Hear it in the one drop

Inveterate meddling #186: using the Flea with heavier loads

The 'flea' is based on the obseration that for very light loads the error amplifier itself can source all the current you actually need. In theory the AD797 can actually supply up to 100mA under suitable conditions, but in practice this is curtailed by internal thermal dissipation. Hence when used as a 5v regulator, we recommended 30mA as a practical maximum - far more than enough for a Tentlabs XO for which it was actually designed.

It has been observed that as this limit is reached or exceeded (for example, using a flea to supply +V_DD to a TDA1541A dac) the DC accuracy begins to suffer, dropping about 40mV at 40mA output - which looks like a massive 1ohm DC output impedance!

There are two reasons for this behaviour. 30mA at 5V approximates a load of 150ohms; we are also effectively driving the opamp with +/-6 v supplies (i.e, the single rail at 13.x v) These two conditions conspire together - first, the AD797's open loop gain drops off sharply for loads below, say, 250ohms (see fig 17 of the datasheet); secondly, we're right at the bottom-end of the recommended supply range, which severely limits available drive into low impedances (figs 2 & 3 of datasheet) , presumably by starving the internal current sources which are key to the chip's operation.

So - one palliative would be increasing the 797s pre-reg supply - which would require a higher raw supply, and would increase the thermal loading in the opamp (which has a negatively afffect on both AC accuracy and longevity). Still, we really can't complain at the audible performance of a part not designed to do the job we're using it for...

Is it broken? No. But let's fix it anyway...

A bit more thinking about getting the flea to happily drive lower impedances inevitably leads to thoughts of adding a pass transistor - with all the potential for instability that brings. But pressing on regardless we now have a runner:

- basically, one cut track, one new resistor, one new transistor.

How it works: rather than have the opamp solely drive the base of an external pass transistor as in most regs, here the 797 drives the output through the new 1200ohm resistor. Once the current into the load exceeds about 0.5mA the new transistor starts to turn on, which removes the heavy current handling (and so thermal loading) from the AD797 (exactly just like Quad's 'current-dumping' scheme) Global negative feedback is taken from the output side still, keeping the new pass transistor inside the loop and output voltage governed by the 797. Effectively the opamp runs in 'Class A' and near constant-current - since loading above this is divided by h_fe of the pass transistor. This negates the DC error sources identified above. In theory, we may even gain slightly better performance, due to the extra gain in the loop and reduced /constant loading on the AD797 error amp.

A handful of notes:

C8, the 50pF decompensation cap, is now connected directly from the final regulator output pin to pin 8 as an overall decompensation loop including the pass transistor. This value is critical to stability - do not increase it. Also see the update notes below.

The pass transistor collector is connected to the 7812 output, and not the opamp supply pin, where the preceeding series 0R5 could provide unwanted positive feedback with reactive loads, inviting instability.

The power handling limit is still only about 80ma, maybe a little more, due to the BC547s dissipation limit. Remember this also affects the gyrator at the input to the Flea! If you must have more it could be worth trying, say, a BD139 instead but

The resistor can be pretty much any value over 470ohms; 1K2 works very well indeed, since it reduces loading on the 797.

Stability depends to an extent on the added pass transistor having a F_t comparable or higher than the opamp's GBW, hence use of generic small-signal transistor. You will not be able to bolt a darlingon in here and go arc-welding with a MondoFlea...( well you can - but it takes a bit more compensation!)

Not shown, but strongly recommended, is adding a small capacitor from the 7812 output pin to ground; a 10uF tantalum works well. This provides HF bypassing for the new pass transistor.

How to implement this: If you wish to modify an existing Flea PCB, work on the back of the board. Cut the short thick track between AD797 ouput (pin 6) and the heavy trace which connects C9 +ve to L1. Bridge the cut with new resistor (1K - 2K2 is fine) Cut the adjoining thin track between opamp Pin6 and R6; connect R6 now to output end of new resistor. Leave C8 connected to opamp pin 6. Add new BC547 (or BD139 or similar) as shown in diagram above. That's all! Adding a cap from output of the 7812 to ground is probably also a good idea to aid stability, but not mandatory in practice. The new resistor can be anything over 1K; 2K2 works well as a default value.)

The payoff A modified pcb has been is up and running and driving 38mA into a TDA1514 with a cool 797 and no hint of RF oscillation on the scope for days; sounds great too. Tracking the operating point overnight (12hours of runnng) and the hotrodded flea powering the DAC held output voltage within 2mV and input current within 0.1mA (!), both of which appear to have more to do with ambient thermal swing than anything else. Looks like this mod is worth trying if you want more than 10mA out of a flea used as a regulator. Please note, though, that if you want 50-60mA or more, you could still be better off with a super regulator - which offers lower dropout voltage and much higher current handling, negative rail options, remote sensing, heatsinking - in other words, much more flexibiltiy and superior performance at these current levels.

One last observation: when using the flea as a regulator for remote/ heavier loads, you can lift the lower voltage set resistor (0v end of R7- the 820ohm resistor) and return this seperately to the 0v pin of the device fed. This will give better regulation, because the 0v wire impedance no longer appears in series (see here for an explanation) Picking-off the output feed for R6(1K3 ohms) from the fed device +v pin will give you full Kelvin sensing, but also firmly risks instability - not recommended without further investigation on a case-by-case basis.

18 Oct 06 update:

Some feedback - ron commented:

Did anyone notice the sound to be a little brighter?
its not a problem at all and still sounds great in my system 
but it seems that theres slightly more highs.

I too thought I heard a slight change in tonal balance that way. On reflection I decided I didn't much care for it - more like added haze, not 'better' treble. So I've had a play, mostly with C8. It seems (to me) the best sounding arrangement when trying an extra pass transistor is to:

1) Remove C8 (47pF)

2) Connect Pin 8 of the AD797 to the output* with a 33 - 47pF capacitor.

Two things: one, this means that the 'decompensation/distortion-correction' of the 797 encompases the new hack and two, it does not as fully decompensate the 797 as previously - which should help tolerate parasitics added with the new parts.

The result does sound different. That difference is subtle, but definitely there - the haze is gone; it makes leaving C8 alone (when adding a the pass transistor) sound slightly muzzy. Note there is no difference in current draw (within <0.1mA that I can measure) which together with a good stare at the scope suggests it's not a matter of stopping rampant oscillation as just...finessing this implementation. The full 47pF acan be used if desired - I've had no stability issues now in over a year of running full decompensation.

I'd appreciate feedback if anyone tries this whole hack.

*e.g on the output side of the new transistor, at its junction with the new resistor. Keep loop area as small as possible though - use the shortest path available for the caps connections.

Back to Part One...

Misc. reference information:

AD797 datasheet

1.09.07 Notes on series resistor valu updated.

18.10.06 Notes about C8 updated.

15.10.06 Page added.