Sorry to jump in here, but lately I've been trying to understand the value choices for R1/R2.
I've got an ne5532 (inverting) that has R1=2k, R2=20k so gain = -10.
I need a unity gain buffer so I changed it to R1=R2=10k but why not 2k? Or 1k? What would be optimal and why?
I've got an ne5532 (inverting) that has R1=2k, R2=20k so gain = -10.
I need a unity gain buffer so I changed it to R1=R2=10k but why not 2k? Or 1k? What would be optimal and why?
It is an audio driver/buffer amplifier to be used after the DAC filter such that a volume potentiometer can be arranged in the interface to the DAC. The TPA6120 can deliver up to 700mA at the output and the minimum loading is 16 Ohm. TI recommends 10-100 Ohm in series with the output (explained in the datasheet / I use 10 Ohm). As it is a current-feedback design, the impedance seen from the input pins should be kept low to avoid considerable offset at the output.
Please "jump in" - that is the purpose of a public forum.
The "optimal" level is difficult to say exact but we can discuss the effects. As also explained by Piotr, the lower impedance the less noise in general. Further, less sensitivity to radiated noise. My own limit here is when I cannot notice the noise anymore. OP-AMPs have an output stage that is designed for a certain minimum load impedance. Most can do 2K loading and many today 600 Ohm. That also means that their performance is slightly improved with some loading (which is not the minimum load impedance). A 600 Ohm OP-AMP I would let operate with some 2-5K loading. That loading may be from the following stage or the feedback network of the OP-AMP. I would avoid using impedance levels so low that my 1/4W resistors cannot be used with a power margin of one third at least. Warm resistors also make noise.
In the past, higher impedances like 47K-200K were used for discrete circuits. That left a bit of noise and more sensitivity to radiated noise. With most modern OP-AMPs the impedance level has reduced to 1K-20K.
On top of that, I follow the advice of some senior English engineers I met: "When it works well, why change it?"
Last edited:
The other thing, as mentioned before, is self-heating of the feedback resistor with signal. Especially at lower frequencies (500Hz and below) that can be significant, to the point that it becomes the dominant distortion mechanism, producing distortions that are higher than the amp itself.
In sensitive test equipment like Audio Precision, they go through special topologies because this effect would be bigger than the instrument distortion itself.
If you use a through-hole resistor of say 1W, it probably isn't a problem, but can be with smaller SMD resistors. Low tempco resistors also help to limit the effect.
Jan
In sensitive test equipment like Audio Precision, they go through special topologies because this effect would be bigger than the instrument distortion itself.
If you use a through-hole resistor of say 1W, it probably isn't a problem, but can be with smaller SMD resistors. Low tempco resistors also help to limit the effect.
Jan
Last edited:
The old German school of engineers will say that too, I think. But young people want to invent something better 🙂
I made a mistake with this warm resistor and 300mW I wrote about earlier. One of its ends is soldered into the ground plane, where all heat escapes to the copper quickly, so it's not hot, just warm.
By the way, why not use several resistors in parallel to reduce the heat in the feedback resistor?
Regarding TPA offtopic: do headphones have a problem with Dampening Factor like speakers? I am asking because I am not a headphone freak at all. I mean this series resistor at the headphone amplifier output and that it increases the output impedance of the amp.
Why below 500 Hz? Is it about the specific nature of the music? That there is statistically more power, i.e. heating?
The distortion occurs because the resistor heats & cools in the rhythm of the signal. That means that the R is larger at the top of the sine wave and the R is lower at a zero crossing. The bipolar nature of a zero-volt centered signal means the distortion happens twice on a cycle, thus is 2nd harmonic.
The heating and cooling of the R needs time - it is completely analogous to the cap in an electrical RC time that needs charging and discharging.
Higher frequencies causes less ripple across the C of an RC. Mechanically, a higher heating/cooling frequency causes lower 'temperature ripple' across the resistor, so the resistor varies less in value with higher signal frequencies.
If you take a physically larger resistor, with more heat capacity, the frequency where the resistance modulation happens goes down.
Just as choosing a higher C value for an RC makes the frequency at which larger ripple occurs, go down.
Jan
Last edited:
If you are concerned about this, better to use multiple R's in series. Take example 20k and 1k feedback network. If you use 21 identical 1k resistor, 20 in series, all resistors see exactly the same voltage. So all resistors will change in value due to thermal modulation exactly the same. That means that the ratio 20k/1k also stays exactly constant, independent from the thermal modulation, which means no distortion due to thermal modulation.
No rocket science.
Jan
Got it! I think I've got it. You explain really well.
20 resistors in series plus one to the ground means the same power per each resistor, which means temperature and their resistance are pulsating exactly the same way, which means our 20/1 divider has no error caused by resistor's TCRs. Resistors must be the same type of course? Is this right? Am I right?
Cool! 20 resistors, quite soldering challenge. But let's say 5 of them in series would still better than one??
Sure would be better, but hard to quantify. If the one is 0.25W and you use 5 others at 0.5 or 1W, who knows? The only thing you can say is that it surely is better than one at 0.25W or 0.5W
I think you can calculate it IF you knew all data like thermal capacity of resistors, but normally you don't know that.
Jan
I think you can calculate it IF you knew all data like thermal capacity of resistors, but normally you don't know that.
Jan
Yes, you will never know the metal film thickness and trim style, it may be thicker for lower resistances. Maybe using one micromelf (0102 case) as R1 and few melfs (0207 case) in series as R2 could give minimum error. Some people prefered thick film Caddocks MK132 in feedback. This makes sense to me now, because thick layer must have bigger thermal capacity.
Yes but thick film resistors are more non-linear than thin film, higher tempco and higher voltco ....
In a feedback position I wouldn't use a thick film.
Jan
In a feedback position I wouldn't use a thick film.
Jan
Thanks for the help everyone! I think the thread answers many questions with feedback in an interesting way. I think it's time to experiment a little. Maybe I will design some matrixes of resistors for the experimentation. Maybe I can also measure some THD's for low frequencies?
As far as I can tell- much-much less. Speakers have large impedance variation and 99% of designs since mid-1930s aim for "low" impedance drive for best flatness. Headphones were traditionally driven with non-zero impedances. And they mostly do not depend so much on resonance, and do not show large resonance rise. The "32r" phones I have tested show very small impedance variation. The "300r" phones do have a rise to 600r. But 10-25 Ohm in series with either does "no harm". And starts to equalize the difference of voltage sensitivity.