Dear J, Markus, LF Group,

I would agree with all Markus says. The current-mode class D configuration
requires a high power supply source impedance at the output frequency -
hence the series choke and no bypass capacitor. Otherwise, you would be
trying to force a square-wave voltage to appear at the input of the
pi-section output filter. If all the components were "ideal", a square wave
voltage would require the filter capacitor voltage to change instantly from
the positive peak value to the negative peak value, requiring an infinitely
large current to flow for an infinitesimal period of time to discharge and
recharge the capacitor. Of course, this can't happen in the real world, and
in practice, if you try to do it, what you get are big current spikes that
cause a lot of heating in all the components involved, and something has to
give way.

With the series choke, the DC supply current is limited to a nearly constant
value during a cycle. This results in an approximately square current
waveform without big current spikes being delivered to the filter input,
with a voltage waveform that depends on the filter component values and
configuration. In an ideal class D circuit, in which the filter has an
infinite loaded Q, the voltage would be sinusoidal. In practice, the loaded
filter Q is usually 1 or less in this type of circuit, and the waveform is a
bit "lumpy", but still a reasonable approximation.

In practical circuits, the components, in particular the transformer, have
stray capacitances and inductances, which, when included in the circuit
diagram, mean the circuit isn't exactly a current-switching or voltage
switching circuit. The result of this is that there usually are both current
and voltage spikes in the waveforms, and multiple resonant frequencies that
give rise to "ringing" but these can be kept quite small with careful
design, and the addition of some RC damping, without reducing efficiency
significantly. The main culprit is the leakage inductance of the
transformer; making the coupling between windings as tight as possible
helps, by using foil windings, bifilar windings etc. Using only the minimum
winding inductance required to avoid excessive core heating helps too.

Some designs do have a pi-section capacitor input filter with a decoupling
capacitor at the centre tap. These can work if the leakage inductance is
high enough to limit the current to a reasonable level, but it is difficult
to achieve a specified value of leakage inductance, and this always seems to
result in unwanted resonances at low harmonics of the switching frequency.
Such circuits have large amplitude odd-harmonic currents, which increase
heating and reduce efficiency in an unpredictable way.

Regarding the spikes in the output waveform - any narrow (a microsecond or
less) spike should be very highly attenuated by the low-pass filter.
Assuming the filter components are OK, it is possible that what you are
seeing are switching transients that are being picked up by the scope probes
or input leads. If you connect the scope input to the ground point at the
filter output, and you still see spikes, this is what is happening. In this
case, don't worry about it too much!

Cheers, Jim Moritz
73 de M0BMU


-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of J. Allen
Sent: 31 August 2006 18:45
To: LF (RSGB)
Subject: LF: Why?

Thinkers,

What are the reasons for not bypassing the center tap on the push-pull amps 
apart from high bypass current?

There are spikes on the rising edge near the positive and negative peaks in 
the waveform of the output from the LPF from the amp I am working on, and 
bypassing the center tap to ground cleared up the spikes.  It also cleared 
up the bypass capacitor  :o).

J.