"Noiseless" input stage

[maartendh]

What about Craig Anderton's idea: guitar signal goes directly to non-inverting input - no c1, c2, r1, r2 or zeners. Can be found in one of the last issues of Device (on Mark Hammer's site).

Maarten

[PRR]

OPA827 http://www.ti.com/lit/ds/sbos376i/sbos376i.pdf  see Fig 41.

For sources over 2K, you can't do better.

Sources under 2K are likely balanced (different ball of worms) or Active (and they should determine your noise figure).

[amptramp]

So it seems apart from Ron the majority opinion favors a simple non-inverting JFET-input opamp stage for the input gain stage. For illustration purposes, I drew one (assuming split power supply):



Couple questions:
1) What opamp? The only one I have tested to any serious extent in such a position is the TL072. And that one is definitely too noisy. It is noisier than a NE5532 even when fed from a high impedance source. The OPAx134 series seems fairly popular and I have an OPA2134 lying around, so I'll definitely give that a try. Does anyone have other suggestions for very low noise JFET opamps? Looking at datasheets, the super-expensive ones like the OPA627 don't really seem to have an advantage here despite nominally better noise voltages because above 10k source resistance, the noise is dominated by the resistance anyway. But maybe someone has an underappreciated favorite.

I have not kept up with the latest in op amps for low-noise applications but the OPA827 listed by Paul seems to be as good as any with a midband noise of 4 nV/root Hz rising to 6 nV/root Hz at the lowest required frequencies.  It also has 2.2 fA/root Hz so it is good for high impedances.  With bass amplifiers, be sure to take into account the rise in noise at low frequencies.  1/f noise can override the midband noise figure to the extent that the best midband performance is not necessarily give you the best bass performance.  The 1/f knee can be anywhere from 1000 Hz down to below 100 Hz (as it is with the OPA827).

2) What about the series Resistor R1? Adds noise, no question. Necessary for protection of the delicate JFET input or not? What is the smallest value that would be recommended here? Adding 1k to a typical passive guitar output impedance should ot make too much difference. 100k on the other hand...

Since this is a guitar input, I would be more concerned about the quality as well as the value of the level pot on the guitar itself.  I would use a shorting jack on the input of the amplifier so the input is shorted any time it is not in use.  I have seen values from 1 K to 47 K here but you have to calculate a value that adds minimal noise and provide adequate protection for the protection diodes.

3) Zeners: I added Z1,2 for JFET protection as per Rons suggestion (if I understood that right). What values to choose here? The OPAx134 datasheet specifies a mix input voltage of supply +0.7V on both ends but surly it would be advisable to set it to much less than the absolute maximum. Z3,4 are to protect the opamp from clipping and will be chosen depending on power supply voltage and  datasheet. I also want to add an indicator circuit that kicks in as soon as the diodes conduct and keeps it glowing for a fraction of a second even after they stop. Have to think about that a bit, can't be too difficult with a transistor as a current source and a cap to keep the charge a bit.

There is one caveat with Zener diodes: they have a lot of capacitance and in fact, I have heard of people using them as voltage-variable capacitors in audio filters with some devices being up to 15 nF.  Use back-to-back 1n4148 diodes in series with the series Zener diodes to keep the capacitance down to the low pF values. In some cases, normal protection diodes strung from the inputs to the supplies would be simpler and better.

4) Gain control: If it is placed as it is in the drawing, then the variable R3 forms a variable high pass with C3, meaning that the bandwidth decreases as gain increases. We might want to keep the bandwidth constant, though. So maybe move the gain control to the path to ground (R4)? Then it only affects the lower end of the bandwidth, which probably wont bother us as much in terms of noise?

If you cut the bass response, it will affect things since this is a bass amplifier and your gain variation requirement may affect your frequency response more than you can tolerate.  I would just do the systems design and select a fixed gain that would work.  You don't want the gain setting to affect the frequency response at this point in the system.

I almost miss the days when I soldered my first bazz fuss and just marveled at the fact that it made any sound at all without bothering why exactly it did what. Growing up is hard godsdamnit!

Cheers,
Andy

[Rob Strand]

I have not kept up with the latest in op amps for low-noise applications but the OPA827 listed by Paul seems to be as good as any with a midband noise of 4 nV/root Hz rising to 6 nV/root Hz at the lowest required frequencies.

Ultimately the pickup thermal noise will dominate.

I have seen values from 1 K to 47 K here but you have to calculate a value that adds minimal noise and provide adequate protection for the protection diodes.

A large proportion of "pro" bass amps are in the 10k zone and that does add a noticeable amount to the pickup thermal noise.  For bass amps I've been using 2k2 to 4k7.

The opamp's 6nV /rtHz pans out at about Requiv = 2k2.

[Fancy Lime]

@maartendh

What about Craig Anderton's idea: guitar signal goes directly to non-inverting input - no c1, c2, r1, r2 or zeners. Can be found in one of the last issues of Device (on Mark Hammer's site).

I can't find the article you mentioned because Marks page does not load (server problem?). I don't think we can do away with C1 and R2, we need to bias the input somehow if unless we can be sure that the preceding device delivers a signal referenced to the right DC level. R1 is for protection of the JFET input from electrostatic discharge and similar nastiness and c2 is to kill radio frequency interference.

@PRR
Yepp, the OPA827 datasheet doesn't look too bad. I'll have to see if I can find it anywhere, though. My usual suppliers don't have them.

@Rob Strand

Ultimately the pickup thermal noise will dominate.

Well that's the goal, right? If the noises I cannot influence (i.e. pickups or whatever else might precede the amp input) clearly drown the noise produced by the amplifier itself, then I have made the, for all practical incense and porpoises, quietest amp possible. The best thing I can do and what I intend to try my best at, is not to significantly add to the noise that is, from the perspective of the amp, god-given. 2k2 for the input resistor sounds like a good place to start.

@amptramp

1/f noise can override the midband noise figure to the extent that the best midband performance is not necessarily give you the best bass performance.

I think in terms of 1/f noise for audio applications, Messrs. Fletcher and Munson are our friends, for a change. And even apart from that, low frequency noise appears to be much less noticeable annoying than high frequency hiss in real-life audio recordings, even if their perceived loudness when soloed is comparable. For scientific instrumentation and the like, this is of course completely different, where you want to have best possible SNR for all frequencies you want to analyze. But in my application, I think it may be worth to consider the "subjective total noise", as perceived by humans, instead of the objective noise as measured by a scope. Of course there is a balance in there somewhere, at some point optimizing hiss and neglecting rumble will also have negative impact on the subjective noise. And of course I'm talking out of my hindquarters a little, because to judge the subjective side I have to build the thing first and compare different versions.

Since this is a guitar input, I would be more concerned about the quality as well as the value of the level pot on the guitar itself.  I would use a shorting jack on the input of the amplifier so the input is shorted any time it is not in use.  I have seen values from 1 K to 47 K here but you have to calculate a value that adds minimal noise and provide adequate protection for the protection diodes.

Right, protecting the protection. https://xkcd.com/917/ Since were not going to blast the input with a serious high current source, 2k2 or 4k7 should probably be ok.

There is one caveat with Zener diodes: they have a lot of capacitance and in fact, I have heard of people using them as voltage-variable capacitors in audio filters with some devices being up to 15 nF.  Use back-to-back 1n4148 diodes in series with the series Zener diodes to keep the capacitance down to the low pF values. In some cases, normal protection diodes strung from the inputs to the supplies would be simpler and better.

Right, forgot about the capacitance. I even looked into the whole "zeners as varicaps" thing at one point and wanted to use it in a voltage controlled variable state filter (which I gave up as too unpredictable and complicated). But isn't the capacitance only present at the cathode? So that when we connect die zeners cathode to cathode instead of anode to anode as in the schematic, the "outside world" doe not see the capacitance until the zener breakdown (at which point we are busy frying bigger fish anyway)?

If you cut the bass response, it will affect things since this is a bass amplifier and your gain variation requirement may affect your frequency response more than you can tolerate.  I would just do the systems design and select a fixed gain that would work.  You don't want the gain setting to affect the frequency response at this point in the system.

If the gain control is dune such that it alters bass response, C4 would of course have to be set so that the high pass cutoff always stays well below the audio band. But it would mean that the make the amplification band unnecessarily large (at the bottom end) at low gains. A fixed, relatively large gain with variable passive attenuation would be a solution as well. Maybe switching between a number, say 6, different gains with a rotary switch would allow to simultaneously change the caps to keep the bandwidth optimal. I wonder if that is "better enough" noise wise to justify the extra hassle.

Cheers,
Andy

P.s.: This thread turn out to be extremely educational for me. Great big thanks to all contributors! 

[Prehistoricman]

You could use LEDs to limit the input swing. Lower voltage LEDs typically have lower capacitances.

https://electronics.stackexchange.com/questions/86717/what-is-the-latency-of-an-led

[Rob Strand]

Well that's the goal, right? If the noises I cannot influence (i.e. pickups or whatever else might precede the amp input) clearly drown the noise produced by the amplifier itself, then I have made the, for all practical incense and porpoises, quietest amp possible. The best thing I can do and what I intend to try my best at, is not to significantly add to the noise that is, from the perspective of the amp, god-given. 2k2 for the input resistor sounds like a good place to start.

The thing is the pickup thermal noise is quite high in comparison to low noise opamps.

The thermal noise generate by the pickup is determined by the resistive part of the impedance looking into the pickup terminals.   With a pickup this isn't just the DC resistance of the pickup.   The resistive part is actually a function of frequency.

If you look here there are *impedance curves* for pickups,
https://courses.physics.illinois.edu/phys406/sp2017/Lecture_Notes/Guitar_Pickup_Talk/Electronic_Transducers_for_Musical_Instruments.pdf

The first thing to notice is how high the impedance gets.  Hundreds of k's.  In a real guitar the tone control and the volume control are in parallel with the pickup which brings that down.  The cable capacitance and the input impedance of the amp also have a small effect.     Very roughly the top of the impedance peaks is an indicator of the resistance at that frequency.    The peak is going to be over 50k ohms.

The second thing to notice is the peak occurs at high frequencies.   So we not only have a high source resistance, the resistance is high in a frequency region contributes more to the overall noise (due to the wide bandwidth).  On top of that it is in a region which stands out more because our ears weight the noise higher there (approximately A-weighted) and for bass there's bugger all signal to mask it.

When you consider those details the effect of the opamp is watered down significantly.   You might get low noise with the guitar volume turned down but it actually goes up quite a bit when the pickups are present - and there's no way to reduce it.

Edit: Here's a short and simplified explanation:
https://pl-1.org/getproductfile.axd?id=238&filename=AN6605.pdf

[Rob Strand]

FWIW, I pulled up some of my old files.   

A pickup with volume control, tone control (set to treble), 500pF cable + amp capacitance + 470k amp input load will produce an A-weighted noise level like an opamp with about 22nV/rtHz noise.

The impedance curves in that link show a large variation between pickups.  The pickup model I used sat somewhere in the middle in height and frequency of the single coils.

So without getting too caught-up with specific pickups you can see the pickup thermal noise is significant.

[amptramp]

If the gain control is dune such that it alters bass response, C4 would of course have to be set so that the high pass cutoff always stays well below the audio band. But it would mean that the make the amplification band unnecessarily large (at the bottom end) at low gains. A fixed, relatively large gain with variable passive attenuation would be a solution as well. Maybe switching between a number, say 6, different gains with a rotary switch would allow to simultaneously change the caps to keep the bandwidth optimal. I wonder if that is "better enough" noise wise to justify the extra hassle.

Don't change the setting while you are on stage or you will be remembered as the guy who was at the epicentre when the earthquake struck!  The large gain with variable attenuation at the output would be a good idea.  Since this is an amp for a bass guitar, the upper bandwidth need not be all that high.  One album with limited bandwidth is "No Jacket Required" by Phil Collins from a third of a century ago.  There is almost nothing above 5000 Hz on it and no one seemed to be upset about it and that was for everything - lead guitar and vocals as well.  You are looking at 41 Hz as the minimum frequency.

[Rob Strand]

Hard to believe but I found someone else (Tilmann Zwicker or M.Zollner?) who computed the thermal noise of a guitar pickup,

https://gitec-forum.de/wp/wp-content/uploads/2017/08/poteg-5-12-noise.pdf

Slightly more capacitive load than my example and slightly lower noise but certainly within the same zone given the wide variation across pickups and loadings.

BTW:   The site actual has a book on offer "Physics of the Electric Guitar".
            Make sure you get the English version.

[Fancy Lime]

@Rob Strand
Thanks for the noise analysis paper! The whole pickup noise thing makes me think there really is no way around just building a few variants of the input stage and throwing all likely relevant sources (active passive, different PU configurations and impedances) at it and measure the noise.

@amptramp

Don't change the setting while you are on stage or you will be remembered as the guy who was at the epicentre when the earthquake struck!

You're trying to make me change the setting, aren't you? I'm too old for a proper Rock'n'Roll death but your suggestion sounds like the next best thing after 27.

About the noise calculations and bandwidth:
In noise calculations we often talk about the noise power, which is calculated by integrating the noise over the frequency spectrum. See http://ww1.microchip.com/downloads/en/AppNotes/01228a.pdf as suggested by Ron R. Therefore, we can get much better numbers if we restrict the bandwidth because there is less to integrate over. So if, for example we limit the bandwidth of out 10MHz opamp to 20kHz, we get a much lower noise power. While that looks nice on paper, I ask myself if that really matters. Don't know 'bout you, but I can definitely not hear 1MHz hiss. So is noise power really all that useful for our particular purpose? Sure, cutting the bandwidth to exactly the upper end of the "useful frequency spectrum" of our instrument, say 6kHz, will remove the his above 6kHz from the line output without altering the sound significantly. But it will not even matter through a guitar speaker, which would not be able to reproduce these frequencies anyway. So, while I get that bandwidth restriction is useful for shutting out Radio Yerevan and potentially suppressing high-frequency oscilations and stabilizing the opamp, does it really make an audible difference to the "perceived noise" such as hiss?

Cheers,
Andy

[Rob Strand]

Therefore, we can get much better numbers if we restrict the bandwidth because there is less to integrate over. So if, for example we limit the bandwidth of out 10MHz opamp to 20kHz, we get a much lower noise power.

When the numbers show something your ears don't hear it's a sign something is wrong!

The way this is handled is by band-limiting before taking the rms and/or by weighting functions.  If you don't band-limit your noise measurement you will get all sort of misleading results and conclusions.

For the simplest case of 20kHz bandlimited and unweighted noise you integrate the output of a 20kHz filter.  or for digital stuff (assuming no aliasing problems due to sampling) you just stop the integration at 20kHz.

A-weight attempts to weight the spectrum to produce a final number that is more like what you hear.  The A-weighting is just a filter.   It has a low-pass filter built into the weighting however the roll-off isn't very steep so signals above 20kHz will still contribute.   So at some point (I can't remember the year) the A-weighting was refined to only integrate up to 20kHz.

A "problem" with A-weighing is the numbers can hide hum levels.   IMHO this is just a problem with trying to represent a complicated thing with a single number.  Obviously if you have hum problems you want to get rid of them, or at least reduce them.

[Rob Strand]

Here's an example of the A-weighting noise from 1k resistor.

The green trace is the cumulative sum of noise (note the scale is 1/1000 so it's in uV).
You can see that the low-pass filter in the A-weighting does not contribute much over 20kHz
and the integrated noise levels-off.  However you can also see it still contributes 0.3dB from 20kHz
to 100kHz.

The red trace is bit more abstract and artificial.  It the *averaged* noise density *upto* that frequency (not at that frequency).    The A-weighted noise at over 20kHz is less than that for the 1k resistor.   The weighting boost some frequencies and cuts others.  Upto 5kHz the two effects balance out.

The main thing to show here is the total noise levels in the green trace and that integrating upto higher frequencies doesn't contribute much.  The rest is just FYI.

[Fancy Lime]

Hi Rob,

ahaa! That might be more relevant to my task, then. I have procured some common low-noise "audio opamps" to test with the design principles suggested in this thread (OPA2134, LM4562, plus the usual suspect NE5532) to see how much difference the FET input makes. Very much looking forward to put these in an A-B test rig, just no time to do that this or next week, unfortunately.

Andy 

[Rob Strand]

Very much looking forward to put these in an A-B test rig, just no time to do that this or next week, unfortunately.

Good luck with it all.  It's quite a few years since I actually sat down and did that type of test. 

My basses were all single coils and I had a lot of problems with buzz around my house.  The room I did the tests in had no fluorescent light either.   It didn't help the cause.   You really need an old pickup sitting inside a steel biscuit tin, maybe wrapped in foil.   A few amps I made ended up with LF356's on the first stage because I had a few of them and they were relatively quiet (IIRC 10 or 12nV/rtHz on paper).


EDIT:
One thing I forgot to mention way back in the thread:  A good many on-board bass preamps use low-power opamps.  Even on high-end basses.  These are inherently high noisy.    Once you have that sitting in the first position of the signal chain  the whole noise optimizing thing falls in a heap!

[amptramp]

The LM318 has an interesting architecture.  You can take both inputs to the negative supply, shutting them off, and connect external FET's to the compensation inputs which are at the collectors of the input transistors.  The LH0061CH was a FET input op amp made with an LM318 and a pair of FET's on a single substrate and it was the first FET input op amp with video speed.  (And I was the first customer for it in Canada.)  You can compare n-channel FET's with a test fixture like this.

@Rob Strand: We tested LF356 amplifiers in one program we were on and their quoted noise level does not stand up.  They quote 12 nV/root Hz at 1000 Hz and 15 nV/root Hz at 100 Hz.  Our tests showed they could barely get down to 20 nV/root Hz.  If they were OK, a TL071 would probably be just as good or better.

[Rob Strand]

@Rob Strand: We tested LF356 amplifiers in one program we were on and their quoted noise level does not stand up.  They quote 12 nV/root Hz at 1000 Hz and 15 nV/root Hz at 100 Hz.  Our tests showed they could barely get down to 20 nV/root Hz.  If they were OK, a TL071 would probably be just as good or better.

Interesting results.  I'm not really surprised.  The LF356's I used definitely beat my TL071/72's.

I just checked the datasheet, and yes the LF356's are quoted at 12 nV/rtHz at 1kHz.   What is very suspicious is the LF355's, which supposed to be uncompensated LF356's, quote 20nV/rtHz.  So maybe some manufacturers, or eras of manufacture, produced that level of noise.   The datasheets don't seem to quote upper limits on the noise so maybe that's how they got away with it.     Devices targeted for low noise often put a max figure on the noise voltage but the generic stuff is open to the evils of the world.

It would be interesting to dig up some old articles which quote noise voltages.  I'm pretty sure there were some with LF356's.   Some well-known phono preamps had LF356's but I can't remember if they used a raw device or they had an added transistor or JFET stage at the front [actually, I'm pretty sure they would have to].

I'm definitely *not* recommending LF356's.   There's plenty of better devices out there.

[bool]

There is not much problem with f.e. 06x bass onboard preamps; many of these have pre-emphasis/de-emphasis "noise reduction" built-in. At least there isn't many problems with "dual-opamp" preamps.

Noise can build-up with "overengineered" preamps (such as schack preamp from 90's with 2x 064), but even that one was quiet enough for serious studio work.

Even for "hi-fi" bass sound you don't need much of content above 12kHz, so you can band-pass and cut off the hiss and chirps.

The recipe for good snr is and always was simple: hot-enough pickups, low-enough noise preamp, bandpass (can be also rfi LCR filtering), ultra-clean supply rails. Copper/alu cavity wrap for live; Graphit 33 spray for "studio bass".

Do you realize how many major releases were cut with basses running noisy lm4250 opamps (musicmans, GLs), thru noisy sansamps (with 10k/22n inputs), basses thru noisy muffs etc ...

If noise is your concern, I would primarily look into getting a higher output pickups. LF356 is "battery hungry", NE5534 too. Try "old technology" MC33078, 33079 (quad).

[dschwartz]

BTW
On LTspice, you can simulate the noise behavior of a circuit..

[Rob Strand]

There is not much problem with f.e. 06x bass onboard preamps; many of these have pre-emphasis/de-emphasis "noise reduction" built-in.

Actually Pre-emphasis can't improve the first stage.  The first stage always contributes the full noise of the chosen opamp.  So those you can't actually hide the high noise level of those low-power opamps.

Where pre-emphasis/de-emphasis works is when you have many opamps.   

For a two stage bass preamp the first stage contributes full noise.  Without pre-emphasis/de-emphasis the second stage opamp would contribute equally so the noise is 3dB higher than the first-stage opamp.   With pre-emphasis/de-emphasis the second opamp contributes a lot less than 3dB.

Suppose your first stage is followed by a graphic equalizer which has say 10 opamps.   Without pre-emphasis/de-emphasis  the noise would go up by a factor of sqrt(10+1) = 10dB.   In an ideal situation pre-emphasis/de-emphasis should keep just above the first stage noise.  In practice it is nowhere near that but the noise increase is lot less than 10dB.

(A real graphic which contribute more than just opamp noise.)