4558 and 072 work but 5532 and 833 do NOT [solved]

[j_flanders]

Hi,
I'm playing around with this very simple circuit, based on the Muff Fuzz. I'm testing different component values, diodes and opamps.
I notice that ne5332 and lm833 are not working properly when using 1M resistors for either R1, R2, or both.
R1 and R2 both 1M: sounds choked/starved
R1:100k and R2:1M sounds ok(ish) when I play hard/full chords, but sounds 'starved' when I play very lightly and there is no tail, it just cuts out.
When I swap out the opamp for a 4558 or 072 any combination of 100k and 1M for R1 and R2 works.

So, I've read that these are very different kind of opamps with regard to: bias current, input impedance, output current capabilities, bipolar vs FET etc.
My question is how do I calculate the 'minimum requirements' for R1 and R2 (and probably also R5 and R6) based on the specs for these opamps in this circuit.
I'm not asking if ne5532 is a good choice as a high impedance unity gain buffer in an inverting stage opamp, I just want to learn how to do the calculations so I get the basics right.

[DrAlx]

If I remember correctly the input impedance of the LM833 is a few hundred kOhm.  So R1 and R2 need to be at least a factor of 10 smaller than that (say 10kOhm). So 1M scales down by a factor of 100 to 10k, and 100nF at input scales up by 100 to 10uF to compensate.
Values of R5 and R6 should not matter.
The question though is what are you plugging in as signal?
If the input comes from some other circuit with output impedance of 1k or below then above changes should be OK.
If the input is direct from a guitar pickup (which will have impedance in the hundreds of kOhm for high frequencies) then you'll end up losing highs.

[anotherjim]

Those 100k bias divider resistors are the trouble.

Change the 100k bias resistors to 10k and it should be happy.

The amps that don't work have input protection clamps between + and - inputs. During power up and before they can act properly as operational amplifiers, the input clamps turn on and place lower resistance between the inputs, preventing the bias voltage from reaching 1/2 supply.
If you look at what happens on power up, chip power will get there before correct input bias does. The bias has to charge the 10uF cap thru 100k - takes time. The + input the bias is feeding will be close to 0v and be below the common mode input range of the chip which stops an op-amp working probably - it can't swing its output to set the - input the same via the feedback resistor, its output can't swing to 0v. The inputs will have different voltages and the clamp turns on.
When working properly, the inputs will have the same voltage and the clamps don't act.

...but as DrAlx says, 5532 and 833 and many other "audio" amps don't really like high input impedances. Can actually be noisier than TL07x in these conditions. So if you insist on them, a buffer in front would be a good idea so you can bring those input R's down.

[amz-fx]

The TI datasheets say that the minimum supply voltage for the 5532 and LM833 are plus/minus 5 volts, or a 10v single supply. Many of us have used them at 9v and they work okay, but the chips will be happier if you power with 12v. This may be an application where in minimum voltage makes a difference.

I would alter the bias resistors as Jim suggested to give the chips the best chance of working well.

regards, Jack

[j_flanders]

Thanks for the replies so far.

If I remember correctly the input impedance of the LM833 is a few hundred kOhm.

I think I read 300k somewhere but I also read somewhere that this is the open loop input impedance and that it will be much higher after you apply negative feedback.

The question though is what are you plugging in as signal?

Good question. I'm plugging in my guitar, as I think at least a guitar by itself should work. The original Muff fuzz(trasistor version) , like a Fuzz face, does not like to see anything  but some passive pickups at the input.
10k as input resitor, which also sets the input impedance in the inverting opamp configuration, would seriously load down my pickups.

Those 100k bias divider resistors are the trouble. Change the 100k bias resistors to 10k and it should be happy.

I soldered 10k resistors parallel to the two 100k's and it didn't make any difference, or at least none that I could hear. It still sounds starved, I really need to strum hard to get some sound/output.

The TI datasheets say that the minimum supply voltage for the 5532 and LM833 are plus/minus 5 volts, or a 10v single supply. Many of us have used them at 9v and they work okay, but the chips will be happier if you power with 12v. This may be an application where in minimum voltage makes a difference.

I tried it with 12v and there seems to be a tiny bit of improvement but I could be imagining things.

I could still go lower on the bias resistors. Is this calculation for absolute minimum correct? Based on 0.250W resistors and 10v (9,6v) power supply:
max Watt= 250mW (resistor limit)
max Voltage= 10V (psu limit)
gives: max Ampere = 25mA (within 200mA psu limit)

R5 + R6 = 10V/0.025A = 400 Ohm
R5 and R6 should not be lower than 200 Ohm.
I guess I can try 1k to be on the safe side.

As for R1 and R2:
both 100k : sounds ok, chords and notes fully ring out
both 1 M : sounds bad, needs lot of input to get any sound/output
R1=100k R2=1M: 10x gain: sounds ok(ish) but chords and notes don't ring out, when the input signal gets too low, it simply cuts out
R1=1M R2 100K: 10x attenuation: sounds ok,chords and notes fully ring out, still even mild some distortion

I'm surprised why that last case works and the before last doesn't.

[anotherjim]

Most odd, 'cause I've known cases where high bias resistors were the issue. 10k should already be low enough. We do know cases where the LM833 at least, works as a substitute in circuits with 33k bias resistors. I don't think you need to go lower than 10k.
Despite all I said, I don't now think the bias source is the problem (but lower than 100k is still advised!).

What are the output DC voltages like?

I think you're right about the 300k input value.

What could be happening with these thoroughbred chips is high frequency (ultrasonic) oscillation, particularly the output amp. This can make the signal sound sputtery.

Do you have power decoupling caps (100nF ceramic) close to the supply pins?
A small ceramic cap across the 100k feedback (10p to 50p).
100R series resistor in the output to quell output cable capacitance.
All the above are data sheet recommendations.

[j_flanders]

Most odd, 'cause I've known cases where high bias resistors were the issue. 10k should already be low enough.

I went as low as 2k and it didn't help. Reverted back to 10k's

What are the output DC voltages like?

Nothing unusual.
With 12,4 volt power supply:
Inverting and non Inverting pins were 6,18V
Vout was 6,40V and 5,80V
Similar 'normal' readings for a 9V power supply

Do you have power decoupling caps (100nF ceramic) close to the supply pins?

Do you mean power filtering caps or the AC-coupling/DC-blocking caps? I assume power filtering.
As this was an experimental setup, point to point, with lots of wires, they were not close, as in: inches/centimeters away.
I added a 100nf ceramic to the power filtering and soldered everything with regard to the bias voltage to the pins of the op amp socket.
Can't get any closer. It didn't help. 

A small ceramic cap across the 100k feedback (10p to 50p).

Closest value I had at hand was 140pF, but it didn't help.

100R series resistor in the output to quell output cable capacitance.

I don't understand where or why I should put it. I have a 10k resistor on the output of U1a and 100k at the output of U1b (although after a cap, but since I don't take a signal from the intersection it shouldn't matter.) I mean, my 10k could be 10,1K and my 100k could be 100,1k so I didn't see the point of an extra 100 Ohm resistor in those places but maybe I misunderstood. 

All the above are data sheet recommendations.

Yeah, I know, but I tend to think you can disregard all that when it comes to distortion devices. 

Just for the record, I don't mind using another op amp instead of the 55532, I just wanted to understand how to calculate the resistors to meet the minimum requirements for a specific opamp.

[reddesert]

Some suggestions that are basically guesses:
- try audio probing the circuit to figure out where the signal disappears.
- the 1M input and feedback resistors on the first op-amp are higher than one normally sees in stompbox circuits. Try scaling them down to 100K or less. The RC constant of the 100nf coupling cap and a 100K resistor is still really low frequency, so you don't need to change the cap.
- the + inputs of the opamps are connected together and both to Vbias. If you put a modest value resistor from Vbias to each + input, the inputs will no longer be directly connected, which is probably good practice.

[DrAlx]

Back to school for me. Opamp input impedance isn't the problem. It's the large input bias currents for the opamp's bipolar inputs that can give problems. These become voltage errors on the outputs. You seem to be reporting quite a difference between the bias level (on the inputs) and the output DC levels.
To minimise effects of input bias currents when using amps with bipolar inputs ...
1) Minimise the impedances presented to both opamp inputs and try to match them. So you have 1M DC impedance on the inverting pin. So you should aim for the same DC impedance on the non-inverting pin. 1M is too big though.
2) Don't try and squeeze out too much gain.

I would start with the second stage and choose 18k||18k for the two bias resistors as its a better match to the 10k||100k you have on the second stage. You will then have about 9k impedance on both inputs. Now using 18k||18k when the first stage has 1M on its inverting input will give a DC error on the output of the first stage so decouple first and second stages with another cap before R3 (1uF should do it).

[rankot]

Thank you all, very interesting topic!

[j_flanders]

Thanks for the additional replies and comments.
They all sound valid but I cannot help but wonder if they apply or help here.
They do correspond with what I read in articles like this :Avoiding Op-Amp Instability Problems In Single-Supply Applications
http://www.analog.com/en/analog-dialogue/articles/avoiding-op-amp-instability-problems.html

I would fully agree for hifi, situations with lots of gain where errors/offsets are multiplied, maximum headroom, lowest distortion etc but I wonder if these offset errors really matter when you are dealing with a gain of 1, headroom of 12volts peak to peak and an input signal of a lets say 100mv to 1V peak to peak.
Having an extra dc blocking cap after the first opamp wouldn't hurt I guess since I have 10x gain in the second opamp.

So I still think this starved sound/ouput is caused by a (too low) current problem somewhere because only lowering R1 or R2 or both has helped so far.

[DrAlx]

Here's an extreme example to clarify the point. Assume 9v supply and equal bias resistors. Consider sine wave at input of 0.2V pk-pk. Lets say offset error from unity gain stage makes that sine move to an offset of 5.5v for example rather than 4.5v as desired.  Now consider that second stage without the diodes. The gain of minus 10 means all input voltages in the range 5.4v to 5.6v will make the output hit the bottom rail (unless some other offset error manages to counteract the fact that the sine has been given a bad bias).  So you would totally lose the sine wave. The clipping diodes don't solve that.  Compare that to the case when the sine wave ripples about 4.5V.  The second stage gain without diodes would give you an output sine of 2v pk-pk centred about 4.5 +/- some other offset error.  Adding diodes would take off the tips and bottoms of the sine.

[PRR]

5532, input bias current: 200nA typ, 1000nA max

Taking the Max, 1000nA is 1uA. 1uA in the 1Meg R2 is a 1V offset.

You got the sim and breadboard. You think U1 is biased at 4.5V. Is its output really sitting somewhere else? It CAN be as much as 1V off.

That alone is still tolerable. But then you DC couple into U2. R4/R3 is dc gain of 10. The 1V error off U1 is multiplied to a 10V error out of U2. Since this is only a 9V system, that can not happen. Output is jammed to the rail.

Taking the Typ input current spec gives 0.2V error at U1 and 2V error at U2. This may not hit a 9V rail. However as Jack says, 5532 is more a 30V (to 44V) part, and does not have swing real-close to rails (not important in 30V work). If we figure it can only swing to 2V and 7V, and the "4.5V" is error-ed to 6.5V or 2.5V, it will "work" but very lame one side of the signal.

There is additional input current error in the 100K dividers. Only 1/10th as much... ah, 1/20th as much. Not big, but enough to eat that last bit of free swing.

The DC analysis is much happier if R1 R2 are 10X even 100X smaller. However R1 is your guitar-loading impedance, and should not be a lot less than 1meg.

Anyway 1Meg resistors alone (not even considering opamp hiss current) are as much hiss voltage as guitar-chain can stand.

The input REALLY should be a Follower, not an Inverter. Then the 1Meg's hiss is shunted by guitar impedance and comes out small. (In unity gain you seem to save one resistor, but going right to an input you should have 5K-33K series to protect against Bad Inputs, so no.)

You should also consider cap-coupling between stages. 1uFd to the right of R3 breaks the DC gain in U2. Direct coupling is elegant but you have to beware DC error build-up. Also bass-loss but this is never a real issue for guitar (10uFd here would be -1dB at 4Hz, four octaves down on guitar, so 1ufd and -1dB@40Hz is probably better than any guitar speaker.)

The hiss-current of 4558 and 5532 ensure poor to bad hiss performance at guitar impedances. TL072 _IS_ the perfect cure. No other change needed _if_ you accept the 1Meg's thermal hiss; else change to non-inverting.

I doubt "phase" is relevant in a distortion. Even an output phase signal does not mix-cancel with itself unless both inputs are same level. In distortion the "effective gain" is different all the time.
___________________

> input impedance of the LM833

Under NFB, this rises. The naked Zin is almost never an issue.

[DrAlx]

As Paul says.  If you want to get a ballpark estimate of the offset error at the output, take the input bias current from the datasheet and multiply by the DC impedance hanging on the inputs and the gain factor.
So using your original circuit and second op amp as example, the error at the inverting input is Ibias * 10k||100k = Ibias * 9.1k.  There's also an error at the non-inverting input of Ibias * 100k||100k = Ibias * 50k.  So the total offset error (assuming equal Ibias on the two inputs) is the difference between the two giving about Ibias * 40.9k.  This then gets multiplied by the gain factor of 10 for that opamp to give output voltage error of Ibias*409k. That's the reason for suggesting balanced impedance hanging on the two inputs.

[j_flanders]

If you put a modest value resistor from Vbias to each + input, the inputs will no longer be directly connected, which is probably good practice.

so decouple first and second stages with another cap before R3 (1uF should do it).

Having an extra dc blocking cap after the first opamp wouldn't hurt I guess since I have 10x gain in the second opamp.

You should also consider cap-coupling between stages. 1uFd to the right of R3 breaks the DC gain in U2. Direct coupling is elegant but you have to beware DC error build-up.

Apparently the dc offset does matter. 
I added a 100nF dc-blocking/ac-coupling cap after R3 and the starved sound is totally gone.
I switched back from 12V to 9V power and did some readings again of all the pins:

1) OUT 1: 4,70V
2) -IN 1: 4,46V
3) +IN1: 4,46V
4) Vee: 0V
5) +IN 2: 4,46V
6) -IN 2: 4,46V
7) OUT 2: 4,49V (without the extra coupling cap between U1 and U2 it is 4,04V)
8 ) Vcc: 8,93V

The actual resistances:
R1: 1,038M
R2: 1,007M
R3: 10,07k
R4: 106k

I recorded a clip with my phone, bypassing the extra coupling with an aligator clip. The difference is obvious:
http://s000.tinyupload.com/index.php?file_id=51877524222895296642
http://s000.tinyupload.com/download.php?file_id=51877524222895296642&t=5187752422289529664231603

Thanks for all the help and insights!

I might go back and try to solve the offset error through the bias network and see it that would also solve the issue (and not create another one, since the compensating resitor needs to big if I keep those 1M's). At the moment the entire bias network is directly soldered (p2p) to the pins of the opamp socket so experimenting there isn't so easy.

[anotherjim]

Before your last answer, I typed this!

Quote from: anotherjim on Today at 05:48:26 AM

What are the output DC voltages like?

Nothing unusual.
With 12,4 volt power supply:
Inverting and non Inverting pins were 6,18V
Vout was 6,40V and 5,80V
Similar 'normal' readings for a 9V power supply

I'm going back to your comment that the sound was different in some way. I wonder if that's just due to increased DC offset making on of the clipper diodes closer to turn-on and the other  harder to turn on.  Just one side clipping rarely sounds good.
"Vout was 6,40V and 5,80V" - meaning 0.6v difference across the last amp. A diode is turned on and clamping the gain of the last amp down. Removing the diodes will show the true extant of the offset at the final output.

Comparing your DC readings to a good sounding chip might be instructive?

So I still think this starved sound/ouput is caused by a (too low) current problem somewhere because only lowering R1 or R2 or both has helped so far.

This ties in with the offset thing.
Lower R1 & 2 to reduce offset?
For a distortion, high 1M input impedance isn't strictly necessary. Lower impedance dulls a guitars output, but then the distortion adds new harmonics back. Some inverting schemes out there use 100k input resistors. They depend on this effect and sound harsh with a buffered input. However, they usually apply a lot more gain than x10, as distortion must be maintained during note decay otherwise it quickly gets dull.

The usually way of breaking the DC offset has already been pointed out -  a DC blocking capacitor between the stages. That will settle the second amp back to the bias and more importantly, centre the clipping diode action.

Quote from: anotherjim on Today at 05:48:26 AM

    100R series resistor in the output to quell output cable capacitance.

I don't understand where or why I should put it. I have a 10k resistor on the output of U1a and 100k at the output of U1b (although after a cap, but since I don't take a signal from the intersection it shouldn't matter.) I mean, my 10k could be 10,1K and my 100k could be 100,1k so I didn't see the point of an extra 100 Ohm resistor in those places but maybe I misunderstood. 

Sorry, I assumed your output was  a simulated potentiometer shown as 2 resistors -the schematic looks like simulation software. The 100R would not be needed with a fixed resistor already there.

[j_flanders]

Comparing your DC readings to a good sounding chip might be instructive?

Good idea. Here are the readings compared to a good sounding 4558 (the diodes, paralleled with a 140pF cap are still in the circuit)

5532:
1) OUT 1: 4,70V
2) -IN 1: 4,46V
3) +IN1: 4,46V
4) Vee: 0V
5) +IN 2: 4,46V
6) -IN 2: 4,46V
7) OUT 2: 4,49V (without the extra coupling cap between U1 and U2 it is 4,04V)
8 ) Vcc: 8,93V

4558:
1) OUT 1: 4,44V
2) -IN 1: 4,46V
3) +IN1: 4,46V
4) Vee: 0V
5) +IN 2: 4,46V
6) -IN 2: 4,46V
7) OUT 2: 4,46V (without the extra coupling cap between U1 and U2 it is 4,65V)
8 ) Vcc: 8,92V

Sorry, I assumed your output was  a simulated potentiometer shown as 2 resistors -the schematic looks like simulation software. The 100R would not be needed with a fixed resistor already there.

Ok, I see what you mean now. With a real pot, full up, there would be no series resistance (to form an rc filter with the capacitance of a cable)
It is a simulated potentiometer but the actual circuit also has those since I used up all my real potentiometers in various places in the circuit for variable gain, additional resistance in series with the diodes, etc. There are also switches to include more or less diodes, symm and asymm. The shown circuit in my first post was a much simplified version.
The simulation software is an online java applet: http://www.falstad.com/circuit/
But it's not always useful as for example it shows zero bias current into the inputs of an ideal opamp (unlike in my real life 5532 version)

For a distortion, high 1M input impedance isn't strictly necessary. Lower impedance dulls a guitars output, but then the distortion adds new harmonics back.

The original EHX Muff fuzz, where this all started, has 100k and it sounds great. I just wanted to see what would happen if you changed that to 1M.

[DrAlx]

The original EHX Muff fuzz, where this all started, has 100k and it sounds great. I just wanted to see what would happen if you changed that to 1M.

Ah OK.   The 4588 has bipolar inputs also but Ibias is smaller than the 833/5532 and about 30nA typically (although I have seen large variation between different datasheets!!!).
Also looking at the EHX schematic they have 470k DC resistance on the inverting input of the first opamp (i.e. just the feedback resistor) and 680k||680k = 340k on the non-inverting input (the bias resistors).  So impedance mismatch at first opamp in EHX circuit is 130k.  Multiply that by  Ibias and the 4.7 gain factor of that first opamp stage and you get voltage offset error of 30nA * 130k * 4.7 = 18.3 mV at output of  first stage.
You changed the first stage to unit gain but impedance mismatch at the inputs was greater (1M - 100k||100k) = 950k, so you would get about 30nA * 950k * 1 = 28mV offset error at output of the first stage when using the 4558 (you reported 20 mV).   Not too far from the EHX circuit and still small compared to guitar signal pk-pk, hence the circuit should cope without decoupling cap if using a 4558.

[j_flanders]

I might go back and try to solve the offset error through the bias network and see it that would also solve the issue (and not create another one, since the compensating resitor needs to big if I keep those 1M's). At the moment the entire bias network is directly soldered (p2p) to the pins of the opamp socket so experimenting there isn't so easy.

Ok, so I removed the coupling cap between U1 and U2 and tried to solve the issue with a compensating bias resistor between +In1 and Vbias.
+In2 is still connected directly to Vb without a resistor.
I thought the correct value was R1||R2 (Rin||Rf) or 500k in this case since Rin and Rf are both 1M but strangely this didn't solve anything.
Next, I used a 1M pot as rheostat instead of the 500k resistor and kept turning it up until the issue was reolved.
At that point I measured its resistance and found it was 990k... More like 1M than 500k...
I did some readings of all the pins again:

1) OUT 1: 4,46V
2) -IN 1: 4,25V
3) +IN1: 3,85V
4) Vee: 0V
5) +IN 2: 4,46V
6) -IN 2: 4,46V
7) OUT 2: 4,46V
8 ) Vcc: 8,92V

Any ideas to explain this?

Current circuit:

[anotherjim]

I've never yet completely understood the +input bias resistor calculation. Given that R1 is AC coupled, I don't see how it can affect bias current on the -input. Surely there is only the feedback R2? That happens to be 1M.

BTW, you can see the loading effect of your meters input impedance on pin 3 with the 1M fitted. If you had 2 meters and measure the output at the same time, you would see that also drop.