Howdy,
I'm a retired analog IC designer who has finally found the time to play around with an interest in audio. I just found this forum, and suspect that I can learn a lot from you folks.
I've made a first cut at an all-bipolar power amplifier using LTspice. I've run into the problem of choosing various commercially available transistors that have Spice models. How do audio designers generally go about choosing them?
What is a "CFA" amplifier?
Alan
I'm a retired analog IC designer who has finally found the time to play around with an interest in audio. I just found this forum, and suspect that I can learn a lot from you folks.
I've made a first cut at an all-bipolar power amplifier using LTspice. I've run into the problem of choosing various commercially available transistors that have Spice models. How do audio designers generally go about choosing them?
What is a "CFA" amplifier?
Alan
Hi alanF00
Look at the web sites of the big manufacturers e.g. ST, ON Semi and Diodes (zetex transistors). They usually provide links to their audio power transistors and then to SPICE models.
ON Semi models have not always been the most reliable, they seem to have outsourced the modelling to a third party which worked from the data sheet limits. As you no doubt know we need typical as well as worst case models. So it is well worth checking each device you choose individually in test circuits to see if you think the model is correct, first.
I suspect many on here don't do that though ... I'm guilty too until things don't seem right.
Look at the web sites of the big manufacturers e.g. ST, ON Semi and Diodes (zetex transistors). They usually provide links to their audio power transistors and then to SPICE models.
ON Semi models have not always been the most reliable, they seem to have outsourced the modelling to a third party which worked from the data sheet limits. As you no doubt know we need typical as well as worst case models. So it is well worth checking each device you choose individually in test circuits to see if you think the model is correct, first.
I suspect many on here don't do that though ... I'm guilty too until things don't seem right.
Regarding On Semi power devices and other common types used in DIY audio amplifiers, member Bob Cordell has modified and posted quite a few verified models for LTSpice on his website. CordellAudio.com - Home
Current Feedback Amplifier:
https://www.ieee.li/pdf/viewgraphs/current_feedback_vs_voltage_feedback_amplifiers.pdf
http://buildaudioamps.com/wp-content/uploads/2015/04/Alexander Current FB Amp.pdf
(PDF) Current-feedback architecture for high-slew-rate and low-THD high-end audio amplifier
http://www.analog.com/media/en/reference-design-documentation/design-notes/dn46fa.pdf
It can be a controversial topology for an audio power amplifier. One of the things that sometimes gets argued about around here.
Last edited:
Thanks, huggygood!
john_ellis wrote:
I'll do that. So far, I've looked at the offerings of the big distributers like Mouser and Digikey, but been baffled at the wide variety of devices.
In the companies I've worked for, modeling of IC processes was a very big deal, and much effort was put into making worst case models. The IC industry would fail without it.
In the commercial world, where IC volumes can run into the tens of millions per year, worst case design is crucial, but in the world of DIY audio, I suspect that it's nowhere nearly as important.
At two large-volume IC manufacturers where I've worked, if wafer yields dropped to below 99%, people would start to panic. But at the relatively low-volume Tektronix (chip volumes ranged from a few hundred to a few tens of thousands per year), up until the early 1990s, yields of 50% to 75% were acceptable. With low volumes, putting a lot of effort into worst case modeling was not considered cost effective.
Ian Finch wrote:
Also, I dabbled with PSpice a bit, but found LTspice much more friendly. I found that there was a big library of devices that came with the PSpice/Oracle Lite suite, which contain models ending with the name "_Cordell". The models directly from Cordell's website end in "C", and some of the internal parameters are different between the two sets of devices. Any comments?
Markw4 wrote:
In the op-amp world, CFA's have certain advantages and disadvantages. Their loop response is pretty insensitive to external gain setting resistor values, which is good. But they don't usually have as good a set of important specs as regular op-amps.
I used to work for a guy who, at Comlinear Corp. in the 1980s, developed some of the earliest CF op-amps. I believe they were somewhat disappointed that CFA's didn't take over the world.
Alan
john_ellis wrote:
I'll do that. So far, I've looked at the offerings of the big distributers like Mouser and Digikey, but been baffled at the wide variety of devices.
So I've gathered. I stumbled on the Cordell models by a circuitous route several days ago.
Yes, that's good advice. But after poking around in model files that I got from PSpice and Cordell's website, I don't see any models labeled as typical or worst case. Comments?
In the companies I've worked for, modeling of IC processes was a very big deal, and much effort was put into making worst case models. The IC industry would fail without it.
It's a lot of effort doing worst case modeling, so that's understandable. One problem, though, seems to be that manufacturers like ON Semi don't seem to provide the data to do that.
In the commercial world, where IC volumes can run into the tens of millions per year, worst case design is crucial, but in the world of DIY audio, I suspect that it's nowhere nearly as important.
At two large-volume IC manufacturers where I've worked, if wafer yields dropped to below 99%, people would start to panic. But at the relatively low-volume Tektronix (chip volumes ranged from a few hundred to a few tens of thousands per year), up until the early 1990s, yields of 50% to 75% were acceptable. With low volumes, putting a lot of effort into worst case modeling was not considered cost effective.
Ian Finch wrote:
Yes, I stumbled on that a few days ago.
Also, I dabbled with PSpice a bit, but found LTspice much more friendly. I found that there was a big library of devices that came with the PSpice/Oracle Lite suite, which contain models ending with the name "_Cordell". The models directly from Cordell's website end in "C", and some of the internal parameters are different between the two sets of devices. Any comments?
Markw4 wrote:
I'll definitely have to look into that.
In the op-amp world, CFA's have certain advantages and disadvantages. Their loop response is pretty insensitive to external gain setting resistor values, which is good. But they don't usually have as good a set of important specs as regular op-amps.
I used to work for a guy who, at Comlinear Corp. in the 1980s, developed some of the earliest CF op-amps. I believe they were somewhat disappointed that CFA's didn't take over the world.
Alan
This is a very handy link for selecting "popular" transistors for driver, VAS, small signal (input & current/voltage sources)
users.tpg.com.au/gerskine/greg/driver%20transistors.htm
I copied this with fingers from another computer, so if it doesn't work look up the greg erskine transistor tables using a search engine.
These popular values are often stocked, but a bit expensive.
Other transistors are often cheaper, or much easier to get.
For example, 2sc/2sa numbers are generally NOT AVAILABLE in the USA, but fairchild sometimes sells similar transistors with a KSC/KSA prefix.
Some of us primitives that build projects with a soldering iron, reading glasses, a saw & a drill, don't simulate anything, and just buy whatever has right Vceo, Ic, power disippation, package, gain, Ft & stock level. My board really doesn't sound that bad - - -
You've heard from the real experts above, though: Ellis, Finch, I'm quite impressed. I'm just a record/CD fan who hates running to the store to buy another wonderproduct that is designed to blow up in five years. My projects wear like iron, and stopped going up in flames when I started reading around here.
users.tpg.com.au/gerskine/greg/driver%20transistors.htm
I copied this with fingers from another computer, so if it doesn't work look up the greg erskine transistor tables using a search engine.
These popular values are often stocked, but a bit expensive.
Other transistors are often cheaper, or much easier to get.
For example, 2sc/2sa numbers are generally NOT AVAILABLE in the USA, but fairchild sometimes sells similar transistors with a KSC/KSA prefix.
Some of us primitives that build projects with a soldering iron, reading glasses, a saw & a drill, don't simulate anything, and just buy whatever has right Vceo, Ic, power disippation, package, gain, Ft & stock level. My board really doesn't sound that bad - - -
You've heard from the real experts above, though: Ellis, Finch, I'm quite impressed. I'm just a record/CD fan who hates running to the store to buy another wonderproduct that is designed to blow up in five years. My projects wear like iron, and stopped going up in flames when I started reading around here.
Last edited:
Hi Alan
I agree, the device models out there do not seem to cover a range of values. I also worked in the semiconductor industry where a lot of effort went into worst case and typical models. On the other hand manufacturing always aimed to target centre of spec. to maximise yield.
Fortunately the majority of discretes do seem to be represented by "typical" models these days. I reduce the gain variation by choosing e.g. "B" grades for small signal, e.g. BC547B/BC557B and where manufacturers specify the gain groups in higher power transistors, those too e.g BD139-16, BD140-16 for drivers.
Though you might always get devices that are worst case, on occasions. For serious production that is a QA job in incoming.
I agree, the device models out there do not seem to cover a range of values. I also worked in the semiconductor industry where a lot of effort went into worst case and typical models. On the other hand manufacturing always aimed to target centre of spec. to maximise yield.
Fortunately the majority of discretes do seem to be represented by "typical" models these days. I reduce the gain variation by choosing e.g. "B" grades for small signal, e.g. BC547B/BC557B and where manufacturers specify the gain groups in higher power transistors, those too e.g BD139-16, BD140-16 for drivers.
Though you might always get devices that are worst case, on occasions. For serious production that is a QA job in incoming.
This is a rather wierd URL. A search turned up this:
404 Not Found transistors.htm
The space between "vas" and "transistors" gets translated by a copy/paste to "%20", like this:
VAS Transistors
Thanks for that information! I wondered about the 2sc versus ksc nomenclature, since some datasheets from Fairchild are title with both.
If it works . . .
Having been in the analog chip industry for more than 30 years, simulation has been the name of the game. You cannot design working chips without it. So I'm quite prejudiced in favor of simulating discrete audio designs. Assuming the models are valid, as an old colleague from Tektronix used to say: "Spice knows all!"
One company that I worked for was a bit lazy about simulation, and its modeling group produced somewhat deficient models. Most chips had 3-5 go-rounds as a result. They saw nothing wrong with their tradition of silicon breadboarding. The other companies I worked for had a culture where it was a black mark against you if your chip didn't come out perfectly the first time. Time lost in go-arounds is extremely expensive in lost sales.
Well I'm just playing around with audio because I've always had an interest, and now have the time to delve more deeply into it.
Standard industry practice. Occasionally, though, for experimental purposes designers might request that a particular fab run be skewed in certain directions. Sometimes such variations became standard.
I'll keep that in mind.
Obviously the playing around most people here are doing is rather different from commercial production. I'm used to simulating all manner of worst case corners, so the lack of good, trustworthy models is a bit bothersome.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.