ESP Logo
 Elliott Sound Products Class-D Amplifiers 

Class-D Amplifiers - Part 2

Copyright © June 2022, Rod Elliott

HomeMain Index articlesArticles Index

Class-D amplifiers are now one of the most popular audio amps, and are used in a vast number of consumer products.  One of the reasons for this is that heatsink can be much smaller, and for low power the PCB often provides an adequate heatsink for normal listening.  On-line sellers offer a range of different boards, and many cost less than the parts used to build them.

Not all are usable though, with some being so bad that anyone who is serious about sound quality would be unable to listen to them.  This can be the case even when apparently identical parts are used.  Because Class-D amplifiers operate at very high switching frequencies (usually greater than 300kHz), a small error in the board layout can make a big difference to the end result.  I have a selection of Class-D amps that were purchased for evaluation and with this article in mind.

Some are very good, with low distortion and a flat frequency response, although many are load dependent and the high frequency response will change depending on the load impedance.  Consider that almost all speakers will have an impedance that's well above the rated/ nominal impedance at frequencies above 10kHz or so.  This can make the result something of a lottery with a Class-D amp that is highly load dependent.

I doubt that any of the Class-D ICs currently made are inherently 'bad'.  It's a foregone conclusion that some are better than others, but the primary cause of poor sound quality is the PCB layout.  Most 'linear' amplifiers are at least passably tolerant of board design, but if it's not done properly you may end up with twice as much distortion than you expected.  With a poor layout, it's easy for the distortion from a Class-D amp to be 10 times that of a good layout, even using the exact same parts.

If you're unsure about how these amps work, I suggest that you read Class-D (Part 1), which has been on the ESP site since 2005.  It's a contributed article, written by one of the owners of ColdAmp (based in Spain), but the company has since ceased business.  Importantly, it covers the operation of a 'standard' Class-D amplifier, but concentrates on fixed frequency types.  These are now in the minority, with variable rate switching being more common now.  These are often classified as using '1-bit' Sigma-Delta modulation.  A very common (and popular) IC is the IRS2092, and while it's a fairly early IC (introduced ca. 2007) it is still used in many Class-D designs.

One thing that's more than a little annoying is the insistence by many Class-D vendors on quoting output power at 10% THD (total distortion and noise).  A common way to claim the maximum power is to simply use the power supply voltage.  For example, if the supplies are ±40V, with zero losses the RMS voltage is 28.3V, so power is claimed to be 200W into 4Ω.  A more realistic figure is about 2dB less, or 160W, but that still assumes a regulated power supply that maintains the voltage under load.  In most cases, the 10% THD output power should be divided by two (-3dB) so the claimed 200W output is more realistically only 100W.  Some Class-D amps get very 'ragged' as the output voltage approaches the rail voltage(s), as evidenced by the screen-shot shown further below.

Some also quote the 1% THD power output, which is when the amplifier is (supposedly) just on the verge of clipping.  It's uncommon for most people to run an amp at full power, and one has to troll through the datasheet to find THD figures at realistic output power.  You can be sure that the figures quoted are for a PCB that's been very well designed, with first-rate parts used throughout and regulated supplies.  If you buy something from eBay or similar sites, you get what you get.  Sometimes quite alright, but other times a disaster.  I have examples of both.

I've not published a Class-D project, and after reading this article you'll know why.  Some of the parts are troublesome, with the output inductor being one.  The best performance can only be achieved by using SMD parts, which minimise stray inductance that causes problems with very high switching speeds.  The PCB has to be perfect, which often means several iterations to get it right.  Unless a constructor can get the exact parts specified, there's no guarantee that performance will be acceptable.  It becomes a minefield, where the smallest construction error can cause instantaneous failure, and it's simply not something I'm willing to try to support.

In the following descriptions and circuits, there are often multiple supply voltages referenced.  +VDD is the upper MOSFET's drain voltage and -VSS is the lower MOSFET's source voltage.  The PWM signal switches between the two - there is no intermediate state other than the dead-time, where both MOSFETs are turned off.  Additional supplies may be referred to by a number of different terms, but they are usually easy enough to identify.  Anything that includes 'A' (e.g. VDDA) means it's power for analogue circuitry (input stages, modulators, etc.).  There are no conventions, even from the same IC manufacturer, so where necessary they are explained in the description for each circuit presented.

1   Basic Principles

Class-D was invented by British scientist Alec Reeves in the 1950s [ 1 ].  Strictly speaking, he invented pulse-code modulation (PCM), the underlying principle of Class-D.  As with so many things we take for granted, PCM was developed for telephony, with the first patent taken out by Reeves in 1938 (using valve circuitry).  Class-D wasn't practical until the MOSFET was developed.  This 'new' device was presented in 1960, a year after its development.  The idea was proposed in 1926, but it was not possible to fabricate the device at that time.  The term 'Class-D' came about because it was the next letter in the alphabet, and we already had Class-A, B and C.  The 'D' does not mean digital, but that distinction has become blurred over time.  While some Class-D amplifiers may use digital processing internally, the operation is completely analogue.  For those Class-D amplifiers with digital inputs, after any internal signal processing there's a DAC (digital to analogue converter) before the power amplifier itself (for example, see the SSM2518 - Digital Input Stereo, 2 W, Class-D Audio Power Amplifier Data Sheet [Rev. B] datasheet).

fig 1.1
Figure 1.1 - Derivation Of PWM Signal From Audio And Reference

The 'standard' fixed-frequency PWM waveform is derived from the comparison of the input (audio) signal and a triangle (or sometimes a ramp) reference waveform.  This is the switching frequency, and it should not be less than ten times the highest audio frequency.  The two signals are applied to a comparator, which outputs a 'high' or 'low' voltage, depending on the relative amplitudes of the two inputs.  With no input, the output is a 1:1 squarewave, leaving a net output voltage of zero.  However, there is always some switching signal breakthrough, and in an ideal case it's a sinewave at the switching frequency.  An increased switching frequency makes the output filter less critical, but increases switching losses.  Low switching frequencies reduce switching losses but makes the output filter more difficult.  Most modern Class-D amps use a switching frequency of at least 300kHz.  In the case of self-oscillating types the switching frequency is generally highest at low input levels, and reduces as the amp approaches clipping.

The way a Class-D amplifier operates is fairly simple in principle, but getting it right is not.  The first commercially available Class-D amp was a kit made by Sinclair, designated X10, which was closely followed by the X20.  When the X10 was launched in ca. 1965, it was the first to use Class-D, but it had many problems.  The output stage used bipolar transistors that weren't fast enough to switch cleanly, and because there was no 'real' output filter it radiated harmonics of the switching waveform.  This caused radio frequency interference, and that (along with dubious quality semiconductors) caused its demise.  The X20 was no better, and while it was claimed to deliver 20W, that was simply not possible.  The kits disappeared very rapidly once the problems were discovered.  Many other manufacturers followed, but Class-D remained something of a niche product until the early 2000s.

fig 1.2
Figure 1.2 - Sinclair X20 Class-D Amplifier (ca. 1965)

The above is adapted from an original Sinclair circuit.  (Sir) Clive Sinclair was nothing if not modest (not!), hence the photo included in the schematic, which I retained because it's part of the legend.  The amp itself was a disaster, not only because of radiated RF interference, but also due to Clive's penchant for getting the cheapest transistors available at any one time.  That's why no transistor types are shown, because they were likely to change.  Note that the amp uses a negative supply voltage (not uncommon with germanium circuitry at the time).  However, it's very unlikely that germanium transistors were used in the X20.  A PCB photo I've seen puts TR11 and TR12 at odds with what's shown in the schematic, with TR12 being the TO-3 device, with no TO-66 transistor to be seen.

While many Class-D amps use PWM (pulse-width modulation), there are a couple of alternatives [ 2 ].  These include Sigma-Delta (ΣΔ aka Delta-Sigma) and so-called 1-bit modulation.  A variation sometimes seen is 'PDM' - pulse density modulation, where the number of pulses depends on the signal level.  Many new designs use a 'self-oscillating' converter, which solves some of the issues but introduces others.  When multiple Class-D amps are combined in a chassis, there is always the chance that the difference between oscillator frequencies causes audible whistles (sometimes referred to as 'birdies').  With designs using a fixed modulation frequency the oscillators in each amp can be connected to an external 'master' oscillator, and some ICs have clock synchronisation inputs and outputs.

The designs shown below are a combination of fixed and self-oscillating types.  Self-oscillating Class-D amps cannot use clock synchronisation, because there is no 'clock' as such.  Self-oscillating amplifiers generally have a switching frequency that changes with signal level.  The amount of variation depends on the design.  Using a modulated clock frequency reduces the radiated emissions (RF interference) because the interference is spread out, rather than concentrated at a single frequency.

One of the major claims is that Class-D amps are very efficient, but that requires some qualification.  When operated at (or near) full power, they are more efficient than Class-B (including Class-AB) designs, typically up to 90%.  However, at (say) one tenth power that may fall dramatically, depending on the quiescent current.  MOSFETs are incapable of instantaneous switching, and at low power the switching losses and operating current for the modulator become significant.  At very low power, they are usually no more efficient than a low quiescent current Class-B amplifier.  For home use, it's unusual to operate any amplifier at close to full power unless it's only a low-power design, but this also depends on the loudspeaker efficiency, the type of music and the listener's preferences.  A great deal depends on the design of course, so you need to look at efficiency graphs in the datasheet.

There are many things that must be considered in the design of a Class-D amplifier, most of which were ignored completely with the Sinclair designs.  While they were ground-breaking at the time, the required technology wasn't available to make them work well.  Most of the designs covered here are capable of distortion levels below 0.1%, which doesn't match most of the better Class-B (including Class-AB) amplifiers, but there are other designs (mainly proprietary) that achieve noise and distortion levels that rival anything else available.  Even the IRS2092 IC is easily capable of distortion well below 0.05% at any frequency, but the PCB layout has to be perfect.

It's important to understand that the power supply is more critical.  Where a Class-B amplifier's supply only has to supply current, the supply for Class-D both sources and sinks current.  If it's unable to sink (absorb) current from the amplifier, the supply voltage will increase (bus pumping).  This effect can be reduced by using large filter capacitors.  The supply still has to be able to provide the maximum peak current demanded by the load.  While the switching operation does reduce the supply current at lower output levels, at peak amplitude (at or near clipping) the supply must deliver V/R amps (assuming a resistive load).  A ±50V supply must be capable of delivering ±12.5A peaks, and if it can't, the amplifier will either clip prematurely or the power supply may shut down (if it's a switchmode type).

In many ways, it can be helpful to think of a Class-D amplifier as a '4-quadrant switchmode buck converter', with the instantaneous output voltage (and polarity) determined by the audio input.  '4-quadrant' simply means that the amp can supply and sink current of either polarity.  A 'conventional' amplifier is different, and the PSU only has to supply current, and any power returned from the (reactive) load is dissipated as heat in the output transistors.  Output device dissipation in a linear amp depends on the voltage across and the current through the output devices.  For a Class-D amp, MOSFET dissipation is a combination of switching losses and RDS-on (the MOSFET's 'on' resistance).

fig 1.3
Figure 1.3 - 'Contemporary' Class-D Amplifier With Bootstrap Circuit

One thing that you almost always see with Class-D amps is a bootstrap circuit.  This is used to provide a 'high-side' voltage that's greater than VDD (positive drain voltage).  Every design described in this article uses the bootstrap principle to enable the high-side MOSFET to be driven with a positive gate voltage.  It is possible to use a P-Channel MOSFET, but they invariably have lower specifications than the N-Channel 'equivalent'.  To provide optimum performance, almost all Class-D amplifiers use only N-Channel MOSFETs.  The principle of the bootstrap circuit is shown above.  The waveform also shows dead-time, exaggerated for clarity.  Dead-time is very important.  Too little and you get cross-conduction as both MOSFETs conduct at the same time, too much and you get high distortion.

When the output (VS) is low (either ground or -VSS, Q2 turned on), Cboot is charged via the high-speed diode (Dboot which is forward biased, with optional current limiting by Rboot).  When the output switches high (VDD), Dboot is reverse biased (no current flow), and the voltage held across Cboot is used to provide the upper MOSFET (Q1) with a gate voltage that's 12V greater than the +VDD supply voltage.  This extra voltage is necessary to switch the gate high, to 12V above the source ('VS', which is the output).  Bootstrapping is not needed for the lower MOSFET (Q2) because that's provided by the 12V supply referred to -VSS (VCC).

The bootstrap principle is not particularly intuitive, and you may need to sketch out the circuit and solve for the two output conditions (high and low).  The simplified circuit shown should allow it to make sense though, but the VS + +12V voltage is relative to VS (the output).  The voltage across Cboot is relatively constant at a little under 12V, but the voltage (VB) referred to GND varies from -38V to +62V.  In many cases, the value of Cboot appears to be much too small, but it only needs to supply current for a short duration as the upper MOSFET's gate capacitance is charged.  The current peak may only last for a few nanoseconds.

A few manufacturers have experimented with 'tri-level' Class-D, with a number of possible implementations.  The general idea is that the MOSFETs don't have to switch between the two supply rails, only between zero and positive/ negative as demanded.  There isn't a great deal of information on this scheme, but there are a number of patents that describe the principles.  I don't know of any commercial offerings, but Crown did release an amplifier it called 'Class-I' which uses "symmetrical interleaved PWM" (See White Paper).  I'm unaware of the current status of this, but given the wild claims and lack of any updated information it can probably be ignored until further notice.

Some of the terms used with Class-D can be perplexing at first.  The datasheets usually explain what everything means, but it can be hard to find.  The most common are as follows ...

PWMPulse Width Modulation, as shown in Fig 1.1
SESingle Ended.  Either using two supplies [+Ve and -Ve] or an output capacitor for single-supply amps.
BTLBridge-Tied-Load.  Two power amps with the load connected between the outputs.  The two amps operate in anti-phase (180° phase shift/ inversion).
Peak-peak output voltage swing is twice the supply voltage, so a 50V supply gives a 100V P-P output voltage (70V RMS)
PBTLParallel BTL.  Two amps are operated in parallel to double the available current.  Usually requires that the IC is designed to be paralleled.

There are also many different terms used to describe the supply voltage(s), along with any other voltages either generated by the IC or required for it to work.  A sample of these is shown in the following drawings, but other devices will often used different terminology for the same voltage, even from the same manufacturer.

One of the things that nearly all high-power Class-D use is a level shifter.  This translates a voltage in the normal operating range (typically around ± 5V) to a higher (or lower) voltage, which can be as much as 200V.  Manufacturers are very cagey about disclosing the details of the circuitry used, but it's not particularly difficult for low-speed circuits.  This changes when the IC is switching at 300kHz or more, especially since the rising and falling edges are so critical.  A displacement of just a few nanoseconds may cause the switching waveform to create shoot-through current if the two MOSFETs are turned on simultaneously.  Fortunately, this is all handled by the IC itself, and the user doesn't have to worry about it too much.

2   IRS2092

The IR (international Rectifier) IRS2092 has been around for a long time.  While it cannot be considered 'state of the art', with a well-designed PCB it works very well indeed.  It's not in the same league as some of the best examples around, but for low-frequency drivers in particular, it can match many of the other offerings.  One down-side is that it requires a separate regulator - it's not complex, but is a nuisance to include.  It also requires external MOSFETs, which are surprisingly critical.  Because the gate drive current is rather limited (+1A, -1.2A), you can't use nice big MOSFETs, as the maximum recommended gate charge (Qg) is only 40nC (nano-coulombs).  To put that into perspective, the (now) rather lowly IRF640 has a total gate charge of 63nC and an IRF540 has 94nC.

fig 2.1
Figure 2.1 - IRS2092 Amp Schematic (From IRAUDAMP7D Reference Design)

The circuit is deceptively simple.  Working out some of the resistor values is a minefield though, as there are interdependencies that make it a complex process.  The CSH and OCSET pins are used to program the current limiting.  The dead-time - a mandatory period where both MOSFETs are turned off - is also programmable.  Dead-time prevents 'shoot-through' current that would flow during the small period where both MOSFETs are (partially) conducting.  If the dead-time is too great distortion performance is seriously compromised, if too short, output stage failure is likely.

I don't propose to go through all of the options here, because the datasheet, application note [ 3, 4 ] and other published material (by IR) go into everything very thoroughly.  There's probably more information available for this IC than for any other, and I expect that's one of the reasons it's remained popular for so long.

Suitable MOSFETs are the IRF6645, with a gate charge of 14nC, rated for 100V and 25A (@ 25°C), allowing for supply voltages up to ~±45V.  Another is the dual IRFI4019, 13nC gate charge, 150V and 8.7A (@ 25°C), which can use supplies up to ±70V.  However, the limited current means that only high-impedance loads can be used with the maximum voltage (8Ω minimum).  As shown in Fig 1, 4Ω loads will be alright, but only if 'benign'.  If it's expected that the amp (as shown) will be driven hard into 4Ω, the supply voltage should be reduced.

Note that the IRS2092 is inverting by default, so the speaker should be wired with the 'positive' terminal grounded.  If two amplifiers are used, one should have a unity-gain inverting stage in front of one channel.  This places the two amplifiers in 'anti-phase', which minimises the 'bus pumping' effect.  This condition arises because the speaker load is reactive, and is made worse at low frequencies and/ or by inadequate power supply filter capacitance.  One or both supplies can have their voltage increased sufficiently to cause an 'OVP' (over-voltage protection) shutdown.  If properly configured, this will be activated before the voltage is high enough to cause MOSFET failure.

Bus-pumping is a potential issue with all Class-D amplifiers, and most stereo configurations will invert one channel.  This is shown with some of the other circuits seen in this article.  The IRS2092 reference designs (there are several) show additional circuitry, which is not needed for basic operation.  If it's not used, there's no over-temperature cutout, so heavy usage with low impedance loads can cause the output stage to fail.  IR has published a number of compete designs, including additional protective circuitry, and extensive measurements.  In most cases, it should be possible to get less than 0.05% distortion with an excellent PCB layout.  Unfortunately, many of the PCBs you can buy don't qualify.  One that I've tested has over 3% distortion even at modest output levels, which is completely unsatisfactory ... and very audible!

Another, using almost the exact same parts, has distortion that remains well below 0.1% at any level below clipping.  The PCB layout is only one factor though.  A miscalculated output inductor and (to a lesser extent) an inappropriate output filter capacitor can easily wreak havoc with the performance.  If the inductor saturates, distortion is increased dramatically.  The inductor also needs to have low resistance, otherwise it will get hot, the ferrite characteristics will change, and it wastes power.

fig 2.2
Figure 2.2 - IRS2092 Amp Performance (The Good)

The capture above is from an IRS2092-based amplifier, and distortion is below 0.1%.  The PCB appears to be well laid out, and it has decent-sized filter caps on board.  It both tests and sounds like any other amplifier.  There is a limitation in my workshop speaker system that precludes 'audiophile' comparisons, but I listened at various levels and didn't detect anything 'nasty'.  Overall, this is what I'd expect from a budget amp using the IRS2902 IC.  I have another that's better, but the capture above is indicative of what you should expect.  In these (and the next) traces, the violet trace is the distortion residual from my distortion meter, and the yellow trace is the audio.

fig 2.3
Figure 2.3 - IRS2092 Amp Performance (The Ugly)

In stark contrast is the Fig 2.3 capture.  This board also uses an IRS2092 IC and the same dual MOSFET, but the distortion is considerable, and very audible.  The majority of the parts are much the same, but the 'designer' chose to omit the gate resistors and any form of supply bypass.  The test was done after I'd added gate resistors and supply bypass caps, but it's still awful.  This is the difference between seemingly similar amp boards, where you'd normally expect them to be almost identical.  Note the ragged audio waveform, which is a give-away that all is not well.  The layout and component choices make all the difference!

fig 2.4
Figure 2.4 - IRS2092 Clipping Performance

The scope capture above shows what happens as a self-oscillating amp clips.  It's easy to see that the modulation frequency falls as the amp's output approaches the supply rails.  In 'full' clipping, the oscillation stops completely, which should come as no surprise.  As the modulation frequency falls, its amplitude increases because the output filter is less effective.  This is roughly the 10% figure that's often quoted for output power, and as you can see it's unacceptable as a 'figure of merit'.  The distortion trace isn't shown because my meter was unable to make sense of the waveform with its superimposed oscillator residuals.

IR (International Rectifier) probably has more detailed information on the design and implementation of Class-D amps than any other manufacturer.  A lot of it is fairly old now, but the documents published are very comprehensive.  Naturally, the emphasis is on IR devices throughout, but for an understanding there's nothing better that I've found.  If you do a search for 'classdtutorial.pdf' and 'classdtutorial606.pdf' you'll see what I'm talking about.  These documents go into a lot of detail about things you probably don't need, but they also cover the things you do need to know.

3   TDA8954 (NXP)

The NXP (Philips Semiconductors) TDA8954 is a popular IC, and it's theoretically capable of 120W output.  This is highly optimistic though, as the limit with ±30V supplies is 112W into 4Ω (120W is claimed into 2Ω).  In reality, expect around 100W at the most (4Ω).  These ICs are used in a wide variety of different configurations, including parallel BTL (two BTL amps in parallel for double the output voltage and current, resulting in up to 400W into 4Ω.  While it's claimed to be high efficiency (83%), it has relatively high quiescent dissipation at about 3W, and it runs warm at idle.  Many Class-B amps will be lower than that, but of course they will dissipate far more at any significant output power.

fig 3.1
Figure 3.1 - TDA8954 Power Amp Schematic

The schematic is adapted from the datasheet, and it's somewhat inscrutable because almost everything is internal.  While you can't see the internal functions, the basic diagram shown in Fig 1.3 is sufficiently generic that you can work out what happens internally.  The IC has differential inputs, but single-ended operation is obtained by grounding the inputs as shown above.  Note that the two channels are operated in 'inverse phase' to prevent bus-pumping.  This approach is common, and is seen with other examples as well.  If connected as BTL, the positive input of one channel is connected to the negative input of the other and vice versa.  The input signal can be balanced or single-ended.  It has the modulator, level-shifters and gate drive circuits shown in Fig 1.3, and adds thermal and overcurrent protection circuits, as well as the differential inputs and standby/mute functions.

In many cases, the 'Mute' and 'Standby' functions aren't necessary, in which the 'Mode' pin is simply pulled high (+5V).  The datasheet is quite extensive, and has many graphs of distortion, power dissipation, frequency response and anything else you may find interesting.  Because the IC is SMD only, it's expected that most of the support resistors and capacitors will be SMD as well.  The IC has a thermal pad on the top, so a heatsink is simply clamped onto the top of the package (with thermal compound of course).  It's performance is surprisingly good, as shown next.

fig 3.2
Figure 3.2 - TDA8954 Power Amp Performance

There's not much switching frequency residual, and the distortion residual shows no sign of harmonics.  That doesn't mean there aren't any of course, but my distortion meter gets 'confused' when there's a high-frequency present along with the audio.  The meter reading was below the minimum the meter can show reliably, but I used the same output voltage and load that was used for the two captures shown above.  Overall, this is a good result, and the sound quality seems to be very good (my workshop speaker systems are not true hi-fi though).

While one could certainly build an amplifier from scratch, the TDA8954 is listed as 'no longer manufactured', which makes things harder.  However, there are many complete amps from China that still use it.  Unfortunately, many ICs of this type have depressingly short production runs, and in some cases it's possible for the ICs to become unavailable before a PCB can be designed and manufactured by a hobbyist or 'small-scale' supplier.

4   TPA3251 (TI)

This IC (along with the next) is from TI (Texas Instruments), and is a single-supply BTL Class-D amplifier.  The recommended supply voltage (PVDD) is 36V, and it requires a separate 12V supply (VDD).  Almost everything needed for a Class-D amp is internal, but as you can see, there are many external support components.  These are predominantly capacitors for supply rails, bootstrap and input coupling.  The inputs can be used as balanced or unbalanced, with a 24k input impedance.  The DC voltage at each IC input pin is not specified, but I'd expect it to be 6V.  The datasheet shows the input capacitors as non-polarised (presumably ceramic), but electrolytic caps will probably have slightly lower distortion.  High-K ceramic capacitors have a considerable value variation with applied voltage and temperature.

The amp can drive 4Ω loads in BTL, with a 1% THD claimed output of 140W.  The claimed output power at 10% THD is 175W, but that's an unacceptable amount of distortion.  The THD at 1W is said to be 0.005%, and if that's achieved it's a very good result.  Unfortunately, I don't have a board using the IC to test, so I can only quote the datasheet figures.  While the datasheet claims that no supply sequencing is necessary, it also say that form minimum noise the '/RESET' pin should be pulled low for power-on and off.  The other supervisory pins indicate a 'FAULT' and clipping or over-temperature warning ('CLIP-OTW').  If the IC produces an over-temperature shutdown, a '/RESET' must be applied to enable operation.  The IC also has protection for over/ under voltage, overcurrent for both high and low-side MOSFETs.  The 'Mode' pins ('M1, 'M2' & 'M3') are shown for standard BTL operation.

fig 4.1
Figure 4.1 - TPA3251 BTL Power Amp Schematic

This IC can also be used in single-ended mode, but due to a DC offset of 1/2 PVDD (nominally 18V) the speakers must be coupled via capacitors.  The value depends on the impedance, but I wouldn't recommend anything less than 2,200µF (-3dB at 18Hz with a 4Ω load).  Another option is PBTL (parallel BTL), which couples the two outputs together in parallel to allow the load impedance to be as low as 2Ω.  IMO this is not useful in most cases, because the speaker leads have to be big to prevent significant power losses.  It's less of a problem for powered boxes, with the amplifier directly connected to the speakers in the enclosure.

5   TAS5630 (TI)

The TAS5630 is another IC from TI.  It has a higher power rating (higher supply voltage at up to 50V), and can be used in single-ended mode, BTL or PBTL.  The maximum output is claimed to be up to 480W (1% THD) into 2Ω when used in PBTL, or an output of 240W in BTL into 4Ω.  Rated distortion is 0.05% at 180W output (4Ω) or less.  There are many similarities with the TPA3251, but for reasons that I find somewhat mysterious, the pinouts are different.

fig 5.1
Figure 5.1 - TAS5630DKD BTL Power Amp Schematic

The schematic shown is for the TAS5630DKD (HSSOP package) version.  There's an alternative package (TAS5630PHD, HTQFP package) which has 64 pins.  I don't know which is the most common, but the schematic shown is still representative, although there are a few pins missing that are present on the 'PHD' version.

Input impedance is 33k, and the DC voltage at the input pins is 6V (estimated, as it's not disclosed).  The inputs have series resistors that are missing on the TPA3251 circuit, as are the 100pF capacitors which will reduce the amount of HF noise reaching the input pins.  It seems that the schematic and overall design were done by different people, with no reference to other ICs or schematics (within the same company), where you'd expect the designs to be almost identical.

Like the previous example, I don't have one of these to test, so I have to rely on the datasheet.  The supervisory pins ('/RESET', '/SD', '/OTW' and 'READY') should be self-explanatory.  Unlike the TPA3251, there's no clipping indication.  The 'Mode' pins ('M1, 'M2' & 'M3') are shown for standard BTL operation.  These pins (for both the TPA3251 and TAS5630 ICs) are used to select SE (single-ended), BTL or PBTL operation.

6   TPA6304 (TI)

The next drawing is yet another from TI, but this time it's a dedicated automotive IC, the TPA6304-Q1.  Automotive ICs are a very cost-competitive product, so the support parts are the minimum possible.  It's rated at 25W/ channel, but of course that's highly optimistic (1% THD claimed, but I'm doubtful).  Like more 'traditional' automotive power amps, each channel is BTL, but it does have the capability to use PBTL to drive lower impedance loads, down to 2Ω.

fig 6.1
Figure 6.1 - TPA6304 Automotive Quad BTL Power Amp Schematic

The real output power (at 13.8V) will be closer to ~18W/ channel at around 1% THD, and I include the following quote from the datasheet ...

The TPA6304-Q1 device is a four-channel analog input Class-D Burr-Brown audio amplifier that implements a 2.1 MHz PWM switching frequency that enables a cost optimized solution in a very small 2.7 cm² PCB size, high impedance single ended inputs and full operation down to 4.5 V for start/stop events.

The TPA6304-Q1 Class-D audio amplifier has an optimal design for use in entry level automotive head units that provide analog audio input signals as part of their system design.  The Class-D topology dramatically improves efficiency over traditional linear amplifier solutions.

The IC has extensive diagnostics built-in, and details can be obtained using the I²C interface.  Many aspects of the ICs operation can be changed as well, but there are presets (aka default) values that will work for most purposes.  Consider that the datasheet is 122 pages, so the amount of information is vast.  Not all of it is essential of course, but to get the maximum performance some degree of configuration (via the I²C bus) is essential.  Since it's designed for automotive applications, it's protected against transients up to 40V (typically caused by a 'load dump', when the electrical system disconnects a high current load).

7   TA2020 (Tripath)

I've included the Tripath TA2020 because for a while, everyone seemed to think it was the greatest thing since sliced bread.  Released in 1998, it even managed to be listed in the 'Chip Hall of Fame (by IEEE Spectrum), although their description was wrong, claiming a 50MHz sampling rate (it's not stated in the datasheet, but it's doubtful if it exceeded 400kHz).  The topology doesn't show it, but these ICs relied on a self-oscillating architecture that was sufficiently different at the time to be 'noteworthy'.  At one stage I had a pair of larger versions (4 x TA2022 in BTL if I recall) installed in a chassis that was intended to be used for high-power testing, but the first time it was called upon to 'do its duty' it promptly blew up.  At the time, I was testing subwoofer speakers, and while the power supply had very large filter caps, it apparently 'pumped' the supply voltage high enough to cause IC failure.

fig 7.1
Figure 7.1 - TA2020 Single-Ended Power Amp Schematic

The Tripath company filed for (US) Chapter 11 bankruptcy in 2007, only 9 years after the first patent was filed.  Their 'claim to fame' was the modulation technique (dubbed Class-T, but it's still technically Class-D).  Class-T was a registered trade mark, not a 'new' class of amplifier.  There was a great deal of hype for a few years, with many claims that they "sounded like valve (vacuum tube) amplifiers".  Having used one occasionally (until it blew up) I can safely say that that particular claim was just nonsense, but the myth persisted nonetheless.  As I recall, at sensible listening volumes in my workshop it sounded pretty much like any conventional Class-AB amp, which few other Class-D amps I tested could manage at the time.

The ICs were used in a number of commercial products, and although Cirrus Logic purchased the Tripath company, they never returned to production.  All that's left are a few recollections, which in all cases (including my own) should not be considered as particularly useful.  The TA2020 is rated for 20W into 4Ω.  There were a number of versions, with a fairly wide range of output powers, and these were available for some time after Tripath folded.  As near as I can tell, there is no longer any stock, but a few still pop up from time to time.

The specifications for most of the Tripath ICs are easily matched or beaten with other ICs now, so there's no need for anyone to try to get one.  Like any Class-D amp, the PCB layout is critical, as is the selection of the output inductor.  If you get either of these wrong, then the performance will be awful.  Unfortunately I can't provide any 'scope captures as I no longer have any Tripath boards.


This article is not intended as a series of projects, but is only intended to show some examples of current ICs.  The exception is the TDA8954 which is now obsolete but still readily available in complete PCBs available from China.  The amps are not intended to demonstrate 'state-of-the-art', but if the PCB is well designed and a high-quality output inductor is used, they will equal or surpass many Class-AB amplifiers.  There are other proprietary designs that one can purchase, but they are generally fairly expensive.  I expect that many readers will know about them, but I'm not in the habit of providing free advertising.

There's no doubt that Class-D has become mainstream, but there's also no doubt that some implementations are worse than useless.  One I tested has a major mistake in the PCB, and has an output Zobel resistor in series with the output on one channel.  Other errors include badly laid-out PCBs, the wrong type of inductor (causing saturation) and a multitude of other problems.  These errors will not be apparent until you've bought the board, so it's very much a case of 'buyer beware'.  Sellers on auction sites don't care if the product is crap, because people will buy it anyway.  It's not even possible to 'name and shame', because they just close the account and open a new one with a different name.  Fortunately, I had no expectations for the boards I bought, because it was in anticipation of writing this article.

In the midst of testing the amps I have to hand, I did a comparison with an early version of the Project 68 subwoofer amp.  It makes no pretense of being 'hi-fi', as it's intended for subwoofers, where the (tiny) amount of crossover distortion cannot be heard.  When compared to the 'good' Class-D amps it was marginally worse at very low volume, but it trounced the mediocre and 'ugly' (poorly designed and executed Class-D) with the greatest of ease.  Comparing the good Class-D amps with a low-distortion power amp revealed no audible differences on my workshop systems.  A more revealing loudspeaker may betray a difference in sound quality, but once the distortion is below 0.1% it's difficult to hear a difference - provided the frequency response is the same.  My aging ears don't work at 20kHz any more, so I rely on instruments which not only show any difference, they also quantify any difference that exists.  Hearing a 0.1dB difference isn't easy, but a measurement is precise.  The same applies for distortion of all kinds.


Note that most of the references are not linked directly, because manufacturers keep changing the location of reference material (for reasons I cannot fathom) and the links will break. 

  1. Class-D amplifier - Wikipedia
  2. Application Note AN-1138 IRS2092(S) Functional Description (International Rectifier)
  3. Class D Audio Amplifiers: What, Why, and How (Analog Devices)
  4. IRS2092 - Protected Digital Audio Amplifier (IR)
  5. IRAUDAMP7D Reference Design (IR)
  6. TDA8954 Datasheet (NXP)
  7. TPA3251 Datasheet (TI)
  8. TAS5630 Datasheet (TI)
  9. TA2020 Datasheet (Tripath Technology)


HomeMain Index articlesArticles Index
Copyright Notice. This article, including but not limited to all text and diagrams, is the intellectual property of Rod Elliott, and is © 2022.  Reproduction or re-publication by any means whatsoever, whether electronic, mechanical or electro-mechanical, is strictly prohibited under International Copyright laws.  The author (Rod Elliott) grants the reader the right to use this information for personal use only, and further allows that one (1) copy may be made for reference.  Commercial use is prohibited without express written authorisation from Rod Elliott.
Change Log:  Page published June 2022