Is an op amp an op amp an op amp?

[Al (astral highway)]

Hi folks,

Here's one to throw into the ring!

My first exposure to op amps was LM741, with its simple inverting input, non-inverting input and output. No offset null and no frills.

Things seem to be a lot more complex now, with very much higher bandwidth and slew rates and all sorts of bells and whistles, and devices that can run on very much higher voltage rails very comfortably. All good...

But have things really changed, fundamentally? Can we do the same basic things, if we choose not to do the advanced things? Like design a simple amplifier stage with a gain of two with a couple of resistors? (Or a gain of 10,000, equally, just as we could before? Does the little triangular symbol always embody the same capacities, at heart? as well as being used in any number of more recent applications (as video amplifier, etc.)

[TonyDuell]

In theory, yes. The 'ideal op-amp' (infinite input impedance, zero output impedance, infinite gain, etc..) hasn't changed as far as I know. The simple applications are based on that ideal.

BUT...

Some op-amps, particularly high-bandwidth ones, are not stable at low overall gains (that is with a lot of -ve feedback, the hardest being the unity gain buffer where all the outut signal is fed back). Phase shifts in the op-amp can turn that -ve feedback into +ve feedback at high frequencies, and if the op-amp still has enough gain at said frequency, the thing turns into an oscillator. I can remember many late nights in the lab at Bristol University taming such devices.

Some will be OK though. You have to read the data sheets.

[Ed_Dinning]

Hi Al, the theory stays the same and they generally work fine unless you are trying to get large gain bandwidth products, unity gain buffering or very high slew rates.
These modern amps often need careful PCB layout and veroboard can be a no no due to the intertrack capacitance.
If you want use one as a comparator, do not use an op amp, use something like the LM311 that is designed for comparator use...


Ed

[Radio Wrangler]

The humble 741 could do a lot of things quite well, and it still can. BUT it became such a standard that it became a commodity. Prices fell to pennies and the manufacturers found it hard to turn any profit on them. So they brought out loads and loads of improved ones, all different. Life got confusing.

But let's look at the 741 and see what it can and can't do.

It has a DC gain of about 100dB. Zin is about 100k, Zout about 100 Ohms and the inpi=uts take about 0.3uA of bias current.

It's flat from DC to 10Hz (Yes, ten Hertz) Then the gain falls at 20dB/decade until it hits 0dB at about 1MHz.

Its output can move no faster than 0.5v/us

With so much DC gain, it has to have feedback around it to control its own DC conditions. It steers the output to try to get its two inputs to be at equal voltages, plus an error in the region of +/-5mV.

The feedback acts to improve the bandwidth and the gain/BW tradeoff follows that 20dB/decade line. It increases Zin and lowers Zout

It can drive its output within a few volt of its supplies.

You can use it for various DC and low frequency jobs. For audio, it's OK-ish but not hifi.

You can get amps which trounce it on any of the above parameters, and many of these better parts have become affordable commodities too. You have too much choice!

So if you have a basic job which needs only a jellybean opamp you could use any one of a lot of types, but not any one of any type.

Tony's explained that some of the fast ones aren't stable in low-gain circuits. You can only use these if you're careful.

Then there are the weird ones:

Current feedback opamps have very low input Z on the inverting pin only. This confers a speed advantage but you can't use them in integrators and filters are difficult.

Operational Transconductance Amplifiers.... sort of like opamps but control the output current instead of output voltage.

Norton Operational Amplifiers.... use current signals on both inputs and steers the output to balance the input currents.

OTAs and Nortons are rarely seen. Current feedback ones are used in fast circuits. Mostly these type are avoided.

Horowitz and Hill hav some good selection tables in their book, and they are a good guide to what opamps are industry standards and are good choices.

David

[Al (astral highway)]

Originally Posted by Ed_Dinning

These modern amps often need careful PCB layout and veroboard can be a no no due to the intertrack capacitance.

Hey Ed, thank you and good heads up there for people on veroboard...
I tend to shun veroboard except for DC or very low frequency circuits, never for ICs, and instead use a big old ground plane (0.3mm thick copper foil) and stick-on pads from QRP-me, so a form of surface-mounting. I haven't used a SOIC style op-amp yet but I also have a special mounting pad for those devices, too...

I do have an LM2904DG from TI that I'll be investigating. What a huge transformation since the days of LM741! I mostly want one as a simple voltage amplifier on a 24V rail right now, but also in dynamic filters like notch filters etc. All part of the learning curve!

[Tyso_Bl]

LF351 Nuff said?

[Al (astral highway)]

Originally Posted by Radio Wrangler

The humble 741 could do a lot of things quite well, and it still can. BUT it became such a standard that it became a commodity. Prices fell to pennies and the manufacturers found it hard to turn any profit on them. So they brought out loads and loads of improved ones, all different. Life got confusing.

That's fascinating! Really interesting to see how plummeting manufacturing costs can play out in an industry like this!

Thanks for your tour of the 741 too, David. Will read an inwardly digest at more leisure, but already very illuminating. Cheers!

Originally Posted by Radio Wrangler

Horowitz and Hill hav some good selection tables in their book, and they are a good guide to what opamps are industry standards and are good choices.

Hah, will still have to add HH to my book list. I only have 'Practical Electronics for Inventors' atm. (Scherz/Monk - pretty good... anyone else have it?)

[Radio Wrangler]

Some methods of construction like Veroboard or having power transistors on long wires from boards to heatsinks are considered no-nos for high frequency operation, but it's not so much a matter of what frequency range you want the thing to work at, but also of the frequency range you don't want it to oscillate at.

So if you use a device with gain to hundreds of MHz to do a DC task (common enough nowadays) then the style of layout has to be good for hundreds of MHz, not just DC.

A lot of transistors have been 'improved' to the point where replacement devices will hoot in some classic circuits. The Quad triplets are a well known case.

I remember chatting with Ian, G3ROO, some years ago. Ian's a prolific builder of homebrew HF radio equipment, including several complete transceivers. He was fancying getting a spectrum analyser and said he was looking for something which went up to about 100 MHz - enough to see a few harmonics because he only built HF stuff. Perfectly sensible, but a major benefit of having one is in seeing that there are no spurious oscillations. Having coverage to several hundred MHz is reassuring.

So don't just think of the frequency range you want things to happen at, also consider the range you want nothing to happen at!

David

[Craig Sawyers]

First time I used opamps was to build the somewhat famous Texan audio amp back when I was 16. There were three per channel - one 741 and two 748. I recall that these were 60p each in 1972 - or about £8 each in today's money. That is the blunt end of fifty quid now for six op amps.

The standard vanilla audio opamp now is the 5532. Dual in an 8-pin dil or sm, 10MHz gain-bandwidth product, 9V/us slew rate. About a quid.

Even better the LM4562. Dual in 8-pin. Gain-bandwidth of 55MHz, 20V/us slew, 2.7nV/rootHz noise, 0.3ppm distortion at unity gain driving 600 ohms. About £2.50.

The ultimate for low noise is the LM797. A single amp per 8-pin package. Gain-bandwidth 450MHz, 1ppm distortion at unity gain driving 1kohm. But the par excellence is the noise - a mere 0.9nV/rootHz (the same noise as a 50 ohm resistor). Expensive - about £9.

[MrBungle]

TBH you can actually use veroboard for quite high frequency stuff in the 100's of MHz range. You have to do a lot more than just break the tracks though. Lots of isolation, peeling and solder filleting. However don't expect the design to be repeatable or not give you a headache.

On the subject of opamps, fundamentally they are the same beasts as they always have been but there are new characteristics and trade offs and nice ways of solving problems. Art of Electronics, already mentioned has probably the best set of explanations on the matter and on selection. Personally, I have found that apart from very specialist applications, it makes sense to grab a tube of opamps with GBW of about 2Mhz and not worry about it too much. TLC272CP are nice: cheap, dual, rail to rail, single rail and low bias current. I have a tube of those and a few single opamps from Linear with some varying special powers

RF ones are finicky and I'd rather just use discrete stuff there.

[Ed_Dinning]

Hi Gents, a good precursor to H&H is Electronic Circuits (Fundementals & Applications) Mike Tooley, pub Newnes.

Ed

[Diabolical Artificer]

Why do some audio designers prefer opamp's made from discrete components rather than opamp's themselves? This question was prompted by a comment that Rupert Neve made in a video. He said he'd had a look at just about every opamp for a new design he was working on, but dismissed them all in favour of the discreet approach.

A.

[MrBungle]

There is no reason to build an opamp with discrete components in a new product. Absolutely none. You instantly lose thermal compensation and low offset voltage. Noise figure is worse as well.

It's all high end audio woo. Neve is full of it. Regularly goes to talk about being able to hear spurs at 54KHz, golden ears and being frustrated if the product sounds wrong. Human perception is considerably more variable than measurement so unless you can back the former up with the latter and rigorous application of scientific method. you're talking rubbish.

[Radio Wrangler]

Discrete opamps?

Once upon a time the only way to make PNP transistors on a wafer that went through standard IC processes was the 'Lateral PNP' its performance was far inferior in gain and bandwidth to the NPNs which the same process made. This walloping great mismatch made a bit of a joke out of any attempt at using complementary symmetry.

Since then the playing field has been levelled by using some dramatic processes like making the IC upside down, depositing what will be its substrate on top, and then grinding off the other side!

A second advantage to the discrete opamp is that higher voltages can be used without as much loss of performance. In preamps, this was used to help score more brownie points for overload margin. This was overcome if you put the gain in the right places and planned signal levels properly. Active gain controls are a good weapon iin this armoury.

A third advantage is simply freedom. If someone doesn't already make what you want, you'll never afford the up front costs of getting a custom IC made, so just do it in discretes.

In power stages, the voltage issue wins. Very high power and very high current also favour the discrete approach.

On the other hand, the discrete approach misses out on some nice matching tricks the IC designer has on-tap. But the IC designer hasn't got accurate absolute values of resistor without difficulties and laser trimming.

You pays yer money....

David

Crosspost with MrBungle! Yup, you can do it either way now and make it work so the deciding factor is which will open the punter's wallets the furthest. A) "We bought this opamp from Imperial Circuits for $3.49" or B) "Our skilled designers spent many years listening to each and every discrete component in our artistically crafted preamp. Every capacitor has a 5-bar gate printed on it, every resistor is deposited on glass recovered from classic Leica lenses. The design team ate only wholefoods for a year before starting work"

B is the obvious winner

[Craig Sawyers]

Originally Posted by MrBungle

There is no reason to build an opamp with discrete components in a new product. Absolutely none. You instantly lose thermal compensation and low offset voltage. Noise figure is worse as well.

It's all high end audio woo. Neve is full of it. Regularly goes to talk about being able to hear spurs at 54KHz, golden ears and being frustrated if the product sounds wrong. Human perception is considerably more variable than measurement so unless you can back the former up with the latter and rigorous application of scientific method. you're talking rubbish.

The only way you can get extremely low input noise voltage is using discrete, at least in a front end. The best you can do with an op amp is the 797 I mentioned earlier at 0.9nV/rootHz. If you need less than that you have to start parallelling - and parallel 797's is an extremely expensive way to do it.

So you can either use current production FET's, such as the Linear Systems LSJ74, also 0.9nV/rootHz, but parallel four running at Idss and you get to 0.45nV/rootHz cost effectively. Or the THAT 320, which is four matched PNP in a single package each with 0.75nV/rootHz, so in parallel gets 0.375nV/rootHz. Finally a less well known low noise device is the BF862 JFET which costs very little but has 0.8nV/rootHz and eight are easily paralleled to get down to 0.28nV/rootHz - the same noise generated by a 4.7 ohm resistor.

Whether you stay discrete from then on, or simply follow the sort of front end above with a suitable op amp is up to you.

So it is not all audio woo. If you have a very low source impedance, and need a low noise figure you absolutely have to use discrete.

[MrBungle]

yes aware of this. The application is usually woo which is the problem. You see lots of people selling discrete op amps with low noise Japanese transistors blessed by the tears of Eno but if you look closely at the design, all those nice resistors contribute more thermal noise than the rest of the system and they don't publish noise figures, merely statements of experience using non quantifiable terms.

If you give me a noise figure and repeatable methods to test such as what you see towards the end of opamp data sheets (Linear for example usually have statistical distribution of characteristics) then you win.

[Craig Sawyers]

Originally Posted by MrBungle

yes aware of this. The application is usually woo which is the problem. You see lots of people selling discrete op amps with low noise Japanese transistors blessed by the tears of Eno but if you look closely at the design, all those nice resistors contribute more thermal noise than the rest of the system and they don't publish noise figures, merely statements of experience using non quantifiable terms.

Oh I totally agree. Any audio amplifier manufacturer, whether at a sub-system level such as discrete op amps through to the complete product are extremely cagey about detailed specifications. In the past that was not the case - Quad for example published detailed graphs showing distortion with power level and frequency, and signal to noise ratios back in the 50's and 60's.

Case in point - distortion in class AB power amps. If it is quoted at all it is usually at 1kHz. Which neatly gets around the fact that distortion rises by at least an order of magnitude by 20kHz.

However a good counterexample is the Hypex N-core class D module https://www.diyclassd.com/product/nc400/11 which is exceptionally detailed.

[emeritus]

The noise that can be generated by resistors does tend to get overlooked. In the 1970's Plessey had problems with a phase locked loop circuit that turned out to be caused by the noise generated in a resistor that was at a point in the R-C loop filter where the signal level was very low. The engineer who discovered the cause, made his name by replacing it with a low-noise equivalent that magically fixed the problem.

[Hartley118]

Originally Posted by Dr Wobble

Why do some audio designers prefer opamp's made from discrete components rather than opamp's themselves? This question was prompted by a comment that Rupert Neve made in a video. He said he'd had a look at just about every opamp for a new design he was working on, but dismissed them all in favour of the discreet approach.
A.

Unlike some of its competitors, the Neve company was originally nervous of using op amps in studio mixing desks: proprietary discrete op amp PCB modules had always been used.

In 1977, we were designing a new range of compact transportable desks (the 5422 and its successors), so ICs were very desirable from the packaging point of view. Rupert himself was involved in that design and had good contacts in Philips who had just released samples of its TDA1034 op amp, specifically designed for professional audio, with low noise, high slew rate and a 600 ohm load drive capability. It was at least as good as our discrete amps, so Rupert and I gave it the thumbs up. We quickly adopted it not only for new designs, but, by means of pin-compatible PCB modules, to replace the old discrete amps. It's now renamed the NE5534, NE5532 etc -power hungry, but a good performer.

Incidentally, the pin compatibility of the NE5534 and 741 enables us to carry out an interesting experiment on the subjective importance of high op amp slew rate.

We equipped a channel amplifier with 741s and compared it with the adjacent identical channel amp fitted with its usual NE5534s. We did this at George Martin's AIR studios in London using a close-miked 'hi hat' cymbal as a full bandwidth live sound source with lots of HF content with an aggressive slew rate to fully push the limitations of the 741s.

The surprising result was that none of the 'golden ears' present in the control room, listening on top quality Tannoy Dual Concentric monitors could hear any difference at all between the NE5534s and the cheapo 741s.

But we could never have sold top quality equipment full of 741s!

Martin

[MrBungle]

Originally Posted by Hartley118

The surprising result was that none of the 'golden ears' present in the control room, listening on top quality Tannoy Dual Concentric monitors could hear any difference at all between the NE5534s and the cheapo 741s.

This is the thing that gets me.

My brother was in a cheesy goth band in the late 1990s that released a couple of singles but didn't get much further. They used a Korg Trinity, a drum machine, a couple of guitars and a Mac all wired up with to a cheapo mixer. Just a couple of years back he got himself into the "high end audio" rubbish and apparently spending £10k brings new depths to the music they created (yes he's a narcissist as well).

I just dropped him an email and have asked if he's still got the mixer and can I borrow it. I'm going to see if it's got 741's in it. I bet it has. It was from Tandy. I'm sure that the reproduction off a digital PCM synthesizer, guitar, drum machine, routed through a Realistic mixer and into a cheap ADC, then burned to CD really has so much more detail, depth than it had when they were recording it