Current mirror is done incorrectly for led driver?

**broken link removed**

...in the above schematic , are Q4 and Q5 supposed to be current mirrors?
There base's are connected together, but shouldn't their emitters also be connected together?

It is fine. The emitter degradation resistors help to reduce the mismatch in current due to the fact the transistors aren't matched. They are physically separate devices so without the emitter resistors the current matching could be very bad.

Keith

Is the attached circuit ("current reg") not an improvement?, as it regulates the total led current...wheras in the ti.com one, one of the branches is essentially unregulated.

I mean, say one BJT has a bjt of 0.55V , and the other 0.7V.
Also, since they are sot23 bjts, (sorry I didn't tell you this before) they will not be very well thermally coupled, even if next to each other on the pcb.

(LTspice sim also attached)

- - - Updated - - -

In fact, if you run this simulation, ("CURRENT REG_SOT23") with just 0.4V difference between the led strings, then one of the sot23 bjt's dissipates 120mW...and that's above the maximum allowable dissipation for a sot23 which is 100mW.
I think this circuit is just not reliable, would you agree?

It's far from an improvement. Besides current mirror matching which has been already adddressed, it's no longer fail safe. If the left LED string gets disconnected, the new circuit drives the same current to one LED instead of two.

Regarding transistor rating, you are free to choose a suitable type. Expectable forward voltage differences should be determided from LED data.

I see your point, "one off-all off" is a good thing here.

But ok, the schematic in the top post is , I believe inadequate, and I believe you will agree with me...
They are SOT23's used for Q4 and Q5.
Clearly these are over-powered with mildly different total string vfs....the leds are LA B6SP from osram, and these are matched but each led can differ in vf by 0.15V.
Clearly, sot23's are inadequate?

Yes I agree that bigger transistor packages can be used for Q4 and Q5...but bigger BJT package transistors are not likely to be nearly so well vbe matched....this is due to the difficulty of making matched power transistors due to the greater volume of silicon in these bigger packages.

Therefore, do you not think that the following solution, (called "CLAMPED CURRENT") involving FET, DPAK (or sot89) current clamps is far far better.....?

The FETs are virtually zero resistance when the led strings happen to be near perfect matches....but if there should be significant un-sharing of current, then the respective fet will clamp the current to 120mA maximum.....and being a dpak, or sot89 package, it is far far more able to handle this power.

- - - Updated - - -

Incidentally, the following attached ("Constant off time boost") drives the exact same load with a constant off time boost converter, and doesn't bother with the sot23 bjts which are going to be over-powered when poorly matched led strings occur.
-Just puts the entire 26 LEDs in one long series string.....so much simpler!
Why did ti.com not just do this?