Which OEMs Actually Use The OPA627 / Other Uber-High Slew Rate OAs?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

thermionic

Well-known member
Joined
Jun 3, 2004
Messages
1,671
Excuse my ignorance, but off top of head I can't think of too many OEMs who use the OPA627. In fact, I can't think of many OEMs who use much tamer modern OAs such as the LM4562. It seems to me that modern uber-fast OAs require so much effort in terms of board layout and general design skill that if an OEM has the knowledge, they're likely to go with a discrete design - it would make for better ad copy than IC-based, right?

I wonder how many DIY-ers who fit these OAs have access to before-and-after FFT plots, or even a basic THD+noise average figure? The uber-opamps *should* give better figures than the 5532 etc, but in reality, whose design can exploit it? How many people have shown you plots / figures to prove the design benefits the uber-OA? There are many in the DIY fraternity who claim they can *hear* an improvement, but did they really lower THD /noise? Many claim they can hear cables, so subjectively-evaluated claims aren't remotely plausible. 
 
Distortion measurements for the OPA627 (and the LM4562/LME49860) are available in my opamp distortion paper (opamp_distortion.pdf). In fact the LM4562 is, overall, the better audio opamp unless you need the low current noise/input bias.

The OPA627 is a relatively old part and thus expensive (probably manufactured on an old process). I'm using the OPA827 instead which is, overall, a step forward for most applications.

Samuel
 
PS: High slew-rate is not usually a suitable indicator for good audio performance. Perhaps rather the opposite--high slew-rate amplifiers usually have to accept compromises which reduce audio performance.
 
Samuel Groner said:
PS: High slew-rate is not usually a suitable indicator for good audio performance. Perhaps rather the opposite--high slew-rate amplifiers usually have to accept compromises which reduce audio performance.

Could you elaborate on the compromises as you see them, Samuel? Are you suggesting that, in order to keep the OAs from oscillating, they have to compromise THD and / or noise? This would mirror my own experience. Twitchy ICs need filtering and compensation, which often comes at a price. 

All the data I've read suggests that talk of slew rate is marketing. I mentioned it in the thread title as it does seem that certain OAs are marketed by SR specs. I would assert that, in the hands of the average engineer, high-SR OAs (such as the 627) are likely to oscillate and deliver inferior performance to jellybean OAs such as the 5532.

BTW - Examples of gear that uses relatively exotic ICs would be welcome. Most gear I can think of tends to use 5532/34, OPA2134 and the like (not that there's anything wrong with these ICs - that's not my point)
 
The tendency for oscillation is more related to high unity-gain bandwidth than high slew-rate; of course opamps with high unity-gain frequency tend to have higher slew-rate (because of the smaller compensation capacitor), but that's more a side-effect.

There are numerous subtle trade-offs in operational amplifier design; a fundamental consideration is the selection of the input stage topology. For low noise and good low-/audio-frequency precision/distortion performance, an undegenerated or only modestly degenerated bipolar differential pair is hard to beat. Also the collector current of the input stage is fixed to, say, the 0.5-3 mA range in order to avoid excess current noise. With the with a current 36 V process achievable unity-gain bandwidth of perhaps 20-50 MHz, the slew-rate is bound to be in the 10-30 V/us range.

For higher slew-rate, the degeneration of the input stage can be increased (and thus the compensation capacitor decreased). This however increases voltage noise and pronounces distortion contributions from following stage (e.g. crossover distortion from the output stage). The situation is similar with JFET input stages. Then there are more advanced input stages which e.g. dynamically boost their bias current to achieve high slew-rate. Typically these have again secondary effects on (low-/audio-frequency) distortion and noise which limits their usefulness.

Samuel
 
In addition to what Sam has posted I recall seeing some dedicated S/H op amps decades ago that were optimized to slew blazingly fast when a "sampling" step is presented to the input, then settle quickly for the "hold" after it gets there. This has precious little to do with actual audio performance, since during the step the op amp is slewing open loop as fast as it can, without actual NF. IIRC the internal design of dedicated S/H op amps actually bypasses around the input LTP for large steps. (Note: this was back when they still published schematics for inside the ICs).

Extremely high slew rate op amps are like bayonets for your rifle... It means you have screwed up and run out of ammunition. Good design IMO involves rise-time limited audio paths that never come close to slew limiting your chosen op amp. Slew limiting is distortion, or clipping in the "rate of change" domain. More is generally better but when you give something else up to get more speed than you need that is bad engineering. The large signal power bandwidth will generally tell you when you have enough slew rate, and even that is conservative since very high frequency audio signals are rarely loud too, so you probably won't encounter 20kHz full scale audio signals, while a premium design should pass that cleanly. IMO

I defer to Sam regarding the state of the modern uber op amps, I have long since stopped chasing the next best thing. In my judgement by the 1980 op amps were fast enough for audio, while they have gotten better in other ways. YMMV

JR
 
Excellent explanations as usual, Samuel, as usual.

I have been told from trustworthy source that Merging (Pyramix) Horus uses newer TI Opamps from the OPA 16xx range - also in their mic preamp stage. They claim superior performance (of course), but published specs at least are excellent.

on many designs the well chosen compromise between voltage noise and current noise of a NE5532 / 5534 is hard to beat, unless you can tweak your design for lowest impedances. THD of well designed NE5532 stages is really low as well, so that in most cases PCB and circuit design will degrade audio performance first... on application with higher output current requirements other opamps will beat the oldie though, or if you need a FET input.

- Michael
 
audiomixer said:
I have been told from trustworthy source that Merging (Pyramix) Horus uses newer TI Opamps from the OPA 16xx range - also in their mic preamp stage. They claim superior performance (of course), but published specs at least are excellent.

In the last few years, I've extensively used OPA1611/1612/1632 for commercial designs as well. I find that they have, for the typical ~1k source Z we find in line-level audio applications, a good voltage-current noise compromise. The distortion performance is also pretty good, for most applications.

Samuel
 
thermionic said:
... if an OEM has the knowledge, they're likely to go with a discrete design - it would make for better ad copy than IC-based, right?
If an OEM has the knowledge, they are likely to use that know how to concentrate on layout, decoupling & grounding with 553x.

Ad copy is better served by claiming your Mains Cables are Hand Carved from solid Unobtainium by Virgins.  8)

Kingston had an excellent thread on what is REALLY important for noise & THD; opamps and local decoupling of rails, some questions

http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

Da uber OPAs will only show their stuff on the very simplest circuits.  Once you get beyond a single OPA, the earthing, layout & decoupling take priority.  Here's a simple circuit which demonstrates this.  Note that it achieves SOTA performance with what many consider yucky OPAs.

http://nwavguy.blogspot.com.au/2011/07/o2-design-process.html

You should follow his links on OPAs too.

Some of the uber OPAs have yucky stuff which isn't shown on the datasheets ... or even picked up by the gurus.  eg LM4562 & its more expensive siblings have a nasty latching behaviour on single supply applications.

But OPAx134 though it's damned with faint praise by many gurus, has subtle characteristics which often gives it better THD bla bla in real life.  It is VERY resistant to RFI and is probably the least fussy of the uber OPAs to decoupling.  Doesn't have 1nV/rtHz noise though.

There's good reason why 553x is pre-eminent.

As always, horses for courses.  I can only think of one obscure application where OPA627 would be preferred.
 
If your slewrate, V/us, is as big as your peak signal level, V, and you are not dragging a lot of ultrasonics around, you are fine.

This suggests 3V/uS to 20V/uS as all the slew you need for line-level audio.

Speaker amps might use more; they often don't; which makes uber-slew in the console a bit moot.

Aside from the re-compensation to get big slew numbers, there are (as Samuel says) hot-rod input stages which gear-shift to "Class C" to get big jolts of slew current. These give better specs when the job is to slew, settle, and then measure. They must do awful things to fast audio.
 
But OPAx134 though it's damned with faint praise by many gurus, has subtle characteristics which often gives it better THD bla bla in real life.  It is VERY resistant to RFI and is probably the least fussy of the uber OPAs to decoupling.  Doesn't have 1nV/rtHz noise though.

I really don't get why the OPA2134 gets so little praise. It's quiet enough for most applications, is capable of respectable THD performance and can drive well below 600R without any nasties. I regard it as a classic opamp. I've also found it to be very docile; you have to be really clumsy with implementation to get it to misbehave. It seems quite happy on single-sided PCBs with basic decoupling. Subjectively, I think of it being totally 'vanilla'. Projects I've used it in have made their new owners very happy. I'd be very upset if they discontinued the 2134.

Some of the uber OPAs have yucky stuff which isn't shown on the datasheets ... or even picked up by the gurus.  eg LM4562 & its more expensive siblings have a nasty latching behaviour on single supply applications.

Experience would suggest this to be the case. I'm convinced that few uber OAs are as stable as the data sheet would suggest.

I've spent the last few days working on a design that has a single gain stage. I tried 5532 (Signetics and TI - I've found the Sig part consistently to measure better THD, rather depressingly...), OP275, OPA2134 and LM4562. I've had to revise the layout 3 times. On the first 2 layouts, the FET OAs were stable, passing a good square, and supplying 'ok', if not stellar, noise / THD figures. The 5532s were twitchy on the 1st layout, but on the 2nd layout played ball nicely, coming in around 3-4dB quieter than their FET-input siblings (as expected for the design).

As for the 4562... On the first 2 layouts it was far noisier and higher in THD than either 5532 type, particularly the Sig (18 Mhz low-level oscillation!). This became frustrating as I'd studied the datasheet and couldn't think of what I hadn't done. In theory 'it should be better'. It was twitchy, with ringing evident. For the 3rd layout, I didn't change any values, and even used a lower value cap in the FB loop, but I did a much neater layout... Voila! The 4562 is now around 3-4dB better than the Sig 5532 in all regards, but I really had to work to get this improvement (I was expecting a tiny improvement, not 3-4 dB).

Dropping a 4562 into a circuit that's been designed around a 5532 is asking for trouble. I would expect noise to increase and stability to decrease. Modern opamps need the layout designed around them. Dropping them into a layout designed for a 5532 is akin to putting a Porsche engine in your Escort. It won't have the brakes...
 
thermionic said:
I've spent the last few days working on a design that has a single gain stage. I tried 5532 (Signetics and TI - I've found the Sig part consistently to measure better THD, rather depressingly...), OP275, OPA2134 and LM4562. I've had to revise the layout 3 times. On the first 2 layouts, the FET OAs were stable, passing a good square, and supplying 'ok', if not stellar, noise / THD figures. The 5532s were twitchy on the 1st layout, but on the 2nd layout played ball nicely, coming in around 3-4dB quieter than their FET-input siblings (as expected for the design).

As for the 4562... On the first 2 layouts it was far noisier and higher in THD than either 5532 type, particularly the Sig (18 Mhz low-level oscillation!). This became frustrating as I'd studied the datasheet and couldn't think of what I hadn't done. In theory 'it should be better'. It was twitchy, with ringing evident. For the 3rd layout, I didn't change any values, and even used a lower value cap in the FB loop, but I did a much neater layout... Voila! The 4562 is now around 3-4dB better than the Sig 5532 in all regards, but I really had to work to get this improvement (I was expecting a tiny improvement, not 3-4 dB).
two questions on this

- how do you check for 18Mhz oscillation? would probing with a scope not affect that?
- could you help educate us all and post a snippet of the three layouts? just the immediate surrounding of the opamp and may be your initial thoughts? there is a lot of info on topology, grounding and so in books and the internet, but very little on real PCB implementation...

- Michael
 
Hi Micheal,

The oscillation was seen by connecting a probe to the o/p socket, with the i/p socket shorted. My audio analyser picked up broadband fuzz at around -50-ish. I then hooked the scope probe and saw 18mV of something sine-shaped at 18Mhz.

The layout is on stripboard! The next stage is to give it to the PCB designer I work with. It seems daft to prototype anything with a 'relatively' quick opamp such as the 4562 on strip-board, but you have to get proof of concept somewhere, right? With anything faster I suspect you'd have to go for a multi-layer PCB from the start. Just the inductance of an IC socket can set off some OAs.

edit - trust me, you don't need to see the layouts. The moral of the story is that just a couple of mm here and there can trigger bad behaviour if the OA is quite fast.
 
I know stripboard ugliness in full detail -  ;)

I have had a few go's at the same LM4652 on stripboard - I would wonder if it oscillates as well....
I have been trying to get a quicker turn-around well homemade PCBs, but not really successful up to now. I dream of a UV laser based printer to have photo sensitive PCBs directly exposed.... I have started to go for lately more then through hole, but that makes testing a bit less straight froward difficult, and modding component values even more so.

I have heard from Merging that small component placement optimization can be beneficial as well... there is a paper in some of these aspects at AP, but again on theoretical aspects without one to one examples....

to bad you can't dig in deeper on the changes you made, but I understand, stripboard is always a mess (may be besides uwe beis's work.... )

- Michael
 
How old are your TI 5532s? I've found the ones I bought this year to be much better than the ones from 3-6 years ago for some reason. So much so that I exchanged them in all my gear. According to Douglas Self they're also easily damaged by applying only + or - voltage accidentally. The part still works but specs are degraded.

 
living sounds said:
How old are your TI 5532s? I've found the ones I bought this year to be much better than the ones from 3-6 years ago for some reason. So much so that I exchanged them in all my gear. According to Douglas Self they're also easily damaged by applying only + or - voltage accidentally. The part still works but specs are degraded.

More likely process improvements in modern IC foundries might be responsible, while I would expect the performance benefits to be modest. Some of the same process parameters that affect yield and feature size, could benefit old parts too.

I can imagine audible but modest noise floor improvements. Back while I was working at Peavey we actually had a special house part number for a 5532 that was selected for noise spectrum by the manufacturers, so process apparently can affect the noise floor. A much cleaner more uniform process might do even better than our decades old selected parts. 

JR

PS: DS may be talking about how bipolar junctions when allowed to zener can lose Hfe and get noisier. These degraded junctions can be annealed after injury to somewhat restore previous performance but good practice is to just not allow low noise transistor junctions to zener. You may notice small signal diodes, reverse connected across base-emitter junctions in low noise input designs to prevent reverse voltages larger than a diode drop (zenering generally requires 6V reverse bias or more). Sometimes parts can be degraded by things as simple as poor PS cycling when one rail comes up before the other, stressing some junctions in reverse inadvertently. Mic preamps with phantom blocking caps need to be clamped to prevent mischief.
 
living sounds said:
How old are your TI 5532s? I've found the ones I bought this year to be much better than the ones from 3-6 years ago for some reason.
What was better?

Doug Self did a 10W amp for Elektor which used a zillion 5532s as the output stage.  As part of that exercise, they measured all available 5532s.  The TIs had slightly worse THD20k.

Sam Groner measured a slightly smaller sample and found similar results.

Apparently, there are only 3 sets of 5532/4 masks in circulation.  The TI set is known to be slightly inferior.

I had a similar experience in Jurassic times when all 5532/4s came from the Mullard, Southampton plant.  There was a shortage of 5534ANs, the selected LN version, so we set up a jig to screen them.  We found every single 5534/2 easily met the tighter AN spec.  I must have measured at least a hundred myself.

A decade later, when I was again involved in electrical stuff, I was disappointed to find you couldn't rely on this.

Mullard (Philips, UK) is long gone so no more 5532/4s hand carved from solid Unobtainium by Southampton virgins  :(
 
The mask is not the only variable... While in an ideal world every part from a production run would be identical and with modern process today that is closer to the case but we are talking about an IC design that has been manufactured for some 4 decades or so. I would be surprised if the part has not been retooled multiple times for die shrinks and/or to tweak for process changes. While I doubt any manufacturer wants to spend too much money on an old proved design, but the process lines have probably changed a lot since back then (just wafer diameter for one obvious change). 

Back in the day ( a mere 2 decades ago) performance within a given lot was variable enough that we paid extra for selected parts. As I've mentioned before that house number was not selecting for absolute amount of noise but a noise spectrum shape or ratio between spot frequency noise, I suspect to weed out elevated 1/F suspects (? I did not spec that particular part, I just used it because it was already in the system when I was hired). 

Back in the '70s for my kit company I performed 100% testing on all***ICs that I resold and I did it because I found significant numbers that did not work up to expectations, By the mid-late '80s the reliability coming from major IC makers (Like TI, Signetics, et al)  was good enough that I dropped 100% screening (20 years ago).

I am not making a strong argument that modern IC process is variable, but IMO all bets are off for measurement or comparison experiments made a few decades ago, unless they tested statistically significant numbers of parts.

JR

*** All ICs from US chip companies, I never had a single bad part from Matshustia (japanese) the company I bought early BBD delay chips from since the mid '70s. Reticon a US company made a higher performance BBD chip that was noticeably lower quality/reliability typical of US IC companies back then. 
 
Dunno whether its true today but certainly for LN, cleanliness is everything.  It's all about sh*t in the chip.

The same Mullard Southampton factory made the Mullard BFW11, my old favourite FET for condensor mikes.  No other maker's BFW11 was anywhere near its LN performance.  Today, no BFW11 is a LN device.
 
ricardo said:
What was better?

Doug Self did a 10W amp for Elektor which used a zillion 5532s as the output stage.  As part of that exercise, they measured all available 5532s.  The TIs had slightly worse THD20k.

Sam Groner measured a slightly smaller sample and found similar results.

Apparently, there are only 3 sets of 5532/4 masks in circulation.  The TI set is known to be slightly inferior.

I had a similar experience in Jurassic times when all 5532/4s came from the Mullard, Southampton plant.  There was a shortage of 5534ANs, the selected LN version, so we set up a jig to screen them.  We found every single 5534/2 easily met the tighter AN spec.  I must have measured at least a hundred myself.

A decade later, when I was again involved in electrical stuff, I was disappointed to find you couldn't rely on this.

Mullard (Philips, UK) is long gone so no more 5532/4s hand carved from solid Unobtainium by Southampton virgins  :(

I was referencing what Douglas wrote about how a 5532 can be easily damaged by voltage applied the wrong way. From what I remember this supposedly not only increases noise, but also THD.

The new ones measured better in terms of THD in audio circuits. And they were audibly better, clearer, more defined low end etc. This was universal from console to power amp input stage to compressors, EQs,...

Maybe TI refined the changes that had to be made due to ROHS. These new op amps also tested and sounded better than the chips from the 80s I've got.
 
Back
Top