From: "James Moritz" <J.R.Moritz@herts.ac.uk>
To: rsgb_lf_group@blacksheep.org
Date: Wed, 12 May 2004 12:18:30 +0100
Message-ID: <DC9AE469E761D611BD420002A5EC855C7726F1@cerberus.herts.ac.uk>
MIME-Version: 1.0
Importance: Normal
In-Reply-To: <DC9AE469E761D611BD420002A5EC855C7721E0@cerberus.herts.ac.uk>
Subject: LF: RE: G0MRF transmitter problems- investigations (long!)
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 8bit
Sender: owner-rsgb_lf_group@blacksheep.org
Precedence: bulk
Reply-To: rsgb_lf_group@blacksheep.org

Dear LF Group,

I built one of the G0MRF TX boards a few years back, with the intention of
using it for /P operation, but other things have prevented me doing that
yet... Reading about PA4VHF's problems persuaded me to try running my board
at full output. I found I had similar problems to the ones Dick describes -
the DC supply current was rather high, and components got hot, especially
the output transformer. Increasing the DC input to more than about 100W led
to things getting excessively hot. My board is quite faithful to the
original G0MRF article, using the 2x 42mm 3C85 toroid cores, which made it
unlikely that Dick's LOPT core was the source of the trouble. Although these
salvaged cores are not an exact equivalent, the ferrite material and the
other core parameters are certainly very similar, so one can expect very
similar performance.

There are 2 basic types of class D amplifier, the voltage-fed type, where
the power supply is a low impedance voltage source and the load is series
resonant at the operating frequency (the Decca PAs are this type), and
current-fed type, where the supply is a current source (in practice fed
through a choke) and the load is parallel resonant at the operating
frequency. Ideally, the voltage-fed type has a square-wave voltage at the
MOSFET drains, and a sinusoidal drain current, while the current-fed type
has sinusoidal drain voltage and square-wave drain current.  Looking at the
circuit, David's design seems to be half way between the two; the supply
decoupling capacitor has a reactance comparable with the load impedance, so
the supply voltage to the transformer primary is by no means constant, but
neither is it a high impedance, constant current, so one would expect the
voltage and current waveforms to be some kind of distorted sine wave.
However, the low-pass output is very low loaded Q, and so is not really
series or parallel resonant... I wasn't sure what to expect!

Connecting up the scope showed actual waveforms that reminded me of those
bizarre spiky deep-sea fishes you see on TV nature programs, with very heavy
ringing with high frequency spikes and lower frequency undulations. Ringing
was occurring basically at 2 frequencies; roughly 300kHz and 1.6MHz, with
the ringing currents actually larger than the desired 136kHz current. This
explains the high losses in the transformer. Some experimentation showed
that the 1.6MHz resonance was due to a combination of the MOSFET output
capacitance and the leakage inductance of the transformer primaries. The
300kHz ringing was associated with the primary and secondary leakage
inductance, the input capacitor of the LPF, and the 470n "decoupling"
capacitor. The output waveform of the transformer was a totally different
shape, supporting the idea that leakage inductance was the main culprit for
the peculiar waveforms.

To improve circuit operation, I did 2 things. First, I wound a new output
transformer with lower leakage inductance. The primaries are 2 x 4 turns,
wound in bifilar fashion to minimize leakage inductance between them. The
secondary is 16 turns to give the same 4:1 ratio as before. The windings are
made as compact as possible rather than distributing them around the whole
core, with the primaries wound directly on top of the secondary to minimize
leakage flux. I used a salvaged line output transformer core of identical
dimensions to the one Dick describes. The new transformer has much lower
winding inductance than the original - the secondary is about 500uH instead
of a couple of mH of the original - but the inductance is still more than
adequate. The reduced leakage inductance shifts the 1.6MHz ringing to about
8MHz; I replaced the original 10nF capacitors between the MOSFET drains and
ground with a series combination of 5ohm/4.7nF, which largely damps this
out. I also modified the circuit so that it is definitely voltage-fed, by
replacing the power supply choke with a short circuit (decoupling the
transformer centre tap to ground with about 5uF total, and replacing the
pi-section output filter with a series-tuned tank circuit of 11.5nF/110uH.
The leakage inductance of the secondary is absorbed into the series tank
circuit inductance, which effectively kills the 300kHz resonance.

The new circuit has voltage and current waveforms which are much closer to
the "text book" voltage-fed class D, apart from some switching spikes where
the MOSFETs turn on and off. Running with 40V, 10A DC input, the efficiency
seems to be over 90%, and after 20 minutes key-down operation, the MOSFETs
and transformer were certainly warm but still acceptable temperature. The
damping resistors connected to the MOSFET drains dissipate a few watts each
I guess. So this seems to improve things greatly. There are still things to
be done; at the moment the loaded Q of the series tank circuit is only about
2 (the components were to hand...), so there is some visible distortion of
the output waveform with a resistive load, so I might increase the Q or
change the output network design. It should be possible to further reduce
the leakage inductance of the transformer primary by using metal strip
windings. It might be interesting to try a current-fed variant, although the
tank circuit component values will be more awkward. 

I hope this is enough to suggest a way forward, although I don't have time
to go into more detail at the moment - I hope to write up some more info on
this later, perhaps at the weekend. Obviously, the operation of the original
circuit will be very dependent on the exact transformer construction - I
think this may explain why some builders have had trouble, whilst others
have had good results from this type of circuit.

Cheers, Jim Moritz
73 de M0BMU