R11 and R33 are supposed to be 47.5K - it's the correct value, straight from Grey's original schematic. I have built the prototype channel over a year ago with these values and it has always worked properly. Note that R11/V1's only function is to supply a wee bit of extra bias current to the base of Q3. The more current is provided at Q3's base, the lower the current through Q1 will be. However, 4.7K seems a fair bit too low to me. Grey used high values specifically so that this little resistor network wouldn't load and interfere too much with the voltage dividing duties of R8 / R12.
Note that MPSA18 has a DC h(fe) of 500 - 1500, while BC550 has 110 - 800 (fairchild datasheets), so it is possible that you got a part with low h(fe) and this contributed to the problem. I suspect there is still another reason, though. Have you actually measured all the other component values carefully? I'm thinking especially of R14 and R15 here, but it wouldn't hurt to check R7 and R8 as well... it's possible some of them have been damaged and are no longer their proper value. You may also want to re-solder the joints in this part of the circuit, in case you've got a cold joint somewhere.
One other noteworthy point: Q4 and Q9 will normally kick in to protect Q2 and Q11 from damage, if you've got the voltage dividers set up properly for the desired current limit (R10 & R13, with R6 being the current sensing element). If this happens, and the current source is not happy, it may start dumping current through the load, and subsequently activate the current protection on the other side as well. When this happens, the errant current source will not be able to deliver the current it wants to, and thus your gate voltage may 'run away' to some high value... like, say, 14V... Keep in mind, the max Vgs for these power mosfets is around 20V, and once you've exceeded the SOA or punched a hole through the delicate oxide, all sorts of bad things will happen to your circuit.
Note that MPSA18 has a DC h(fe) of 500 - 1500, while BC550 has 110 - 800 (fairchild datasheets), so it is possible that you got a part with low h(fe) and this contributed to the problem. I suspect there is still another reason, though. Have you actually measured all the other component values carefully? I'm thinking especially of R14 and R15 here, but it wouldn't hurt to check R7 and R8 as well... it's possible some of them have been damaged and are no longer their proper value. You may also want to re-solder the joints in this part of the circuit, in case you've got a cold joint somewhere.
One other noteworthy point: Q4 and Q9 will normally kick in to protect Q2 and Q11 from damage, if you've got the voltage dividers set up properly for the desired current limit (R10 & R13, with R6 being the current sensing element). If this happens, and the current source is not happy, it may start dumping current through the load, and subsequently activate the current protection on the other side as well. When this happens, the errant current source will not be able to deliver the current it wants to, and thus your gate voltage may 'run away' to some high value... like, say, 14V... Keep in mind, the max Vgs for these power mosfets is around 20V, and once you've exceeded the SOA or punched a hole through the delicate oxide, all sorts of bad things will happen to your circuit.
Yup. I changed back to 47K and this does not give me enough headroom for adjustment.
OK, Here are the results for the moment:
Successes:
- 0,02V absolute offset
- less than 0,01V offset between + and -
- with a simple choke-input-supply 30mV ripple in the rail which results in
- 0,18mV ripple at the speaker, I guess it is already very quiet
Still challenges:
- On one side I have two 044n in the gain stage. Now, one gives me 0,33V, the other 0,5 V over the 0,33 resistor. Now, when I exchange both with each other, the problem does not move with the mosfet, the values stay ! Ok, move all resistors and cabeling around as well: The problem stays at the same location. OK, dis-mounted the mosfet with the lower value: Now we have normal values of 0,5V...Ok, changed the insulator to mica...stays..back to silpad...stays, anyhow no connection can be measured to the heatsink...but when mounting it back to the heatsinks, it behaves again strange...
...to be precise: It starts fo a short time with the right value and than the voltage drops from 0,45V to 0,33V over 60 sec.
Ok, may be ocillation, so put in a 47pf mica capacitor between gate and collector...same behaviour...
By the way: The differences between voltages over the 392R-resistors are still there and don't move as well when swapping the mosfets against each other. So, THere's still something in the circuit which does not work as intended...any suggestions on how to trace it back are highly welcome.
OK, Here are the results for the moment:
Successes:
- 0,02V absolute offset
- less than 0,01V offset between + and -
- with a simple choke-input-supply 30mV ripple in the rail which results in
- 0,18mV ripple at the speaker, I guess it is already very quiet
Still challenges:
- On one side I have two 044n in the gain stage. Now, one gives me 0,33V, the other 0,5 V over the 0,33 resistor. Now, when I exchange both with each other, the problem does not move with the mosfet, the values stay ! Ok, move all resistors and cabeling around as well: The problem stays at the same location. OK, dis-mounted the mosfet with the lower value: Now we have normal values of 0,5V...Ok, changed the insulator to mica...stays..back to silpad...stays, anyhow no connection can be measured to the heatsink...but when mounting it back to the heatsinks, it behaves again strange...
...to be precise: It starts fo a short time with the right value and than the voltage drops from 0,45V to 0,33V over 60 sec.
Ok, may be ocillation, so put in a 47pf mica capacitor between gate and collector...same behaviour...
By the way: The differences between voltages over the 392R-resistors are still there and don't move as well when swapping the mosfets against each other. So, THere's still something in the circuit which does not work as intended...any suggestions on how to trace it back are highly welcome.
Thanks, will try that. What types do you use ? Mica ? Ceramic ?
Just for the records (in Hifizen's schematic):
- Exchanged R14/R31
- Exchanged R15/R32
- Exchanged R8/R37
- Exchanged Q3/Q8
No effect.
In the meantime I made the following observations:
1. On the one side, the current sources work nicely at the same current, but as described above the voltage across the 0,33R of the gain 044ns the situation becomes worse: Now I have 0,50V at one mosfet and 0,21V across the other, so one mosfet has to work harder and harder
2. On the other side, the gain side is nice, but I have similar phenomens on the current side: there I have one mosfet working with 0,6V very hard while the other one with 0,3V is a lazy guy.
Differences across the 392 stayed as before.
One question: I measure the supply voltages at the mosfets and they are all the same. Nevertheless, I made them p2p with strained Teflon cable, specified for 6A. May this be a source of problems as well ? Or are the 044n in combination with the BC550C something bad ?
Just for the records (in Hifizen's schematic):
- Exchanged R14/R31
- Exchanged R15/R32
- Exchanged R8/R37
- Exchanged Q3/Q8
No effect.
In the meantime I made the following observations:
1. On the one side, the current sources work nicely at the same current, but as described above the voltage across the 0,33R of the gain 044ns the situation becomes worse: Now I have 0,50V at one mosfet and 0,21V across the other, so one mosfet has to work harder and harder
2. On the other side, the gain side is nice, but I have similar phenomens on the current side: there I have one mosfet working with 0,6V very hard while the other one with 0,3V is a lazy guy.
Differences across the 392 stayed as before.
One question: I measure the supply voltages at the mosfets and they are all the same. Nevertheless, I made them p2p with strained Teflon cable, specified for 6A. May this be a source of problems as well ? Or are the 044n in combination with the BC550C something bad ?
I just found this Aleph-X-schematic:
http://www.myclie.info/images/ALEPHXSCH.pdf
There the Anti-Osci-Cap is not from collector to base but from collector to emitter. What is adviseable ?
Best Regards
http://www.myclie.info/images/ALEPHXSCH.pdf
There the Anti-Osci-Cap is not from collector to base but from collector to emitter. What is adviseable ?
Best Regards
RE: Osc.
I have used 1nF MKT. From C to B
I also tried using a capacitor for C to E. I had to use 47nF to tame
the oscillation. This big a cap will have an influence on the
operation of the Aleph current source. So I switched to the smal
valued 1nF on C to B at the bipo.
Try sticking to a single schematic. HifiZen's for instance. Your link
points to a pdf with a modified version. No values in the
schematic. I don't even know if there is an amp build based on
this modified version. I would not use it, if I were you.
Regards
I have used 1nF MKT. From C to B
I also tried using a capacitor for C to E. I had to use 47nF to tame
the oscillation. This big a cap will have an influence on the
operation of the Aleph current source. So I switched to the smal
valued 1nF on C to B at the bipo.
Try sticking to a single schematic. HifiZen's for instance. Your link
points to a pdf with a modified version. No values in the
schematic. I don't even know if there is an amp build based on
this modified version. I would not use it, if I were you.
Regards
Blitz:
The stabilizing caps should go between base and collector. This forms the local feedback loop around the transistor which lowers it's gain at high frequencies, thus taming the oscillations. Also, the Miller effect essentially multiplies the cap's value by the gain of the transistor, so a much smaller value can be used than would otherwise be necessary for a given pole frequency. I'd use a good ceramic type in this position.
1nF might work OK, but I'd try for to find the smallest value that works properly. If you're in the sub-100 pF range, you can use a 'gimmick' to find what value works... just take apart some cat5 cable or telephone wire and use one of the twisted pairs as a really long capacitor... usually they'll specify the capacitance per length in the datasheet, or you can make your own gimmick from solid-core hookup wire. 1pF per inch is approximate, if you don't have a cap meter. Keep trimming bits off the end of the gimmick until the oscillations reappear, then you know where the stability limit is. That'll give you the capacitance needed for a teeny-tiny phase margin. Multiply the value a few times, and test with a real cap of that size. If you have a square wave generator and oscilloscope, you'll be able to see overshoot and ringing, which is more useful than just watching for full-blown oscillations.
Sort out the oscillations first, then deal with the DC problems. My bet is, if your speaker output has a low differential DC offset (and it does), then the diff pair is doing it's job just fine, and the difference in voltage you see across the 392 ohm resistors is what the diff pair is doing to compensate for differences in the output stages on each side of the amp. I'd be interested to know what happens when you remove the lower BC550s (Q4/Q9)... be careful when you do this, as these transistors protect your lower output devices from too much current! Watch the voltage across R6 as you power up and if it gets too high, then shut it off before anything blows up.
If your amp is built exactly as in Netlist's previously posted schematic, then the output transistors on one side should share current very nicely. If not, then you have some matching / heatsinking problems (one transistor is getting hotter than the others), or possibly a source resistor which is damaged, and changing value as it heats up. Have you also checked the 221-ohm gate resistors to make sure they're still OK?
The stabilizing caps should go between base and collector. This forms the local feedback loop around the transistor which lowers it's gain at high frequencies, thus taming the oscillations. Also, the Miller effect essentially multiplies the cap's value by the gain of the transistor, so a much smaller value can be used than would otherwise be necessary for a given pole frequency. I'd use a good ceramic type in this position.
1nF might work OK, but I'd try for to find the smallest value that works properly. If you're in the sub-100 pF range, you can use a 'gimmick' to find what value works... just take apart some cat5 cable or telephone wire and use one of the twisted pairs as a really long capacitor... usually they'll specify the capacitance per length in the datasheet, or you can make your own gimmick from solid-core hookup wire. 1pF per inch is approximate, if you don't have a cap meter. Keep trimming bits off the end of the gimmick until the oscillations reappear, then you know where the stability limit is. That'll give you the capacitance needed for a teeny-tiny phase margin. Multiply the value a few times, and test with a real cap of that size. If you have a square wave generator and oscilloscope, you'll be able to see overshoot and ringing, which is more useful than just watching for full-blown oscillations.
Sort out the oscillations first, then deal with the DC problems. My bet is, if your speaker output has a low differential DC offset (and it does), then the diff pair is doing it's job just fine, and the difference in voltage you see across the 392 ohm resistors is what the diff pair is doing to compensate for differences in the output stages on each side of the amp. I'd be interested to know what happens when you remove the lower BC550s (Q4/Q9)... be careful when you do this, as these transistors protect your lower output devices from too much current! Watch the voltage across R6 as you power up and if it gets too high, then shut it off before anything blows up.
If your amp is built exactly as in Netlist's previously posted schematic, then the output transistors on one side should share current very nicely. If not, then you have some matching / heatsinking problems (one transistor is getting hotter than the others), or possibly a source resistor which is damaged, and changing value as it heats up. Have you also checked the 221-ohm gate resistors to make sure they're still OK?
After some searching I found the post that led to the explanation of the famous compensation caps in my first WIKI write-up.
There has been a whole lot of discussion in the AlephX thread long ago about this issue.
Finally, the answer came from Nelson himself and is here:
http://www.diyaudio.com/forums/showthread.php?postid=33550#post33550
/Hugo
There has been a whole lot of discussion in the AlephX thread long ago about this issue.
Finally, the answer came from Nelson himself and is here:
http://www.diyaudio.com/forums/showthread.php?postid=33550#post33550
/Hugo
Somehow the link does not lead to a statement of Nelson, so what was the conclusion ?
Here are the latest observations:
- Implemented 1nF ceramic caps at c9 /C10. Have not measured oscillations yet, but this has no influence on the strange DC-behavior.
- Removed (again) the protection transistors as requested. No change as well.
- As reported, I have
- one mosfets in the current source of side A which works hard (0,6V across 0,33R, while the other guy is lazy (0,3V)
- one mosfet in the gain of side B which is lazy (0,22V) while his colleague is busy (0,5V).
I exchanged them with each other, so that the lazy guy goes on the position of the hard worker and vice versa. Result: NOw the lazy guy becomes busy as his colleage before and the busy guy becomes lazy.
What now ? What effects can come across when mouting mosfets on heat sinks or in cabeling ? Clearly the issue is not mosfet matching. I start to have my doubts that the issue can be found on the pcb as well...
Here are the latest observations:
- Implemented 1nF ceramic caps at c9 /C10. Have not measured oscillations yet, but this has no influence on the strange DC-behavior.
- Removed (again) the protection transistors as requested. No change as well.
- As reported, I have
- one mosfets in the current source of side A which works hard (0,6V across 0,33R, while the other guy is lazy (0,3V)
- one mosfet in the gain of side B which is lazy (0,22V) while his colleague is busy (0,5V).
I exchanged them with each other, so that the lazy guy goes on the position of the hard worker and vice versa. Result: NOw the lazy guy becomes busy as his colleage before and the busy guy becomes lazy.
What now ? What effects can come across when mouting mosfets on heat sinks or in cabeling ? Clearly the issue is not mosfet matching. I start to have my doubts that the issue can be found on the pcb as well...
I swapped the source resistors in the same way as the mosfets (as descripted above). No effect. The voltages stay at the location of the mosfet. So, are there any capacitive effect which can happen between heatsink (in my case a copperbar which serves all current sources and a separate bar which serves the gain mosfets; then mounted per side on 4 huge sk109 heatsinks) and mosfet ?
All drains of the current source mosfets should have the same voltage. Same with the negative rail at the connection point of the source resistors of the gain devices.
Further, regarding the capacitance effect, this is behind my knowledge. The only thing I know is that Nelson recommended not to use too long wires. Something like 10 inches should be the maximum if I recall correctly.
/Hugo
Further, regarding the capacitance effect, this is behind my knowledge. The only thing I know is that Nelson recommended not to use too long wires. Something like 10 inches should be the maximum if I recall correctly.
/Hugo
Well...found at least one solution: The differences in the Side A in the current source are not longer there. I soldered everthing new and the problem was gone.
Did the same thing at the gainstage...including exchange of all insulators...problem persists...
As well I can bring the amp currently only a good low DC at the speaker output when I run one side much hotter than the other one...for today it is enough...are there any voodoo-masters out there who have a sufficient enough spell for these kind of issues ?
Did the same thing at the gainstage...including exchange of all insulators...problem persists...
As well I can bring the amp currently only a good low DC at the speaker output when I run one side much hotter than the other one...for today it is enough...are there any voodoo-masters out there who have a sufficient enough spell for these kind of issues ?
Blitz: did you try changing those 221-ohm gate-stopper resistors? If at some point you punched through the oxide layer in a MOSFET when it blew up, one of these may have been damaged. Then, you'd effectively have a floating gate (!) on one of the gain transistors. At the very least, you should measure them all with a DMM to make sure they're still good.
Victory !
After my immense frustration at the weekend, I rewired today everything:
- Got rid of the Teflon stranded cable, and bought solid copper 2,5mm2 cable to wire the 044n,
- 4mm2 solidcopper cable from the power supply to the board,
- re-assembled the mosfets on the heatsink so that those who are a group are pretty next to each other so that thermal drift is equalized between them,
- exchanged the grid resistors and now:
Taaaataaaaa ! The 044ns show only small differences which can come perfectly from matching.
Now we can climb the next stage...a few questions:
- Right now I have 4,1 Volt over the 392 Rs, too high ?
- I have 21,7 Volts and I see DC moving up and down for approx. 30mV, is that normal or has the power supply not enough uF ?
- Ripple of the power supply is as well 30mV, which results in 0,3mV ripple at the Speaker output is that already dead quiet or do we have to become better there ?
- What needs to be measured now to bring it to 50% Ac current gain ?
Thanks again for your help & Best Regards
After my immense frustration at the weekend, I rewired today everything:
- Got rid of the Teflon stranded cable, and bought solid copper 2,5mm2 cable to wire the 044n,
- 4mm2 solidcopper cable from the power supply to the board,
- re-assembled the mosfets on the heatsink so that those who are a group are pretty next to each other so that thermal drift is equalized between them,
- exchanged the grid resistors and now:
Taaaataaaaa ! The 044ns show only small differences which can come perfectly from matching.
Now we can climb the next stage...a few questions:
- Right now I have 4,1 Volt over the 392 Rs, too high ?
- I have 21,7 Volts and I see DC moving up and down for approx. 30mV, is that normal or has the power supply not enough uF ?
- Ripple of the power supply is as well 30mV, which results in 0,3mV ripple at the Speaker output is that already dead quiet or do we have to become better there ?
- What needs to be measured now to bring it to 50% Ac current gain ?
Thanks again for your help & Best Regards
Excellent! I wouldn't worry about slow DC fluctuations, unless they're really large. The easiest way to check if the hum is acceptable is simply to hook it up to a speaker and listen (if you haven't already done this). Now, to get your amp to 50% current gain, have a look at the procedure in the wiki under the 'Bring-up and Adjustment' section.
Some useful tips:
- Be sure to do all your final adjustments on a fully warmed-up amp. A good 30 min of idling should be sufficient. If in doubt, watch for drift in the DC parameters, and wait for them to settle. I've heard reports that opening the chassis can affect the bias points a little, due to cool air getting in. So if possible, you should try and keep the chassis fully assembled while you do the adjustments. To allow for this, you can attach thin wires and run them out through a vent hole or something, then use them as remote probe points, or a place to put the temporary trimpots for the adjustment phase. Just be careful you don't let any of these dangling wire-ends short to each other! Keep them organized, separated, and secured in place with some tape.
- Before adjusting the current gain, get all of the DC bias parameters set where you want them... adjust the bias current on each side, and get the absolute and differential DC offsets trimmed. Once you've completed the current gain adjustment, you can replace the temporary trimpot with a fixed resistor, and should never have to worry about it again, even if you have to come back for minor adjustments to the DC parameters. It should only need adjusting if you make major changes to the bias current.
Some useful tips:
- Be sure to do all your final adjustments on a fully warmed-up amp. A good 30 min of idling should be sufficient. If in doubt, watch for drift in the DC parameters, and wait for them to settle. I've heard reports that opening the chassis can affect the bias points a little, due to cool air getting in. So if possible, you should try and keep the chassis fully assembled while you do the adjustments. To allow for this, you can attach thin wires and run them out through a vent hole or something, then use them as remote probe points, or a place to put the temporary trimpots for the adjustment phase. Just be careful you don't let any of these dangling wire-ends short to each other! Keep them organized, separated, and secured in place with some tape.
- Before adjusting the current gain, get all of the DC bias parameters set where you want them... adjust the bias current on each side, and get the absolute and differential DC offsets trimmed. Once you've completed the current gain adjustment, you can replace the temporary trimpot with a fixed resistor, and should never have to worry about it again, even if you have to come back for minor adjustments to the DC parameters. It should only need adjusting if you make major changes to the bias current.
Here is the intermediate report for those who are interested: I glued the aluminum profile with rectifiers on to the main heat sink as these become hotter than the mosfets themselfs.
With this I started my longer term system test: 4 hours of operations at 22V and 6A. Nothing broke down, Heat measured at the copper bar where the mosfets are mounted to: 68 degree celsius / 154 degree fahrenheit. I guess good enough for a 120W at 6 ohm.
I have still 0,1V DC now absolute and relative. When swapping thetwo 9610 with each other, this resulta in a swap from the positive to the negative binding post and vice versa, so the origin of it is the matching of the 9610. I could get it down to 0V with the trimmers, but this would result as well into a significant mismatch in bias between the two sides as the trimmers as well change significant the bias on each side.
I can confirm Netlists experience: Zou need to match the 9610 in the actual amplifier; i did that externally first and dont have the same result now in the amp
With this I started my longer term system test: 4 hours of operations at 22V and 6A. Nothing broke down, Heat measured at the copper bar where the mosfets are mounted to: 68 degree celsius / 154 degree fahrenheit. I guess good enough for a 120W at 6 ohm.
I have still 0,1V DC now absolute and relative. When swapping thetwo 9610 with each other, this resulta in a swap from the positive to the negative binding post and vice versa, so the origin of it is the matching of the 9610. I could get it down to 0V with the trimmers, but this would result as well into a significant mismatch in bias between the two sides as the trimmers as well change significant the bias on each side.
I can confirm Netlists experience: Zou need to match the 9610 in the actual amplifier; i did that externally first and dont have the same result now in the amp
OK, 9610 matching finished. Uffff. I got 25 pcs and only 3 pairs came out. I have not yet mounted the two 9610 on a heat sink back-to-back. Currently I have (absolutely and relative) between -20mV and +40mV DC at the output, fluctuating whithin 10 minutes after 1 hour of operation (the amp is definetly hot). Is this good enough ?
Second observation: within the group of gain-mosfets I have per side two devices. The total sum of current per side (3A) is very much the same. Nevertheless the current per devices in a side is still different: I have over the 0,33 source-resistor in one extreme case 0,47V over at the first mosfets, while the other contribures with 0,53V. How important is it that these two value are identical when the DC-conditions are already OK ? Do you anticipate distortion because they run with (slightly) different biases and therefore their loadline goes in a different angle through the curves which would mean higher distortion or is this only theory ?
Second observation: within the group of gain-mosfets I have per side two devices. The total sum of current per side (3A) is very much the same. Nevertheless the current per devices in a side is still different: I have over the 0,33 source-resistor in one extreme case 0,47V over at the first mosfets, while the other contribures with 0,53V. How important is it that these two value are identical when the DC-conditions are already OK ? Do you anticipate distortion because they run with (slightly) different biases and therefore their loadline goes in a different angle through the curves which would mean higher distortion or is this only theory ?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.