Yeh, I printed out the whole article and going to read it. There is a lot of truth about not trusting the theoretical circuits from book. I have question about another circuit of Mr. Cordell in post 5121 http://www.diyaudio.com/forums/solid-state/171159-bob-cordells-power-amplifier-book-513.html. He still have not got back to me. Maybe he ran out of patience with me already. I suspected there is a problem, I did the simulation and seemed to back up my concern.
That also was my point when I brought up the Threshold/Pass circuits. The circuit is everything that theory in the book is trying to improve. But this is Nelson Pass!!! Threshold amp is recognized as one of the top high end amp by a lot of people. The few schematics I got are nothing more than typical blameless circuit. Mr. Pass actually responded to one of my post on darlington and said he used higher value RE for the power transistor, totally opposite to Oliver's optimization. Hey, his Threshold speak louder than any book.
The reason I still want to do a complementary IPS is because my all time favorite amp YBA uses that......even without cascode with resistor load. That was the sweetest amp I ever heard and that's the one that got me into chasing for the sound here. I used to believe the sound is all from the speaker. That's why I spent so much more on the pair of speaker with just the Acurus. Then.......I listened to the YBA driving my speakers, that changed everything I believed in!!!
Last edited:
I don't really follow Nelson Pass' work however every one of his circuits that I have seen uses MOSFET outputs; gm doubling is not much of a problem and the Oliver criterion does not apply here. In that case he might use the larger Re (indeed Rs) values to improve bias stability or use the increased feedback to help linearise the output stage.
One thing I have learnt is that if you are chasing a particular sound then you won't get it by building sophisticated circuits designed for low low THD: in that category they all tend to sound the same, and if they don't I would suspect some errant behaviour such as oscillation or clipping/recovery behaviour.
The ones that have a 'sound' are the simple circuits like the P3A, TGM8, quasi's, simple MOSFET stages, with minimal to moderate feedback.
One thing I have learnt is that if you are chasing a particular sound then you won't get it by building sophisticated circuits designed for low low THD: in that category they all tend to sound the same, and if they don't I would suspect some errant behaviour such as oscillation or clipping/recovery behaviour.
The ones that have a 'sound' are the simple circuits like the P3A, TGM8, quasi's, simple MOSFET stages, with minimal to moderate feedback.
Knowing that you love the YBA is a great starting point. Why don't you develop an input stage based around the YBA, with a couple of small tweaks to make it your own, which will then serve as a terrific reference point for other topologies and circuit embellishments you wish to experiment with?
You have a passion for the YBA and want to add a twist, to make it your own and compare the differences. Sounds like a terrific DIY project and an interesting discussion, worthy of its own dedicated thread on here.
These discussions about super-complex designs are only interesting on an intellectual level, because you soon realise that if properly executed, it will sound identical to the next super-duper one sitting next to it, that you built a month ago.
You have a passion for the YBA and want to add a twist, to make it your own and compare the differences. Sounds like a terrific DIY project and an interesting discussion, worthy of its own dedicated thread on here.
These discussions about super-complex designs are only interesting on an intellectual level, because you soon realise that if properly executed, it will sound identical to the next super-duper one sitting next to it, that you built a month ago.
Why not using a blameless design? There is nothing wrong with that.
Here an example of a very good sounding high power "Blameless" with nearly symmetric slew and clipping, excellent overload behaviour, ultra low distortion, low noise, high PSRR, TMC compensation, very stable Bias, wide power supply voltage range. You may simply reduce output bjt pairs if using lower supply voltages:
2stageEF high performance class AB power amp / 200W8R / 400W4R
used devices:
- BC550C for LTP
- KSC3503D for Cascode
- BC560C, KSA992 for CM
- BC560C and dual KSA1381E for VAS
- KSC3503D for CCS
- 2SC4793 for Bias generator
- dual 2SC4793/2SA1837 as OPS drivers
- TTC5200 and TTA1943 as OPS power devices (lower Cob types of 2SC5200/2SA1943)
The few Threshold schematics are all BJT. He even replied my post attaching his Darlington power BJT with high value emitter resistor!!!
Chasing YBA is one of my goal, that's where the symmetric pcb comes into play!!! jumper the cascode and get rid of the current mirror, that will get closer already.!!!
I have to go to bed, it's 3AM now!!! Talk to you tomorrow.
I am looking at matched pairs for the input differential pair of IPS. My issue is the beta of PNP is only 80 vs NPN of over 200. That by default putting a DC offset at the ouput of the amp already as the input of the IPS is going to to be negative due to the excess of base current drawn by the PNP.
One of the solution is to use discrete transistors and hand match them before putting them in. Would this gives better result. eg. betas of BC550 and BC560 are a lot closer. The only thing is the thermal matching is not going to be as good as the matched pair on a single die.
One other thing, I simulate the complementary IPS 1) with fancy current mirror, 2) just a 2.2K resistor as load for IPS. And 3)YBA simple resistor load with NO cascode in IPS. The resistor load has lower 3rd order distortion!!!! There is no difference between with or without cascode. No difference in distortion at 1KHz and more importantly, no difference in frequency response as shown in the simulation.
One of the solution is to use discrete transistors and hand match them before putting them in. Would this gives better result. eg. betas of BC550 and BC560 are a lot closer. The only thing is the thermal matching is not going to be as good as the matched pair on a single die.
One other thing, I simulate the complementary IPS 1) with fancy current mirror, 2) just a 2.2K resistor as load for IPS. And 3)YBA simple resistor load with NO cascode in IPS. The resistor load has lower 3rd order distortion!!!! There is no difference between with or without cascode. No difference in distortion at 1KHz and more importantly, no difference in frequency response as shown in the simulation.
Forgot the 20KHz. Here's the FFT of 20KHz running for 5mS and step of 0.0005uS. Again, there is no observable difference.
YBA might know what they are doing!!!
YBA might know what they are doing!!!
I use a LTP jig to match transistors.
The emitter resistor must be very low or zero to get adequate sensitivity to balance/imbalance.
The jig was also shown by Anatech.
With BJTs the base resistor value also affects the sensitivity, one needs the base resistor to measure the hFE. But that resistor needs to be very low or zero when the Vbe is being compared across the pair.
jFETs aren't affected by the value of the gate resistor 100r to 1k is OK.
I start with Id = Idss and Vds = manufacturer's test voltage (sk170 uses 10Vds).
Then I sweep the gate voltage from 0Vgs to sufficient to bring Id down to about 20% of Idss. The discrete steps in my Id Vs Vgs are usually 100%, 70%, 50%, 30%, 20%.
If the Id match across the LTP pair all the way down then you have a pair that are a very good match. My looser sk170b pairs were sold off in 5%, 2% and 1% grades. I kept my 0.5%, & 0.2% supermatched pairs for use in amps like the Roender.
I apply a very similar strategy for BJTs, keeping in mind the need to switch out base resistor for the second stage of the match comparison.
These matched pairs are then super-glued together and 4 or 5 turns of 0.6mm diam copper wire wound around to bind them together as a near thermally coupled pair.
I have 20 sk389. They get nowhere near "matched pair" performance. They are thermally coupled but their matching is only guaranteed to 10% and that is at an unknown Id. Neither Toshiba nor Linear give the matching jig details, I suspect they are assuming from selected sampling that adjacent devices from the silicon slab must be "matched" !
The emitter resistor must be very low or zero to get adequate sensitivity to balance/imbalance.
The jig was also shown by Anatech.
With BJTs the base resistor value also affects the sensitivity, one needs the base resistor to measure the hFE. But that resistor needs to be very low or zero when the Vbe is being compared across the pair.
jFETs aren't affected by the value of the gate resistor 100r to 1k is OK.
I start with Id = Idss and Vds = manufacturer's test voltage (sk170 uses 10Vds).
Then I sweep the gate voltage from 0Vgs to sufficient to bring Id down to about 20% of Idss. The discrete steps in my Id Vs Vgs are usually 100%, 70%, 50%, 30%, 20%.
If the Id match across the LTP pair all the way down then you have a pair that are a very good match. My looser sk170b pairs were sold off in 5%, 2% and 1% grades. I kept my 0.5%, & 0.2% supermatched pairs for use in amps like the Roender.
I apply a very similar strategy for BJTs, keeping in mind the need to switch out base resistor for the second stage of the match comparison.
These matched pairs are then super-glued together and 4 or 5 turns of 0.6mm diam copper wire wound around to bind them together as a near thermally coupled pair.
I have 20 sk389. They get nowhere near "matched pair" performance. They are thermally coupled but their matching is only guaranteed to 10% and that is at an unknown Id. Neither Toshiba nor Linear give the matching jig details, I suspect they are assuming from selected sampling that adjacent devices from the silicon slab must be "matched" !
For differential pair, how about making a plug wire type breadboard that connect both emitter of the two transistors together, both base grounded through identical resistor and with the same collector resistor load.
Say for NPN, tie a resistor from both the emitter to -V to set up tail current. Connect the two collector resistor to +V.
Now you can measure the voltage drop across the two collector resistors. pick the two transistors that have the same voltage drop across the collector resistors.
If you are picky, measure the voltage drop across the base resistor and match.
I am planning to buy like 50 each of the BC560, BC550 and find the matched pairs out of them.
Can do the same thing on power BJT. Tie the base to ground, put current from the emitter and measure the Vbe ( ground to emitter). Then label each transistor to find 5 of them that match.
Say for NPN, tie a resistor from both the emitter to -V to set up tail current. Connect the two collector resistor to +V.
Now you can measure the voltage drop across the two collector resistors. pick the two transistors that have the same voltage drop across the collector resistors.
If you are picky, measure the voltage drop across the base resistor and match.
I am planning to buy like 50 each of the BC560, BC550 and find the matched pairs out of them.
Can do the same thing on power BJT. Tie the base to ground, put current from the emitter and measure the Vbe ( ground to emitter). Then label each transistor to find 5 of them that match.
you need to match both hFE and Vbe. and both need to be measured at operating current.
Then there is a follow up.
Do they "track each other" as the voltage is swept over the range that they see in the amplifier.
Measuring all of those "in the amplifier" will not be easy, or may be impossible.
One of the most difficult parts of this measuring testing is knowing the device Tj.
Keeping Tj to a constant or known value is nearly impossible.
For that reason I only attempt comparison measurements.
The jig allows various currents to be measured. But much more importantly one can ensure that I is the same between the pair and V is the same and they are close coupled and in the same draught free location.
Same I and same V results in same Pdiss. Same Pdiss and no draughts means virtually identical Tj. Comparison while Tj is the same effectively removes that almost impossible to control variable.
When you have batched same hFE devices, then select from that batch until you find potential matches.
And when you compare those "matches", it becomes very obvious that Vdiff @ collectors stays near constant and tiny while the Vgs is swept.
Measuring singletons needs quite sophisticated equipment and exemplary method.
Then there is a follow up.
Do they "track each other" as the voltage is swept over the range that they see in the amplifier.
Measuring all of those "in the amplifier" will not be easy, or may be impossible.
One of the most difficult parts of this measuring testing is knowing the device Tj.
Keeping Tj to a constant or known value is nearly impossible.
For that reason I only attempt comparison measurements.
The jig allows various currents to be measured. But much more importantly one can ensure that I is the same between the pair and V is the same and they are close coupled and in the same draught free location.
Same I and same V results in same Pdiss. Same Pdiss and no draughts means virtually identical Tj. Comparison while Tj is the same effectively removes that almost impossible to control variable.
When you have batched same hFE devices, then select from that batch until you find potential matches.
And when you compare those "matches", it becomes very obvious that Vdiff @ collectors stays near constant and tiny while the Vgs is swept.
Measuring singletons needs quite sophisticated equipment and exemplary method.
Last edited:
My method measure both hfe and Vbe. I plan to measure at current used in my IPS......1mA each transistor. That's what it matters, I don't really need to run through the whole current range.
The separate the hfe and Vbe, I guess ground the base and measure the voltage drop to match the Vbe first. Then put in the base resistance and measure the voltage drop on the base resistor. You pick the pair that match in both test.
Point is to make it easy to do, don't need to buy fancy stuff to do the test.
The separate the hfe and Vbe, I guess ground the base and measure the voltage drop to match the Vbe first. Then put in the base resistance and measure the voltage drop on the base resistor. You pick the pair that match in both test.
Point is to make it easy to do, don't need to buy fancy stuff to do the test.
Last edited:
The currents through the two halves of the LTP change with signal level !
The total current stays nearly constant due to the effect of the LTP CCS.
You are talking about hand match transistor of the same kind!!! They by default have the similar VAF NF, NE, IKE etc.!!! You are not matching two totally different kind of transistors like trying to match 2N2222 to 2N4124.
The important thing is to match the Vbe at quiescent condition when the base of both are at equal voltage ( ground). That's where the biggest error occurs. Remember current ratio doubles to every 18mV difference between the Vbe. Matching the Vbe is the most important thing.
Then match the hfe comes next as there is always resistance at the base. Base current cause voltage drop. If beta is not equal, that creates offset voltage just like Vbe mismatch.
So if you use my method to match both, then you get about the best matched condition with two discrete transistors.
Last edited:
I am looking at the data sheet of BC550 and BC560, the ones from Fairchild is very different from On-Semi!!! The one in Fairchild have higher Vce and lower beta than On Semi.
I looked at the BF parameters on both Cordell and Linear technology, the BF are all over about 500 to 600!!! what is the story?
I looked at the BF parameters on both Cordell and Linear technology, the BF are all over about 500 to 600!!! what is the story?
That is unusual.
All clones of the Philips originals generally have the same specifications.
I don't remember a clone giving different specs.
But the clones will almost certainly be made using different processes and that may lead to different performance/parameters. But the cloners still state the same specs.
I have the Onsemi and Fairchild datasheets open.
Which specs are "very different"?
Last edited:
I found the problem. You cannot look at the graph of Fairchild. You have to pick the "C" grade to match On Semi. You can't pick the "A" or "B" of Fairchild.
Last edited:
- Status
- Not open for further replies.