In Part 1 of this article, I detailed the theoretical background to the design. This part will detail the practical build process of the circuit as well as the enclosure.
The idea of the circuit is to use an input buffer, a step up interstage transformer for gain, and an output buffer. The circuit used for both buffers is a modification of the so-called Diamond Buffer that utilizes a servo to control for output DC offset. Everything is DC coupled.
The first step in implementing the amplifier was to design a small PCB for a diamond buffer that would allow some experimentation. I designed the boards such that they could be built for either input (small signal) or output (large signal) use depending upon parts selection. The output transistors in the power buffer configuration are Sziklai (compound) pairs in part because I think they sound good, and in part because it made biasing the circuit a tad easier. These worked great, and they allowed for experimentation of different transformers, power supplies, etc.
I found with the initial breadboard version that both Cinemag (CMMI-5C) and Lundahl (LL1544a) microphone input transformers worked equally well in the interstage position. The sonic signature is slightly different owing both to different winding techniques and to different core material (the Cinemag is high nickel, the Lundahl is amorphous cobalt), but neither was “better” in any real sense. Further, I tried numerous power supplies: a switcher, a well regulated linear supply, a poorly regulated linear supply, and this also did not make a lot of difference, presumably owing to the high PSRR of the circuit.
Once this experiment was complete and I was sure the concept worked, I set about to designing a full PCB for an entire amplifier. For this, I chose to use the Lundahls both for space saving reasons (the Cinemags are slightly larger) but also because I had several pairs on hand that I was not going to use for anything else.
Additionally, while the PCB has positive and negative regulators on board (basic LM317/LM337 circuits) I also included spots for a power transformer plus rectifier or a modular switching supply. It should be noted that these switchers are not the horrible $5 switchers that come for free with a router, but they are actually quite high quality. And a 20W switcher uses almost exactly as much PCB space as a 20W power transformer.
Part of the impetus for this design was that I was trying to come up with an interesting amplifier that did not require a lot of careful parts matching. On designs like the DSHA-1, I go through piles of MOSFETs, Zener Diodes, etc. to find sets that are very closely matched. It is a time consuming and expensive process. My hope was that the self-balancing of the diamond buffer circuit would eliminate this need. I mention this because once the full amplifier PCB was assembled, I found that while everything was running as expected, it did not sound as good as the bread board version. By that I mean that it was a little harsh and edgy in a way that a lot of solid state based amps can be. This, it goes without saying, was quite disappointing. There was also a tiny bit of scratchy noise when turning the volume from 0 to on, which was a good hint as to what the problem was.
When I built the breadboard version, I built it without the volume pot. This was because my sources all have volume controls which made it unnecessary, and pots are expensive enough that being able to not use one in an experiment is a good thing. However, because the parts in the amp were not matched, the input buffer was drawing a tiny amount of current.
On a bipolar transistor (BJT) the base, which is where the input signal goes, draws a small amount of current. The amount of current is defined by the Hfe of the transistor where the amount of current from the collector to the emitter is Hfe times the current through the base. So if a transistor has an Hfe of 100, and a current source draws 100mA across it, the base will contribute 1mA. In the case of the diamond buffers in the Black Diamond, if the CCSes are perfectly matched, and if the Hfe’s of the input transistors are perfectly matched, then theoretically the input currents perfectly cancel out. However, since there was no matching done, this did not actually work. In reality, even if they were matched at some operating point and temperature, that would surely not last.
In this case, the Hfe’s are within about 10% of 600, and the CCS current is about 5mA, meaning that the imbalanced current draw of the base is tiny. However, it was enough to cause a tiny bit of harshness when dragged across the volume pot. This actually strikes me as important for two reasons. First, it would account for why stepped attenuators often seem to sound better than standard audio pots, but not in all cases. Nearly all DC coupled sources will have some offset, even if it is just a fraction of a mV. While this will not likely cause issues with a stepped attenuator made from discrete resistors, it does seem to be enough with a pot. Second, it accounts for why sometimes it does sound better to put coupling caps on the output of a source. And frankly may account for a certain amount of “system matching”.
At any rate, eliminating the DC offset is not that hard, but it does require more exotic parts than the $0.07 BJTs I initially used. I replaced the input transistors with JFETs which do not draw gate current. To this end I used the ubiquitous 2SK170/2SJ74 pair. However, I did not use carefully matched pairs. Instead, keeping with the umatched parts mantra of the design, I grabbed random ones out of a bin of similarly graded parts. I also needed to replace the emitter/source resistors since Vgs of the JFETs is smaller than Vbe of the BJTs. I was pleased to discover that within a certain grade of parts, Vgs is actually pretty consistent – consistent enough for this design anyway.
That small fix eliminated the harshness completely. I had initially been concerned that a small amount of current across the secondary of the interstage transformer might be saturating the core and thus contributing to the harshness, but that does not seem to be the case. Thus, the input side of the output buffers remained BJTs. The rule for these transformers is that they do not want to see any DC, but this seems to be relaxable to “very little DC”. A µV or so was OK in this case.
I decided to do the case a little differently from what I had done before. I recently joined a hacker space in Chicago which gave me access to a laser cutter. This gave me a unique tool to make a unique case.
Laser cutters are very good at cutting precise pieces out of acrylic, but they only cut in two dimensions. As cases are three dimensional this creates a bit of a challenge. I started with a pattern for side pieces with the idea that I would stack them up.
The acrylic is about 3mm thick. In order to center the volume pot on the front panel, I wanted an overall height of about 45mm. That should mean 15 layers, but inexactness in the actual thickness pushed this to 16. This also necessitated cutting the sides and measuring them before cutting the front and back to ensure the height of the front and back panels are correct.
I also wanted to do a more traditional wood and aluminum case using the same technique. Here I used the laser cutter to cut side pieces out of 3mm plywood which are glued together. I also made 2 modifications to the design. The first was to cut thin strips of wood that I glued around the top and bottom edges. I also cut hex holes rather than round ones in the center layers. This allowed me to embed a standoff inside the sides and thus use screws from the top and bottom to hold the case together. On the acrylic, there are long screws that go all the way through from the top to the bottom.
I also had aluminum panels made. Unfortunately, I did this a little out of order which meant that I needed to size the wood to the metal, and not the other way as I had done with the acrylic. Fortunately, this is possible. The acrylic sides could not be shimmed on a table saw, but the wood can. It is still off by a fraction of a millimeter, but that is easily fixed using a sander of some sort.
In the end, these amps came out pretty well. They sound great, they are very quiet, and they are unique in the headphone world so far as I am aware. The gain is about 4x which works well for most dynamic and orthodynamic headphones, and they can play plenty loud for just about any taste.
Both of the built amps are (or will shortly be) available from my odds and ends page for essentially the cost of the parts, as are a small number of extra circuit boards.