Hi Daniel
It's a long process of unravelling all the many inter-related techniques of harmonic distortion "profiling" that are now used by designers and DIYs alike to arrive at worthwhile improvements in our listening experience. Efforts tend to be focused on the power amplifiers, where the effects are generally more pronounced due to higher currents but this is probably wandering too far off-topic. Suffice to say that IMHO, basic stereo and larger 2.1 systems are as far as it is worth applying this sort of skulduggery before you start conflicts with DSP and multi-channel imaging.
It's a long process of unravelling all the many inter-related techniques of harmonic distortion "profiling" that are now used by designers and DIYs alike to arrive at worthwhile improvements in our listening experience. Efforts tend to be focused on the power amplifiers, where the effects are generally more pronounced due to higher currents but this is probably wandering too far off-topic. Suffice to say that IMHO, basic stereo and larger 2.1 systems are as far as it is worth applying this sort of skulduggery before you start conflicts with DSP and multi-channel imaging.
I use monophonic for soundfield size observations, with 1 speaker. Adding the distracting complexity of more channels would not be conducive to accuracy in observations.
You're not off topic. Mooly's resistor and my very similar contraption both do make noise. However, both do make for gigantic and more involving presentations, well able to fill a house from just 1 speaker located far off in the spare room, yet crystal clear audio all over the house. That's a good compromise.
Mine is doing multi-compensation as the pre-existing compensations of the design were not removed. As you turn up the volume, it starts to engage dynamically. My circuit is "off" during quiet transients. Well, it is not quite off--Those diodes are capacitive. It is a miniature RC. Except for that bit, I've altered only louder transients.
It was like removing earplugs, or a lot more like turning off a fan.
And, like Mooly indicated:
This is quite puzzling for an already excellent amplifier.
P.S.
Audio like strong wind, full blast, without discomfort, is usually an impedance thing and usually done with some sort of multi-pass circuit. Does either Mooly's circuit or mine have two paths for the source signal to get amplified? I didn't do this intentionally (this time), but would like to know if it exists.
I built a preamp back in '83 with all 5534's. The top notch EE I worked with at Tektronix at the time told me to just put a 50k R from the output to the negative input instead of a short, for the non-inverting topology, which depended on the internal capacitance of the chip and it worked real good. Then I worked in Engineering at Dolby Labs, and learned a few more tricks.
If you're source is digital, put a passive Rf filter ahead of the opamp circuit. I usually use a 1-10K R in series and a roughly 300pF - 1nF cap to Gnd. Keep all leads as short as is practical. Put 0.1uF caps within an inch of the chip bypassing the power supply (closer is better). Put a 100-200 ohm R in series with the output of the opamp circuit, outside of the feedback loop, to reduce the effects of reactance in the load. Only then are you ready to choose a feedback C, or whatever, to get the best phase margin.
The idea is to roll off the "loop gain" (open loop gain minus closed loop gain) with one pole before the other poles come in to play causing additional phase shift. You want the open loop gain to be rolled off to what the closed loop gain will be, with only the phase shift of a one pole filter.
Most people set the cap across the feedback R to cause a -3dB point at roughly 200kHZ, so there's still a lot of distortion correcting feedback at 20kHZ. If you set that higher than 200kHZ, you start to run the risk of other poles getting involved and causing the phase shift to be too much. Once you think you have a reasonable guess at what the cap should be, stick it in and drive the circuit with a 10kHZ squarewave, and look closely at the leading edge corner of the waveshape. Compare it with the output of the generator in case the generator isn't perfect. It may be wise to do this at both highlevel and low level signal drive. Ringing should be very minimal. If it's barely stable, ringing will be substantial.
Then, push the opamp circuit just barely into clipping with a sinewave signal. Move it in and out of clipping slowly. Feedback turns to shyte when the opamp is in clipping, and sometimes it will spuriously oscillate as it comes out of clipping (poor recovery). If so, you might want to experiment further.
In one case I had to also put a tiny cap across the two inputs of a poweramp IC (LM3886) because on the negative half cycle there was a small amount of spurious oscillation as the signal came out of clipping. That's a different mechanism, but still important and affected by the linear mode phase compensation.
That's how I do it. I hope this helps.
Ever since I heard the Linkwitz Labs Orion speaker system, which has many OPA2134's in the signal path, I'm sold on the OPA2134 dual opamps. If they don't sound perfect, it's not likely the opamp itself that's at fault. The 553X's are great too, but have bipolar input transistors, so will load the source a little more, which may not matter in most cases.
If you're source is digital, put a passive Rf filter ahead of the opamp circuit. I usually use a 1-10K R in series and a roughly 300pF - 1nF cap to Gnd. Keep all leads as short as is practical. Put 0.1uF caps within an inch of the chip bypassing the power supply (closer is better). Put a 100-200 ohm R in series with the output of the opamp circuit, outside of the feedback loop, to reduce the effects of reactance in the load. Only then are you ready to choose a feedback C, or whatever, to get the best phase margin.
The idea is to roll off the "loop gain" (open loop gain minus closed loop gain) with one pole before the other poles come in to play causing additional phase shift. You want the open loop gain to be rolled off to what the closed loop gain will be, with only the phase shift of a one pole filter.
Most people set the cap across the feedback R to cause a -3dB point at roughly 200kHZ, so there's still a lot of distortion correcting feedback at 20kHZ. If you set that higher than 200kHZ, you start to run the risk of other poles getting involved and causing the phase shift to be too much. Once you think you have a reasonable guess at what the cap should be, stick it in and drive the circuit with a 10kHZ squarewave, and look closely at the leading edge corner of the waveshape. Compare it with the output of the generator in case the generator isn't perfect. It may be wise to do this at both highlevel and low level signal drive. Ringing should be very minimal. If it's barely stable, ringing will be substantial.
Then, push the opamp circuit just barely into clipping with a sinewave signal. Move it in and out of clipping slowly. Feedback turns to shyte when the opamp is in clipping, and sometimes it will spuriously oscillate as it comes out of clipping (poor recovery). If so, you might want to experiment further.
In one case I had to also put a tiny cap across the two inputs of a poweramp IC (LM3886) because on the negative half cycle there was a small amount of spurious oscillation as the signal came out of clipping. That's a different mechanism, but still important and affected by the linear mode phase compensation.
That's how I do it. I hope this helps.
Ever since I heard the Linkwitz Labs Orion speaker system, which has many OPA2134's in the signal path, I'm sold on the OPA2134 dual opamps. If they don't sound perfect, it's not likely the opamp itself that's at fault. The 553X's are great too, but have bipolar input transistors, so will load the source a little more, which may not matter in most cases.
I'm in agreement with sensible bandwidth limiting of the applied signal too.
I'm confident that everything I said above is accurate, except for one part. The part where I said, "Most people set the cap across the feedback R to cause a -3dB point at roughly 200kHZ, so there's still a lot of distortion correcting feedback at 20kHZ." You could also adjust the cap across the feedback R to be -3dB at 20kHZ, and you would still have close to the same distortion correcting feedback at 20kHZ.
I'm not the all knowing expert on this, but I believe it's true that pretty much all opamps have a built in pole at a low frequency, maybe around 100HZ, which is in the mix as well. When you add another pole (cap across Rf), you add phase shift. Each pole introduces up to 90 degrees of phase shift; 45 degrees at the -3dB frequency. In theory, you would have better stability by positioning your added pole (cap across Rf) higher in frequency, so at the frequency where the loop gain goes below one, the phase shift will not be too close to 180 degrees, thereby causing oscillation. In practice it doesn't seem to work this way. I'm not sure why. It would seem that adding the cap across Rf pole would cause the total phase shift to get close to 180 degrees by the time the loop gain rolled off to below one. Apparently most opamps are designed to expect the added pole across Rf, and not oscillate. Maybe someone else could shed more light on this.
I'm not the all knowing expert on this, but I believe it's true that pretty much all opamps have a built in pole at a low frequency, maybe around 100HZ, which is in the mix as well. When you add another pole (cap across Rf), you add phase shift. Each pole introduces up to 90 degrees of phase shift; 45 degrees at the -3dB frequency. In theory, you would have better stability by positioning your added pole (cap across Rf) higher in frequency, so at the frequency where the loop gain goes below one, the phase shift will not be too close to 180 degrees, thereby causing oscillation. In practice it doesn't seem to work this way. I'm not sure why. It would seem that adding the cap across Rf pole would cause the total phase shift to get close to 180 degrees by the time the loop gain rolled off to below one. Apparently most opamps are designed to expect the added pole across Rf, and not oscillate. Maybe someone else could shed more light on this.
Interesting experiments you're doing Mooly! It would be interesting to look at the output spectrum of the OPA134 with the reduced loop gain. If I recall correctly, the last time I looked at the distortion of the OPA134 and the OPA2604 it was predominantly 2nd harmonic (fun fact, you can actually adjust the level of the 2nd harmonic of the OPA2604 by playing with the power supply voltage) coming from the output stage. The reduced loop gain will obviously increase that second harmonic. However, the increased differential voltage at the inputs will make the input stage less linear as well, contributing odd harmonics. I'll see if I can toss one on the Audio Precision when I get back to Tucson (on international travel at the moment).
I've wondered for awhile why no one has used this approach on the LM3886, it should be fairly simple to stabilize that part for low gains using a noise gain approach. I used a similar technique to stabilize difference amplifiers here: http://www.ti.com/lit/an/slyt630/slyt630.pdf
I've wondered for awhile why no one has used this approach on the LM3886, it should be fairly simple to stabilize that part for low gains using a noise gain approach. I used a similar technique to stabilize difference amplifiers here: http://www.ti.com/lit/an/slyt630/slyt630.pdf
Thanks John. It would be interesting to see how the spectra changed using this technique. I'm afraid I haven't that kind of equipment available to me.
And something else that would great to see if you had any... the LM833 from TI vs the original type (still also available from TI. Post #72 here explains it... screen shot at post #78
http://www.diyaudio.com/forums/solid-state/274575-about-op-amps-use-4.html#post4361276
And something else that would great to see if you had any... the LM833 from TI vs the original type (still also available from TI. Post #72 here explains it... screen shot at post #78
http://www.diyaudio.com/forums/solid-state/274575-about-op-amps-use-4.html#post4361276
Very interesting! I didn't realize those two parts had different output stage topologies, I'd be happy to toss them on the bench and see how the output spectrums compare. Looking at the two datasheets, I don't see any specs for output current capability on the LM833-N, it is possible that it has a more robust output stage than the TI LM833.
I should add that I don't support those parts, they're handled by our SLL group (Standard Linear and Logic) in Dallas, so don't read too much into my ignorance of their differences
I'm happy to take a look at them though!
Last edited:
I couldn't quite believe it either but it really does seem to be the case. I can't just think of any part where its 'different different' yet retaining the same base number. Package differences, temp range, yes, but not a change to the internal structure.
The later second sourced TI version is massively better in its drive ability... and if it really does use a quasi output stage then the distortion spectrum is probably going to be totally different as well.
Maybe the TI active volume control would reveal differences
The later second sourced TI version is massively better in its drive ability... and if it really does use a quasi output stage then the distortion spectrum is probably going to be totally different as well.
Maybe the TI active volume control would reveal differences
I couldn't quite believe it either but it really does seem to be the case. I can't just think of any part where its 'different different' yet retaining the same base number. Package differences, temp range, yes, but not a change to the internal structure.
The later second sourced TI version is massively better in its drive ability... and if it really does use a quasi output stage then the distortion spectrum is probably going to be totally different as well.
Maybe the TI active volume control would reveal differences
I just re-read that post, so you found the LM833 had better output drive than the LM833-N? I initially thought you were saying the opposite (jet lag). That wouldn't really surprise me, knowing that the semiconductor industry is a never ending arms race. I'm sure when they defined the TI LM833 they felt the need to provide some benefit over the original, if for no reason other than to become the "first source". The real surprise is that they don't brag about it more in the datasheet.
Obviously, this would have occurred before the National acquisition. This type of stuff is still happening, although it's become much less popular to use the same part number. Personally, I feel that duplicating a part number is a great way to commoditize your part and no one is going to see the value in the product that you offer over a competitor.
In general, I think the quasi-complementary output stage can be really appealing. I've wondered if its responsible for the favorable listening impressions people have of the LM3886, and the Burr-Brown op amps (OPA134, 2604, 227, etc).
It is confusing, the improved part is marked and sold as LM833P in DIP form and LM833DR in SOIC outline. LM833P is the only marking on the actual package apart from production codes. The SOIC package says only LM833 plus codes but is fully described as LM833DR for ordering.
This is the TI response,
LM833-N is an op-amp originally from National Semiconductor. NS named this device as LM833. As for the package, LM833N means a PDIP package LM833-N. NS calls PDIP package as “N” package.
TI developed another LM833 as 2nd source device and named it LM833. When TI made LM833, TI changed its output stage to NPN type from VPNP.
After TI acquired NS, TI renamed NS’s LM833 to LM833-N in order to differentiate the LM833-N from TI’s LM833. Please don’t be confused by LM833-N and LM833N
This is the TI response,
LM833-N is an op-amp originally from National Semiconductor. NS named this device as LM833. As for the package, LM833N means a PDIP package LM833-N. NS calls PDIP package as “N” package.
TI developed another LM833 as 2nd source device and named it LM833. When TI made LM833, TI changed its output stage to NPN type from VPNP.
After TI acquired NS, TI renamed NS’s LM833 to LM833-N in order to differentiate the LM833-N from TI’s LM833. Please don’t be confused by LM833-N and LM833N
Good thing we tell you not to be confused...
This happened with a lot where there was portfolio overlap when the two companies merged, the national version of the part got a "-N" appended to the part number. There are numerous reasons why you can't simply obsolete one of the two overlapping products. For example, it might force a customer to re-qualify their product with the new, supposedly identical, part. Also, despite the part numbers being the same, as you have found "identical" parts can be quite different between suppliers. An end product that worked fine for years may not function properly with a different supplier's version. For these reasons, many semiconductor companies are very hesitant to obsolete products.
Good thing we tell you not to be confused...
Lol
absolutely.Well if you've got some of them there new little blighters it would be interesting to see how they compare distortion wise.
I can just imagine some application that 'needs' that higher drive ability and has been designed using the correct part and then some repair or whatever takes place and the old part is fitted (hey, its an LM833, what's the problem) and it doesn't work. Geez, that could cause some head scratching.
Lol
Well if you've got some of them there new little blighters it would be interesting to see how they compare distortion wise.
I can just imagine some application that 'needs' that higher drive ability and has been designed using the correct part and then some repair or whatever takes place and the old part is fitted (hey, its an LM833, what's the problem) and it doesn't work. Geez, that could cause some head scratching.
That can happen! Normally it comes up when someone from the purchasing department notices that these LM833s are cheaper than those LM833s...
But fun stuff like that keeps applications engineers employed, so I can't complain too much!
from post61
Edit:
Since John's post predates your further explanation:
The NatSemi version LM833-N is the older NS part,
whereas the older Ti part LM833N is also the poorer drive capability part.
The only high capability output part is the LM833P in DIP form and LM833DR in SOIC
Is this correct?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.