"The Bog Standard" - A good enough amplifier for the rest of us

Some of you asked for schematics and details. Everything, including CAD files and SPICE simulations is in this Github repo. Keep in mind that this is a work in progress and there might still be bugs to work out. Preliminary tests have been successful so far, but proceed with caution if you want to use the design!

https://github.com/prydin/the-bog-standard

If you see something amiss, please raise an issue in the repo, or even better, fix it and submit a PR. 🙂
 
@njswede
I was at Github PRydin.
Your amplifier in the Bog Standard.
Most of it is good - it is working.

I think a possible weak spot is the Vbe multiplier.
Here is my take on this:
Vbe Suggestion.jpg
 
  • Like
Reactions: STOXX
You can reduce your component count by dropping the two transistors in the CCS for the input stage.
This will balance the H3 dominant output of the LTP with a little more H2. Supply the LTP with two resistors in series, with a 47uF from the midpoint to ground for smoothing. More H2 will add a touch of warmth, making the amp sound very slightly tubey.
Then you can drop the gate drivers. You should use about 10mA through the VAS stage, two comp IRF240/9240 gates are easy to drive and you are not moving 200KHz.
Finally, the bias generator is best used with another small mosfet with an identical tempco. The suggestions using bipolars are more complex, and the tempco will not identically match.

HD
 
Adding a second diode-coupled transistor for thermal compensation doesn't seem necessary from what I've seen so far. Mounting Q7 on the same heatsink as the output devices seems to makes it fairly thermally stable. It does briefly spike to about 150-170mA bias current on startup when everything is cold, but after a minute or so, it settles nicely at 100mA. Once the MOSFETs warm up, they seem to have roughly the same tempco as the VBE, since the bias current only varies by about 10% even if I crank up the volume a lot. This is with a proper heat sink fitted. When I ran it with just a piece of sheet metal, it had a very significant negative tempco.
 
You can reduce your component count by dropping the two transistors in the CCS for the input stage.
This will balance the H3 dominant output of the LTP with a little more H2. Supply the LTP with two resistors in series, with a 47uF from the midpoint to ground for smoothing. More H2 will add a touch of warmth, making the amp sound very slightly tubey.
I'm still ruminating on some contrarian Le Monstre / ACA Mini -like topologies, taking local feedbacks from source resistors in the output stage.

Apart from source degeneration, taking further steps to linearise the output MOSFETs individually is rare. But I think it could make them perform a bit more like single-ended designs. One general "push-pull problem" I see is that the cancellation of harmonics is uncontrolled. There's nothing to prevent a polarity reversal / "zero crossing" of harmonics.

E.g.: the IRFP240 generates some H2, gradually sloping UP in amplitude. The IRFP9240 generates a similar amount of H2, sloping DOWN. For a brief moment there is perfect cancellation as the amplitudes are equal. Then the amplitude bounces back. And the overall envelope of this H2 modulation is unrelated to the sine wave that it's a harmonic of, but may be influenced by something else like a bass line playing at other frequencies.

A related curiosity is that, in some simulations, MOSFET source resistors showed around ~6% HD in an open loop test. Of course tying the drains together reduced the combined distortion to below 1%, but how steady is that in response to DC (low freq.) offsets? At the same time, the speaker is loading down the output by a slightly-mysterious amount, while the NFB is trying to sense the voltage. So there's a lot going on. This makes me wonder if that's the secret sauce behind Nch-Nch designs, or single ended, where the CCS load does not cancel anything. Food for thought.
 
I’ll gladly admit that I didn’t understand much or that post, especially not above the paragraph.
I can attach a sketch, when I'm not on the go. Maybe the example wasn't clear? Rephrasing: the IRFP(9)240 transistors both generate distortion, but in opposite polarity, so the combined total is smaller than either on its own. That's the idea anyway. However, due to real-world modulation they may switch places:

IRFP9240: reduces from 1% H2 ---> to 0.9%,
IRFP240: increases from 0.9% H2 ---> to 1%

Somewhere in the middle, on a time scale, there is a perfect null: 0.95% - 0.95% = 0.0000...%.
For multi-tones, e.g. 100Hz and 1kHz, the 2nd harmonic @ 2kHz would have a 100Hz beat frequency, aka amplitude modulation. One funny thing is, by slight de-tuning or adding a small amount of even HD at the input, the dynamic range of the beats (max-dB minus min-dB) could be reduced.
 
with the resistor 'tail' for the input LTP. And you could vary the length of the tail by using a string of resistors and compare the sound with the capacitor connected at different voltage points.
See here, crazy stuff.. :)
ccs_w_bloom_control_002-png.149388

ps. OStripper tried to simulate it but used a very small cap, C1 (C5 0,1 uF in his schematics), I think it should be much larger, unfortunately I never tried this circuit in reality so I don't know the result.. :(
 
Last edited:
  • Like
Reactions: abstract
I tried to replace the CCS with a simple resistor for the LTP and got some interesting results. The total THD is about the same, but there's a significant (albeit probably inaudible) difference in H3. It's a fair bit lower without the CCS. The quiescent current through the LTP is about 1.4mA in both cases.

Conventional wisdom seems to be that a CCS improves OLG and therefore helps smoothing out crossover distortion, but I guess there's something else going on here.
 

Attachments

  • with ccs.png
    with ccs.png
    23.7 KB · Views: 43
  • without ccs.png
    without ccs.png
    20.3 KB · Views: 42