I came here by way of an existing discussion on the design and function of the Jenson 990 line of discrete op amps, a truly insightful piece of engineering for its time. For those not familiar, Jensen used a matched NPN pair which (of course) is no longer manufactured and being the cheapskate/financially limited soul that I am, thought I'd have a crack at re-working some of the internal design with modern (and lower-cost) parts.
https://www.technicalaudio.com/pdf/...ted/Jensen_JE-990_opamp_JAES_reprint_1980.pdf
According to some at least, the Jensen is a mediocre performer by today's standards: in a simulation but by reputation from people who have "heard" it there's a stark difference.
https://www.diyaudio.com/community/threads/simulation-of-the-je-990-op-amp-by-deane-jensen.107404/
I'll take a pair of ears over a computer simulation any day thank you. Over the years, some improvements have been made along the way, a current mirror was added to the LM394, improved current sinks and one thing that's truly baffled me: the inclusion of two diodes in the collector of the "inverting" input, I've not come across anything like this (except for voltage references and current sources) and I'll have to admit defeat, even in the face of the Art of Electronics 3rd edition.
My tinkering has resulted in component changes appropriate to the mic pre-amp that I'm making and with the inclusion of relatively good, low-cost transistors (MMBT3906/6) and a MAT12 for the differential stages and MMBT390x for the push-pull output. I know I won't be driving a capacitive load, etc. so I can stick to what I know there too.
And here's where things go south for me. My (poor) knowledge of calculus means that I can't work in the time domain as Deane did, only in the frequency domain so (unless there's a trick I'm missing) it's not as easy to plot the various poles and zeros for critical analysis. "So what," you might say.
If you're sitting comfortably, I will begin. I'm going to assume that the r'bb is such that it doesn't matter for something with a 100-300R[?] load (condenser mic) but Deane specifically discusses the effect of the emitter resistance at the design current of 1.6-mA is 16 R, resulting in "a first stage single pole response of 83 KHz". And that, naturally, is where my brain takes its ball and goes home.
Firstly -- where does that pole even come from? There's a capacitor I'm missing here (base, perhaps?) I struggle with this stuff.
I (broadly) understand the use of the inductors in the emitter path thanks to the explanation in the paper but that r'bb won't be the same for my design. A 1mA (for the MAT12s) I figure that's 25-R, which isn't bad, but at 100-uA required to keep the jellybeans quiet that rockets to 250-R and that's really going to upset everything.
I rarely ask for help but this stuff has me stumped on the last leg of a design journey that's taken me a year of free time and a lot of cash. This board is the most expensive part and I need to try and get it right first time.
I've pretty much exhausted my foundational knowledge and something more meaty is required with equations my cat could understand because I'm very seriously weak at math. By the same token, I don't ask nor expect anyone to "do my homework" as it were, just help me with these pesky humps.
Thanks
https://www.technicalaudio.com/pdf/...ted/Jensen_JE-990_opamp_JAES_reprint_1980.pdf
According to some at least, the Jensen is a mediocre performer by today's standards: in a simulation but by reputation from people who have "heard" it there's a stark difference.
https://www.diyaudio.com/community/threads/simulation-of-the-je-990-op-amp-by-deane-jensen.107404/
I'll take a pair of ears over a computer simulation any day thank you. Over the years, some improvements have been made along the way, a current mirror was added to the LM394, improved current sinks and one thing that's truly baffled me: the inclusion of two diodes in the collector of the "inverting" input, I've not come across anything like this (except for voltage references and current sources) and I'll have to admit defeat, even in the face of the Art of Electronics 3rd edition.
My tinkering has resulted in component changes appropriate to the mic pre-amp that I'm making and with the inclusion of relatively good, low-cost transistors (MMBT3906/6) and a MAT12 for the differential stages and MMBT390x for the push-pull output. I know I won't be driving a capacitive load, etc. so I can stick to what I know there too.
And here's where things go south for me. My (poor) knowledge of calculus means that I can't work in the time domain as Deane did, only in the frequency domain so (unless there's a trick I'm missing) it's not as easy to plot the various poles and zeros for critical analysis. "So what," you might say.
If you're sitting comfortably, I will begin. I'm going to assume that the r'bb is such that it doesn't matter for something with a 100-300R[?] load (condenser mic) but Deane specifically discusses the effect of the emitter resistance at the design current of 1.6-mA is 16 R, resulting in "a first stage single pole response of 83 KHz". And that, naturally, is where my brain takes its ball and goes home.
Firstly -- where does that pole even come from? There's a capacitor I'm missing here (base, perhaps?) I struggle with this stuff.
I (broadly) understand the use of the inductors in the emitter path thanks to the explanation in the paper but that r'bb won't be the same for my design. A 1mA (for the MAT12s) I figure that's 25-R, which isn't bad, but at 100-uA required to keep the jellybeans quiet that rockets to 250-R and that's really going to upset everything.
I rarely ask for help but this stuff has me stumped on the last leg of a design journey that's taken me a year of free time and a lot of cash. This board is the most expensive part and I need to try and get it right first time.
I've pretty much exhausted my foundational knowledge and something more meaty is required with equations my cat could understand because I'm very seriously weak at math. By the same token, I don't ask nor expect anyone to "do my homework" as it were, just help me with these pesky humps.
Thanks
I figured some of that. I'll have to go look back over the Miller effect and see what's going on there. I can check the input capacitance of those transistors, perhaps that what he used?
I'm using LTSpice for simulation - it allows me to see what my changes do, and if they follow what I expect, but I'm nervous relying on it for this since parasitic oscillations aren't something I've seen in any Spice package; and they've bitten me once before when I turned an amp into an oscillator! DERP.
The input diodes I got, but someone (Deane I assume) added some diodes (forward biased of course) between the current mirror - also not shown in this schematic - and the collector of Q1. There's a single diode shown here, CR3.
@MarcelvdG I'm reading that directly from Deane's paper, page 28, right-hand paragraph. I know OF base spreading resistance but it's something I've only discovered recently and I've never had to consider it a parameter.
@PB2 Sorry I'm still learning how to work this forum. I'm working with SMD and factory-supplied parts so I'm pretty limited in choice. They're a known quantity. The supply voltage can go lower as I'm only applying a small amount of gain so I don't need the swing Deane originally envisaged. It's Frankensteined by necessity, not choice I'm afraid.
This amp is primarily a dominant pole or Miller (Miller with pole zero) compensated amp.
Look it up, I'm sure that Bob covers it in his book but the dominant pole is formed by the
current available from the diff pair output to drive the Miller capacitance as multiplied by
the gain of the Vas stage.
You should keep the diff pair biased at the same current or scale the Miller cap (RC) to adjust
for changes. I would not make a 10X change in diff pair current, that's too much.
Look it up, I'm sure that Bob covers it in his book but the dominant pole is formed by the
current available from the diff pair output to drive the Miller capacitance as multiplied by
the gain of the Vas stage.
You should keep the diff pair biased at the same current or scale the Miller cap (RC) to adjust
for changes. I would not make a 10X change in diff pair current, that's too much.
Bob being? (Sorry)
Factory parts are from JLC. I don't have the facilities to solder SMD and cost is a serious option for through-hole parts.
When you mention the 10x change in current that's the for the alternative - 2mA for the MAT12 and 200 uA for a classic LTP variation for experimental (ultra-cheapo version) use.
Thank you for that explanation of the Miller effect going on - I'll go stick my nose in my recently (last week) copy of AoE 3rd Edition and look up what you've said! So thanks that's brilliant, really pushed me in the right direction!
Factory parts are from JLC. I don't have the facilities to solder SMD and cost is a serious option for through-hole parts.
When you mention the 10x change in current that's the for the alternative - 2mA for the MAT12 and 200 uA for a classic LTP variation for experimental (ultra-cheapo version) use.
Thank you for that explanation of the Miller effect going on - I'll go stick my nose in my recently (last week) copy of AoE 3rd Edition and look up what you've said! So thanks that's brilliant, really pushed me in the right direction!
Agh. See what you mean about the "not being able to edit" PB. You meant Bob Cordell of course, it's been weeks and weeks since I read the thread, I've been trying to get some other stuff working in the mean time - and I'm doing OK all things considered - but I had forgotten about Bob. I just scanned through again and it hit me right away, My bad.
The 2n390x series are listed in AoE as a good low-noise component (not claiming its the lowest at all!) but it's affordable and available.
That said, I'm happy to do a through hole design if I can find more suitable parts. I don't have to use LCSC/JLC parts but it is convenient to get a board back from the board house all built up.
The MAT12 (presumably the replacement to the MAT01) is available in TO-78-6 (IIRC) but I have so much to learn. I know about Miller and cascodes, etc. but I had no clue about calculating that value as part of the phase response. Make sense when I think of it though.
The problem with being largely self-taught is that I had a terrible tutor. (There's a rimshot on that joke but it might offend, so I've omitted it.)
The 2n390x series are listed in AoE as a good low-noise component (not claiming its the lowest at all!) but it's affordable and available.
That said, I'm happy to do a through hole design if I can find more suitable parts. I don't have to use LCSC/JLC parts but it is convenient to get a board back from the board house all built up.
The MAT12 (presumably the replacement to the MAT01) is available in TO-78-6 (IIRC) but I have so much to learn. I know about Miller and cascodes, etc. but I had no clue about calculating that value as part of the phase response. Make sense when I think of it though.
The problem with being largely self-taught is that I had a terrible tutor. (There's a rimshot on that joke but it might offend, so I've omitted it.)
I'm going from memory here, and thinking again most dominant pole designs have the first
pole below 100Hz, you gave a figure of 83 KHz so not sure what Jensen was talking about.
I've not read that paper carefully in over 40 years.
Also, while parasitics come into play at times, you want the performance of the design to be
dominated by actual components. Note that the Cdom cap (C1) is 150pF or roughly ten
times the internal capacitance of a small signal transistor.
Also, using the terms correctly, the Miller capacitance is internal to the device and the correct
term for C1 is the Cdom capacitor.
Fairly sure that Miller was they guy who discovered the effect probably in tubes at first.
pole below 100Hz, you gave a figure of 83 KHz so not sure what Jensen was talking about.
I've not read that paper carefully in over 40 years.
Also, while parasitics come into play at times, you want the performance of the design to be
dominated by actual components. Note that the Cdom cap (C1) is 150pF or roughly ten
times the internal capacitance of a small signal transistor.
Also, using the terms correctly, the Miller capacitance is internal to the device and the correct
term for C1 is the Cdom capacitor.
Fairly sure that Miller was they guy who discovered the effect probably in tubes at first.
If you are looking for a very cheap, reasonably well-matched and fairly low noise NPN transistor pair, I can recommend the Nexperia BCM56DS. The datasheet doesn't say anything about noise, but I used a dozen of them in a low-noise bandgap reference and was happily surprised by its noise performance.
Not meaning to cheapen the design - but when it comes to simulation (in LTSpice) the phase shifts remain pretty consistent - but with the corrective components it's all over the place. I know others have done simulations but I haven't actually tried them. Hell, I could have just adapted an existing Op Amp guts, or even a DC coupled AB power amp, I picked this because of the thought that went into it; because I specifically want to understand how everything works.
I could use a modern op amp, even an INA designed with mics in mind like the THAT 1512 but the whole point of this is a learning experience - hence why I want to know how to calculate and work around the imperfections forced upon me by virtue of a budget. The 2N390x line don't have particularly good hFE, especially when operated for low noise (100uA) where the gain just takes a nose dive. Although I could use Darlington to boost that (again, simulation is rather unhelpful). I'll have to try some of the other SPICE programs I have to see if the issue is with the models.
If I screw up the schematic capture, this is only part of a larger overall board, then won't have wasted money on a more expensive matched SMD device. This is all experimental for me - mostly breaking new ground in my own noggin. Stuff that's everyday bread and butter to an engineer is often well beyond me. That's why I came seeking help. I mean no offense.
@MarcelvdG Thanks. I did spec one of that family in an earlier iteration but IIRC the table in X-Chapters mentions them for low noise work but favours these jellybeans, the seem to be excellent performers, beaten hands down by some ZTX offerings. I've found both through hole and SMD matched pairs and some intrinsically matched ones for the mirror.
Sadly no one has been able to offer an explanation of what those extra diodes do in the B and C revisions. Damned if I can figure out why, but the will be a reason.
Sadly no one has been able to offer an explanation of what those extra diodes do in the B and C revisions. Damned if I can figure out why, but the will be a reason.
I guess the extra diodes are there to reduce the difference between the collector voltages of both transistors of the differential pair, in order to reduce systematic offset. (The single diode in the original 1980 circuit is simply part of the current mirror, used instead of a diode-connected transistor.)
Where does that 100 uA come from?
I've always been taught that to get minimal noise from a single bipolar transistor, you have to find one with low base resistance, low 1/f noise and no popcorn noise, and bias it such that the reciprocal of its transconductance re = |Zsource + rb'b|/√hFE. The corresponding collector current is kT/(qre). Whether that is 100 uA depends on the source impedance and current gain. This optimum is a balancing act between collector shot noise and base shot noise.
For a differential pair with one side driven from the source impedance and the other from a very small impedance (low-impedance feedback network), the optimal current per transistor is √2 times as high.
Transistors with strong hFE roll-off at low currents usually have bad 1/f noise.
Noise optima are broad, so when the current is off by a factor of two, it hardly makes any difference.
Last edited:
The local series feedback in the first stage with an inductor and a 30 ohm resistor per side causes this. The zero is determined by the 30 ohm resistor and its parallel inductor, the pole by the inductor and the parallel combination of the 30 ohm resistor and the re ~= 1/gm of the transistor.
Ah, I've been given an approximation to work on which is re ~= 2R5 at 10-mA and increases roughly tenfold with each order of magnitude reduction in collector current - so 25-R at 1mA, 250-R at 100-uA and so on.
But a better approximation is to use the reciprocal of the transconductance? Works for me... But this is only way we can learn and I appreciate everything people contribute to help this doddering old fool get by! Someone else might read this at a later date and not have to ask the same question.
An annoyance with modern search engines hit me again to day. I asked the BING (which is ChatGPT) AI where to find that schematic and it opened by telling me the 990 was designed by ... John Hardy. I screamed back at it (in text) and it apologised and corrected itself, but it remains to be seen if this "sticks". The fact people are relying on LLMs for actual scientific papers and more now is terrifying.
Also @mods sorry I didn't see the edit button until just now.
EDIT: @MarcelvdG another thank you for pointing out what my brain did a "WTF?" at now I look at it, CR3 is just forming the other part of the mirror as you said. I've "diode wired" transistors before but never considered using a diode as part of a mirror. The connection between the two should have been a clue I guess, but no... right over my head. And then you mention it, in passing almost, in a different context and now I can't "unsee" it.
Brains are amazing.
Last edited: