Blue Hawaii BJT Build thread

[JimL]

On 4/24/2017 at 5:44 AM, kevin gilmore said:

the closest thing to a perfect current source is a high voltage tube with a cathode resistor, i.e. megatron

T2 current source is actually not really very good because the resistor divider for the pnp current source causes a significant change in current

for example 700v peak to peak on the plate the current varies from 13.3 to 15.8ma

same thing for a cascoded depletion mode is 14.8ma to 14.85ma

a better idea would be to replace the stacked pnp and resistor with a very highly isolated dc to dc converter, something good for 2kv of isolation and you would need 4 of them per amplifier.

So the T2 current source calculates out to an effective impedance of around 280 kilohms, the cascoded DN2540 to around 14 megohms.  Wonder how good the performance would be if you replaced the stacked pnp/resistor combo of the T2 with the cascode MOSFET current source feeding D3/D4?    

[sorenb]

30 minutes ago, JimL said:

So the T2 current source calculates out to an effective impedance of around 280 kilohms, the cascoded DN2540 to around 14 megohms.  Wonder how good the performance would be if you replaced the stacked pnp/resistor combo of the T2 with the cascode MOSFET current source feeding D3/D4?    

...and how does the numbers look like at, lets say 4kHz?

[kevin gilmore]

I will synthesize that and get back to you

[kevin gilmore]

twice the impedance at 10k than a cascade current source alone, 20 times better than the t2 current source

but by specs, only good to about 930 volts

 

 

Edited May 2, 2017 by kevin gilmore

[JimL]

So basically very little advantage over the cascode current source alone.  My measurement of the cascode using a 10M90S upper and DN2540 lower gave somewhat better results that a dual DN2540 cascode, I wonder if the simulation would also show better results.

[sorenb]

the fet (cascoded or not) will deteriorate across the audio region where as most bjt CCS does not.
If you just want high numbers, fets seems to be the answer.
If one want linearity bjt seems to be a better option.  

 

 

[JimL]

I think we're going around in circles on this.  

 

Here's my argument.  We can roughly model the roll-off at higher frequencies as a capacitor.  So let's say the BJT current source is equivalent to a 300 kilohm resistor and a parallel 2 pf cap, whereas the MOSFET current source is like a 15 megohm resistor and a parallel 2 pf cap.  The MOSFET impedance will start to bend over earlier in the audio spectrum because its DC resistance is higher to begin with, and the BJT will look flatter over the audio spectrum.  At the highest frequencies they will both look like a 2 pf cap so they will eventually meet.

 

BUT, within the audio spectrum, the BJT will burn up more signal current because it has a lower resistance, so the output device will have to work harder with the BJT than the MOSFET, and therefore will cause more distortion due to the output device working harder.  

 

Again, if we model the headphone as a resistor and capacitor in parallel (the resistor being there to model the fact that we're using energy to make sound), then the output device sees the headphone load (resistor and capacitor) and the current source (resistor and capacitor) all in parallel.  The amount of signal current it has to generate depends on the the resistors in parallel and the capacitors in parallel.  The higher the signal current generated the higher the distortion.  If the current source "resistance" is higher it wastes less signal current in it.  If the current source "capacitance" is small, the amount of signal current it uses to swing the capacitance doesn't really differ much from that needed for the headphone alone.  The cascoding of the MOSFETs ensures that this extra capacitance is on the order of a few pf, much less than that of the headphones alone, so it really doesn't make much difference.

 

The ideal current source has infinite impedance, and wastes no signal current, so it should be sonically invisible.  My gut instinct is that the less signal current it uses, the less sonic imprint the current source should have.  A flatter impedance may result in a more consistent sonic imprint, but a higher impedance should result in a lesser sonic imprint, so my argument is, higher is better.

 

 

[sorenb]

11 minutes ago, JimL said:

 My gut instinct is that the less signal current it uses, the less sonic imprint the current source should have.

Agree

12 minutes ago, JimL said:

A flatter impedance may result in a more consistent sonic imprint

Agree 

so I guess we just disagree on the importance of consistency then.
Since acoustical instruments have their fundamentals associated with some complex mix of harmonics I'd say it is important to treat all those with consistency.
There might be more to it though, as I didn't really hear much of a difference in my Carbon as I did on my BH ...
Some might also remember the difference stated between KGSShv IXYS/fet) and Sanyo(bjt) ... most prefer the latter for some reason ...

[JoaMat]

I think I like the c4686 with two LEDs and a 9 volts battery current source.

[JoaMat]

DC/DC converter followed by LT3045 regulator.

If this works I've well-regulated 9 V well isolated and hopefully it could be used with NPN current sources. If not I've two dozen of L3045 for sale.

[spritzer]

Make sure to measure the noise as that's the main problem with these

[JoaMat]

 

 

35 minutes ago, spritzer said:

Make sure to measure the noise as that's the main problem with these

Which part is the noisy one?

[spritzer]

The DC-DC converter. 

[JoaMat]

The output from DC/DC converter is pretty ugly. How much of this that will be stopped by LT3045 is to be seen. LT3045 datasheet looks rather nice though.

P.S.
Should have ordered LT3042 instead.

Edited May 5, 2017 by JoaMat

[JoaMat]

Seems Birgir is right about noise. I tested it with a scope. As expected the output from the converter is ugly. It’s better after the LT3045 but there is too much noise passing through.

[kevin gilmore]

I think the cascaded fets instead of the 2 x pnp and resistors etc is the easiest to increase the performance.

[PretentiousFood]

On 5/6/2017 at 4:08 AM, JoaMat said:

Seems Birgir is right about noise. I tested it with a scope. As expected the output from the converter is ugly. It’s better after the LT3045 but there is too much noise passing through.

What kind of noise? I've had good luck filtering switchers with a CLC with -3db a decade below fosc (so ~10khz), to pick up where the regulator starts to drop off. Gives values in the low mH and uF range. 

[JoaMat]

1 hour ago, PretentiousFood said:

What kind of noise? I've had good luck filtering switchers with a CLC with -3db a decade below fosc (so ~10khz), to pick up where the regulator starts to drop off. Gives values in the low mH and uF range. 

It’s spikes from the DC/DC converter. The way I built this the spikes are passing through the LT3045 more or less unaffected. I lack understanding of and knowledge how to filter the spikes and also how to design the board so it will work properly.

Any help is appreciated. I find the idea with isolated DC/DC converter interesting.

[JoaMat]

Just had to try the DC/DC converter. Converter in red circle.

According to the datasheet the converter is rated at 1400VAC 60 Hz for one minute. So one of four current sources has the converter. Seems to work.

Regarding spikes mentioned earlier I’ve done some "googling" and I believe the issue can be solved.

[Kerry]

On ‎4‎/‎24‎/‎2017 at 4:16 PM, kevin gilmore said:

here are 3 current sources at 10khz

 

currentsource.PDF

I've been playing with the current sources in spice.  This one can cover 1000V and models really well.  I'm not convinced the performance is not just LTSpice and the model I found for the IXTP01N100D.  The LT1117 could be a LM317.

Edited June 1, 2017 by Kerry

[JoaMat]

Interesting current source.

Would 01N100D/DN2540 cascoded work (in LTSpice)?

[Kerry]

Yes, but the regulators perform much better in spice, especially at higher frequencies.

[JoaMat]

Thank you, good to know.

[JoaMat]

Two current sources. One “normal CCS” and one “Kerry’s in progress regulator CCS”.

[kevin gilmore]

really think you should add the t2 current source modified with 2sc4686 and cascade current source to drive the led reference, that still synthesizes the best of everything