Hi Andy,
Many thanks for your thoughts.
Off topic for a ribbon tweeter amp ..... THD is meaningless here so we are without a written standard for performance comparison.
I have always regretted not writng notes of my 'young ear' observations when inserting passive buffered filters ahead of a 500kHz JLH amplifier, this after I had read of some users becaming concerned about such b/w in audio applications. I concluded that no filtering - no degradation - was best. No other amplifier came near for a cleanliness of sibilant reproduction; and that kind of b/w is not a problem to those who understand the precautions necessary.
____________________________________________________
Hi jcx.
Just a couple of thoughts.
Q2 collector potential can be set just below input overdrive saturation level for U3.
Could Q6 pulldown damage U3 ?
____________________________________________________
My bipolar circuit is a 'concrete block' to crack a walnut.
A Mosfet class-A would be a 'half block'.
I am not in a position to experiment or simulate the curvilinear Mosfet circuit I posted above, but I do feel that a little class-A 'brick' would be possible; with a potential for simplicity too.
Cheers ......... Graham.
Many thanks for your thoughts.
Off topic for a ribbon tweeter amp ..... THD is meaningless here so we are without a written standard for performance comparison.
I have always regretted not writng notes of my 'young ear' observations when inserting passive buffered filters ahead of a 500kHz JLH amplifier, this after I had read of some users becaming concerned about such b/w in audio applications. I concluded that no filtering - no degradation - was best. No other amplifier came near for a cleanliness of sibilant reproduction; and that kind of b/w is not a problem to those who understand the precautions necessary.
____________________________________________________
Hi jcx.
Just a couple of thoughts.
Q2 collector potential can be set just below input overdrive saturation level for U3.
Could Q6 pulldown damage U3 ?
____________________________________________________
My bipolar circuit is a 'concrete block' to crack a walnut.
A Mosfet class-A would be a 'half block'.
I am not in a position to experiment or simulate the curvilinear Mosfet circuit I posted above, but I do feel that a little class-A 'brick' would be possible; with a potential for simplicity too.
Cheers ......... Graham.
Charles Hansen said:
testing my sim with a 1 A 5KHz sine current source on the output gives ~ 6 uOhm Zout @ 5KHz with the feedback loop gain helping out of course
looking at U3 gate voltage i get ~ 60 S tranconductance for the output stage, 30 S per device isn't quite bipolar performance but it seems adequate, particularly as the drive circuit is a to-92 bjt buffering an op amp
there seems little reason for output Q impednce to be much lower than the current sense resistors - we are talking class A here so i assume some degeneration/current sense of >> 10 mOhms will be necessary
WHAT?!
"looking at U3 gate voltage i get ~ 60 S tranconductance for the outputstage, 30 S per device isn't quite bipolar performance but it seems adequate"
Unless I have missed something, you were talking about a pair of IRL3103s............... The minimum transconductance in the data sheet is 22......... FOR AN Id OF 34 AMPS!
At 2 or 3 amps of bias current you are probably looking at a transconductance of something in the range of 5 to 8 for most of the high transconductance vertical mosfet I have played with The graph doesn't show this in the data sheet for the IRL3103 since it does go down to a few amps. Look at the trends of Gf verses Id for a few other Hexfets like the IRF150, IRF240, ect. Without a feel for transconductance as a function of temperature and Id you have no reality check for Spice results. When you get results verging on an order of magnitude different from realistically expected it is time to find out why.
http://www.irf.com/product-info/datasheets/data/irl3103.pdf
You guys are turning Spice into a CHD tool. Time to turn off the computer and fire up the scope and soldering iron.
CHD: Computer hindered design.
"looking at U3 gate voltage i get ~ 60 S tranconductance for the outputstage, 30 S per device isn't quite bipolar performance but it seems adequate"
Unless I have missed something, you were talking about a pair of IRL3103s............... The minimum transconductance in the data sheet is 22......... FOR AN Id OF 34 AMPS!
At 2 or 3 amps of bias current you are probably looking at a transconductance of something in the range of 5 to 8 for most of the high transconductance vertical mosfet I have played with The graph doesn't show this in the data sheet for the IRL3103 since it does go down to a few amps. Look at the trends of Gf verses Id for a few other Hexfets like the IRF150, IRF240, ect. Without a feel for transconductance as a function of temperature and Id you have no reality check for Spice results. When you get results verging on an order of magnitude different from realistically expected it is time to find out why.
http://www.irf.com/product-info/datasheets/data/irl3103.pdf
You guys are turning Spice into a CHD tool. Time to turn off the computer and fire up the scope and soldering iron.
CHD: Computer hindered design.
hi Fred,
certainly i've pushed the sim to the point where some “ground truth” would be the next step, i've mentioned my mistrust of the model several times
but 30 S doesn't seem impossible, as you say the 22 S is a min # and the transfer characteristic graph in the data sheet fig 3 shows gm beginning to level off in the high amp range, it looks to me like there is room for the curve to steepen below 10 A
in fact, looking at fig 1 i think you can find the current going from 4 A to 14 A for a 2.7 to 3 Vgs change, …that sounds like 33 S to me
certainly i've pushed the sim to the point where some “ground truth” would be the next step, i've mentioned my mistrust of the model several times
but 30 S doesn't seem impossible, as you say the 22 S is a min # and the transfer characteristic graph in the data sheet fig 3 shows gm beginning to level off in the high amp range, it looks to me like there is room for the curve to steepen below 10 A
in fact, looking at fig 1 i think you can find the current going from 4 A to 14 A for a 2.7 to 3 Vgs change, …that sounds like 33 S to me
I have spoken about propagation delay affecting first cycle treble responses, but little feedback has developed.
The original JLH class-A has only 400mR of imbalanced (CC/CE) output impedance, yet it satisfies listeners in much the same way that a non-feedback amplifier can.
If you examine its response from t=0 at mid band you will see that its NFB improved damping characteristic is totally coherent with the examining 'back emf' current application; not just at the steady state point, from where an 'impedance' figure would be derived.
Hi Jcx,
Your circuit is clever, with local feedback at the output stage, plus a global 3x gain controlling loop; ie. an awful lot of NFB.
Theoretically 6uR is way below any requirement for the most difficult loudspeaker, and totally unnecessary for a ribbon tweeter, but this figure cannot be derided.
Any chance of you running your 1A-5kHz sinewave into the output stage and letting us know how it copes from t=0 to say t=100nS, checking whether it retains accurately coherent and non-spiky control which might be affected due to the NFB loops reacting differently with gate capacitance. Does the voltage at the output node remain negligible, or does the amplifier generate new rf waveforms as it attempts to correct the error ?
Just interested !
Are you going to build it, or does Fred have the last laugh ?
Cheers ......... Graham.
The original JLH class-A has only 400mR of imbalanced (CC/CE) output impedance, yet it satisfies listeners in much the same way that a non-feedback amplifier can.
If you examine its response from t=0 at mid band you will see that its NFB improved damping characteristic is totally coherent with the examining 'back emf' current application; not just at the steady state point, from where an 'impedance' figure would be derived.
Hi Jcx,
Your circuit is clever, with local feedback at the output stage, plus a global 3x gain controlling loop; ie. an awful lot of NFB.
Theoretically 6uR is way below any requirement for the most difficult loudspeaker, and totally unnecessary for a ribbon tweeter, but this figure cannot be derided.
Any chance of you running your 1A-5kHz sinewave into the output stage and letting us know how it copes from t=0 to say t=100nS, checking whether it retains accurately coherent and non-spiky control which might be affected due to the NFB loops reacting differently with gate capacitance. Does the voltage at the output node remain negligible, or does the amplifier generate new rf waveforms as it attempts to correct the error ?
Just interested !
Are you going to build it, or does Fred have the last laugh ?
Cheers ......... Graham.
Graham Maynard said:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=32137
Hi GM,
post something contentious here, its got very boring.
sreten.
We will have to agree to disagree
"hi Fred,
certainly i've pushed the sim to the point where some “ground truth” would be the next step, i've mentioned my mistrust of the model several times
but 30 S doesn't seem impossible, as you say the 22 S is a min # and the transfer characteristic graph in the data sheet fig 3 shows gm beginning to level off in the high amp range, it looks to me like there is room for the curve to steepen below 10 A
in fact, looking at fig 1 i think you can find the current going from 4 A to 14 A for a 2.7 to 3 Vgs change, …that sounds like 33 S to me"
Fig. 1 is for a temperature of Tj = 25 degrees C. I (and I would imagine Hugh and Grey as well) would sure like to see to cooling system for that amp! Liquid nitrogen perhaps.........
Fig.3 is a linear to log graph and estimating the trend of the curve past what is shown on the graph can be Very misleading.
In school this is what we use to call cooking the data.
I am reminded of a story told by my Dynamics professor. They were doing a study on auto collisions with an experimental collapsible lightpole. 
They were using a very large steel ball swung on a chain to simulate the cars mass and velocity.
The only thing they could find to weigh it was a truckstop scale. They explained to the scale operator what they needed and he said "How much do you want it to weigh?"
"hi Fred,
certainly i've pushed the sim to the point where some “ground truth” would be the next step, i've mentioned my mistrust of the model several times
but 30 S doesn't seem impossible, as you say the 22 S is a min # and the transfer characteristic graph in the data sheet fig 3 shows gm beginning to level off in the high amp range, it looks to me like there is room for the curve to steepen below 10 A
in fact, looking at fig 1 i think you can find the current going from 4 A to 14 A for a 2.7 to 3 Vgs change, …that sounds like 33 S to me"
Fig. 1 is for a temperature of Tj = 25 degrees C. I (and I would imagine Hugh and Grey as well) would sure like to see to cooling system for that amp! Liquid nitrogen perhaps.........
Fig.3 is a linear to log graph and estimating the trend of the curve past what is shown on the graph can be Very misleading.
In school this is what we use to call cooking the data.
I am reminded of a story told by my Dynamics professor. They were doing a study on auto collisions with an experimental collapsible lightpole. They were using a very large steel ball swung on a chain to simulate the cars mass and velocity. The only thing they could find to weigh it was a truckstop scale. They explained to the scale operator what they needed and he said "How much do you want it to weigh?"jcx said:
Out of interest, do you have a model which backs this up? In other words, given no more that X% THD with a single tone, how much more than X% error can you have on a complex signal? Many of the proposed models (thermal distortion, slew-limiting distortion) will actually show up in suitable single-tone tests.
Considering, for example, an MP3 codec, it will reproduce single tones with almost arbitrary accuracy, but fall apart badly on a random input signal. Clearly, therefore, you can mathematically create distortion which 'hides' from a single-tone test (or even n-tone), but can these be realised using a handful of transistors and passives?
Cheers
IH
hi Fred,
i think your current objection is that the data is from an "undercooked" transistor
turning this to a discussion of thermal dependence of gm when you originally claimed the sim # had to be an order of magnitude off
the fig 1 data can be interpolated well enough to show ~ 30 S @ 5-14 A, admittedly pulsed to make a Tj=25 C measurement
unfortunately the threshold shift prevents reading similar Id, Vds from the 175 Tj curves
how much should we expect gm to drop at 120-150 C Tj, 30% ?
Ian,
i'm not an expert in nonlinear modeling but i think you can find more info in “Distortion Analysis of Analog Integrated Circuits” P Wambacq & W Sansen 1998
in the audio world there has been more interest in modeling dynamic speaker distortion with Volterra series
i think your current objection is that the data is from an "undercooked" transistor
turning this to a discussion of thermal dependence of gm when you originally claimed the sim # had to be an order of magnitude off
the fig 1 data can be interpolated well enough to show ~ 30 S @ 5-14 A, admittedly pulsed to make a Tj=25 C measurement
unfortunately the threshold shift prevents reading similar Id, Vds from the 175 Tj curves
how much should we expect gm to drop at 120-150 C Tj, 30% ?
Ian,
i'm not an expert in nonlinear modeling but i think you can find more info in “Distortion Analysis of Analog Integrated Circuits” P Wambacq & W Sansen 1998
in the audio world there has been more interest in modeling dynamic speaker distortion with Volterra series
Attachments
......is wrong I don't wanna be right.
"turning this to a discussion of thermal dependence of gm when you originally claimed the sim # had to be an order of magnitude off
the fig 1 data can be interpolated well enough to show ~ 30 S @ 5-14 A, admittedly pulsed to make a Tj=25 C measurement
unfortunately the threshold shift prevents reading similar Id, Vds from the 175 Tj curves
how much should we expect gm to drop at 120-150 C Tj, 30% ?"
Read the data sheet. Your playing a game af apples and oranges. I am talking about trying to figure out what the tranconductance in real life under real operating conditions. Don't try and guess data that is not on the datasheet. Go look at the data sheet for something like the IRF150 to see how these parameters change from a couple of amps to a couple of tens of amps.
The data you assume is with the transistor at 5-14 amps and Tj=25 C. That is not where the transistor is biased. Base your
measurements on actual bias currents and operating temperatures. I would assume that one would want to know what the small signal transconductance is rather than what it is doing pulling the load close to the rail. Also Figure 1 shows a pretty serious change in transconductance with with drain to source voltage and doesn't look too linear.
Maybe a new thread or even a forum: "CHiD AMPS" for people who want only to run Spice without making measurements or even comparing results with the data sheet. I think we will have hundreds of post since this seems to be a very popular pastime on DIYaudio in the last few months.
"turning this to a discussion of thermal dependence of gm when you originally claimed the sim # had to be an order of magnitude off
the fig 1 data can be interpolated well enough to show ~ 30 S @ 5-14 A, admittedly pulsed to make a Tj=25 C measurement
unfortunately the threshold shift prevents reading similar Id, Vds from the 175 Tj curves
how much should we expect gm to drop at 120-150 C Tj, 30% ?"
Read the data sheet. Your playing a game af apples and oranges. I am talking about trying to figure out what the tranconductance in real life under real operating conditions. Don't try and guess data that is not on the datasheet. Go look at the data sheet for something like the IRF150 to see how these parameters change from a couple of amps to a couple of tens of amps.
The data you assume is with the transistor at 5-14 amps and Tj=25 C. That is not where the transistor is biased. Base your
measurements on actual bias currents and operating temperatures. I would assume that one would want to know what the small signal transconductance is rather than what it is doing pulling the load close to the rail. Also Figure 1 shows a pretty serious change in transconductance with with drain to source voltage and doesn't look too linear.
Maybe a new thread or even a forum: "CHiD AMPS" for people who want only to run Spice without making measurements or even comparing results with the data sheet. I think we will have hundreds of post since this seems to be a very popular pastime on DIYaudio in the last few months.
Attachments
SPICE MODELS FOR Sanken 2SC3264/2SA1295
A bipolar output stage appears to be the most efficient topology for a 1ohm load.
The Sanken 2SC3264/2SA1295 appear to be the best available transistors. Any better?
http://www.profusionplc.com/scripts/wsisa.dll/pcatdtl?ipartno=2SC3264
http://www.profusionplc.com/scripts/wsisa.dll/pcatdtl?ipartno=2SA1295
COULD SOMEONE POST SPICE MODELS FOR SANKEN 2SC3264 2SA1295 ....Google came up empty. Thank you.
I will Spice Graham's quasi-complementary design with the 2SC3264, and a JFET input design using a complementary 2SC3264/2SA1295 output.
A bipolar output stage appears to be the most efficient topology for a 1ohm load.
The Sanken 2SC3264/2SA1295 appear to be the best available transistors. Any better?
http://www.profusionplc.com/scripts/wsisa.dll/pcatdtl?ipartno=2SC3264
http://www.profusionplc.com/scripts/wsisa.dll/pcatdtl?ipartno=2SA1295
COULD SOMEONE POST SPICE MODELS FOR SANKEN 2SC3264 2SA1295 ....Google came up empty. Thank you.
I will Spice Graham's quasi-complementary design with the 2SC3264, and a JFET input design using a complementary 2SC3264/2SA1295 output.
Hi jcx,
You are discussing device characteristics for an unproven architecture !
I always found that topology was far more important that device characteristics.
Because of this, I drew your circuit on my simulator, and, as I expected from imagining the current flows - it oscillates. Circa 500kHz.
Is there something special about your IC ( parameters not available to me ), I used TL081C which is not too bad.
Don't trust simulators.
Use them for guidance after you have proved a circuit works in real life.
I have quoted the figures from my simulator for my amplifier circuits. Are these correct ? Surely the distortion must be affected by far more significant physical lead and layout characteristics. My amplifier does not make its presence audibly identifiable at reasonable sound levels with multi-driver loudspeakers, and to me, this is what really counts.
____________________________________________________
Hi LineSource,
Quasi-complementary is NPN+NPN / PNP+NPN output pairing.
The JLH - ALL NPN output stage is NON-COMPLEMENTARY.
If I was buying new output devices I would go for 2SC5200, and that is what in my prototypes. I do not know the 2SC3264, so my second choice would be 2SC3281 which I still have.
Can anyone confirm whether the OnSemi and Toshiba '3281' devices are actually the same. Is it possible the Toshiba were slightly less rugged but had a lower internal capacitance.
LineSource, the NPN/PNP circuit you are proposing to 'Spice' must be completely different from my JLH variant. If your output devices are to be emitter followers on a VAS collector then you are also moving to three stages of amplification, which might lead to a degration of high frequency first cycle distortion.
Cheers .......... Graham.
You are discussing device characteristics for an unproven architecture !
I always found that topology was far more important that device characteristics.
Because of this, I drew your circuit on my simulator, and, as I expected from imagining the current flows - it oscillates. Circa 500kHz.
Is there something special about your IC ( parameters not available to me ), I used TL081C which is not too bad.
Don't trust simulators.
Use them for guidance after you have proved a circuit works in real life.
I have quoted the figures from my simulator for my amplifier circuits. Are these correct ? Surely the distortion must be affected by far more significant physical lead and layout characteristics. My amplifier does not make its presence audibly identifiable at reasonable sound levels with multi-driver loudspeakers, and to me, this is what really counts.
____________________________________________________
Hi LineSource,
Quasi-complementary is NPN+NPN / PNP+NPN output pairing.
The JLH - ALL NPN output stage is NON-COMPLEMENTARY.
If I was buying new output devices I would go for 2SC5200, and that is what in my prototypes. I do not know the 2SC3264, so my second choice would be 2SC3281 which I still have.
Can anyone confirm whether the OnSemi and Toshiba '3281' devices are actually the same. Is it possible the Toshiba were slightly less rugged but had a lower internal capacitance.
LineSource, the NPN/PNP circuit you are proposing to 'Spice' must be completely different from my JLH variant. If your output devices are to be emitter followers on a VAS collector then you are also moving to three stages of amplification, which might lead to a degration of high frequency first cycle distortion.
Cheers .......... Graham.
I like to buy 'murcan.
When I started using bipolars, the Toshibas were not easily available. I've actually never tried the Toshiba parts. On paper they have very slightly less Cob. On the other hand they are rated at 25% less power dissipation, so the die size must be smaller (which explains the capacitance reduction). Also, I've had good luck with the On-Semi and Motorola products -- very consistent and reliable with excellent designs.
When I started using bipolars, the Toshibas were not easily available. I've actually never tried the Toshiba parts. On paper they have very slightly less Cob. On the other hand they are rated at 25% less power dissipation, so the die size must be smaller (which explains the capacitance reduction). Also, I've had good luck with the On-Semi and Motorola products -- very consistent and reliable with excellent designs.
Hi Graham,
I could have predicted the circuit would oscillate with a 3-4 MHz op amp without spice, the compensation networks are still shaping the loop gain in that region
The LT1115 is a variation of the LT1028, more like 40–60 MHz GBW
I’ve zipped up the necessary SwCADIII files below for anyone who finds some value in playing with sims
I haven’t great faith in modern low voltage mosfet models’ analog application accuracy but despite Fred’s sputtering the data sheet graphs suggest the sim # are not off by an order of magnitude
Fred,
You seem to be the one the major issue with the sim #, why don’t you measure one of these devices for us? You seem to be hung up on comparing a 4 yr old, low voltage optimized mosfet with short channel length, fine mesh pitch (possibly even trench rather than hex or rect mesh – IRF seems pretty loose with HEXFET, using it as a trademark rather than descriptive term today) against decades old higher voltage designs with 10x channel R. It’s great to have rules of thumb, particularly simplified device equations, but I see no reason to believe these rules don’t need updating and adjusting for new generations of mosfets that have specifically been tweaked for application in regions not covered by simple long channel fet equations
I certainly am not suggesting simulations are the end of design, they are a tool for exploring and sharing ideas and only the beginning of the engineering process, anyone spending >10% of their engineering effort on simulation are probably wasting that time
I could have predicted the circuit would oscillate with a 3-4 MHz op amp without spice, the compensation networks are still shaping the loop gain in that region
The LT1115 is a variation of the LT1028, more like 40–60 MHz GBW
I’ve zipped up the necessary SwCADIII files below for anyone who finds some value in playing with sims
I haven’t great faith in modern low voltage mosfet models’ analog application accuracy but despite Fred’s sputtering the data sheet graphs suggest the sim # are not off by an order of magnitude
Fred,
You seem to be the one the major issue with the sim #, why don’t you measure one of these devices for us? You seem to be hung up on comparing a 4 yr old, low voltage optimized mosfet with short channel length, fine mesh pitch (possibly even trench rather than hex or rect mesh – IRF seems pretty loose with HEXFET, using it as a trademark rather than descriptive term today) against decades old higher voltage designs with 10x channel R. It’s great to have rules of thumb, particularly simplified device equations, but I see no reason to believe these rules don’t need updating and adjusting for new generations of mosfets that have specifically been tweaked for application in regions not covered by simple long channel fet equations
I certainly am not suggesting simulations are the end of design, they are a tool for exploring and sharing ideas and only the beginning of the engineering process, anyone spending >10% of their engineering effort on simulation are probably wasting that time
Thanks for pointing those differences out. It suggests some application specific variability that's useful. I've been using the On Semi parts, too, but was looking at the 2SC52/2SA1962 equivalents as drivers, since they're also in a smaller variant of the TO3P package, not the TO264. They must be a smaller chip still, though they're rated at 15A, they're spec'd at 130W and 200 pF Cob. Something rather arbitrary or funny about the specs, I'd say.
There's quite a bit of discrepency in the Cob ratings, but then it's just a max rating at On, so who knows?
600pf vs. 270pF? But then On may be using the same number as a max for both npn and pnp. Thanks again.
Regards,
Jon
The Toshiba specs are for typical values, while the On Semi specs are for maximum values. However, if you get the latest (rev D) spec sheet for the On Semi parts, there are now typical capacitance curves on the last page. Then we can see that things are closer:
NPN Toshiba = 200 pF
NPN On Semi = 280 pF
PNP Toshiba = 360 pF
PNP On Semi = 550 pF
If you factor in the difference in power ratings (presumably indicative of die size), they offer basically the same capacitance per watt.
NPN Toshiba = 200 pF
NPN On Semi = 280 pF
PNP Toshiba = 360 pF
PNP On Semi = 550 pF
If you factor in the difference in power ratings (presumably indicative of die size), they offer basically the same capacitance per watt.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.