[NelsonVandal]

Ha! the bloody thing is playing music! It was hard work to match and solder everything. It took some creative bending and "bugs" on the bottom. It's very very tight.

I haven't the zobel ready yet since I forgot to order SMD resistors. The filmcaps aren't mounted either, and I haven't even started on the box or the batteries.

I was optimistic about the compensations caps in the ground channel. It's oscillating with the 10 pF's. With 43 pF it's DC-stabel. It's not easy to make a unity gain stable amp. I think I'm going to use 100 pF caps. I don't really know how to simulate the ground channel in a proper way, that's why I got it wrong.

It's funny how well the simulation matches reality. Everything measures exactly as in LTSpice except I had to use 180R instead of 160R from positive rail to diff pair/cascode, but I knew that would be a matter of Hfe. But then everything except the output transistors are matched to within 1 %.

Pictures to come later.

 

[MASantos]

pics! pics! pics!

 

[NelsonVandal]

Pictures can now be seen at: I guess I should clean it from flux.
Folded cascode headphone amplifier eSnips Folder
I guess I should clean it from flux.

There was still oscillation in the ground channel with 43 pF (as heard when listening with IEM's), with 100 pF it seems very stable. L/R channels seem stable with the original 10 pF caps paralleling the feedback resistor, after the zobel was added. The offset is easily adjusted, and seem stable to within a couple of mV's depending on the voltage supply.

The total current draw is kind of massive ~100 mA.

The amp is very quiet. The noise is on par with Mini3, and the gain is exactly the same.

How it sounds is hard to say. I have to listen to it for a while to be sure, and to let it "burn in". I think it's a bit soft. At least compared to Mini3, but AD8397 is sharp, aggressive, accentuates bass and is very forward sounding. There's no strong coloration, and the imaging seems fine. If I find it too soft in the long run, I'll build another one with 2SA970/SC2240 as cascodes and input transistors for the diamond buffer. They sound sharper and gives the impression of a more defined sound. I'll also use other output transistors as well, since ONSemi's BD137/138 is on the soft side of neutral.

 

[Hayduke]

I know I'll sound like a newb, but what do you mean by "folded cascode"?
I've never seen those terms.

Thanks

 

[NelsonVandal]

Quote:

Originally Posted by Hayduke /img/forum/go_quote.gif I know I'll sound like a newb, but what do you mean by "folded cascode"? I've never seen those terms. Thanks

It's a common opamp topology used in many (of the best) monolithic opamps like the champ AD797, AD829, AD823, ADA4841, ADA4899, AD8099, LT1469 etc. The cascode stage (common base) is connected by their emitters to the input and the collectors are folded towards the "opposite" rail. The amp is often reffered to being a one stage amplifier, meaning there's only one voltage stage. You can look at the data sheets for the opamps mentioned above, and I think you'll get the picture.

 

[NelsonVandal]

I've finally got the amp to sound the way I want. Some errors had to be corrected. The updated schematic can be seen at Folded cascode headphone amplifier eSnips Folder

(7. Increasing LTP bias from 2 to 4 mA, no certain improvement, I'll probably go back)
(6. Going from unity to gain 2 in the ground channel. Maybe more stable, maybe not, gonna find out later)
(5. Adding compensation capacitors to L/R, not sure it did anything to the sound or stability)
4. Increasing the compensation capacitor in the ground channel seems to make it more stable.
3. Lowering the cascode bias (to the casdode bases) improved the sound.
2. Removing the LTP degeneration resistors lowered the noise a good deal and improved the overall sound a bit.
1. CHOOSING THE RIGHT TRANSISTORS made all the difference - from a mediocre to a good sounding amp. BC549/559 from Philips (or 550/560) sound neutral but grey, soft and uninvolving. SA970/SC2240 from Toshiba sound lively and detailed but overly bright and a bit harsh. SA1016/SC2362 from Sanyo sound neutral, detailed and smooth. They were the missing link. Without them I would have stuck this amp in a drawer.

The amp sounds very clean, "liquid" and smooth. I tend to play louder than I used to, maybe because of lower distortion. Sometimes I miss the more aggressive SA970/SC2240. Noise level is lower than monolithic opamps, but not as low as in my JISBOS-, JLH- and Hiraga-like amps. To voltage freaks, I can say the amp doen't sound a bit better running from an external 24V TREAD-clone and internal L7815 compared to battery operation at 9.6 V.

I guess it could be improved further with "audio grade" resistors and better output transistors.

If I were to redo this amp (and I probably will, but as a speaker amp or desktop headamp), I would add CFP to the input stage and cascode the current sources/sinks.


This is not a beginners type of amp, and some tweaks and bending have to be done to get everything in place. If anyone would like to build it, you're welcome to buy a board, but it's quite fiddly. It's not tested in any way except by ear and DMM, so it's a risky project, and it's not worth the effort unless very good transistors are used.

I'm going to rebuild my LISAIII clone and optimize it for AD797 and AD829. It'll be a cool shootout - discrete folded cascode against (the best!?) monolithic folded cascode opamps.

 

[NelsonVandal]

I prefer this amp to the new chip-amp-amp, and it doesn't matter if I use AD797, AD825 or AD744 (comp pin out) as L/R and AD825 or AD829 (comp pin out) as ground. The chipamp sounds better with diamond buffers than diode biased emitter followers (LISAII type), but the discrete amp is just more my taste - warm and full and still with all the details.

I've ordered parts to build 5 or 6 amps - the last stock of SK170 from ELFA and potentiometers from Tangent. From BDENT I've ordered the fabulous Sanyo SA1016/SC2362 and I'm taking a chance on Sanyo SA1538/SC3953 as output devices (they look very promising with both very low Cob and seems linear up to enough currents to drive headphones).

I'm going to make one of those a desktop amp and maybe one as a preamp. With higher voltage supply I can increase the diode voltage drop to the cascode, and get even lower distortion.

I also ordered some SK246 to see if cascoding the CCS to the LTP make any difference.

I'll probably experiment with CFP input stage and maybe cascoding the folded cascode. My intuition tells me CFP will be it. CFP input has been a major improvement in other amps I've built. I can't understand why not more people use it.