DIYEFI.org Forum

TonyS wrote:

Fred wrote:Dual reg + 5V6 zener + 470 ohm inputs + schottky diodes = bad conditions for the CPU. 5V1 might be a better choice IF it doesn't conduct too much at idle. If that doesn't solve it, we need to rethink the protection as it's currently not 100% safe. I measured 5.6 at the pin with 12v applied to outside. BUT, I measured 5.3 on the rail. This screws all of the ADC readings pretty badly and puts the pin at 0.6V over the digital supply which it's attached to.

Fred.

Fred, it is a bit hard for this old brain to understand the specifics of your observations. So you applied 12V to one of the analog input pins, measured 5.6V at the input to the cpu and measured 5.3V on the "5vc -analog" net / rail?

I have a bit more to say but perhaps this should be the start of a new thread in Hardware as I see that the topic of power supply design is a ToDo item for Ravage.

Thanks,
Huff

Well, Fred suggested that a new thread might be appropritate to continue this discussion - so here it is.

1. Based on Fred's observations, I would guess that the amount of current being shunted to the "5vdc-analog" bus during his test on his Jaguar board exceeded the current requirements for the bus under his test conditions. Since the 5V regulator can only source current (no sinking capability), I believe that it was effectively "off" otherwise the bus voltage would have been clamped to the regulator's 4.75 - 5.25V spec. So I believe that the 5.6V zener clamp on the bus was responsible for the 5.3v clamp voltage that was observed. I personally am not comfortable with the use of a 5.6V zener for this function (and definately not a 5.1V) as zeners are rather "sloppy" devices. Perhaps the input clamps could be connected to the "5vdc-cpu" bus instead (higher current requirements).

2. It's important to remember that the A/D inputs on the cpu are referenced to the "analog" bus voltage. So as long as the sensors are fed from the same supply, the resistive sensors and most likely the active sensors will provide a ratiometric value of the "analog" supply voltage with the actual value of the supply not being that important (within reason).

3. Another thing to remember is that a short to battery on the analog inputs should only be a temporary condition, tolerated by the hardware but recognized by the software as an "out-of-range" value and notify the user (via lights, sound, behaviour, ...) that there is a problem that must be resolved.

Thanks,
Huff

1. You're right that there was almost no load on the analogue rail. Ideal test conditions, actually. Not acceptable to pump a noise source into the CPU rail, though, just attach 10, and same behaviour. Something is required to keep the Voltage on the rail regulated from both directions. You're right that zeners are sloppy devices, what have you got in mind that is better?

2. Whilst this is true, there are other constraints that mean that the readings are affected.

3. Agreed. It just needs to be robust to damage.

The question is: How much current will a 5V1 Zener draw at 5V, and is that amount acceptable heat wise.

How about using something like this:
http://circuits.linear.com/all--surge_s ... ection-529
Maybe even in combination with one of these:
http://circuits.linear.com/all--automot ... c%7Cdc-522

The links above are for a switch mode power supply. (buck-boost).

Very noisey, too many discretes, not that suitable unfortunately.

Naughty naughty!

I know late 90's and early 2k bosch and denso stuff uses linear (style, not brand) regulators.

Switchers can be used with great success, however getting the design right is non-trivial, and out of scope for first-gen FreeEMS boards due to how much time it would add to the development cycle.

KISS for now :-)

Fred.

Fred wrote:1. You're right that there was almost no load on the analogue rail. Ideal test conditions, actually. Not acceptable to pump a noise source into the CPU rail, though, just attach 10, and same behaviour. Something is required to keep the Voltage on the rail regulated from both directions. You're right that zeners are sloppy devices, what have you got in mind that is better?

I don't really know if what is currently in the design is acceptable or not.
The design really is dependant on the requirements / specifications.
If you really want it to tolerate 10 inputs shorted to battery continuously, thats one thing. If you make the assumption that everything is wired right and then one input wire accidently is shorted (pinched wire?) that's a bit different.
My only related reference is "shorts to battery / shorts to gnd" testing that I was involved with.
Acceptable (defined) operation when a single wire was shorted to bat or gnd for 1 - 5 minutes. No permanenet damage and normal operation restores when short is removed.

Nothing better (or as "clean") comes to mind at this time to replace the zener, but I'll dream about it for a while : )

Fred wrote:The question is: How much current will a 5V1 Zener draw at 5V, and is that amount acceptable heat wise.

I would be cautious performing this experiment as my guess is that regulator may / will go into short circuit protection due to the regulator output voltage range / zener range overlap.
Thanks,
Huff

I think that I just came up with a good solution to this.

5V1 zener + current limit resistor on the analogue rail will provide enough load for down regulation for X pins. 5V6 zener on rail will provide hard clamp in case of extreme dodginess.

Unsure on R value, but low would work as the voltage across it would be low too.

------

Huff, I look forward to your ideas :-)

Fred.

Hey Fred,
There is a class of devices called "shunt regulator"s that could generally be described as precision zeners. I don't think they are typically used for high power and personally, I've never used one, but it may be something you might want to take a look at.

Fred wrote: Not acceptable to pump a noise source into the CPU rail, though, just attach 10, and same behaviour.

I was going to argue that "noise" from a short to battery should be minimal, but then realized that because there are seperate 5V supplies, the differenct output voltages might be far enough apart to cause the clamp diodes to start conducting even with no fault.

Thanks,
Huff

Poor choice of words on my part. I was thinking intermittent short, but also noise to me meant "voltage fluctuations" even if slow/low speed/constant.

Got a link to one of those parts?

Fred.

The TL-431 is the most used programmable shunt regulator
http://www.fairchildsemi.com/ds/TL/TL431A.pdf