units of , or as a single number representing the total noise, expressed in Vrms. To convert from a noise density function to Vrms, you must integrate the noise density function over your bandwidth of interest. In the case of wide band (flat) noise, this integration breaks down into simple multiplication where

In general, you can derive the RMS noise of a device from the noise density function, but you cannot get the shape of a graph from a single number. The figure below illustrates a typical measurement device's noise density curve at low frequencies. This curve consists of the two sections in the figure. The steeply sloping portion to the left of the point, known as the  corner is referred to as  noise. To the right of the  corner, the noise level flattens out and is known as wide band noise.

Figure 1: Noise Spectral Density Curve

Components of the Noise Floor

Wide band noise generally appears flat in the frequency domain, meaning that equal energy occurs in every Hz of bandwidth. This type of noise can result from almost every component in the measurement device, whether they reside in the signal path, or you use them for reference. These components include op amps, resistors, voltage references, and analog-to-digital converters (ADCs). Using post-processing techniques, such as averaging, helps minimize the effects of wide band noise on your measurement accuracy. We will discuss these techniques further in this article.

 noise, however, is more complex. It is referred to as  noise because its voltage density is proportional the square root of frequency. You may also hear  noise referred to as "flicker noise" because it causes the least significant bit on a DMM to flip or "flicker." Thermal gradients among board components and contamination during IC manufacturing processes are the primary causes of this noise. These causes make  noise difficult to predict and control, and IC manufacturers generally do not adequately specify the impact of  noise. End users of data acquisition (DAQ) devices may find this especially troublesome because you cannot remove this uncertainty with any post-processing operations. For example, the longer you average, the more opportunity the board has to drift. Therefore, depending on the  slope, you may never converge on the "true" value, regardless of how much you average. In fact, as far as it has been proven, the  spectrum continues its upward slope to the left, limited only by the aperture of your measurement. IC manufacturers have seen strong correlation between  noise levels and the long term drift of the voltage references they manufacture.

Minimizing Noise Floor

NI has minimized wide band noise on data acquisition devices by designing them with high-quality amplifiers with a high Common Mode Rejection Ratio (CMRR). This means that the devices reject a significant portion of the noise experienced on both terminals of the amplifier, making your measurements less susceptible to common mode noise that can decrease accuracy.               

NI designs multifunction data acquisition devices with separated ground planes that connect to a single ground reference. Analog-to-digital and digital-to-analog converter chips are commonly designed with analog signals on one side of the chip and digital signals on the other. By placing the converter chips so they straddle the barrier between the analog and digital ground planes, noise generated on the digital side of the data acquisition board does not affect the analog side of the chip or the traces residing around the analog ground plane.

Thermal gradients in the measurement device can often induce noise. To combat this, NI has implemented several features to ensure our measurement devices experience minimal temperature drift. NI uses matched, temperature tracking circuits and custom resistor networks restrict temperature drift to 6 ppm/°C on all data acquisition hardware. In addition, NI uses high-quality components with  and drift characteristics that have been well characterized. Finally, NI self-calibration circuitry, accessible by a single function call, references a highly stable voltage source that drifts at a rate of only 0.6 ppm/°C.

Further Decreasing Your Noise Floor

While some types of noise result from imperfections in ICs or environmental factors, such as temperature, the resolution of the board can also create noise. This is known as quantization error.

NI driver software allows you to enable dithering through software. When you enable the software, it adds approximately 0.5 LSBrms of Gaussian white noise to the input signal. This noise is added to the signal before the input to the ADC. As a result, a signal that might fall somewhere in the smallest voltage difference that the board can detect (known as code width and defined by the formula ) now randomly bounces above and below the boundaries of that code. When sampled, points now appear on both the top and bottom boundaries, and the number of points on either the top or bottom of the code width are weighted based on the location of the actual signal. You can then use averaging to essentially zoom in past the specified resolution of the board, providing more accurate measurements that are less influenced by wide band noise. For instance, a 12-bit board can perform with 14-bit resolution with dithering enabled. You can also disable dithering for high-speed applications that do not use averaging.

Figure 2: You can decrease quantization error on 12-bit devices using dithering

These techniques not only reduce noise caused by nonideal components on the measurement device, but they also help reduce noise originating from other components of the measurement system. In addition,  noise at the system level can result from the sensor, so you should choose a high-quality sensor to help insure a lower noise floor for your overall system. The environment, long wires, nearby electromagnetic fields, and other sources can also induce noise on the system. To reduce noise from external sources, you should ground your system properly and use shielded cables. For more information on these topics, please see the related links below.