On choosing opamp feedback resistor values.

Please forgive the simple opamp design question... I'm sure...

What are the factors when deciding the feedback resistor value? Not the "Gain"... I beleive I understand the gain is the ratio of the two resistors?

What I'm wondering is there any difference between a 10K-10K for a gain of (2) and a 100K-100k for a gain of (2)?

As a net effect, smaller values will yield lower noise figure and less phase smear (caused by opamps input's transistors parasitic capacitance) compared to larger resistors.

If your opamp has BJT inputs, there will be also less DC offset at the output generated by the fact that transisors sink some base current. FET input opamps are usually more immune to this..

As a general rule, use the lowest resistances you can get away with.

tv said:

As a general rule, use the lowest resistances you can get away with.

Click to expand...

This is kinda what I'm wondering... how do I know what I can get away with?

jwhmca said:

tv said:

As a general rule, use the lowest resistances you can get away with.

Click to expand...

This is kinda what I'm wondering... how do I know what I can get away with?

Click to expand...

The lower limit is set by distortion caused by drive limitations. For the op-amp's output the resistance of your feedback network is placed in parallel with the load.

JDB.

So I need to calculate the load...? then what?

Sorry for the questions, just really trying to understand.

Then what?

Click to expand...

You have to know the opamp. Most audio parts are happy to drive 2k, less needs verification particularly at high noise gains.

Your ground resistor generally sets input impedance, and feedback resistor value is chosen for gain (ratio).

Click to expand...

Sorry, doesn't make any sense to me... If you're thinking of a noninverting configuration (that's where on resistor connects to ground) then the input Z is independent of the feedback network Z. In an inverting configuration the input Z is indeed determined by the feedback network, but there is no resistor which connects to (non-virtual) ground.

Samuel

Before you go crazy putting the lowest value resistor you can find everywhere, it is useful to hold everything in perspective.  For a simple unity gain inverter, the opamp's input noise will be multiplied 2x, so making resistor noise more than a few dB less than the opamp input noise is probably a wasted effort, and opamp distortion will increase the heavier load it must drive. Also electrolytic capacitors behave better when passing less current, better yet a higher impedance resistor value might support a film instead of an electrolytic cap forming a HPF pole. No design decisions ever hinges on just a single aspect of performance.

Resistor self noise makes a lot more of a difference in very high gain stages, because input signal voltages are so much lower. Also if the high gain stage uses bipolar input devices there will be an input noise current term that gets converted to noise voltage by the resistance. Resistance value makes less of a difference for line level unity gain stages.

This is all useful to understand if/when you do any high gain/low noise design.

JR

> is there any difference between a 10K-10K for a gain of (2) and a 100K-100k for a gain of (2)?

Take it to an absurdity. 1 ohm and 1 ohm. The opamp has to drive 2 ohms.... it often can't do well. So try 1Gig+1Gig. That's practically no conductance. Most practical opamp leakage, and universal stray capacitance, will overwhelm 1Gig.

For jellybean wall-power NON-inverting design, make your feedback resistor 10K and your shunt resistor whatever the gain wants.

10K is comfortably above usual minimum load (you need some spare for your Real Load), yet comfortably below most audio noise and bias-current troubles.

Then see where you are.

If you hope for gain of 100 down to 1Hz you need a 1,700uFd cap, maybe non-polar. In a jellybean project, that may be large or costly. Take the series resistor to 220K, shunt 2.2K, you only need 80uFd.

Say my PRR741 opamp has input bias current of 1uA. Since there's a cap, this flows in the 220K and induces 0.22V of offset. In a single-supply system, that's often "nothing". Cap-coupled likewise. With no coupling caps, the next stage may be upset about 0.22V DC into it.

Pure bias current on a differential input can be balanced by making the other input's resistor the same, in so far as both inputs suck the same current. They never do. My PRR741 uses floor-sweepings input transistors. I might give you hFE=450 on one side and hFE=27 on the other side. Bias currents won't cancel. (Real opamp makers do much better than my hypothetical opamp.)

You can't reduce all errors to zero. Many errors are not important to a specific design. Figure out what matters (experience helps) and seek to balance errors and costs. If you MUST use 1Gig resistors (as for condenser mike), you can pay for low-low current chips and get <<<1V offset. Though as it happens, a BiFET (TL072) in shirt-sleeve temps will take 1Gig with 0.01V-1V offset, and in audio we just cap-couple that away.

Hi PRR,


    thank you so much for that simple explanation. marvelous!



    kindest regards,



    ANdyP

For jellybean wall-power NON-inverting design

Click to expand...

And for INVERTING design?

What's your desired input impedance? What's your desired voltage gain? Most times, your Rf is fixed by these considerations.

If I am using a balanced line receiver design, and I want to use it to drive a meter, I am far less concerned with noise and distortion (since I can't really see any typical distortion on a meter) and far more concerned with keeping the bridging load low enough not to disturb the Common Mode Rejection of the link I am metering.

So in that case in the past I have used 200K matched resistors (which I matched using a wheatstone bridge) to reduce the CMR impact on the input.  Later stage (precision rectifier) used more typical values where frequency response matters.

PRR/John Roberts is using high impedence like this a reasonable practice for a metering circuit?  how high can I go?  As high as I can get away with without breaking the meter performance?

bruce0 said:

If I am using a balanced line receiver design, and I want to use it to drive a meter, I am far less concerned with noise and distortion (since I can't really see any typical distortion on a meter) and far more concerned with keeping the bridging load low enough not to disturb the Common Mode Rejection of the link I am metering.

So in that case in the past I have used 200K matched resistors (which I matched using a wheatstone bridge) to reduce the CMR impact on the input.  Later stage (precision rectifier) used more typical values where frequency response matters.

PRR/John Roberts is using high impedence like this a reasonable practice for a metering circuit?  how high can I go?  As high as I can get away with without breaking the meter performance?

Click to expand...

The calculus is pretty simple for a meter... what is the sensitivity of the meter? If it's a VU meter that only indicates down to -20VU who cares if the noise floor is -40VU? OTOH if the meter reads down to -40 dBu you want some noise margin below that.

Without resorting to actual math this early in the morning I suspect 200k is fine. An upper limit may be more limited by stray capacitance that could cause HF loss, than thermal noise in the resistor.

If you do a lot of design you will find yourself using a lot of the same values in arbitrary applications where a range of values work... I still have a partial reel of 20k 1% TH resistors that I used for precision circuits, lots of 100k for general purpose line level.

Sharp pencil math, and device spec sheets will generally advise us about upper and lower limits for values, the actual values can be whatever you want within that range. In a large console another factor is current consumption from 100 opamps driving 1k loads. 

Note: There can be a temptation to optimize every circuit block for noise floor, but generally total path noise is dominated by input or early gain stages, so much of this optimization is just for personal satisfaction. Ir doesn't matter like in the meter example, if low enough. 

JR

Thanks, that makes a lot of sense. My lowest metered range is -20dBu, so for my meter application it is working.

I realize that it is not a big contributor to noise, it is metering at the output's of preamps, compressors and filters, levels are high and impedances are low.  But I am putting them across every channel in a meter bridge for a 51x rack, and so I am optimizing for the one case where it does end up mattering some day (because I can't switch it out of the circuit).

200K works fine, and the meter response rolls off at around 15Khz, but I think that is because of the rectifier performance.

And yes, I admit it, I am tempted by the noise floor  :  ... peace and quiet is not such a bad vice I guess.

I was looking at this : http://www.thatcorp.com/datashts/THAT_2162-Demo_Datasheet.pdf
U4b and U11b have 316r feedback resistors, isn't this supposed asking too much current to the op amp ?

This probably should have started a new thread... (I may move it later).

The resistor and op amp in question are feeding a VCA control voltage port, in a dynamics side chain. The voltages expected are probably tens of mV, so currents will be well within the drive capability for a 553x. I expect the low resistance was selected to keep noise on the control voltage port as low as practical. Noise on that line with be multiplied by VCA action to show up in outputs.

JR

Of course ! thanks

JohnRoberts said:

This probably should have started a new thread... (I may move it later).

Click to expand...

I f you do it I'll have another question