General attributes: Single-ended input, derived balanced output. Bad part is any headphones need to have the 3 conductor jack replaced with a 4 conductor jack to be used with this design. Contains Jan Meier's crossfeed circuit, switchable on/off. Nice big ground plane on the bottom of the PCB, with virtually zero current flowing through the ground. Using CadSoft Eagle PCB Layout program with the freeware license. Thus any board is limited to 4" x 3" maximum. So my design had to fit physically. No way to do anything of decent complexity in that area without going to SMD. All of the parts I use are still sufficiently large to be solderable by normal (but still highly skilled) humans. Plan to use
www.custompcb.com to make the boards. They can take the CadSoft Eagle board file directly. No conversions required. Two boards with soldermask is $58. Or $78 with silkscreen too, which I don't care about enough to pay an extra $20.
Input section: One AD8620 input opamp for both channels. Assumes some external volume pot. Only there to provide consistent low source impedance to Jan Meier's crossfeed circuitry. I scaled the capacitance to 400nF based on SMD component sizes. A convenient size is SMD 2416 footprint. The most capacitance I could find with PPS film type in this size is 100nF. I have two placement pads per channel. To get the 400nF, I would have to stack two of them on each pad for a total of four per channel. You could not stack and use only two per channel, but you would have to multiply the two resistors in the crossfeed circuitry by two as well. Tried to keep those resistors as near to 1k as I could. I find typically resistances above roughly 1k add to the noise of the circuit, where below 1k, the opamp noise dominates. Also more current which is to be avoided. So 1k seems to be the ideal compromise between noise and current.
The next section is another AD8620 driving the OPA1632 feedbacks. They provide also provide a consistent low source impedance for sourcing/sinking current through the dual feedback loops. The audio signal only passes through one side of the AD8620. The other side is essentially buffering ground (to keep current off the ground).
The single-ended to balanced conversion section is controlled by the TI/BB OPA1632. This part take a differential input (between the single-ended input and ground) and produces a differential output (with gain). In this design, I have not connected the Vocm pin to anything. It control the center voltage around which the different outputs swing. It defaults to the midpoint of the power supply. Which is just what I wanted anyways. The enable pin is also not connected. It default to enabled. A little capacitance shorting the feedback loop is required for stability. Not terribly sure how much is needed here. In spice simulations, only 5 or 10pF is sufficient. The largest Mica SMD 0805 footprint I could find is 30pF, so that's what I put in the schematic. It has negligable impact on the frequency response for audio.
The output stage is five Intersil HA-5002s per phase per channel, for a grand total of 20 chips. The documentation for this chip says it has some stability issues with load capacitance between 25pF and 150pF. That should not be a problem, because most headphone cables of any length should exceed 150pF lump capacitance. Just to be sure, I use 1k source impedance to stabilize it as recommended by the data sheet. In all simulations, even with 50pF load capacitance (worse case), the circuit seems to be completely stable. Depends on how much you trust Beige Bag spice I guess. There is one odditity with the simulation. No matter what I do, there seems to be a tiny high megahertz oscillation in the buffer outputs and even power supply current inputs. Think this is an issue with the spice model. Even in the most possible ideal conditions, the buffer still does it. Like I said, think it's a spice issue. Maybe some kind of noise generator in the model. Anyone have any ideas on this point?
The gain is set to x2. I chose this value for my intended source and headphones. My DVD player outputs 2Vrms max. My headphones are AKG K701s. Max power is 200mW. A gain of 2x gives mv 4Vrms or roughly 260mW. Wanted a little headroom for those recordings which are pretty far below max output level, like Steve Winwood for example. For HD600/650s, I would recommend a gain of x5. If you are using a lower output source, like perhaps phono, you would need more gain. Replace the four 2k resistors with basically 1k times whatever gain you want.
Why so many output buffers? Let's do some math. My rails will be roughly 13.2Vdc each because of the lead-acid batteries. Using AKG K701s with an impedance of 62ohms, a 2Vrms source, and a gain of x2 represents a near worst case in terms of amplifier efficiency. The differential output is 4Vrms, but each phase is 2Vrms. At that voltage, the current through the headphones is roughly 65mA. The voltage drop across the active output transistors in the output buffers is 11.2V. That means the buffers are dissipating roughly 145mW each. The thermal resistance of each SO-8 SMD packaged buffer is 157C/W. With an ambient temp of 25C, each buffer would run at roughly 48C. Perfect. Audio DIY God Nelson Pass has said if you want your active components to last a while, don't let them go over 50C for any period of time. In the case of HD650s with an impedance of 300ohms, and a gain of x5, the power dissipation would only be 55mW each. Secondary reasons, more buffers appear to have less distortion, and it's the most I could fit on a maximum board size of 4" x 3".
I chose the AD8620 because it is a highly regarded chip, and I have had good luck with it in the past. Also because Analog Devices chips tend to be a little more bright and agressive, which I thought would be nice synergistic match with the TI/Burr Brown OPA1632, since Burr Brown chips tend to be a little mellow and laid back. The OPA1632 was chosen mostly because it does what I want (SE to balanced) and has incredibly good specs. Hope it sounds as good. The HA-5002s were chosen because they are a good replacement for the old EL2002s, which were highly regarded. It was in some article on these older chips were I saw more of them in parallel decreased distortion. Spice simulations agree.
The design is setup for a bipolar power supply, just to make sure the rails are equal. Otherwise there could be non-symmetrical clipping. But unlikely since even with relatively high impedance headphones, each phase only has to output half the differential voltage, which should be a good bit less than say +- 12Vdc. I intend to use an external (to the board) TLE rail splitter. There should be almost no ground current, so the little TLE will suffice. Plan to use a STEPS power supply setup for 26.4 volts to keep the circuit continuously warmed up. But when I want to listen, switch over to the two big lead-acid batteries. As for the TLE2426, I plan to use the 8pin dip version, since it has a noise reduction pin which I plan to use. The buffers will run straight from the rails. I want to use the N channel JFET transistor isolation technique (like the PPA designs) for the opamps. But one pair for all 3 AD8620s and 2 OPA1632s. Thus I need roughly 50mA Idss. Does anyone know a good part for this?
One improvement I can think of right away, it biasing the opamps into pure class A operation, but there just wasn't room on the board for this right now. Perhaps if I put the jfet cascode current sources on the bottom of the board it might fit. It would have been a lot more tricky to place and route the parts (which I did all by hand - no autoplace or autorouting).