Yes, it is. You can see that it is match terminated with the 50 ohm resistor but I was thinking of leaving it out and just building it on a BNC to banana jack adapter so there is no need for a cable and termination. Also, this oscillator has a DC offset control, so with the 1K base resistor we get 1mA base bias current per volt of DC offset across that resistor eliminating the need for the negative supply. I'm also planning to use .1 uF for the bypass as you mentioned.
2.5 mA? Do you mean ball park? Ie would be about 1.9v/510 = 3.7 mA
Last edited:
I just read through the last few pages, and noted Ostrippers comments on the 340/350. I did not enjoy his success with these parts in the VAS stage - I almost always got HF parastic oscillation just as the amp was comming off its peak -ve output voltage (which is where I had it configured as a current source). I tried loads of things to try to cure it (base stoppers, changed Iq, played with loop comp etc etc). It was particularly bad when driving IRFP250/9240. Switched to BF469/470 (I got a pile from a friend at Philips) and the problem was cured. I suspect now that what was hapening was that Ccb was changing very dramatically through the voltage swing, and on the signal peaks its value was much higher. The 469/470's are HF video driver devices (now obsolete), and they never had this issue. I'd be interested to here any other feedback (BTW, after this I abondoned 'blameless' style amps and went FBS - but I guess the same issues would apply)
The end of an era
I learned yesterday that TI is acquiring National Semiconductor. This indeed would seem to be a sad end of an era. National has always been one of the top, premier IC companies in linear ICs.
Perhaps this should be no surprize, given the layoffs that happened awhile ago at National of some top analog designers, including those involved with their high-end audio ICs. I hope that their legendary products will not be discontinued. I guess there is some hope there if one looks at what happened to BB when they were acquired by TI.
I guess the great National Semiconductor name will disappear after about a year. What a pity.
I know, I have a soft spot in my heart for the way things were.
Best,
Bob
I learned yesterday that TI is acquiring National Semiconductor. This indeed would seem to be a sad end of an era. National has always been one of the top, premier IC companies in linear ICs.
Perhaps this should be no surprize, given the layoffs that happened awhile ago at National of some top analog designers, including those involved with their high-end audio ICs. I hope that their legendary products will not be discontinued. I guess there is some hope there if one looks at what happened to BB when they were acquired by TI.
I guess the great National Semiconductor name will disappear after about a year. What a pity.
I know, I have a soft spot in my heart for the way things were.
Best,
Bob
TI's acquisition of Burr Brown didn't kill the products. Nor the name.
The volume of analog EVMs (including BB) I have to test each day
is becoming an extreme challenge. I'm sure its the same at all of TI's
CMs. Analog is huge at TI, and they are not going to mess this up.
My personal reaction was, "there go my weekends"...
The volume of analog EVMs (including BB) I have to test each day
is becoming an extreme challenge. I'm sure its the same at all of TI's
CMs. Analog is huge at TI, and they are not going to mess this up.
My personal reaction was, "there go my weekends"...
Last edited:
I work on RF. When Ti bought some years ago the small Chipcon we wonder what the product were going to be. In fact TI maintained the development center in Norway with the line of product written" Chipcon from TI". New products are an evolution of the product line with CCxxxx reference. in fact portfolio has increased an impressive way with TI as if additionnal resources have been injected. From a competitor point of view (I work for an old american semiconductor company that makes also mobile phones...) I can say the result is very good. Sometime I dream we would make such win-win deals...
Hi Bonsai,
Its true that the Ccb will increase as the VAS transistor has less Vce, but I'm not sure the 340/350 would be so much worse that it would provoke or allow oscillation in a circuit that was otherwise adequately stable. In other words, the overall circuit might have been on the edge even with the VAS transistors that worked. It sounds like the amp you had this trouble with was a MOSFET design. MOSFETs can be more prone to oscillation because they are faster, and if the gate stopper resistors (you mentioned base stoppers, so there may be some confusion on my part) are not large enough and/or other HF precautions for the MOSFETs are not in place, there could be interactions with the VAS current source transistor. Also, check what kind of current source you are using and whether it might have some possible Miller effects going on from its collector to base.
Also, try to determine if this is a global oscillation or a local oscillation. If you double the closed loop gain (e.g., double the value of the feedback resistor), does the oscillation go away? If it does, that suggests that the oscillation is in the global feedback loop. If it does not go away, it may be a local oscillation resulting from the interaction of the current source and the output stage. Also bear in mind that the drain-gate capacitance of the output MOSFETs gets quite large when you get near the rails and Vdg gets small.
Cheers,
Bob
MJE340/350 are good for higher voltage.
Otherwise I prefer BD139/140.
Think it is the higher gain that makes BD139/140 win over MJE340/350.
Many amplifiers have lower voltage than 80 Volt, the max of BD139/140
I have tried both options in SPICE amplifiers.
And BD139/140 gets substansially lower THD distortion than MJE340/350.
Otherwise I prefer BD139/140.
Think it is the higher gain that makes BD139/140 win over MJE340/350.
Many amplifiers have lower voltage than 80 Volt, the max of BD139/140
I have tried both options in SPICE amplifiers.
And BD139/140 gets substansially lower THD distortion than MJE340/350.
Yes I experienced the same, a quite stable design with the BD's (all BJT, no mosfet) turns unstable when replacing BD139/140 with MJE340/350, it is necessary to add lots of Miller capacitance to stabilize, I guess that the MJE's has a less doped collector (350V Vceo) that induces huge Miller capacitance variations. So for less than 80V swing go BD's, also THD can be 1 order of magnitude lower.
Hi lineup,
Yes, the BD139/140 are good VAS devices. I'll be including models of them in my upcoming next release of device models to my web page at CordellAudio.com - Home.
Cheers,
Bob
Poor models
As many of you know, I often complain about the shortcomings of transistor SPICE models provided by vendors. Here's an example that certainly competes for the prize. This morning I was working on the BD139/140 models. I usually start with models from one or more manufacturers and SPICE them on several testbench circuits and compare the results to the datasheet and/or measurements I have made. The SPICE models for these devices from a well-known semiconductor company that will remain nameless had the usual set of numbers to four or so significant digits, but did not contain at all one of the most parameters: TF, which defines ft. What frustration!
Cheers,
Bob
As many of you know, I often complain about the shortcomings of transistor SPICE models provided by vendors. Here's an example that certainly competes for the prize. This morning I was working on the BD139/140 models. I usually start with models from one or more manufacturers and SPICE them on several testbench circuits and compare the results to the datasheet and/or measurements I have made. The SPICE models for these devices from a well-known semiconductor company that will remain nameless had the usual set of numbers to four or so significant digits, but did not contain at all one of the most parameters: TF, which defines ft. What frustration!
Cheers,
Bob
Design Support
The factors leading to this shameful situation are legion. About 15 - 20 years ago I got the impression that some manufacturers wanted to have SPICE models of their parts just so marketing could say, "We have SPICE models.".
A few incarnations ago I worked with an engineer who said that composing SPICE models had been his assignment in a summer job while a college student. He worked from Data Sheets, and wondered if anybody had ever verified his work against real parts.
In 1975, there may have been some justification for publishing "abbreviated" models. Since your department was probably charged by the millisecond for CPU time on the System 360/65, it made sense to have a simplified model that would compute quickly and tell you if things like bias points and basic operation were correct . . . and then you'd use a more complete model to investigate the places where your sliderule and nomographs said there might be a problem. That work environment ended about the time a math coprocessor became available for the PC/AT.
With RARE exceptions (Burr-Brown comes to mind, but I'm too lazy to verify it at the moment) SPICE models almost never tell you what characteristics are, or aren't, modeled - or how accurately. Users are left to wonder if a simulation anomaly is real, or an artifact of the simulation process, or the result of some un-modeled behavior. I believe some of Analog Devices' products have sets of models covering nominal, worst-case, and temperature extremes of behavior, but I don't recall others who followed suit. Again, the end user is (at best) stuck with creating and executing a process for model verification before he can use the model for serious design work.
It's bad enough that simplified, and inaccurate, models from 20 or 30 years ago will probably survive (thanks to the Internet) through a Nuclear Winter. Some seem to have acquired the ability to spontaneously regenerate when they were handed out with the "free" versions of P-CAD, Electronics Workbench, et al. Data Sheets seem to get revised every few years (at least to correct typos), but once a SPICE model has been created it's probably never revised. What irks me even more is that even rather recent models suffer the deficiencies of the early efforts. C'mon, guys - it was in the early 1990's (!!) when Alexander and Bowers (at PMI/Analog Devices), and Biagi, Baker et al (at Burr-Brown) published App Notes that make significant corrections and much-needed enhancements to the Boyle modeling approach for opamps.
p.s. - I have started putting the oscillator section of your THD Analyzer into LTspice.
That spec seems to be missing from several prominent manufacturers' Data Sheets, too - so perhaps the model maker is simply reflecting that omission.
The quality of SPICE models in general could become a never-ending rant. On more than one occasion I have been heard to say something like, "Yes, I know that the CAD/CAE program cost as much as an engineer's salary, and you ran it on a Personal Confuser that cost more than my car . . . BUT . . . No, in spite of what the simulation says, that uA741 will NOT supply enough current to jump-start a locomotive!". (Yes, there ARE opamp SPICE models that have no limits on output voltage or current.)
The factors leading to this shameful situation are legion. About 15 - 20 years ago I got the impression that some manufacturers wanted to have SPICE models of their parts just so marketing could say, "We have SPICE models.".
A few incarnations ago I worked with an engineer who said that composing SPICE models had been his assignment in a summer job while a college student. He worked from Data Sheets, and wondered if anybody had ever verified his work against real parts.
In 1975, there may have been some justification for publishing "abbreviated" models. Since your department was probably charged by the millisecond for CPU time on the System 360/65, it made sense to have a simplified model that would compute quickly and tell you if things like bias points and basic operation were correct . . . and then you'd use a more complete model to investigate the places where your sliderule and nomographs said there might be a problem. That work environment ended about the time a math coprocessor became available for the PC/AT.
With RARE exceptions (Burr-Brown comes to mind, but I'm too lazy to verify it at the moment) SPICE models almost never tell you what characteristics are, or aren't, modeled - or how accurately. Users are left to wonder if a simulation anomaly is real, or an artifact of the simulation process, or the result of some un-modeled behavior. I believe some of Analog Devices' products have sets of models covering nominal, worst-case, and temperature extremes of behavior, but I don't recall others who followed suit. Again, the end user is (at best) stuck with creating and executing a process for model verification before he can use the model for serious design work.
It's bad enough that simplified, and inaccurate, models from 20 or 30 years ago will probably survive (thanks to the Internet) through a Nuclear Winter. Some seem to have acquired the ability to spontaneously regenerate when they were handed out with the "free" versions of P-CAD, Electronics Workbench, et al. Data Sheets seem to get revised every few years (at least to correct typos), but once a SPICE model has been created it's probably never revised. What irks me even more is that even rather recent models suffer the deficiencies of the early efforts. C'mon, guys - it was in the early 1990's (!!) when Alexander and Bowers (at PMI/Analog Devices), and Biagi, Baker et al (at Burr-Brown) published App Notes that make significant corrections and much-needed enhancements to the Boyle modeling approach for opamps.
p.s. - I have started putting the oscillator section of your THD Analyzer into LTspice.
According to info provided to me from phillips the BD139 is BC639 in a TO126 case. The ft of bc639 is 100 Mhz at 10ma and cob figures are also available for the BC639 which is around 7pf at -10v Vcb. Likewise the BD140 is the BC640. This was done to achieve higher dissipation. These transistors might be old but they arent half as bad as some think them to be.
Hi dchisholm,
I agree with everything you have said. The situation is probably worse with things like op amp macro-models than with transistors. I believe that with transistors, they just subcontract out the model generation and a machine with some algorithm that does not necessarily apply any common sense just spits out silly numbers with many significant digits. Then the vendor just throws the model over the wall without even checking to see if the model yields results that are in reasonable accordance with the device spec sheet. No regression testing is done, and much better results would be achieved if even a college summer student was hired to look at the models. I guess the vendors just don't care.
Cheers,
Bob
Hi David,
I'm guessing that is what the manufacturers are doing, but without much success. Maybe they just need better algorithms. Even if they did not do modeling measurements and just extracted the data from the curves in their datasheets, and could create model parameters that would reasonably match the datasheet info, that would be nice.
Cheers,
Bob