[aos]

That's only about 42kHz square signal, right? Pretty badly damped... doesn't look good at all at such low frequency.

 

[jeffreyj]

Quote:

Originally posted by eric343 Oy. On ppl's reccomendation I removed the 22pf caps...

I hate to say I told you so, but I told you so.

There's a good reason so many engineering textbooks advise putting those capacitors there...

 

[ppl]

jeffreyj> There are just as manny tech docs that also advise the reasons for not using Compensation caps. eric told me the sq wave was seriousl tilted with the caps in place as one would expect at this frequency with this much compensation capacitence.

I have got perfect 100 KHz. sq wave response without compensation capacitors with this Buffer. it involves chousing an op amp with 5 times less Bandwidth then the el-2008 and selecting proper value resistors around the feedback loop and the Inputs. Large value resistors so as to bring board parrisitics into play. anyway since eric is just starting to learn i guess it is best for him to use compensation capacitors untill he fully understands the reasons somthing is done and compensation capacitor use is to compensate for a circuit that by nature is unstable and have all the Poles in the wrong place. placing compensation capacitors to make a unstable circuit stable is bad design. But this is just me i do not claim or do i generaly follow conventional wisdom. so put caps in don't put caps in Hell i dont care but your ears will

 

[jeffreyj]

I have a pc board here for a mixer/amplifier that has 75k feedback resistors shunted by 10pF NPO ceramics. RMAA 5.0 reports response being down 0.35dB at 20kHz. Changing the caps to 2.2pF flattened the drooping out to 0.05dB down at 20kHz. A 40kHz square wave went from looking fairly rounded (tr=2uS) to squared up but with soft shoulders (tr=500nS).

Without the shunt capacitors, ringing persisted for approximately 1/3rd of each half-cycle with a resonant frequency of 24MHz (the op-amp's GBW is 25MHz... surprise, surprise). Rise time was only slightly longer than the orginal signal (52nS vs. 38nS). The 2uS rise time looks like an eternity on the scope, but it still translates into a 500kHz bandwidth...

Of course inexperienced and/or incompetent designers will have more trouble if they omit the shunt capacitor, but even the most experienced engineer can have troubles trying get an op-amp with a GBW north of 10MHz to behave!

 

[jeffreyj]

eric343: the ringing in your scope shot has a period of about 3.5 cycles in 5uS, so its frequency is ~700kHz. A shunt capacitor of 2.2 or 3.3pF ought to knock it out completely for the common range of the feedback resistor values (less than 220k, more than 47k, let's say). A more precise, but still approximated, way of determining how much shunt capacitance is needed is to make the capacitor's reactance equal to the feedback resistance at twice the cutoff frequency desired (when Xc=Rf, the output will be down 6dB). Practically speaking, though, if the equation says you need a capacitance of less than 3pF, then you might want to lower the values of the gain setting resistors because parasitic capacitances will contribute a major, if not dominant, amount of the total capacitance present.

Oh, and ppl... leaving the shunt capacitance out is not the sign of a good designer, it's the sign of a designer who has left his or her circuit's high-frequency behavior to chance.

 

[eric343]

Well, my feedback resistors are 4.75K and 330 ohms, so that won't quite work

 

[jeffreyj]

Quote:

Originally posted by eric343 Well, my feedback resistors are 4.75K and 330 ohms, so that won't quite work

Feeling a bit froggy today, aren't you? (first the lead-acid battery thing, now this, eh?)

You have access to a nice scope, apparently, so why not use it? Put some different value caps in parallel with Rf and bang the input with 20kHz, 40kHz, whatever frequency square waves you want your amp to be able to reproduce without either ringing or drooping.

But the ability of the amp to reproduce even 20kHz square waves is, in itself, meaningless given that the usual source devices are quite unable to reproduce a square wave of so high a frequency. I do mean unable, too.

In fact, give it a try by recording a 20kHz square wave onto a CD with the best sound card you can beg, borrow or steal and then look at what plays back on your scope.

 

[aos]

jeffreyj, you are forgetting that at sufficiently high frequencies the board parasitics can take over. The reason a designer might have ommited the feedback capacitor is because he designed the PCB (thickness, ground plane layout etc.) so that there is in fact capacitance there to compensate. If you have measured it and saw that it works fine, why do you say it would be leaving it to chance? It's a nice way of having a safety if you're using an unknown board but if it works, why change it?

 

[eric343]

Froggy? Heh, interesting term

*ribbit* *ribbit*

In any case, yeah I know that it isn't realistic to assume my amp will ever have to see a square wave in real life... but I could certainly imagine some signals or transients having a rise similar to the leading edge of a square wave! And I don't want any distortion on those...

Besides, if I can get my amp to reproduce a near-perfect square, then at least I'll know it will reproduce perfectly anything else that I throw at it, and isn't that the whole point to audiophilia...?

Regarding the trial and error thing, yeah, I'd do that except I don't have any caps in a value other than 22pf.

(the other ones I've got are all in the uF range)

 

[aos]

Jeffreyj is right in that you'll never see 20kHz square coming from the DAC, in fact at most you'll get a sine wave at 20kHz. Though if you had 20kHz square sampled at 192kHz - if using DVD-A for example, you might get a more complex form than just a simple sine since you'll see at least 3 harmonics (at 40, 60 and 80kHz). Actually I can't remember any more but a square wave might only contain odd harmonics. Yes, just checked on the web, only odd harmonics, so you'd only get a 60kHz (in addition to 20kHz) if you sampled at 192kHz (everything above 96kHz is cut). But you're also right in that there is no reason why your amp should not be able to have enough bandwidth to transfer a 20kHz square unaltered (except for amplitude).

Oh, here's something I just found, showing the effect of adding those harmonics one by one. You can click on "+" sign above each harmonic to see how the signal looks like so far.

http://psych.hanover.edu/Krantz/fourier/square.html

 

[jeffreyj]

aos: Good point! I should reiterate, then, that I meant my comment to apply only to the topic of audio amplification, in which I feel it is most practical to not attempt to incorporate circuit parasitics, rather, to swamp or avoid them.

eric343: the point of audiophilia is the reproduction of sound, not square waves. Square waves are composed of the addition of the odd harmonics to a fundamental in the proportion of one over the harmonic number (the third harmonic would be 1/3rd the level of the fundamental, the fifteenth would be 1/15th, etc.) This is known as the Fourier Transformation of the square wave, btw, and keeping it in mind will go a long way towards tempering your thirst for bandwidth and slew rate

To keep your square wave looking reasonable requires that the rise time not take up more than 5% of the period (we'll assume that fall time is the same, for a total of 10% of the period spent in transition). 5% of the period of a 20kHz square wave is 2.5uS. To convert this into bandwidth, you can just take the reciprocal of the period: 400kHz. And, mind you, your amp's response must be flat out to this frequency.

Now what do you suppose are the chances your CD player - in which a brick-wall filter chopped off the signal at 22,050Hz somewhere along the way - will be producing 20kHz square waves accurately? Sorta takes the breath away to look at it from this perspective, doesn't it? But don't feel dejected, virtually nothing on the planet you'd want to listen to generates a square wave, which is why, along with the above explanation, I think banging square waves through an amp is good only for testing its stability.

edited because I just saw aos' post and wanted to say: cool link! I was looking for something like that to illustrate my point!

edited again because I couldn't resist tossing one more jibe eric343's way... hehe. Quote:

Besides, if I can get my amp to reproduce a near-perfect square, then at least I'll know it will reproduce perfectly anything else that I throw at it

A 7404 Hex inverter will do the same thing, but I wouldn't want to use one as my headphone amp!

'Course, there was that link awhile back that had a CMOS hex inverter chip being used as a headphone amp... maybe not so crazy?

Tomo ought to know

 

[MERTON]

is the 2009 better than the 2001?

 

[eric343]

Merton- Yes.

Jeffrey - what might also be contributing to the ringing/overshoot is that I'm running it into a 1M scope probe as a load- not the <50ohms that the EL2009 requires to be run into if you don't want to use a snubber. So I think I may need to use a snubber...

(incidentally, do snubbers go inside or outside the feedback loop?)

 

[MERTON]

should i upgrade my 4x el2001's to el 2009's? price? and why do none of the diy builders offer anything above the 2002?

 

[eric343]

Merton, please start your own thread instead of hijacking this one. This thread is a discussion of the issues involved in using an EL2009. If you have an EL2009-based amp and wish to discuss ways of improving its performance, feel free to do so in this thread. Otherwise you will find it considerably more productive and less ire-attracting to start another thread.

My apologies if I seem a bit harsh, but if I remember correctly you've done this kind of thing before.

NOW, back to the DIY talk!