What is your favorite most versatile op-amp?

[Neomys Sapiens]

My preferred OPA in the specified class is the LT1468/1469. It did prove repeatedly to be care free and rock solid when used to the full extent of it's capabilities. It is available in PDIP. But you would have to compromise on some parameters: Input bias current is slightly higher and output current is +/-22mA nominal.

I would fully endorse the AD845 too as well as the OPA2727.
Some other components to look at: ADA4898, LM6142, OPA602, OPA827, TLE2027, ISL55001 (I have not checked all parameters, but they are >10MHz, around 1 mV or better and they satisfy your +/-15V requirement.)
Maybe LM7332, if current is critical.

[MasterT]

Recently I discovered LM4562, it has 100x (!!!) lower distortion level than NE5532. And 5.5 times wider bandwidth and available in DIP8 package.
 IC is good with 5V power, capable to output about 3 Vp-p no distortion, for comparison NE5532 outputs 2 Vp-p.

[exe]

There are several NE5532 on the market, at least two from on semi and TI. Rumors say they are quite different. They are also different from different years of manufacture. I think I found on the Internet that TI did at least once shrinkage of the die. This badly affected noise (according to the.

As for distortion, I wouldn't care that much. According to Rod Elliot distortion of any good sound amplifier (not just opamp, but the whole circuit) is inaudible. So, 100x reduction of inaudible distortion is still inaudible .

PS did someone mention this document? http://www.nanovolt.ch/resources/ic_opamps/pdf/opamp_distortion.pdf . An interesting reading with tons of plots showing how opamps compare to each other in terms of distortion under different operating conditions. I refrain myself from drawing any conclusions as final performance is depends a lot on the circuit.

[Zero999]

Recently I discovered LM4562, it has 100x (!!!) lower distortion level than NE5532. And 5.5 times wider bandwidth and available in DIP8 package.
 IC is good with 5V power, capable to output about 3 Vp-p no distortion, for comparison NE5532 outputs 2 Vp-p.

According to the data sheet, it's got over double the input current noise of the NE5532 at 1kHz, so it's not necessarily an improvement.

[MasterT]

From NE5532 DS  TI : Equivalent Input Noise Voltage: 5 nV/√Hz Typ at 1 kHz
                      OnSemi: Input Noise Voltage: 5.0 nVń Ǹ Hz (Typical)

LM4562 TI : Input Noise Density: 2.7nV/√Hz (typ)

Regarding distortion, audio is not the only area for OPA application, measurements is another. To have anything above 4.5 digits, OPA must be very low distortion. THD level at 0.002% put a limit same as 15-bits ADC 

[Zero999]

From NE5532 DS  TI : Equivalent Input Noise Voltage: 5 nV/√Hz Typ at 1 kHz
                      OnSemi: Input Noise Voltage: 5.0 nVń Ǹ Hz (Typical)

LM4562 TI : Input Noise Density: 2.7nV/√Hz (typ)

You've misread my post. I know the voltage noise is lower. I was taking about the current noise, which is higher, than the NE5532 at 1kHz. The LM4562 has 1.6pA/√Hz vs only 0.7pA/√Hz for the NE5532. If the impedance seen by the input is greater than about 2k925, the LM4562 will be more noisy, than the NE5532.

The current and voltage noise densities add at the op-amp inputs. The current noise forms a voltage noise source, when if flows through the resistance, connected to the input.

V_{N_TOT} = √(V_N²+(I_N×Z_IN)²)

So if you have an inverting amplifier.


R_in = 10k
I_IN = 0.7pA√Hz
V_N = 5nV/√Hz

V_{N_TOT} = √(10k×0.7p)²+5n² = 8.6nV/√Hz

If R_f = 100k then the gain will be 100 so the output voltage noise will be 860nV/√Hz.

[nemail2]

OPA2197 it is, for me...

[MasterT]

You've misread my post. I know the voltage noise is lower. I was taking about the current noise, which is higher, than the NE5532 at 1kHz. The LM4562 has 1.6pA/√Hz vs only 0.7pA/√Hz for the NE5532. If the impedance seen by the input is greater than about 2k925, the LM4562 will be more noisy, than the NE5532.

I see, the point of this thread is to define a priority list of the OPA's parameters. Since there is no perfection in this world, no OPA that would satisfy all requirements. Current noise is the last line of all spec I'd care about.
Voltage offset is close to the middle of the list. Its' easy to fix in hardware or software. The THD is the most important, difficult to straight up and it's directly limit accuracy of the circuits. DC accuracy is no exception, basically THD is equivalent to INL / DNL in the ADC-DAC specification.  Long time THD is associated with audio, or AC in general, but IMHO it's a big mistake to use LM358 at the front end of the anything with >12-bits resolution. And input offset is not an issue.

[exe]

The THD is the most important, difficult to straight up and it's directly limit accuracy of the circuits. DC accuracy is no exception, basically THD is equivalent to INL / DNL in the ADC-DAC specification.  Long time THD is associated with audio, or AC in general, but IMHO it's a big mistake to use LM358 at the front end of the anything with >12-bits resolution. And input offset is not an issue.

Isn't THD frequency-dependant? If so, can lm358 still be good for low-frequency signals?

As for offset, why do you say it's not an issue? If DC accuracy is required, it is an issue.

[SiliconWizard]

One thing to note about the NE5532 is that (at least in the TI datasheet for the NE5532 that I have) there is no figure to be seen regarding distortion. Did I stupidly miss it? Does anyone have a datasheet stating distortion? And if we're going to use NE5532's made by TI, how are we going to know what to expect in terms of distortion?

I've used the LME49720 with very good results for current-to-voltage output stages for audio DACs.

[Zero999]

You've misread my post. I know the voltage noise is lower. I was taking about the current noise, which is higher, than the NE5532 at 1kHz. The LM4562 has 1.6pA/√Hz vs only 0.7pA/√Hz for the NE5532. If the impedance seen by the input is greater than about 2k925, the LM4562 will be more noisy, than the NE5532.

I see, the point of this thread is to define a priority list of the OPA's parameters. Since there is no perfection in this world, no OPA that would satisfy all requirements. Current noise is the last line of all spec I'd care about.
Voltage offset is close to the middle of the list. Its' easy to fix in hardware or software. The THD is the most important, difficult to straight up and it's directly limit accuracy of the circuits. DC accuracy is no exception, basically THD is equivalent to INL / DNL in the ADC-DAC specification.  Long time THD is associated with audio, or AC in general, but IMHO it's a big mistake to use LM358 at the front end of the anything with >12-bits resolution. And input offset is not an issue.

If depends on what you're doing.

Noise can't be accounted for in software, so it's pretty important in my opinion. Current noise dominates voltage noise, at higher input impedances and voltage noise is dominant at low impedances. This is very often overlooked. Someone might choose the NE5532 over the TL072, on the bases it has a much lower voltage noise of 5nV/√Hz vs the seemingly massive 18nV/√Hz of the TL072, but the TL072 has a current noise of just 0.01pA/√Hz vs the 0.7pA/√Hz for the NE5532. Then they wonder why it's so noisy, as the input impedance of the amplifier is 100k: the correct choice would have been the TL072 of course.

The THD is the most important, difficult to straight up and it's directly limit accuracy of the circuits. DC accuracy is no exception, basically THD is equivalent to INL / DNL in the ADC-DAC specification.  Long time THD is associated with audio, or AC in general, but IMHO it's a big mistake to use LM358 at the front end of the anything with >12-bits resolution. And input offset is not an issue.

Isn't THD frequency-dependant? If so, can lm358 still be good for low-frequency signals?

As for offset, why do you say it's not an issue? If DC accuracy is required, it is an issue.

Voltage offset is often unimportant nowadays, as the op-amp is more often than not, before an ADC. The circuit can be zeroed in software by shorting the inputs, taking a reading, storing it in memory and subtracting it from subsequent readings.

The one thing to watch out for is drift, which can't be corrected for, unless it's convenient to frequently zero the circuit.

[magic]

Isn't THD frequency-dependant? If so, can lm358 still be good for low-frequency signals?

CMRR spec is 85dB, presumably only valid with light load, meaning that 20V of input swing produces 1mV change in offset voltage. And it's not even guaranteed to be linear, its slope may be flat close to ground and get steeper towards VCC or vice-versa or any other bizarre thing.
If you look at the schematic, the output is class A up to 50µA. If you want to sink more, the voltage gain stage needs to swing 3 diode drops down to turn off the upper darlington and turn on the lower sink transistor. Given 100dB open loop gain at DC, that's a 15µV step in offset voltage just to respond to changing polarity of load current.
But you are right that crossover distortion at 1Hz or so will not be as bad as at audio frequencies, where open loop gain is lower.

That's also a problem for most of those rail-to-rail input amps, like OPAx192. If you read the datasheet, they very carefully avoid specifying any CMRR number within that problematic region from 3V to 1.5V below VCC and instead tell you to have a look at figure 13-14 in "typical characteristics", which doesn't look pretty to say the least.

One thing to note about the NE5532 is that (at least in the TI datasheet for the NE5532 that I have) there is no figure to be seen regarding distortion. Did I stupidly miss it?

Nope. I have datasheets from TI, Philips, Signetics, JRC, Fairchild, ON Semi - no one specifies it.

Does anyone have a datasheet stating distortion? And if we're going to use NE5532's made by TI, how are we going to know what to expect in terms of distortion?

One Samuel Groner published quite comprehensive distortion measurements of a few dozen opamps. I think I saw a link earlier in this thread or just search for it. Beware it's 35MB.

Or, as Douglas Self puts it:

The 5532/5534 is made by several companies, but they are not all created equal. Those by Fairchild, JRC, and ON-Semi have significantly lower THD at 20 kHz and above, and we’re talking about a factor of two or three here.

[exe]

If you look at the schematic, the output is class A up to 50µA. If you want to sink more, the voltage gain stage needs to swing 3 diode drops down to turn off the upper darlington and turn on the lower sink transistor. Given 100dB open loop gain at DC, that's a 15µV step in offset voltage just to respond to changing polarity of load current.

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

[Zero999]

If you look at the schematic, the output is class A up to 50µA. If you want to sink more, the voltage gain stage needs to swing 3 diode drops down to turn off the upper darlington and turn on the lower sink transistor. Given 100dB open loop gain at DC, that's a 15µV step in offset voltage just to respond to changing polarity of load current.

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

I don't see why not. Crossover distortion is only irrelevant for DC or when the output is continuously sourcing/sinking.

[Neomys Sapiens]

If you look at the schematic, the output is class A up to 50µA. If you want to sink more, the voltage gain stage needs to swing 3 diode drops down to turn off the upper darlington and turn on the lower sink transistor. Given 100dB open loop gain at DC, that's a 15µV step in offset voltage just to respond to changing polarity of load current.

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

I don't see why not. Crossover distortion is only irrelevant for DC or when the output is continuously sourcing/sinking.

Crossover distortion doesn't make it into the discrete samples.

[MasterT]

Isn't THD frequency-dependant? If so, can lm358 still be good for low-frequency signals?

As for offset, why do you say it's not an issue? If DC accuracy is required, it is an issue.

It is affected by frequency, but it does not go to 0% (THD) at DC level. So, no, LM358 and NE5532 are not good above 12-bits.

Noise can't be accounted for in software, so it's pretty important in my opinion.

Noise easy to null out by averaging in software. Though ADC sampling rate needs to be higher.

One thing to note about the NE5532 is that (at least in the TI datasheet for the NE5532 that I have) there is no figure to be seen regarding distortion. Did I stupidly miss it? Does anyone have a datasheet stating distortion? And if we're going to use NE5532's made by TI, how are we going to know what to expect in terms of distortion?

I've used the LME49720 with very good results for current-to-voltage output stages for audio DACs.

I love LME49721, pity charts depicted THD just to 6-7 kHz.
NJM5532 has THD at page 3,  THD vs. Output Voltage

TI has data for NE5534, I'd think 32/34 as a twins.

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

 And this is most interesting parts. Interfacing to high resolution ADC is tricky, +-15V power is not applicable since next question would be safety of the ADC. Tweaking with high end adc (16 /18 bitters from TI and AD, ad7988, ad7984, ads8866) I came to conclusion that all worlds leaders failed to make R2R ultra low THD (0.0001%)  OPA.
AD on their AD7988 evaluation board, follow old fashion design - demanding -2.5 +7.5V power for OPA just to keep THD under control. So, crossover still in there, and it's not the only one of source THD to keep an eye on. 

[David Hess]

Noise can't be accounted for in software, so it's pretty important in my opinion.

Noise easy to null out by averaging in software. Though ADC sampling rate needs to be higher.

If flicker noise could be averaged away, then there would be no need for chopping.

[H713]

LM4562 has come down in price considerably over the last few years, so that has a considerable impact. I didn't realize just how much its price came down over the last few years. IMO, (again from an audio point of view), applications where a 5532 or 5534 is suitable, the LM4562 is probably a better choice, because if the relatively high current noise of a 5532 isn't a problem, the LM4562's even higher current noise may not be an issue either.

Again, this really does come down to application. A microphone preamplifier has different requirements from an EQ circuit, and the digital world (which I don't deal with very much) will likely have different (in some cases more stringent) requirements.

The discrete op amp is a great example. It seems foolish, after all it is a costly and large, and the DC precision is poor compared to even a 741. The noise performance and linearity, however, can be excellent, and they can be designed to handle very high supply rail voltages. For some applications it is a poor choice, but for others it makes perfect sense.

Still, I think if one is going to make a list of good, versatile op-amps to stock, the LM4562 (not a whole lot more costly than a 5532 these days) and perhaps something like an OPA2134 would be on my list.

With that said, there are a ton of comparably decent FET input op amps, some of which may or may not be relevant now. I still have a large stock of OPA604s and AD713s, but their cost makes them hard to recommend at this point over something like an OPA2134. Even the TL072 is perfectly adequate for many applications, and they're cheap as chips.

[magic]

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

CMRR is still crap, though.

NJM5532 has THD at page 3,  THD vs. Output Voltage

Damn, I missed it

TI has data for NE5534, I'd think 32/34 as a twins.

They aren't, 5532 has its open loop gain reduced to stay stable in voltage followers and it shows as more distortion at audio frequencies. Also lower unity gain frequency and slew rate.

I came to conclusion that all worlds leaders failed to make R2R ultra low THD (0.0001%)  OPA.

OPA365 comes close by using internal charge pump to generate beyond-the-rails voltage supply for its input stage. Other similar opamps may exist, but most R2R input designs combine two input stages and have bad "input crossover" distortion.

[RES]

LT6015 (SOT-23-5)

https://www.analog.com/media/en/technical-documentation/data-sheets/601567ff.pdf

[Zero999]

Noise easy to null out by averaging in software. Though ADC sampling rate needs to be higher.

Not true. If you're talking about averaging, then all you're doing is reducing the bandwidth, which is only acceptable in a DC application. Then there's 1/f noise, which increases at lower frequencies and equates to offset drift, which needs to be frequently nulled.

If you look at the schematic, the output is class A up to 50µA. If you want to sink more, the voltage gain stage needs to swing 3 diode drops down to turn off the upper darlington and turn on the lower sink transistor. Given 100dB open loop gain at DC, that's a 15µV step in offset voltage just to respond to changing polarity of load current.

Just don't make it sink . I mean, if we are talking about ADC, crossover distortion is irrelevant, isn't it?

I don't see why not. Crossover distortion is only irrelevant for DC or when the output is continuously sourcing/sinking.

Crossover distortion doesn't make it into the discrete samples.

It does if some of the samples are taken at the point when the output current is changing direction.

[exe]

It is affected by frequency, but it does not go to 0% (THD) at DC level. So, no, LM358 and NE5532 are not good above 12-bits.

DC offset can be nulled. I did a couple of experiments on lm358 and ne5532 from TI measuring offset voltage vs input vs supply voltage on a breadboard at room temperature (unity-gain configuration, no load). In addition, I also heated the opamp with hotair gun to ~110C. I include the best and the worst opamp. I used soic opamps, mounted on a dip adapter, the circuit assembled on a breadboard using crappy wires, flux wasn't cleaned. Please find the data attached. In total, I tested two samples of ne5532 from TI, two samples from OnSemi, and four samples of lm258/lm358 from various vendors. All bought from reputable distributers (Farnell and TME). For NE5532 I include my best sample, as well average sample. For lm358 I just include "typical" performance.


I did measurements in 2017, so don't remember all the details. The goal was exactly to understand if they are suitable for 12bit ADC. My NE5532 samples showed that Vos is stable within 0.05mV if enough voltage headroom is given (as measured by UT61C). Temperature coefficient is about 1.5-3uV/C (but I measured only two datapoints: at 25C and ~100C on a breadboard, thermocouple issues etc were not taken into account, I let the temperature stabilize for ~1min to avoid die stress due to non-uniform temperature).

It's up to the reader to decide how good they are. But I disagree that 12bit is their limit.

[SiliconWizard]

Averaging for noise reduction only works as long as the noise is reasonably random, or at least uncorrelated to the signal (probably a workable assumption here) AND the signal of interest is identical from one capture to the next. Otherwise you're just basically low-pass filtering and will lose bandwidth.

Averaging is often used for repetitive signals triggered by some event, not as a noise-reduction technique for non-repetitive or unsynchronized signals.

[MasterT]

Noise can't be accounted for in software, so it's pretty important in my opinion.

Noise easy to null out by averaging in software. Though ADC sampling rate needs to be higher.

If flicker noise could be averaged away, then there would be no need for chopping.

"Hardware averaging" - putting a few devices in parallel ?
What I'm getting at, there are solutions to minimize noise & offset, but distortion /non linearity are different species. This is why I'd move to #1 in my priority list.

Probably, DC accuracy may get better by increasing open loop gain using 2+ OPA as combine or composite whatever it's called. Anyway it's not so easy like paralleling for noise reduction.

[Zero999]

Some types of non-linearities can be corrected in software, as long as they are known and repeatable. For example, a single supply op-amp might be non-linear near the negative rail, even at DC, but if we can compile a table of the input  and output voltages, compared to the expected (ideal) output voltages, then we can apply a correction factor later.