Being "concerned" seems suggesting being overcharged with emotion Hans. Rather, having two amplifiers sharing the same common nodes across their inputs, both operating open loop, just appeared "highly problematic" at the time, and still does. To use trimming components, or alternatively using lower offset opamps as Drbuj suggests, is a patch that doesn't directly address the significance of sharing the same nodes, rather your solution does, as it breaks those direct connections.
Basically I don't use coupling capacitors unless unavoidable, rather DC servos if needed. This is mainly because they can have sonic characteristics as much as cables. Okay... now for my official disclaimer... before the forces of the double blinders descend like resplendent emancipators of the universe... I am completely dillusional... I hear nothing... I hear nothing...
Last edited:
Sure investigated the peak velocity versus frequency on LP's, performed by L.R. Happ
Assuming that the usually 5cm/sec is the rms velocity, if so this can be translated in 7cm/sec peak velocity.
Now looking at the image below with recorded peak velocities as published by Sure, at 1Khz peak is at ca. 30cm/sec and around 10Khz ca 60cm/sec,
Versus the 7cm/sec peak this means a factor 8.6 or 19dB more output at the 10Khz region as maximum peak velocity, this being caused by the anti-Riaa emphasis.
So for a Cart specified at 0.35mV@5cm/sec, this is not the maximum output, because around 10kHz level can go up to 3.0mV rms.
To give the amp some extra headroom for scratches or dust, an extra 10dB margin can be added, thereby increasing the maximum input level to 10mV rms, or almost 30 times the original 0.35mV@5cm/sec.
Hans
Assuming that the usually 5cm/sec is the rms velocity, if so this can be translated in 7cm/sec peak velocity.
Now looking at the image below with recorded peak velocities as published by Sure, at 1Khz peak is at ca. 30cm/sec and around 10Khz ca 60cm/sec,
Versus the 7cm/sec peak this means a factor 8.6 or 19dB more output at the 10Khz region as maximum peak velocity, this being caused by the anti-Riaa emphasis.
So for a Cart specified at 0.35mV@5cm/sec, this is not the maximum output, because around 10kHz level can go up to 3.0mV rms.
To give the amp some extra headroom for scratches or dust, an extra 10dB margin can be added, thereby increasing the maximum input level to 10mV rms, or almost 30 times the original 0.35mV@5cm/sec.
Hans
Daag Hans,
I think this chart you posted is main picture from Shure document I mentioned earlier. Also fully agree with your comments.
Still, what you also described; 0.35mV x10 (20db increase for RIAA) x 3 (for margin) results in about 10mV RMS, , all good Than if we , as I suggest, apply "mad" gain in first stage of x1000 or 60 db, it still result in worst case at 10 V RMS, not a big deal neither for amplifier if +-18V PS is used, neither for resistors and capacitors in RIAA EQ filter.
Then in practice , these high frequency high level signals cannot be and will not be at any continuation, this can be only an burst of triangle in worst case, most likely an record scratch, so no high power calculations are needed., at least I think..
I think this chart you posted is main picture from Shure document I mentioned earlier. Also fully agree with your comments.
Still, what you also described; 0.35mV x10 (20db increase for RIAA) x 3 (for margin) results in about 10mV RMS, , all good Than if we , as I suggest, apply "mad" gain in first stage of x1000 or 60 db, it still result in worst case at 10 V RMS, not a big deal neither for amplifier if +-18V PS is used, neither for resistors and capacitors in RIAA EQ filter.
Then in practice , these high frequency high level signals cannot be and will not be at any continuation, this can be only an burst of triangle in worst case, most likely an record scratch, so no high power calculations are needed., at least I think..
You mean not as "mad" as applying 76 db first stage gain with +- 8 V PS's? Certainly gain can be dropped to most any extent, to that of a conventional head amplifier perhaps. This suggests that noise, distortion and overload exist as mostly the only considerations of sonic implications in the design. Do you think this true?
This would take it far from safety even for my taste, even it cold be that one will not hear clipping even in this case
Sorry Hierfi, but I did not understood fully what you mean?
I can say that for whatever reason in blind test inferior vinyl sounds much better than technically advanced digital record of the same piece of music. This is probably reason why we discuss this in first place
How is limiting overload to 10dB unsafe? The volume control is set to produce whatever level you want regardless of this gain setting.
I am only discussing vinyl in this thread.
I am only discussing vinyl in this thread.
Last edited:
Here is a thought experiment Hans. What if you took all the records you liked and you found that they all sounded their best with a gain setting of 70dB with whatever headroom that turned out to be. Would you still lower the gain to increase headroom based on accurate data presented by Sure?
If so, I would use a supply voltage high enough to enable processing the highest expected input levels still with acceptable distortion, but when this is not possible sound perception would of course prevale over headroom.
Hans
Salut, still thing we misunderstood somewhere between lines , 76db gain is something like 6309V/V x 0.35mV typical output x 20db (x10) for HF RIAA = 22 VRMS, cant be handled by +-8 VPS....
Did you mean 76db gain in total, after HF cut of -20db was applied somewhere along line? 2,2 VRMS is good and easily handled by +-8V PS.
By the way I like your approach that high as possible gain as soon as possible is good stuff for our purpose!
Indeed.
Dick Burwen's 1984 design, that ultimately became the Cello Audio Palette, used 5532A's for the gain stages (a boatload of them), with FET input LF412A/s for the servos. See https://www.burwenaudio.com/images/CELLO_AUDIO_PALETTE.pdf
There are better choices now of course, although the 5532A is still a tough act to follow. And because that paleolithic IC does not have bias current compensation, it behaves very politely as regards noise.
In contrast the LT1028 datasheet explicitly states that, although the voltage noise is an impressive 0.85nV/rootHz, that is only reached if the resistance seen at the + and - inputs is identical - bias current compensation injects two correlated noise signals into both inputs. Use if for an RIAA stage, always with unequal resistances, and the noise performance is truly lousy.
Lots of audio IC's now have bias current compensation to reduce offsets. It would be interesting to know if the current crop of audio IC's suffer in the same way as the LT1028
Craig
Although I tried numerous dual op-amps (as outlined previously in this thread) I eventually ended up back with NE5532A as the output buffer. There was nothing much wrong with the others, rather mostly had a neutral character. On the other hand the 5532A was not as neutral sounding, yet had a seemingly darker/blacker background with greater presence, digging deeper in harmonic structures and vocals. The bottom line is that there is no desire to go back, or to try others yet.
Of note, however, it was found that this odd character is reduced while maintaining a darker/blacker background by using a pull-up resistor or current source from the positive supply to the output of the device. In my view it is critical for best performance of the 5532. The value and nature of the pull up was found being dependant upon the nature of the surrounding network. For low signal/low output drive, a higher resistance value (lower current perhaps 1mA) works better. In the output buffer network as previously shown a resistance was chosen to cause about 5mA or so at the output.
From examining the internal structure of the 5532 this reveals the output is basically driven in the negative direction through a diode into an NPN darlington sink, being pulled up by a current source feeding an NPN transistor. The output network frequency compensation is applied around the darlington as not including the diode, causing questions regarding if this diode is causing its odd character, perhaps in the crossover. Notwithstanding the true cause of this character, the 5532 device is helped by driving more current through this diode. In other words by increasing current the signal developed across this diode becomes more linear with signal current variations.
As an aside, the LT1115 (similar to the LT1028) shows a FET current sink from its output to the negative power supply rail in one of its applications.
Of note, however, it was found that this odd character is reduced while maintaining a darker/blacker background by using a pull-up resistor or current source from the positive supply to the output of the device. In my view it is critical for best performance of the 5532. The value and nature of the pull up was found being dependant upon the nature of the surrounding network. For low signal/low output drive, a higher resistance value (lower current perhaps 1mA) works better. In the output buffer network as previously shown a resistance was chosen to cause about 5mA or so at the output.
From examining the internal structure of the 5532 this reveals the output is basically driven in the negative direction through a diode into an NPN darlington sink, being pulled up by a current source feeding an NPN transistor. The output network frequency compensation is applied around the darlington as not including the diode, causing questions regarding if this diode is causing its odd character, perhaps in the crossover. Notwithstanding the true cause of this character, the 5532 device is helped by driving more current through this diode. In other words by increasing current the signal developed across this diode becomes more linear with signal current variations.
As an aside, the LT1115 (similar to the LT1028) shows a FET current sink from its output to the negative power supply rail in one of its applications.
Hello Hierfi,
You almost lost me again, by carefully reading I'm thinking if you refer to technique to force opamp output stage into SE class A by bypassing one of output transistors with resistor or current source? I read about it but have no hands on experiance, Rod Elliot wrote about it, here I found one article at this site https://www.diyaudio.com/community/threads/running-op-amp-in-class-a.32226/post-371261
This could be very interesting as SE class a will have different sonic qualities. As you seem to be experimenting with this I would encourage you to share more experiences, if you wish...
Hello Craig,
Hans was talking about opamp for DC servo compensation, not as audio gain.
Nevertheless discussion about 5523 is lovely, hard to beat chip as you sad, agree fully.
Hans was talking about opamp for DC servo compensation, not as audio gain.
Nevertheless discussion about 5523 is lovely, hard to beat chip as you sad, agree fully.
Then there is the whole sub topic of enhanced output stage bias current in opamps by connecting a ccs from the output to the negative rail, putting the output stage into class A operation. See eg https://www.audioasylum.com/audio/tweaks/messages/31186.html
Care is of course needed not to exceed the power dissipation of the opamp!
Craig
Care is of course needed not to exceed the power dissipation of the opamp!
Craig
That is my thinking as well Hans. Though what if increasing the supply voltages deteriorated sonics? I would think not, though anything seems possible.
Another factor that I am considering is that the gain level set for best sonics may not be determinant by the SSM2019, rather the subsequent network(s) may require such levels to produce best results. If that is improved, as to take in lower input signals to better function, then gain settings can be reduced on the input device to increase overload margins.
Last edited:
In the thread you pointed, Nelson Pass in his post #4 back 20 years ago goes:
"I find that a 10K to V- is a good general value for audio
circuits. The V- rail is preferred because in most op amps
we would rather work the N type output device rather than
the P or the quasi-complementary "P", and 10K is high
enough not to significantly load the output and low enough
to give you SE Class A bias with most audio type loads."
There seems no real need to think of a network as class A or otherwise. To me it doesn't matter... only sonics matters. This seems as also implied by Pass, in that there can be negative sonic affects by pushing into class A bias too far. This is something that I have found and can concur. Secondly, adding current via resistors, or CCS's, doesn't do much of anything sonically for many op-amps. The NE5532 (and probably would be too for the NE5534 as well) seems most sensitive of any tried thus far.
The original Cambridge Audio P40 from the late 60's had an interesting approach to overload margin. This was to use a two transistor "op amp". This had a flat frequency response, with 47k in series with the MM input, and a feedback pot of 1M as the volume control in a shunt feedback arrangement. Now this had the noise penalty of the 47k resistor in series with the cartridge, so it is around 10dB noisier than series feedback, but it has the advantage that it was impossible to overload the small signal chain before the power amp clipped.
The flat input stage fed a shunt feedback unity gain stage with mono switching, then a single transistor - again shunt feedback - RIAA stage. Baxandall tone controls and filters followed. Then the power amp.
Design award winning, it was revolutionary in many ways for a design now around 60 years old.
Craig
The flat input stage fed a shunt feedback unity gain stage with mono switching, then a single transistor - again shunt feedback - RIAA stage. Baxandall tone controls and filters followed. Then the power amp.
Design award winning, it was revolutionary in many ways for a design now around 60 years old.
Craig
