5 or 10 sec here to get ~-300dB or so math noise error floor. I have never gotten PSPICE to do just that, we have a mode to force a transient analysis to behave as an ideal jitter free A/D.
We need to pick an output stage and bias it for max load wanted, if I read Dick's schematic correctly the output devices are several tens of mA quiescient.
Its 45mA per MJE device. it should be SIM'ed to death and compared. It is simple and very effective but may not be optimized.... I never sim'ed it nor the complete circuit. BTW - I would be honored to have my output stage combined with your gain stages. Should make a unique and killer design performance. -Dick Marsh
Last edited:
Oh come on. Are you saying that you get wildly divergent results for faster sims? If so, something is wrong with your simulator or your use of it.Brad
I said trustworthy.
BTW: my computer and simulator is extremely fast. If you believe that a trustworthy simulation takes a second or two, that is you problem. I've seen enough of such nonsense, be realistic.
agree with Scott: lets pick an ops.
Stein
It is not a problem whatsoever, let alone MY problem. I am confident that Wurcer's results are reliable (there's a lot riding on it!), his software better, and as well he undoubtedly has access to faster machines. My own sims sometimes take several minutes, but not hours, AND they are cross-checked extensively. In many cases the settling of the circuit is crucial before Fourier is even attempted, and there may be some initializations applied to that end. Sometimes step sizes are adjusted downward to see how it affects the results, and of course that slows things down.
Particularly w.r.t. distortion/Fourier, I also look at reality checks like the curve of growth of the distortion with level. For smooth variabilities and nonlinearities, a system with predominant second harmonic will tend to double distortion with a doubled level for example. For a system that is still slowly settling due to long time constants, the Fourier output for a too-short-sim will tend to show nearly equal harmonics of all orders --- a sure indication that something is wrong.
You have to know what you are doing, and as well know what the simulator is leaving out, thermals one of the biggest gotchas in this regard (Scott's simulators however incorporate them explicitly, an enviable adjunct).
The Ti Ni Ag as interface between Si and metal connections is one of the most used vacuum metal deposition in power semiconductor industry.
Who said that Ni is not good for audio?
--- one has to also consider the tolerances and precision being asked for in the time of SIM processing. Check that there isnt somethat is unrealistic in this regard. -RNM
isnt relavent
A semi or tube can be made with just about anything .... they are the amplifying device and so it doesnt matter, esp. with nfb. But when outside the amplifier circuit and with induced signals that werent a part of the amplification chain.... it can matter.
The first comment was from myself about 'induced' currents from nonlinear materials produced distortion in the signal carrying wire. This was reported by researchers in Japan decades ago when steel was within 3 cm of a signal carrying wire. [even though they they couldnt measure the residual harmonic levels that we can today]
As far as Nickel in connectors et al... same issue.... if we are dealing with a SOTA circuit with distortion as low as was simulated here then such things matter or at least be considered if the final result and actual build from using that circuit is to stay as clean of harmonics as shown in the sim. - RNM
The Ti Ni Ag as interface between Si and metal connections is one of the most used vacuum metal deposition in power semiconductor industry.
Who said that Ni is not good for audio?
A semi or tube can be made with just about anything .... they are the amplifying device and so it doesnt matter, esp. with nfb. But when outside the amplifier circuit and with induced signals that werent a part of the amplification chain.... it can matter.
The first comment was from myself about 'induced' currents from nonlinear materials produced distortion in the signal carrying wire. This was reported by researchers in Japan decades ago when steel was within 3 cm of a signal carrying wire. [even though they they couldnt measure the residual harmonic levels that we can today]
As far as Nickel in connectors et al... same issue.... if we are dealing with a SOTA circuit with distortion as low as was simulated here then such things matter or at least be considered if the final result and actual build from using that circuit is to stay as clean of harmonics as shown in the sim. - RNM
Last edited:
I never said seconds by the way. But an hour and a half ---- that's reminiscent of batch processing on a mainframe, and where you come back for output to see people standing around grumbling and wanting to abort your run.
When my previous machine was cluttered with other programs and a polluted hard drive, and the sims required a lot of "out-of-core" operations, it got a little sluggish. But the main delay was usually in the plotting routine, which is frequently the Achilles' heel of many programs.
And you did not answer my question: what happens when you do a shorter run? At what point do you get caca instead of believable results??
No need for any hair raising here, I took your comment literally. From a cold start I would expect to fiddle away an hour or two before I would have presentable results. That said, once I'm happy changing the load (for instance) and doing another run is a matter of seconds.
The thermal sim is nice but you have to compute the coupling matrix yourself and enter it as another layer. Every circuit component has a thermal node (like Qxx_t_deg_k) and once you place a thermal resistance between two devices you have to compute a new thermal resistance to ambient for both and on and on. Someone needs to couple an all physics engine to the process to automate it. Still waiting.
I'll fiddle with sims somemore this weekend. BTW the models Fairchild provides for Dick's trannies have nonsense values for some of the parameters.
Last edited:
A simple transient run takes about 5-10sec.
The result is almost the same if I run it for 10 sec ,2min. or 30min.
BTW that was not the task. I'm not going to quote myself, but to run the whole analysis within 15 sec. is nonsense and wrong, no matter what program or puter you have.
That is what you have to do if you are going to analyse the circuit. (and it takes more than 15 sec.) (starting from scratch)
Stein
Last edited:
doesnt surprise me -
That does not surprise me. Its discusting that we arent shown the correct parameter data. Which lead one (in)famous person to say: "Trust but verify".
That does not surprise me. Its discusting that we arent shown the correct parameter data. Which lead one (in)famous person to say: "Trust but verify".
Last edited:
Onsemi's models of the MJE devices seem much better. I have to laugh though, everyone hooks up their HP semiconductor parameter analyzer and records all the parameters to 8 figure accuracy.
One big problem for audio, extraction of rbb from the terminal I/V characteristics rather than the noise can be way off the right value.
best way to find Rbb
Can you teach the rest, the better and more accurate way via noise? equiv Rbb based on noise values. first and second order.
Pls.
Can you teach the rest, the better and more accurate way via noise? equiv Rbb based on noise values. first and second order.
Pls.
Last edited:
Midband, and at "everyday" temperatures, two primary voltage noise sources, uncorrelated: shot noise in collector current appearing as equivalent input noise via division by transconductance (noise resistance expression involves temperature because of the transconductance dependence); thermal noise in effective base spreading resistance. If beta is quite low, shot and excess noise in base current producing voltage across base resistance.
Since the "half-thermal" of the voltage noise from shot noise is pretty reliable by itself, the thermal noise from rbb' (and hence the value of rbb') can be backed out by root-difference-of-squares from the total voltage noise behavior.
Among other corrections: emitter contact resistance, i.e., not part of the equivalent emitter resistance that is reciprocally proportional to emitter current.
All of the above presumes separating out the parallel noise or "current noise" from the measurements.
Ref. for thorough discussion, including parallel noise and low/high frequency effects: Motchenbacher and Fitchen, Low-Noise Electronic Design, pp. 61-84 (ISBN 0471619507)
Van der Ziel has some good discussion too. If you are lucky the data sheet will have noise figure vs Rs and Ic curves and from this you can de-embed rbb. I showed the fab guys how to do it with a Quantek. Circa 1970 big round emitters were fashionable for matching, very bad for noise (Grey and Meyer has an easy to understand rbb computation from geometry).
Last edited:
I have the ref. Low Noise Electronic Design in my private tech library.
And, i just gave away my transistor Quantek to local college :-(
It used to be that shot noise was considered a process issue and that manufacturing improvements would minimise it. The original LH0002 was a good example of a device with a lot of shot noise. Any thoughts on that?
And, i just gave away my transistor Quantek to local college :-(
It used to be that shot noise was considered a process issue and that manufacturing improvements would minimise it. The original LH0002 was a good example of a device with a lot of shot noise. Any thoughts on that?
Last edited:
Maybe thinking of popcorn noise Richard?
Shot noise is just due to the particulate nature of charge carriers as they cross a potential barrier. In vacuum tubes that are not temperature-limited, there is a space charge smoothing effect that reduces the noise a bit. In semiconductors I don't think there is a comparable reduction effect. JFET gate leakage has at least full shot noise, 2*qe*Io per square root Hz, where qe is the electronic charge in coulombs and Io is the magnitude of the current.
Interesting sidelight: when I was trying to find my way in the early days with charge preamps, within the narrow field of instrumentation for astronomy there was precious little published. A paper that managed to get into a prestigious Academic Press collection, on vidicons and their preamps, from a couple of guys at Princeton, had a totally wrong theory of JFET noise which I trusted initially, and which said that the channel noise was due to shot noise. Somehow they missed the paper by van der Ziel from 1962 with the correct theory
I finally found lots of good work that was applicable to my problem in publications from nuclear science, although their detectors had typically a unipolar fast charge output, and I needed to handle bipolar signals with high accuracy.
Shot noise (sqrt(2*qI)) comes from first principles, the statistical limit on noise from uncorellated events. Many devices have excess noise usually at low frequencies, or just noise in general from poor design or necessary tradeoffs.
We called them "Dirty noises", i.e. noises caused by non-essential (what is English term?) charge carriers, as result of odd dopes. I heard an anecdote, when suddenly "dirty noises" increased in production, and during the investigation they found finally that the only thing that changed, was boyfriend of one lady. New boyfriend did not like her perfume. After that usage of perfumes had been banned.