AD8397-class A?

Jul 27, 2006 at 1:35 PM Post #3 of 26
Quote:

Originally Posted by MASantos
Would the AD8397 OPAMP benifit from class A biasing


Why not try it and see?

Quote:

would this cause any oscilation/instability problems?


If that's all it takes to destabilize your configuration, you didn't have enough margin to begin with.

Quote:

Originally Posted by robzy
The amount of current you would have to draw might be in the order of hundreds of milliamps


I can't see why. Even with an unrealisticially high load of 2V peak into 32 ohms, you only need 62 mA. That's well within the 8397's abilities.

Now, problems I do see are:

1. How do you keep the op-amp cool? You'll definitely have to use the EPAD version, and it'll have to be connected to a substantial copper plane.

2. What do you use for the CCS? A lot of the favorites around here won't scale up that high.
 
Jul 27, 2006 at 1:58 PM Post #4 of 26
Quote:

Originally Posted by tangent
I can't see why. Even with an unrealisticially high load of 2V peak into 32 ohms, you only need 62 mA. That's well within the 8397's abilities.


Hrm, good point.

A bit off topic then - is the reason the PIMETA/MINT have buffers after the class-A biased opamps because the opamp did not have the current capability for both the load and the CCS?

Thanks,
Rob.
 
Jul 27, 2006 at 3:11 PM Post #5 of 26
Quote:

Originally Posted by tangent
Why not try it and see?

If that's all it takes to destabilize your configuration, you didn't have enough margin to begin with.



I don't understand your point. Could you explain this?

Quote:

Originally Posted by tangent
Now, problems I do see are:

1. How do you keep the op-amp cool? You'll definitely have to use the EPAD version, and it'll have to be connected to a substantial copper plane.

2. What do you use for the CCS? A lot of the favorites around here won't scale up that high.



Cooling: Using the EPAD version connected to the ground plane and use some sort of heasink on top of the IC.

CCS: either a CRD or jfet cascode configured for more current output.

the reason for this thread is that the PINT(mini3fied) has amazing sound for it's size and I was thinking what could be done without size limitations. A home version with beefier PS section using a TLE2426 to create the virtual ground, Maybe isolated power rails a la PPA and M3, Alps RK27 pot etc.

It would be a killer amp!
 
Jul 27, 2006 at 4:08 PM Post #6 of 26
Quote:

Originally Posted by robzy
Class A biasing is not really possible for opamps driving a headphone load directly. (I assume you dont have a buffer?)

http://tangentsoft.net/audio/opamp-bias.html

To summarise: The amount of current you would have to draw might be in the order of hundreds of milliamps, which is unrealstic.

Rob.




Unrealistic but quite doable- isn't that what this hobby is about?

http://www6.head-fi.org/forums/showt...pa#post2082883

(scroll up page to see pics)

My proto is about Class A output operation over the full range with op amps - yes it does require "order of hundreds of milliamps" - I used ~220mA push-pull (440mA pk) to get 3 Wrms into 32 Ohms - as per Gilmore's recommendation of required headphone power


most "class A op amp biasing" you see in headphone op amp circuits is only a few mA for internal stage use

it is likely the AD8397 is already working with 3-5 mA pp output stage bias internally to get good MHz distortion specs - so the 1st 5-10mA of output is probably "class A"

if you really want to go full ouptut swing Class A you're limited in sucking power out thru the pcb with the EPAD to a few W - heatsink on top of the package won't help much - after all the die/leadframe is deliberately set downward to be exposed on the bottom of the package so you have nearly the full package thickness of molding compound in between - this isn't like a large chip area/thin molding memory or cpu thermal path - the op amp die is small and the plastic molding compound thickness is large

which is why I went "belly up" with the heatsink clamped to the exposed thermal pad - I am sucking 6 W from each package from the 2 most heavily loaded chips in my cascaded/paralleled amp - the bigger TPA6120 op amp packages in my amp have the Cu slug Zalman cpu heatsink clamped to the TPA power pads with a few mil smear of Artic Silver in between - pcb power plane heat sinking is for wimps!
 
Jul 27, 2006 at 4:44 PM Post #7 of 26
Could you post some pics of the internals and how you secured the heatsink to the IC?


Quote:

Originally Posted by jcx
Unrealistic but quite doable- isn't that what this hobby is about?

http://www6.head-fi.org/forums/showt...pa#post2082883

(scroll up page to see pics)

My proto is about Class A output operation over the full range with op amps - yes it does require "order of hundreds of milliamps" - I used ~220mA push-pull (440mA pk) to get 3 Wrms into 32 Ohms - as per Gilmore's recommendation of required headphone power


most "class A op amp biasing" you see in headphone op amp circuits is only a few mA for internal stage use

it is likely the AD8397 is already working with 3-5 mA pp output stage bias internally to get good MHz distortion specs - so the 1st 5-10mA of output is probably "class A"

if you really want to go full ouptut swing Class A you're limited in sucking power out thru the pcb with the EPAD to a few W - heatsink on top of the package won't help much - after all the die/leadframe is deliberately set downward to be exposed on the bottom of the package so you have nearly the full package thickness of molding compound in between - this isn't like a large chip area/thin molding memory or cpu thermal path - the op amp die is small and the plastic molding compound thickness is large

which is why I went "belly up" with the heatsink clamped to the exposed thermal pad - I am sucking 6 W from each package from the 2 most heavily loaded chips in my cascaded/paralleled amp - the bigger TPA6120 op amp packages in my amp have the Cu slug Zalman cpu heatsink clamped to the TPA power pads with a few mil smear of Artic Silver in between - pcb power plane heat sinking is for wimps!



 
Jul 27, 2006 at 6:10 PM Post #8 of 26
maybe pics soon -

if I forget the weeks of all niters required to build it

if I figure out the digital camera at work

if I don't play outdoors weekends - its summer in case you hadn't noticed

I may open it up to rework/optimize some compensation before the Boston meet
 
Jul 27, 2006 at 6:50 PM Post #9 of 26
jcx, maybe it's just me but that sounds pretty crazy
tongue.gif
 
Jul 27, 2006 at 7:33 PM Post #10 of 26
Quote:

Originally Posted by robzy
is the reason the PIMETA/MINT have buffers after the class-A biased opamps because the opamp did not have the current capability for both the load and the CCS?


Bingo. Not only are most op-amps limited to around 40 mA into a dead short circuit, they'll actually go into current limiting, harming their sound quality, well before that. There's just no good way to push such op-amps into class A if you want them to drive a load, too.

Quote:

Originally Posted by MASantos
CCS: either a CRD or jfet cascode configured for more current output.


The biggest CRDs I'm aware of are around 5mA.

As for raw JFETs, when you cascode JFETs, the total current is some fraction of the Idss of the lower JFET, so you want one somewhere up in the 100mA range to hit my 62mA number, and more like 200-300mA to make jcx happy. I did some searching, and couldn't find such a beast.

This is exactly why I said what I did above. You're going to have to look beyond the old favorites. As a starter, watch the "Precision Current Source" show here. There are several more that Pease didn't cover there. In fact, I believe I recently saw a book just on this topic.

Quote:

Originally Posted by jcx
I used ~220mA push-pull (440mA pk) to get 3 Wrms into 32 Ohms - as per Gilmore's recommendation of required headphone power


Hmm...I see 125 mA in the standard Gilmore dynamic amp article on Headwize. You need twice my recommended voltage for this, and I thought I had it overspec'd pretty wildly already.

All three numbers probably give class A, but the sound is no doubt different. The main points are: 1) yes it can be done with this chip; 2) no it can't be done with most other chips.
 
Jul 27, 2006 at 7:36 PM Post #11 of 26
Quote:

Originally Posted by MASantos
Quote:

Originally Posted by tangent
If that's all it takes to destabilize your configuration, you didn't have enough margin to begin with.


I don't understand your point. Could you explain this?



A CCS presents only two minor problems to its current source: a static (DC) current load, and probably a bit of input capacitance. The 8397 is a cable driver IC, so high currents and a fair bit of capacitive load are what it is designed to drive. You would have to be driving the chip right on the edge of oscillation already for this to send it over the edge.
 
Jul 27, 2006 at 8:59 PM Post #12 of 26
Quote:

Originally Posted by tangent
As for raw JFETs, when you cascode JFETs, the total current is some fraction of the Idss of the lower JFET, so you want one somewhere up in the 100mA range to hit my 62mA number, and more like 200-300mA to make jcx happy. I did some searching, and couldn't find such a beast.


just for info: BF246C is such a beast, followed by BF246B and pn4391. Nevertheless I'd rather pick a good buffer than deal with the all the hassle of this route.
 
Jul 28, 2006 at 4:10 AM Post #13 of 26
OK a little “moonlite” photo session at work:

amp_front.jpg



amp_guts.jpg


The heat sink is a //Zalman//[edit: Thermaltake] Golden Orb II, drilled and tapped for a ¼-20 ss socket head screw which protrudes from the center of the Cu slug bottom and to which the pcb carrying the TPA6120s is clamped

The 6 TPA prints of Artic Silver are visible on the Cu slug of the Zalman

The Al tripod has nylon pcb standoffs cut down to press on the pcb – the conical spring washers maintain clamping force

The TPA6120 op amps were laid down belly up on the pcb and each leg was bent down and soldered (how? – very carefully, with the aid of optics) – You can see I didn’t bother to bend down all of the legs – some are not connected internally in the TPA

The pcb turned out to be flat enough that even with the 6 packages there was only a few mil of unevenness – plenty small enough gap given the Artic Silver thermal conductivity (don’t get it on the circuitry – not formulated as a electrical conductor it still can be a problem)

I’m sure there are simpler solutions possible but I wanted a “universal” proto amp +/-24 V swing for Hi Z cans and +/-440 mA pk Class A for Lo Z cans – these #s give ~ 25 W total power dissipation in Class A for the 2 channels/6 TPA6120s
 
Jul 28, 2006 at 2:35 PM Post #14 of 26
Quote:

Originally Posted by steinchen
just for info: BF246C is such a beast, followed by BF246B and pn4391


Not really. I can't find any sources for the 246C in the US, but I did find the B, and its minimum Idss is 60mA. That means you'll have to cherry-pick parts to get a cascode with >60mA using this JFET. You'd need a minimum Idss somewhere above 100mA to be able to use arbitrary JFETs.

Look, I'm not trying to say this is impossible. Just impractical.
 
Jul 28, 2006 at 3:05 PM Post #15 of 26
I don't think it will be practical to implement this feature acording to the replies I am receiving here.

Other questions:
The datasheet says that the AD8397 benefits much from well regulated power. Since batteries aren't really desired for a home amp, I tought about a LM317 based power supply with a Filtered IEC receptacle.
Would there be any other regulator IC that would be a good improvement without increasing the price of PS section a lot?

As for the amp's power section I want to use a TLE2426 as voltage divider and probably more capacitance. Would there be any benefit of using a buffer(probably opa551) to increase current capability? I believe that using jfet to isolate the power rails would be useful? Or does this only apply if there are two different power sections such as in the PPA?

I read in the PPA page that is uses diferential drive topology bt haven't found anything about it in the other documents. Could you explain this better?

Tangent, apart from your article regarding virtual grounds and related articles, what other documents are worth reading for this subject?

Manuel
 

Users who are viewing this thread

Back
Top