New Discrete OpAmps

[Samuel Groner]

Hi

Been a while since we've seen new elaborated discrete opamps; about time to present some recent work from my side.

First we have a successor to the SGA-SOA-1: SGA-SOA-2_r1.pdf

The changes are subtle, but worth a more detailed discussion. First we now have inclusive Miller compensation, i.e. the output stage is included in the local loop of the compensation capacitor C2. This frees the second stage (Q4) from providing the current to charge C2; benefit is much reduced distortion at high frequencies (100 kHz THD+N dropped ten times) and lower output impedance. The former is also a result from somwhat higher slew-rate. The input stage degeneration resistors R1/R2 are increased in value which in turn allows reduction of C2. R4 is introduced to balance the collector voltages of Q1 and Q2 which somewhat improves CMRR; same applies for the use of a transistor (Q3) with higher hFE for the tail current source and the use of a single inductor in the input stage. To reduce the short-circuit output current the output stage emitter resistors are increased (R10, R11); to get the quiescent current back to 15 mA R9 is added.

All the improvements have been achieved at unchanged parts count and quiescent current. The resulting typical specifications for 600 Ohm load and ±18 V supplies are as follows:

Input offset voltage: < 20 mV
Input bias current: -1.1 uA
Input voltage noise: 1.5 nV/sqrt(Hz) at 1 kHz
Current noise density: 0.6 pA/sqrt(Hz) at 1 kHz
Open loop voltage gain: 80 dB
Gain-bandwidth product: 40 MHz at 10 kHz
Unity-gain bandwidth: 15 MHz
Slew rate: ±14 V/us
Output swing: ±17 V
Class A output drive: 600 Ohm for full voltage swing
Maximum output current: ±230 mA
Supply current: 21 mA (no load)

R6 should be adjusted if different supply voltages are used. Suggested values are 18 kOhm (±15 V), 15 kOhm (±12 V) and 12 kOhm (±10 V).

Second an opamp with very low voltage noise: SGA-LNA-1_r1.pdf

A fully complementary topology is used; this effectively reduces voltage noise by 3 dB and cancels input bias currents. Resulting voltage noise is low enough for a MC preamp. The design is stable at noise gains of about 3, and no adaptations for lower supply rails are needed. The typical specifications for 600 Ohm load and ±24 V supplies are as follows:

Input offset voltage: < 10 mV
Input bias current: ±100 nA
Input voltage noise: 0.5 nV/sqrt(Hz) at 1 kHz
Current noise density: 1.7 pA/sqrt(Hz) at 1 kHz
Open loop voltage gain: 80 dB
Gain-bandwidth product: 145 MHz at 100 kHz
Slew rate: ±48 V/us
Output swing: ±22 V
Class A output drive: 1.4 kOhm for full voltage swing
Maximum output current: ±50 mA
Supply current: 25 mA (no load)

Third a general-purpose opamp designed to operate at high voltages up to ±40 V: SGA-HVA-1_r1.pdf

The design uses the generic two-stage topology which is well know from D. Self's work and the JE-990 opamp. There are a few unusal details though which might be interesting; first the base-current errors from the input stage collector load mirror (Q3 and Q4) are first-order cancelled by operating Q6 at twice the collector current of Q3/Q4; the voltage divider formed by R6 and R7 allows the use of transistor types with low breakdown voltage but high hFE for Q3-Q5. D3 is used to operate Q1 and Q2 at similar collector voltage (this is maintained over temperature) for improved CMRR. C1 provides a high-frequency feedforward path around Q5. This prevents gain peaking at some hundred MHz and keeps the inclusive Miller loop formed by C3 stable. R6 should be altered for lower supply voltages to maintain the 1.6 mA bias through D3/D4. Suggested values are 33k (±30 V), 27k (±24 V), 20k (±18 V), 16k (±15 V), 12k (±12 V) and 10k (±10 V).

I'm pretty pleased with the performance of this amplifier; the input offset measured at the prototype was just half a mV, indicating that second order effects from the current mirror and second stage are well addressed. Distortion is very low at least in inverting mode, rivaling that of the very best IC opamps. And the quiescent current is low for a discrete implementation. The typical specifications for 600 Ohm load and ±40 V supplies are as follows:

Input offset voltage: < 5 mV
Input bias current: -1.1 uA
Input voltage noise: 1.5 nV/sqrt(Hz) at 1 kHz
Current noise density: 0.6 pA/sqrt(Hz) at 1 kHz
Open loop voltage gain: 130 dB
Gain-bandwidth product: 65 MHz at 10 kHz
Unity-gain bandwidth: 10 MHz
Slew rate: ±21.7 V/us
Output swing: +37.3/-37.5 V
Class A output drive: 1.9 kOhm for full voltage swing
Maximum output current: +85/-95 mA
Supply current: 18 mA (no load)

I have little intention to design suitable 2520-style PCBs, but if someone want's to go ahead feel free to do so. At least the first two designs should not present serious problems as their parts count is relatively low.

Samuel

 

[bcarso]

Have you found the 3329 and 1316 parts to be readily available?

 

[JohnRoberts]

Nice...  I like the symmetry of #2, but would caution the parallel input devices cuts both ways... you decrease noise voltage while increasing noise current, but as long as you keep resistor impedances around it low, it should be fine.

Selecting the input devices is obviously to keep input bias current low. If you don't select, does distortion degrade, or just DC characteristics?

I wasn't familiar with the 3329, glad to see MC level impedance parts available, since ROHM dropped the 737/786 several years ago. The 3329/1316 look like a better match to each other than the old 737/786. 

JR

 

[Samuel Groner]

Have you found the 3329 and 1316 parts to be readily available?

No, I use parts bought some years ago.

You decrease noise voltage while increasing noise current.

Sure, but the 1.7 pA/sqrt(Hz) is still pretty low. Many ICs have such or higher current noise densities for higher (and much higher) voltage noise figures.

If you don't select, does distortion degrade, or just DC characteristics?

The hFE match on the input devices is unlikely to have significant effect on distortion. The tail current match might however influence the CMRR/common-mode distortion. Unlike conventional input stages this design shows common-mode distortion with almost only odd order harmonics.

Samuel

 

[Ptownkid]

Well then, I'm gonna get right on this with regards to making some board layouts. If they sound anything like the current SGA-soa's, we're all in for a treat.

Awesome Sam, way to go.

 

[JohnRoberts]

You decrease noise voltage while increasing noise current.

Sure, but the 1.7 pA/sqrt(Hz) is still pretty low. Many ICs have such or higher current noise densities for higher (and much higher) voltage noise figures.


Samuel

I didn't do a survey of modern opamps and that noise voltage is indeed much better than any IC I've ever seen, but even the old 5534 is on the order of .6 pA/ rt Hz @ 1kHz. In the grand scheme of things this (noise current) is not a big deal for low Z sources (like mics or MC carts) since ein is the combination of all noise sources, and factored by impedance, but just a FWIW.

For GP applications what do you think of something like the THAT 340 array http://www.thatcorp.com/datashts/300data.pdf ? It might work for a drop in without selecting parts. Higher noise voltage (1 nV) but still respectable for all but very high gain apps. I didn't see a spec for noise current on the THAT parts. The nominal beta of the THAT devices are lower than your MC parts, and mismatch nominally 100x/75x so there would be some uncanceled input bias current, but probably less than your other DOA (and most opamps- just speculating). 

I just don't like selecting devices. In part from 15 years working with volume production and from memories of an old phono preamp kit I sold where I supplied matched pairs of selected JFETs...  never again...

JR

 

[bcarso]

The lower beta will of course entail higher current noise.  I spoke to them once about that lower beta, and they said that when they had done some attempts at higher beta, the breakdown voltage variations they saw were prohibitive. 

Easy for me to say, not having to make and sell such things, but I asked how Toshiba managed to do it.

Their parts do have very good log conformance, and in some of the applications where that is crucial, like VCAs, you can design around the lower beta.

 

[bigugly]

Thanks for the new op-amps Samuel. What is the purpose of C2 in the "SGA-SOA-2 r1" schematic? Here is a preliminary pcb layout for the SGA-SOA 2.

 

[Ptownkid]

That's funny, my layout looks almost the same...weird. ;D

 

[bigugly]

well, ya know they say that great minds think alike.  ;D

 

[Sredna]

Are you both thinking of using these long pins?
And where do you get hold of them?

Sorry for OT.

Regards,

Anders

 

[Ptownkid]

No, I will be making kits with the standard millmax pins that i use in my other doa kits

 

[bigugly]

Are you both thinking of using these long pins?
And where do you get hold of them?

Sorry for OT.

Regards,

Anders

the long pins are needed in my design because the components mount to the underside of the PCB. Millmax has these longer pins also.

 

[Samuel Groner]

There was a minor error in the SGA-SOA-1 schematic, the second BC560C should read Q5 not Q4. Corrected now, thanks for pointing out Bryan.

I didn't do a survey of modern opamps.

The closest parts I know of are the LT1028/LT1128 (0.9 nV/sqrt(Hz) and 1 pA/sqrt(Hz)). AD797 and ADA4898-1 have the same voltage noise specification but higher current noise (2 pa/sqrt(Hz) and 2.4 pA/sqrt(Hz)). However, these ICs might have additional unspecified common-mode current noise from input bias cancellation.

For GP applications what do you think of something like the THAT 340 array. It might work for a drop in without selecting parts.

It would surely work, at least at lower supply voltages. IMO the topology is not particularly well suited for an GP opamp though; the CMRR of that complementary input stage just isn't that great. And increasing the compensation for unity gain stability will probably require the use of those RL-networks at the emitters of Q1-Q4, which makes at least six more parts which is inconvenient.

I didn't see a spec for noise current on the THAT parts.

With the nominal hFE being six times lower than for the Toshiba parts I'd expect an increase of sqrt(6) = 2.45.

What is the purpose of C2 in the "SGA-SOA-2 r1" schematic?

That's just the miller compensation capacitor.

Samuel

 

[JohnRoberts]

The lower beta will of course entail higher current noise.  I spoke to them once about that lower beta, and they said that when they had done some attempts at higher beta, the breakdown voltage variations they saw were prohibitive. 

Easy for me to say, not having to make and sell such things, but I asked how Toshiba managed to do it.

Their parts do have very good log conformance, and in some of the applications where that is crucial, like VCAs, you can design around the lower beta.

I noticed the lower beta, so all things equal worse noise current term, but wouldn't they also be dialed in at a lower current density somewhat mitigating THAT?

Your design looks great for a conventional (virtual ground inverting) console summing bus amp with a lot of inputs, where the feedback network impedance would be low due to all the inputs in parallel, and stability constraints would also be met easily.

Sorry, I don't mean to hijack your thread... nice work...  Just a personal dislike for selecting parts, acquired from experience doing it. The low noise parts are generally graded for beta so performance might not be horrible w/o selection. 

JR

 

[bcarso]

You know what I don't know is the bulk emitter resistances of the Toshiba parts.  That would be something to determine and take into account.

The rbb of the THAT parts is going to be a significant limitation compared to the Toshibas, once a design runs enough current to make the half-thermal noise of of the equivalent emitter resistance negligible---and provided that the external impedances are made low enough to make the current noise contribution small.  But since they don't seem to be made anymore  :'(  it may be a moot point for any production quantities, until/unless some outfit like LIS figures out how to make them and decides to do so.

I have bags of the Toshibas but they are too good to use 

BTW a bit OT but Dennis Colin has measured more of the LIS LSK389 dual JFETs (see the Feb aXp) and finds them to be within spec now for noise.  This is a relief---I really want LIS to succeed.

 

[JohnRoberts]

Nice part (389). I used a single 1 nV JFET in an old phono preamp a couple decades ago and that was sweet.  Back in the '70s I experimented with 2n3957 duals, in an attempt to use one half inside a DC feedback loop, to linearize the other half to make a reliable voltage controlled variable resistance. I eventually bailed and just used a VCA for high end, and OTA for low end apps. With well matched 1nV parts it might make sense to revisit (if I was still messing with such things). 

I didn't grok that the Toshiba parts were obsolete.. one data sheet I found was < 10 years old.. I guess there aren't many new MC preamps coming out these days.

I still have a modest stash of 2sb7373s.. but designing with them is limited in application.

JR

 

[bcarso]

Based on my experience, the low noise short-channel JFETs like the 2SK170 and 2SK389/LSK389 are basically lousy for VCR applications, no matter now ingeniously (or InGenius-ly  ;D) deployed.

I even tried to use them in a quarter-square multiplier, only to find that their conformance to a square law was iffy at best.

 

[Ptownkid]

Hey Samuel, why the switch from 2n4403 to bc560C in the SOA?

 

[Samuel Groner]

Why the switch from 2N4403 to BC560C in the SOA?

As implied in the first post their higher hFE are benefficial for CCS output impedance, which in turn gives better CMRR.

Samuel