There's no issue with using either of the three pairs; only difference will be the bias setting due to the different gains. I suggest the MJ15030 be used if possible for the bias transistor for the chosen resistor and trimpot values; but of course changing these will make the other ones just as fine. There shouldn't be any audible difference.
MJ21194/21193 is a good idea for a prototype since their lower speed makes them less prone to oscillation issues due to long wiring.
BTW, has anybody ever done side-by-side tests of the various output transistors, not necessarily on a KSA? Lots of talk over the different ones, and the only definite is that the MJ15003/4 is outclassed by all the modern ones. The ones I'm interested in however is MJ(L)21194/3, 2SC3281/2SA1302 and their Toshiba successors, MJL4281/MJL4302, and some of those lovely Sanken things.
Same with small-signal transistors. Besides speed and other ratings, has anybody ever listened to different ones? MPA06/56, MPS8099/8599, MPSA42/92, 2SC2240/2SA970, BC546/556, 2Nxxxx, some of the newer Japanese ones etc.? I remember early on the KSA50 forum that somebody mentioned he likes the sound of MPSA06/56 but if that's your cup of tea then the 8099/8599 is a better choice. Spec-wise I personally like the BC546/556 ones, and they're cheap and easily available.. in some countries the Japanese stuff are hard to get.
MJ21194/21193 is a good idea for a prototype since their lower speed makes them less prone to oscillation issues due to long wiring.
BTW, has anybody ever done side-by-side tests of the various output transistors, not necessarily on a KSA? Lots of talk over the different ones, and the only definite is that the MJ15003/4 is outclassed by all the modern ones. The ones I'm interested in however is MJ(L)21194/3, 2SC3281/2SA1302 and their Toshiba successors, MJL4281/MJL4302, and some of those lovely Sanken things.
Same with small-signal transistors. Besides speed and other ratings, has anybody ever listened to different ones? MPA06/56, MPS8099/8599, MPSA42/92, 2SC2240/2SA970, BC546/556, 2Nxxxx, some of the newer Japanese ones etc.? I remember early on the KSA50 forum that somebody mentioned he likes the sound of MPSA06/56 but if that's your cup of tea then the 8099/8599 is a better choice. Spec-wise I personally like the BC546/556 ones, and they're cheap and easily available.. in some countries the Japanese stuff are hard to get.
Matching of output transistors
Hi
Has anybody ever heard a difference between using matched output transistors and using non-matched transistors?
Does it really make a difference?
Mark- I think most of us cannot wait to see how the new boards look like aor how they perform once build up.
Jozua
Hi
Has anybody ever heard a difference between using matched output transistors and using non-matched transistors?
Does it really make a difference?
Mark- I think most of us cannot wait to see how the new boards look like aor how they perform once build up.
Jozua
Matching output devices is close to impossible.
I strictly use the same batch and only check for bad ones.
Even within the same batch code there's an ocassional rotten critter.
I've fumbled a lot of line and headphone amp circuits, tried a bunch of BC types, MPSAs, 2N5s, and Jappon devices.
Stuff like BC414/416, BC550/560, BC545/556, BC547/557 are still very nice for the money.
No such thing as the perfect high voltage small signal device, MPSAs are amazing for their voltage capability but at lower voltage levels there are choices that are better soundwise and don't cost anything extra.
Same goes for high power devices, imo.
I've only tried Motorola TO3s once, just a pair in parallel, sounded kind of strained.
(Motorola TO3s were pretty expensive overhere in the previous century)
Trick is to use plenty of those in parallel for the familiar effortless sound, like the JRs, MLs, Krells and Threshold had and did.
I still have documentation that refers to the OnSemis(Motorolas) as Super High Ft devices
Compared to the crappy 1-2 MHz power devices of yesteryear they were.
The high voltage routine goes for products of the same brand as well: if i don't need to turn to high voltage 2SA1295 Sanken devices it is better to use a lower voltage 2SA1216.
And sound with a 2SA1215 is better overall than 2SA1216, downfall is the lower power rating.
I don't have enough MJE15034/35 to do any selecting, as i do have of the 32/33.
Swapping 34/35s in place of the lower voltage types was not an improvement. The difference in the datasheets between the 32 and 34 seems to back that up. Don't think i'll ever go for 34/35s, unless i start using them as power devices instead of drivers.
I strictly use the same batch and only check for bad ones.
Even within the same batch code there's an ocassional rotten critter.
I've fumbled a lot of line and headphone amp circuits, tried a bunch of BC types, MPSAs, 2N5s, and Jappon devices.
Stuff like BC414/416, BC550/560, BC545/556, BC547/557 are still very nice for the money.
No such thing as the perfect high voltage small signal device, MPSAs are amazing for their voltage capability but at lower voltage levels there are choices that are better soundwise and don't cost anything extra.
Same goes for high power devices, imo.
I've only tried Motorola TO3s once, just a pair in parallel, sounded kind of strained.
(Motorola TO3s were pretty expensive overhere in the previous century)
Trick is to use plenty of those in parallel for the familiar effortless sound, like the JRs, MLs, Krells and Threshold had and did.
I still have documentation that refers to the OnSemis(Motorolas) as Super High Ft devices
Compared to the crappy 1-2 MHz power devices of yesteryear they were. The high voltage routine goes for products of the same brand as well: if i don't need to turn to high voltage 2SA1295 Sanken devices it is better to use a lower voltage 2SA1216.
And sound with a 2SA1215 is better overall than 2SA1216, downfall is the lower power rating.
I don't have enough MJE15034/35 to do any selecting, as i do have of the 32/33.
Swapping 34/35s in place of the lower voltage types was not an improvement. The difference in the datasheets between the 32 and 34 seems to back that up. Don't think i'll ever go for 34/35s, unless i start using them as power devices instead of drivers.
Thank you Jacco, fully agreed. The few I tried to match didn't yield consistent results anyway.
Besides, since we need so many output devices it's very likely that we'll get a reel from the same manufactured batch that's often more than adequately matched.
As I understand the prototype boards will be bare ones without soldermask or silkscreen and with thin copper. Thus not much to look at, but more than sufficient to test whether it works, what heatsink sizes are necessary and the results of different transistors and FET's. So far Mark, Flodstroem and I will each build one of these boards before the actual Group Buy ones are manufactured.
Besides, since we need so many output devices it's very likely that we'll get a reel from the same manufactured batch that's often more than adequately matched.
As I understand the prototype boards will be bare ones without soldermask or silkscreen and with thin copper. Thus not much to look at, but more than sufficient to test whether it works, what heatsink sizes are necessary and the results of different transistors and FET's. So far Mark, Flodstroem and I will each build one of these boards before the actual Group Buy ones are manufactured.
In a class A power amp I would think that they should just be matched so all share the same amount of current. The circuit for matching like that is quite simple. I have a Tektronix 575a curve tracer if someone here really wants to get into matching devices. I've not had much tome to learn how to use it properly but it is functioning 100%.
Mark
matching output devices...
...could have benefits, but don't forget to include the surrounding components in the matching efforts...
No point in having transistors that are within 1% if the emitter resistors are 5%. Current sharing in the output stage is encouraged through increasing the emitter resistors, not statically matching the transistors and definitely not adding base resistors.
My 2c
Stuart
...could have benefits, but don't forget to include the surrounding components in the matching efforts...
No point in having transistors that are within 1% if the emitter resistors are 5%. Current sharing in the output stage is encouraged through increasing the emitter resistors, not statically matching the transistors and definitely not adding base resistors.
My 2c
Stuart
Hi,
measure gain of all your output devices @ Iq.
Group them into high and low.
Use the low PNP and NPN groups in one channel and the high groups in the other channel.
A couple of extra devices will allow one to discard the oddballs if they exist.
Definitely match the emitter resistors within a group and preferably within an amp stage. Go for 1% or better in your matching.
I set up a constant current source of about 200 to 300mA and string the power resistors in series (two terminal blocks allow ten resistors to be measured at the same current) and mark them accordingly. I assume one as my reference and calibrate ALL the others to it (a DMM set to 200mVdc can achieve matching to better than 0.1% if you set your current to drop near 150mV across the calibration resistor).
measure gain of all your output devices @ Iq.
Group them into high and low.
Use the low PNP and NPN groups in one channel and the high groups in the other channel.
A couple of extra devices will allow one to discard the oddballs if they exist.
Definitely match the emitter resistors within a group and preferably within an amp stage. Go for 1% or better in your matching.
I set up a constant current source of about 200 to 300mA and string the power resistors in series (two terminal blocks allow ten resistors to be measured at the same current) and mark them accordingly. I assume one as my reference and calibrate ALL the others to it (a DMM set to 200mVdc can achieve matching to better than 0.1% if you set your current to drop near 150mV across the calibration resistor).
Well the original used 1ohm resistors which is more than enough to take care of the current sharing. I too have a curve tracer (and an expensive one at that), and found that the ONSemi stuff was close enough anyway. In a large batch there may be a few oddballs but that's much more dominant (and critical) with the small-signal stuff.
AndrewT is definitely correct that if any attempt at matching is to be made it must be at Iq or at least close to it. Matching at e.g. 10mA in this application will not tell much. If the AC current swing is large compared to the bias current, this is difficult since the parameters change according to current. Therefore high-power class-AB amps at low bias have much more to gain than Class-A amps by matching, and since very few people building these (e.g. subwoofer or PA amps) have current imbalance issues, its negligible except in extreme cases. Buy different transistors from different vendors and use the most shaky resistors you can find at the lowest resistance possible and MAYBE there will be mentionable imbalance.
Besides, I used non-matched MJ21193/MJ21194's (same reel) in a KSA50 clone with 0.5ohm 5% non-matched emitter resistors, and the bias in the final circuit was all withing 1mA. At typical bias levels that is less than 1%. Still I suppose matching the emitter resistors to 1% is a good idea and it's easy enough.
AndrewT is definitely correct that if any attempt at matching is to be made it must be at Iq or at least close to it. Matching at e.g. 10mA in this application will not tell much. If the AC current swing is large compared to the bias current, this is difficult since the parameters change according to current. Therefore high-power class-AB amps at low bias have much more to gain than Class-A amps by matching, and since very few people building these (e.g. subwoofer or PA amps) have current imbalance issues, its negligible except in extreme cases. Buy different transistors from different vendors and use the most shaky resistors you can find at the lowest resistance possible and MAYBE there will be mentionable imbalance.
Besides, I used non-matched MJ21193/MJ21194's (same reel) in a KSA50 clone with 0.5ohm 5% non-matched emitter resistors, and the bias in the final circuit was all withing 1mA. At typical bias levels that is less than 1%. Still I suppose matching the emitter resistors to 1% is a good idea and it's easy enough.
Hi,
the easy way to measure Iq is by measuring the voltage drop across the emitter resistor.
If the emitter resistors are +-5% then there must be another method to identify bias matching within 1mA in 600mA.
I find it easier to match and then measure across known Re.
the easy way to measure Iq is by measuring the voltage drop across the emitter resistor.
If the emitter resistors are +-5% then there must be another method to identify bias matching within 1mA in 600mA.
I find it easier to match and then measure across known Re.
Need for a group buy for Transistors?
Gentlemen
Would it be worthwhile to explore the possibility organizing a group buy of output transistors in order to have the volume's to enable one to sort and match transistors ? The alternative is that everybody sources his own and hopes for the best?
Compared to the KSA50 clone this is not a low powered amp- these device are going to work much harder- hence my concern that some transistors might be cruising whilst others might be at risk of breaking down.
What are the risks?
Jozua
Gentlemen
Would it be worthwhile to explore the possibility organizing a group buy of output transistors in order to have the volume's to enable one to sort and match transistors ? The alternative is that everybody sources his own and hopes for the best?
Compared to the KSA50 clone this is not a low powered amp- these device are going to work much harder- hence my concern that some transistors might be cruising whilst others might be at risk of breaking down.
What are the risks?
Jozua
Since almost everybody are going to use ONSemi (Motorola) devices and thus samples, and the others a lot of different types and models I seriously doubt if a group buy will work. In addition to the last few posts there're also more than enough evidence posted elsewhere on the web that for most applications matched output devices isn't necessary.
Besides, so far I followed nobody on the KSA50 forum complained about current imbalance in the output devices and few of them bothered to match theirs, and I've never measured an amp with that problem.
Each output transistor of the original KSA50 dissipated the same wattage as each of the KSA100's - check the math.
Besides, so far I followed nobody on the KSA50 forum complained about current imbalance in the output devices and few of them bothered to match theirs, and I've never measured an amp with that problem.
Each output transistor of the original KSA50 dissipated the same wattage as each of the KSA100's - check the math.
Exactly... You have twice the output devices and twice the value emitter resistor..... My KSA-50 is extremely well balanced as far as each device sharing the current. It also has virtually no DC drift while warming up....
Mark
I purchased 40 each of the 2SA1943 and the 2SC5200 to match at 500mA for my KSA50 amps using the "Elliot Sound" method and found that about 5% were rejected as they got very hot under test and the current increased significantly.
All other transistors were very close to 500mA and as Andrew suggested I grouped them in two groups high gain and low gain.
I do not know if matching has a significant effect on sound but you do get all transistors opperating at close to the same temperature.
Matching the emmiter resistors is no problem if purchased in large numbers or if they are puchased as 1% tolerance.
All other transistors were very close to 500mA and as Andrew suggested I grouped them in two groups high gain and low gain.
I do not know if matching has a significant effect on sound but you do get all transistors opperating at close to the same temperature.
Matching the emmiter resistors is no problem if purchased in large numbers or if they are puchased as 1% tolerance.
This article is pretty much a good summary of matching:
http://reviews.ebay.com/Classic-Ara...0001149810?ssPageName=BUYGD:CAT:-1:LISTINGS:2
Two paragraphs of interest:
"The original metal case outputs [output transistors] of the 4004 were made by RCA or Motorola. Because of the great variability of the Vbe and Beta characteristics of these transistors, matching was essential. Without matching, one or more of the parallel connected output transistors would conduct all of the bias current, leaving the others off. Later, before the introduction of the MK II, Toshiba metal transistors began to be used. These were supplied by the manufacturer in sorted beta groups, which made matching much less critical... [the Toshiba metal case transistors] were also high quality, tightly matched, and designed specifically for audio amps. They were much better than the earlier RCA and Motorola outputs..."
and...
"The Toshiba 2SA1302 and 2SC3281 are the plastic case transistors used in the Aragon 8000 series, the Klipsch 2000 and 3000 series, and all the Acurus amps. These were specifically designed for audio power amp use, and in my opinion are the finest bipolar output transistor made. They were expensive to manufacture, and Toshiba superseded them with the 2SA1943 and 2SC5200. These have a much smaller internal structure, poor matching, and are of generally lower quality. Toshiba tried to get us to switch to the newer transistors and sent a large number of samples which we used in building a few sample Acurus amps. It was impossible to accurately set the bias on these amps without individually matching the transistors. We therefore chose to continue with the original 2SA1302 and 2SC3281 by buying up existing inventories worldwide, even after they were no longer being manufactured by Toshiba. It is interesting to note that there is such a worldwide demand for these original transistors that they are actively being counterfeited. As far as I know, all Aragon amps, including the Klipsch built units were made with the original plastic case Toshiba 2SA1302 and 2SC3281 transistors."
If it's Motorola devices from the same batch, matching is unnecessary and that's exactly my experience with all the output stages I've built so far - since all used these. Add the large emitter resistors and it's even more negligible.
http://reviews.ebay.com/Classic-Ara...0001149810?ssPageName=BUYGD:CAT:-1:LISTINGS:2
Two paragraphs of interest:
"The original metal case outputs [output transistors] of the 4004 were made by RCA or Motorola. Because of the great variability of the Vbe and Beta characteristics of these transistors, matching was essential. Without matching, one or more of the parallel connected output transistors would conduct all of the bias current, leaving the others off. Later, before the introduction of the MK II, Toshiba metal transistors began to be used. These were supplied by the manufacturer in sorted beta groups, which made matching much less critical... [the Toshiba metal case transistors] were also high quality, tightly matched, and designed specifically for audio amps. They were much better than the earlier RCA and Motorola outputs..."
and...
"The Toshiba 2SA1302 and 2SC3281 are the plastic case transistors used in the Aragon 8000 series, the Klipsch 2000 and 3000 series, and all the Acurus amps. These were specifically designed for audio power amp use, and in my opinion are the finest bipolar output transistor made. They were expensive to manufacture, and Toshiba superseded them with the 2SA1943 and 2SC5200. These have a much smaller internal structure, poor matching, and are of generally lower quality. Toshiba tried to get us to switch to the newer transistors and sent a large number of samples which we used in building a few sample Acurus amps. It was impossible to accurately set the bias on these amps without individually matching the transistors. We therefore chose to continue with the original 2SA1302 and 2SC3281 by buying up existing inventories worldwide, even after they were no longer being manufactured by Toshiba. It is interesting to note that there is such a worldwide demand for these original transistors that they are actively being counterfeited. As far as I know, all Aragon amps, including the Klipsch built units were made with the original plastic case Toshiba 2SA1302 and 2SC3281 transistors."
If it's Motorola devices from the same batch, matching is unnecessary and that's exactly my experience with all the output stages I've built so far - since all used these. Add the large emitter resistors and it's even more negligible.
Hi P,
those two extracts fit my understanding of the bias balancing model that's inside my brain.
I believe that some amount of measuring and selecting is essential to ensure some semblance of balance @ Iq.
Whether this selecting is done at the manufacturing plant or at home is irrelevant. As long as it is done when many devices are used in parallel.
The emitter resistors then try to keep a modicum of balance as output curent rises. The increase in Vre is what helps high current balance.
those two extracts fit my understanding of the bias balancing model that's inside my brain.
I believe that some amount of measuring and selecting is essential to ensure some semblance of balance @ Iq.
Whether this selecting is done at the manufacturing plant or at home is irrelevant. As long as it is done when many devices are used in parallel.
The emitter resistors then try to keep a modicum of balance as output curent rises. The increase in Vre is what helps high current balance.
According to a discussion I once had with an ONSemi's engineer their tolerance on same-batch products are tight enough for most purposes, and that is exactly what I've found with all the ones I've measured so far. With parameters such as heatsink placement and contact resistance being the same the case temperature also remained the same. It appears that Toshiba's quality control has slacked a bit lately, as indicated in the article and by Harry3's claims, and that the 2SA1943 and 2SC5200 are not the magic panacea to transistor heaven. IMO MJL4302/4281 are a better solution in most regards - quality, quality control and tolerance-wise.
PWatts said:
With a Cob of 470pF for a 2SA1302 and 360pF for the 2SA1943 it's obvious that the die of the new series is smaller.
In contrast to OnSemi specimens, all Japanese power devices are hFe classed. Toshiba uses 2 hFe classes, R and O selected. Sanken devices can go up to 4 ; O, Y, P, and G.
That doesn't mean you can get a tight selection from devices which are from the same batch and hFe class.
I've never managed to get below 2% hFe selection on any of the Japanese power devices. Neither Toshiba nor Sanken, Fujitsu or ETC.
For NPN/PNP hFe matching it's even worse, i've never managed to get a decent fit.
From 2 same hFe class batches of 100 exotic Toshiba RET power devices i wasn't able to get two 1% matching sets of 4 output transistors. (Cost in 1987: $1500. Go figure, those Sanken case style devices had 3 hFe classes; R, O and the 120-240 hFe Y type which i used )
Vbe selection of Japanese devices is pretty good.
An attractive thing of the OnSemi products: 10% NPN/PNP hFe matched.
I'm constructing fully balanced Leach amps with the 6-device boards made by Jens. With lower rail voltages of 35Vdc i'm using 120 watt Toshiba 2SA1301/3280, instead of the 1302/3281 pair i thought of using at first.
"Best bipolar power transistor" depends on the voltage and power rating imo.
Here's the hFe curve of a 2SC3280
Attachments
Hi Jacco,
the hFE starts to drop off @ 3A.
For a three pair output stage that will give you 9Apk.
With a 4r load that allows full voltage drive from your rails.
That indicates to me that 4ohm is a sensible limit for speaker impedance and that 6ohm or 8ohm would be a better fit.
Hope that fits with your plans.
However, I ran Bensen's spreadsheet and it showed very good performance on 4ohm 60degree phase angle and still well below the 50degC 1second SOAR. Seems like they will drive 4ohm all day.
I would be tempted to use +-40Vdc supply rails unless you plan very high Iq.
the hFE starts to drop off @ 3A.
For a three pair output stage that will give you 9Apk.
With a 4r load that allows full voltage drive from your rails.
That indicates to me that 4ohm is a sensible limit for speaker impedance and that 6ohm or 8ohm would be a better fit.
Hope that fits with your plans.
However, I ran Bensen's spreadsheet and it showed very good performance on 4ohm 60degree phase angle and still well below the 50degC 1second SOAR. Seems like they will drive 4ohm all day.
I would be tempted to use +-40Vdc supply rails unless you plan very high Iq.
In case you forgot : the Leach boards have an integrated 3-stage V/I limiter. Michael Kiwanuka sent me his paper on SOA protection, and the revision paper.
The baby Toshibas are good for 24 amps peak.
Fully balanced means something like bridged, i have plenty of 6-Leach boards, 12 devices per channel and 70Vpp is +250 watts.
Prefer to use these despite having 800 OnSemi power devices last time i counted.
Motorola TO3s are hFe linear to about 1.5 A, some TO3 designs were fitted with a high frequency/high output current boost.
Yesterday i posted this link:
www.brinkmann-usa.com/monoamplifiers.php
These amplifiers use 4 Sanken devices per channel for 250 watts in 4 Ohms. This must be one of the longest running amplifier models. Considering their list price i suppose the designer must be a scrooge.
The baby Toshibas are good for 24 amps peak.
Fully balanced means something like bridged, i have plenty of 6-Leach boards, 12 devices per channel and 70Vpp is +250 watts.
Prefer to use these despite having 800 OnSemi power devices last time i counted.
Motorola TO3s are hFe linear to about 1.5 A, some TO3 designs were fitted with a high frequency/high output current boost.
Yesterday i posted this link:
www.brinkmann-usa.com/monoamplifiers.php
These amplifiers use 4 Sanken devices per channel for 250 watts in 4 Ohms. This must be one of the longest running amplifier models. Considering their list price i suppose the designer must be a scrooge.