Class D OEM product FAQ

From: "Saved by Internet Explorer 11" Subject: Class D OEM product FAQ Date: Mon, 20 Jan 2014 09:55:34 -0800 MIME-Version: 1.0 Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Content-Location: file://C:\Users\R&D 1\Documents\Research\Power Amplifiers\Class D Design Info\Class D OEM product FAQ.htm X-MimeOLE: Produced By Microsoft MimeOLE V6.1.7601.17609 Class D OEM product FAQ=20 =20 =20 =20

Class=20 D Product FAQ

Dear=20 Reader,

Here's version = "0" of a =20 class D product FAQ file. It was not my intention to give any = DIY =20 information on building amps (that will be another document) but = to give =20 some background information on the OEM class D market, for use = by=20 people who are looking to improve their audio power products = without=20 having to start their own research, as well as for plainly=20 interested folks (like myself). Someone had to stand = up and=20 distill the info from the noise. I hope you'll find it=20 useful!

Tip Voigt (tv@classd.org)

What = are the =20 reasons to move to class D ?

Class D amplifiers = are =20 chiefly chosen for their power efficiency and all subsequent = advantages:
small physical volume, low weight, low air-flow = =20 requirements, lessened power-supply requirements. All these = properties add=20 up to vastly expand the product concept and design freedom. = In a few=20 cases the high-tech appeal of a class D amplifier (then = called a=20 "digital amplifier") is a deciding factor. Also, some class = D=20 offerings exhibit sound quality otherwise attainable only = with much=20 more expensive means.
Contrary to popular belief, class = D=20 amplifiers should not be chosen for cost reasons. The most=20 cost-effective class D amplifiers only come close to the = price of=20 the most cost-effective linear amplifiers. This is mainly = because=20 heatsinks are not as expensive as they look.

What = kind of =20 amplifiers should be called "class D"?

In the light of the = varying =20 terminology and trade names concerning class D amplifiers, = here is the=20 best working definition of "a class D power amplifier" that = I've=20 come across:

"A power = amplifier =20 operating with all its power stage elements alternating between on = and off=20 states"

Note that this = definition =20 does not specify actual power stage topologies or modulation = methods. It=20 also explicitly excludes improved-efficiency amplifiers in = which a=20 linear power stage is embedded (Sunfire, Indigo BASH, = Himmelstoss et=20 al, ...). Through their relative obiquity they are known as = "poor=20 man's class D" not for financial reasons but because they = are=20 invariably proposed by people who are unable to build a good = class D=20 amp.

Is a = class D =20 amplifier a digital amplifier? When does it qualify as=20 one?

There is quite a = tradition of=20 attributing the word "digital" to any class D amplifier. = This is=20 because of the fact that the output devices in a class D = amplifier=20 are principally operated in the ON and OFF states. However: = =20 =20

they spend = between 1% and =20 10% of their time transitioning between those two states. = What=20 happens in this transition interval has a defining = impact on the=20 amplifier's performance. These errors are called = timing=20 errors and constitute very analogue = behaviour. =20 =20
the amplitude of = the =20 switched voltage is not fixed. Power supplies have a finite = output =20 impedance and tend to have ripple on them. Class D power = stages per=20 se have no power supply ripple rejection. Switch-induced = ringing=20 on the power supply and power stage similarly cause = amplitude=20 errors.

These=20 two error categories often make the intuitively apparent = (and often=20 quoted) linearity of the output stage worse than that of = properly=20 designed linear power stages, which can be very linear = indeed. If=20 half-decent performance is required from even a = well-designed power=20 stage it will inevitably require error control, mainly = feedback in=20 its many forms. This has to encompass monitoring the actual =20 non-ideal power stage waveform, making it an inherently analogue = process. =20 =20

A good definition = of a =20 digital power amplifier would be:

"A class D power = amplifier=20 where: =20

the switching = pattern =20 is a bit pattern =20
- generated = entirely =20 using digital means, independently of the output=20 stage =20
- that is in = its own =20 right a high-quality digital representation of the = wanted =20 signal
the power = stage is =20 designed with inherently minimal timing and amplitude errors = such as=20 to directly offer high-quality reproduction of the drive = signal" (i.e. exhibiting minimal analogue=20 nonidealities!)

Conversely, an = amplifier is =20 not digital if any of the following is true: =20

analogue = processing is =20 present =20
the amplifier is = not class=20 D =20
the input signal = is =20 analogue =20
feedback or = other error =20 correction is employed =20
large timing = and/or =20 amplitude errors are present

It is interesting = to find =20 that only one commercially available product (Tact Millennium) = and none=20 of the OEM offerings qualify by this definition. Those = designs=20 operating in open-loop with digital PWM but with normally = performing=20 power stages (hence offering poor performance) will be = called=20 digital-PWM amplifiers.

What = are good =20 reasons to choose for/against digital-PWM = amplifiers?

For: =20

A controller = built into an=20 already used DSP with only power components to be added = is the=20 cheapest way possible to achieve a power signal = reminescent of=20 an audio signal. If the controller is a separate chip = (equalling=20 extra component cost), as is currently the case with all = known=20 offerings, this argument becomes void. =20
Considerable = tech appeal =20 of the words "digital amplifier" on the box. As said = earlier, most=20 companies quite unashamedly use this nomenclature for = perfectly=20 analogue amplifiers as well.

Against:=20 =20

Obtaining = acceptable =20 performance (THD<0.1%, hardly enough to classify as a = digital =20 amplifier) is complicated and expensive. It requires an = enormously =20 well-regulated power supply and a blazingly fast switching = power=20 stage (expensive electronics, hard to tame EMI problems = in the=20 FM tuner range). Conversely, a power stage simple enough = to=20 realise the aforementioned low cost will have long = switching=20 times (100ns) and use an unregulated power supply. The = resulting=20 distortion will be well above 1% (even at low output = powers) and=20 the output signal will be modulated by the 100Hz ripple = on the=20 power supply. This will be very audible as = power-modulated hum=20 in the output signal.

Unless=20 such low performance is acceptable, a simple analogue class = D with=20 analog control plus a very cheap DA convertor (often offered = on-board DSPs as well) will provide much higher performance = at the=20 same cost. This means that the argument of = "digital all=20 the way improves signal integrity because of elimination of = DA=20 conversion" does not hold true.
The reason is = that the=20 power stage has extremely analog behavior, and it is easier = and more=20 economical to control nonideal behavior in the small-signal path = =20 instead of the large signal path, in the same way as loudspeaker = crossover=20 filtering is easier to do in the small-signal path (active). = =20 =20

Are = class D =20 amplifiers as reliable as conventional amplifiers?

There is no reason = why class =20 D amplifiers should be any less reliable than linear = amplifiers. Some=20 people have expressed wonder over whether they never = mis-switch.=20 They don't. Unlike software, hardware does not have a = propensity to=20 crash or hang up - unless when overheated... which takes us = to the=20 main topic:
The 3 most important factors dictating = reliability=20 in any electronic product are temperature, thermal = behaviour and heat management. I hope I have your = attention now. Class D amplifiers are used for their low = heat output=20 but this is not an excuse for improper heatsinking. In their = drive=20 to demonstrate the efficiency of their products many class D = vendors=20 underdesign their heatsinks or dispense with any altogether. = Yet, by=20 the laws of thermodynamics a very small quantity of heat in = a very=20 small heatsink translates in very high temperatures. As a = result=20 many class D executions are unable to sustain their rated output = =20 power for much longer than necessary to quickly measure it. This = is done =20 under the premise that real music never drives an amplifier = into rated=20 output power for longer than a few seconds. Unfortunately = the=20 ingenuity of the proverbial User in abusing their equipment = is=20 unlimited and many properly engineered products are known to = have=20 burned out in their hands.

For this and other = reasons =20 many countries (including the US) have regulation in place = requiring=20 audio products to be able to deliver their rated output = power for at=20 least 5 minutes. In terms of the physically small class D = amplifiers=20 this is equivalent to "indefinitely". Don't go for = less.

Why = is a ClassD =20 amplifier a potential EMI source ? What can be done about it? = Is it =20 difficult?

EMI is caused by = rapid =20 changes in voltages and currents. When the power stage is made to = =20 transition (a process taking anywhere between 1ns and 200ns) the = switch =20 outputs change across the entire power supply voltage and = (barring one =20 exception) the entire loudspeaker current is re-routed = through the=20 output stage. This is the root cause of EMI in class D = amplifiers=20 and it is inevitable as well. Contrary to popular belief, = the=20 voltage change (dV/dt) is not a big issue as long as = capacitively=20 coupled currents can be returned directly to the source = using=20 electrostatic shielding. High dI/dt values on the other hand = will=20 readily provoke magnetic radiation, which in the far field = turns=20 into electromagnetic waves. Working this EMI mode is done on = 3=20 fronts: =20

Minimise dI/dt = as much as =20 possible, i.e. relax switching speeds (if distortion is=20 controlled by feedback). Luckily the belief that high = efficiency=20 requires blazingly fast switching is quite incorrect. A = deep=20 understanding in MOSFET physics helps to determine = correct=20 operating parameters. =20
Puzzle your = lay-out =20 meticulously. It is not uncommon and totally acceptable to work = on a=20 new power stage lay-out (only the switches, drivers and = output=20 filter) for several weeks on end. After that, exercise = similar=20 care about the rest of the board lay-out. =
Be very careful = about =20 using standard "EMI remedies". Most work counterintuitively = and hence=20 tend to be counteractive (or have no effect at = best). =20 =20

back to top

What = is the =20 difference between meeting legal EMC requirements and being tuner = =20 compatible?

Legal EMI = requirements are =20 quite relaxed. The main idea is thou shalt not disturb thy = =20 neighbour, where the latter is presumed to be well outside = breathing =20 distance. To this end all equipment is required to sustain = full=20 specified operation at a certain irradiation level and that = it may=20 not produce radiation which at 3 metres distance would = exceed the=20 same level minus a 10-20dB safety margin. Legal EMI tests = should be=20 performed with the amplifier driven to 1/8 of its rated = output=20 power.

"Tuner = compatibility" is a =20 loosely used term to say that a [power amplifier] module inside = a=20 product does not disturb the operation of the AM/FM/TV tuner = inside=20 that same product.
The most important interference mode = is noise=20 radiated outward and back into the external antenna. This = mode can=20 be tested with a bare amplifier module and an external = tuner, the=20 requirement being that the effective sensitivity (antenna = level=20 required to obtain a specified SNR at the tuner output) is = not=20 significantly affected. Such problems can be solved only on = the=20 amplifier circuit itself.
Internal interference modes = are=20 conducted (via ground and power supplies) and radiated from = board to=20 board. Experience has shown that these can be worked either on = the=20 amplifier, the tuner or on the product architecture. A designer = =20 well-versed in EMI theory who has full control over the entire = =20 architecture and layout can make a tuner operate fine together = with a =20 perfectly foul amplifier. For real-world product designers this = is not an=20 option so in this context we should define "tuner = compatibility" as=20 being integratable by a normal product designer into a = fairly=20 standard architecture.
AM-interference can be avoided to = a large=20 extent when the switching frequency can be controlled and is = kept=20 away from the receiving frequency. In TV and FM bands only = good EMI=20 control will do and about 20dB extra margin should be = calculated=20 relative to the legal limits. The output power up to which = this=20 should be true is determined by the applicant. The same 1/8 = of rated=20 output power should be a healthy guess.

What = are the=20 pro's and con's of having class-D amplifiers with = digital=20 input interface ?

In the case of real = digital =20 amplifiers the interface will be digital by default. This = question =20 concerns analogue class D amplifier subcircuits or modules with = local=20 D/A converter. This option would be considered mainly for = cases=20 where disturbances on analogue input signals are expected. = Moving=20 the D/A converter away from the DSP section and onto the = amplifier=20 indeed excludes any interference on the analogue signals. = Care=20 should be taken before deciding for digital input though.=20 Transmitting the high-speed data and clock from the digital = circuit=20 to the amplifier section is likely to produce a fair amount = of=20 radiation. Coupling of data and other noise into the clock = will=20 cause jitter, making the option unsuitable for real = high-quality=20 audio.
Reducing the radiation would either require = slowing down=20 transitions (further degrading jitter performance) or moving = to=20 differential low-voltage signalling. The latter would probably = both =20 reduce radiation and maintain good jitter performance but is quite = =20 complicated and expensive. A more effective method seems to be to = transmit=20 the analogue signal together with its reference (D/A = converter local=20 ground) and use a simple op-amp+4 resistor circuit ("diff = amp" or =20 "translator") to translate the signal to the amplifier's local = =20 reference.

Scanning the =20 market for various high-efficiency amplifier solutions yields a = host of =20 names. What are the technologies behind them and what do they = =20 offer?

Sharp =20 1-bit

Over the past few = years Sharp=20 has been working the Japanese audio market with a type of = class D =20 amplifier called "1-bit amplifier", riding on the waves of the = 1-bit=20 audio (DSD-SACD) hype. Apparently they have gathered large = following=20 in the industry.
The 1-bit amplifier operates like a = 1-bit A/D=20 converter where the analogue input signal is summed with the = output=20 from the switching power stage to form an error value. The = error=20 value enters a high-order loop filter (in Sharp's case a 7th = order)=20 to form a correction value. The correction value is = quantised to 1=20 bit and sampled at a clock rate of 2.8224MHz, effectively = yielding a=20 1-bit data stream at this bit rate. This bit stream controls = the=20 power stage. The principal advantages of this technique are: = =20 =20

Very high = distortion =20 rejection of >120dB across the audio band owing to the = high-order =20 loop filter. =20
No loop gain = compensation =20 is required with power supply variations (a one-bit = quantiser has=20 undefined gain)

The=20 principal disadvantages are: =20

The undistorted = output =20 swing is hardly more than 50% of the power supply voltage, = limiting =20 available power to just over 25% of the normal maximum = (with=20 switching losses virtually the same as for maximum = power).=20 Larger signal levels cause progressive overload on the = loop=20 filter resulting in noise-like distortion. = =20
The high = switching rate =20 causes increased switching losses.

These=20 two problems add up to an efficiency which is only as good = as more=20 efficient class B power amplifiers. =20

The vendor also = quotes the =20 "spread spectrum" nature of the switching waveform and hence = the EMI =20 pattern. This is incorrect: 1-bit convertors (deltasigma = converters)=20 have a strong discrete frequcncy component idling at half = the=20 sampling rate and varying linearly (!) downward with = absolute=20 modulation index. The inordinate amount of shielding on the = vendor's=20 commercial products bears witness of remaining EMI = problems. =20 =20

This amplifier is = touted as a=20 digital amplifier for direct DSD amplification. It isn't, = for the =20 following reasons: =20

The input signal = is =20 analogue. Even when a DSD source is used, it is effectively = converted=20 to analogue using analogue lowpass filtering prior to = entering=20 the amplifier. =20
The control loop = is an =20 analogue feedback loop. =20
The "bit-stream" = is not an=20 accurate representation of the wanted signal. Instead, = by the=20 action of the feedback loop it contains the wanted = signal plus=20 "compensation distortion" which will cancel the = distortion=20 previously created in the power stage. =20
The power stage = is =20 operating at a high switching frequency, exacerbating the = relative =20 impact of the timing errors. =20
The power supply = is not =20 regulated.

A remaining note = concerns the=20 distortion. It would be reasonable to expect better than = -120dB=20 distortion figures owing to the control loop's gain (120dB). = In=20 reality, figures along the lines of -70...-80dB are found. One = =20 should wonder what the open-loop distortion of the system would = be... +40 =20 dB (10000%)?
(To save the reader from sleepless nights: = this =20 distortion is caused outside the control loop by the toroidal = inductors=20 in the output filter) =20

All-Digital:=20 TACT, TI, Pulsus and others

I have grouped = these names =20 together (and might have forgotten a handful) as they share a = common=20 take: open-loop amplification and fully digital PWM = generation. As=20 said earlier, TACT on their Millennium product have pulled = it off=20 very successfully indeed, with distortion figures = consistently below=20 0.02%. Not surprisingly, this is done by thorough work on = every=20 detail. The power supply is regulated and has an output = impedance of=20 roughly 4milliOhms. The dead time is less than 10ns. The = latter goes=20 at the expense of idle losses (and hence somewhat of = efficiency) and=20 definitely EMI but what the heck, they pulled the trick. = This=20 product also shows that getting good performance out of this = mode of=20 operation comes at a cost: complicated drive circuitry, very = tough=20 lay-out work, nonsaturating (i.e. huge) filter coils and a = regulated=20 power supply. All in all a lot of analogue design expertise. = A real=20 digital amplifier is not for everyone.
TI has purchased = Tact in=20 order to make this technology available as chips. Doing so = with the=20 PWM generator is a cinch, given Tact's fine VHDL work. Making = power=20 stage chips to deliver the kind of performance to make "digital" = =20 real is quite another matter. Worse still, the power supply will = most =20 certainly be of a garden variety and will spoil what's left of = decent =20 audio performance. Otherwise put, a digital PWM generator does = not make =20 for a good digital amplifier, if at all worth the = name...
The =20 digital PWM generator is the easy part of a digital = amplifier,=20 which goes a long way explaing why so many people are into=20 it.

Apogee =20 DDX

Apogee DDX is = another player =20 in the All-digital league but it commands a separate entry. = Not for=20 their intelligence alas... The DDX full-bridge power stage = is=20 operated in a kind of 3-level mode (class BD), but not in = bi-phase=20 PWM. The DDX power stage does offer the same efficiency = advantage as=20 other class BD implementations but it introduces a new, = previously=20 wholly nonexistent problem.
Around zero modulation = bi-phase PWM=20 produces two nearly 50% switching waveforms which are = practically=20 in-phase. The difference signal (as seen by the load) looks = like=20 extremely narrow spikes, even though they are not physically = created=20 as such (rising and falling edges of the narrow pulses are = on=20 alternate half bridges). Apogee DDX explicitly try to produce = these =20 very narrow impulses physically (having rising and falling edges = performed=20 by the same half bridge with the other half bridge static). = This=20 obviously a tough job, akin to modulating a normal PWM = system near=20 full scale. The digital noise shaper is therefore amended = with an=20 approximate model of the timing and amplitude errors = committed=20 trying to get those narrow spikes out. The resulting system = is=20 enormously complicated and does not perform any better than = a=20 straight digital amp without these problems.

Crystal =20 Semiconductors (Cirrus Logic)

At this stage = they're "just" =20 another runner in the full-digital race, with the first chips = offering hardly more than a digital PWM generator. Their act = is more=20 interesting than the others though, owing to their teaming = up with=20 International Rectifier to develop gate drivers and MOSFETs=20 specifically targeted at class D. According to = representatives,=20 Crystal realise that commercial products will in fact = require a more=20 tolerant amplifier concept than a plain open loop system. = Instead of=20 moving to analogue (simpler and more effective but far less = sexy)=20 they are working to implement correction measures for the = two chief=20 distortion mechanisms. This consists of monitoring the power = supply=20 voltage with an A/D converter and measuring the switching = delay=20 error, correcting for both in the digital domain. With due = care I=20 should expect such a design to perform in the -80dB THD region, = =20 which is indeed what they are claiming. While this system is = definitely =20 the closest thinkable cross between digital amplification and = analogue =20 realpolitik, there is more analogue processing to it = than to=20 a straight D/A converter plus analogue class D = amplifier.

Bang = &=20 Olufsen PowerHouse, ICE Power

B&O shun the = term class =20 D, instead calling their amplifiers Intelligent, Compact and=20 Efficient, conveniently acronymating as "Ice". Their most = important=20 products are the 250 and 500W "sandwich" modules, based on a = properly optimised full-bridge power stage operating in a=20 self-oscillating mode called Phase Shift Controlled PWM = (PSCPWM), in=20 which the idle frequency is that at which the combined phase = shift=20 of output filter, loop filter + compensation and power stage = delay=20 goes 360=B0. The loop filter is high-order and an on-board = zobel=20 network keeps the system stable under odd or no-load conditions. = =20 Feedback is taken after the output filter, resulting in low output = =20 impedance and good rejection of distortion caused by the toroidal = output =20 choke. Performance figures are very respectable (among the = best on the=20 market), though independent assessment rates the sound = performance=20 as less than high fidelity. EMI on the module is within = legal limits=20 but only just, potentially resulting in problems on the = product=20 level.
Other offerings and developments by this company = are=20 kilowatt power amplifiers operating in bi-phase mode (dubbed = class=20 BD) and pseudo-digital amplifiers (dubbed PEDEC). Class BD = reduces=20 switching frequency and hence switching losses by a factor = of 2 and=20 improves open-loop distortion by a factor of 2 as well. The = PWM=20 drive is based on a fixed clock frequency. This allows for = less=20 tight control over output impedance but otherwise performance = =20 figures are still very respectable.
PEDEC is based on a = digital PWM =20 generator and an analog feedback loop that corrects for timing = errors=20 by adjusting the edges of the incoming PWM.

Class T, =20 Tripath.

Tripath employs a = rather =20 strict definition of class D (namely a fixed-frequency PWM based = one) in =20 order to award themselves a new class name, class T. = Nonetheless, for=20 all intents and purposes it may be classified as class D = without=20 omission.
They were the first to deliver a serious chip = to the=20 class D scene, a 10W device targeted at multimedia. Another = product=20 is a driver module to build higher power amplifiers with. = Power=20 stages found in Tripath's products are half- and full-bridge = setups.=20 The control system is a 3rd order switched-cap loop = controller (the=20 first order forcibly converted to continuous-time to prevent = aliasing of switching errors and noise) with hysteresis = modulation=20 to create a time-quantised PWM (of variable repetition rate) = signal.=20 The time-quantisation implies that the PWM will have = quantisation=20 noise on it. For a third order loop (now a noise-shaper) to=20 sufficiently control this noise inside the audio band a high = sampling =20 rate (switching frequency) is required. Indeed, the (idle) = switching rate=20 is frighteningly high at 1.4MHz. The result is that even = Tripath are=20 reluctant to claim much better than 80% = efficiency.

Distortion figures = are good, =20 but are usually spoiled by the coils of the output filter = which is=20 outside the control loop. The figures inspire Tripath to = make claims=20 toward "audiophile" sound quality, but support of this = notion is not=20 unequivocal.

EMI of Tripath = products is =20 reportedly troublesome, with the company unwilling/unable to = offer =20 application support for EMI problems.

Jam, = Class =20 J

Technology start-up = Jam has =20 come up with a more unusual power stage approach, namely one = which can=20 produce voltages both below and above the supply voltage = (i.e. a =20 buck-boost converter). Although class D by the definition given = earlier=20 it is very different from the text-book power stage, = meriting the=20 company their new class name more than Tripath theirs. The = basic big=20 thing about it is that it'll put out higher voltages than = the power=20 supply rail, allowing high output powers into normal load = impedance=20 out of low supply voltages.
Unfortunately, while = innovative the=20 technology does not offer a practical breakthrough. Some=20 back-of-envelope work suggests that the number of switching=20 components remains identical to that of a boost converter = plus a=20 normal class D, with silicon area also identical.
The = company=20 prefers to operate the power stage with an open-loop digital = controller, but this is not a defining feature of Class = J.

Philips

Philips is a = fragmented =20 company and not surprisingly, has several class D products = developed =20 quasi-independently.

TDA89xx

These are two-chip = amplifiers=20 (one stereo modulator and one stereo 100W power stage) based = on more=20 or less text-book PWM. The loop filter is second order and = together=20 with the tightly switching power stage, offers good THD = results=20 (rising at higher frequencies). The output filter is outside = the=20 control loop. EMI results are among the best on the chip = market,=20 with Philips Semiconductor's application boards operating = very=20 comfortably below any conceivable legal limits (barring = automotive=20 maybe).
Power ratings are quite optimistically = specified,=20 requiring a bone-hard power supply delivering the chip's = absolute=20 maximum rating.

SODA

This is a discrete = circuit =20 with a half-bridge power stage operating in a hysteresis = switching =20 (self-oscillating) mode. Feedback is combined from before and = after the =20 output filter. Interestingly the entire modulator/feedback = chain is=20 built with only passive components directly connecting to = the=20 comparator. MOSFET drivers are a discrete affair and the = power stage=20 consists of one N-channel and one P-channel MOSFET. = Distortion is=20 commendable and virtually independent of frequency, making = THD at=20 10kHz among the lowest seen (0.02%). Output impedance is=20 significantly better than that of designs without feedback = take-off=20 after the output filter (the majority) but higher than the = B&O=20 modules.
In listening tests, the sound quality of the = design=20 lab's own 2x100W sample was consistently found to be quite = stunning,=20 sitting among devices whose cost per channel alone warrants = a=20 mortgage.
EMI is comfortably below the legal limit.
The = =20 circuit seems quite fiddly in optimisation so it would be best to = copy it =20 straight from the reference design and resist making changes = to it=20 for whatever reason. The technology can be licensed from = Philips but=20 given the above it would seem wisest to try and persuade = them to=20 make modules for you (the company not having announced any = formal=20 OEM module plans yet)

UCD

This circuit is = intended to =20 replace SODA in the short run. Somewhat immodestly called = "Ultimate=20 Class D", it sports milliohm scale output impedance (even at = 20kHz),=20 very good THD figures and a healthy efficiency that could = allow it=20 to operate without a heatsink in normal operation. Yet it = has one.=20 The circuit is a half bridge with two N-MOS this time and = fully=20 discrete drive circuitry (including the comparator).
The = modulation scheme is a Phase-Controlled self-oscillating = design with=20 feedback taken only after the output filter. Around this, = one or two=20 further orders of control are built. EMI results have yet to = emerge,=20 but supposedly the target is to improve on their previous = work. As=20 this is a discrete circuit even more than SODA, buying = modules is=20 advisable.

PP-A (Phased = Power =20 Analogue)

PP-A is a = no-holds-barred =20 design that uses a 4-phase (5-level) power stage driven by 6 = active and=20 2 passive orders of loop control (the 2 passive ones = apparently=20 accounting for the output filter). Combination of these 4 = switching=20 signals is done using a patented inductor-summation "tree" = built=20 around center-tapped coils as nodes. While the control of = the 4=20 power stages requires a large amount of digital logic, the = amplifier=20 is definitely analogue. To contrast this design with = Sharp's, it=20 uses the multilevel output stage to improve undistorted = modulation=20 index to 88% and to reduce switching frequency to a = manageable level=20 (around 200kHz, promising very good efficiency). Also, the = control=20 loop operates solely on the basis of the filter output, = promising=20 negligible output impedance and distortion. Nicer still, the = spectral behaviour of multilevel noise shapers is truly = spread-spectrum =20 with dominant tones virtually absent.
The concept suggests=20 ridiculously low distortion figures could be attained = (-130dB??),=20 but one should remain skeptical of the importance of such = low THD.=20
A discrete implementation is not for the financially = faint of=20 heart (i.e. only for the audiophile community), mainly = because of=20 the 4 output stages and the FPGA based logic block. A chip = version=20 is planned that should significantly slash costs while = retaining=20 most of the performance.

Trying to=20 measure distortion figures gives me nowhere the kind of = results that=20 vendors are specifying. Why is this?

You are probably = measuring =20 distortion and noise with the switching residual still present. = The=20 output filter of a class D amplifier is intended to reduce = the HF=20 switching noise such that it no longer produces significant=20 dissipation in the intended load, not to reduce it until it = vanishes=20 in the noise floor. When making measurements a measurement = filter=20 should be inserted that does attenuate the switching = component ad=20 oblivion. Normally the 22kHz or 30kHz filter fitted as = standard on=20 most audio analysers should do the job. A side-effect is = that THD=20 readings suddenly "improve" beyond 7kHz as the dominant = harmonic=20 (third) moves out of band. Extrapolating the readings is = generally=20 valid.

A minor snag on the = Audio =20 Precision analysers is that the autoranging circuitry on the = "reading" =20 detector (which sits after the filter) is set on the basis of = levels =20 measured before the filter, resulting in the distortion only = occupying=20 lowest few bits of the post-detector-A/D. This gives = typically=20 "jumpy" readings at low amplitudes (the instrument sometimes = failing=20 to put out a reading altogether). The only option is to = defeat the=20 autoranging feature.

How = can I infer =20 sound quality from measurement data?

The best way to = determine the=20 sound quality of an amplifier is to hook up a source and a = pair of=20 good speakers and listen. Some even go so far as to claim=20 measurements are uncorrelated with audible quality. The = reality is=20 "yes and no".
Yes, there are amplifiers with = excellent=20 figures that put out a perfectly foul sound. And yes, = there=20 are amplifiers with very mediocre figures that sound quite = fabulous.=20 But no, there are some important things to be learned = from=20 measurements.

Firstly, it is true = that =20 absolute distortion figures tell little about how an amp will = sound. The =20 behaviour of distortion versus frequency does.
Assume a = power=20 stage operated in "open loop" (this may be hypothetical, not = all=20 amplifiers can work in open loop at all). For such a power = stage,=20 the ideal condition would be the one which has the same = voltage=20 transfer all the time, regardless of the rate of change of = the=20 signal processed. Such an amplifier will have THD totally=20 independent of frequency. Typical examples are zero-feedback = triode=20 amplifiers but also some very carefully designed zero = [overall]=20 feedback solid-state amplifiers. Both categories are known = to offer=20 fabulous sound, with higher distortion levels only producing a = =20 certain "colouration" without affecting transparency.

An open-loop = amplifier that =20 produces frequency-dependent (ie. increasing with frequency)=20 distortion levels has underdesigned driver stages (in linear = amplifiers) or badly chosen output coils (in class D = amplifiers,=20 such as toroids or too small ferrites). Distortion typically = rises=20 at a rate of 6dB/oct. Amplifiers with such problems can = still attain=20 good (but not excellent) sound performance, but with varying = success=20 and dependent of the type of music used.

When feedback = (first order =20 assumed) is next applied, distortion products will be = attenuated better=20 at frequencies where loop gain is high (low frequencies = normally),=20 and will remain progressively higher at increasing = frequencies.=20
Amplifiers that had a flat distortion characteristic in = the open=20 loop will then exhibit a distortion characteristic rising at = 6dB/oct. Even though the distortion will have diminished, = the=20 amplifier will have acquired the glassy, sticky sound, which = seems=20 only to worsen as the dominant pole nears DC, and which is = commonly=20 known as "the negative feedback sound". Practice has shown = that it=20 is wiser to limit loop gain such that it does not rise any further = =20 below 20kHz. This indeed retains the "zero-feedback sound" while = =20 simultaneously improving significantly on neutrality.
Of = course, this =20 is bad news for specmanship. Usually distortion at 1kHz is = the most =20 proudly noted number on an amplifier's data sheet, whereas the = 10kHz =20 distortion (typically 20dB worse) is tucked away in a graph = that may=20 or may not have been reproduced.
Now, describing the = sound of an=20 amplifier with rising distortion in the open loop with extra = feedback added is left as an exercise to the = reader.

How = do class D =20 amplifiers sound?

OK, I admit. I = edited that =20 question. Class D amplifiers have a reputation for = unsatisfactory sound=20 quality. Presumably this is caused by a couple of very early = products that were released in the 70's and 80's and that = sounded=20 quite abysmal. The current state of technology is quite = different.=20 Practically any half-decent implementation of a simple class = D=20 amplifier executed using today's components sounds warm,=20 dynamic and musical, having the kind of direct = appeal=20 also found in vacuum tube amplifiers. It is this discovery = combined=20 with the previous bad reputation that has led most budding = class D=20 designers to conclude they had something wonderful and = "audiophile"=20 on their hands (and proclaim it in their flyers). = Unfortunately, an=20 appealing sound is just a fraction of what constitutes truly = audiophile-quality sound (neutrality and = transparency =20 to name a few). Of the class D amps auditioned so far only 2 = pulled it off=20 (incidentally the ones with a flat THD response).

Watt's behind=20 the power rating?

(Sorry, stupid pun) = The power=20 rating of finished and boxed amplifiers is already a = contentious=20 issue, let alone that of chips and modules. Let's have a = look at the=20 product case first:
One amplifier could have the = following power=20 ratings: =20

50W 1%, = 60Hz-20kHz =20 (IHF) =20
65W 10% THD = (EIAJ) =20 =20
80W Music Power = =20 (DIN) =20
130W Peak Power = =20 (Sales) =20
1300W PMPO (Yahu = =20 Wahu)

where: =20 =20

IHF, the = Institute =20 of High Fidelity tests rated power by first warming up = the=20 amplifier by letting it cook at 1/3 of the rated power = for an=20 hour and then testing, at rated power, for 5 minutes, if = the=20 amplifier is capable of delivering quoted distortion = over the=20 quoted test band (whereby strictly speaking THD is = measured by=20 harmonics only). This is the mandatory rating system for = products sold in the US (enforced by the FTC). =20
EIAJ = allows power =20 to be rated on a non-warmed-up device, right after = power-on. The=20 output power at 1kHz for 10% THD (heavily clipping), is = the=20 rated power. =20
Music = power tries =20 to fathom dynamic behaviour of the power supply by = measuring=20 average ("RMS") power on a single-cycle 1ms tone burst. = =20 =20
Peak = power is twice=20 the average power, assuming a non-clipped sinewave. Even = if=20 average power was measured at 10% THD this assumption is = made. =20
PMPO = ("Product =20 Manager's Power Output") is a nondescript item that allows = any number=20 to be attached to it.

Things are not that = sticky =20 when dealing with chip or module vendors. Still there are a few = caveats: =20

What power = supply voltage =20 is stated for rated output? All too often it is the = breakdown=20 voltage of the device. Real life power supplies = (especially=20 transformer-based ones) have outputs varying according = to load=20 and line conditions. Check very carefully what your = power supply=20 will deliver under nominal-line, full-load conditions = given that=20 it should never exceed the breakdown voltage at line = overvoltage=20 and no-load conditions. Derate the power according to = the square=20 of the ratio between those two. With chips and a = traditional=20 power supply you'll quite often end up cutting power rating = by=20 half. Switchmode power supplies normally have much better = regulation, =20 allowing higher output power to be delivered = reliably. =20 =20
What distortion = spec is =20 used to rate power at? =20
Can the = amplifier deliver =20 rated power long enough to pass the tests in vigour in = the areas=20 where you will sell your product? =20
back to top

I = asked my=20 vendor if he couldn't re-engineer his 250W/4 ohms power amp = for=20 250W/8 ohms and he swallowed uneasily. What's the=20 problem?

Well, 250W into 4 = ohms =20 requires a peak voltage of roughly 44V. With some margin this can = be done =20 with 55V or 60V MOSFETs. To get 250W into 8 ohms requires at = least=20 63V, which means 100V MOSFETs. The MOSFET market is not = driven by=20 class D amplifiers, but by computers and cars. This means = that the=20 low-voltage fets are usually far more evolved technically = than the=20 high voltage parts. It may well be that the move from a 60V = to a=20 100V FET pushes your vendor's existing design over the EMI = limits,=20 and increases switching losses by a factor of two. In = general, if=20 someone has a properly functioning unit with certain power = and load=20 specs it is often more economic to modify the application = than it is=20 to modify the amplifier module.
This is quite a = difference from=20 linear amplifiers where scaling is just a matter of changing = output=20 devices or adding some more.

Where does one =20 start to select an amplifier or vendor for a specific =20 application?

Get down to the = tough matters=20 first. In which aspect are you requiring extreme performance = from=20 your amplifier (or your vendor)? =20

EMI = =20
Sound = quality =20 =20
Efficiency =20
Power = =20
Tech = appeal =20
Cost = =20
Time to = market =20 =20
(etc). = =20

Start with = the one you=20 value most. Before you're 3 items down your list you will = have=20 pinpointed your vendor (and possibly the product = too). =20

=20 =20