The first part of EEWorld’s two-part “virtual roundtable” discussion on Class D audio considers technical challenges and design trade-offs for engineers using Class D audio technology. Joining us for this virtual roundtable are: Joshua LeMaire (JL), Audio Systems Architect at Analog Devices; Steve Colino (SC), Vice President Strategic Technical Sales with Efficient Power Conversion; and Jens Tybo Jensen (JTJ), Head of Application Engineering for Class D Audio at Infineon Technologies.
JS: What is usually the biggest challenge engineers face when first using Class D?
JL: EMI and optimal layout.
JTJ: Setting the right specifications to get the most out of the amplifier; that is, separating the target design specs from the test specifications. Setting the design specifications should be about ensuring enough headroom and dynamic range for the amplifier to operate efficiently in its linear region giving sufficient dynamic range (i.e., before too much distortion and signal clipping sets in) to hit the required sound pressure level with the selected speaker driver load (i.e., with realistic use case conditions playing real music driving a real speaker). In contrast, the test specification should aim to ensure that the amplifier will operate safely without interruption under the use cases it can be subjected to (which may entail carefully selected stress tests).
SC: Going to Class D is going from analog to digital. Sorry, digital people, but in power, an analog engineer can do digital, so it is not the engineer that we have to convince. The difficulty in converting high-end companies to digital is the stigma. I did a demo at a high-end linear amplifier company. Whenever I do a demo, I stand behind the amplifier and observe the listeners. At one high-end customer, I saw somewhere between confusion to amazement in their eyes. At the end of the demo, I asked what they thought, and they told me that it was good. To be honest, the high end is not going to drive Steve’s retirement. Yes, while I appreciate the technology, we need to convince mainstream audio companies to take advantage of our technology. This is converting designers to move from silicon MOSFETs to GaN devices. Much of this has to do with changing from a traditional power device to one with RF capability. Fundamentally lower switching charge and eight bucks get you a latte if you don’t design your PCB for low inductance. EPC has developed and documented multiple techniques to maximize performance. EPC’s local FAEs are happy to review schematics and layout, as well as assist in assembling the wafer-level packages. The biggest challenge, as is the challenge in power conversion, is convincing customers’ management to invest in a redesign.
JS: What is the most misunderstood aspect of Class D technology and design? How does that translate into specific design challenges?
SC: I don’t think class D is misunderstood. I think that its advantages and disadvantages are well understood, but the understanding is based on old technology. We are already seeing some high end converting to class D with our technology, but that is not our primary target. What is most understood is that GaN can give cost-effective, high quality, high-density audio to the listener. The fundamental open-loop linearity benefits of a GaN class D amplifier are obvious if you take a look and listen.
JL: Filterless Class D doesn’t mean filterless. You still often use an EMI filter (such as a ferrite bead, common-mode choke, whatever) to pass radiated and conducted emissions testing. Also, people don’t think to minimize the capacitance at your switching outputs, causing your idle power to really creep up on you. And also for the very novice – Class D doesn’t imply Digital. Yet another thing – Class D’s are very efficient, but please do remember to calculate your power supply budgets!
JTJ: I would say that the most misunderstood aspect of Class D technology is the simplification that an audio input signal can be represented by a 1 kHz sinus, that speaker drivers have nominal and purely resistive impedances and that highest continuous (RMS) power (often at completely unrealistic THD+N levels for the use case) should be a design target. Going back to the previous question, the challenge for the designer is to carefully set target design specifications separate from the test specifications to get the most out of the Class D amplifier product of choice.
JS: Is there still a trade-off between PWM and PDM modulation techniques for Class D, or has one become dominant?
SC: PWM is required when an analog signal is used. PWM turns the analog into digital, which is the basis of class D amplification. A digital signal can drive the power stage directly with PDM. The advantage of an analog loop around the class D amplifier is that feedback can be used to compensate for non-linearity of the class D amplifier. A digital system is limited to injecting pre-distortion. GaN systems give much better analog systems due to requiring less feedback. GaN systems enable all digital systems because the open-loop linearity is so good (pre-distortion can only go so far. With many music sources being digital, we are seeing more all-digital systems. As vinyl sounds awesome, a good digital system can always have a good A/D on the front end.
PWM running on a fixed frequency is digital implementation friendly, easier to handle errors from finite switching times, and EMI emission characteristics. PDM is great for having higher loop gain hence achieving better audio performance, also has an advantage of reduced switching loss and voltage utilization ratio. A novel modulation techniques and output topologies that seem to provide an additional picture to this. I am here referring to Infineon MERUS™ multilevel Class D, which both provides higher signal density (due to the inherent frequency multiplication of the carrier frequency) and varying pulse with up to 5 levels of output. This topology can be combined with PWM or PDM and bring benefits to them.
JL: I think ADI might be the only silicon vendor using PDM modulation for Class D. Our core clock is operating around 6MHz, and we intelligently switch down to a more manageable <1MHz range, doing lots of cool stuff in between to maintain audiophile-grade sound quality and minimize real-world power consumption. We have a completely flat frequency response in the audio band, support extremely fast sample rates, and always win the listening test shoot-offs. In many ways, PDM is similar to DSD audio. PWM dominates; in particular, those 24V and above amplifiers requiring LC filters are all PWM.
JS: I understand that so-called multilevel Class D topologies have emerged, are they displacing conventional half-bridge topologies?
JL: The bridge configuration is a separate issue from multilevel. Half-bridge means you are connecting one switching output to a single-ended load connection. This single output would be filtered and effectively ground-centered at the load. In a BTL configuration, you have a differential connection to the load, doubling the voltage swing, quadrupling the power. Mostly you’ll see half-bridges used to power speakers with very long cable runs and BTL used when you can place the amplifier close to the driver and hopefully run it filterless (EMI filter only) if you can help it.
Multilevel means your power stage can switch at different levels – for example, with small signal inputs, you might be only switching at PVDD/2 to minimize your losses and keep your low-level power dissipation minimized. You don’t want to be switching crazy voltages and only driving 1/100th of your total power levels. There are other incarnations too. For example, you can adopt ternary-style modulation with lots of different ways of switching at the output, including introducing some common-mode switching (sometimes desired) so both outputs are kept at the same level, and there’s zero differential across the load (not burning power at idle input). That’s kind of oversimplifying it.
JTJ: Yes. Due to the distinct advantages, the multi-level topology is displacing conventional Class D topologies in an increasing range of applications. Note that the multi-level topology, which includes 4 smaller instead of 2 larger transistor elements per half bridge (at approximately same effective die area) is also applied in half-bridge configurations and combined (with a 90 degree phase shift) in full-bridge (BTL) and Parallel Bridge Tied Load (PBTL) topologies.
The advantages of multilevel class D topology include: Absence of switching activity at idle which entails much lower switching loss when there is no amplifier input (which is the case most of the time for most audio signals), inherent frequency multiplication of the multilevel topology (4 x frequency multiplication between carrier and output in a BTL configuration), increased dynamic properties of being able to dynamically create the required output voltage swing over more transistors in up to 5 output levels. In combination, this entails lower overall power dissipation, superior peak power to idle power ratio, fewer filter components requirements, smaller solution size, longer battery life, and better EMI performance.
SC: Multilevel topologies are used to overcome voltage limitations of semiconductor technology. If meeting the voltage requirement of a system with a particular technology becomes prohibitive, a multilevel topology becomes a good answer. Silicon MOSFET technology is good at low voltages, but switching losses become unwieldy as voltage increases. The question is whether a multilevel topology using one technology is “better” (sound quality, cost, size) of another technology. To be honest with you, we are pushing multilevel topologies in offline PFC and inverter applications. We don’t have the voltage capability to do single level, but our solution is still more efficient lower cost, and smaller. At this point, the comparison is to various high voltage topologies, and we still win. EPC’s technology is currently up to 200 V. This covers much of the audio requirement. We have not looked at multilevel amplifiers above this level, but it is something that I am sure would work well. The question is whether higher voltage technologies would work “better.” I am not sure what the answer is, but I am sure that within our voltage range, an EPC based class D amplifier will have better sound quality, be cheaper, and be smaller than a multilevel amplifier. Also, note that crossover distortion is multiplied with a multilevel topology due to having double dead-time periods per PM cycle.
JS: What are the pros and cons of so-called filterless Class D amplifier designs?
JTJ: One of the inherent benefits of multilevel amplifiers is that they exhibit less out of band noise and hence require less output filtering – also at higher peak output power levels and while operating with low idle power consumption. With shorter speaker wires, just a few inexpensive ferrite beads for EMI filtering is usually enough. This is in contrast to conventional Class D amplifiers, which are sometimes “tweaked” to also run filter-less but then only under a certain peak power level and in so-called BD modulation. With conventional Class D amplifiers, the designer has to make trade-off choices between output power level, overall power consumption (which can be lowered somewhat in so-called hybrid modulation mode), and filter requirements. With multilevel amplifiers, on the other hand, the designer does not need to make this trade-off but reap all benefits at once.
JL: Pros – The ability to eliminate the need for an LC filter is a huge advantage. Selecting an inductor with a high enough saturation current rating can be challenging, along with dealing with large passive components taking up more PCB area than the amplifier itself. The distortion induced from such filtering may deteriorate audio quality. The same can be said for the ferrite bead EMI filter in “filterless” designs but often to a lesser extent.
Filterless class D can implement clever switching techniques to minimize idle and low power losses. Also, much effort has been spent to reduce radiated emissions with techniques such as spread spectrum switching and edge rate control.
The Cons – They are only suitable for shorter speaker cable runs. If you need to run meters of speaker cable, you are going to want to use beefier filtering. Due to some of the switching methods incorporated in filterless designs – if you do need to resort to standard LC filtering, it may not be such a straightforward filter design, and you may encounter difficulties handling some of the switching behavior designed to optimize around the low power states.
JS: Thank you to our three Virtual Roundtable participants for their insights into Class D audio technology and design considerations! You might also be interested in reading, Class D Audio, Gallium-Nitride Versus Silicon – Virtual Roundtable (part 2 of 2)