In the design of analog front ends that require low noise, it is important that all aspects of the design are considered. This is especially true when looking at design targets with 20+ bits of resolution for, say, power metering or other low-frequency audio applications. When measuring a circuit capable of producing such high performance, creating a board with the proper references is a critical component of success.
The design of an evaluation board to properly measure a part with high dynamic range and low noise requires that the board include a low noise reference. But how low is low? To address this question, I will present a case where we experienced degraded performance due to issues with reference noise. To explain the issue, I will provide some background information on sampling analog-to-digital converters.
For a sampling ADC, the reference voltage supplied to the part determines the ultimate quality of the converted signal. In most ADCs used for general-purpose applications in the range of 10 bits of resolution, creating a board reference is not too difficult if attention to detail and proper bypassing rules are followed. To illustrate this point, Figure 1 shows a typical reference that a board design may use to develop the voltage the ADC will use for conversion. As can be seen in the picture, the board is designed with regulators that can tolerate large decoupling capacitors. In addition, the decoupling caps are placed on the supplies to the regulators.
Notice, if you will, that the bypass caps are made up of a series of smaller caps in parallel, with at least one of those smaller caps placed as close to the DUT as possible. For more information on this topic, refer to my previous blog. Once these caps are properly placed on the board and are targeted for adequate filtering, is the reference design adequate? It depends. What about the low-frequency noise coming out of the regulated voltage used for the reference?
To answer these questions, I must talk briefly about what happens to noise that exists on the golden reference voltage of the ADC. The reference and its noise is sampled just as if it were another input, as shown in Figure 2 for a typical input sampling stage of an ADC. Therefore, any noise or unwanted energy is sampled along with the signal. The noise sampling is more or less of a problem depending on the architecture chosen. Nevertheless, the existence of noise is problematic for very high-performance ADCs with 21+ bits of resolution and bandwidths that process signal on the DC+ to audio bandwidths, as is the case for power metering front ends mentioned in this example.
The data shown in Figure 3 is taken from a 21-bit analog front end design for exactly this market. The ADC that was part of the analog front end was a sigma delta ADC that used the regulated supply as the reference for the ADC. Extreme attention to detail was paid when designing the board. This detail included the choice of regulators for the references, as well as the placement of capacitors to filter the supply. In addition, special attention was given to the routing of the traces on the board. The results shown in Figure 3 explain what happened.
The plot shown provides the FFT of the output of this front end before and after a solution was provided to fix the problem. The trace in blue on the plot shows a strange noise shelf around the signal process. This noise shelf accounts for ∼10 dB in loss performance. Why? To answer this question, remember my previous statement about the sampling stage shown in Figure 2. In the case of the integrator sampling stage of the sigma delta ADC shown in Figure 2, the integrator design used the reference as another input that is sampled just like the signal. Therefore, any noise on the reference is sampled just like the signal. As can be seen in the plot, there is low-frequency noise on the reference that is sampled and shows up in the FFT. To combat this problem, I had to design a 0.5Hz filter into the regulated reference to reduce the noise before it was sampled and folded into converted bandwidth (shown in Figure 1).
In most cases, regulated noise is usually not a problem unless these voltages are used as references for high-performance ADCs. In this case, every aspect of the board design is not only important but crucial.
Have you experienced similar situations where the unexpected noise from a reference caused issues?