I dont know if this have been discussed before, but i did a sim on MJL21193/MJL21194 against MJL3281A/1302A in a emitterfollower outputstage with perfect +/-40V rails (zero impedance) and a perfect generator to see wich kind of distortion they make.
I wanted to make my design as good as possible without negative feedback before adding it.
I have searched on this site for a previous discussion but did not find any.. So sorry if i make a repost on this subject.
Back to the compare.
The MJL21193/94 generates a dominating odd order distortion.
The MJL3281A/1302A generates a dominating even order distortion.
Idle current is 550mA in each transistor. output is 50Vpp into a load of 8Ohm.
Look for you self in the attached zipfile.
I wanted to make my design as good as possible without negative feedback before adding it.
I have searched on this site for a previous discussion but did not find any.. So sorry if i make a repost on this subject.
Back to the compare.
The MJL21193/94 generates a dominating odd order distortion.
The MJL3281A/1302A generates a dominating even order distortion.
Idle current is 550mA in each transistor. output is 50Vpp into a load of 8Ohm.
Look for you self in the attached zipfile.
Thanks for posting the results. Lacking any technical competence to comment, I would like to ask some questions -
How does the spectrum look with a 4 ohm load?
Are the constant beta charateristics of the 3281/1302 pair the causal effect?
Given the constant beta characteristics, how do the spectrums look with a less than perfect source driving the followers?
The spectrums are quite interesting, especially given Self's writings on 'load invariant' output stages.
Thanks,
Paul
How does the spectrum look with a 4 ohm load?
Are the constant beta charateristics of the 3281/1302 pair the causal effect?
Given the constant beta characteristics, how do the spectrums look with a less than perfect source driving the followers?
The spectrums are quite interesting, especially given Self's writings on 'load invariant' output stages.
Thanks,
Paul
There's an old saying in the computer business:
"Garbage in, gospel out!"
A simulation is only as good as its models - and it seems your models weren't very close.
I would be surprised to see a qualitative difference between two sets of BJT devices, and that's what your initial results suggested. I would not be surprised to see a modest quantitative difference.
There's another old saying:
"The purpose of computing is insight, not numbers."
What can you learn from the incorrect model? Specifically, what has to change in a real-world transistor to get the kind of results you want?
"Garbage in, gospel out!"
A simulation is only as good as its models - and it seems your models weren't very close.
I would be surprised to see a qualitative difference between two sets of BJT devices, and that's what your initial results suggested. I would not be surprised to see a modest quantitative difference.
There's another old saying:
"The purpose of computing is insight, not numbers."
What can you learn from the incorrect model? Specifically, what has to change in a real-world transistor to get the kind of results you want?
Hi chucko.
For the models .. They are good, but it is done by hand and therefore subjected to human error. Wich is the problem in this case.
It is not some homebrew ones but from onsemi.
The models are made by engineers not software designers.
"I would be surprised to see a qualitative difference between two sets of BJT devices" - If you take another BJT like TIP31 and 32 you will get another result.. So it will give you a indicator how the transistor behaves.
The sims also showed me that the higher the idlecurrent is => lower distortion to some degree and lot of other things.
This you only will see if you have done your PCB and maybee ended up making a new one.
Wich is a lot cheaper than make 10 different PCB's just to discard 9 of them.
You can test your design down the level just before the PCB to see if you have done any miscalculation.
At this point sims are good.
And when do not have a audioanalyzer this is a really good tool. You are not in total darkness.
(Sorry for english and my temper)
Sonny
For the models .. They are good, but it is done by hand and therefore subjected to human error. Wich is the problem in this case.
It is not some homebrew ones but from onsemi.
The models are made by engineers not software designers.
"I would be surprised to see a qualitative difference between two sets of BJT devices" - If you take another BJT like TIP31 and 32 you will get another result.. So it will give you a indicator how the transistor behaves.
The sims also showed me that the higher the idlecurrent is => lower distortion to some degree and lot of other things.
This you only will see if you have done your PCB and maybee ended up making a new one.
Wich is a lot cheaper than make 10 different PCB's just to discard 9 of them.
You can test your design down the level just before the PCB to see if you have done any miscalculation.
At this point sims are good.
And when do not have a audioanalyzer this is a really good tool. You are not in total darkness.
(Sorry for english and my temper)
Sonny
Sonny
The OnSemi model for the MJL1302A is not necessarily bad. Agreed the bf is set at 10000, but this is modified by other factors during use. If you simulate gain v collector current for the MJL1302A and the 2SA1302 you will find that the gain is not too dissimilar. These figures are simulated at 27degC and rounded to the nearest 5:
Ic (A) ... MJL1302A ... 2SA1302
0.1 .......... 100 ............ 105
0.5 .......... 120 ............ 115
1.0 .......... 115 ............ 115
2.0 .......... 100 ............ 110
5.0 ............ 75 ............. 90
Before using any new model, I always run this check and compare the results against the datasheet graphs for a typical device. Some models are very close to the datasheet figures and others show significant discrepancies and must be used with caution.
Geoff
The OnSemi model for the MJL1302A is not necessarily bad. Agreed the bf is set at 10000, but this is modified by other factors during use. If you simulate gain v collector current for the MJL1302A and the 2SA1302 you will find that the gain is not too dissimilar. These figures are simulated at 27degC and rounded to the nearest 5:
Ic (A) ... MJL1302A ... 2SA1302
0.1 .......... 100 ............ 105
0.5 .......... 120 ............ 115
1.0 .......... 115 ............ 115
2.0 .......... 100 ............ 110
5.0 ............ 75 ............. 90
Before using any new model, I always run this check and compare the results against the datasheet graphs for a typical device. Some models are very close to the datasheet figures and others show significant discrepancies and must be used with caution.
Geoff
I am a big idiot!!
The first sim i have done is OKAY. The MJL3281A/1302A pair do generate lower odd order harmonics.
And Geoff your are right about that the simulation model is OKAY.
I have not taken the low va into acount.
The gain of 240 was calculated at a vce of 40V. Lowering it to 10V shows a gain of ~130.
The MODEL IS OKAY!!!! for the MJL1302A.
Regarding the MJL21193(PNP) high Ic (When driving low impedance loads) it shows off a rise in vbe drop.
It is not my day .. Maybee i should get some sleep now!
Sonny
The first sim i have done is OKAY. The MJL3281A/1302A pair do generate lower odd order harmonics.
And Geoff your are right about that the simulation model is OKAY.
I have not taken the low va into acount.
The gain of 240 was calculated at a vce of 40V. Lowering it to 10V shows a gain of ~130.
The MODEL IS OKAY!!!! for the MJL1302A.
Regarding the MJL21193(PNP) high Ic (When driving low impedance loads) it shows off a rise in vbe drop.
It is not my day .. Maybee i should get some sleep now!
Sonny
pmkap said:
It looks like it is "constant beta" characteristics of the 3281/1302 pair. The current gain peaks at ~640mA and is nearly flat over an area of 1Amp. (1 - 2% gain change) where the MJL2119X has a gain who falls all the time. It has no flat "spot".
I have observed the same thing with smallsignal BJT in a emitterfollower.
If you can place the idlecurrent at the peakpoint and the transistor is flat over a wide area then the harmonics falls off in a nice slope at a relative "heavy load".
When you increase the load to 4 ohm or lower. Then you will see a rise in odd order harmonics.
At this point it matches D.self's comment about loading ,currentgain curve and multiple pairs of outputtransistors.
All this is when perfect source driving the followers.
All this would also apply to driverstage because of the heavy change in base current from the outputpair.
The driverstage also have to have a high early voltage. If it is low it will lower the input impedance on the driverstage.
I have not seen the article on 'load invariant'? But i would like to.
I can be wrong about all this. It would not be the first time today.
Sonny
Sonny,
Thanks for the response. The reson for my query is that at this very moment a pair of AKSA 100watt amps are wending their way from Australia to me in the Nasty Apple. Hugh's amps use that very same 3281/1302 pair (or possibly their equivalent 5200/1943 pair). As to why I'm seeking technical validation of my subjective view, it is simply a reflection of my typical 'audiophile' insecurity. I wouldn't buy the AKSAs until I heard them, and I was very impressed, to say the least.
With regards to Self's article on Load Invariant Amplifiers, its not available at his site -
http://www.dself.demon.co.uk/ampins.htm
like his article on Sources of Amplifier Distortion, but was published orignally in EW Jan 97. My reference was to the 1st part of 'Load Invariant Power Amplifiers' in the current, March 2002 "Audio Xpress" magazine. I don't know if they're the same article. If you can't find access to the article, email me with your address (my email server is currently down), and I'll send you a copy, or if you want to wait, I could email you the full article when the second part comes out.
To torture you, I'll quote the last part of the first part -
""increased driver nonlinearity is caused by beta-droop, which is caused by increased ouput device collector current" There are three points of attack here. First, increase the linearity of the driver system; this can't be done if it's a single transistor, as is usually the case. Secondly, the output device types can be selected for the least beta droop. This is very device dependant... Third, the per-device collector current can be reduced by using paralell output devices..."
Just as you said.
Thanks,
Paul
Thanks for the response. The reson for my query is that at this very moment a pair of AKSA 100watt amps are wending their way from Australia to me in the Nasty Apple. Hugh's amps use that very same 3281/1302 pair (or possibly their equivalent 5200/1943 pair). As to why I'm seeking technical validation of my subjective view, it is simply a reflection of my typical 'audiophile' insecurity. I wouldn't buy the AKSAs until I heard them, and I was very impressed, to say the least.
With regards to Self's article on Load Invariant Amplifiers, its not available at his site -
http://www.dself.demon.co.uk/ampins.htm
like his article on Sources of Amplifier Distortion, but was published orignally in EW Jan 97. My reference was to the 1st part of 'Load Invariant Power Amplifiers' in the current, March 2002 "Audio Xpress" magazine. I don't know if they're the same article. If you can't find access to the article, email me with your address (my email server is currently down), and I'll send you a copy, or if you want to wait, I could email you the full article when the second part comes out.
To torture you, I'll quote the last part of the first part -
""increased driver nonlinearity is caused by beta-droop, which is caused by increased ouput device collector current" There are three points of attack here. First, increase the linearity of the driver system; this can't be done if it's a single transistor, as is usually the case. Secondly, the output device types can be selected for the least beta droop. This is very device dependant... Third, the per-device collector current can be reduced by using paralell output devices..."
Just as you said.
Thanks,
Paul
Hi paul.
I will go looking for audioXpress tomorrow .. I think i know where to get a magazine. I will mail you if can't get my hands on a magazine.
Thanks a lot!!
By the way MJE15032 and MJE15033 have the same curve as MJL2119X.
So another driver is required .. Maybe Toshiba,Zetex,Rohm,philips (Philips get more and more japanese parts!!)..
My favor right know is Zetex... They are easy to get here in denmark, but they only make parts up to around ~3Watt.
Sonny
I will go looking for audioXpress tomorrow .. I think i know where to get a magazine. I will mail you if can't get my hands on a magazine.
Thanks a lot!!
By the way MJE15032 and MJE15033 have the same curve as MJL2119X.
So another driver is required .. Maybe Toshiba,Zetex,Rohm,philips (Philips get more and more japanese parts!!)..
My favor right know is Zetex... They are easy to get here in denmark, but they only make parts up to around ~3Watt.
Sonny
Sonny,
I really wouldn't spend much time trying to track down the article. The first 3/4s of it is simply a reprise of Self's article on amplifier distortion which is accessable at Self's website. So here is what you're missing -
"Supringly, the output devices do not contribute increased distortion with increased loading- it all comes from the drivers.
There are good SPICE-simulator evidence that this expanation is correct and LSN (Large Signal Nonlinearity) is entirely due to the beta-droop [what an unfortuneate choice of terminology] causing exta current to be drawn from the drivers, degrading their linerarity. Here is the evidence:
- Simulated output stages buit from output transistors modified to have no beta-droop at all (simply by increasing the SPICE model parameter IKF) show no LSN...
-Simulated output devices driven with zero-impedence voltage sources instead of transistor drivers show no LSN. This shows that LSN does not occur in the output transistors themselves, but in the drivers.
- If you look at the output stage distortion as an error voltage between input and output, the double emmitter follower (EF) stage errror is error=(driverVbe + OutputVbe +Re drop)
A simulated EF output stage with the usual drivers shows that of these three terms, it is primatily nonlinerarity in the driver Vbe that increases as the load resestance reduces, while th output Vbe nonlinerarity is hardly changed. The voltage drop across Re is essentially linerar; as it is in series with the load."
Thats about it, until the second part is published.
I do find Self's comment about the inviolate nature of a single transistor driver somewhat specious. Among those who've incorporated Self's findings, AKSA & LC Audio, both have adopted the EF topology (including the speedup cap accross the resistor linking the emmitters of the driver) through the unique approach of actually listening prior to choosing an output topology that incorporates extended beta outputs. In addition to that adoption, LC has chosen to address the current drive limitations of the driver current by using a cascade of 4 emmitter followers for each rail....
http://www.lcaudio.com/temil.htm
Golly, talk about brute force!
I'm certainly not criticizing Self's (or anyone's) simulation approach to design. It is through the use of these powerful tools that the state of the art is advanced. I just don't happen to agree that objective measures should necessarily superceed subjective metrics.
I really wouldn't spend much time trying to track down the article. The first 3/4s of it is simply a reprise of Self's article on amplifier distortion which is accessable at Self's website. So here is what you're missing -
"Supringly, the output devices do not contribute increased distortion with increased loading- it all comes from the drivers.
There are good SPICE-simulator evidence that this expanation is correct and LSN (Large Signal Nonlinearity) is entirely due to the beta-droop [what an unfortuneate choice of terminology] causing exta current to be drawn from the drivers, degrading their linerarity. Here is the evidence:
- Simulated output stages buit from output transistors modified to have no beta-droop at all (simply by increasing the SPICE model parameter IKF) show no LSN...
-Simulated output devices driven with zero-impedence voltage sources instead of transistor drivers show no LSN. This shows that LSN does not occur in the output transistors themselves, but in the drivers.
- If you look at the output stage distortion as an error voltage between input and output, the double emmitter follower (EF) stage errror is error=(driverVbe + OutputVbe +Re drop)
A simulated EF output stage with the usual drivers shows that of these three terms, it is primatily nonlinerarity in the driver Vbe that increases as the load resestance reduces, while th output Vbe nonlinerarity is hardly changed. The voltage drop across Re is essentially linerar; as it is in series with the load."
Thats about it, until the second part is published.
I do find Self's comment about the inviolate nature of a single transistor driver somewhat specious. Among those who've incorporated Self's findings, AKSA & LC Audio, both have adopted the EF topology (including the speedup cap accross the resistor linking the emmitters of the driver) through the unique approach of actually listening prior to choosing an output topology that incorporates extended beta outputs. In addition to that adoption, LC has chosen to address the current drive limitations of the driver current by using a cascade of 4 emmitter followers for each rail....
http://www.lcaudio.com/temil.htm
Golly, talk about brute force!
I'm certainly not criticizing Self's (or anyone's) simulation approach to design. It is through the use of these powerful tools that the state of the art is advanced. I just don't happen to agree that objective measures should necessarily superceed subjective metrics.
Hi paul.
Your comment :
"I'm certainly not criticizing Self's (or anyone's) simulation approach to design. It is through the use of these powerful tools that the state of the art is advanced. I just don't happen to agree that objective measures should necessarily superceed subjective metrics."
I agree a lot in this because there is a lot of things we do not know about the human ear.
I only use the sims to get up with something who is not totally wrong in terms of stability,distortion etc.
The rest has to be testet in real life > listning tests.
I am going to work now...
See you in this space later...
Sonny
Your comment :
"I'm certainly not criticizing Self's (or anyone's) simulation approach to design. It is through the use of these powerful tools that the state of the art is advanced. I just don't happen to agree that objective measures should necessarily superceed subjective metrics."
I agree a lot in this because there is a lot of things we do not know about the human ear.
I only use the sims to get up with something who is not totally wrong in terms of stability,distortion etc.
The rest has to be testet in real life > listning tests.
I am going to work now...
See you in this space later...
Sonny
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.