Diamond Cascode, a Diamond-like buffer/current booster

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I've mentioned (and posted schematics) of what appears to be a simple and novel diamond-like buffer in other threads (Discrete Opamp Open Design, etc.). I've now built an opamp current-booster using this output stage, in the DIP-8 form factor. This thread is for the purpose of collecting the info about this topology and implementation(s) in one location, for easy reference.

1. LTSpice simulation schematic with LME49860, with and without the diamond cascode current booster.
2. THD20 FFT at 4v output amplitude into 600R load, with and without the current booster.
3. Eagle schematic.
4. Eagle layout.
5. Component-side picture of prototype.
6. Solder-side picture of prototype (with SOIC-8 LM318M installed).

Implementation notes:

1. The most difficult part of the assembly is the SOT363 BC847BPN dual NPN/PNP. However, it's manageable with a toothpick, standard soldering iron (could also be used as a diagnostic test for Parkinson's), and some no-clean liquid flux.
2. All other actives on component side are SOT23.
3. All passives are 0603 with large pads, which will also accommodate 0805 with some care.
4. From the schematic, all the SOIC-8 opamp pins are terminated at the DIP8 pins, except pin 6 (output), which drives the diamond cascode current booster, whose output goes to the DIP8 pin 6.
5. As a consequence, just about any SOIC-8 single opamp could be used - an LME49990 will probably be the king-of-the heap.
 

Attachments

  • lf07_dlb.jpg
    lf07_dlb.jpg
    85.7 KB · Views: 1,229
  • lf07_dlb_4v600r_thd20_fft.jpg
    lf07_dlb_4v600r_thd20_fft.jpg
    71.2 KB · Views: 1,198
  • lf07_sch.png
    lf07_sch.png
    5.4 KB · Views: 1,162
  • lf07v10_brd.png
    lf07v10_brd.png
    41.2 KB · Views: 1,106
  • lf07_lm318_cs_12x9.jpg
    lf07_lm318_cs_12x9.jpg
    254.8 KB · Views: 1,070
  • lf07_lm318_ss_12x9.jpg
    lf07_lm318_ss_12x9.jpg
    175.5 KB · Views: 436
Last edited:
Wonderful job. Are you able to measure this device? DC offset?
BOM?
Q1= SST4393?
Q2=BC847PN
Q3=MMBT5550?
Q4=MMBT5401?
U1= LME49860,LME9990

Thanks - I haven't specifically measured the DC offset, except at the output of a MyRef Rev C with an LM318M installed as U1. In that case, it was negligible - 0.6 and 1.4 mV respectively in the two channels at the main speaker output.

Preliminary BoM:

R1: 1k to 1k5/1%/0603 (for Idq of ~1 to 0.6 mA in the MMBF4393)
R2, R3: 3k3/1%/0603
R4, R5: 220R/1%/0603
R6, R7: 22R/1%/0603
Q1: MMBF4393/SST4393 (there may be better SOT23 NJFETs for a CCS)
Q2: NXP/Infineon BC847BPN SOT363
Q3: MMBT5551/BC817/MMBT2222 (any medium power NPN)
Q4: MMBT5401/BC807/MMBT2907 (any medium power PNP)
U1: Almost any SOIC8 single opamp - LM318, LT1028, LF356, OPA627/827, LME49710MA, LME49990, etc.

The existing prototypes use LM318M (~6 qty) and LME49990 (1st unit assembled this morning, tested working - will do another one and audition shortly).
 
Hi Guys

I do not see this as a diamond at all. Q3,4 are merely a bias regulator. Drive from the opamp is applied to the base of the PNP EF. You might find an improvement in performance if you added a cap across the EF bases.

To have this really be a diamond buffer in any form, the opamp output would be tied to the bases of Q3,4.

Also, the no-clean solder (Kester 245) is fine if you simply leave the flux residue in place - chip it off mechanically if you want to after it dries, but totally stay away from the water-clean solder. This garbage leaves residue that eats the board.

Have fun
Kevin O'Connor
 
What strikes me as interesting is the change in the relative levels of the 2nd,3rd (etc) harmonics in the simulation. The circuit seems to cancel some 3rd and make the 2nd higher, which may mean that (assuming such things are audible) the circuit will sound "better" than the opamp alone!

Also wondering what the performance would be if Q3/4 were not bootstrapped to the outputs and instead went to the rails?

If the function of Q3/4 are only for bias regulation, then the above connection ought not alter the distortion products.

Also why not use a standard VBE multiplier or other more or less passive WRT the signal bias method and see how that looks?

I keep looking at J1 and thinking "mirror"... not sure why, or if that would be of any benefit...

The other thing is, don't drink too much coffee before soldering!! :D
Linuxguru, so far I think this is an interesting Diamond Like circuit.
 
Last edited:
What strikes me as interesting is the change in the relative levels of the 2nd,3rd (etc) harmonics in the simulation. The circuit seems to cancel some 3rd and make the 2nd higher, which may mean that (assuming such things are audible) the circuit will sound "better" than the opamp alone!

Yup, the discrete output stage is biased with about 5-10 mA quiescent current, so it's in Class-A for typical high-Z loads and line-level or similar swings. So it tends to have Class-A sonics. The second part is that the CCS also biases the opamp output stage to Class-A, thus improving the audible sonics of those opamps that benefit from sinking constant current into the output pin (classic ones like NE5534, LF356, LM318, as well as some recent ones).

I've physically built and auditioned it with an LM318 - the sonics become Class-A, i.e. airy, smooth, less fatiguing.

Also wondering what the performance would be if Q3/4 were not bootstrapped to the outputs and instead went to the rails?

If the function of Q3/4 are only for bias regulation, then the above connection ought not alter the distortion products.

Q3, Q4 also have a (weak) lower cascode effect through the emitters of the output stage BJTs. I didn't actually sim it with the collectors tied to the rails, but it should also work with bias adjustments, without the cascode effect.

Also why not use a standard VBE multiplier or other more or less passive WRT the signal bias method and see how that looks?

It's a bit trickier to trim a standard Vbe multiplier in SMD, ie. to set the quiescent operating point. There are some alternatives that may work OK.

The other thing is, don't drink too much coffee before soldering!! :D
Linuxguru, so far I think this is an interesting Diamond Like circuit.

Thanks - a shot of Jack Daniels may work better :)
 
Last edited:
I do not see this as a diamond at all. Q3,4 are merely a bias regulator. Drive from the opamp is applied to the base of the PNP EF. You might find an improvement in performance if you added a cap across the EF bases.

Good idea - it's reasonably high-Z at the CCS, so a small X7R may work - say 100 nF.

Also, the no-clean solder (Kester 245) is fine if you simply leave the flux residue in place - chip it off mechanically if you want to after it dries, but totally stay away from the water-clean solder. This garbage leaves residue that eats the board.

I use an IPA-soaked tissue for manual cleaning, and it seems to work - I don't use water-clean stuff.
 
Good idea - it's reasonably high-Z at the CCS, so a small X7R may work - say 100 nF.

I checked it in the sim with 10 - 100 nF between the bases of the EFs, and H2 (and the evens) reduce quite dramatically with 100 nF. Unfortunately, H3 and the odds increase slightly, so there may be no net gain in sonics, even if THD20 decreases overall.

Nevertheless, it's a useful option, especially if the output stage is biased heavily into Class-A and there's sufficient room to decrease H2 while still allowing it to stay dominant.
 
Hi Guys

Good suggestions, Bear.

I would think that the main benefits you are seeing in any sim of this circuit compared to the raw opamp results is mostly the result of the bias shift of the opamp output due to the current source. This is an old trick, beginning simply as a resistor to one rail from the output.

Have fun
Kevin O'Connor
 
Hi Guys

That's essentially my point. The EF is just one stage - not a diamond - so it does give the benefit of current boosting. An open-loop single-ended EF produces mostly 2nd harmonic THD. One would expect some canceling of evens with a push-pull EF but this is not always the case. So, there may in fact be some harmonic enhancement with the EF on its own basis.

How did a brown bear get to the east coast? We only have black bears around here.

Have fun
Kevin O'Connor
 
FWIW, I added the base-to-base capacitor for the EFs and redid the layout, squeezing it in the same area as before - however, only a 0603 will fit for now, which is fine for small X7Rs. There's an added bonus - a zero-ohm resistor can be optionally fitted there to allow modifying the assembly (with the nJFET CCS installed, but the rest of the discretes omitted) for opamp + CCS Class-A operation alone.
 

Attachments

  • lf07v11_sch.png
    lf07v11_sch.png
    13.9 KB · Views: 367
  • lf07v11_brd.png
    lf07v11_brd.png
    40.4 KB · Views: 351
Disabled Account
Joined 2013
Thanks - I haven't specifically measured the DC offset, except at the output of a MyRef Rev C with an LM318M installed as U1. In that case, it was negligible - 0.6 and 1.4 mV respectively in the two channels at the main speaker output.

...

The existing prototypes use LM318M (~6 qty) and LME49990 (1st unit assembled this morning, tested working - will do another one and audition shortly).

Hi Siva, may I know what do you expected with your prototype installed as 'LM318' in MyRef Rev C?
Does it give better sound?
 
Hi Siva, may I know what do you expected with your prototype installed as 'LM318' in MyRef Rev C?
Does it give better sound?

I certainly prefer it to the plain LM318 for the kind of music I listen to - the audible sonics of the LF07 are distinctly Class-A in a MyRef Rev C: better definition, resolution, micro-detail, air, smoothness, etc.

However, for the differences to be audible, the other components in the MyRef also have to be of sufficiently high quality - C13, C9, C12, C7, R12, C10, C32, C30, C34, C6, C11, etc. have an impact on sound quality in a MyRef.
 
Have you compared it to the Chris Paul/Bob Cordell cross coupled output buffer? (Cordell uses that connection, but Chris Paul published it I think earlier... still trying to get to the bottom of that... like where that circuit came from)

Is there a link that shows or discusses this topology? Bob has posted a lot of stuff here, so it's daunting to wade through a zillion posts.

Meanwhile, the base-to-base capacitor for the EFs suggested by Kevin does not appear to hurt audible sonics, and it sims with lower H2 and THD20 as I mentioned earlier. I kludged a 100nF Panasonic ECHU stacked PPS film cap into a couple of LF07 modules and auditioned them in a MyRef Rev C. The sonics are the same as before, except that the definition and resolution of the highs seem enhanced. I'll listen to it for a while, but it already looks like a worthwhile enhancement.
 
Yeah... ummm... it's used in the output stage of his power amps, iirc... in his book. I mentioned it some years ago in a thread about output buffers, and suggested the connection instead of having the drivers collectors going to the rails, go to the junction of the output's emitter + emitter resistors on the opposite side.
So instead of going to the + rail, the lower driving transistor would connect to junction of the output run off the + rail and its emitter resistor...

dunno if that makes any sense...

I'll try to find an example I can link or upload.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.