Analog_guy/Ben, here is the link to my amp that uses a symmetrical vas and cascodes.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=89380
I have since changed the vas, and drive the cascodes via a darlington rather than a single transistors. This gives about 20dB more OLG and I get close to that in simulated distortion reduction at 20Khz.
It was on this amp (spider wired) that I tried the fully balanced differential topology driving the cascodes and got the latch-up. There are some proposals earlier in the thread to use a potentiameter to bring the vas into balance on the fully differntial topology, but I have reservations about this approach wrt to temperature changes, load changes etc. I'll have to think about th e problem some more to come up with a foolproof apporach - ideally, no pots.
With the present topology on my amp, there are no problems and the amp sounds very good (I'm driving B&W 703's). My Tokyo neighbours are not impressed, but what the hell.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=89380
I have since changed the vas, and drive the cascodes via a darlington rather than a single transistors. This gives about 20dB more OLG and I get close to that in simulated distortion reduction at 20Khz.
It was on this amp (spider wired) that I tried the fully balanced differential topology driving the cascodes and got the latch-up. There are some proposals earlier in the thread to use a potentiameter to bring the vas into balance on the fully differntial topology, but I have reservations about this approach wrt to temperature changes, load changes etc. I'll have to think about th e problem some more to come up with a foolproof apporach - ideally, no pots.
With the present topology on my amp, there are no problems and the amp sounds very good (I'm driving B&W 703's). My Tokyo neighbours are not impressed, but what the hell.
Hi Suzy
If you are considering a second shot at getting your 50 W board working well, I reckon it might be a good experiment to try soldering in a pair of these (dead bug style) to replace your current individual SOT-23 devices.
They’re BC857BS (PNP) and BC846BS (NPN) dual transistor pairs with tightly matched Hfe, housed in a 6 lead version of the SOT-23 package. Being duals, each transistor shares a common chip, so they are tightly thermally bonded – and they cost next to nothing!
http://au.farnell.com/jsp/Semiconductors/Transistors/PHILIPS/BC847BS/displayProduct.jsp?sku=1081234
http://au.farnell.com/jsp/search/productdetail.jsp?sku=1081247
Cheers,
Glen
If you are considering a second shot at getting your 50 W board working well, I reckon it might be a good experiment to try soldering in a pair of these (dead bug style) to replace your current individual SOT-23 devices.
They’re BC857BS (PNP) and BC846BS (NPN) dual transistor pairs with tightly matched Hfe, housed in a 6 lead version of the SOT-23 package. Being duals, each transistor shares a common chip, so they are tightly thermally bonded – and they cost next to nothing!
http://au.farnell.com/jsp/Semiconductors/Transistors/PHILIPS/BC847BS/displayProduct.jsp?sku=1081234
http://au.farnell.com/jsp/search/productdetail.jsp?sku=1081247
Cheers,
Glen
Actually I'm having a fair degree of success redoing the layout using the MJE340/350 devices, as used in the 100W version. I started by lining all ten transistors in two vertical rows of five, such that they can be bolted to a short strip of metal, and all end up at the same temperature. I took the large electrolytics off the board, and moved a whole pile of the SMD bits to the bottom of the board. The original had almost all the parts on the top.
The result (so far!) is a titchy board just 3.2 by 2.1 inches. Component heights are 0.5 inches on the top side, and 0.2 inches on the bottom (for a total module thickness of 0.8 inches or so):
As you can see I'm still working on it. There's a fair few nets still to be routed, plus pouring planes etc. I'm actually kinda enthused by this one, as it really does end up pretty small. FWIW, the original was 4.4 by 3.4 inches, and (due to the large electros) 1.4 inches thick.
Just the ticket for bi/tri amping, methinks.
The result (so far!) is a titchy board just 3.2 by 2.1 inches. Component heights are 0.5 inches on the top side, and 0.2 inches on the bottom (for a total module thickness of 0.8 inches or so):
An externally hosted image should be here but it was not working when we last tested it.
As you can see I'm still working on it. There's a fair few nets still to be routed, plus pouring planes etc. I'm actually kinda enthused by this one, as it really does end up pretty small. FWIW, the original was 4.4 by 3.4 inches, and (due to the large electros) 1.4 inches thick.
Just the ticket for bi/tri amping, methinks.
I suggested the larger devices and heatsinking as a way to test your theory, as I'm not convinced that the problem is a thermal issue. However, you don't seem to offer all the results of your debugging so it is hard to know exactly what is going on. I have a feeling you do. I have to wonder if the smaller and faster devices have introduced an oscillation, but I'd guess that you looked for this already.
It is a complex design, overly complex in my opinon but that is just my opinion.
The second dual comp-diff stage, and the rest of the design look like a more conventional amp. It looks as if the FET front end was just tacked on, if you know what I mean.
Seems it would make sense to contrast this design with Bob Cordell's classic.
Just my opinion, I'm looking at the old Tiger amps just because I think they have potential, and I'd like to fully understand the instabilities so I understand being interested in the classics.
Pete B.
It is a complex design, overly complex in my opinon but that is just my opinion.
The second dual comp-diff stage, and the rest of the design look like a more conventional amp. It looks as if the FET front end was just tacked on, if you know what I mean.
Seems it would make sense to contrast this design with Bob Cordell's classic.
Just my opinion, I'm looking at the old Tiger amps just because I think they have potential, and I'd like to fully understand the instabilities so I understand being interested in the classics.
Pete B.
suzyj said:
Suzy,
I have had a bit better look at your schematic.
The circuit appears to be highly sensitive to VAS thermal VBE
modulation and 1st stage CM variation.
Starting at IP dif pair, you are running each device at about 0.7mA.
This, through the 1k2 collector load R's, will set the VAS current at
approx 1.5mA / side.
This is way too low for the size of R33 and R34.
Consider a thermally induced VBE change of just 0.05V in the VAS
transistor(s) which will result in a huge change in that small 1.5mA
/device.
It's worth trying the following:
- Delete ground conn between R17 / R18. This will allow dif pairs
to track more closely
- Run IP dif pair at 2.0mA / device.
- Change R27/28/29/30 to 860 ohms.
- Increase R33/34 to75R
- Increase R51/52/53/54 to 20 or 30R
- Thermally join (or keep very close) VAS pairs.
This should set up VAS at around 6mA / device which, in conjunction
with added degeneration, will keep them in a much more linear operating region.
cheers
Terry
Looking at this again, it seems fairly clear that, roughly, D3 should be thermally coupled to Q9, and D1 and D2 to Q17 and Q18, for example. Similarly with the other side.
I agree that the design is too sensitive to thermal drift and component matching in general. I would not be surprised if the design works well when the diff pair devices (Q15-16, Q17-18)come from the same lot, or wafer, and very poorly when not. A prototype does not prove out a design unless mismatched semis are deliberately used to represent the expected allowable spread in tolerance, when selected/matched devices are not used. Simulation and/or analysis should be done with mismatched devices representing the allowable spread in the design. Really, we should step back and consider if the diff amp VAS makes sense? I think not since the design cannot afford significant drop in the emitter path to provide a good current source, and it is far too sensitive to device matching as Terry pointed out.
IMO, when diff pairs are used an effort should be made to have them come from the same lot in any critical application. I also wonder why dual devices are not used more often in diff pair designs.
Harman Kardon has been using duals since the 1970s with their old Citation 12, makes a lot of sense. There's really no excuse not to use duals, when a suitable one is available, in high performance designs such as these.
Pete B.
I agree that the design is too sensitive to thermal drift and component matching in general. I would not be surprised if the design works well when the diff pair devices (Q15-16, Q17-18)come from the same lot, or wafer, and very poorly when not. A prototype does not prove out a design unless mismatched semis are deliberately used to represent the expected allowable spread in tolerance, when selected/matched devices are not used. Simulation and/or analysis should be done with mismatched devices representing the allowable spread in the design. Really, we should step back and consider if the diff amp VAS makes sense? I think not since the design cannot afford significant drop in the emitter path to provide a good current source, and it is far too sensitive to device matching as Terry pointed out.
IMO, when diff pairs are used an effort should be made to have them come from the same lot in any critical application. I also wonder why dual devices are not used more often in diff pair designs.
Harman Kardon has been using duals since the 1970s with their old Citation 12, makes a lot of sense. There's really no excuse not to use duals, when a suitable one is available, in high performance designs such as these.
Pete B.
suzyj said:
Looks good.
I think that some good points have been made with regards to Ic balance, Vbe variations etc and it will be interesting to see if you need to do any device matching / tail/emitter degeneration resistor tweaking to tune Ic imballances in order to match the THD performance of your 100W version.
If I can make another BJT recommendation, these are complementary TO-126 devices with a specified fT of 150MHz and only a few pF of Cob. They would make an excellent replacement for the MJE340/MJE350’s in your amp. The spec summary on the Farnell site is wrong WRT the Pdiss rating, they are actually 10W devices.
http://au.farnell.com/jsp/Semiconductors/Transistors/SANYO/2SA1209/displayProduct.jsp?sku=1208641
http://au.farnell.com/jsp/search/br...nsearch_001&Ntt=2sc2911&Ntx=&_requestid=77966
Cheers,
Glen
syn08 said:
Kiwi land?? I think you mean Skippy land
And yeah I know, Fennel is a jip. I haven't really looked for a cheaper supply yet, but I probably should since I'm going to need a few for my current project.
suzyj said:
Dunno. I haven't looked for one though. My guess though is probably not.
I'll try Google.
Cheers,
Glen
Hey, does anyone out there have a datasheet for the Fairchild version of these devices?. I have tried the Fairchild website, but they don't seem to list/manufacture them anymore and google drew a blank.
The Sanyo datasheet has a really stoopid/useless 1W SOA graph - odd for a 10W transistor.
Cheers,
Glen
The Sanyo datasheet has a really stoopid/useless 1W SOA graph - odd for a 10W transistor.
Cheers,
Glen
AndrewT said:
Yep, you are right. Not enough coffee in my veins this morning. It's though strange that such a low Cob is not correlated with a much higher Ft.
For VAS, pre-driver and EC diff amp applications I'm now using with excellent results 2SA1407/2SC3601 which, while having the Cob in the same range as 2SA1209, are spec'd at Ft=400MHz. 2SA1407/2SC3601 are expensive though, I was able to get them from http://tenfourltd.com/ for about $3/pair is 50's quantity.
G.Kleinschmidt said:
Glen:
I've noticed that Mouser stocks and sells (for 25 cents in 10's quantities) the Fairchild KSA1381/KSC3503. These are some decent video amp devices, with Cob under 3pf. You may want to give them a shot. According to my measurements they are not quite at the performance of 2SA1407/2SC3601 but, from a price/performance perspective, they are still excellent devices for a high performance VAS.
Datasheets are available on the Fairchild web site.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.