Now that is an interesting question..... I guess my build would drift too much to be used without servo and a coupling cap.
right you just consolidated my point.
I mean +/-10mV with Servo at the outpu is it too much for a phono amp since the signal will be amplified at leat other 30dB.
Not that 10mV static offset is too much rather the opposite, but how fast drifs is the problem.
It works because it works in your case and in mine too, but it is not optimal.
Which also brings to the point that if you strip the servo off, the circuit wonders really bad which means that servo is working really hard, thus affecting performance.
It would be interesting to get some other people's here experience.
I hope that this aspect wasn't overlooked, but I am pretty sure it wasn't and then I would like to understand how I could fix this issue.
To make the drift smaller the servo can be fitted with smaller resistors. The resistor in the back to back pair sets the current, the servo then injects the needed adjustment current, by making the resistors to the servo smaller, it's possible to keep the drift in the low micro-volt range.
I have made some of my paradise variants with mirror and current source formed by quad SMD transistor-bricks. The thermal lock here improves DC performance quite a bit. But makes the PCB more critical as the stray capacitance between the transistors tend to trigger oscillation (cured with ceramic decoupling at each transistor)
I have made some of my paradise variants with mirror and current source formed by quad SMD transistor-bricks. The thermal lock here improves DC performance quite a bit. But makes the PCB more critical as the stray capacitance between the transistors tend to trigger oscillation (cured with ceramic decoupling at each transistor)
I know that, you are basically setting the servo authority higher and higher.
Basically the servo would need to work very hard.
My quesiton connected to your observation was: what is your offset before servo i.e. stripping the servo off?
I feel like that if servo has to work so hard something is not quite right and needs to be improved.
The servo is working through 100K... looking at the output at the op-amp it.s almost a steady voltage (On my own PCB with SMD) app 2-3V. meaning that the injected adjustment current is is 2/100000=0.02mA-0.03mA. Compared to the 15mA driving the circuit. the servo cant be judged as working Hard.
If your circuit is a tad bright, have you checked for oscillation..?? I mean we have had trouble with the first generations due to sub-optimal PCB layouts.
If your circuit is a tad bright, have you checked for oscillation..?? I mean we have had trouble with the first generations due to sub-optimal PCB layouts.
oh no, absolutely not at all.
The circuit doesn't have bad brights at all. I am saying they are a tad glossy a tad and that is likely have to do with the quality of my build now that is missing the high quality PSU.
There are not oscillation, hoever if I increase the voltage rail over 12V I have oscillation that I will deal with.
My own PCB is SMT too.
The servo is not considered to work hard only on the amount of current injected but, and you have checked this, by the variation.
The sensitivity of that folded current mirror, is very high and few uV from the servo can adjust few volts at the output.
Do you have matched devices on your board?
I am wondering if you have built it with first hand devices and offset was significanlty different.
I am sure distortion wise the matched devices will yeald to a better result, but just from a offset's standpoint, what is the effect of matched devices.
Moreover, if you take your servo off, what is your drifting (not static offset introduced by tolerance and buffer)?
I have boxed it, so I can't make that observation. the offset is placed in the center of very very high impedance mirrors and the high impedance input of the Jfet buffer. If you disconnect the servo even micro-volts of current drift will throw the offset way off... that is the basic idea. The close to Ideal mirrors makes the distortion spectra clean.
I don't consider the servo working hard, as the current injection is so tiny compared to the driving current.
My mirrors are matched as they are on the same chip, and my input is HFe matched to app 5%
I don't consider the servo working hard, as the current injection is so tiny compared to the driving current.
My mirrors are matched as they are on the same chip, and my input is HFe matched to app 5%
Last edited:
Ok so you technically answered to my question, due to the very high impedance morrors, the smallest variation of current die to temperature or mis-matching or whatever else, will cause several volts of offset.
Agree with you.
Saying that servo doesn't work hard is probably not corrected just because the amount of current injected is small.
It is determined by how much cirrection you need to apply at the output not at the source.
I will try to see how I can make the servo working properly and set output down to +/-1-2mV which is more than acceptable.
To me hard is the amount of correction current needed, compared to the driving current. Here the needed current is so low, that the transistors working points is so minutely affected that it's insignificant.
If you lover the resistors to 22K or even 10K the servo will have much more power to adjust. Making this circuit without the servo is not possible. Well it is if you AC-couple it with at small cap before the buffer.
If you lover the resistors to 22K or even 10K the servo will have much more power to adjust. Making this circuit without the servo is not possible. Well it is if you AC-couple it with at small cap before the buffer.
I respectly disagree with you on both points:
1) if current needed is small doesn't mean that servo is not working hard. If the voltage that needs to adjust is 10V of constant drifting, then servo is working hard. But that might be debatable it's like the glass half full or half empty.
2) Circuit can't work with HP filter if the drifting is high and frequency involved are pretty high, also because you will find bias of your gainstage sometime at +10V and some other time at -10V which deeply affects distortion's performance.
Also, if the variation is so significant that not only involves frequency below 2Hz or so, then bandwidth for the servo has to be wider which in turns definitely affecs sonic performance as the servo will go correct audio signal as well by a small amount
I am hope that Joachim or some other guys who took part to the development of this design could comment on it, especially considering that the servo has marked up a resistor value of 150K which means that it should work properly with that value which suggest that something is wrong with my implementation or this aspect might have been overlooked a bit.
But the drift is not jumping up and down....variations are slow, but due to the high impedance of the mirrors voltage may be quite high-ish.
I made the house keeping to work with a servo. But in a way so the servo is taken extremely out of the signal loop. mind you that the 22K or what you need is way way higher than the current setting resistor. The servo is only dealing with DC. or app 0.1 Hz.
I made the house keeping to work with a servo. But in a way so the servo is taken extremely out of the signal loop. mind you that the 22K or what you need is way way higher than the current setting resistor. The servo is only dealing with DC. or app 0.1 Hz.
Last edited:
In that case it would be ok.
In my build the variation are pretty fast and that is why I think the servo doesn't handle the offset.
I am wondering why I would experience this strange offset drifting.
Could it be caused by low hfe BJTs on the mirror? I don't have an explanation to that.
Matching should affect DC drifting little IMHO.
I mean drifting is not pretty slow in my opinion and yes is mainly thermal as always.
However I need to know a reference to understand if that is normal somehow.
Like if you trim the offset either by changing current on the mirrors or unbalancing top against bottom size, what would be your thermal drift? let's say 10mV is 1 seconds 100mV in 5 seconds?
Also does anybody here have a screen shot of the residual noise with the scope?
I have a little residual noise that doesn't look like it is broadband noise.
I am thinking if the resistors I am using at the top of the 2 current mirrors which sees a current of 35mA is rated for 1/4W or 1/8W.
I am using a small SMD resistor there and if rating is 1/8W it could possibly cause more residual noise and more drifting.
I will also look into that later.
However I need to know a reference to understand if that is normal somehow.
Like if you trim the offset either by changing current on the mirrors or unbalancing top against bottom size, what would be your thermal drift? let's say 10mV is 1 seconds 100mV in 5 seconds?
Also does anybody here have a screen shot of the residual noise with the scope?
I have a little residual noise that doesn't look like it is broadband noise.
I am thinking if the resistors I am using at the top of the 2 current mirrors which sees a current of 35mA is rated for 1/4W or 1/8W.
I am using a small SMD resistor there and if rating is 1/8W it could possibly cause more residual noise and more drifting.
I will also look into that later.
Stefanoo, i was not watching the thread for some time. First you say you run the Paradise on plus-minus 12V from a battery. That is not enough, it needs plus-minus 18V.
Then you say you use an AD797 as servo. That does not work because it is a bipolar. You need a high impedance Fet like OPA134.
Then you say you use an AD797 as servo. That does not work because it is a bipolar. You need a high impedance Fet like OPA134.
Right I kind of figure that.
Why do I need 18V there? Technically you get enough voltage to bias input devices.
If I increase voltage over 12V I get oscillation, I will try to fix that.
For the AD797 I agree with you.
The discrete module I am using in place of the OPA, though, is JFET input pair.
I am using Burson Audio Discrete op amp (since I am out of opamps now.)
Would you see any problem using that one?
I am going to put on order tonight some OPA134.
Question: how is the offset without servo? Is it possible to run the preamp without servo and use coupling cap just to test the actual effect of the servo?
The offsety without the servo is hard to predict. Some comes from the input stage and some comkes from the buffer and is dependend how well you match the sides. Also there is a high impedance point between the buffer and the mirrors ( where the RIAA is ). Even tiny missmatch in the mirror gives a big offset. Made correctly you can expect around 10mV offset ( with servo ) that needs some stabilizing. Directly switched on it needs some time to come down. That happens also when the elcaps are new. Their leackage current goes down after a while. That behaviour disappears with use. For absolute minimum offset you have to run the Paradise say for one week continous and then trim the servo.
You talk about safety. The only safe way is a coupling cap. This circuit has an enormous gain in only one stage and even our ex professor from Canada that chalenged me to a point where i closed the thread for some time put a coupling cap after his phono stage although on his website it says : DC coupled without caps in the signal path, he simply lied. DC coupling from cartridge to speaker is posibble but NOT SAFE>
You talk about safety. The only safe way is a coupling cap. This circuit has an enormous gain in only one stage and even our ex professor from Canada that chalenged me to a point where i closed the thread for some time put a coupling cap after his phono stage although on his website it says : DC coupled without caps in the signal path, he simply lied. DC coupling from cartridge to speaker is posibble but NOT SAFE>
Last edited:
I can not say if the Bursons can be used for servo. I only know that they performed poor in the tests that Samuel Groner made.
Download the file : Opamp Distortion
SG-Acoustics · Samuel Groner · IC OpAmps
Download the file : Opamp Distortion
SG-Acoustics · Samuel Groner · IC OpAmps