Hi Guys
The amps I've been working on recently use a 10k feedback resistor and 10k base-leak on the input side, so matched impedances for the base current paths of the diff amp. These are typical values - some might say obsolete values. All my comp-diff amps use resistive loading for the input stage.
With complementary diffs, most of the base current from the NPN passes through the PNP - ideally all of it does - so the actual base difference current through the two Rs in question should be quite low and thus the DC offset should be quite low. The sims show extremely tiny offset, as do the physical amps.
Self promotes much lower R values to push noise as far down as possible. I'm happy if S/N is >100dB. The ideal is to drive the input of the PA from a low impedance, allowing both Rs to be made smaller. At the same time, if the input buffer also has a small amount of gain, the closed-loop gain of the PA proper can be reduced, further reducing overall THD.
There are lots of ways to do things in electronics, which provides a lot of leeway for individual tinkering.
Still love the book!
Have fun
The amps I've been working on recently use a 10k feedback resistor and 10k base-leak on the input side, so matched impedances for the base current paths of the diff amp. These are typical values - some might say obsolete values. All my comp-diff amps use resistive loading for the input stage.
With complementary diffs, most of the base current from the NPN passes through the PNP - ideally all of it does - so the actual base difference current through the two Rs in question should be quite low and thus the DC offset should be quite low. The sims show extremely tiny offset, as do the physical amps.
Self promotes much lower R values to push noise as far down as possible. I'm happy if S/N is >100dB. The ideal is to drive the input of the PA from a low impedance, allowing both Rs to be made smaller. At the same time, if the input buffer also has a small amount of gain, the closed-loop gain of the PA proper can be reduced, further reducing overall THD.
There are lots of ways to do things in electronics, which provides a lot of leeway for individual tinkering.
Still love the book!
Have fun
If you break the connection shorting out the input capacitors, offset at output gets much better. If you change the resistors at base of the input transistors from 25k to 19k, you further improve the offset. So, to me, that simulation seemingly did not prove much about your concern of offset or its drifting, but instead demonstrated the amount of degeneration does not impact offset when circuit is set up in a balanced way.
Short the input capacitor just change the bias offset due to the base current that just happen to help. Problem is not the input LTP, it is the lack of gain of the total stage to compensate any mismatch of the current of the two LTP and the mismatch of the push-pull VAS.
Of cause if you match everything, you cure the offset!!! The only true help is reducing the degeneration resistance, but then you get distortion.
Really the bottom line is how low the degenerate resistor can be lower before it start to be a factor in the distortion of the power amp. As I said, the major contribution of distortion is in the OPS, the crossover distortion being the major factor. All the simulations, the distortion of the IPS/VAS combo is an order lower to any distortion from the OPS. So putting a 300ohm degeneration resistor might be an unnecessary over kill.
Besides, remember, lower the degeneration resistor increase loop gain and that lower distortion also.
Last edited:
It does not "happen" to help, it is engineered to help. BJT LTPs are usually designed to see identical DC resistance on their front and back sides to cancel out the voltage drops caused by the base current, or the non-zero base current differential in case of a mismatched complementary IPS as you showed. The stages have HUGE open loop gain at DC regardless LTPs degeneration, you don't need more of it.
Take a look at the attached slightly modified schematic. The offset at output with heavy LTP degeneration is only a few tens of mV with DC coupling at both input and feedback. Note the added R15 and R16 balance out the resistance on the back side of the LTPs, you need them for an input low pass filter anyways in an actual amp.
If a "true help" does not serve a purpose well yet creates unwanted problems it is not a help.
Again, as many others have pointed out for times, a DC servo is almost a necessity for a DC coupled amp as a good safety practice.
"a 10k feedback resistor and 10k base-leak on the input side" these two resistors have no bearing with the noise or S/N. The input resistor noise contribution is shorted by the source impedance (much lower), while the feedback resistor contribution is divided by the loop gain, resulting in pretty much nothing in the audio band. It is the grounded side of the feedback network that contributes to noise, e.g. if you chose Rf=9k and the closed loop gain 20dB, then the grounded part of the feedback network results 1k, that is about 4nV/rtHz noise contribution.
100dB S/N requirement, input ref. to what? Otherwise, it doesn't make any sense.
Anyway, in the above example, 4nV/rtHz of noise at the inverting input would already be almost good enough for a moving magnet cartridge, nothing to worry about.
That's exactly what I said, you short out the cap and use the low impedance of the voltage source to reduce the voltage drop due to the base current on one side that happens to help. It is standard design to try to have equal resistance on both sides so the offset is equal.
Again, the problem I brought up and Mr. Cordell agreed is NOT from the input of the LTP and matching of the two differential transistor. The amp has DC gain of unity. This means if the transistor pair has offset of 10mV, the output will only have 10mV offset. That is not even close to the issue.
Again, the worst is when the current source of the top and bottom LTP is imbalance, you can see huge offset at the output. Try put one side to 1.5mA instead of 2mA, you have over 100mV of offset at the output.
This whole problem has NOTHING to do with the input base current that you mentioned, nothing to do with the Vbe matching of the input differential pair. It has everything to do with the current source imbalance and the mismatch of the push pull VAS transistors.
Try lowering the degeneration resistors, the offset will get better as you have more loop gain to compensate for any imbalance of the rest of the circuit.
I am sure Mr. Cordell can explain it better than me.
I think the best solution is to lower the degeneration resistor a little, like to 100ohm and simulate to see how is the distortion. As I kept repeating, the distortion of the OPS is an order of magnitude above any reasonable IPS/VAS, I don't think we need to sweat too much over this.
Then I am going to hand pick 10V zener diode to bias the current source. This way I have 9.3V across the emitter resistor to control the current. This will make it much more predictable and consistent. It was a big mistake for me to layout the board using the green LED that produce only 1.9V to set up the current source. Any variation of the voltage drop across the LED or Vbe change of the current source transistor will cause big drift in the current. I don't think there is a way out of this one.
It is much easier to match devices than it sounds. Only took me 10 minutes to matched out 4 pairs of BC550 and 4 pairs of BC560 to less than 1mV offset on the Vbe AND more important I matched to within 40mV between the two collectors of the differential pair using 2mA tail current and 2.2K collector resistor. I am sure I can match two zener diodes to 0.1V at a given current for the two current sources. This will be matching the two current sources to less than 2%.
Then I am going to hand pick 10V zener diode to bias the current source. This way I have 9.3V across the emitter resistor to control the current. This will make it much more predictable and consistent. It was a big mistake for me to layout the board using the green LED that produce only 1.9V to set up the current source. Any variation of the voltage drop across the LED or Vbe change of the current source transistor will cause big drift in the current. I don't think there is a way out of this one.
It is much easier to match devices than it sounds. Only took me 10 minutes to matched out 4 pairs of BC550 and 4 pairs of BC560 to less than 1mV offset on the Vbe AND more important I matched to within 40mV between the two collectors of the differential pair using 2mA tail current and 2.2K collector resistor. I am sure I can match two zener diodes to 0.1V at a given current for the two current sources. This will be matching the two current sources to less than 2%.
Hi Chris,
That yields very nice numbers, but I believe that the more common S/N quoted is x dB below 2.83V, which is 1 watt into 8 ohms. That is the one most often measured and quoted by John Atkinson, for example. This makes for an apples-apples comparison between the actual audible noise level for amplifiers of different power capabilities when connected to the same loudspeaker.
Cheers,
Bob
Here is the problem as I see it. Bear in mind I am not disagreeing with what you have said.
If one has to go to such lengths of matching the current sources et al, then this is a serious disadvantage of the chosen topology or arrangement. Hand matching like this simply should not be necessary to build a good amplifier with reasonable output offset, even without a DC servo.
Cheers,
Bob
Hi Mr. Cordell
I think this problem is not limited to complementary IPS, even the blameless topology has part of the problem though not quite as bad. Any current mismatch (at no signal) between the current source in the VAS with the transistor driven by the IPS will produce offset. The advantage of blameless is you don't have two current sources at the IPS that need to be match and it's the dominant cause of offset in my simulation.
Regarding to hand matching. I made all the pcb already, too late to worry about the DC servo. I am building my ultimate amp, I don't mind hand picking parts. Actually it's very easy This is a post I posted in matching MJL3281 and 1302. Took me about 1 hour to match out at least 50 of each. I have enough to build 5 amp minimum ( assuming I use 5 transistor pairs per OPS, more if anyone use less). So it's not much of an issue.
http://www.diyaudio.com/forums/solid-state/275364-mjl3281-vbe-matching.html
It seems transistor is a lot more consistent than 30 years ago. I remember when I was designing semi custom IC for Exar in the 80s, we can only count on 3mV Vbe matching on a simple transistor without any fancy technique. You can see the matching is quite well in the MJL.
Thanks
You are right on target with most of what you have said. However the "Blameless" topology with the IPS current mirror load is extremely tolerant of differences in the VAS current source value. It almost doesn't care whether the VAS current is 5mA or 25mA in regard to DC offset. This is simply because the DC gain in front of the VAS is so high. All of this is doubly the case when the VAS transistor has a Darlington arrangement (2T VAS).
When a current mirror load is not used for the IPS, and a single-ended resistive load is used instead, then the value of the IPS current source becomes much more influential on DC offset, especially when there is significant IPS emitter degeneration.
I didn't mean to imply that a DC servo is necessary to get good DC offset. I have built many amplifiers without DC servos that have not had DC offset problems.
You are quite right about the relative ease of matching transistors that come out of the same tube.
When at Exar, did you interact much with Alan Grebene?
Cheers,
Bob
Absolutely right. It's only when you move from the Blameless configuration to one that is apparently more sophisticated, with double input stages/push-pull VAS, that you really appreciate how tolerant Blameless is in various ways. In Audio Power Amplifier Design (6th edition) I spent the whole of Chapter 8 looking at various double input stages/push-pull VAS and none of them were as good as a Blameless in any parameter, and there were usually unwelcome extra problems like drift of the input pair balance.
You gotta ask yourself, what are you hoping to gain with a double input stages/push-pull VAS configuration? I gather Alan0354 has already had some experience with the problems. Not wishing to discourage you, but it might be worthwhile to read Chapter 8 before putting too much time into this.
Hi Bob,
I might be a little out of time these days, but power amplifiers often did rate from full published power ratings. Taking the number from the one watt point makes it easy for the consumer to compare numbers in a meaningful way.
Taking the reference as the one watt power point is how I was taught to make the measurements back in the Superscope Marantz days. I was affiliated with that company the most back then. They were serious about how their numbers came about, and I was running a higher level service shop in their service network. Those were very good days as my parts discount allowed me to sell to other shops in their network at a profit. Anyway, I had to work in a cleaner manner than most for that position and my training reflected that.
Hi Douglas,
Excellent books, I highly recommend them to anyone interested to know how amplifiers actually work in the real world. I learned a lot, even after decades of servicing audio products. Many thanks for writing them.
-Chris
Isn't the nice number the entire point of the advertizing industry in home audio?
I might be a little out of time these days, but power amplifiers often did rate from full published power ratings. Taking the number from the one watt point makes it easy for the consumer to compare numbers in a meaningful way.
Taking the reference as the one watt power point is how I was taught to make the measurements back in the Superscope Marantz days. I was affiliated with that company the most back then. They were serious about how their numbers came about, and I was running a higher level service shop in their service network. Those were very good days as my parts discount allowed me to sell to other shops in their network at a profit. Anyway, I had to work in a cleaner manner than most for that position and my training reflected that.
Hi Douglas,
Excellent books, I highly recommend them to anyone interested to know how amplifiers actually work in the real world. I learned a lot, even after decades of servicing audio products. Many thanks for writing them.
-Chris
Is single digit distortion our only goal ? The blameless natively has the worst overload
behavior. But , it is the best documented (resilient) design out there. ALL is a compromise.
nearly the same. Symmetric cascoded VAS's are much more durable and many
prefer the SQ ( yes , there is difference).
I have both blameless's and symmetric designs , no issues with either.
BTW , a blameless without TMC is about even with all the symmetric designs
@ 20K THD. In this case, I know of 2 designs that beat a blameless in all
parameters !
Edit - your book allowed the forum to hear the "blameless" in the form of the
Badger and Wolverine , no disrespect for the design or analysis. But don't
discount the symmetric designs , they have some superior attributes.
OS
behavior. But , it is the best documented (resilient) design out there. ALL is a compromise.
No way , the blameless (only) "takes the show" at 5ppm. Overload , drift , PSSR -
nearly the same. Symmetric cascoded VAS's are much more durable and many
prefer the SQ ( yes , there is difference).
I have both blameless's and symmetric designs , no issues with either.
BTW , a blameless without TMC is about even with all the symmetric designs
@ 20K THD. In this case, I know of 2 designs that beat a blameless in all
parameters !
Edit - your book allowed the forum to hear the "blameless" in the form of the
Badger and Wolverine , no disrespect for the design or analysis. But don't
discount the symmetric designs , they have some superior attributes.
OS
No I don't recognize the name. It's been 30 years since I left Exar. I think the person in charge was Yukyo or something. I was young at the time, my immediate boss was John Ng. It was like an international melting pot there. They had people from Turkey, Japan, Chinese............ I can't really spell their names. It's common like this today where all these races mix together, but back in the 80s, it's very different. I was with Exar for a year and half only.
I already have the pcb made and partly stuffed. I never power anything up yet, just some nagging thoughts about the DC offset that made me simulate the offset under different mismatch.
I remember I did simulation before on single LTP IPS vs complementary IPS. I remember the distortion is slightly lower with the complementary. BUT I was totally surprised that the distortion of the OPS is at least an order of magnitude higher. That's where I agree with you, what difference does it make!!! Any reasonable IPS/VAS is going to be much lower in distortion than the best OPS stage. So why even haggle over what IPS topology is better.
The only reason I did the complementary IPS/VAS was because my all time favorite amp YBA using this. The famous Threshold amp by Nelson Pass use Blameless in one of his design, It's Nelson Pass!!!
I separate the IPS/VAS onto a different small board, when I design another one, I definitely simplify the design to a Blameless like. I just increase the VAS current to get back the slew rate. My OPS is a tripple Darlington, it's not going to present a heavy load to the VAS anyway.
The blameless concept is really god and simple way to build an amp with very low distortion. I built my TT -TMC amp according to that concept and it is very god and very nice sounding amp. It has very low THD1k but not so low THD20k. Symmetric concept with symmetric IPS and VAS and the same OPS as in my TT-TMC amp shows lower THD20k thanks to the loop gain of more then 80 dB at 20 kHz. The THD1k is little worse as the loop gain is the same as at 20 kHz. Here is the schematic and it is CFA. http://www.diyaudio.com/forums/solid-state/253039-unique-cfa-120-230w-amp-15.html#post4362435
BR Damir
BR Damir
Well, as I've argued elsewhere, frequency response, slewrate, noise etc are all more than good enough. If we seek the perfect amplifier, then distortion looks like the unsolved problem.
It does? In what way?