Class D at low volume

I cannot speak of very high power class D, but my Rowland 102 producing 150W with my 6 ohm speakers is wonderful at low volume. In spite of hearing curve I don't feel any lack of lows or highs, resolution is there and image is even better than one at higher volumes (that might be related to some echo in my room). It behaves much better at low volume than my 50W Cambridge integrated, I had before.

Also, I hear much less noise/hiss while bringing my ear to the tweeter, in spite of higher power.

Class D (PWM) can be easily constructed without negative feedback at all. NGF improves linearity, bandwidth and output impedance like in any other class of amplification.

While global negative feedback in class A, AB is going over many stages of amplification, that create delay (thus producing TIM distortions), class D amplifiers have only one stage - a modulator that drive output switches (Mosfets). This modulator converts voltage to duty cycle of the output frequency. In simplest case it can be created using ramp generator and comparator but current modules resemble more of Delta-Sigma A/D converters. In short one analog quantity (voltage) is converted to another analog quantity (duty cycle) to end up with voltage again by obtaining average value of duty cycle. It is usually done by common mode choke and capacitors (Zobel Network), leaving about 1% of switching noise on the speaker wires. Amount of output power is controlled simply by setting amplitude of switched DC voltage. Frequency of this remaining noise is too low for the speaker cables to become antenna for electromagnetic coupling, but direct capacitive coupling is still possible.

As for the type of power supply - first generations of class D amp also has SMPS power supplies REF1000 being one example and my Rowland 102 being another. Ice power modules were available from the start in two different varieties - with or without SMPS. These switching supplies used to operate around 50-100kHz to preserve efficiency, but newer designed by Rowland run at 1MHz to make filtering easier (very difficult to design). I would say that good efficiency, very quiet SMPS operating at 1MHz able to deliver close to 1kW is a masterpiece.

One advantage of SMPS powered amplifier is often ability to operate at universal voltage and to tolerate any amount of DC. In fact my amplifier can be supplied by DC only up to 400V. Another advantage is regulation. SMPS have line and load regulation. Amplifier with SMPS keeps composure during power peaks since voltage does not sag, like it often happens in linear power supplies.

Bombaywalla, fig. 2 shows exactly how it can be done. For
any input voltage corresponding duty cycle is created.
Feedback would only improve performance. Modulator itself
might be not 100% linear but most likely linearity error is
created by the Mosfet switches. Their positive and negative
slew rates are different and vary a bit with the load
(creating timing errors). Negative feedback improves timing
but also lowers output impedance increased by the presence
of common mode choke. Early D class amps (tripath) had this
choke outside of the feedback and had loading problems with
many speakers. Icepower uses two different feedbacks. One
is from the output of modulator (improving timing), the
other from the speaker output. At least that's what I
remember from Karsten Nielsen doctorate.

Icepower uses single supply. Speaker (thru Zobel network)
is placed in the bridge made of 4 N-channel Mosfets. Bridge
is connected to GND and +Vs. At any given moment speaker is
always connected between +Vs and GND only direction changes.
Mosfets are turned on in pairs diagonally. At 50% duty
cycle average, filtered voltage on the speaker is zero. Half
of supply voltage is always present on the speaker's
terminals and the output cannot be bridged. Hypex used only
two Mosfets switching speaker between GND and +Vs or -Vs.
This allows output to be bridged.

As for the phase shift - 20 deg is a substantial phase shift
but I have no idea how much it (summing of harmonics) is
audible. Obviously Jeff Rowland wants no phase shift,
extending bandwidth to 350kHz.

I've read few times people saying "I know that class D
requires a lot of negative feedback" Now I hear that it
cannot even operate without feedback. How do they know
that??? If anything, class D requires less feedback since
voltage-duty cycle conversion seems to be more linear than
nonlinear bipolar transistors in class AB output stage.

SACD is an example of class D without any feedback.
Incoming 2.8MHz train of pulses is averaged and becomes a
sound.

Tim, GNFB serves the same role in class D as it does in class AB - reduces distortions, widens bandwidth, reduces output impedance. It is not necessary for operation in general but particular circuit can be designed to depend on it.

Since even shallow 20dB of GNFB will improve everything tenfold it is very tempting to use it. It can be done without negative effect if certain rules are obeyed but then GNFB will be shallow. Designer has to start with good design and quality components to start with and not to fix everything with GNFB.

As for new Benchmark amp - it is class AB with SMPS supply.
http://www.enjoythemusic.com/magazine/equipment/1214/Benchmark_Media_AHB2_Stereo_Power_Amplifier_Review.htm

Pavpet, yes TIM distortion can be fatiguing. It is basically an overshoot, in time domain, of the rapidly changing signal. It won't show with any THD or IMD measuring. In freqeuncy domain it translates to very audible odd order harmonics that our hearing is sensitive to. It makes sound bright. In extreme cases (perhaps some SS amps in 70s) overshoot was so big that it could choke output transistors that remained choked for a while (charge trapped at the junction) creating short gaps in music. These gaps were not audible, since our brain compensates for that, but made listeners very tired.

Bombaywalla, Linear power supplies have regulation in preamps, DACs etc. but MOST of power amps are unregulated because of the amount of heat dissipated in power supply (and loss of efficiency).

AFAIK older REF1000 and newer REF1000M use the same module, but REF1000M has additional separate power factor correction module.

Look at fig. 2 showing how PWM signal can be obtained by just using linear ramp. Icepower modulator uses sinewave, has two feedbacks etc. - but only for improvements.
http://en.wikipedia.org/wiki/Pulse-width_modulation

One limiting factor in class D is bandwidth. Latest Rowland class AB amps have bandwidth of 350kHz while newest class D model 925 50kHz. My small 102 extends to 60kHz. It is limited by the speed of Mosfets currently available especially high current devices, but it is improving every single year. We definitely cannot hear 350kHz but wide bandwidth prevents phase shift thus preserving harmonics within audible band. My amplifier with bandwidth of 60kHz shifts phase of the 20kHz signal by 20 degree.

The difference in sound, with class D, I observed is that sound is more immediate, faster but at the same time there is less of brightness in high frequencies. Cymbals sound less "splashy" but more brassy with more body. Perhaps it is related to lower TIM distortions or it is just the nature of the beast. It is definitely different but I like it.