One compressor design I'm working on currently employs the above mentioned pair of transistors in every stage. There are quite a few: a buffer stage, a gain stage, further gain+rectifying stages. For peace of mind, I settled on the most general purpose among general purpose transistors, but I see many use something from the BC5xx series for audio applications, and supposedly they have better noise figures. Also, the 3904/3906 have relatively low current gain, that is, an hFE around 100 typical (which, as I understand it, is still a good, workable value), whereas other transistor types start from 200 or 400. So, how much does it matter?

Quote from: fryingpan on October 06, 2024, 03:37:23 PMOne compressor design I'm working on currently employs the above mentioned pair of transistors in every stage.

I presume you mean either Complementary Feedback Pairs and/or Sziklai pairs, no..??
Or n-p-n to p-n-p (and vice versa) directly coupled pair, perhaps..??

If yes, there should be no significant difference of performance between different h_FE implemetation, as long as those stages are well designed..

P.S.
As R.G. said: When in doubt, use a 2N5088..

Quote from: antonis on October 06, 2024, 04:14:00 PM

Quote from: fryingpan on October 06, 2024, 03:37:23 PMOne compressor design I'm working on currently employs the above mentioned pair of transistors in every stage.

I presume you mean either Complementary Feedback Pairs and/or Sziklai pairs, no..??
Or n-p-n to p-n-p (and vice versa) directly coupled pair, perhaps..??

If yes, there should be no significant difference of performance between different h_FE implemetation, as long as those stages are well designed..

What about noise?

Quote from: fryingpan on October 06, 2024, 04:16:26 PMWhat about noise?

Are you building some Hi-End effect..??

If cost is not that different, better is better than worse, innit?

Of course, a guitar or bass signal will have, what, 60dB SNR on the best of days? It doesn't matter that much for this. But if you're using an effect on quieter sources...

More seriously, there isn't any "general purpose quiet" BJT..

Collector current and signal source resistance must be taken into account along with Base spreading resistance..

https://electronics.stackexchange.com/questions/192451/understanding-noise-figures-for-bjt-transistor#:~:text=I%20like%20to%20achieve%20the%20optimum%20noise%20given%20a%205V

Quote from: antonis on October 06, 2024, 04:26:25 PMMore seriously, there isn't any "general purpose quiet" BJT..

Collector current and signal source resistance must be taken into account along with Base spreading resistance..

https://electronics.stackexchange.com/questions/192451/understanding-noise-figures-for-bjt-transistor#:~:text=I%20like%20to%20achieve%20the%20optimum%20noise%20given%20a%205V

I know, it depends on several factors. Still, there must be a rule of thumb.

I usually default to BC550C (NPN) and BC560C (PNP) because they are high hfe (>400 or so), quiet, and the cheapest BJTs money can buy from my usual sources (< 5 cents). For high input resistance and specific currents it is possible to design less noisy stages with 2N5089 or MPSA18 but I would argue that this only matters if you are designing high end studio hardware. Especially because those transistors are much more expensive and hard to get here in Europe. Many fakes around, too. For high current applications, I like me the BC327-40 and BC337-40 or 2N2222 and 2N2907.
Use what you can get easily, it hardly ever really makes an audible difference in our genre of diy.

Quote from: Fancy Lime on October 07, 2024, 11:17:17 AMI usually default to BC550C (NPN) and BC560C (PNP)

Me too but with BC549C & BC559C (both because of lower V_CEO and I've a couple of hundrends of them..)

@fryingpan: About "rule of thumb":
Without known signal source resistance, proceed to high h_FE (and don't bother with r_bb value) and medium to low Collector current..

It's a matter of objectives, which often translates to "how will you know when you have succeeded?"

If your success scoring polynomial includes terms about total cost of parts, then design optimization should address making the thing work with any random transistor you can get on the surplus market. This was, by the way, how Electro Harmonix did pedals in their early years. Maybe still, don't know.

If your scoring polynomial is heavy on "lowest noise", be prepared to really dig through source impedances, resistor materials, transistor types, and so on.

If your scoring polynomial is heavy on quickest time to market, choose the most widely available parts.

If political relevance is important (don't laugh; I once worked at a place where only preapproved parts could be used), then study the preapproved parts A LOT.

My personal view is that I should design so that almost any old NPN or PNP in the correct package will work, at least mostly. It likely won't be the quietest, highest gain, cheapest, whatever-else, but if it works at all, you can improve it with fancier parts.

Keen's Second Law says when in doubt, whip in a 2N5088. If I just had to use a PNP, I would use either a 2N5086, or a 2N3906, as low noise nearly never has a PNP solution. But times change, and the world moves on. Mouser now charges more for the 2N5088 than the BC5xx series.

Shrug. Give me a hundred of those BC thingies and let me get on with it. If it's too hissy, I can fix it. I guess the bottom line is to not rely on getting one specific magic part, or you'll be in trouble when you can't get it.

Quote from: antonis on October 07, 2024, 01:26:22 PM

Quote from: Fancy Lime on October 07, 2024, 11:17:17 AMI usually default to BC550C (NPN) and BC560C (PNP)

Me too but with BC549C & BC559C (both because of lower V_CEO and I've a couple of hundrends of them..)

@fryingpan: About "rule of thumb":
Without known signal source resistance, proceed to high h_FE (and don't bother with r_bb value) and medium to low Collector current..

How is lower Vceo beneficial?

Quote from: R.G. on October 07, 2024, 04:43:19 PMIt's a matter of objectives, which often translates to "how will you know when you have succeeded?"

If your success scoring polynomial includes terms about total cost of parts, then design optimization should address making the thing work with any random transistor you can get on the surplus market. This was, by the way, how Electro Harmonix did pedals in their early years. Maybe still, don't know.

If your scoring polynomial is heavy on "lowest noise", be prepared to really dig through source impedances, resistor materials, transistor types, and so on.

If your scoring polynomial is heavy on quickest time to market, choose the most widely available parts.

If political relevance is important (don't laugh; I once worked at a place where only preapproved parts could be used), then study the preapproved parts A LOT.

My personal view is that I should design so that almost any old NPN or PNP in the correct package will work, at least mostly. It likely won't be the quietest, highest gain, cheapest, whatever-else, but if it works at all, you can improve it with fancier parts.

Keen's Second Law says when in doubt, whip in a 2N5088. If I just had to use a PNP, I would use either a 2N5086, or a 2N3906, as low noise nearly never has a PNP solution. But times change, and the world moves on. Mouser now charges more for the 2N5088 than the BC5xx series.

Shrug. Give me a hundred of those BC thingies and let me get on with it. If it's too hissy, I can fix it. I guess the bottom line is to not rely on getting one specific magic part, or you'll be in trouble when you can't get it.

Well, only the first stage has unknown impedance (it could be 100ohms, it could be 10kohms). The second should have medium source impedance (the second stage sees either a jfet-pnp buffer, or a npn-pnp buffer, and both simulate with low output impedances, and that goes into a medium impedance attenuator, right now it works out at, dunno, 2kohms?). The sidechain circuitry, I'm not too bothered. Noise will be heavily filtered by the envelope generator anyway.

Oh and why does the simulator show huge noise with a jfet as the first semiconductor? I've tried all sorts, from J113s to J201s (too low a Vgsoff for my tastes) to 2N5459 and the sort. Biased with either a 1M resistor to ground or two 2.2M resistors to both rails. Noise for the whole circuit jumps from around 24uV to at least ten times that. According (not only to) Amos's Principles of Transistor Circuits JFETs literally shine as a first stage because of their low noise (and tendency for nonlinearity with high signal levels, which I'm not bothered about).

Quote from: fryingpan on October 07, 2024, 05:12:04 PMNoise for the whole circuit jumps from around 24uV to at least[...]

OK, so your scoring polynomial has a big coefficient on noise performance, yes?

Quote from: fryingpan on October 07, 2024, 05:12:04 PMWell, only the first stage has unknown impedance (it could be 100ohms, it could be 10kohms).

I translate that as "this is not a quitar pickup input". Guitar pickups are about 4K to 18K or more resistive in series with 1 to 4 Henries of inductance. You need upwards of 100K to 1M or more input impedance to avoid selective loss of treble with a guitar signal directly from the guitar. If you're intending to allow 100 ohms to 10K, guitar input doesn't fit.

I recommend you read Henry Ott's book, "Low Noise Electronic Design", which is where some of this comes from. There are two variants of noise from sources, voltage noise and current noise. Different sources have different mixes of these. Minimizing input noise requires adapting the input impedance of your device to the impedance and noise mixture of the source. To grossly oversimplify, source impedances down in the sub-100 ohm range want a low input impedance; the reigning king was at one time a low-rbb' bipolar. 10K sources work well with bipolar inputs, and 100K and up tend to work best with FET inputs. To really optimize noise performance, you usually have to optimize the input stage, and its biasing, to the source impedance and its noise characteristics.

Whipping in a JFET is not always a slam-dunk way to minimize noise. The simplest case is something like a gate leak resistor to ground, so the resistor thermal noise is added to the noise from the source. If you don't get any gain from the FET, its bias noise and internal noise is then amplified by following stages. Biasing techniques matter too. Single resistor biasing gives you the resistor noise. Bias chain strings gives you the noise of the two resistances in parallel. "Noiseless" biasing sets the bias voltage with a string, shunts the thermal noise to ground with a BFC, then leaves in only the noise of the series resistance to the active device control node.

QuoteThe second [...]

In general, the noise performance of a circuit is dominated by its input stage, as this stage takes the input noise, adds to it any biasing noise, adds any amplification noise, then sends this along to the rest of the circuit to be further amplified. Later stages usually can't help but make the sins of the input stage worse.

I have not been too happy with the performance of simulators in estimating actual noise performance of circuits, although they are a useful indicator.

Quote from: R.G. on October 07, 2024, 07:40:02 PMHenry Ott's book, "Low Noise Electronic Design",

Low noise electronic system design by C. D. Motchenbacher is another good one. Pre-owned paper books are available. There seems to be an academic copy here.

If you must use JFETs..... at least get fresh ones, not warehouse dregs. JFE2140 or JFE150. Yeah, costs as much as a filling lunch.

SPICE would work, if you ask the right question, and if the model is accurate. Most are NOT. rbb is 'always' 10 ohms, which is BS. 1/f rumble is not clearly simulated in any model I looked into. (1/f is often unimportant in BJTs and JFETs in the audio band, but can dominate in vacuum tubes and in MOSFETs, and <<1Hz measurements.) I don't think anybody figured flicker noise, only try to keep the wafer clean.

Noise is not *that* important for this particular circuit. But it's an opportunity to learn.

When I say that the source impedance could be anything from 100 ohm to 10kohm with a lot of reactance, I mean that the circuit should work well with a buffer before it, with a guitar, with a -10dBV line input, so it should be minimally flexible.

Quote from: PRR on October 07, 2024, 10:31:04 PM

Quote from: R.G. on October 07, 2024, 07:40:02 PMHenry Ott's book, "Low Noise Electronic Design",

Low noise electronic system design by C. D. Motchenbacher is another good one. Pre-owned paper books are available. There seems to be an academic copy here.

If you must use JFETs..... at least get fresh ones, not warehouse dregs. JFE2140 or JFE150. Yeah, costs as much as a filling lunch.

SPICE would work, if you ask the right question, and if the model is accurate. Most are NOT. rbb is 'always' 10 ohms, which is BS. 1/f rumble is not clearly simulated in any model I looked into. (1/f is often unimportant in BJTs and JFETs in the audio band, but can dominate in vacuum tubes and in MOSFETs, and <<1Hz measurements.) I don't think anybody figured flicker noise, only try to keep the wafer clean.

Honestly, I'm trying to avoid JFETs except where it really makes sense (the gain control element, a booster stage where its high input impedance is more useful than part consistency. I'm designing a circuit to be assembled by the PCB manufacturer (it's actually quite cheap, all things considered) and I'll be using mostly SMD parts. My next stage of the design is to minimise the number of different parts, in fact.

Quote from: fryingpan on October 08, 2024, 03:13:50 AMWhen I say that the source impedance could be anything from 100 ohm to 10kohm with a lot of reactance, I mean that the circuit should work well with a buffer before it, with a guitar, with a -10dBV line input, so it should be minimally flexible.

You know, buffers just kick the (noise)can down the road..
(S/N ratio is passed almost unaltered to succeding stage..)

As another rule of thumb (you're fond of rules of thumb, don't you..?? ), for signal source impedance up to 50k (or so) use BJTs..

Quote from: antonis on October 08, 2024, 04:45:48 AM

Quote from: fryingpan on October 08, 2024, 03:13:50 AMWhen I say that the source impedance could be anything from 100 ohm to 10kohm with a lot of reactance, I mean that the circuit should work well with a buffer before it, with a guitar, with a -10dBV line input, so it should be minimally flexible.

You know, buffers just kick the (noise)can down the road..
(S/N ratio is passed almost unaltered to succeding stage..)

As another rule of thumb (you're fond of rules of thumb, don't you..?? ), for signal source impedance up to 50k (or so) use BJTs..

Of course. The worst case scenario, at that point, is a magnetic guitar or bass pickup. That said, the circuit already features a buffer because it's better to attenuate with low value resistors than with high value ones, and with a buffer I can avoid large resistances in the signal path. A buffer should produce less noise than a 1M series resistor or so. I know that you need high value resistors for a high input impedance, but those are shunted to ground, and a CFB buffer gets a lot of its high impedance through feedback, which should lower noise too.