Haven't posted much on this for a while because I have spent lots of time thinking and experimenting. The circuit presented here looks like it will be pretty much the final one. It sounds so clean. Transients are sharp and clear. Vocals are very smooth. My standard test CD is the 1984 version Metropolis soundtrack by Giorgio Moroder, originally recorded on a Sony PCM-something digital recorder.
All I have to do now is build the thing, and AKSA, if you read this I will then haul it over to your place for an audition. This will definitely be fun.
GP.
All I have to do now is build the thing, and AKSA, if you read this I will then haul it over to your place for an audition. This will definitely be fun.
GP.
Attachments
Yep. At this point I plan it to be 4x 6 amp 600v TO-220 rectifier diodes (BYV249-600) in series, 2 in proximity to each fet. Obviously I don' t need diodes this big because I am only going to put 100mA through them, but the fact is I have just over 900 of them, and also the metal tab allows good thermal contact.
I am going to try a variation of the bias voltage circuit I suggested in another thread a while back that has 2 pots, 1 for offset and the other for slope. Put the slope at zero, and set the offset for the right quiescent current with the amp cold. As it heats up the current will increase, so adjust the slope pot that set the right amount of downward slope of mV per deg C till the current gets back down to where it should be. Finally when the amp is fully hot, tweak the slope pot till the current is exactly right. Now the bias should more or less track the temperature over the whole range. The only error will be if the diode forward voltage tempco or the fet threshold voltage tempco is not a straight line. Actually the threshold voltage is tempco slightly curved so with a bit of luck the diode curve will be too. We'll see. Anyway, it should be a whole lot better than that ugly hit-and-miss "amplified diode" setup using a single transistor and a pot. Anyone still using that setup in this day and age should be put in jail.
GP.
I am going to try a variation of the bias voltage circuit I suggested in another thread a while back that has 2 pots, 1 for offset and the other for slope. Put the slope at zero, and set the offset for the right quiescent current with the amp cold. As it heats up the current will increase, so adjust the slope pot that set the right amount of downward slope of mV per deg C till the current gets back down to where it should be. Finally when the amp is fully hot, tweak the slope pot till the current is exactly right. Now the bias should more or less track the temperature over the whole range. The only error will be if the diode forward voltage tempco or the fet threshold voltage tempco is not a straight line. Actually the threshold voltage is tempco slightly curved so with a bit of luck the diode curve will be too. We'll see. Anyway, it should be a whole lot better than that ugly hit-and-miss "amplified diode" setup using a single transistor and a pot. Anyone still using that setup in this day and age should be put in jail.
GP.
janneman said:
Actually, in general terms it is called "slope" and "intercept". Sounds much more impresseive to me
Jan Didden
Hmmm... perhaps "slope and intercept" for a downward sloping line as in this case, and "slope and offset" for an upward sloping one?
Go Aussie? Well, the whole world knows Australia has the worlds best rocket scientists - a la scramjet. For a fleeting moment anyway.
heh heh Anyway, here's the cct. I haven't built it yet. I'll have a go on Wednesday night.
Hooroo.
GP.
Attachments
Go Aussie......!
Graham,
I will be very interested to hear this beast.
I agree with Jan; your ingenuity is just incredible. You have more good ideas in a day than I've had hot dinners......
When you do show it to me, I shall collectively toast you on behalf of all the contributors here, with 324 cups of strong coffee.
Cheers,
Hugh
www.aksaonline.com
PS Jan, thank you for your IEEE paper, I really appreciate it!
Graham,
I will be very interested to hear this beast.
I agree with Jan; your ingenuity is just incredible. You have more good ideas in a day than I've had hot dinners......
When you do show it to me, I shall collectively toast you on behalf of all the contributors here, with 324 cups of strong coffee.
Cheers,
Hugh
www.aksaonline.com
PS Jan, thank you for your IEEE paper, I really appreciate it!
Dept. of thinking-out-loud
Just ruminating over a new gate drive cct using a long tailed pair with an opamp to make sure everything stays balanced to the nth degree. Avoids the need for HV opamps which is a good idea for when I want to scale up the power output. Gate drive fets are just some TO-220 thingies with enough voltage rating. Only a general idea presented here. I think I'll have to wind back the driver mosfet gain with some source resistance perhaps. Hmm... I feel like a cup of tea. Where's my car keys. Oops! Forgot! People listening to me thinking.
GP.
Just ruminating over a new gate drive cct using a long tailed pair with an opamp to make sure everything stays balanced to the nth degree. Avoids the need for HV opamps which is a good idea for when I want to scale up the power output. Gate drive fets are just some TO-220 thingies with enough voltage rating. Only a general idea presented here. I think I'll have to wind back the driver mosfet gain with some source resistance perhaps. Hmm... I feel like a cup of tea. Where's my car keys. Oops! Forgot! People listening to me thinking.
GP.
Attachments
Design unfrozen...
The gate drive cct has drifted to something like this now. I would call it a "cross-coupled paraphase long-tailed pair". That's a mouthfull! The top rail would be 110v, allowing for 10v across the constant current source at the bottom. I have seen some low level ccts using a current mirror at the collectors of a normal long-tailed pair. What is the purpose of that?
The way I have things set up in my cct here is that whenever either driver fet starts to turn on it gives a very low impedance drive to the output fet gates. What's more, the low impedance is symmetrical, i.e. there is no hard pulldown by a fet and only a soft pullup by a resistor, it should be hard driven both directions.
GP.
The gate drive cct has drifted to something like this now. I would call it a "cross-coupled paraphase long-tailed pair". That's a mouthfull! The top rail would be 110v, allowing for 10v across the constant current source at the bottom. I have seen some low level ccts using a current mirror at the collectors of a normal long-tailed pair. What is the purpose of that?
The way I have things set up in my cct here is that whenever either driver fet starts to turn on it gives a very low impedance drive to the output fet gates. What's more, the low impedance is symmetrical, i.e. there is no hard pulldown by a fet and only a soft pullup by a resistor, it should be hard driven both directions.
GP.
Attachments
Back to reality.
Seems I have being doing too much theorizing and to little "get your hands dirty" experimenting lately. I was looking for linearity and symmetry to the 15th decimal place, on paper at least. Then I rigged up a little mosfet on the bench as a phase splitter and I could barely believe the results. Driving the gate of the phase splitter from a 50 ohm source and 10v p/p, and the top and bottom outputs only driving the cro probes, the response was flat to 5 Megahertz! It then started to peak up around 7 megs or so. At normal audio frequencies the lower output was within 7mV of the input voltage (10v p/p) upper output was similar. Lower output had a 200 picosecond(!) delay in it's sinewave zero crossing referenced to the input. The upper ouptut had a 100nS delay; that's 0.8 deg at 20kHz. Not too shabby for just half a dozen bit and pieces.
Then I put a x2 gain stage ahead of it. The capacitive loading of the phase splitter input on the gain stage causes the whole thing to be 3dB down at 460 kHz. Drat! I might just try some lower rated fets that have smaller junctions and therefore lower capacitance. IRF540's are rated at 100v 28 amps, and seeing there is only about 50 mA going through them they are a bit of an overkill. I used them because I had some.
I wasn't really looking forward to using constant current sinks and balanced circuitry and as a result the amplifier was becoming decidely un-fun. Seeing I'm making it just for fun, it was time to make a change in direction. It has been simplified greatly, always a good thing if you have to do point to point wiring! It's just after lunch now, so when I get home tonite, Wednedsay here, I'll throw it together quick and see just what happens.
GP.
Seems I have being doing too much theorizing and to little "get your hands dirty" experimenting lately. I was looking for linearity and symmetry to the 15th decimal place, on paper at least. Then I rigged up a little mosfet on the bench as a phase splitter and I could barely believe the results. Driving the gate of the phase splitter from a 50 ohm source and 10v p/p, and the top and bottom outputs only driving the cro probes, the response was flat to 5 Megahertz! It then started to peak up around 7 megs or so. At normal audio frequencies the lower output was within 7mV of the input voltage (10v p/p) upper output was similar. Lower output had a 200 picosecond(!) delay in it's sinewave zero crossing referenced to the input. The upper ouptut had a 100nS delay; that's 0.8 deg at 20kHz. Not too shabby for just half a dozen bit and pieces.
Then I put a x2 gain stage ahead of it. The capacitive loading of the phase splitter input on the gain stage causes the whole thing to be 3dB down at 460 kHz.
Drat! I might just try some lower rated fets that have smaller junctions and therefore lower capacitance. IRF540's are rated at 100v 28 amps, and seeing there is only about 50 mA going through them they are a bit of an overkill. I used them because I had some.I wasn't really looking forward to using constant current sinks and balanced circuitry and as a result the amplifier was becoming decidely un-fun. Seeing I'm making it just for fun, it was time to make a change in direction. It has been simplified greatly, always a good thing if you have to do point to point wiring! It's just after lunch now, so when I get home tonite, Wednedsay here, I'll throw it together quick and see just what happens.
GP.
Attachments
current mirror
Usually if there is a current mirror in the collectors of a differential pair, it is there because the following stage is single-ended, and it gives a way to double the effective transconductance of the diffamp, by capturing the signal current from the opposite side that would have otherise just "gone to waste" to the power rail.
Another nice benefit of using a current mirror is that it sort of enforces balance on the differential pair if the whole thing is wrapped in a feedback loop with enough gain.
Usually if there is a current mirror in the collectors of a differential pair, it is there because the following stage is single-ended, and it gives a way to double the effective transconductance of the diffamp, by capturing the signal current from the opposite side that would have otherise just "gone to waste" to the power rail.
Another nice benefit of using a current mirror is that it sort of enforces balance on the differential pair if the whole thing is wrapped in a feedback loop with enough gain.
Faster fets
Mirlo, what do you think a source follower would do in this situation? More able to drive the capacitance? Un-fun means getting all stressed out from my imagination, thinking it will be all just too hard
I removed the IRF540's and put in some STP3NB60's. TO-220, 600v 3.9 amp, 3R9 and much smaller capacitance. Most importantly though, I actually had some in my seemingly bottomless junkbox. That made the choice easy. Now the output is 3dB down at 3.1 Mhz, same conditions as before. That will set my tweeters on fire for sure. Actually the main reason for trying to get this good response is so the opamp is the dominant pole no matter what. Then the feedback loop should behave itself. It will be the source of any drooping frequency response and therefore phase shift, not anything else. Of course it won't be quite as good when I connect the output devices, but anyhoo...
Put a square wave in and at 300kHz and 40v p/p output from the second fet drain it still looked quite reasonably square. The waveform from the generator had about 20nS rise and fall time, and the 40v p/p output actually measured 103 nS rise and 86 nS fall. That should be enough I think. Time to go home and stick some music through it!
GP.
Mirlo, what do you think a source follower would do in this situation? More able to drive the capacitance? Un-fun means getting all stressed out from my imagination, thinking it will be all just too hard
I removed the IRF540's and put in some STP3NB60's. TO-220, 600v 3.9 amp, 3R9 and much smaller capacitance. Most importantly though, I actually had some in my seemingly bottomless junkbox. That made the choice easy. Now the output is 3dB down at 3.1 Mhz, same conditions as before.
That will set my tweeters on fire for sure. Actually the main reason for trying to get this good response is so the opamp is the dominant pole no matter what. Then the feedback loop should behave itself. It will be the source of any drooping frequency response and therefore phase shift, not anything else. Of course it won't be quite as good when I connect the output devices, but anyhoo...Put a square wave in and at 300kHz and 40v p/p output from the second fet drain it still looked quite reasonably square. The waveform from the generator had about 20nS rise and fall time, and the 40v p/p output actually measured 103 nS rise and 86 nS fall. That should be enough I think. Time to go home and stick some music through it!
GP.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.