by Maxwell Kreizinger
This summer, as a rising junior studying electrical engineering at Cal Poly San Luis Obispo, I interned at Analog Devices with the Central Applications group in San Jose. The team wanted to investigate how different voltage regulator topologies affect the performance of an audio amplifier and tasked me with this as my 12-week project. In this post, I’m going to share some of the enlightening discoveries that I made while working on this project.
Throughout the first week of my experimentation, I believed that the voltage regulators I had selected to test demonstrated comparable performance relative to one another. I thought to myself, “Am I wasting my time? Should I spend my time working on something else?” That is until I made distortion vs. frequency measurements (see Graph 1) using the LT8608S and the LT8610 evaluation kits. The humps in the graph baffled me. In comparison to a benchtop power supply, these voltage regulators performed terribly. What was causing these voltage regulators to distort my audio output?
Graph 1. THD+N Ratio vs. Frequency (Unoptimized, 3.1 Vrms Input)
I discovered my first clue when I input a sinusoid to the audio amplifier and noticed that the current drawn from the voltage regulator by the audio amplifier was a full wave rectified sine wave. (Figure 1)
Figure 1. Current Draw of the Audio Amplifier, 3.1 Vrms, 1 kHz Input
The goal of the voltage regulators is to guarantee a fixed 5-volt output, regardless of the current drawn. However, as the load current rapidly varies, the regulator struggles to compensate for this variation. This results in an imperfect DC rail with noticeable voltage ripple, as shown in Figure 2 — the LT8608S and the LT8610 have 301.5 mV (6.03%) and 248.6 mV (4.97%) of ripple respectively, much more than what is acceptable for this application.
Figure 2. Voltage Ripple of Unoptimized LT8608S (Top) & LT8610 (Bottom), 3.1 Vrms Input, at peak distortion
As shown in Figure 3, this ripple lowers the effective voltage rail of the amplifier, distorting the output by causing the top of the output waveform to be cut off.
Figure 3. Clipping of Unoptimized LT8608S. (Voltage Rail in Yellow, Out+ in Blue, Out- in Green), 3.1 Vrms, 3.8 kHz Input
The voltage regulator is stressed by the amplifier’s continuously varying current draw, failing to adequately respond to high-frequency disturbances at its output. To solve this problem, the transient response must be improved. While the LT8608S and the LT8610 are internally compensated, which means that the control scheme is nearly entirely internal to the integrated circuit, we still have control over the feedback network, as Figure 5 illustrates. The goal of the feedback network is to connect the voltage output to the control scheme so that the regulator can adjust to a transient output. Modifying the feedforward capacitor (Cff) allows the regulator’s control scheme to address high-frequency disturbances more effectively, increasing the high-frequency component which travels though the capacitor and into the feedback pin. Therefore, the transient response can be improved by increasing the feedforward capacitance because this increases the influence of high-frequency disturbances. However, care must be taken not to increase the feedforward capacitance too much because this can cause the voltage regulator to become unstable. Taking this newfound knowledge into consideration, I summoned my soldering skills and increased the feedforward capacitors (from 10 pF to 100 pF for the LT8608S, and 10 pF to 47 pF for the LT8610).
Figure 4. Feedback Network of the LT8608S and the LT8610
With the modified feedback loop, I took new measurements of THD, and found that the distortion "humps" were eliminated! As shown in Graph 2 below, the improved voltage regulator circuit now features comparable performance to that of the benchtop power supply.
Graph 2. THD+N Ratio vs. Frequency (Optimized, 3.1 Vrms Input)
In addition, I measured that the voltage ripple has improved significantly (533% and 178% for the LT8608S and the LT8610 respectively) as shown in Figure 5. It is possible that the voltage ripple could be decreased further for both voltage regulators; however, this is not necessary because the voltage regulators already demonstrate superb performance, as shown by Graph 2.
Figure 5. Voltage Ripple of Optimized LT8608S (Top) & LT8610 (Bottom), 3.1 Vrms Input, at peak distortion
To put the cherry on top, the last thing that I wanted to see was how much harder I could push my amplifier while still delivering a clean output. As Graph 3 shows, I measured distortion vs. input signal level, and for both regulators, I gained over 100 mVrms of headroom. In a real-life application, this allows for a speaker to be driven louder and cleaner! What an exciting discovery!
Graph 3. THD+N Ratio vs. Level (3.8 kHz Input, LT8608S)
Graph 4. THD+N Ratio vs. Level (5.2 kHz Input, LT8610)
in reply to mcorriga