You have describe all of the basics (points 1-7) when dealing with IC op-amp implementation. That is beginners stuff. Point 8 is incongruous.
I use to believe the same thing as you and tried everything I could to optimise them, then built a discrete circuit that basically made that all a waste of time in terms of obtaining good sonic performance.
You quote Douglas Self - is that the same Douglas Self that did some papers on optimal biasing of output stages? With drawings showing the crossover region transfer function with various levels of bias current and the optimal amount? Around 30mA or in that region IIRC was optimal?
At the moment that i haven't a serious FFT analyzer, my only trustable measuring instruments are: a 50MHz Hameg DSO, a 10MHz Hameg programmable function generator and a precise Thandar true RMS DVM.
With certainty, i am not at the same scientific level in audio like you.
OTOH, my experience and my experiments showed that the response of a circuit in square waves is a very good way of extracting useful conclusions.
Well, according to your view, I must revisit (revise???) my test (measurement???) setup.
In what please?
Last edited:
I see you are using smaller film cap in parallel with an electolytic in your amplifier schematic for the power supply rail decoupling.
Do you do that with all of your circuits, as recommended in many data sheets an application notes from IC op-amp manufacturers?
Last edited:
I've offered one - see post 19. We have this observation that the JE-990 sounds better than opamps. One crucial difference is those inductors - opamps can't have on-chip inductors of sufficiently high value. Don't you think that's at least worth investigating?
abraxalito, how does that make it 'sound better'?
There are a few clear things that make some designs better sounding than others, and we know what they are - adequate bandwidth, adequate slew rate, low noise, no PSU artifacts and so on. I listed 8 or 9 key points above that need to be met in order to have half a chance in creating a decently engineered audio amplifier.
I have built discrete designs and plenty of op-amp based circuits, and other than times when I made a mistake or used the wrong type of op-amp, never (no never) heard anything like that which some of the detractors claim. I have also offered up the fact that almost all recorded music goes through opamps nowadays (dozens's of them, and electrolytic caps too by the way . . . ), and yet we cannot say that almost all recordings suffer from 'the opamp sound'. To then argue that the signal passing through just 1 or 2 opamps in the listeners living room screws the sound is just plain fallacious - its already been through dozens and dozens of them. Then we have the claim from Craig above that you can even hear op-amps when used in the servo circuits . . . well, you have to doing some pretty fundamental things wrong to have that situation. So, I just don't accept the notion that opamp based audio sounds fundamentally flawed or bad if you do it right. Again, let me stress that I am not against discrete circuits - I have an open mind on the subject - the door on opamps is not shut for me.
I don't want to get into a circular argument here, and I will probably not stay with this thread for much longer because like many things on diyAudio.com, people have very entrenched positions on these things and to argue is futile.
There are a few clear things that make some designs better sounding than others, and we know what they are - adequate bandwidth, adequate slew rate, low noise, no PSU artifacts and so on. I listed 8 or 9 key points above that need to be met in order to have half a chance in creating a decently engineered audio amplifier.
I have built discrete designs and plenty of op-amp based circuits, and other than times when I made a mistake or used the wrong type of op-amp, never (no never) heard anything like that which some of the detractors claim. I have also offered up the fact that almost all recorded music goes through opamps nowadays (dozens's of them, and electrolytic caps too by the way . . . ), and yet we cannot say that almost all recordings suffer from 'the opamp sound'. To then argue that the signal passing through just 1 or 2 opamps in the listeners living room screws the sound is just plain fallacious - its already been through dozens and dozens of them. Then we have the claim from Craig above that you can even hear op-amps when used in the servo circuits . . . well, you have to doing some pretty fundamental things wrong to have that situation. So, I just don't accept the notion that opamp based audio sounds fundamentally flawed or bad if you do it right. Again, let me stress that I am not against discrete circuits - I have an open mind on the subject - the door on opamps is not shut for me.
I don't want to get into a circular argument here, and I will probably not stay with this thread for much longer because like many things on diyAudio.com, people have very entrenched positions on these things and to argue is futile.
Bonsai, it could be that the system you use to listen to differences between IC op-amps and discrete designs is masked by its own distortions. Or maybe your hearing is not sensitive to certain types of distortion. Or your discrete designs were not of adequate quality to rival an IC op-amp.
IC op-amps have a lot of design limitations compared to discrete.
Firstly everthing is on one die. That die is subject to thermally induced distortion from the output stage if it starts to drive any current. With a discrete design, any instantaneous power disipation from the output stage is isolated to the output transistor and not thermally coupled to any of the other transistors or components in the circuit.
Secondly, IC op-amps rarely use resistors as resistors. They use P and N junctions. They do this because resistors take up a lot of space. Using the P and N substrates that are already there makes it more efficient to do it that way. Those IC resistors do not perform as well as good discrete resistors.
Thirdly, with IC op-amps you cant get inside the things and start optimising different areas like you can a discrete design. For example, with a discrete design you can bump the power supply rails up to get around some of the device parastics that are going to make your circuit less linear.
Then what about electromagnetic coupling in an IC?
There are some good recordings out there that haven't gone through op-amps. There are some people on the recording side that strive to improve sound quality by using good equipment.
There was nothing wrong with the implementation of the servo IC op-amps. They were done as per manufacturers application notes, thoroughly scoped and distortion analysed and then optimised in addition to manufacturers recommendations.
Bonsai, are you using silver mica caps for compensation in your amplifier or does the "SM" mean something else?
IC op-amps have a lot of design limitations compared to discrete.
Firstly everthing is on one die. That die is subject to thermally induced distortion from the output stage if it starts to drive any current. With a discrete design, any instantaneous power disipation from the output stage is isolated to the output transistor and not thermally coupled to any of the other transistors or components in the circuit.
Secondly, IC op-amps rarely use resistors as resistors. They use P and N junctions. They do this because resistors take up a lot of space. Using the P and N substrates that are already there makes it more efficient to do it that way. Those IC resistors do not perform as well as good discrete resistors.
Thirdly, with IC op-amps you cant get inside the things and start optimising different areas like you can a discrete design. For example, with a discrete design you can bump the power supply rails up to get around some of the device parastics that are going to make your circuit less linear.
Then what about electromagnetic coupling in an IC?
There are some good recordings out there that haven't gone through op-amps. There are some people on the recording side that strive to improve sound quality by using good equipment.
There was nothing wrong with the implementation of the servo IC op-amps. They were done as per manufacturers application notes, thoroughly scoped and distortion analysed and then optimised in addition to manufacturers recommendations.
Bonsai, are you using silver mica caps for compensation in your amplifier or does the "SM" mean something else?
Indeed SM = Silver Mica
Craig,
I have a reasonably good system, and I believe I am very even handed and honest in my assessment. I can easily discern the differences between good and bad recordings (where there is a huge amount of variability BTW). I place a lot of value in big, deep sound stage and a smooth top end - but that's about as subjective as I am going to get though.
I work in the semi industry, though not as an IC designer (I did work as an industrial instrumentation designer for 7 years earlier in my career), so some of the points you raise I recognize, but I believe they have little or no impact on opamp performance for audio.
Thermal distortion has been aound for a long time . . . but its not an issue on modern op-amps. In designs from the 70's and early eighties, the open loop gain (and therefore loop gain) was indeed modulated by the device dissipation (See e.g James Solomon's op-amp tutorial for a discussion on this), and there were offset/drift related issues as well, although these were not really a concern for audio. But, op-amps from especially the 70's were notoriously bad for audio, other than the 5532 and 5534 which changed everything when they came along in the late '70's. I still remember a pre-amp I designed and built in '77or 78 using LF356's which I later modified to use 5534's - the improvement was huge. Now, the original preamp was gritty and not good - but who designs with devices like that today? No one - things have moved on. After all, I dont see people complaining about their crappy Comodore 64's or Sinclair 80's anymore - we're all on Apples or Window's machines in 2011. Same thing with opamps - things have moved on!
EMC/EMI. The op-amps have the advantage here. The circuit loops (and thus radiating and receiving areas) are minute and measured in general in microns, other than the bond outs (S08 package) which of course are measured in single digit mm's. You cannot match this with any discrete design.
Linearity. The LM4562 is around 300ppb or close to -140db. And, there are quite a few other devices from ADI, LT and Maxim that are also down near this level. There are very few discrete circuits that are going to better that if any for $4 (theres a challenge for someone!). Even the 30 year old 5532/5534 is no slouch when dealing with reasonable loads (i.e. 600 ohms and above). Therefore, I'd say the opamps have the advantage here too. And, forget that these devices in general have very superior PSRR going all the way out to >20KHz.
Resistors. Most IC resistors are of the polysilicon variety. But, in an IC process with well matched transistors easily acheived on the same die, there is no need for highly accurate resistors, and in most cses they don't even have to track accurately. Accurate, stable current sources and mirors are easily acheived - you do not need accurate absolute resistor values for this. In many cases, a single resistor will set up a current reference into a mirror that all the current source loads in an op-amp are referenced off. As for things like degeneration resistors, where these are occasionaly used in an IC opamp, they will be phyically next to each other and laid out symmetrically - i.e., they will track over time an temperature very closely. Take a look at the paramter spreads on some opamp data sheets to get a feel for just how tight they are. And this is repeatable across thousands of devices over long periods.
I will agree that on discrete designs you can acheive higher output levels and are not limited to the typical +-15V; further, you can drive more output current and for ultra low noise applications, you can acheive better performance (see Syno8's designs for example). But, these are specialist application areas. You need low noise for MM and MC input stages, and for things like headphones, the ability to drive substantial current into the load is important. For higher supply rails - well this requirement is questionable when almost every power amp on the planet will easily be driven to full output with 2VRMS input signal. I do see high rails on discrete designs but this is to achieve high linearity usually - something good opamps can do easily at +-15V.
Craig,
I have a reasonably good system, and I believe I am very even handed and honest in my assessment. I can easily discern the differences between good and bad recordings (where there is a huge amount of variability BTW). I place a lot of value in big, deep sound stage and a smooth top end - but that's about as subjective as I am going to get though.
I work in the semi industry, though not as an IC designer (I did work as an industrial instrumentation designer for 7 years earlier in my career), so some of the points you raise I recognize, but I believe they have little or no impact on opamp performance for audio.
Thermal distortion has been aound for a long time . . . but its not an issue on modern op-amps. In designs from the 70's and early eighties, the open loop gain (and therefore loop gain) was indeed modulated by the device dissipation (See e.g James Solomon's op-amp tutorial for a discussion on this), and there were offset/drift related issues as well, although these were not really a concern for audio. But, op-amps from especially the 70's were notoriously bad for audio, other than the 5532 and 5534 which changed everything when they came along in the late '70's. I still remember a pre-amp I designed and built in '77or 78 using LF356's which I later modified to use 5534's - the improvement was huge. Now, the original preamp was gritty and not good - but who designs with devices like that today? No one - things have moved on. After all, I dont see people complaining about their crappy Comodore 64's or Sinclair 80's anymore - we're all on Apples or Window's machines in 2011. Same thing with opamps - things have moved on!
EMC/EMI. The op-amps have the advantage here. The circuit loops (and thus radiating and receiving areas) are minute and measured in general in microns, other than the bond outs (S08 package) which of course are measured in single digit mm's. You cannot match this with any discrete design.
Linearity. The LM4562 is around 300ppb or close to -140db. And, there are quite a few other devices from ADI, LT and Maxim that are also down near this level. There are very few discrete circuits that are going to better that if any for $4 (theres a challenge for someone!). Even the 30 year old 5532/5534 is no slouch when dealing with reasonable loads (i.e. 600 ohms and above). Therefore, I'd say the opamps have the advantage here too. And, forget that these devices in general have very superior PSRR going all the way out to >20KHz.
Resistors. Most IC resistors are of the polysilicon variety. But, in an IC process with well matched transistors easily acheived on the same die, there is no need for highly accurate resistors, and in most cses they don't even have to track accurately. Accurate, stable current sources and mirors are easily acheived - you do not need accurate absolute resistor values for this. In many cases, a single resistor will set up a current reference into a mirror that all the current source loads in an op-amp are referenced off. As for things like degeneration resistors, where these are occasionaly used in an IC opamp, they will be phyically next to each other and laid out symmetrically - i.e., they will track over time an temperature very closely. Take a look at the paramter spreads on some opamp data sheets to get a feel for just how tight they are. And this is repeatable across thousands of devices over long periods.
I will agree that on discrete designs you can acheive higher output levels and are not limited to the typical +-15V; further, you can drive more output current and for ultra low noise applications, you can acheive better performance (see Syno8's designs for example). But, these are specialist application areas. You need low noise for MM and MC input stages, and for things like headphones, the ability to drive substantial current into the load is important. For higher supply rails - well this requirement is questionable when almost every power amp on the planet will easily be driven to full output with 2VRMS input signal. I do see high rails on discrete designs but this is to achieve high linearity usually - something good opamps can do easily at +-15V.
Last edited:
Well, if the last one or two opamp stages in the listners amp are badly engineered then you are correct. If they are well engineered, all other things being equal, I believe my point stands.
I was referring to the imperfect medium inbetween the two.
Not equivalent to a series string of opamps.
Other than that, given the huge improvement of IC opamps over the last 20-25 years, it's questionable why some folks bother with discrete opamps.
Even if there's a measurable/audible difference, the majority doesn't have the means for the entire required chain to make it noticeable.
As convincing as the techy ins&outs talk on CD versus TT, with medium grade consumer gear as reference.
Cost is the number 1 criterium for any application, for every thousand minus one, anywhere.
Very likely, even a fair share of the OPA627/AD797/LM4562 fiddlers could do with a serious investment in peripherals.
(personally, i don't give a hoot whether it's SS or valve, digital or analogue, global or local nfb, discrete or integrated circuit, merely eyeball what exits)
Not equivalent to a series string of opamps.
Other than that, given the huge improvement of IC opamps over the last 20-25 years, it's questionable why some folks bother with discrete opamps.
Even if there's a measurable/audible difference, the majority doesn't have the means for the entire required chain to make it noticeable.
As convincing as the techy ins&outs talk on CD versus TT, with medium grade consumer gear as reference.
Cost is the number 1 criterium for any application, for every thousand minus one, anywhere.
Very likely, even a fair share of the OPA627/AD797/LM4562 fiddlers could do with a serious investment in peripherals.
(personally, i don't give a hoot whether it's SS or valve, digital or analogue, global or local nfb, discrete or integrated circuit, merely eyeball what exits)
Last edited:
Well I'm speculating rather as to the reasons as I've no hard data in terms of measurements yet. But its quite possible the presence of the inductors reduces the susceptibility to out of band interference.
I've no beef with your list of points myself, I would merely note that you didn't talk about grounding issues - these I've found to be crucial when aiming for the best sound. I'm not the only person who's noticed this - Bas for example concurs that the grounding arrangements make a significant difference to the sound quality.
I agree, and have worked for a console company in the past. There's a crucial couple of differences - the first is that professionals work with balanced inputs, so grounding is far easier to get right. The second is that in a console, the opamps are in general (prior to the mix bus where most of them reside, in the channel strip) dealing with just the sound from a single instrument or voice because that's how recordings are made. So intermodulation performance is not such a huge issue as with the mixed recording.
Your argument against looks fallacious to me - it assumes that the problem lies in the opamps, not the way the opamps are employed in the two scenarios. But I'm a big fan of opamps myself - there's nothing intrinsically wrong with the chips, just the issues with regards to layout to obtain a clean sound with single ended inputs aren't so widely known.
I'd want to see more of the topology - layout and grounding to comment on that.
On this then we're in agreement, just we disagree about what particular precautions are needed to indeed 'do it right'. Your list is not exhaustive but its a good start.
IMHO the two inductors are likely to be a trick to get a higher phase margin (or to get a reasonable phase margin at all). With the values shown the input LTP works with no degeneration at all below ~20 kHz, and with full degeneration beyond 1MHz or so - probably out of band interference could be better - and easily - addressed by proper input low-pass filtering.
Disregarding the inductors, the JE-990 is a pretty standard design: high open loop gain (~120dB), with a -3dB corner in the 20-30Hz range, and a GBW product in the MHz range - a plain 5534 properly compensated is faster; and noisier, probably, although not an order of magnitude noisier. A 5534 certainly has less driving capabilities, but if you are not planning to direct drive low impedance headphones rail-to-rail, or unless you don't need very low noise, then there's no point in going discrete to achieve overall * less * speed. Moreover, an amplifying module based on the JE-990 relies strongly on feedback, and layout, and... to perform well - the same golden rules apply to a well designed 5534-based module: it's very hard to figure out how two identical topologies, based on the same key-concepts design, could sound perceptibly different in discrete or monolithic form.
L.
Last edited:
'Proper' being the operative word here. Its often the case that designers implement low-pass filters with a series R and a C to a dirty ground. That dirty ground comes about because there's common-mode noise coming down the screen of the cable cable from the source component and onto the 'cold' side of the capacitor. Such an LP filter makes matters worse than merely a series resistor because the cap is a very low impedance at RF.
"I've no beef with your list of points myself, I would merely note that you didn't talk about grounding issues - these I've found to be crucial when aiming for the best sound. I'm not the only person who's noticed this - Bas for example concurs that the grounding arrangements make a significant difference to the sound quality."
I do not disagree, but I assumed this was a given in the discussion. Its important to note that good grounding and layout apply equally to discrete and opamp based designs if you are going to get a good result.
WRT balanced inputs, yes, in the professional world this is the standard, but is lack of balanced inputs in the last 1 or two stages of the signal chain be the reason for the 'harsh, gritty' sound of op-amps? I don't think this can be a factor, assuming good layout.
I would say the use of cheap, sub-optimal opamps in low end consumer gear (e.g. on the output of a DAC, or in a tone control stage - 4558 comes to mind - oftem running of low supply rails) can be a cause of bad sound. And to Jacco's point, in consumer equipment every single cent counts - I work with some of these companies (very high volume) and 1c is 1c - they'll flip to the cheaper solution at the drop of a hat. But, you cannot condem opamps here - its the engineer who is forced to select a low grade opamp to keep within budget that's the problem.
I do not disagree, but I assumed this was a given in the discussion. Its important to note that good grounding and layout apply equally to discrete and opamp based designs if you are going to get a good result.
WRT balanced inputs, yes, in the professional world this is the standard, but is lack of balanced inputs in the last 1 or two stages of the signal chain be the reason for the 'harsh, gritty' sound of op-amps? I don't think this can be a factor, assuming good layout.
I would say the use of cheap, sub-optimal opamps in low end consumer gear (e.g. on the output of a DAC, or in a tone control stage - 4558 comes to mind - oftem running of low supply rails) can be a cause of bad sound. And to Jacco's point, in consumer equipment every single cent counts - I work with some of these companies (very high volume) and 1c is 1c - they'll flip to the cheaper solution at the drop of a hat. But, you cannot condem opamps here - its the engineer who is forced to select a low grade opamp to keep within budget that's the problem.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.