Digging a new hole to throw money into

Well, I guess there's always room for one more project...    I've caught the 51x bug and have started planning my entry into said territory.

First items on the list:  A few API-312s and a couple 1176-Ds.  I'm picking these because they're easy designs.  I'm not so worried about the build difficulty, but I have it in mind to customize them with the exact options I want -- so, no off-the-shelf kits for me.  This implies a few iterations of the design phase, and so the first order of business will be a power supply that is suitable for testing purposes.

To that end, I'm looking at putting together a small test harness, with the EDAC port backplane and XLR/TRS I/O for two cards (in order to handle builds with 2U width or stereo links).  The PSU rails will be limited to 150mA (16/24V) and 50mA (48V), with switchable current limiting at lower levels (e.g., 10, 25, 50, 100mA) for smoke testing, and a simple LED display of current consumed for a quick heads-up of what's going on.

Attached is a high level block diagram of what I have in mind.  The TI LM5002 is a configurable switching regulator IC capable of 0.5A peak at up to 75V.  The Linear Tech LT3471 is a dual-channel buck/boost/inverter that is good for +/-40V at a peak of 1.3A.  Both should be able to provide more than enough current to downstream linear regulators for the positive (LT channel A boost to ~25 or 26V), negative (LT channel B inverting to -25V) and phantom (TI boosting to ~50V) rails.  This makes is super easy to power the whole deal with a typical (though reasonably stout) wall wart of 12V or 20V (laptop) DC output.

The linear regulators will be designed around a P-MOSFET pass transistor, with a few op-amps in the feedback loop providing a constant voltage control scheme up to the selected current limit, and then constant current thereafter.  Current consumption will be displayed by a simple resistor ladder + comparators driving some LEDs.  I'm resisting the urge to drop in an ATtiny and reduce the parts count by a third, figuring some analog design might be nice for a change. 

OK, that's all for now.  Schemos and such to follow.  (Only get to post one file at a time.)

These faster (~1MHz) regulators are a step into new territory for me, as I'm used to working with mid-100s of KHz.  The latter are a little more forgiving in terms of compensation networks, while the higher speed stuff should do a better job regulating, and with smaller magnetics.  Converting +12 to -26V and +51V at somewhat substantial current is no trivial thing.  So, I've broken out the switching pre-regulators to their own board to minimize PCB and parts investment, with the assumption it may take a spin or two to get these right.

The linear regs will also be their own board at first, since I'm building those as discrete circuits vs. using a jellybean linear reg IC.  Ergo, further tweakage is expected.

Attached is an image of the OSH Park PCB rendering of the switching board.  It measures in at about 1.9x1.6" (48x41mm), and I'm hoping I can pull at least 30W out of it.  We'll see.

Sounds like a good approach - I'm building something similar for a project I'm working on. The advantage of having one external supply is significant - you can ignore finding an odd custom supply, and if you lose the main input, all outputs can go down at the same time, which is nice - no need to fry your circuit when one rail goes down.

One thing I'm trying to do is to sync all of the switchers so that they can't generate any unfortunate beat frequencies. Some switching regulators can accept a sync input, and that can be derived from one of the regulators as master, or in my case, from the micro controller which will be doing some other PWM stuff. If I understand correctly, the sync input controls the period, but not the duty cycle, so you actually have separate regulators, but they share a common period and avoid beat frequencies if they would otherwise run open loop at close frequencies to each other.

Your design using separate boards for the linear and switchers can provide good isolation too, especially if you use some common mode chokes and caps to connect the switched supplies to the linear board. It also makes it easier to put a shielding can around the switchers if you need more isolation.

Anyway, just a note of encouragement as I'm planning to do something very similar for a lot of similarly good reasons. Best of luck!

Good feedback - thanks!

The main regulator shares an internal clock for both its A and B outputs, which in this case provide the + and - master rails that are then fed to the 16/24V linear regulators.  Of course, the +48V rail has its own clock.  It's user-selectable by choice of resistor, and IIRC can be treated as a clock input as well, if one is so inclined.  In this case, I don't think it really matters.  I am open to suggestions if I'm neglected something, though. 

You touched on an important point with the synchronization of rails.  I'm trying to design this with the concept of sequencing in mind.  For e.g., the switching regs all require the +5V supply to provide an enable input, so that rail is generated via a standard linear regulator pulling directly from the main Vin.  It doesn't need to provide much current -- I think it was something like 50mA when I added it all up, so I wasn't concerned much with efficiency there.

The linear regs require +5V and -5V for the feedback op-amps and various reference voltages.  The -5V rail is generated by another linear IC, but it requires the inverting reg to generate the main negative rail first.  This is a rather extreme voltage drop (-26V to -5V), but again if memory serves, under worst case, the current requirements on this rail are about half that of the +5V rail, so I think it's still OK.  I may need to revisit the thermals if the PCB area doesn't dissipate enough heat.

By the time the switchers are up, and the main linear regs are starting to see voltage on their inputs, the +5V rail is already long since up and running, and the -5V rail should be following close behind.  I don't expect it to be any problem, as w/o the references and the op-amp power supplies, the pass transistors should never turn on.  (They're pulled up/down to their respective rails.)

This doesn't cover EVERY conceivable failure, but it does catch a few of them. 

I'll take this post as an opportunity to attach the schematic for this board.

As usual, there's not as much time to make progress on these projects as I would like.  But it's still moving forward, so here's a small update.

I got a couple PCBs in today.  The first is the switching PSU board from above (with a couple minor tweaks).

The second is a prototype simple mic pre channel with phantom power, and a comparator-based -30/-12/0/clip LED indicator.  The pre is an adaptation of ESP's Project 66.  I just want a simple design with reasonably good performance that I can put into an 8-ch 1RU box as part of my live rig.

I currently use a trio of Presonus Digimax FS 8-ch pre + D/As connected to a Firestudio Lightpipe (Firewire) interface box.  This is both my live and "studio" rig, which I'm hoping to separate and replace.  Live does not need D/A -- just some simple, portable, analog processing (EQ + compression and mixing) would be fine.

If the pre works out well, I'll be looking for a simple EQ module I can put into another 1 or 2RU box, and then something to replace my ACP88 and dbx 1066/1046 compressors.  Ideally, I would have per-channel processing, and be able to use simple DB-25 cables between them, instead of a ton of 8-ch TRS snakes.  Finally, I'm looking to build a mixer with 16 or 24 channels, a couple monitor sends, and four buses.  That would replace 2x Behringer RX1602 16-ch line mixers (which actually work well enough, but lack the sends and buses) and a Furman headphone distro box.  For the latter part, I'm looking at A-to-D and multiplexing over CAT5 to a stand-mount or maybe even belt pack receiver.

For recording, I'll be using various 500-series modules as a front-end, and my MOTU 24io for D/A until I can afford to replace it with 3x Metric Halo LIO-8s.  Might re-use the mixer design for cue sends as well.

Lots to do.