Time to start planning the final build

With most of my design work complete and measurement and listening tests of my prototype showing nice results, it's time to start planning the final build.  That means final part selection, wood frame prototyping and layout planning.  I've made probably 15 orders in the last 6 months from 6 different US vendors, 1 Chinese, and the rest on Ebay.  Lots of part swapping means my parts bins are brimming with parts to be used on future prototypes and projects.  My current prototype uses all of the components of my most resent design revision except for the power switch, jacks, etc. and it still uses carbon resistors which will be swapped with metal film.

For the wood frame, I've decided to use some Ipe (aka Brazilian Walnut or Iron Wood) boards which my coworker had left over from his very nice deck (Thanks Jeff!).  It's super dense, fairly unforgiving to work and it creates a nasty dust which I shouldn't have been breathing, but it's fricken beautiful wood.  The boards were roughly 6" wide by 3/4" thick.  I had a 2.5' board and a 1.5' board.  I ripped them (roughly) in half and had 4 2.5" strips.  Perfect!  I even justified a new carbide tipped ripping blade which did an amazing clean cut.

I wanted to finally try my hand at some nice half blind dove tail joints.  I've had a jig to do those for years.  I spent a many hours one weekend learning how to set it up and use it.  There's about a dozen adjustments and it's imprecise enough that the only way to really get a tight joint is by iterative trial an error.  Keeping everything aligned and square is the key.  Also logging measurements for each trial.  I had some hemlock blanks which had very similar dimensions to my new Ipe blanks.  I used those to hone my dovetailing skills and made a few small frames.  Here's a "half size" frame about 7" x 5" with a kerf (slot) for the aluminum top panel to slide into:

Driver Design and Fun with Distortion

My original *mono* prototypes had the 6B4G output tube biased using "Grounded Cathode biasing" or "Auto Biasing".  They included a 6SL7 driver tube configured in SRPP (Series Regulated Push-Pull).  The driver preformed pretty well: decent gain and distortion, but because SRPP has the cathode of one triode at half the HT voltage (125v in my case), I had to drive it's heater from a supply separate from the heaters of the output tubes or else I'd exceed the maximum heater-to-cathode voltage of the driver tube.  In the end, my power transformer had 2 heater windings, so I probably could have made it work by using one for the power tubes and one for the driver tubes, but I had already set off into the land of fixed bias for my output tubes with a simple 6SL7 common cathode driver (see my previous post).  This reduced my tube count to three (2 output tubes plus the 6SL7 with one triode to driver each channel). 

My first test with that configuration in October resulted in a much better overall sound.  The details were nice, gain was lower than expected but the distortion profile was generally more pleasing (~4.5% THD@1kHz, -27dB 2nd order, -35dB 3rd - see red spectrum graph below).

Still not outstanding, but better.  However, that was just one channel.  Unfortunately, duplicating the circuit to the second channel resulted in almost double the distortion, but more gain.  WTF!  Both channels were using identical components.  The only difference was the cathode bypass capacitor on the driver triode.  I had used a junk box generic 47uF cap on the original channel (the one with the good distortion) and a decent Nichicon VX series 47uF on the second channel.  Sure enough, swapping the caps pretty much swapped their distortion and gain profiles. So the crappy generic cap had some magic mojo for reducing distortion...

Well, kindof.  Turns out, it's value measured by my multimeter was about 0.02uF.  For a bypass cap, that's essentially not bypassing much other than very high frequencies.  It wasn't a magic cap, it was a blown cap which wasn't far from no cap at all.   Many designers believe that with a proper design, a bypass cap shouldn't really be necessary.  It increases gain, but also increases distortion.  It's used most everywhere because we generally don't want to "throw away" gain.  So to not include a bypass cap means that the lower gain has to be designed into the whole amp.  Fortunately, the 6SL7 driver I'm using has a high transconductance and yields about 25x gain even without a bypass cap.  Thus with an input signal of 1V RMS, I could expect about 25V RMS (~ 35V peak to peak) on the grid of the output tube which is just below the maximum for the operating point that I selected for my output tube.  Almost like the 6SL7 was born to drive the 6B4G!

Look Ma, no driver bypass caps!

Output Section Design

The output section includes the output tubes, the output transformers which drive the speakers, and the biasing circuitry.  The job of the output transformer is to take the high-voltage/low-current/high-impedance output of the output tube and convert it to a low-voltage/high-current/low-impedance output for driving the low impedance (4/8/16 ohms) speaker.  Early on, I settled on a pair of James output transformers, partly because they look awesome in their potted cans, but also because budget conscious Single Ended amp builders seemed to like them, they pair well to the 6B4G output tubes I chose and gave me an option of 3.5kOhm windings and 5k.

In my first mono prototypes, I biased the output tube the good old fashioned way, Grounded Cathode (aka Auto bias).   Because I didn't have enough 6.3v heater windings on my power transformer to heat two output tubes plus two driver tubes in a SRPP configuration, I decided I needed to be able to power both output tube heaters from one 6.3v winding.  That suggested that I should consider using Fixed biasing of the output tubes so that both cathodes could be tied to the same potential (ground).  Fixed biasing adds a negative DC potential to the grid of the output tubes.  That allows a positive AC voltage (the signal) to applied to the grid and still be at a lower potential than the cathode.  The main reason fixed biasing isn't often used is that it requires an extra power supply.  The supply draws almost nothing though (a small handful of microamps in my tests) so the only real design negative is the extra parts count and associated real-estate they will occupy.

I built a quick and dirty (yet remarkably clean) negative supply, reconfigured the output stage, and was amazed by my first listen.  The upper-mid distortion that had bugged me for weeks was almost gone.  When I looked at how other designs (like my Cary SE-1) implemented fixed bias, I was surprised that they usually use the same negative supply to bias both channels.  That seemed odd to me since that basically means they are tied together.  Granted there's 400k ohms or so of resistance between the two grids, but still the signal will take that path with a resulting crosstalk.  I tried shorting the Right channel of my SE-1 to ground and sure enough the Left channel bled to an audible level to the Right speaker.  Granted, it was maybe only 2 to 5% the volume of the Left channel, but still, that's substantial for "Hi-Fi".

I decided I could do better, and I did.  I searched online for hours for a small power transformer with dual primaries.   It's common on large trannies, but not small (5 to 12 volt) trannies.  My plan was to not connect the transformer to the main voltage, but to connect it backwards to a heater winding of my main power transformer keeping it as THE isolation unit and not having to worry about fusing two (or three) power trannies, etc.  So the voltage gets stepped down by the main power tranny from 110v to 5v, then back up to about 90v by the small biasing tranny.  Since it has dual primaries (which I'm using as secondaries), I get two -90v adjustable supplies, one for each channel - zero crosstalk introduced by the biasing.  This was the one idea in my whole design which was completely my own.  I'm sure someone has done it before, but still, it felt great to have that spark of a novel idea and avoid the need for a second transformer.  Below is what the final dual -90v supply looks like:

I needed about -90v so that I could slap a high resistance potentiometer on the end and have the resulting voltage divider produce my target biasing voltage (roughly -40v) somewhere in the middle of the potentiometer's range.  A fixed bias configuration really simplifies the output stage circuitry.  Below is my final output stage schematic with the bias supply applied where you see "-39 v".