Cool stuff, but very difficult to assemble 0402 devices, especially between pins. The smallest I'm willing to go for actives are those 6-pin dual-devices from Rohm and so on. I'll revisit this again in due course - I've been distracted for the last year or so with chipamps and full-range/2-way bookshelf speaker designs.
The (simulation) schematic of the LF08 is at post #3291 on this very thread:
http://www.diyaudio.com/forums/analog-line-level/218373-discrete-opamp-design-330.html#post5094617
It's conceptually derived from the venerable Melcor 1731, but with modifications including a current-mirror load for the LTP, a cascoded VAS, and altered compensation. It might have a completely different sonic signature compared to the Melcor - no way to tell without building one and doing an A-B. It appears to work in simulation, at least. I need to dig out the layout and refine it before sending it out for fabrication (eventually).
Minor update: LTSpice simulation reveals that XRK's idea of adding an additional pair of output NPN/PNP for driving low-Z headphone loads is a good one - there's some THD20 reduction on the higher harmonics, as well as better current-handling characteristics. You can also add a 1uF X7R or similar to bypass the base-to-base diode pair between the drivers, i.e. from the base of the NPN outputs to the base of the PNP outputs. It works remarkably well in simulation for clobbering the higher-order harmonics, when driving standard line-level loads, as well as lower-Z loads.
(I also have a N-JFET LTP version in simulation - seems OK, but needs higher tail currents and has higher THD20, but with desirable H2-dominant harmonics and higher unity-gain BW. It will also need careful hand-matching of N-JFETs to minimize input offsets. We'll see how it goes, but I would recommend staying with the all-BJT version if anybody wants to do an initial layout and build. There are also a lot of choices of BJT duals for SMD layouts and so on.)
(I also have a N-JFET LTP version in simulation - seems OK, but needs higher tail currents and has higher THD20, but with desirable H2-dominant harmonics and higher unity-gain BW. It will also need careful hand-matching of N-JFETs to minimize input offsets. We'll see how it goes, but I would recommend staying with the all-BJT version if anybody wants to do an initial layout and build. There are also a lot of choices of BJT duals for SMD layouts and so on.)
If you don't mind performing a 2-variable optimization, I think you may get even further improvement if you add resistor "R8" in series with base-to-base capacitor "C4". I expect you'll see some improvement in the frequency domain and even greater improvement in the time domain. Figure below.
At least in my experiments, the optimum value of C4 when R8 is present, is very different than the optimum value of C4 when R8 is absent (shorted). So it's a full 2 variable optimization, not a sequential pair of 1 variable optimizations.
_
Attachments
Cordwood design. Brings back memories of CDC 6600 and 7600 supercomputers. Not trivial to build, impressive to look at and impossible to repair.
I have seen some in ear Bluetooth headphone electronics done as folding flex PCBs, similar but more manageable. About the same footprint.
Looking forward to reports on the performance.
I have seen some in ear Bluetooth headphone electronics done as folding flex PCBs, similar but more manageable. About the same footprint.
Looking forward to reports on the performance.
Actually, I already did a first-pass layout of a dual LF08 in DIP8 with cordwood-style layout, with one board for each channel, but with *no* connections between the two boards except via an edge connector which mounts on a base board with the pins that plug into the DIP8 socket. There are two structural struts that keep the two opamp boards in place. That's now more or less standard in many of the commercial discrete dual-opamps on the market like the Burson HD, Earth, Moon, etc. The advantage is that it uses an existing base board which I already have for another discrete opamp, the LF03, and each opamp channel uses an identical board, so rework is fairly simple if needed - a defective channel can be swapped out with a good one with about a dozen desoldering/soldering operations (maybe 5-10 minutes work).
I just have to add a few things (optional 2 pairs of BJT outputs for headphone amp applications and the base-to-base bypass cap) and it's probably good to go, but I'll just fiddle around in simulation for a while to see if I've missed something important.
Edit: There's a series of NXP TO92 pairs that may work well at the output, with or without the need for a second pair - PBSS4350/5350, etc.:
PBSS4350 Nexperia USA Inc. | Transistors - Bipolar (BJT) - Single
However, BC639/640 are ubiquitous and inexpensive, so that would be the first choice at any rate.
I just have to add a few things (optional 2 pairs of BJT outputs for headphone amp applications and the base-to-base bypass cap) and it's probably good to go, but I'll just fiddle around in simulation for a while to see if I've missed something important.
Edit: There's a series of NXP TO92 pairs that may work well at the output, with or without the need for a second pair - PBSS4350/5350, etc.:
PBSS4350 Nexperia USA Inc. | Transistors - Bipolar (BJT) - Single
However, BC639/640 are ubiquitous and inexpensive, so that would be the first choice at any rate.
Last edited:
One valid reason for a discrete design rather than using an op-amp is that it can be designed for wider power supply range. Unless their exists an op-amp I don't know about, there are none capable of greater than +/-18v power. There are a couple of reasons to pursue this. One is greater dynamic range, two, it could be used as the front end to drive a class A/B amp output stage directly for reduced component count. A powered PA speaker with a single mic input comes to mind.
I have such a design but but rather than offer it here, I'd like to see what a consortium comes up with. Perhaps its' performance will beat mine.
One thing I learned - use differential topology until the output stage whereby class A will be necessary. You will get incredible common mode power supply rejection to the extent that no power supply regulation will be required. The nice thing about PS ripple is it is very symmetrical so CMRR kills it completely.
I have such a design but but rather than offer it here, I'd like to see what a consortium comes up with. Perhaps its' performance will beat mine.
One thing I learned - use differential topology until the output stage whereby class A will be necessary. You will get incredible common mode power supply rejection to the extent that no power supply regulation will be required. The nice thing about PS ripple is it is very symmetrical so CMRR kills it completely.
Cool - it seems a good idea to place the small-signal and output stages on separate boards and stack them, cordwood-style, as you've done. For higher-current/high-dissipation applications, including low-Z headphone amps and small speakers, it might be prudent to have an output stage with through-hole TO92/92L output pairs. Here's a first-pass layout of such an output stage, with two pairs of TO92 outputs as suggested by xrk971. The dimensions are 0.65" x 0.85" (~16.5mm x 21.6mm) - maybe a bit large, but it allows for easier assembly of the SMD components (all 0805 or larger). If you're really good at it, it can be done as a toner-transfer PCB with (H202 + HCL) etchant.
R1..R4: 8.2E (0204 MELF should be usable)
D1, D2: BAV99 or equivalent.
Q1, Q3: BC639 (lots of other possibilities for high Ic/low Vceo or low Ic/high Vceo)
Q2, Q4: BC640 (")
C1: 1uF/16v X7R or similar (0805 or 0603). Some SMD Tantalums up to 10uF may also fit.
Edit: It uses pins 5 and 8 as inputs from the VAS/driver stage. That may cause various incompatibilities
in substitutions for legacy DIP8 single opamps which use one or both of these pins for Offset null,
Overcompensation and so on. For such substitutions, just cut pins 5 and 8 on the assembled module
- i.e. ensure that pins 5 and 8 don't make contact with the legacy socket.
Attachments
Last edited:
There are several monolithic opamps with +/- 22V maximum rails, including LME49860 (probably defunct now), NE5532 and so on.
The discrete designs on this thread can probably be tweaked for +/- 25V with minor difficulty. If you want to go higher than that, you'll have to give up JFETs (often used in the input LTP, cascode or CCS to save parts count) and stick with BJTs. For a serious high-voltage discrete opamp module capable of +/- 50V, you'll probably have to go with low-current biased BJTs, maybe all through-hole, for the the small-signal stages, and omit the output stage - i.e. make the output stage entirely discrete and external. This is feasible with the LF08 schematic, if the VAS/cascode PNPs and the driver NPNs are spec'ed for say 120 to 150V Vceo and the input LTP NPNs for say 100V. There are some Japanese small-signal low-noise BJTs that may make the cut - A1845/A992, C2240/A970 and their SMD equivalents. The drivers can be TO126 or similar devices like C669, C2953 or similar video/chroma transistors from the CRT era - they're biased to just ~2 mA in the LF08, so they're really just pre-drivers. You'll have to use a conventional external driver stage, maybe a discrete darlington or diamond-buffer or similar.
In short: yes, it's feasible to make a +/- 50V discrete BJT module with the LF08 by separating the small-signal and the output stages. +/- 60V would be really pushing it, but just barely feasible. There was a monolithic 20-pin DIP, the TDA7250, which fulfilled most of this functionality up to +/- 40V, but which has been defunct for over a decade.
Last edited:
Here is an SOIC opamp with up to 140V (+/-70V) just to show that IC's can swing a lot: http://www.analog.com/media/en/technical-documentation/data-sheets/6090fe.pdf I don't think it has the slew rate or full power bandwidth to make a good amplifier but there are others. There is a monolithic power amp that has similar output swing from Apex: Apex Microtechnology - Power Operational Amplifers - PWM Amplifiers - Voltage References up to 2500V.
You can use JFet inputs with high voltage, Cascode them. I use depletion mode Mosfets for cascoding. It makes it very simple and improves the performance of the input a lot. You can get 350V depletion mode N channel Mosfets. Power dissipation will be an issue at high voltages. I like to run the Jfets at 5 mA so 5 mA X300V would be 1.5W in the cascode device. The parts count becomes an issue for this project but still possible.
You can use JFet inputs with high voltage, Cascode them. I use depletion mode Mosfets for cascoding. It makes it very simple and improves the performance of the input a lot. You can get 350V depletion mode N channel Mosfets. Power dissipation will be an issue at high voltages. I like to run the Jfets at 5 mA so 5 mA X300V would be 1.5W in the cascode device. The parts count becomes an issue for this project but still possible.
Here's a single-sided SMD-only layout of the small-signal portions of the LF08, including the VAS and pre-drivers. It's in the same form-factor (0.85" x 0.65") of the EF08 output stage I posted earlier, and the two can be stacked to form a complete single opamp module.
Everything is SOT23 and 0603 (or 0102 micro-melf) - it could be made smaller by using SOT323 and double-sided SMD, but it would be difficult to assemble. This size is about the right trade-off between compactness and ease of assembly. With the right choice of devices, it can probably run with +/- 60V rails - in which case Q10 can be omitted and the LTP tail current source is purely resistive (R1 + R2).
Of course, the EF08 output stage can be omitted entirely, and the pre-driver outputs from pins 5 & 8 can be used to drive an external Vbe multiplier and Darlington power output stage. Numerous options are possible including CFP, Diamond buffer, Quasi Complementary, etc. with an external power output stage. The pre-drivers can sink/source only ~2 mA, so a 2-stage external power output stage is pretty much mandatory except maybe for low-rail, low-power applications like headphone amps.
Everything is SOT23 and 0603 (or 0102 micro-melf) - it could be made smaller by using SOT323 and double-sided SMD, but it would be difficult to assemble. This size is about the right trade-off between compactness and ease of assembly. With the right choice of devices, it can probably run with +/- 60V rails - in which case Q10 can be omitted and the LTP tail current source is purely resistive (R1 + R2).
Of course, the EF08 output stage can be omitted entirely, and the pre-driver outputs from pins 5 & 8 can be used to drive an external Vbe multiplier and Darlington power output stage. Numerous options are possible including CFP, Diamond buffer, Quasi Complementary, etc. with an external power output stage. The pre-drivers can sink/source only ~2 mA, so a 2-stage external power output stage is pretty much mandatory except maybe for low-rail, low-power applications like headphone amps.
Attachments
Last edited: