Current projects
PMP, a Poor Man's Power amplifier
Actually, it is not really a poor man's amplifier, rather a simplified version of the PGP amp. Nevertheless, it is targeted  
at the ultra high-end market. In terms of distortion, it's main competitor is Halcro.

In the following aspects, it differs from the PGP amp:

1. No NDFL stage.
2. A current feedback (CFB) input stage.
3. Double Transitional Miller compensation (DTMC)
4. Local feedback around the output stage instead of error correction (HEC)
5. A 'saturation sensor' on the drivers to ensure clean clipping instead of a local acting clamp on the VAS..
To do:
1. description front-end
2. gain and phase global NFB loop
3. gain and phase compensation NFB loop
4. distortion figures, THD, IMD
5. PSRR figures
6. DC servo loop
7. Over current protection.
Fig. 1 Block diagram, shownig feedback and compensating loops
Although an additional NDFL stage is capable of reducing the distortion considerably,  the PMP amp uses a so
called Transitional Miller Compensation (TMC) to achieve a comparable improvement. This is far simpler to
implement, just three additional components, R4, R5 and C3. However, it should be mentioned that this technique -
opposed to NDFL- only reduces the distortion of the output stage. As the latter is normally the biggest source of
distortion, this is not a serious limitation and a distortion reduction of say 15dB is quite feasible. Since the output
stage has given some gain (1.333x), node C2-C3 has to be tied to a resistive divider (R4-R5) in order to attenuate
the output voltage to the same level as seen at the VAS output.
For more information on TMC look here, here and here  (

The principle of TMC has been extended to the input side of the compensation loop (node R3,C1,C2), hence its
name Double Transitional Miller Compensation (DTMC). At audio frequencies, the Miller capacitor C2 'sees' not
only the OPS output (instead of the VAS output), but also the NFB node of the input stage (instead of the VAS input).
This arrangement reduces the distortion a little bit further. Also the drive requirements of the input stage is less
demanding, as most of the 'Miller current' goes through R3 in stead of C1. Notice that at frequencies far above the
audio spectrum (>1MHz), the compensation loop only encompasses the VAS, along the path of C1, C2 and C3.

Notice that in the third version,  TMC is also applied to the output stage.

To ease the implementation of the output stage, the technique of error correction (see Hawksford) is replaced by a
ordinary local feedback, as the latter don't need precision resistors. But that's not the only advantage, one can also
omit the boosted power supply for the front-end. Moreover, NFB also allows to give the output stage some voltage
gain, so the VAS, even with an additional cascode circuit, can now operate at a lower supply voltage, without
compromising the headroom. Also, this bit of local gain makes room for a voltage regulator between the main supply
rails and the front end supply lines.
There was a lengthy debate on the diyaudio forum about merits of HEC vs NFB, but I concluded that both
approached perform equally well in terms of distortion, speed and phase lag. If TMC is applied, then the distortion is
even lower and the NFB topology has definitely an advantage.
For more information on this topic look here, here and here.  (

Bootstrapped PSU
The power supply of the front end has been bootstrapped with a copy of the input signal. In doing so, the Early effect
of all the transistors of the input stage and current sources has been eliminated. So no additional cascode
transistors are needed. In the MCP and PCP version however, I've partly abandoned this approach and reverted
back to the traditional technique of cascoding. But for a totally different reason: here a low noise transistor array is
used that cannot withstand the full supply voltage.



Output Stage


Power Supply