What opamp?

[Morocotopo]

Hi guys. A question:

I´m designing a super duper delay with four PT2399´s. With that much delay chips, controlling noise is very important, it has NE571 companding, filters galore, all kinds of noise controlling stuff. The signal path starts with a BJT buffer (emitter follower) to get low impedance, then, besides the PT´s circuitry, it has five opamps configured as unity gain inverting amplifiers (so actually buffers...) for summing signals and different tasks. All have 10K resistors in the inv input R and feedback R, so the input impedance is 10K, as far as I can tell. I´m trying to extract the absolute best performance in terms of noise. All the opamps are TL071 or TL072, but I believe they are not the best option. Since all opamps "see" a low impedance at their input (BJT buffers or other opamp´s output), maybe other non FET opamps will give me less noise? 4558´s? NE5532´s? Any other recommended opamps?

Thanks.

[duck_arse]

lm833 is bipolar, was used in diy hi-fi preamps years ago. old hat now.

[R.G.]

lm833 is bipolar, was used in diy hi-fi preamps years ago. old hat now.

Well, maybe. The TL07x series is older hat. But the LM833 is very good for audio, just didn't get as wide an exposure as the TL series or the NE553x. I like it very much in many designs.

The NE5532 is good, low noise and low distortion, but suffers from an input impedance at the + input of perhaps only 100K. If the input is driven from a low impedance source, this is not an issue.

The LM833 is low noise, wide bandwidth, low distortion and higher input impedance than the 5532, although lower than the FET input opamps.

It is difficult to beat the TL07x series for their performance per cost ratio. They may not be perfect at anything, but they are quite good, and cheap. That's why they are so widely used, and so cheap. And yes, that is circular, and intended that way.

@Morocotopo:
Opamps are not the only thing that affects noise. The sampling noise of the PT2399 is pretty much fixed as a part of the digital sampling process, and cannot be reduced except by low-pass filtering at its output after the D-A converter. There is a fundamental tradeoff in sampling noise and bandwidth, in that you can't eliminate sampling noise, you can only make it better by finer-step sampling and fancy internal digital processes, and then filter out the remaining noise at the output.

If the majority of your noise is coming from the sampling path, even zero noise opamps (which do not exist) would not improve it. And real-world opamps may not add enough noise to be noticeable. Correspondingly, if you have a poorly set up opamp stage with opamps and input/bias/feedback resistors that have a lot of thermal noise, the opamp again may not matter much.

"Very best noise performance" is not a simple thing to do. There are engineers that spend their entire careers trying to get better noise performance. Unless there exist a few extra high noise generating places in your circuit, just replacing an opamp won't help much. It's like getting a smooth floor: all the polishing in the world won't help until you sweep the rocks out and hammer down the nails that are standing above the surface.

[Mark Hammer]

Your answer is on page 10 here: http://hammer.ampage.org/files/Device1-8.PDF
When input Z is low, 5532/833 is better for noise.  When input Z is high, bi-fet is better.

[Morocotopo]

lm833 is bipolar, was used in diy hi-fi preamps years ago. old hat now.

Well, maybe. The TL07x series is older hat. But the LM833 is very good for audio, just didn't get as wide an exposure as the TL series or the NE553x. I like it very much in many designs.

The NE5532 is good, low noise and low distortion, but suffers from an input impedance at the + input of perhaps only 100K. If the input is driven from a low impedance source, this is not an issue.

The LM833 is low noise, wide bandwidth, low distortion and higher input impedance than the 5532, although lower than the FET input opamps.

It is difficult to beat the TL07x series for their performance per cost ratio. They may not be perfect at anything, but they are quite good, and cheap. That's why they are so widely used, and so cheap. And yes, that is circular, and intended that way.

@Morocotopo:
Opamps are not the only thing that affects noise. The sampling noise of the PT2399 is pretty much fixed as a part of the digital sampling process, and cannot be reduced except by low-pass filtering at its output after the D-A converter. There is a fundamental tradeoff in sampling noise and bandwidth, in that you can't eliminate sampling noise, you can only make it better by finer-step sampling and fancy internal digital processes, and then filter out the remaining noise at the output.

If the majority of your noise is coming from the sampling path, even zero noise opamps (which do not exist) would not improve it. And real-world opamps may not add enough noise to be noticeable. Correspondingly, if you have a poorly set up opamp stage with opamps and input/bias/feedback resistors that have a lot of thermal noise, the opamp again may not matter much.

"Very best noise performance" is not a simple thing to do. There are engineers that spend their entire careers trying to get better noise performance. Unless there exist a few extra high noise generating places in your circuit, just replacing an opamp won't help much. It's like getting a smooth floor: all the polishing in the world won't help until you sweep the rocks out and hammer down the nails that are standing above the surface.

Yep R.G, lowest noise is an ideal to get close to, not an absolute goal. I want the best possible s/n ratio for the design I have. It started almost unusable, little by little I made it pretty good, but I want to get "as good as possible within my limited knowledge".
The thing is non true bypass, and in bypass I was getting some noise (so the dry path was at fault), that I managed to reduce quite a bit, the opamp variable is the last thing I want to try to get that last bit of performance. Maybe a different opamp will make a difference or not, but it doesn´t hurt to try. I believe I have reduced all sources of noise outside of the opamps. When finished, I´ll show the schem and maybe get some more criticism / advice.
The wet path is quite good noise wise, considering there are four PT´s. Each one has, besides the in/out 2nd order MFB LPF, a BJT 3rd order Sallen Key LPF at the output, and the whole wet path has NE571 companding. Also, the mixing opamps have feedback caps with -3db freq around 2.5-3K. A LOT of filtering! At the longest delays I get the inevitable grainy sound, but that´s to be expected, yet hiss is quite controlled. It´s quite a beast, the thing. 100 rersistors, 50 caps...

Your answer is on page 10 here: http://hammer.ampage.org/files/Device1-8.PDF
When input Z is low, 5532/833 is better for noise.  When input Z is high, bi-fet is better.

Mark, yes, I read the articles from Craig Anderton, they are the reason I posted the question, the opamps he mentions there are mostly obsolete. But the conclusion is clear.

Just read the 833 datasheet. Wasn´t familiar with it. I´ll try that I think.
Regarding the 5532, I read that it can be difficult to use, that it´s not a direct replacement for other opamps. True?

[PRR]

IMHO, your hiss is mostly from the delay lines, not the buffers.

S/N is how much the signal is *above* the hiss. Don't (just) lower the river, raise the bridge! Get the signal level as HIGH as possible through the delay lines. Boost until it comes out bent (then back-off a bit). This may mean trading your unity gain follower at the input for an adjustable gain booster, and maybe throwing some loss at the end to get back to reasonable level.

Measurably better performance comes with better delay lines (this IS the 21st century!(*)), or with companders (though they may add artifacts, bobbles and surges).

(*) Think. With skill and modern tech, we can "delay" a performance an hour or a year with insignificant degradation. High-quality digital recorders are 24-bit (>100dB S/N) and fit in the hand. Handy-recorders do not delay less than the length of the track, but that is a mild technical problem (being able to read and write at arbitrary points). I was wrassling with the first BBDs 37 years ago; it is disgusting that we still struggle with hissy delays.

[Mark Hammer]

[Mark, yes, I read the articles from Craig Anderton, they are the reason I posted the question, the opamps he mentions there are mostly obsolete. But the conclusion is clear.

Just read the 833 datasheet. Wasn´t familiar with it. I´ll try that I think.
Regarding the 5532, I read that it can be difficult to use, that it´s not a direct replacement for other opamps. True?

A 5532 is just a dual 5534.  The 833 is aimed at more or less the same sort of application, low-noise, low input Z and low-Z loads.  And I will assume that "bi-FET" is TL071/81/074/etc.  The 833 and 5532 both draw more current, from what I understand, but then if your intent is to pack a quartet of 2399s in one circuit I assume you are not looking for ways shave off 2ma of current draw here and there. 

[karbomusic]

The NE5532 is good, low noise and low distortion, but suffers from an input impedance at the + input of perhaps only 100K. If the input is driven from a low impedance source, this is not an issue.

Why is the non-inverting input (+) impedance so low? I thought the non-inverting input was ideally infinite and practically 10's of MOhms?

[R.G.]

Why is the non-inverting input (+) impedance so low? I thought the non-inverting input was ideally infinite and practically 10's of MOhms?

Ideally, yes, practically infinite. Real-world amplifiers are not ideal. The 553x amps do in fact have impedances of about 100K for both inputs listed on their datasheets.

This is often good enough when dealing with pro-audio stuff with incoming signal impedances of 600 ohms to 10K, but falls woefully short of the 1M that has come to be standard for guitar inputs.

The 553x are the darlings of the pro-audio world, where the lower input impedance can be ignored, and where low resistances all around are sought for low noise.

In fact, the very existence of JFET input and MOS input opamps is a consequence of the input impedance of a bipolar input opamp not being good enough in some circumstances.

Mother Nature is wacking you over the head right now.   

[karbomusic]

Ideally, yes, practically infinite. Real-world amplifiers are not ideal. The 553x amps do in fact have impedances of about 100K for both inputs listed on their datasheets.

I see 300k in the TI datasheet which is the other reason I asked.  That and it doesn't make sense to me, to see so many references of "even in practical applications non-inverting is still 10s of mega ohms" yet we wind back up at a measly 100k. Seems it be better to just never make any of the statements outside of working theory on paper, especially the but it's still megaohms piece. I'll dig around and see if I can find the original reference.

In fact, the very existence of JFET input and MOS input opamps is a consequence of the input impedance of a bipolar input opamp not being good enough in some circumstances.

:

Yep, I made that deduction before posting but there are too many exceptions (this being one of them) to not ask. Assumptions tend to bite back when it's too late.   This isn't the first time I've asked, the current poll is running about 50/50. Reminds me of my real job debugging, I spend more time debugging the debugger than the code it is supposed to debug. 

[Paul T]

Have you tried the Burr-Brown OPA2604 ?
The specs are quite fantastic in terms of noise, input/output impedance and frequency range.

[Morocotopo]

IMHO, your hiss is mostly from the delay lines, not the buffers.

S/N is how much the signal is *above* the hiss. Don't (just) lower the river, raise the bridge! Get the signal level as HIGH as possible through the delay lines. Boost until it comes out bent (then back-off a bit). This may mean trading your unity gain follower at the input for an adjustable gain booster, and maybe throwing some loss at the end to get back to reasonable level.

Measurably better performance comes with better delay lines (this IS the 21st century!(*)), or with companders (though they may add artifacts, bobbles and surges).

(*) Think. With skill and modern tech, we can "delay" a performance an hour or a year with insignificant degradation. High-quality digital recorders are 24-bit (>100dB S/N) and fit in the hand. Handy-recorders do not delay less than the length of the track, but that is a mild technical problem (being able to read and write at arbitrary points). I was wrassling with the first BBDs 37 years ago; it is disgusting that we still struggle with hissy delays.

Well Paul, I just tried to raise the bridge. Gave +3dB to the input opamp and -3dB to the out opamp(s). But distortion is creeping in... damn.
There is a NE571 compander in the delay lines path.
And yes, we are still dealing with hissy delays... at least the DIYers who don´t want to go to digital programming.

[Mark, yes, I read the articles from Craig Anderton, they are the reason I posted the question, the opamps he mentions there are mostly obsolete. But the conclusion is clear.

Just read the 833 datasheet. Wasn´t familiar with it. I´ll try that I think.
Regarding the 5532, I read that it can be difficult to use, that it´s not a direct replacement for other opamps. True?

A 5532 is just a dual 5534.  The 833 is aimed at more or less the same sort of application, low-noise, low input Z and low-Z loads.  And I will assume that "bi-FET" is TL071/81/074/etc.  The 833 and 5532 both draw more current, from what I understand, but then if your intent is to pack a quartet of 2399s in one circuit I assume you are not looking for ways shave off 2ma of current draw here and there. 

Well Mark, the power consumption is around 85 mA.  So a measly 2 mA won´t make a difference... only thing is, I don´t want to exceed 100 mA or I won´t be able to use my PS to power it, it has 78L09´s for the separate outputs so 100 mA max. Or, you know, six or seven batteries will do...for about three minutes.

Have you tried the Burr-Brown OPA2604 ?
The specs are quite fantastic in terms of noise, input/output impedance and frequency range.

I´ve never bought those here in my country (Argentina), don´t know if they are available, will inquire. Today I got some 5532´s and 833´s, I´ll try those in the meantime.

I just tried the thing in my amp´s FX loop... too hissy still...   and it´s coming from the delay path. Grrr. More aggressive filtering needed.

[karbomusic]

Have you tried the Burr-Brown OPA2604 ?

I have one of those in an orange squeezer circuit, sounds amazing.

[karbomusic]

Yep, I made that deduction before posting but there are too many exceptions (this being one of them) to not ask. Assumptions tend to bite back when it's too late

I'm a big fan of closure so I wanted to follow up with how I solved my (not directly related to this thread) problem... I had circuit where I removed a redundant OPA2134 buffer and the existing clipping op amp (post buffer) was fine impedance wise (CMOS) but I wanted the ability to go back and swap ICs around.

So... I changed my PCB layout so that there is a spot to solder in the buffer IC (if needed). If I don't want/need the extra buffer, I just don't add it and jumper pins 2/3 on the PCB where it would have went which bypasses the buffer section giving me the choice of either/or.