Headphone Amp Redux

I don't know why you think I'm just randomly pasting values - it would seem a fools errand to attempt to get a circuit to work that way. If the amp is going to be DC coupled and there's going to be a servo then one pretty much has to work out what all the bias points are all the way back through the circuit to make it work properly.

When it comes to implementing things like the feedback loop I think it's more difficult; I've practiced deriving transfer functions enough that I could attempt to calculate the phase margin through the feedback loop by hand, but I think it will be messy and probably not a great reflection of reality. I've seen it suggested in at least one of the books I've read on amplifier design that new designs often oscillate when feedback is applied, and tips are given on how to stabilize them through empirical means. So, for now if I have to tweak values in the simulator to get it to work properly I am not going to lose sleep over it. If it means that I am currently excluded from the "real analog designer" club so be it, I still have plenty of learning left to do. I have a formidable-looking textbook on control theory with my name on it...

And you don't need separate signal and servo loops.

Bye. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |

      Remember: Once you go over the hill, you pick up speed

I said un-usable not un-stable. I read what you said about fixing that problem :-)

Unusable in the sense that no one (else) will consider building it. You can't be remembered for a design like that.

You have parts in your circuit that do not make a positive contribution to the quality. Feedback hides the defects, but these defects also limit the lowest distortion figures you can obtain without adding more feedback. And adding more feedback makes it harder to stabilise the design.

To get low distortion you need a stable high gain circuit with loads of bandwidth and then you use bags of feedback or you use ultra linear design blocks with less feedback.

For ultra low distortion you use both ideas. Any component that doesn't contribute positively has to go because it is only contributing distortion.

In case your worried that I will force you into a particular design I won't, but to keep something unchanged I think you should have a reason (and get a reason if it should be changed). FYI my BOE scratchings suggest to me that, if it were my design and using discrete components, my cheapest but not impolite headphone amp would have about 4 active devices, something I think would be appropriate for most hifi and professional use would be around 7 active components and perhaps up to

14 for an attempt at an "ultimate" amp (but that needs to be reviewed on the basis of performance)

You've designed it and simulated it. Now I think you should breadboard it and see if there are any surprises. If so, find out what they are and fix the circuit or the simulation.

Go for it!

Cheers, John

To combine both ideas was sort of the idea I was going for - use a cascode for a wide bandwidth input stage, bootstrap it to give it lots of gain, then use local degenerative feedback to linearize the gain of that block before global negative feedback is applied around the whole thing. It does use a lot of components, so it's hard to say whether the benefits of this arrangement outweigh the distortion added by the excess components used in creating it. I suppose the only way to find out would be to run a bunch of THD tests on various gain stages in isolation. I should do that.

As a challenge to myself I'm going to try to construct the simplest headphone amp I can think of and see how it does. I'm not sure I can get down to 4 components, but I'll try to keep to a limit of 5 or 6.

e:

original

and

gn is

s to be

red. The

4 ohm g

pg

g

t of the

servo to

s DC

ut anyway

ore

(by

ave done

dea but

ppens -

timely

servo

per

hings

find

servo,

t a 64

s. I

rrent

s with

r

mplex

and it

does

ou

offset..

able

the NG

f

of

t to

hat

r;

that

and

down

I'm

"free"

nds

to

ve

k d

on

ood

ith

d

ly

s f

It seems you are the only one here with their pantyhose in a bunch. Did I say output protection is job number one? The answer to that would be no.

pg

jpg

pg

e

ur

you

of

l

of

or;

ave

ck

s k

ad

on

ed

rly

r e

ts

of

g r

Just for the record here, I didn't add a second feedback loop. Rather, I suggested using an op amp for the common mode feedback. It is hard to get op amp quality offsets with random discrete devices. Spice has no mismatch in it.

The only problem with FFT analysis of THD is it doesn't indicate what you do to improve (reduce) THD. For a real (physical) circuit, you display the residual signal from a THD analyzer along with the input signal on a dual channel scope, then see where in the signal waveform the residual increases. That is, you can see crossover distortion versus the onset of clipping.

No they don't. You have some complex and ill-considered interplays. I know you railed at JT suggestion that it looks like you have just cut and pasted bits together but it really does look that way

I suppose the only way to find out would

A cascade stage should already have improved bandwidth and linearity. Trying to do more with it may just screw it up (when you increase gain you often decrease bandwidth and so with feedback the surplus gain available at the frequencies of interest is not improved at all - which means it can provides no reduction in distortion - and that is the

*best* result. You can still screw it up royally.

Also a system with many local feedback loops turn out to be equivalent to a system with 1 global feedback loop. If you are going to add local feedback there needs to be a specific reason for a result that could not have been achieved with the extra gain and global feedback

This is very much a case where more complexity is not better (in terms of better THD slew rate etc) in exactly the same was as striping the thing too far towards simplicity is not better.

So keep the flights of fancy under control. There are many smaller steps to take yet. Testing stages in isolation is a Douglas Self idea - it's a pity it doesn't take into account how the stages may interplay (for example how they load each other)

4 active components - 4 transistors, Jfets, FETs, etc. If you want to be ultra daggy you could go for a one or two transistor amplifier. Unless you are designing a replacement circuit for a 1970's transitor radio there is no need to go that far.

But first, cut the servo stuff out of your circuit. Find an efficient way to bias it, and post it so we can see where to go.

How about two discrete active components, or maybe one?

John

original

be

The

the

to

anyway

done

but

-

timely

servo,

offset..

NG

Gee, let me see. At best then you're talking about designing a circuit to fix a problem that may not exist (excess offset) in a circuit that hasn't even got a solidly designed input or output stage.

Do you normally design like that? Do you think that approach will help Bitrex who already has trouble with design simplicity and discipline?

And why would I be all in a bunch about that? Perhaps because inappropriate expressions of opinions on subjects not relevant are distractions that have to be answered (to avoid a negative effect on the student) and that consume my time. Also, even if debunked or properly answered they may obscure to the student the points being made, distract the student, or prevent him from developing a more correct and disciplined design approach as he runs off on unrelated tangents.

And you didn't offer to help Bitrex and Bitrex hasn't asked you the question.

Here is my solution. You want to teach, just say so and it's all yours, but you had better be prepared to do a good job and finish when Bitrex is ready.

Bitrex - when I say "a solidly designed input or output stage" I am not talking about "simplicity")

Don't take the helpfull criticism personally. Groups are rough and ready and filled with people of all skill sets, moods and pejudices. They often don't suffer fools gladly and the atmosphere can get quite heated at times. Experienced engineers know that the best way to a sound design is to take a minimalist approach and only include stuff that's needed to meet the required spec. Anything more is unnecessary embellishment, even if it is a pet idea. It's wastefull of resources and only adds to the eventual cost of the item, which matters if you are building lots of them.

You want to build the perfect headphone amp, right ?. You obviously have an enquiring mind and are willing to learn, so you will get there, even if a far simpler circuit would get the job done. Sound basic tech knowledge and experimenting with different ideas, together with the experience gained from such work is how practical, competent engineers are grown and should be encouraged as much as possible. Nurturing the enquiring and creative mind and learning something new every day are some of the best reasons to work as a design engineer.

Perhaps do a bit more time with the hardware, scope and sig gen etc and less time playing with the simulator would help, but hey, who am I to tell you ?...

Regards,

Chris

I haven't said that very well. Yes, you should test stages in isolation, and then also consider how it will react with other stages - that's the part that Self ignores.

I'm not a big fan of DS. For one thing he suggests that instruments only tell pert of the story and that the human ear / perception is much more sensitive i.e your insturments might not tell the difference between an amp with .001% distortion and .0012% distortion but you might. I don't have a problem with that argument. But that "better sensitivity" argument is then used to justify "subjective" sound tests where amps with higher levels of distortion sound "better" than measurably better amps - which makes a mockery of the whole basis of it's very own argument. The instruments may not tell the whole story, but they don't lie.

te:

y original

il and

sign is

ars to be

tered. The

64 ohm

jpg

.jpg

jpg

act of the

ne

.

C servo to

is DC

hout anyway

more

lp (by

have done

idea but

happens -

a timely

C servo

roper

things

I find

he servo,

nst a 64

nes. I

current

ies with

our

complex

- and it

t does

you

lt offset..

veable

by the NG

of

al

e of

ext to

that

for;

gn that

ou and

t down

f. I'm

s "free"

tends

.

ve to

have

ick

is

ok

ead

n on

good

e

with

med

e

erly

a

ur

he

uts

of

ng

s r

or

Waste your time? Get a clue. This is usenet. It is a waste of time! Oh, and participation is optional.

Usenet is not email. You pose a question, it goes out to the aether, not a specific person. Answers arrive. Some excellent (like mine), and then well ... lesser answers.

Yep, your pantyhose is in a bunch.

my

f

.jpg

q.jpg

.jpg

ine

d.

DC

r n t t n

DC

g I . -

it

s

ral

s o

nt

as

e

d.

is

ook

d

od

he

e

He

e a d

our

the

ing

is

or

oor

o,

l t

ign.

ion.

e n

Running stages in isolation is fine, but you have to maintain the same load conditions. That is, if you test stage one by itself, it needs to drive stage two anyway. You also need to drive the amp from something other than a voltage source with zero impedance.

With the retarded attitudes in this place, it is no wonder.

If you don't like the attitudes here, you know what you can do.