Please NOTE: This section on the MOSFET option for the Millett MAX diamond buffers has been superceded. Please refer to the MOSFET-MAX website for details on the new, JFET-mod MOSFET-MAX. This section will remain here for posterity, or in case the market's supply of JFETs runs out for the JFET-mod version.
Neil Rothschild - DIYForums user "NeilR" - was instrumental in vetting
the MAX's MOSFET output stage option. Forums user "bperboy" was the first to try a MOSFET MAX, but had run into problems. Neil's testing eventually verified a group of parts that will work. Based on this, bperboy was able to complete his version, too. The current MAX BOM was updated to reflect those changes and many MOSFET MAXes have been built since. The following is a narrative of Neil's efforts plus suggested tweaks.
-- All photos courtesy Neil Rothschild. --
NOTE: My own MAX is built with the MOSFET option per the standard BOM you should be very happy
with the sound. There is no need to deviate from that. These notes are provided for the tinkerers among us. If you
like to tinker, you may want to try some different variations and report your results and impressions back to the user
community. That's what DIY is all about! - NeilR
CeToole's Millett Hybrid Max is designed to be configured with either a Bi-polar Junction Transistor (BJT) Diamond
Buffer (DB) output stage or a MOSFET output stage (also a Diamond Buffer arrangement). Of the two options, the BJT Diamond
Buffer is a proven design used on Steinchen's dDB add-on buffer boards for the DrewD/N_Maher revMH Millett
and Tangent's PPL-inspired PPAV2.
For the MOSFET option, the MAX uses the same MOSFETs used on AMB's M3. However, unlike the M3, the MAX's MOSFET Diamond
Buffer is based on a candidate buffer circuit published by Amb, but never before used on a production DIY board. It is
similar to the DB configuration in the PPA/Steinchen circuit but uses a different and more complex biasing arrangement
than the M3.
Amb's MOSFET Diamond Buffer (shown at right) has some other important differences. In addition to the
obvious difference of the MOSFETs, these include:
different values for the current mirror (multiplier) resistors
different values for the current multiplier resistors
deletion of the signal diver stage (QB2 QB3 in the BJT Diamond Buffer
Rather than attempting to prove the circuit for the first time on the 2nd MAX proto board, I decided to
configure several breadboard circuits, instead. (Webmaster's Note: Again, Neil was
experienced enough to see the pitfalls of the un-tried MAX MOSFET circuit, and welcomed the challenge of
vetting the circuit - to the benefit of us all.)
I breadboarded several different circuits and part values in order to choose a specific starting configuration for my own
build. After some experimentation, I found that the major differences between configurations was the "spread" of the bias
settings. The MOSFET bias settings are much more sensitive, with a narrower spread of the trimmer settings, than the BJT version.
This became the focus of my tests. Interestingly, my breadboard prototype behaved as a PTC (Positive Temperature Coefficient)
design instead of an NTC (Negative Temperature Coefficient) design, as the M3 behaves. Luckily, my actual MAX Millett build
(Proto #2 board) behaves like an NTC. I never resolved that discrepancy, but it is probably attributable to some aspect of
my breadboard and heat simulation device. Because MOSFETs need to be biased much higher, they must dissipate much more power
and heat. Therefore, it's important that it operates as an NTC circuit. Otherwise, thermal runaway would
be a problem - especially in a small, un-vented case typical of the Hammond 1455N160x series.
The various circuits I tried generally required trimmer values greater than 500 ohms in order to provide a setting that resulted
in zero or near zero bias, which is required for initial configuration. Actual values that resulted in a bias of 106ma varied
from 171 to 647 ohms, allowing the change to the 1K trimmer specification.
I measured the swing in resistance that resulted
in a change of bias from 60 to 106mA; these values ranged from as little as 17 ohms (almost impossible to bias accurately) to
78 ohms. I favored configurations with larger bias swings because they are easier to set and most likely will be more stable
over time because they will be less affected by part drift with time and temperature.
There are several configuration options, each discussed in greater detail below:
Choice of JFET (QB1)
Choice of mirror resistors (RB4, RB5, RB6, RB7)
Choice of mirror transistors (QB4, QB5, QB6, QB7)
Inclusion of signal driver stage (QB2, QB3)
Choice of JFET
The BJT part PN4392 results in a much narrower trimmer swing of 17-27 ohms. The 2N5486 resulted in a broader swing and was my
part of choice for that reason alone.
Choice of resistors
Here you have two options:
1. Absolute values of the parts
2. The ratio of RB6-RB7 / RB4-RB5
RB6 and RB7 are complimentary pairs of emitter resistors that must be the same value. Same for RB4 and RB5. If RB6-7 and RB4-5
are the same values (e.g. 220 ohms) the circuit operates as a 1:1 mirror. If RB6-7 is double the value of RB4-5, the circuit acts
as a multiplier, with twice the current flowing through RB4-5 as RB6-7, per Ohm's Law.
Higher absolute values result in lower maximum output signal voltage swings. In my tests, where RB6-7 = 200 ohms, I measured
approximately 1V P-P greater maximum output swings, as measured on a Fluke 85 digital scopemeter.
This is an approximate value
because it is difficult to evaluate the precise clipping point of a sine wave though an amplifier circuit. I did achieve
16-16.5V P-P output with 220 ohm (RB6-RB7) mirrors, verses 17.5V P-P using 100 ohm (RB6-RB7) mirrors. Lower value mirror
resistors resulted in significantly narrower trimmer swings. Note that AMB's circuit uses 100/50 mirror resistors. I modeled
AMB's circuit exactly per schematic (including the 2N2904/3906 transistors) and measured a swing of only 24 ohms from 60-106ma.
I found that insufficient for comfortable biasing.
I recommended using mirror resistors in the order of 220/100 based on my experiments and the actual values I used are 220/107. I
felt that the "standard configuration" should be as easy to bias as possible. To date I have not had a chance to try 100/50 in
the actual circuit. My belief is that the 1-1.5V PP signal loss is not worth a temperamental bias. Further, for simplicity, my
breadboard circuit drove the buffer with an OPA134 opamp, not a tube. The 16+V P-P signal swing probably exceeds the capability
of the Millett voltage gain tubes so the entire issue of voltage swing may be a moot point but is worthy of future experimentation.
Choice of Mirror Transistors
AMB's circuit specified 2N3904/3906 small signal transistors for the current mirror (QB4, QB5, QB6 and QB7). The standard Millett
Max BOM specifies 2N5087/88. I tried both and saw little or no difference in biasing behavior. The choice of transistors is
probably similar to the choice of capacitors- a matter of personal taste. I went with the 2N5087/88 parts mainly to simplify the
BOM and many people consider these parts to have excellent sound. The mirror transistors are not directly in the signal path but
the QB2/QB3 drivers are directly in the path.
Inclusion of Signal Driving Stage
The standard BJT buffer requires a signal driver stage (QB2/QB3) as input to the output transistors (QB8/QB9). The MOSFETs do not.
Note that neither the M3 nor AMB's circuit includes these drivers. The drivers can be eliminated by simply jumpering the emitter
to base for each of these two parts, which connects the tube output directly to the junction of RB2/RB3.
The signal drivers do result in a higher (easier to bias) trimmer swing. I socketed my QB2/QB3 parts so that I could easily swap
the signal drivers in and out (Mouser part# 538-10-18-2031 fits here but fits few other TO-92 parts on the board due to space
limitations). You could also use SIP strips but TO-92 transistor leads don't fit tight enough for my own comfort level. I have
run the amp in both configurations. Both sound very good, and I don't hear any significant differences but you very well may.
Either configuration is a viable and workable build; the standard MOSFET BOM currently includes QB2/QB3, mainly because it is
easier to bias. Anyone interesting in tweaking their builds should socket these two parts and try both configurations. After
inserting or removing these parts, you MUST crank down the trimmers to avoid over-biasing the MOSFETs on the subsequent power up.
The bias will be a bit touchier but not unduly so.
Note: signal drivers (QB2/QB3) bypassed.
Note: signal drivers (QB2/QB3) in place.
I also socketed RB4, RB5, RB6 and RB7 for further experimentation, which has not been conducted on the actual build at this time.
Since this is the first group buy implementation of this muling of the traditional diamond buffer mirror/signal driver and MOSFET
outputs, this entire area is ripe for experimentation. Please report your results and sonic impressions!
file last changed:Sunday, September 7, 2008 12:34:54 PM
Please contact the MAX webmaster for questions about these web pages.