The use of capacitors in an audio signal chain is often fraught with mysticism and little quantitative analysis to justify capacitor selection. With many capacitors costing more than the integrated circuits they serve, it is a challenge to determine a solution that balances cost, size and performance.

This article offers a brief overview of capacitor technology and compares the use of various capacitors in a real-world audio application. The capacitors in this comparison are configured to AC-couple the inputs of a precision audio data converter. Analysis of the data converter output provides an easy way to compare capacitor contributions to signal-path distortion. The goal is to help you make informed decisions about capacitor selection in audio signal-chain applications.

Multilayer ceramic capacitors (MLCCs) are immensely popular in many applications because of their volumetric efficiency and relatively low price. The advantage of this capacitor technology lies primarily in its use of special dielectric materials. To understand why, recall that the capacitance of a simple parallel-plate capacitor follows Equation 1.

where k is the relative permittivity of the dielectric material placed between the plates (otherwise known as the dielectric constant), ε₀ is the permittivity of free space, A is the area of the capacitor plates and d is the distance between the plates.

Equation 1 shows that materials with higher dielectric constants enable smaller capacitor volumes for a given capacitance value. This accounts for the large variations in the size of a 10-µF capacitor with a particular voltage rating, since it all depends on the capacitor dielectric.

MLCC capacitors are organized into different classes depending primarily on their thermal range and stability over that range. Class II ceramics are often referred to as “high k” because their relative permittivities range from 3,000 (X7R) up to 18,000 (Z5U). By contrast, Class I C0G/NP0 capacitors tend to have relative permittivities in the range of 6 to 200. They are “high-performance” ceramic capacitors because their capacitance is more stable than most other dielectrics.

Plastic film capacitors that use materials like polyethylene or polypropylene tend to have even lower relative permittivities, typically less than 3, and also offer very good stability. Table 1 shows the relative permittivity of some common dielectric materials used in capacitors.

In portable electronics like smart speakers,^[1] it is tempting to use high-k MLCCs due to their small size and low cost. It is important to remember, though, that while their relative permittivity is very high, their capacitance changes significantly over applied voltage and temperature, which can degrade signal-chain performance. This variation in capacitance is primarily due to the use of heavy concentrations of barium titanate in the dielectric. Barium titanate is ferroelectric in nature, which means that increasing the electric field intensity inside the material decreases its relative permittivity. According to Equation 1, this will also lead to a decrease in capacitance. Thus, applying a time-varying voltage to the capacitor results in a time-varying capacitance, distorting the current flowing through the capacitor. The change in capacitance with applied voltage is known as the capacitor’s voltage coefficient, and it can be the dominant source of distortion in the low-frequency spectrum where capacitor impedance is relatively high. Furthermore, as the signal amplitude increases, greater distortion occurs. At higher frequencies, the distortion is less noticeable due to lower capacitor impedance, leading to a negligible voltage drop across the capacitor.

Minimizing the voltage drop across the capacitor will mitigate distortion. One approach to do this is to increase the impedance in series with the capacitor to limit the current flowing through it. To demonstrate this, total harmonic distortion-plus-noise (THD+N) measurements were taken on the Texas Instruments TLV320ADC5140 audio analog-to-digital converter (ADC) evaluation module (EVM),^[2] with 4.7-µF X7R 0805 AC-coupling capacitors on the input. This ADC has a programmable input impedance that can be set to 2.5 kΩ, 10 kΩ or 20 kΩ. Figure 1 shows the results from a single channel with this capacitor while varying the input impedance with a 1-V_RMS input signal.

Figure 1 shows that distortion increases at lower frequencies and peaks at the –3-dB cutoff frequency of the high-pass filter formed by the capacitor and the ADC input impedance. This is because the –3-dB point is where capacitor impedance equals the load impedance, which is when the voltage across the capacitor is highest. For cases with a 2.5-kΩ input impedance, the cutoff frequency is around 13 Hz, but the distortion starts much further out—around 400 Hz—and peaks around 13 Hz.

As expected, the change in distortion scales fairly linearly with increasing impedance. The drawback of this approach is that increasing the device’s input impedance degrades the signal-to-noise ratio performance by a few decibels. Moreover, adding series resistance externally also leads to additional noise and increases the gain error. The added gain error increases system complexity, since it may require calibration. For cost- or area-sensitive designs, increasing the impedance may work fairly well, but performance-focused designs will suffer with this approach.

Another option is to increase the value of the capacitor until its impedance is low enough in the band of interest to reduce its distortion. Figure 2 shows the change in distortion over frequency, while Figure 3 shows the change in distortion over input amplitude as a result of varying the input capacitance for similarly rated capacitors. Note that the 47-µF capacitor tested was in a 1206 package, and increasing the package size will also tend to slightly reduce distortion. However, the dominant effect is still the dramatic increase in capacitance.

The data in Figure 3, taken with a 100-Hz input signal, shows that increasing the capacitance works fairly well for AC-coupling applications. MLCCs with capacitances greater than 47 µF are readily available in 0805 or 1206 packages.

Both of the approaches to mitigating distortion involve significantly reducing the cutoff frequency of the high-pass filter, either by increasing resistance or increasing capacitance. When using MLCCs for AC coupling, a good rule of thumb is to place the corner frequency two decades away from the specified passband. Or, worded differently, make the capacitor 100-times larger than needed or as large as can be reasonably supported. When using excessively large capacitors, even small amounts of series resistance will result in very low cutoff frequencies. This means that undesirable low-frequency noise, such as the typical 1/f noise of a complementary metal-oxide semiconductor front end, will not be filtered. Applying digital filtering further down the signal chain will suppress typical 1/f noise. However, it is still difficult to completely eliminate the influence of the ceramic capacitor in the bass region of the audio band. Applications demanding high performance will still benefit from the use of better technologies.

In many audio applications such as analog filter design, the preferred solution is to use C0G/NP0 ceramic capacitors because they have much better performance and are still available in small packages. While this is a great solution, it is not always feasible. In audio signal chains, keeping resistances low minimizes noise, but doing so requires increased capacitance for a given cutoff frequency. C0G/NP0 capacitors are not easily found in values greater than 1 µF, so they have limited utility in AC-coupling and other high-pass filter applications.