Class D Amplifier PCB Layout
A class D amplifier is efficient because its output devices spend almost all their time either fully on or fully off, and that is exactly what makes the layout difficult. Every switching transition is a fast edge carrying a large current, and the board is the only thing standing between a clean audio output and a radio transmitter. Layout is not a refinement in these designs; it is part of the specification.
Why Class D Layout Is Different
In a linear amplifier the output devices dissipate the difference between the supply and the output, so the layout mainly affects thermal performance. In a class D amplifier the output devices are switches, the audio information is encoded in the width of the pulses, and any disturbance of those edges appears directly in the demodulated output as distortion or noise.
Efficiency also means current. A 100 watt channel running from a 36 volt rail draws close to three amperes on average and considerably more at the peaks of the modulation, so the layout has to carry that current without significant voltage drop and without radiating it. The same copper that solves the drop problem can create a loop that radiates if it is arranged badly.
The third difference is that the amplifier is a source of broadband noise. Switching edges contain energy from the audio band up to hundreds of megahertz, and the demodulation filter only removes the part above its own cutoff. What remains between the audio band and the filter corner is the part that shows up as an audible residue or as a conducted emission on the supply leads.
The Switching Loop and Its Area
The loop formed by the high side device, the low side device and the local decoupling capacitor carries the full switching current, and its area determines how much magnetic field it radiates and how much inductance it adds to the edges. Make that loop as small as the layout permits, with the capacitor directly across the two devices rather than somewhere else on the board.
The loop has to be traced on the same layer, right under the devices, rather than routed around the edge of the circuit. A connection that looks short on the schematic can be several centimetres long on the board, and at these edge rates that length behaves as an inductor that produces overshoot and ringing at every transition.
The output side is a second loop, this time formed by the switching node, the LC filter and the load. Keep the switching node copper small, because it is the noisiest node on the board and it couples capacitively to anything nearby, and run the filter inductor close to the output pins. The reasoning is the same as for a switching converter, and our guide to converter layout and routing covers the underlying rules.
Dead Time and Shoot-Through
Both output devices must never conduct at the same time. If they do, the supply is shorted through the two switches for the duration of the overlap, an event called shoot-through, and the resulting current spike can destroy the devices or at least trip the protection on every cycle. Dead time is the deliberate delay inserted between turning one device off and turning the other on.
Dead time solves shoot-through but creates a new error. During the delay both devices are off, so the output voltage is determined by the current direction in the inductor rather than by the modulator, and the resulting error is proportional to the dead time and to the load current. Too much dead time produces crossover distortion that is clearly visible in a distortion measurement.
Layout contributes to the problem in two ways. Inductance in the gate loop slows the switching edge and effectively increases the dead time, while inductance in the power loop extends the ringing after each edge, so the device is still conducting when the complementary device starts to turn on. Both are layout problems with the same solution: short gate loops, short power loops and short returns.
The Output LC Filter
The LC filter removes the switching component and leaves the audio signal. Its corner frequency is set well above the audio band and well below the switching frequency, typically at 30 to 50 kilohertz for a switching frequency of 300 to 500 kilohertz. A second order filter rolls off at 40 decibels per decade, which is usually enough to reduce the residual carrier below audibility.
Inductor choice is critical because it carries the full output current without saturating. A core that saturates at the peak output current stops behaving as an inductor, the filter corner moves, and the residual carrier rises abruptly at high volume. The saturation rating should be specified at the highest temperature the inductor will reach, which is well above ambient in a compact design.
Capacitor choice matters for the same reason. A filter capacitor with a lossy dielectric will dissipate power at the switching frequency and warm up, changing its value and shifting the response. The capacitor also presents a low impedance to the amplifier at high frequency, so the feedback network has to take the output from the correct side of the filter to avoid instability.
Supply Decoupling and Bulk Capacitance
The local decoupling capacitor carries the switching current, so it has to be a low equivalent series inductance type placed directly across the output stage. A large electrolytic cannot supply the edge current because its inductance is too high, so a ceramic capacitor of a few microfarads next to the devices does the work while the bulk capacitance further away supplies the lower frequency component.
The bulk capacitor value follows from the modulation. At low audio frequencies the amplifier draws a current that varies with the signal, and the supply has to source that without collapsing. A rough starting point is 1000 microfarads per 50 watts of output, distributed around the board rather than concentrated in one place, and with a small series resistance in the connection to damp the resonance with the ceramics.
Trace impedance between the bulk capacitance and the output stage matters more than the capacitance itself in many designs. A long, thin supply trace adds resistance and inductance that the local capacitor cannot compensate for, so the supply pins of the output stage should see a wide, short path to both the ceramic and the bulk capacitance.
Grounding and Thermal Design
Use a single ground reference for the power stage and return the speaker current on its own path, separate from the small signal ground until they meet at one point. Mixing the output current with the input reference produces distortion that varies with the programme material, which is one of the most frustrating faults to diagnose because it appears only at high volume.
Thermal design and layout are the same problem in a class D amplifier. The output devices dissipate only a small fraction of the output power, but that fraction is concentrated in a small package, so the copper under the device is the heat sink. Use as much copper as the board allows on the thermal pad and connect it to the ground plane with a grid of vias rather than a single connection.
A multilayer board is the natural home for this kind of design, because it allows a dedicated ground plane and a separate power plane close to the output stage. The advantages of that arrangement are described in multilayer board advantages, and class D amplifiers benefit from them as much as digital designs do.
Debugging EMI on a Class D Board
Start by measuring the switching node with a short ground lead probe and looking at the ringing. Overshoot beyond the supply rail and a slow decay indicate too much loop inductance, and the fix is to shorten the loop and improve the decoupling rather than to add components to the output. Many designs add an EMI filter to the output when the real problem is the switching loop.
If the emissions are conducted along the supply cable, the fix is usually a common mode choke in the supply line together with the local decoupling already discussed. Our article on EMI suppression design sets out the source and return path reasoning that makes the difference between a filter that works and one that merely moves the problem.
For radiated emissions, the speaker cable is often the antenna. A ferrite sleeve close to the connector, a twisted pair rather than parallel wires, and a short cable all reduce the radiated field. Testing with the final cable and the final enclosure is important, because an amplifier that passes on the bench with a short cable can fail in the product with a two metre lead attached.
Finally, check the residual carrier at the speaker terminals with an audio band filter in front of the scope. If the residual is large at low volume but small at high volume, the dead time is too long. If it is small at low volume and rises at high volume, the filter inductor is approaching saturation. Both symptoms point directly at a specific fix, and neither requires guesswork.
FAQ
How much dead time is correct? Enough to prevent shoot-through under worst case temperature and device variation, and no more. Many designs use 15 to 30 nanoseconds. Measure the distortion at low output levels, where the dead time error is proportionally largest, and reduce the setting until the distortion stops improving.
Can I use a ferrite bead instead of the LC filter? A ferrite bead raises the impedance at high frequency but does not provide the second order roll off that removes the carrier. It can supplement the filter but cannot replace the inductor.
Why does the amplifier run hot at idle? Usually because the switching frequency is high and the gate charge is large, so the drive losses dominate. Check the gate resistor value and the device gate charge before adding more copper.
