Near disaster with PCB design

It isn't too bad actually--just an LMR23630 buck switcher (3 mm SON package), an inductor, and a couple of extra resistors so that the processor can dork the output voltage.

The Class-H bit is a variant on the one-transistor simulated inductor--it's a PNP with an inductor in its emitter rather than a resistor, and its base bypassed to the SMPS end of the inductor. (The switcher runs at 2.15 MHz, so I don't care that the low frequency ripple rejection is poor.)

The processor's DAC output controls an NPN current source that supplies base bias to the PNP. It does need hardware current limiting, but that's no big deal.

The voltage control, current control, and temperature control loops are all done in software, and the buck is servoed to keep the dissipation in the PNP down to something a SOT-23 can cope with (150 mW or so) without trashing the ripple rejection too badly. We can trade that off by changing the set point in software, which is nice.

Another approach is to let the simulated inductor do its own thing and do the voltage control to get the desired current, but that would at least quadruple the dissipation and degrade the ripple rejection on account of the low value resistor from base to collector.

The one in our ultraquiet diode laser product servos near room temperature, so it needs to heat as well as cool and also needs four-quadrant capability, which is a bit more complicated.

One approach to that is to use a couple of FETs to switch one end of the TEC between the supply and ground, and use a +- current source on the other end. That would be OK for the MPPC detector, because it doesn't dissipate much power and doesn't really need super accurate temperature control.

For the laser product, that approach is a nonstarter because the laser can dissipate up to a watt or so, and its output spectrum is very sensitive to temperature and bias current. _That_ Class-H has quite a few parts--a BJT current conveyor on one side, and an inverting voltage follower on the other.

It needs a lot more cooling power than heating, so we run the controls off +12V so that V_CE is minimized in the cooling direction. All of that, plus the fancy sub-Poissonian laser driver, power control, all sorts of circuit protection, processor, USB serial, and +5 -> +12V boost converter with two-pole cap multiplier, all fits on a board the size of a credit card, with a $40 BOM cost excluding the box. We think it's a pretty nifty device.

Cheers

Phil Hobbs

There are two categories of people: those who make backups and those who will be making backups.

Best regards, Piotr

So do I. Have a look at the sizes of the plugs - I am sure you are mixing something up here. Or try it, and you'll see that with a bit of enthusiasm, you can stuff a USB B plug into an Ethernet socket.

yep, I just tried, both USB-A and USB_B easily fit an ethernet socket

the other side of the common FTDI is a uart, easy to galvanic isolate that

What a jerk.

It sounds like you only work with 3 other people with a small code base, with infrequent changes if you haven't seen it fall apart yet.

Which isn't hard to trigger and not far in between.

That's pretty far from your original claim, which was that checking out an old commit destroyed subsequent work.

I get that you don't like git, but given that half the world uses it, and considering Torvalds' famously low tolerance for broken tools, I really don't understand your issue. Maybe there are fatal flaws that I haven't run into yet. Could you explain a little more fully?

Cheers

Phil Hobbs

Yes true if you have access to those pins, but I had a complete purchased instrument, which had "warranty void if removed" stickers on the box, and with just a USB socket on the instrument and a FTDI part inside it.

I wanted to locate it about 30 metres from a computer, which was a pain. If I had wanted to optically isolate it without opening the box, or control it from a microcontroller / PLC instead of a PC, those things are a pain too. A proper serial port on the box, (perhaps fitted with external, removable USB to serial dongle) would have been much easier. If higher speed were required, then ethernet is a better standard than USB.

The one fatal flaw of git is that the archives do not have a transparent data format. It is RCS/CVS any time for me. If the worst comes to the worst those files can be fixed manually, or I can write scripts for them. (I think RCS could be The Last Source Control if version 3.0 identified the files with consequitive numbers instead of names, and multiple file commits would be added.)

Groetjes Albert

s. (Git was written by and for the Linux kernel developers.)

git

flicts

on a

of

I used to manage version control software for a couple of teams. The softw are was not too hard to use and the concepts were straightforward. I tried using one of the open source VCS on my machine a few years ago and it was a disaster because the fundamentals were changed. Check in one file and th ey all get a new revision number! What is up with that??!!

The systems I used gave each file a separate revision history with incremen ting numbers for each file. Then we could establish baselines with any fil es and any revisions we chose. It seems like tools like github want to do things very differently.