It's time to start fresh, rethink things. Here's what PIMETA v2.0 looks like today:
(Schematic)
What will not be changing:
- Serpac H-65 compatible board size
- ALPS RK27 and RK097/8 support
- Thru-hole where possible
- Jung multiloop topology, with class A biased op-amp
What has changed:
- 4-layer, with near-complete ground plane! (Bottom layer, blue above, does have some traces running through it in the few cases where running them in the top layer isn't possible work.)
- Thicker traces in several places
- Switched from BUF634 to LMH6321. The availability question is forcing the issue, but these chips sound a lot better, even when you compare to BUF634s biased to draw the same current. There are downsides: requires SMT, can't get down to lower supply currents for long-running amps, may be tying ourselves into another chip that will go away.
- Main rail cap bank goes from rail to rail only, like on PPA. This puts more burden on the ground channel to maintain low distortion. On the plus side, it allows the ground channel more freedom to do this, without getting swamped. It also increases the effective size of the rail cap bank, even though I had to drop 1 of them to achieve other goals.
- Bypass caps at op-amp pins, with choice of either PCM-5 thru-hole or SMT 1206 style packages.
- More mounting holes. (Unfortunate consequence: the outer holes aren't 1.4" apart any more. To mount a crossfeed board, you have to use holes 1 and 3, or 2 and 4.)
- Front shoulders moved back, inner corners rounded, notch added in rear edge, and main cap bank moved in slightly to make board fit better in Serpac H-65 enclosure
- Output pads are no longer clustered. My philosophy of putting the pad nearest the last place in the circuit where the signal naturally exists was being violated in PIMETA v1.
- Several other pad sets are clustered better now, where it doesn't hurt. Allowing room for Molex KK 0.1" pitch connectors in these places.
- Both ALPS pots are centered now, not just RK097
- Dropped Panasonic EVJ-C footprint. I have seen hardly any of these being used since I started carrying the RK097s. If I could have kept it without difficulties in routing, I might have, but since it costs, it goes.
- PG pad placement more useful for ALPS RK27.
- Bigger C1s allowed. (Still jumpered by default.)
- Only one set of input and switch pads. (PIMETA v1.1 used separate ones for RK097 pot case, a big source of confusion.)
What's still up in the air:
- Keep the 4-layer design? This will double the board cost.
- Dropped R8. (Output resistor inside feedback loop.) I can't remember the last time I saw this used at all, or recommended adding it. Unless someone comes up with reasons it's necessary, backed by actual experience with PIMETA v1, it stays out. Note that it's still possible to add an output resistor outside the loop, just by putting one between each O pad and the output jack.
- C4s are no longer on the center line. I can't seem to work out a layout that allows this, without having worse consequences than mere esthetics.
- The current layer stack-up doesn't entirely make me happy. On the plus side, it keeps all traces on the outer layers so you can follow then without looking at CAD drawings. On the downside, it doesn't put ground between the power planes as makes intuitive sense, and I'd prefer V- to be on top to allow better heat sinking for the buffers. That suggests a stack-up of V-, GND, signal, V+, but then you're blind to the traces. Not sure which option makes more sense here...
- Bring back low-voltage LED cut-off? I haven't seen many people use it, so I dropped it. Conflicts with one of the new CCS options, too.
- Class A biasing CCS type. The current one has a number of problems:
-- requires buying lots of JFETs to get a particular current level, and each one has to be hand-tested
-- consequently, almost no one does this, requiring me to do the matching, which for efficiency reasons still has a lot of drawbacks
-- once you select the current, changing it requires going clear back through the process, and replacing the current CCS
-- some say it doesn't perform as well as other possibilities
So, I'm looking at other circuits:
LTspice says they have this dropout behavior:
(LTspice IV file)
Given how soggy the bipolar ones are compared to the JFET ones, you might be wondering why I'm even talking about this option, much less have bothered to do the layout for one of them. They have several merits:
- The first one uses the fact that you almost certainly want a power LED, using it as a voltage reference to set the bias point for the main CCS. That current is set by a single resistor, and then gets mirrored twice to tug on the left and right channels. This means the two channels' bias points are always matched, and can be changed with just one resistor, currently a trim pot with series resistance to set a maximum bias of around 10 mA. The main downside of this option is that there's a waste path through Q2, equal to the op-amp bias current: you multiply the bias current by 3, not 2. This is still better than a naive implementation of this CCS, which would have an LED per channel, and two waste paths, plus the need to bias both channels separately.
- The second one trades performance for less waste current by taking advantage of the fact that we typically bias op-amps with around the same amount of current as we drive LEDs. By using diodes instead of the power LED to bias the main CCS and moving the LED into that Q2 waste path, we make the Q2 path do real work, while allowing less current use in the main CCS biasing path. This is likely to outperform the PIMETA v1.1 current draw when comparing the random JFETs you can get from my shop now, as those tend toward higher currents than you might choose if you had the power. We also get constant LED brightness with batteries as a nice side effect. The main downside is that the LED Vf increases the dropout voltage. Note that in the graph above, the diode-biased CCS cuts off just a little above 3.6 V, that being the Vf of the LED I'm using in the simulation. This is a minor matter, though, given that the amp as a whole isn't likely to be able to get down to this level.
Comments, questions, complaints?
posted 02-08-2009 01:49 PM CST (US) 
