Quote from: fryingpan on October 14, 2024, 04:38:27 PMBear in mind that bootstrapping may cause a low frequency peak (like at 1Hz or so). You might want to simulate AC response starting from, say, 0.1Hz. When bootstrapping, you can't use too large a capacitor because this will cause a larger peak at a very low frequency. Better, if you want to use a very large cap, you need to limit low frequency extension with an input cap so that the peak is well below unity.

I have +6 to +9db of gain at around 1hZ alright. I'm not entirely sure why I was using such a large input cap - dropping it down to 100nF got rid of most of that gain without sacrificing the large output cap (which I have for the reasons Douglas Self outlined in that book). I think I was think of the multiplicative effect of stacking up multiple highpass filters in a signal chain - forgetting, of course, that I have control over the input impedance and can set it wherever I need it to be at. I am now at a more manageable +3db gain at most in the subsonic frequencies. I think this should suffice to prevent against the build up of those potentially very problematic frequencies. If not, I may have to refine this design further. I will attempt Rob's suggested method for reducing voltage offset and come back with the results

The higher the hfe of the input devices, the lower the bias currents for a given collector current, to restate the obvious. Each device type (actually, fab process doing the diffusions and epitaxy) will have a specific current point where its input noise is minimized, and for best noise performance it needs to be run there. This may not be the best gain or bandwidth point. But low input bias currents let you use higher value resistors to get higher input impedance - suffering only the added thermal noise of the higher input resistors to get there.

For lowest offset, you really want either hand-matched or monolithic input differential transistors. Even if you get well matched discretes, they may not be matched under thermal drift effects. Gluing a pair of TO-92s face to face and enclosing them in a styrofoam blob helps as long as they don't dissipate much. Ideally, you'd get discretes from the same wafer, or even better, a dual monolithic pair.

For a quick(ish) intro to noise considerations, find a pdf of National Semiconductor's AN104.

In general, it's possible to get better noise performance ICs than you can make unless you're really good with discrete design and have good lab equipment, even trying for lower cost. The LM833 comes to mind here. It's 1/4 the noise per root hertz of a TL07x, which is already pretty good.

Quote from: Rob Strand on October 14, 2024, 04:00:14 PMThe existing design can sink 6mA, so lets say you want to source 6mA as well, then Q3 needs to supply 12mA:  IC=12mA, hFE=300 => 40uA.  So a 200uA bias will be enough but the diff-pair will bias asymmetrically at idle.

Circling back to this, what determines this circuits ability to sink current vs its ability to source current? If the diff pair can source 6mA with a bias current of IC = 200uA, why would Douglas's original configuration require it to draw 3mA? Was it the low gain transistors he used (MPSA42)? And if in that original configuration, Q3 is biased for IC = 6mA and can indeed sink 6mA, why does changing the IC on Q1 and Q2 require Q3 IC to equal 12mA as you suggested?

Posting my schematic of this (no bootstrapping or bias offset, as the offset voltage is relatively low, though one of those is probably worth doing on a lower voltage power supply).




P.S. Yes Q7 absolutely is cooking itself beyond its rated power dissipation. 

> Each device type (actually, fab process doing the diffusions and epitaxy) will have a specific current point where its input noise is minimized, and for best noise performance it needs to be run there.

There's two minimums. Lowest hiss voltage, lowest hiss current. Hmmm, V/I is an impedance (nearly a resistance). Audio designers speak of "OSI", optimum source impedance--- for a specific source impedance. That source also has noise voltage and noise current. We want the amplifier's noise source to BOTH be low (as possible) compared to source impedance noise. This wraps into a concept of Noise Figure. NF=0dB is unobtainable, 2dB is really good, an astonishing amount of music passes through preamps with NF of 10dB and more.

What is the noise impedance of an e-guitar? There's R and L and C and more Rs and Cs..... We can load guitar with 10k and the bass and mids are fine but the treble is dull, even with 100k load. There's a treble peak, mode`rated by volume and tone pot action.

All modern clean transistors have similar noise plots. 2N3904 can proxy for any of them. If the source is 150r we want to run near 0.5mA. If the source is 10k then 50uA may be better. This was obvious on a few of the early phono preamps with inputs run at mA level.


Uh, that's for midfrequencies. Hiss may rise in the RF band but we hardly care. Hiss (as Rumble) rises below ~~100Hz and that seems to be a fundamental property of matter, though 'enhanced' by dirty processing. That matters for RIAA Phono and for lab mikes, not so much in Real Life.

Quote from: mzy12 on October 14, 2024, 04:35:02 PMMy idea is that this will be lower noise than a tl072 and the like/lower cost than a nice opamp.

It will be quieter on low impedance sources. At 6mA it may be intolerable on guitar.

Sources below 10k, a 5532 will be nearly-best. Above that (to 300Meg condenser!), TL072 is not perfect but very nearly so.

5532 in PDIP is $0.61/ea, discrete NPNs are going $0.10 from mystery-brand makers or $0.24 from On/Semi ($1.48 plus resistors and diodes). That opamp is not well protected against shorts and surges.

Quote from: mzy12 on October 14, 2024, 06:44:01 PMCircling back to this, what determines this circuits ability to sink current vs its ability to source current? If the diff pair can source 6mA with a bias current of IC = 200uA, why would Douglas's original configuration require it to draw 3mA? Was it the low gain transistors he used (MPSA42)? And if in that original configuration, Q3 is biased for IC = 6mA and can indeed sink 6mA, why does changing the IC on Q1 and Q2 require Q3 IC to equal 12mA as you suggested?

Posting my schematic of this (no bootstrapping or bias offset, as the offset voltage is relatively low, though one of those is probably worth doing on a lower voltage power supply).

The way a class-A output stage works is the current source determines the maximum current in one direction. In your case that's the maximum negative output current.  At zero output current the idle current of the top transistor current equals the current source current.  So without driving anything the top transistor is already passing as much current as the bottom current source transistor.   When the output swings positive, if you want the positive output current to reach the maximum negative output current then it must pass twice the current source current.   If the current source passes 6mA then the top transistor must pass 12mA.

All that is only to do with the output buffer.

Where the diff-pair output drive comes in is the ability to supply enough base current to Q3 during the maximum positive swing and load.   If we say the current source is current IQ and the maximum peak current of Q3 IC3_max = 2*IQ, then in order to pass the maximum positive current Q3 needs a base current of Ib3_max = IC3_max / hFE3.    In the example IQ=6mA, IC3_max = 12mA, and suppose hFE=100 then Ib3_max = 12mA/100 = 120uA.   At idle Ib3 = IQ / hFE3 = 6mA / 100 = 60uA.   So in order for the diff pair to supply the enough base current to Q3 it the tail current needs to be more than IEE=120uA.   The maximum output from the diff-pair is equal to the tail current.

Quote from: PRR on October 14, 2024, 07:13:55 PMAll modern clean transistors have similar noise plots. 2N3904 can proxy for any of them. .

If you compare those noise figures to a BC109/BC549 you will find they are *much* higher.

I've got some simple spreadsheets which plot those curves given hFE and rbb'.

The BC109/BC549 curves can be matched with quite feasible looking hFE and rbb' values.

For those non "low-noise" transistors you end-up non feasible hFE values.    What's missing in the theoretical curves based on hFE and rbb' is additional noise from surface effects which are often 1/f in nature.



Notice how the 2N3904 curves don't go below 3dB for any current or source impedance.

---

Here's the theoretical noise fitted to the Mullard data for a BC109 (something like a BC109B):



It's a pretty good match.   At IC=10mA you can see the required hFE is probably smaller than reality.   I'm fairly sure that's
because the excess 1/f noise increases at higher currents like 10mA.

The rbb' figure is quite close to what most papers on noise quote for the BC109.

Quote from: PRR on October 14, 2024, 07:28:47 PM5532 in PDIP is $0.61/ea, discrete NPNs are going $0.10 from mystery-brand makers or $0.24 from On/Semi ($1.48 plus resistors and diodes). That opamp is not well protected against shorts and surges.

Once you bump the quantity up to 100, the transistors I need are around €0.05. You need seven for each buffer. 100/7 = ~15. Lets make 100 of these. (€5/15)*100 = €34 for transistors. With 8 resistors (lets round it up to ten because you'll inevitably need more) at €0.02 each, that €0.02 * 10 * 100, which equals €2, bringing us up to €36.

I'm not including capacitors because that's common to each. Same with those additional protection elements.

Mouser's cheapest Ne5532 is TI's NE5532DR. If you haven't noticed by now, I am a-okay with using SMD parts
We'll go with Douglas Self's suggestion and use the two parts of this dual opamp to create a lower THD buffer. It only seems fair since we're using his suggested design for the discrete transistor buffer. That's gonna add up to ~€24 per 100 buffers. We'll forget about the resistors we need for this design because I was a tad generous on the transistor pricing. We're better off by €12, which is good. What's not so good is, if we are to believe again the measurements of Douglas Self, Texas Instrument's NE5532 opamps suffer from much greater high frequency distortion than their competitors. There's no point in even glancing at the competitors NE5532 offerings if we are worried about price, as they're all 3x-4x the cost. Furthermore, the input impedance of the NE5532 does not reach 1M, which is often a necessity in guitar circuits, but that can become a non-issue if we have total control over the signal path.

In conclusion: I don't know what's better 

Quote from: fryingpan on October 14, 2024, 04:38:27 PMBear in mind that bootstrapping may cause a low frequency peak (like at 1Hz or so).

P.S.
Above formulas derived from Appendix 6 & Chapter 15 of Electronic Designer's Handbook

Quote from: mzy12 on October 14, 2024, 08:11:39 PMIn conclusion: I don't know what's better 

Whip out the wallet and use single OPA2156 buffer I mean, can you beat ESD protection, 0.0001% THd and 3 nV/√Hz noise with discretes?

That sure is an interesting way to avoid using a JFET...

I think present thread started with a discrete unity gain buffer issue (input impedance)..

How did it come to op-amp/THD/noise circuits considerations..??   

Quote from: diffeq on October 15, 2024, 06:14:58 AMWhip out the wallet and use single OPA2156 buffer I mean, can you beat ESD protection, 0.0001% THd and 3 nV/√Hz noise with discretes?

You better believe I can't!  Though as typical with CMOS input devices, voltage noise suffers below 100hZ and is decidedly unimpressive at the lower octave of human hearing.

Quote from: antonis on October 15, 2024, 07:12:29 AMHow did it come to op-amp/THD/noise circuits considerations..??   

Inevitably, one must justify using a particular design when presenting a question on the internet

Quote from: merlinb on October 15, 2024, 06:28:04 AMThat sure is an interesting way to avoid using a JFET...

Speaking of which

JFETs vary too much with Vgs(off). They're more expensive, generally have lower voltage handling capabilities and poorer noise and THD performance. They're great if you need a couple hundred megaohms of input impedance, but require even more wrangling than what I'm working with here.

Quote from: Rob Strand on October 14, 2024, 08:01:59 PMIf you compare those noise figures to a BC109/BC549 you will find they are *much* higher.

...

Quote from: PRR on October 14, 2024, 07:13:55 PMThere's two minimums. Lowest hiss voltage, lowest hiss current. Hmmm, V/I is an impedance (nearly a resistance). Audio designers speak of "OSI", optimum source impedance--- for a specific source impedance. That source also has noise voltage and noise current. We want the amplifier's noise source to BOTH be low (as possible) compared to source impedance noise. This wraps into a concept of Noise Figure. NF=0dB is unobtainable, 2dB is really good, an astonishing amount of music passes through preamps with NF of 10dB and more.

...

This is quite the food for thought when it comes to low-noise designs 

Quote from: R.G. on October 14, 2024, 06:39:20 PMBut low input bias currents let you use higher value resistors to get higher input impedance - suffering only the added thermal noise of the higher input resistors to get there.

Aaaah okay that does make sense to me.

Quote from: R.G. on October 14, 2024, 06:39:20 PMIn general, it's possible to get better noise performance ICs than you can make unless you're really good with discrete design and have good lab equipment, even trying for lower cost. The LM833 comes to mind here. It's 1/4 the noise per root hertz of a TL07x, which is already pretty good.

The LM833 does look very nice alright, but unfortunately it won't work when higher input impedances are required - onsemi suggests the MC33078 as its, "higher performance," replacement, with 175k input resistance, which I'm going to assume the LM833 is similar to.

Quote from: mzy12 on October 15, 2024, 07:51:19 AMJFETs vary too much with Vgs(off). They're more expensive, generally have lower voltage handling capabilities and poorer noise and THD performance. They're great if you need a couple hundred megaohms of input impedance, but require even more wrangling than what I'm working with here.

You already have a bipolar supply, current mirror, and *two* CCS's    I humbly suggest that is more than enough to deal with Vgs variance.

BTW, why +/-24V rails? That's a whole lotta swing for a guitar.

Quote from: mzy12 on October 15, 2024, 07:51:19 AMThis is quite the food for thought when it comes to low-noise designs 

https://www.diystompboxes.com/smfforum/index.php?topic=132504.0

If we are considering input impedance and your source is a guitar, you will have the impedance at the top volume control setting of the resistance of the pickup as modified by inductance and capacitance and the parallel volume control.  If the volume control is at zero, you have zero input impedance.  There is a spot slightly above half volume where you have roughly half the volume control resistance in parallel with the series combination of the pickup and the top half of the volume control, so with a typical pickup of 10 Kohms and a 500 K volume control, you would get half of 510K or 255K as each parallel resistance in the source impedance and therefore, a source impedance of 127.5 K.  This helps with voltage noise because the equivalent resistance of this is in parallel with the input resistors in the preamp as long as the capacitive coupling to the input has negligible impedance (another reason for the high input capacitance in the original design).

If you have no noticeable noise change when running the guitar volume control through its range, then you don't have an impedance problem.

This circuit is already making use of bootstrapping. You can use bootstrapping with an opamp too. Have a look at Rod Elliott's ESP website.

Quote from: merlinb on October 15, 2024, 08:00:53 AMI humbly suggest that is more than enough to deal with Vgs variance.

BTW, why +/-24V rails? That's a whole lotta swing for a guitar.

You're absolutely right, that is enough to deal with JFET weirdness. They're still far more expensive than BJTs though. A single unit of the cheapest JFETs mouser has to offer could get me like ten MMBT3904s.

And the +/-24 is entirely used as an exercise in how far I can push the system. I have no intention of using such a supply in an actual guitar effect.

Quote from: fryingpan on October 15, 2024, 11:59:38 AMThis circuit is already making use of bootstrapping. You can use bootstrapping with an opamp too. Have a look at Rod Elliott's ESP website.

I haven't found as much luck on finding information about how bootstrapping can effect opamps as I have had finding out how they effect BJTs. I will check through that website now.

Quote from: amptramp on October 15, 2024, 10:23:11 AMIf you have no noticeable noise change when running the guitar volume control through its range, then you don't have an impedance problem.

You're quite right. If I can guarantee I have a suitably high input impedance buffer between the guitar and this buffer, then, in general, I don't need to worry about the source impedance being much more than, say, 20k. I'm not saying I always need to have this buffer be at a very high input impedance, my original question was really just the how, not the why that said, I have no problem in discussing the why, especially if that means I can come to a better conclusion on how to do something specific.

Quote from: Rob Strand on October 14, 2024, 04:00:14 PMAnother method is to have the same resistor value on the bases of the two diff-amp transistors, ie add a resistor in series with  Q2

Circling back to this (yet) again, would, having something like a 1M resistors in series with the base of Q2 not introduce a load of unwelcome Johnson noise?


EDIT: Also I don't think I'm even connecting it correctly in my schematic lol ;-;

Quote from: mzy12 on October 15, 2024, 01:22:22 PM

Quote from: merlinb on October 15, 2024, 08:00:53 AMI humbly suggest that is more than enough to deal with Vgs variance.

BTW, why +/-24V rails? That's a whole lotta swing for a guitar.

You're absolutely right, that is enough to deal with JFET weirdness. They're still far more expensive than BJTs though. A single unit of the cheapest JFETs mouser has to offer could get me like ten MMBT3904s.

And the +/-24 is entirely used as an exercise in how far I can push the system. I have no intention of using such a supply in an actual guitar effect.

Antonis is your best friend. Or you are Antonis's best friend, I don't know.

Quote

Quote from: fryingpan on October 15, 2024, 11:59:38 AMThis circuit is already making use of bootstrapping. You can use bootstrapping with an opamp too. Have a look at Rod Elliott's ESP website.

I haven't found as much luck on finding information about how bootstrapping can effect opamps as I have had finding out how they effect BJTs. I will check through that website now.

https://sound-au.com/articles/bootstrap.htm#s2