Spookydd,
The THD specification for this amp by Elektor is 0.01% at 20KHz.The results
reported by ingenious are quite close, but not exactly the same. This may be
due to emitter degeneration you have employed, decreasing feedback. I share
the same opinion with you, some sort of emitter degeneration is a must, but
maybe ingenious can play with their values and optimize their values
somewhat.
An RC filter on the supplies before the VAS will be magical for decreased hum.
You can dispense with the offset adjustment altogether because with
reasonably matched pairs for the diff amps, you will not get an offset much
higher than few mV.
The tail CCS's are best filtered by 2 R's from the collectors of the controlling
transistors to the ground with their midpoint bypassed to the supplies by C's.
Ingenieus,
The current mirrors definitely have to disappear as the amplifier is non-functional as you have posted. Could you please tell me how the VAS
current is determined when the current mirrors are inserted?
Cheers
Selim
The THD specification for this amp by Elektor is 0.01% at 20KHz.The results
reported by ingenious are quite close, but not exactly the same. This may be
due to emitter degeneration you have employed, decreasing feedback. I share
the same opinion with you, some sort of emitter degeneration is a must, but
maybe ingenious can play with their values and optimize their values
somewhat.
An RC filter on the supplies before the VAS will be magical for decreased hum.
You can dispense with the offset adjustment altogether because with
reasonably matched pairs for the diff amps, you will not get an offset much
higher than few mV.
The tail CCS's are best filtered by 2 R's from the collectors of the controlling
transistors to the ground with their midpoint bypassed to the supplies by C's.
Ingenieus,
The current mirrors definitely have to disappear as the amplifier is non-functional as you have posted. Could you please tell me how the VAS
current is determined when the current mirrors are inserted?
Cheers
Selim
Excellent! I chose the values rather arbitrarily, so I expect there is a need for some tweaking.
Would that be on the rail, such as between for example R23 and R25 on my latest schematic? With a diode before it?
There is no adjustment method provided, and I didn't think we'd need it, if we're matching as many transistors as we can.
Are you referring to those added (maybe) current sources in the diff amps?
Spookydd,
Break the supply line between R69 (680R) and R67 (100R). Insert an R (22R
to 47R) at that position. Connect a C from the junction of the new R and R67
to the ground. The bigger the C (within reason) the better. I usually go for
something in the 1000uF range. Do the same for the (-) supply.
Replace R65 (8K2) with 4 3K3. (Let's call them R651, R652, R653 and R654)
Connect the junction of R652 and R653 to the ground. Bypass the junction of
R651 and R652 to the (+) rail and the junction of R653 and R654 to the (-)
rail. The C's in this position are again 'the bigger the better' values within
reason. 220 uF will happily do.
Sorry about offset adjustment.That was in the original Elektor circuit, not
yours.
Cheers
Selim
Break the supply line between R69 (680R) and R67 (100R). Insert an R (22R
to 47R) at that position. Connect a C from the junction of the new R and R67
to the ground. The bigger the C (within reason) the better. I usually go for
something in the 1000uF range. Do the same for the (-) supply.
Replace R65 (8K2) with 4 3K3. (Let's call them R651, R652, R653 and R654)
Connect the junction of R652 and R653 to the ground. Bypass the junction of
R651 and R652 to the (+) rail and the junction of R653 and R654 to the (-)
rail. The C's in this position are again 'the bigger the better' values within
reason. 220 uF will happily do.
Sorry about offset adjustment.That was in the original Elektor circuit, not
yours.
Cheers
Selim
That's what I was thinking and also adding a diode there.
The main reason for that diode would be to prevent some of the plop at shut down, because it would keep the front end going a bit longer when the rails have disappeared. Is that right?
Oh boy, those plus the 2 others on the front end are eating up pcb real estate. It's getting tight on that 200x200mm board now.
I switched the ordinary 3W emitter resistors on the outputs to the non inductive MPC78. Those are 2W but they should do fine there. They're rather rugged and take have a smaller footprint at the same time, which helps now with the lack of room.
We're not even done yet with the bridging. There will be more parts to fit to make it work, hopefully not too many.
I was thinking earlier that I would probably have a little space left to add a speaker plug on the pcb, but that's getting less certain, unless I increase the board size.
I'm wondering though, since it's going to be a grounded bridge and the ground will be floating between the rails, if having added those filters tied to the ground won't be an issue.
Let's take for example a rise in the signal and the gnd is being lifted towards the + rail by the low side. Then what happens with the energy stored in those caps? Having the floor coming up below them, they would probably want to dump their load somewhere before being squashed to the ceiling, if you know what I mean...
How is this going to work?
Changes made, new sch attached:
Attachments
It would try to act as a boot strap. In the QSC "flying rail" amps, this is how the +/-15V for the opamps is maintained. It will increase the dissipation in the resistors, and move the rail voltage to the diff pairs. You don't want this to happen here unless you want to use very high voltage devices in the input stage! For filtering purposes, do not tie them to ground. Tie them to each other. Or, since it becomes two caps in series, use a single cap.
For the overall feedback connection, study the MA2400 schematic. The low side is inverting ad fed from the high side, with a gain less than 1. The + input (normally virtual ground in a normal inverting amp) is referenced to the mid point between the rails with a divider, and further divided down to half the low side's gain. See the network labeled "N102". The resistors need to be well matched. Since the power transistors need almost no voltage swing (with respect to ground) a simple op amp can be used to drive the output stage directly and let the rails fly just like the QSC. The feedback/divider network keeps thsi happening in just the right proportion. I see no reason why a full amplifier cannot be used other than complexity.
Things are beginning to take a better shape now. For this particular
application, as wg_ski has pointed out, it is best to use a single cap
for filtering purposes. Applying same logic to the input stage CCS's,
omitting R652 and R653 and replacing them with a single 6K8 R, thus
breaking the ground connection is a better idea.
Taking a second look at the output stages now, it is better to place R72
between the base and emitter of T16, instead of the (-) rail. This way,
symmetry between upper and lower halves will be preserved.
Also, the emitter resistors R82, R84 and R86 for the lower 3055's should
literally move to the emitters, not the collectors.
The bias pot passes all the VAS current. I have always been skeptical about
currents passing through vipers of the pots. Do you think it is a better idea
to use parallel connected fixed R's in this position and soldering them in
as needed?
Cheers
Selim
application, as wg_ski has pointed out, it is best to use a single cap
for filtering purposes. Applying same logic to the input stage CCS's,
omitting R652 and R653 and replacing them with a single 6K8 R, thus
breaking the ground connection is a better idea.
Taking a second look at the output stages now, it is better to place R72
between the base and emitter of T16, instead of the (-) rail. This way,
symmetry between upper and lower halves will be preserved.
Also, the emitter resistors R82, R84 and R86 for the lower 3055's should
literally move to the emitters, not the collectors.
The bias pot passes all the VAS current. I have always been skeptical about
currents passing through vipers of the pots. Do you think it is a better idea
to use parallel connected fixed R's in this position and soldering them in
as needed?
Cheers
Selim
Spookydd,
Before you embark upon your big project, if you would like to test how this
configuration sounds, I have a few extra PCB's for a stereo version that I can
gladly send to you. This offer is open to anybody who is interested. I have
about 20 PCB's that I can gladly part with for the benefit of fellow DIY'ers.
Just drop me a PM at selim_ardali at yahoo dot com including your postal
address.
Happy weekend everybody
Cheers
Selim
Before you embark upon your big project, if you would like to test how this
configuration sounds, I have a few extra PCB's for a stereo version that I can
gladly send to you. This offer is open to anybody who is interested. I have
about 20 PCB's that I can gladly part with for the benefit of fellow DIY'ers.
Just drop me a PM at selim_ardali at yahoo dot com including your postal
address.
Happy weekend everybody
Cheers
Selim
Alright then, updated sch attached:
That was one of the points I asked for a few times, I moved it now.
One other detail I was also wondering about, and it's true they don't fill their purpose if they're on the collectors.
Updated as well.
Now that's an other thing that bothered me. This bias setup seems a little weird and I have never seen one like this on any other amp before. The main thing is that in case the wiper fails to make contact, we end up with a wide open spreader and max bias, which I think is rather risky and probably not so reliable in the long run.
I'm not in favor of soldering resistors to adjust bias, it's not easy, especially in such a tight layout and it's messy.
Can we come up with something better and sensible, but with a knob to turn instead of soldering repeatedly?
Attachments
I'm interested. I have parts, so that can be a good experiment.
What is it based on? the actual elektor circuit we're using as a base?
It is the actual Elektor circuit but with some tweaks. First, the outputs are
configured to utilize both NPN and PNP parts. Second, the offset adjustment
components are omitted. Lastly, there are few caps added here and there,
mostly as RC filters to the reference voltages and early stage power supplies.
Signal caps are composed of parallel but antiphase connected electrolytics
in parallel with 220n. The bias setting pot is in parallel with a fixed resistor.
RSK has just finished this amp and he omitted the variable resistor altogether,
And he is very happy about the stability of the bias current.
If you already have the parts, it will be an inexpensive experiment for you
to decide if this configuration suits your taste.
Cheers
Selim
configured to utilize both NPN and PNP parts. Second, the offset adjustment
components are omitted. Lastly, there are few caps added here and there,
mostly as RC filters to the reference voltages and early stage power supplies.
Signal caps are composed of parallel but antiphase connected electrolytics
in parallel with 220n. The bias setting pot is in parallel with a fixed resistor.
RSK has just finished this amp and he omitted the variable resistor altogether,
And he is very happy about the stability of the bias current.
If you already have the parts, it will be an inexpensive experiment for you
to decide if this configuration suits your taste.
Cheers
Selim
Ok, so this should be somewhat close to what we're doing now. The fact that you switched from quasi to complementary shouldn't make that much difference for our needs.
Of course it will be stable that way, but less easy to adjust at first.
That pot is of low value and we have about 8mA of current going through there, so the dissipation isn't the big concern, however the reliability of the wiper is what concerns me.
On the new layout, I used a sfernice/vishay T93YB multi-turn cermet trimmer, which should be fairly reliable, but still I would favor something less risky and definitely convenient to adjust.
I have most of the parts, at least the most important ones, so I won't need to go hunting for much, except for a heatsink.
I have several ordinary transformers, non toroids, that can do fine for experimenting.
One thing I forgot to mention in my last posting with the updated schematics, is what to do about the extra decoupling caps that I had added for each power transistor.
I was placing those as close as possible to each power pin, which is not always so easy with the TO3s being on the heatsink.
Now that the V- side emitter resistors have been moved, I question the use of those decoupling caps, so they're not hooked up on the sch right now.
What would be best? Not using them?
I was placing those as close as possible to each power pin, which is not always so easy with the TO3s being on the heatsink.
Now that the V- side emitter resistors have been moved, I question the use of those decoupling caps, so they're not hooked up on the sch right now.
What would be best? Not using them?
elektor's article
I was wondering if someone would have the original full article from elektor in a readable condition (not all yellow and blurry).
I have my elektor collection that goes way back to the late 70s, but in those days I missed quite a few issues and I couldn't find the one with that article.
I was wondering if someone would have the original full article from elektor in a readable condition (not all yellow and blurry).
I have my elektor collection that goes way back to the late 70s, but in those days I missed quite a few issues and I couldn't find the one with that article.
Some more work is indeed needed. With the current mirror in its simple form (as in my first post) it is not possible to calculate the VAS quiescent current from the components and currents shown. Something is needed to establish a more reliable VAS bias current. This should put the output voltage of the current mirror at a defined level when the LTP is in balance.
An approach to solving this problem is presented in Bob Cordell's book. You will notice that each current mirror has an added helper transistor. This is an emitter follower that supplies the base current for the current mirror transistors. intead of it being drained from the incoming current. This improves the DC balance of the current mirror and decreases the influence of transistor hfe. So far, so good. I have kept the resistor values from the book because it seems to work, BTW.
The helper transistor also separates the voltage at the current mirror input by one additional Vbe from the rail. This voltage is now 2Vbe plus the drop across the emitter resistor away from the rail. This is where some change is called for. It is easier to get the same voltage drop from the rail in the VAS by changing to a Darlington transistor. I have made this change in the schematic below.
The resistor connected across the collectors of of the current mirror solves the rest of the problem. When the LTP is in balance, no current flows through it as the voltage on both sides is the same. The voltage at the emitter of the VAS transistor then becomes more or less the same as the voltages at the emitters of the current mirror transistors, which are set by the LTP tail current. The quiescent current of the VAS is now established so that it depends directly on the tail current in the LTPs.
Don't worry too much about the DC offset on the diff amp input. They all do that. As long as they are more or less equal, all is well. This is why there is a 100k resistor on the input to match the 100k in feedback path, BTW. The capacitor on the input keeps the DC voltage away from your preamp or source. On the other side of the LTP, the voltage drop across the feedback resistor keeps the DC away from the output.
It might look like changing the compensation capacitor will have a drastic effect on frequency tresponse, but things are not that simple. There is quite an involved relationship between LTP tail currents, LTP emitter degeneration resistor values and compensation capacitor values. Seeking a good set of values is quite a process. You have too much THD at high frequencies on the one end an instability on the other end. The -3dB frequency is ~100kHz. Everything seems OK but then again I have only checked it against a purely resistive output impedance.
I have changed all the small signal transistors to BC639 and BC640. The capacitor in the feedback network can be made much bigger. This helps to improve THD at low frequencies. Don't worry about the size as it doesn't need to be a high voltage type. It can be as low as 6V but adding diode protection in parallel is a good idea.
With the changes made, THD-1k is 0.0021% and THD-20k is 0.038%. The big drop in THD-1k can be attributed to the Darlington pair VAS. THD-20k is probably now as good as it will ever be without changing the output transistors. With the current mirrors and Darlingtons added, this thing is starting to look more and more like a Blameless...
But let me not detract from the circuit that is under construction. It looks like a perfectly good circuit and should give good service. Look at the suggested enhancements as a way of clarifying my initial remark that current mirrors make a big difference.
Attachments
Last edited:
There is one thing I don't quite understand about the value of that bias pot.
Having a current of about 8mA and the 2 BC547s causing 2 junctions worth of drop, we have 4 junctions plus a little bit from the 0.22ohms to catch up on so we can "spread" those legs for biasing. So I'm thinking we need something around 1.4 or 1.5V on that pot to have a spread that causes a bias current to flow (more or less).
Then with 8mA, that means a value of much more than the 100ohms we have on the sch. How does this work then? What am I missing?
Having a current of about 8mA and the 2 BC547s causing 2 junctions worth of drop, we have 4 junctions plus a little bit from the 0.22ohms to catch up on so we can "spread" those legs for biasing. So I'm thinking we need something around 1.4 or 1.5V on that pot to have a spread that causes a bias current to flow (more or less).
Then with 8mA, that means a value of much more than the 100ohms we have on the sch. How does this work then? What am I missing?
You are forgetting that not all of the 100 ohm will be in use. It is an adjustable resistance after all. My simulation suggest about 20% of that 100 ohm, i.e. 20 ohm * 8 mA = 160mV of voltage drop across the pot. Add it to the ~650mV across each diode-connected transistor and you are in the ballpark of 1.5V.
The bias network is OK, but a Vbe multiplier would work better.
The bias network is OK, but a Vbe multiplier would work better.
spookydd,
The current through the BC639 and 640 is in the order of 1 mA. They have 100
R resistors connected which make for 600 + 100 + 100 + 600 mV = 1.4 V
total bias voltage + whatever you get from the emitter R's of the outputs.
Let's say 1.5 V total. the 2 547s, passing 8 mA, will have about 650 mV drop,
so it makes 1.3 V. The remaining 200 mV is catered for by the variable R,
in this case about 25 R should do. In my PCB, I have a BC550 and a BC560
instead of 2 BC547's and a 22 R fixed resistor in parallel with the pot.
Somehow, even that 22 R seems to be too big a value. Must be due to
Vbe spread between the parts that I use. But RSK has experienced the same
thing and he has installed a 5R6 fixed resistor instead of the pot.
ingenious,
Thanks very much for your efforts. I agree with you that this is as much as
it gets using these components. More complication will not lead to better
performance. If you can kindly comment on spookydd's last circuit and maybe
optimize the values of the degeneration R's for the input pairs, that will be
wonderful.
Cheers
Selim
The current through the BC639 and 640 is in the order of 1 mA. They have 100
R resistors connected which make for 600 + 100 + 100 + 600 mV = 1.4 V
total bias voltage + whatever you get from the emitter R's of the outputs.
Let's say 1.5 V total. the 2 547s, passing 8 mA, will have about 650 mV drop,
so it makes 1.3 V. The remaining 200 mV is catered for by the variable R,
in this case about 25 R should do. In my PCB, I have a BC550 and a BC560
instead of 2 BC547's and a 22 R fixed resistor in parallel with the pot.
Somehow, even that 22 R seems to be too big a value. Must be due to
Vbe spread between the parts that I use. But RSK has experienced the same
thing and he has installed a 5R6 fixed resistor instead of the pot.
ingenious,
Thanks very much for your efforts. I agree with you that this is as much as
it gets using these components. More complication will not lead to better
performance. If you can kindly comment on spookydd's last circuit and maybe
optimize the values of the degeneration R's for the input pairs, that will be
wonderful.
Cheers
Selim
I'm usually not in favor of having such caps there, because they're rarely actually needed. Most preamps will have one on their output, so we're double dipping.
I usually use a mix table to drive my amps, and output circuits on the mix table has caps.
This is not bad at all for 3055s.
I was thinking of using tantalum caps, but they're polarized. However they have a compact size for their value. I just looked in my junk and I have tantalums of 330u / 6V that are no bigger than about 7mm in diameter, for about 17mm in height. This is easily fitted in a tight space.
This board is getting quite crowded, as I do some layout while we tweak the schematic. That board is 200x200mm, and it wouldn't be a luxury to go bigger, but the heatsink more or less fixed the length. I won't go wider unless we really need it. It's big enough already as it is, the cost at the board house would climb. But let's not forget, that single board has everything on it, the 2 amps and psu...
I think it looks great, considering we're using 60s trannies, it would've been considered fairly high end back then.
Those THDs are very good I think, the THD is not as harsh on the ears as TIM.
This brings me to wonder about how we calculate the output impedance and damping factor...
But let me not detract from the circuit that is under construction. It looks like a perfectly good circuit and should give good service. Look at the suggested enhancements as a way of clarifying my initial remark that current mirrors make a big difference.
If those improvements can work and really improve without causing any issues we didn't have before, then we can do that.
Extra trannies, of they're the small signal types, don't take up that much space on the pcb and are cheap.
I don't have any BC639/640 on hand though, so I may have to test with other types.
btw: how did the change from the TIPs to the MJEs in the driver stage affect the amp?