Looking for opinions on low noise preamp design - is my reasoning correct?

[tempus]

Hey all;

I've been working on a few different preamp designs whose main objective is achieving the lowest noise possible. The preamp must be capable of boosting a guitar signal to line level for interconnecting with a recording interface (for use with a computer and amp simulators) and so must have an output impedance of no more than about 1K to minimize loading. Simply using low noise opamps won't do the trick; I'm assuming that the Johnson noise of the resistors used will contribute a significant amount of noise and so must be taken into account in the overall design. Here's what I've concluded:

1. TL072 with 1M input impedance with gain of 2, non-inverting amp configuration (all noise figures in nV/rtHz):
TL072 = 18
1M resistor x 2 = 256
2M feedback resistor = 181
Total: 455

2. J201 JFET, common source configuration:
J201 = 5
1M gate resistor = 128
10K drain resistor = 12
4.7K source resistor = 8
Total = 153        *note: I haven't actually been successful in getting this design to accurately give me a gain of 2, also, I am not convinced that the output Z is in the 1K area necessary. Can someone help out in this area?

3. TL072 follower into a NE5532 non-inverting amp configuration, gain of 2:
TL072 = 18
1M input resistor = 128
NE5532 = 5
2K resistor x2 = 10
3.9K feedback resistor = 8
Total = 168

Although a bit higher in noise than the J201 discreet circuit, #3 seems to be the winner, as it will give an output Z of about 200 ohms with no fuss.

It appears that, due to resistor noise, the best bet for low noise is a design that uses the minimum of high value resistors. I'd be interested in what others have to say about this. Is my methodology in adding these noise sources correct? Since the J201 design is in fact the quietest, I'd also be glad to hear suggestions as to how to ensure a gain of 2 with an output Z of <1K.

Thanks

[antonis]

What minimum input impedance should be considered acecptable..??

[tempus]

I'm glad you asked that, because it's something else that I'm not entirely clear on. The rule of thumb is to make the input Z at least 10x that of the previous stage's output Z to minimize loading. So for a guitar pickup, probably around 200K would achieve that. However, I see a lot of guitar amp/pedal designs that use 1M as an input Z, so I'm assuming there must be a reason for that standard. Finally, and what I really don't understand, is that, if you apply some voltage divider math, a 1:10 output to input ratio gives you a resulting voltage of only 91%, where a 1:50 ratio will give 98%, and a 1:100 ratio 99%. Perhaps this is the reason that it's common to see a 1M input impedance in a lot of circuits.

Given this, why do we only shoot for a 1:10 ratio? Is that remaining 9% of voltage negligible?

[amptramp]

If you are looking at the input resistor noise, remember that within the passband established by the coupling capacitor, the 1 M resistor is in parallel with the volume pot on the guitar, which is usually 500 K or 250 K and changes with control position.  The fact that these noise sources are capacitively coupled does not matter within the passband - the guitar resistance and the input resistor of the preamp are in parallel, making the input resistance smaller than the value you have used.  This might partly answer the question tempus asked - the use of a 10 M resistor might not give you the gains you expected because you have another resistance in parallel.  But I would agree that there is nothing to lose by going up in value - there is no need to protect against switch pop caused by capacitor leakage current because you would use low-leakage capacitors and you would not have bypass switching here anyway.  It would mainly keep the noise down when you plug in or unplug the guitar.

You have a resistance of 5 K - 15 K for the pickup winding and in some cases, you can switch multiple pickup windings in parallel but they have the rather hefty series inductance of 2 - 5 Henries limiting this effect to low frequencies which you may not want to amplify anyway.  If you back off the guitar volume control even slightly, you will add a relatively large amount of series resistance, so your current noise will then increase.  You won't notice it with a TL072.  You might notice it with a 5532.

Using just the input resistor of the preamp will give you a higher noise than you will actually see in service.

[R.G.]

What A.T. said.
As a set of general principles,
> Use a first stage with gain. The idea is to get the signal level up above the device noise without adding in more device noise by additional stages. Using a buffer only makes sense if your signal level is already high enough to make the added noise from the buffer irrelevant.
> Use a single ended first stage. Differential stages have two input devices in the differential stage, and both add noise, as mentioned. This ups the noise by square root of two because there are two devices.
> Select a first amplifying device which minimizes the contribution of the amplifying device's own noise, and its necessary biasing components. As A.T. said, there is a range of input impedances and devices that have different input voltage noise and current noise in and of themselves. Their use also implies the use of resistances to bias the devices, and those resistances add their own thermal noise and current noise for any biasing currents to the input noise from the signal source.

This selection is tricky. In crude generalities, for low source impedances, bipolar devices are a better match for getting lower overall noise. As the source impedance goes up the "best" device changes over to JFETs for higher impedance. There is an overlap in the middle. As noted, a guitar pickup is 4K to 10K+ of wire resistance, in series with one to a few henries of inductance. You are faced with making a compromise between the low resistive source, the high and variable resistance added by any controls on the guitar, and the need to get a really high impedance to prevent treble loss from the interaction of the inductance, the cable capacitance, and the biasing noise added by the first stage.

As a final principle, there's the principle of good enough. It helps in all low noise design to consider what the goal of the quest might be. If it's really to design the lowest possible noise, you're way beyond my help. I go look things up in my textbooks on low noise design. A.T. is a working professional, I think. Get him to guide you in detail, and be prepared to spend a lot of time and probably several iterations. In my experience on my noise quests, I always reached a point where with the amp dimed I could not hear the noise over a couple of feet from the speakers, and decided that "inaudible at reasonable distances" was a good criteria for "good enough" and quit flogging myself.

Quests are noble, and their stories make great epic literature and music, but they do take a lot of time and resources away from other things like normal life, playing guitar, listening to music, and so on.

[PRR]

> 1. TL072 ....with gain of 2, non-inverting .... 1M resistor .... 2M feedback resistor

This is just wrong. The "+" input can have a 1Meg bias resistor (it is bypassed by the guitar which I do not see?). The feedback network can be MUCH lower impedance. Say 1k and 2.2k. What does that hiss work out to be?

> JFET, ....I am not convinced that the output Z is in the 1K area necessary.

It isn't. It is closer to 10k. Does it really matter? How about a buffer after?

> 3. TL072 follower into a NE5532 non-inverting .... #3 seems to be the winner, as it will give an output Z of about 200 ohms ...

Zout will be very near 1 Ohm; if 200r is a goal you need to build-out. Of course any GP opamp facing unknown cable should have a few hundred ohms. But the thread seems to start about "low noise"; how did "200 ohms" become "the winner"?

[jonny.reckless]

I haven't actually been successful in getting this design to accurately give me a gain of 2, also, I am not convinced that the output Z is in the 1K area necessary. Can someone help out in this area?

This is good enough for a guitar preamp input stage. Will be about 10dB quieter at 1kHz than using a TL072 assuming a pickup with 10k ohms and 2 Henries, and will have some second harmonics which will add a touch of warmth to the signal. Gain will be approximately 2. You need to adjust the JFET source and drain resistors for different FETs using the application note I sent you in the other thread. With a 9VDC supply and a middle of the range J113, the source will sit around 1V, with 1mA flowing through the JFET, and will allow 2Vpp input before clipping which is about what I get from a Seymour Duncan JB pickup when I hit the strings really hard.
https://www.vishay.com/docs/70595/70595.pdf

If you need the gain to be exactly 2, well that's a different matter...

[Rob Strand]

Something looks wrong to me.

- not clear if you are using non-inverting or inverting topologies.  I'm assuming non-inverting.

- You are adding the nV/rtHz values.  These need to be combined as a sum of the squares.  Also their effect needs to be referred to the input or output.

- Input resistors contribute very little to the overall noise because, if the source impedance is lowish and the input capacitance is a reasonable size, the source impedance will shunt away the noise.

- Feedback resistors don't contribute individually.    Typically the noise comes from the Thevenin equivalent impedance of the feedback network.

-  There's no need to use 2M feedback resistors on a TL072.   You could use low values, not-unlike the NE5532 values.

- Bipolar input devices usually require inclusion of input noise current, unless the impedance is quite low.


You should look-up some of the old National Semiconductors applications notes about calculating noise.   I'm sure there's many later articles as well.   I think they will help you a lot in getting your head around noise calculations.

[amptramp]

I have some older app notes from Texas Instruments but they are not listed on their site.  There is one called "Low Noise Devices and Circuits" by Bob Crawford dated July 15, 1963.  There are a lot of textbooks and some online resources that can teach you the basics.  Everyone who has posted so far has offered good advice.  What no one has mentioned yet is the concept of noise bandwidth.  It is part of the noise calculation, so you cannot get a nV/(SQRT Hz) value without the bandwidth in Hz being known.

A lot of devices have what is called "excess noise' or "1/f noise" where the noise at some low frequency begins to rise as frequency goes down and you can see it on op amp data sheets where a noise graph is given.  There is also "popcorn noise", a phenomenon where the signal abruptly shifts between two arbitrary fixed levels.  This latter noise is often credited to defects in the semiconductor crystal structure but can appear in things other than semiconductors.

Believe it or not, this gets to be fun after a while.

[antonis]

Believe it or not, this gets to be fun after a while.

Also, believe it or not, I can't utilize a natural number of ++++++ to add on the above said..!!!

P.S.
@tempus: Involvement with all those noise classifications & "chromatic" categorization as well as level up floor & figure interpretations ought to be fun or else it will result into brain stroke..

[tempus]

Thanks for all the replies - it's great to have so many people helping me out with this. I still have a lot of questions and things to sort out in my head, but I wil try to ask only the most pressing ones.

> Use a first stage with gain. The idea is to get the signal level up above the device noise without adding in more device noise by additional stages. Using a buffer only makes sense if your signal level is already high enough to make the added noise from the buffer irrelevant.

Would this be true for a piezo pickup (valid because I need to design a preamp for that too). In my case, I would need an input Z of 10M (so 10M input resistor) and a gain of 8 or 9, which would make my feedback resistor 80 or 90M. Wouldn't the noise contribution from the resistors alone warrant a separate buffer?

> 1. TL072 ....with gain of 2, non-inverting .... 1M resistor .... 2M feedback resistor

This is just wrong. The "+" input can have a 1Meg bias resistor (it is bypassed by the guitar which I do not see?). The feedback network can be MUCH lower impedance. Say 1k and 2.2k. What does that hiss work out to be?

I thought that the - and + input resistors needed to be matched when using an opamp. Is this not the case?

> JFET, ....I am not convinced that the output Z is in the 1K area necessary.

It isn't. It is closer to 10k. Does it really matter? How about a buffer after?

It absolutely matters. This preamp will be driving the input of a recording interface with and input Z of 10k. How did you conclude that the output Z is 10k? What is the equation for the output Z of a common source JFET? I haven't been able to find the answer to this and so I have no idea if any of the JFET designs I've considered/experimented with have the desired output Z.

> 3. TL072 follower into a NE5532 non-inverting .... #3 seems to be the winner, as it will give an output Z of about 200 ohms ...

Zout will be very near 1 Ohm; if 200r is a goal you need to build-out. Of course any GP opamp facing unknown cable should have a few hundred ohms. But the thread seems to start about "low noise"; how did "200 ohms" become "the winner"?

As I mentioned in the 1st post, the output Z has to be in the 1k area, so as not to load the next stage.

This is good enough for a guitar preamp input stage. Will be about 10dB quieter at 1kHz than using a TL072 assuming a pickup with 10k ohms and 2 Henries, and will have some second harmonics which will add a touch of warmth to the signal. Gain will be approximately 2...

If you need the gain to be exactly 2, well that's a different matter...

This is awesome - ty for going to the trouble of designing and drawing the schem for me! Something I didn't mention in my OP was that the pre should really have no distortion or colouration of its own, as all of that is being done in software. Also, the gain accuracy is not unimportant, as it is imperative that I do not clip the ADCs on the way in. A gain of more like 1.7 is probably safer. I'm starting to wonder if, given these parameters, an opamp and its more accurate and reliable gain structure may be my only choice.

Something looks wrong to me.

- not clear if you are using non-inverting or inverting topologies.  I'm assuming non-inverting.

Yes - non inverting. The topology is listed in each calculation.

- You are adding the nV/rtHz values.  These need to be combined as a sum of the squares.  Also their effect needs to be referred to the input or output.

I wasn't aware of that, but isn't the overall principle the same, IOW, isn't the noise contribution by the resistors a much bigger issue than that from the devices themselves?

- Input resistors contribute very little to the overall noise because, if the source impedance is lowish and the input capacitance is a reasonable size, the source impedance will shunt away the noise.

I'm sorry I don't follow. When you say "source resistance" do you mean the resistance from the guitar? If so, is 10-15k considered lowish? And when you say "shunt away" do you mean that the lower source resistance will basically divert the input resistor noise to ground?

Believe it or not, this gets to be fun after a while.

Right now I have a high fun impedance feeding a low frustration impedance and as a result am feeling heavily loaded.

Also, I have to say that I understand the advantages of using a discrete JFET design, and would like to try one, but I lack sufficient information regarding calculating the output Z and calculating /setting the gain. Can anyone give me the equations for these parameters?

One addendum - I tried using a 072 follower with no input resistor, i.e., input connected directly to +input and output directly to - input, which also worked. This would remove the noise contribution by a 1M resistor if the circuit will reliably function without it. Is there any reason not to remove that resistor to ground?

[FiveseveN]

I think your frustration stems from here:

I thought that the - and + input resistors needed to be matched when using an opamp. Is this not the case?

I'm not sure what you're asking, but your non-inverting gain stage should look something like this:

Where the 1M resistor sets the input Z and the feedback resistors have nothing to do with its value.

[antonis]

I tried using a 072 follower with no input resistor, i.e., input connected directly to +input and output directly to - input, which also worked. This would remove the noise contribution by a 1M resistor if the circuit will reliably function without it. Is there any reason not to remove that resistor to ground?

If there is an input capacitor, there MUST be a DC path to ground for the, almost negligible but existent, input bias current..

[PRR]

> I thought that the - and + input resistors needed to be matched when using an opamp.

Why?? (And didn't we discuss this here last week??)

> How did you conclude that the output Z is 10k? What is the equation for the output Z of a common source JFET?

And this also. The plate resistance of a JFET in a resistance-coupled amplifier is almost inevitably >>10X the drain resistor. Zout==Rd to a good practical approximation.

> It absolutely matters. This preamp will be driving the input of a recording interface with and input Z of 10k.

So? If you think you need gain of 2.0 into a 10k load, you can instead make gain of 4 through a 10k source. This "matching" is actually more common in several fields, especially around "long" wires.

> A gain of more like 1.7 is probably safer.

Where do you get this Precision Musician? With output level predictable to 1-1/2dB far in advance of performance?

What would Thomas Edison do? Or Geoff Emerick? Or Tom Dowd?

[tempus]

If there is an input capacitor, there MUST be a DC path to ground for the, almost negligbible but existent, input bias current..

Ty for clearing that up.

> I thought that the - and + input resistors needed to be matched when using an opamp.

Why?? (And didn't we discuss this here last week??)

From https://www.analog.com/en/analog-dialogue/articles/common-problems-when-designing-amplifier-circuits.html:

"To minimize offset voltages caused by input bias currents, which track one another when using bipolar op amps, R1 is usually set equal to the parallel combination of R2 and R3."

I was a little off base there because I forgot to include the ll resistance of R_f. So that's why I thought we needed to use a high value resistor in the feedback circuit. Does this not apply here?

So? If you think you need gain of 2.0 into a 10k load, you can instead make gain of 4 through a 10k source. This "matching" is actually more common in several fields, especially around "long" wires.

I've never heard of this before. Can you explain a little more? What do you mean by 10k source?
edit: nvm i think I know what you mean now - the 10k source is the signal coming from the JFET, correct? And if I amplify it more than necessary, the increased amplification will offset the attenuation resulting from the loading of the input? Isn't loading frequency dependent though, since it's impedance and not straight resistance? Wouldn't I still hear an unflat frequency response compared to an unloaded signal?

Where do you get this Precision Musician? With output level predictable to 1-1/2dB far in advance of performance?

It's not a question of what the predictable output level is, it's a question of what the absolute maximum is that can be tolerated without clipping the inputs of my interface.  I have a few different guitars that I use with this setup. Using an opamp with a gain of 2 I hammered the guitar with the highest output pickups as hard as I possibly could to get the most voltage out of it. With a gain of 2 it clipped the inputs of my interface. With a gain of 1.7 it did not.

The plate resistance of a JFET in a resistance-coupled amplifier is almost inevitably >>10X the drain resistor. Zout==Rd to a good practical approximation.

Ty for that explanation. That helps a lot.

[Rob Strand]

I'm sorry I don't follow. When you say "source resistance" do you mean the resistance from the guitar? If so, is 10-15k considered lowish? And when you say "shunt away" do you mean that the lower source resistance will basically divert the input resistor noise to ground?

Yes I meant the impedance of the guitar (or whatever is driving the input).   

Without even thinking of the impedance of the guitar, imagine putting a 1ohm and 1MEG resistor in parallel.   The parallel resistance is Rparallel = 1/(1/1+1/1MEG) = 1 ohm.    So as far as noise is concerned it's a 1 ohm resistor.   From that we can see that adding the comparatively large noise voltage from the 1MEG with the low noise voltage from the 1 ohms resistor cannot possibly be correct.

"To minimize offset voltages caused by input bias currents, which track one another when using bipolar op amps, R1 is usually set equal to the parallel combination of R2 and R3."

The offset due to input bias currents is almost non-existent for a JFET input opamps.    Even for bipolar input opamps the offset is quite small.   You can see an offset with 1Mohm and a high input bias opamp like a NE5532.  However adding a resistor to the feedback path means you are putting in components which add noise.  For audio you care about noise more than DC offset.   (If you were designing a circuit which measures DC voltages you would need to consider all offsets.)

[tempus]

Thanks for that explanation Rob - makes much more sense now, especially the part about adding noise sources. It makes perfect sense that when they are in parallel you have to treat them as such, but I never would have thought that what applies to parallel resistances also applies to parallel noise sources.

[Rob Strand]

Here's a reasonable article.   
https://www.ti.com/lit/an/slva043b/slva043b.pdf

The points not so great are:

- The maths is presented in a way which is little more complicated than it really needs to be but if you skip to the section "Inverting and Noninverting Op Amp Circuit Noise Calculations" it does show the main points.

- They present the results as large "combined" equations.   The "smaller" equations showing the parts are more useful.

It's often easier to do these calculations by looking at each element or a few elements in isolation, refer the noise to the input or output, then "add" each of the noise contributions using the sum of the squares, 
    v_noise_total = sqrt(v_noise1^2  + v_noise2^2 + ...), 
where  v_noise1, v_noise2  etc. represent the noise from individual causes (in nV/rtHz).
(This is the "smaller" E1, E2 ... equations in the article.)

To be honest, spice simulations can do all the hard work for you.

[amptramp]

Excellent TI reference from Rob Strand and better than the old one I was looking for.

Imagine the noise within a resistor that is made of thin metal film but instead of one spiral conductive resistance path like a screw thread, you have two parallel spirals like the threads on a focus mechanism.  This is effectively two resistors in parallel but for noise purposes, the effective resistance of the device is the single parallel resistance.

Uncorrelated noise sources act like the audience at a football game.  If you have 25 people in a pub watching the game, the cheers are not 25 times larger than one cheer, they are 5 times larger because the cheers are uncorrelated noise that goes up as the square root of the number of people.  If this wasn't the case and a crowd of 40000 people attended the game, you would have the noise level 40,000 times that of a single audience member, not the actual 200 times, a sound power level of 23 db more.

[Rob Strand]

Excellent TI reference from Rob Strand and better than the old one I was looking for.

Thanks for letting me know.  I was thinking that old TI article (which isn't available) was the gold.

The later TI doc is more detailed than the old National Semiconductor notes as well.   

Analog Devices have a lot of stuff but I couldn't find a specific doc on opamps.  They have a lot of more technical stuff dealing with ADC/DACs etc., which makes it hard to find something on basic noise.