Quote:
Well, I already proposed that a device could have two overt addresses. Besides MISO, MOSI, CS, CLK, and IRQ, there would also be a CONFIG pin. The "configuration engine" is selected when (CS AND CONFIG) is true. The "peripheral proper" is selected when (CS AND NOT CONFIG) is true.
It means I'll have to add a gate to the simplest devices so they won't be selected when CONFIG is true, but I can live with that. I'd sure like to use a different scheme so the extra gate (a transparent latch) would go on the smarter devices instead of the simplest ones, but I guess I can't really justify it. Here's how it would work: The configuration selection involves asserting the appropriate SEL line, then asserting CONFIG, then negating the SEL line, using CONFIG to keep an intelligent device's SEL status "latched" until CONFIG is negated. That keeps the simplest dumb devices from being required to have the gate, because when they're selected, the clock will not get toggled until after their SEL line goes false again.
I guess the sub-buses' CONFIG would get set through commands to the hub.
Quote:
The disadvantage is that it consumes "yet another pin" on the serial port
So are we back to a slightly bigger connector, like a 20-pin IDC? That still beats a 26-pin. If there's no other way, I'm willing to go for it to get the maximum range of usefulness. I was considering a couple of things:
--> Five SELect lines may not be enough if we forfeit another one for the CONFIG line. The bigger connector will let us get from six back up to seven and still get the CONFIG, while pairing a couple more lines with grounds (or virtual grounds).
--> Similar situation with the -12V that I sometimes need. If it were derived at the device, the easiest way to get a negative voltage from a positive one is with an inductorless switched-capacitor charge-pump voltage converter. Very few can handle that kind of voltage, but the Linear Technology LT1054 can go to 15V input and deliver enough current for what we've talked about. Still, putting one converter on the master is easier than having to put one in every device that needs it.
Thanks for pointing out the MAX765, as I may use that for something else; but there are a couple of reasons I would not recommend it here. The 765 skips pulses at low load currents in order to be more efficient. This pulse-skipping produces more output ripple on lower frequencies, which will be fine for line drivers and LCD backplanes, but is not suitable for any audio work. Also, these things are not hobbyist-friendly. Take it from someone who has designed these things into commercial products. The first time we used one, I put the MAX732 12V step-up switching regulator on a 6-layer board with ground plane. The 732 only runs at 170kHz so it's already a little easier to design a board for. (Remember that the dI/dt of a switching regulator even at 170kHz is very, very high, and the effect of parasitic inductances on the board and component leads becomes huge.) Still, it was a bear to get the layout right so the part would operate smoothly and not have excessive output ripple at frequencies below the switching frequency. A couple of years into production, Maxim "improved" the wafer-fab process, making the switching faster, and giving us trouble again at high harmonics of the switching frequency. Since we had tons of these multi-layer boards in stock, we opted to do some hand modification since that was cheaper than throwing out otherwise-good boards. We had to solder a capacitor directly to the IC leads with almost no lead length (try getting production people to understand that!), scrape the soldermask off of a few other places and solder chip capacitors on (since chip capacitors have no lead length), and put ferrite beads over certain wires to keep RF noise out of the aircraft radios that operate in the 108-135MHz band. I used the same switching regulator again in other designs later, but by then I knew how to do it right and we didn't have any more trouble. (Part of that, BTW, involved the chainsaw line we talked about earlier.)
Quote:
The total lack of response within a given timeframe (say, half a second) from a peripheral's configuration engine means it's a dumb device.
There probably is no need for a timeout period since it's synchronous and the master dictates the data transfer rate.
