The Cabito: a classic revisited

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Here's a little something I just put together, a classic opamp + discrete buffer with a couple of twists. It simulates very well and sounds fantastic!

The schematic and sim file attached is for the usual +/-15V supplies although I used +/-12V because of what I had in stock. Only one resistor to change for different supply voltages, as shown in the schematic. Attached also a .txt file with the models I've used for the BC183 (a BC182 model I found somewhere and tweaked) and LM4562 (the old one from National, good times!). The BC139/140, 1N4148 and red LED are from Bob Cordell (http://www.cordellaudio.com/book/Cordell-Models.txt). I used the BC183 because I have a bunch and it works well here, but you can use any of the other usual suspects (BC54X, KSC1845, 2N508X) maybe with some minor tweaks. Simulate, build, measure and see.

Simulated THD (real measurements will follow at some point) with a 120R load (the impedance of my AKG K601s, not too high, not too low):

1mW@1kHz/10kHz (Vpeak=0.49V): 0.000001% / 0.000002%
15mW@1kHz/10kHz (Vpeak=1.9V): 0.000005% / 0.000028%
200mW@1kHz/10kHz (Vpeak=6.9V): 0.000018% / 0.000122%

The idea was to make it small but with good performance, in particular:

- A CCS load instead of the single resistor or bootstrapped load commonly used. The simulated improvement from single R to bootstrapped is significant; not so much from bootstrapped to CCS, although there is some and I'm happy to trade a large cap for a couple of small active components. And LED CCS because it looks cool :)

- No heatsinks on the output transistors while having a (close to) optimal Iq Oliver-wise, something that doesn't seem to be of much concern on most designs of this type I've seen.

- To achieve the above with a simple two-diode bias spreader. I was determined to squeeze it into that 6x4cm proto board, so component count counts.

I knew without heatsinks and with the fixed bias spreader temperature would matter, so I simulated with the output transistors at different temperatures (which I guesstimated for different Iq's) and ended up with Iq=11.7mA for 2R2 emitter resistors and a 470R in the CCS. I must have underestimated the temperature a bit because after stabilizing I measure an Iq of 12.5mA in one channel and 15mA in the other. Simulating at 48deg with the 470R CCS resistor I used gives Iq=13.8mA, around the average of what I measure, and increasing it to 560R at that temperature brings it back to the required ~12mA.

In any case the difference isn't huge and I didn't want to get into the wiring mess at the back of the PCB again, so I left the original 470R, but if you build it I suggest to start with 560R and tweak if necessary. If you want more precision you could use a 1k trimmer there and also match the output transistors of both channels for Vbe, which I didn't and must be the cause of the channel difference, all other voltages and currents are pretty well matched.

I've thrown everything I have at it (Senn HD600, Beyer DT880, AKG K601, Shure SRH940) and it drives them all with aplomb, as effortlessly and transparently as my Mytek Stereo DAC 96. In fact right now I'd have to give the edge to The Cabito, it sounds slightly "cleaner" to me, but it's my baby, I just made it and I've seen the impressive numbers, so expectation bias is almost guaranteed and I'd bet good money that I wouldn't be able to tell them apart in a blind test.

More to follow.

Cheers,

Cabirio
 

Attachments

  • The Cabito 1.JPG
    The Cabito 1.JPG
    148.9 KB · Views: 255
  • The Cabito 2.jpg
    The Cabito 2.jpg
    301.8 KB · Views: 252
  • The Cabito Sch.png
    The Cabito Sch.png
    94.6 KB · Views: 272
  • TheCabito.asc
    4.5 KB · Views: 32
  • thecabito.txt
    3.3 KB · Views: 38
Last edited:
Some more design notes:

- Stability: I must say I'm pleasantly surprised. The LM4562 is known to be quirky, but with a 100n cap between rails and one from V- to ground (as recommended elsewhere in this forum), both directly on the pins at the back of the PCB, and that very light 4p7 compensation cap between opamp output and In-, there are no signs of oscillation. It runs at its normal temperature (slightly hot but not unduly so given the idle dissipation), DC offset is stable and there's no harshness, excessive detail or anything of the sort. Simulation says it's fine (~76deg phase margin), but this is obviously layout-dependent, so if you build it, you may need to increase the 4p7 cap and/or move it across the feedback resistor. Try and see.

- Many component choices were based on what I had in stock, like the BC183 and +/-12V supply voltage instead of +/-15V. BTW the power supply is your bog standard LM3X7 affair. Fixed regs would work fine of course, but I didn't have any 7912s in stock.

- I'm not sure if pulling about 2mA from the opamp output to V- is enough to do the Class A bias trick, I haven't looked into this in detail. Anyone?

- The 100k input resistor is in case I want to put a pot before it later (for now I haven't), so the taper won't be affected much. Low value resistors around the opamp to minimize noise and because the LM4562's very low input bias currents let you get away with it while keeping a low DC offset (1mV simulated, 0.5/2.4mV measured) so you don't need a bulky DC-blocking output cap.

- Gain of 4 because it works well for me connected to my computer's line output.

- 10R output resistor for some measure of short-circuit protection without losing too much voltage swing.

- The 100u in the ground leg of the feedback sets the high-pass cutoff frequency, so I went with Nichicon Muse ES, known for their pretty much unmeasurable distortion.

- Wima 2u2 film caps at the input and across the bias spreader. The 2n2 input caps are soldered directly at the RCA input connectors.

- Whatever standard red LEDs I had in stock and the BD139/140 are Fairchild, hfe group 16 (100-250).

Comments welcome!

Cheers,

Cabirio
 
These should help if you want to build it on a similar proto board, and also if you want to make a PCB for it. I show separately the signal traces (what I call "bottom" layer) and the supply and ground ones ("top").

Grounding: solder the shield of the input cable to the wire joining both signal grounds (SGND) and then take a wire from there to the star ground at the power supply PCB, another wire from between the 100u bypass caps on the right and another one from the output jack (or from the shield of the output cable if you use a shielded one, as I did).

If you use RCA input connectors with separate grounds, you can keep them separate, connect each to its corresponding SGND and run two wires from each of those to star ground. Probably overkill but if you can, why not.

If you make a PCB, I suggest providing for compensation caps both from opamp output to In- and across the feedback resistor, as well as bypass caps from both V+ and V- to ground at the opamp, just in case. Maybe also for a trimmer where the 560R goes.

Note that the 2k7 resistors would be 3k6 with +/-15V supplies.

Cheers,

Cabirio
 

Attachments

  • Cabito PCB bottom.png
    Cabito PCB bottom.png
    33.9 KB · Views: 264
  • Cabito PCB top.png
    Cabito PCB top.png
    32.1 KB · Views: 224
Quick update: just for fun and in case someone would like to use the ever popular OPA2134 here, I've given it a try.

Leaving everything as it is, it oscillates: pretty obvious in the hard-to-explain-how-but-clearly-wrong treble, plus it runs noticeably hotter than the LM4562, which it shouldn't. Going with the datasheet-recommended bypassing for the OPA2134, i.e. 10n caps from each rail to ground, doesn't seem to help.

Some simulation and trial and error later, I've found that it's perfectly stable with those 10n bypass caps and a 22p cap across the 1k feedback resistor. Again, this is with my particular layout, so ymmv. Since working on that proto board is such a royal PITA, I haven't really gone back and forth between the two to compare them, but from what I've heard, it sounds just as clean and poised. A good alternative.

Cheers,

Cabirio
 
Nice project!
It doesn't surprise me that you needed to change the feedback cap. I hypothesize the LM4562 rolls off earlier on its' own vs the OPA2134. (Admitting I haven't look at the spec sheets).
There's a rule of thumb:
At the point in the upper frequency domain that an amplifier exhibits more than 40 degrees of phase shift the gain must be 1 or less. This has never let me down. When you think about it - when negative feedback starts to become positive as a result of phase shift you create an oscillator.
I wonder how many designs here have been benched when something as simple as a frequency dependent network hasn't been tried?
 
OPAx134 input capacitance might be more of an issue here, it is a FET input part after all. Still, I would've expected upping the 4p7 to fix the issue.

It's 2018 and people are still putting up with 10 ohms of output impedance? That's not entirely low enough for 'phones as common as the 500-series Sennheisers! (Those can do with as low as 5 ohms, and don't get me started on BA in-ears.) Sure, you may need a Zobel network, but what can you do?

No volume pot? This may end up being a bit on the noisy side, depending on your output. If you've got 10 µV coming out of there (-106 dB from 2 Vrms), those would be amplified to 40 µV all the time - adequate for typical 250-600 ohm full-size 'phones or isodynamics but possibly not dead silent in more sensitive ones.

Honestly, instead of the 2µ2 parallel film, I'd rather use whatever value 10 V electrolytic you can fit there instead. 100µ might do the job. Whatever ESR that has is going to be plenty low enough, and the effect would extend to much lower frequencies.

I'd rather have the diodes in the current source and the LED in bias rather than vice versa. This would allow running the output stage much hotter (adjust emitter resistors as needed, maybe 10R for 30-odd mA, and don't forget some heatsinking). Might help with stability, too.
 
Thanks for the comments guys!

campsquire: if you look at the open loop gain curves in the datasheets, in fact the 2134 rolls off earlier (the simulation models reflect this too), although both opamps have roughly the same phase lag at 0dB, so I suspect it's something else, like physical layout, interaction with the output stage, input capacitance as sgrossklass says or a combination thereof, so my advice to anyone building this is to make sure the supply pins are bypassed properly and experiment from there.

sgrossklass: many thanks for your detailed analysis! I agree that some headphones may not like 10R output resistance, so it may be reduced or replaced with a zobel, but I looked at e.g. the Senn HD598 (I also have them but haven't tried them, they're in a different country, long story) and it looks to me like the effect is negligible (see here).

I don't have any noise issues even with the low-Z, high sensitivity Shure SRH940, but as I said a 10k-22k vol. pot can be used no problem if someone needs it (I've actually included it in the schematic and sim file).

I don't know how accurate simulation is re. the bypass cap on the bias spreader, but I found that there's a trade-off between distortion at 10kHz and 1kHz and 2u2 seems to be a good compromise. When I do real measurements I'll try and see, it's true that most amps of this style I've seen use bigger caps there, so you're probably right that 100u could be better.

It's funny because in my quest to use as few components as possible I did start off the design with a red LED as the bias spreader, but I couldn't find an optimal combination of emitter resistors and idle current (BTW I think you mean 1R rather than 10R for ~30mA, right?) Also I went with 12mA because I wanted to make it small and not use heatsinks, but it looks like 1R resistors and using a 330R in the CCS would give around ~27mA, so that would be a good option if you have the space for heatsinks.

xrk971: Rod Elliott's is one of the ones I investigated, although the closest would be Douglas Self's from his Small Signal Audio Design book (fig. 20.3 in the 2nd edition, 15.7(a) in the 1st), which is basically The Cabito with a single resistor load rather than a CCS and TL072/MPSAx2 instead of LM4562/BD1xx. He also uses an output coupling cap and a much bigger 47R output resistor to compensate for the different voltage requirements of low and high impedance headphones.

That Kirkwood looks great, I have some DRV134s in stock and I see in this thread that it can be adapted... Did I say great? It looks spectacular! That FFT! :eek: Might give it a try sometime in the future...

Cheers,

Cabirio
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.